Tsavdaridis 2015

S
Safety Assessment of Piled Buildings The failures were often accompanied by set-
in Liquefiable Soils: Mathematical tlement and tilting of the superstructure, render-
Tools ing it either useless or very expensive to
rehabilitate after the earthquake. Following the
Subhamoy Bhattacharya 1995 Kobe earthquake, investigations have been
University of Surrey, Guildford, UK carried out to find the failure pattern of the piles.
Piles were excavated or extracted from the sub-
soil, borehole cameras were used to take photo-
Synonyms
graphs, and pile integrity tests were carried out.
These studies hinted at the location of the cracks
Case study of pile damage during the 1995 Kobe
and damage patterns for the piles. Of particular
earthquake; Geotechnical analysis of pile foun-
interest is the formation of plastic hinges in the
dation in liquefiable soils; Numerical Modeling;
piles. This indicates that the stresses in the pile
Pile foundations in liquefiable soils; Reliability of
during and after liquefaction exceeded the yield
pile foundations; Seismic Analysis of Pile Foun-
stress of the material of the pile, despite large
dations; Soil-Structure Interaction
factors of safety which were employed in the
design. Hinges were found to have occurred
Introduction at various depths along the pile: at the pile head,
at the middle of the liquefiable layer, and toward
Piles are routinely used as foundations to support the interface of liquefiable/non-liquefiable layer.
short-to-medium span bridges and buildings typ- In this context it must be mentioned that piles
ically over four stories and other structures. Col- are currently designed with adequate factor of
lapse and/or severe damage of pile-supported safety for geotechnical load-carrying capacity
structures is still observed in liquefiable soils (maximum allowable load in the pile) and against
after most major earthquakes such as the 1995 bending failure. Bending moment can occur in a
Kobe earthquake (Japan), the 1999 Kocaeli earth- pile due to lateral loads arising from: (a) inertia
quake (Turkey), and the 2001 Bhuj earthquake load acting at the pile head and (b) kinematic
(India). The failures not only occurred in laterally loads from the ground. Further discussion on the
spreading (sloping) ground but were also dynamics of the problem can be found in
observed in level ground where no lateral spread- Adhikari and Bhattacharya (2008).
ing would be anticipated. A good discussion on Liquefaction of soil around the pile can
the failure modes can be found in Bhattacharya affect a pile-supported structure in the following
and Madabhushi (2008). ways:
# Springer-Verlag Berlin Heidelberg 2015
M. Beer et al. (eds.), Encyclopedia of Earthquake Engineering,
DOI 10.1007/978-3-642-35344-4
2414 Safety Assessment of Piled Buildings in Liquefiable Soils: Mathematical Tools
1. A pile will be laterally unsupported in the design codes, which advocate bending mecha-
zone of liquefaction, i.e., no lateral restraint nism as the main design consideration, recent
from the soil to the pile in the liquefied zone. research showed that an axially loaded pile can
As a result, a pile may be vulnerable to be laterally unsupported in liquefied soils and is
buckling instability if the axial load is high susceptible to buckling failure.
enough and the unsupported length of pile is Buckling instability under the interaction of
sufficient. axial and lateral loads can be a more critical
2. The pile will lose its shaft resistance in the design consideration because of its sudden nature
zone of liquefaction, i.e., no liquefied soil- and sensitivity to imperfection; see, for example,
pile friction, and as a result the axial load on Dash et al. (2010). More recently, Bhattacharya
the pile in the liquefiable zone will increase. et al. (2009) included the effects of dynamics on
3. The time period of the structure will change due the combined axial and lateral loads on a pile
to liquefaction as the foundation becomes flex- foundation. Essentially, piles in liquefied soils
ible. While calculating the period of a building may be better regarded as columns carrying lat-
under non-liquefied condition, it is assumed eral loads rather than laterally loaded beams.
that foundation is rigid, and as a result only In design, beam bending and column buckling
the dimensions of the buildings are adequate are approached differently. Bending is a stable
to obtain the period. However, as soon as a soil mechanism as long as the pile remains elastic
liquefies, the time period may increase. and secondary failure (e.g., local buckling) is not
4. The damping of a pile-supported structure also a possibility. This failure mode depends on the
increases during and after liquefaction. bending strength (e.g., yield moment capacity
and plastic moment capacity) of the member
The above mechanisms were verified experi- under consideration. In contrast, buckling is an
mentally by carefully designed model tests and unstable mechanism and it occurs suddenly and
can be found in Bhattacharya (2003), drastically when the elastic critical load is reached.
Bhattacharya et al. (2004), Bhattacharya It is the most destructive mode of failure and
et al. (2005), and Lombardi and Bhattacharya depends on the geometrical properties of the mem-
(2014). Bhattacharya (2003) showed through ber, i.e., slenderness ratio, rather than the member
dynamic centrifuge tests that axial load alone strength. For example, steel pipe piles with iden-
can cause a pile to fail if the surrounding soil tical length and diameter but having different yield
liquefies in an earthquake and the mechanism strengths (e.g., 200 MPa, 500 MPa, and
being buckling instability. 1,000 MPa) will buckle at similar axial loads but
Lombardi and Bhattacharya (2014) showed can resist different amounts of bending. In other
through high-quality shaking table tests that the words, bending failure may be avoided by increas-
time period of pile-supported structures will ing the yield strength of the material, but it may not
increase owing to liquefaction. In the experi- suffice to avoid buckling. To prevent buckling
ments, the soil was liquefied progressively failure, there should be a minimum pile diameter
through a broadband white noise signal with depending on the depth of liquefiable soils. There-
increasing magnitude. Furthermore, they also fore, designing against bending would not auto-
showed that the overall damping ratio of the matically satisfy buckling requirements. It is
structures may increase in excess of 20 %. envisaged that there are plenty of existing pile-
These have important design implications. supported structures that may need retrofitting.
All current design methods, such as JRA This entry therefore describes a probabilistic
(1996), NEHRP (2000), IS 1893 (2001), and and a deterministic method to assess the reliability
Eurocode 8 (CEN 2004), focused on bending of pile foundation for a scenario earthquake.
strength of the pile to avoid bending failure due Before the methodology is described, a discussion
to lateral loads (combination of inertia and lateral is presented on the main loading on pile founda-
spreading). In contrast to these conventional tions during earthquakes in liquefiable soils.
Safety Assessment of Piled Buildings in Liquefiable Soils: Mathematical Tools 2415
Methodology Formulation
Bhattacharya (2006) discusses the deterministic
Different Stages of Loading on a Pile- approach to determine the factor of safety of
Supported Structure During Earthquakes pile foundations against the buckling instability
Figure 1 shows the different stages of loading failure. Bhattacharya and Goda (2013)
of a pile-supported structure during a seismic developed a probabilistic procedure for determin-
liquefaction-induced event. Pgravity (Stage I) repre- ing the occurrence of a buckling failure of
sents the axial load on the piles in normal condition. existing piled foundations due to a scenario
This can be estimated based on static equilibrium. earthquake.
This axial compressive load may increase/decrease The methodology is based on assessing two
further by Vinertial due to inertial effect of the super- length parameters:
structure and kinematic effects of the soil flow past
the foundation. Hinertial is the inertial lateral loads 1. Critical length of a pile denoted (HC). Essen-
due to the oscillation of superstructure (Stage II and tially this is the unsupported length of the pile
III). Ground movement causes kinematic loads on that can sustain without collapse due to com-
the pile foundations. This load can be of two types: bined axial and lateral loading. This can be
transient (during shaking, due to the dynamic estimated based on manipulation of the
effects of the soil mass) and residual (after the Euler’s buckling load equation considering
shaking ceased due to soil flow, often known as the correct boundary condition of the pile.
“lateral spreading”) (Stage IV). The various forms Specifically HC depends on the type and
of feasible failure mechanisms of pile foundations dimension of superstructure (bridge or build-
are shear failure, bending failure, buckling instabil- ing), bending stiffness, axial load acting
ity, and dynamic failure. on the pile, dynamic characteristics of
Pgravity Pgravity + Vinertial Pgravity + Vinertial Pgravity + Vinertial
Hinertial Hinertial Hinertial
Loose Liquefied Liquefied

sand
S
sand sand
Stage I Stage II Stage III Stage IV

Before Shaking starts. Soil liquefies. On sloping ground
earthquake on Soil yet to liquefy Vertical inertial Soil liquefies. Lateral
level ground forces act with spreading may combine
gravity. Piles may with behaviour
starts to buckle or in stage III
settle
Safety Assessment of Piled Buildings in Liquefiable Soils: Mathematical Tools, Fig. 1 Different stages of
loading and failure mechanism of pile during earthquake
Safety Assessment of
Piled Buildings in
Liquefiable Soils:
Mathematical Tools,
Fig. 2 Concept of critical
length (HC) and
unsupported length (DL)
superstructure, and boundary conditions of the methodology and tool); and (iii) the probabilistic
pile at the top and bottom of the liquefiable framework provides a straightforward way to
layer. integrate geotechnical assessment tools into
2. Length of unsupported pile due to liquefaction existing probabilistic seismic hazard methods as
(DL): this can be obtained based on the depth well as decision-support tools for implementing
of liquefaction due to the design earthquake. earthquake risk mitigation measures (Fig. 3).
Figure 2 shows the concept of the two lengths. Step-by-Step Methodology

Comparing HC with DL, potential failure of the The steps are:
underground pile due to buckling is predicted
when HC < DL (see Figs. 1 and 2). In reliability (a) Estimation of peak ground acceleration
terms, HC is the capacity variable, while DL is the (PGA) at building/bridge site: The assess-
demand term, and thus the failure criterion can be ment of DL requires (i) estimation of ground
regarded as the limit state function: motion parameters (typically, peak ground
acceleration (PGA)) at a building site due to
g ¼ ðH C D L Þ (1) a specified scenario and (ii) assessment of
liquefaction initiation potential at different
Probabilistic and Deterministic Approach depths along the pile length. Equation 2 is a
In a deterministic approach, we obtain a single typical equation to obtain PGA.
value of the two parameters and can obtain the
factor of safety against failure. However, the Log PGAbldg ¼ f Mw , Rbldg , V S30, bldg þ ebldg (2)
advantages of assessing using probabilistic
approach are (i) the outcome is expressed as a where PGAbldg is PGA value at the building
(estimated) likelihood of failure by taking into site, Mw is the moment magnitude, R is the
account various sources of uncertainty involved distance measured (typically closest distance
in the assessment; (ii) sensitivity analysis can be to fault rupture plane), VS30 is the average
conducted to identify key factors that affect the shear-wave velocity in the uppermost 30 m
outcome (the results are useful to improve the in (m/s), and ebldg is the random error that is
Safety Assessment of Piled Buildings in Liquefiable Soils: Mathematical Tools, Fig. 3 Schematic diagram
showing the methodology
modeled as a normal variable with zero mean studies, Cetin et al. (2002, 2004) developed a
and logarithmic standard deviation. statistical model for calculating the probabil-
(b) Assessment of the depth of liquefiable soil: ity of liquefaction initiation based on SPT
The next step is to conduct probabilistic liq- data, while Moss et al. (2006) developed a
uefaction potential evaluation of a soil column counterpart using CPT data. Using the first-
(along a pile) at the building site to estimate order reliability method, Juang et al. (2005)
DL for a given seismic excitation level. Such developed a similar model based on VS data.
assessment is often conducted by using sim- The significance of these developed models is
plified stress-based methods of Seed and Idriss that key uncertainties associated with the input
(1971) based on standard penetration test data/parameters and the adopted models them-
(SPT) data, cone penetration test (CPT) data, selves are taken into account; the developed
and shear-wave velocity (VS) data. Recently, models can produce unbiased potential of liq-
probabilistic procedures for liquefaction initi- uefaction initiation and are useful to conduct
ation have been developed by considering dif- probabilistic liquefaction hazard analysis S
ferent in situ measures for describing soil (Goda et al. 2011). The probability of lique-
strength. Using a sophisticated Bayesian faction initiation PL at a depth of interest can
regression analysis and well-screened case be estimated as
0 !
N 1, 60 ð1 þ 0:004FCÞ 13:32ln CSReq 29:53lnðMw Þ 3:70ln sv =Pa þ 0:05FC þ 16:85
PL ¼ F
2:7
(3)
where F is the standard normal function, adjusted for fines content), FC is the fines
N1,60 is the corrected SPT counts (but not content (in percentage), CSReq is the cyclic
Piled Buildings in
Liquefiable Soils:
Mathematical Tools,
Fig. 4 Depth required for
fixity a pile
stress ratio (but not adjusted for the moment each pile beneath a building, assuming that
magnitude), and s0 v is the vertical effective each pile is equally loaded during static con-
stress. dition and neglecting any eccentricity of
(c) Unsupported length of the pile (DL): The loading. During an earthquake, inertial action
next step is to determine the unsupported pile of the superstructure imposes the dynamic
length (DL) based on the liquefaction profile. axial load on the piles, which will increase
DL is equal to the thickness of liquefied soil the axial load on some piles. These piles with
layers plus some additional length necessary increased axial loads may be vulnerable to
for fixity at the bottom of the liquefied soils. buckling. An estimate of the maximum axial
Typical calculations show that the fixity compressive load acting on a pile can be
depth is about three to five times the diameter given by
of the pile (see Fig. 4); see Bhattacharya and
Goda (2013). If a relatively thin non-liquefied Pdynamic ¼ ð1 þ aÞPstatic (4)
layer is overlain and underlain by thick liq-
uefied layers, lateral restraint of the pile at the where a is termed as the dynamic axial load
non-liquefied layer might not be expected, factor and is a function of type of superstructure,
i.e., the pile is unsupported. In such a case, height of the center of mass of the superstructure,
the thin non-liquefied layer can be ignored in and characteristics of the earthquake shaking
determining DL, and the unsupported length (e.g., frequency content and amplitude).
needs to be extended until a thick For buckling instability analysis, each pile
non-liquefied layer is encountered. However, needs to be evaluated with respect to its end
if there is a complex soil profile with alter- conditions, i.e., fixed, pinned, or free. Each pile
nating liquefiable and non-liquefiable soil in a group of identical piles will have the same
layers, more detailed analysis is required. buckling resistance as a single pile. If a group of
(d) Assessment of the critical pile length HC: piles is fixed in a stiff pile cap and embedded
This section describes the mathematical sufficiently at the tip, as in Fig. 2, the pile group
background behind the rationale of obtaining will buckle in side sway.
the critical pile depth HC. Before the onset of The elastic critical load of a single pile, Pcr,
shaking, the static axial load Pstatic acts on can be estimated as
Safety Assessment of Piled Buildings in Liquefiable Soils: Mathematical Tools, Table 1 Values of K in Eq. 7
Boundary condition of the pile at the top and bottom
of the liquefied layer Effective
Top Bottom length K Examples
Fixed Fixed [sufficient Leff = 0.5Hc 0.5 Pile groups with raked piles
embedment at the
dense layer]
Free to translate but Pinned [insufficient Leff = 2Hc 2 See, for example, NFCH (Niigata
restrained against embedment at the Family Court House) building in
rotation – sway frame dense layer] Bhattacharya (2003)
Free to translate but Fixed [sufficient Leff = Hc 1 Most cases fall under such category;
restrained against embedment at the see, for example, Fig. 2
rotation – sway frame dense layer]
Fixed in direction but free Fixed [sufficient Leff = 0.7Hc 0.7 Pile groups with raked piles. Improper
to rotate embedment at the pile-pile cap connection
dense layer]
Fixed in direction but free Pinned [less Leff = Hc 1 Pile groups with raked piles. Improper
to rotate embedment at the pile-pile cap connection
dense layer]
Free, i.e., unrestrained Fixed [sufficient Leff = 2Hc 2 Piles in a row such as the Showa Bridge
against rotation and embedment at the piles
displacement dense layer]
p2 EI i.e., Pfailure < Pcr. Equally, the growth of plastic

Pcr ¼ 2
(4) bending zones reduces the effective elastic mod-
ðLeff Þ
ulus of the section, thereby resulting in the
where Leff is the effective length, i.e., Euler’s decreased critical load for buckling (i.e., capac-
equivalent buckling length of a strut pinned at ity). Furthermore, these processes feed each
both ends, and EI is the bending stiffness of the other, and as elastic critical loads are approached,
pile. The effective length of the pile (Leff) can be all bending effects are magnified. Stability anal-
found in Table 1 or from any structural mechan- ysis of elastic columns showed that if lateral
ics textbook or codes of practice. The loads in the absence of axial load would create a
unsupported length of the pile DL is equal to the maximum lateral displacement d0 in the critical
thickness of liquefiable soil plus some additional mode shape of buckling, then the displacement d
length necessary for fixity at the bottom of the under the same lateral loads but with the concur-
liquefiable soil. rent axial load P is given by
The applicability of the elastic critical load, as S
in Eq. 4, to pile buckling failure is an important d 1
¼ (5)
factor. Experiments show that the actual failure d0 1 P=Pcr
load of a slender column is much lower than that
predicted by Eq. 4. Rankine (1866) recognized The term d/d0 is the buckling amplification factor
that the actual failure involves an interaction (i.e., amplification of lateral displacement due to
between elastic and plastic modes of failure. Lat- the presence of the axial load). Figure 5 presents a
eral loads and inevitable geometrical imperfec- graph of the buckling amplification factor plotted
tion lead to creation of bending moments in against the normalized axial load P/Pcr, where
addition to axial loads. Bending moments have P denotes the applied axial load. It can be
to be accompanied by stress resultants that dimin- observed from Fig. 5 and Eq. 6 that if the applied
ish the cross-sectional area available for carrying load is 50 % of Pcr, the amplification of lateral
the axial load; thus the actual failure load is likely deflection due to lateral loads is about two times.
to be less than the elastic critical load, At these large deflections, secondary moments
Piled Buildings in
Liquefiable Soils:
Mathematical Tools,
Fig. 5 Buckling
amplification factor versus
normalized axial load
will be generated due to P-D moment, leading to Determination of “Critical Depth” HC

more deflection. It is therefore important to In the limit state condition of failure,
remain in the linear regime and not in anyway Pdynamic = Pfailure. For the type of structure
near the asymptotic region, where the buckling shown in Fig. 2, Leff = HC in Eq. 4. In order to
amplification factor increases dramatically (e.g., generalize the boundary condition of the pile (i.e.,
P/Pcr > 0.6). pile head fixity condition with pile cap/super-
Moreover, it would also be unwise to use a structure and the fixity at the interface between
factor of safety less than three against the Euler liquefiable and non-liquefiable layers at deeper
load of a pile, i.e., (P/Pcr = 0.33). Such consid- depths), one may write Eq. 7:
eration is consistent with general design practice
where structural engineers use a factor of safety Leff ¼ K H C (7)
of at least three against linear elastic buckling to
take into account the eccentricity of load, deteri- where K is the column effective length factor
oration of elastic stiffness due to plastic yielding, (e.g., K = 1 for free lateral translation but
and unavoidable imperfection. The actual failure restrained against rotation-sway motion). Values
load Pfailure is therefore some factor f (f < 1) of K for other boundary conditions of the pile are
times the theoretical Euler’s buckling load given given in Table 1.
by Eq. 5. With the abovementioned assumptions, Eq. 6
can be rewritten as
Pfailure ¼ fPcr (6)
fp2 EI
Pdynamic ¼ fPcr ¼ (8)
Based on the above discussion, it may be inferred K 2 H 2C
that buckling instability is initiated at around
0.35, i.e., f = 0.35. It is noted that in reality, Rearranging Eq. 8 gives the estimate of the crit-
this factor depends on the axial load, imperfec- ical depth HC for a pile as follows:
tion of piles, and residual stress in the pile due to
driving. The selection of f is one of the signifi- sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
fp2 EI fp2 EI
cant sources of uncertainty in determining the HC ¼ ¼ (9)
2
K Pdynamic K ð1 þ aÞPstatic
2
critical pile depth HC.
Safety Assessment of Piled Buildings in Liquefiable Soils: Mathematical Tools, Fig. 6 Loading condition of pile
foundation in pre-liquefaction and post-liquefaction stages
Therefore, the assessment of HC is based on the the maximum axial force acting on a pile due to
calculation of the critical buckling load for a pile inertia can be calculated as
foundation surrounded by liquefied soils (from
which the pile cannot receive sufficient lateral sup-
b1 ðW=gÞ SA T pre , xpre Dfix þ b2 H B
port). Key parameters in Eq. 9 are f, K, and a. Ppre ¼
Reasonable values of f and K can be selected N p =2 B
based on engineering judgment. (10)
Based on the procedure described above, both
DL and HC can be assessed probabilistically. The
where W is the total weight of a building (note:
occurrence of pile foundation failure due to buck-
W = PstaticNp), SA(Tpre, xpre) (g) is the spectral
ling mechanism is indicated if HC < DL. By
acceleration at Tpre with damping ratio xpre
sampling DL and HC many times, probabilistic
(typically, 2–5 %), b1 is the coefficient to account
assessment of liquefaction-induced pile founda-
for modal mass for the fundamental vibration
tion failure due to buckling mechanism can be
mode (typically, 0.8–0.9), b2 is the coefficient to
carried out.
account for the effective height where the inertia
due to the modal mass acts (typically, 0.65–0.75),
Formulation for Determination of Dynamic HB is the height of building, Np is the number of S
Load Amplification Factor (a) piles (assuming that an equal number of piles are
One of the critical factors is the assessment of a. positioned in two rows), and B (m) is the founda-
This can be facilitated by examining axial forces tion width between the two rows of piles (note:
acting on a pile in pre-liquefaction and post- the foundation width B is along the direction
liquefaction situations as shown in Fig. 6. In a where axial force is induced by overturning
pre-liquefaction stage, the plane of fixity of a moment due to lateral inertia). Tpre can be esti-
building with pile foundation surrounded by mated by using an empirical equation, such as
non-liquefied soils can be estimated by following Tpre = 0.09HB/B0.5; see Anderson et al. (1952)
a similar procedure as shown in Fig. 4, i.e., depth also adopted in IS 1893.
from ground surface to the plane of fixity is In a post-liquefaction situation, the building is
denoted by Dfix in Fig. 6 and is typically a few supported by piles that have relatively long
meters. By assuming that the natural vibration unsupported lengths of DL (as evaluated based
period of the building before liquefaction is Tpre, on the liquefaction initiation analysis). In this
case, the natural vibration period in a post- A case study: collapse of a 5-story RCC
liquefaction stage, Tpost, can be calculated as building in Kobe during the 1995 Kobe earth-
quake: Several buildings with pile foundation
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
W=g had to be demolished due to severe liquefaction
T post ¼ 2p (11) damage during the 1995 Mw6.9 Kobe earthquake.
N p 12EI=D3L
Because of the dramatic consequences, several
detailed post-earthquake investigations were
where 12EI/(DL)3 is the lateral stiffness of each
conducted to examine the cause and failure
pile. Then, the maximum axial force acting on a
mechanism of these cases (Tokimatsu
pile is given by
et al. 1997). For illustration, one of such case
studies (Uzuoka et al. 2002; Bhattacharya 2006)

ðW=gÞ SA T post , xpost ðDL þ b3 HB Þ is focused upon. A 5-story reinforced concrete
Ppost ¼
N p =2 B frame building (total height HB = 14.5 m) was
located at 6 m from the quay wall on a reclaimed
(12)
fill in the Higashinada area of the Kobe City; the
distance from the rupture plane to the building
where SA(Tpost, xpost) (g) is the spectral accelera- site was about 5 km (Fig. 7). The Kobe earth-
tion at Tpost with damping ratio xpost (typically, quake caused a lateral displacement of 2 m to the
10–30 % representing the damping of liquefied quay wall toward the sea, and the building was
soil) and b3 is the coefficient to account for the tilted by 3 due to lateral spreading. The sche-
effective height where the inertia acts in a post- matics of the post-earthquake investigation of the
liquefaction condition (typically, 0.5). The building and pile foundation are shown in Fig. 8.
underlying assumption of Eqs. 11 and 12 is that At the building site, significant lateral spreading
the building is considered as a rigid mass and the was observed (about 1.0–1.5 deformation/move-
pile provides the primary lateral stiffness to ment of the ground toward the sea; Tokimatsu
the building. It must be mentioned that this et al. 1997).
approach is very simple, and many uncertainties, The building was supported on 38 hollow
such as effects of vertical inertia, timing of pre-stressed concrete piles (there were two pile
max inertia and timing of loss of lateral support rows separated by 7.5 m and 19 piles were
due to liquefaction, and potential effects due aligned in each row); the pile length was 20 m
to lateral spreading, are not taken into with exterior and interior diameters of 0.4 and
consideration. 0.24 m, respectively. Figure 9 shows variations of
By defining a = max(Ppre/Pstatic, Ppost/Pstatic), soil profiles and SPT N counts with depth at the
a is given by: building site. The site has fill/sand layers with
relatively low N counts (e.g., 2–9 m and
2
a¼ max b1 SA T pre , xpre ðDfix þ b2 H B Þ, 12–16 m), which are susceptible to liquefaction
gB (i.e., saturated sand layers with low strength); the

SA T post , xpost ðDL þ b3 HB Þ average shear-wave velocities in the uppermost
(13) 12 and 30 m are estimated to be about 147 and
216 m/s, respectively. The water table level was
The additional information needed for evaluating about 2 m below ground surface. The post-
a is the estimated spectral acceleration values at earthquake investigation by Tokimatsu
vibration periods Tpre and Tpost with damping et al. (1997) indicated that soil layers shallower
ratios xpre and xpost (note: Tpre and Tpost are ran- than 9 m were liquefied (based on a simplified
dom variables; in particular, Tpost is significantly stress method). Moreover, the borehole logging
affected by liquefaction initiation analysis). data shown in Fig. 8 suggest that the sandy silt
Safety Assessment of Piled Buildings in Liquefiable Soils: Mathematical Tools, Fig. 7 Location of the building
site and observation site
layer between 12 m and 16 m, having low The 5 % damped response spectra of the two
strength, may be disturbed by strong ground horizontal components and their geometric mean
motion; thus this layer may not offer much fixity are presented in Fig. 9 and are compared with the S
to the pile. median GMPE (Ground Motion Prediction Equa-
The nearby ground motion recording was tions) by Zhao et al. (2006); this relation is used
obtained at the Higashi Kobe Bridge, which is as a representative regional model throughout
away about 0.9 km from the building site; dis- this study to estimate ground motion parameters
tance to the rupture plane for the observation and at the building site. The comparison of the calcu-
building sites is about 5.2 and 4.4 km, respec- lated response spectra with the Zhao et al. relation
tively. The site condition at the observation site is indicates that the observed response spectra
similar to that at the building site (typically, have less spectral content at vibration periods
NEHRP site class D or E). The recorded accelera- less than 1.0 s, while they contain rich spectral
tion time histories at the observation site are shown content at vibration periods greater than 1.0 s.
in Fig. 9 (Public Works Research Institute 1995). The responses at short vibration periods are likely
Piled Buildings in
Liquefiable Soils:
Mathematical Tools,
Fig. 8 Post-earthquake
investigation of a case
study (Tokimatsu et al.
1997)
Piled Buildings in
Liquefiable Soils:
Mathematical Tools,
Fig. 9 Boring log of the
soil at the building site
to be affected by nonlinear site amplification (i.e., hypocenter, and the observation site (Fig. 6) are of
de-amplification), and at such a site, liquefaction- typical “near-fault motions” due to forward direc-
induced ground failure may be expected. Further tivity (Mavroeidis and Papageorgiou 2003); this
to note, geographical positions of the fault plane, can be corroborated by large response spectra
Safety Assessment of Piled Buildings in Liquefiable Soils: Mathematical Tools, Table 2 Summary of proba-
bilistic information of input parameters
Coefficient of [Lower and upper Distribution
Parameter Mean variation limits] type
Moment magnitude Mw 6.9 0.1a [6.6, 7.2] Normal
Peak ground acceleration PGA Equation 2 – Lognormal
(g)
Water table level (m) 2.0 – [1.0, 3.0] Uniform
FC (%) Fig. 9 0.1 – Lognormal
N count Fig. 9 0.15 – Lognormal
Vertical total stress sv (Pa) –b,c 0.1 – Lognormal
Vertical effective stress s’v (Pa) –b,c 0.15 – Lognormal
Pre-liquefaction period Tpre (s) 0.5 0.1 – Lognormal
EI of a pile (MNm2) 32.35 0.1 [24.26, 48.53] Lognormal
Static axial force per pile Pstatic 412 0.1 [309, 618] Lognormal
(kN)
a
This is the standard deviation
b
It depends on the water table
c
Dry and wet soil densities are set to 1.76 and 1.92 g/cm3
Safety Assessment of Piled Buildings in Liquefiable Soils: Mathematical Tools, Fig. 10 Ground motion time
history (a) and 5 % damped response spectra (b) at the Higashi Kobe Bridge observation site
values at long periods (Fig. 9b) and by for DL and HC which shows the high likelihood
inspecting velocity time histories of the acceler- that the building collapsed due to buckling
ation data where large velocity pulses are clearly of the piles (probability of buckling failure is
visible. Table 2 provides a summary of the 0.943). Further details of this methodology can
input parameters used in the analysis, and be found in Bhattacharya and Goda (2013)
Fig. 11 shows the probability density function (Fig. 10).
Piled Buildings in
Liquefiable Soils:
Mathematical Tools,
Fig. 11 Comparison of
probability mass functions
for DL and HC
Summary ▶ Seismic Reliability Assessment, Alternative

methods for
A probabilistic-based method to reassess such ▶ Seismic Vulnerability Assessment: Reinforced
safety has been formulated in this entry. This Concrete Structures
method can easily be coded in a program, such
as MATLAB or Fortran. This method checks
References
the stability of the foundation against buckling
instability at full liquefaction, i.e., when the Adhikari S, Bhattacharya S (2008) Dynamic instability of
soil surrounding the pile is at its lowest pos- pile-supported structures in liquefiable soils during
sible stiffness. Two parameters, namely, “crit- earthquakes. Shock and Vibration 16(6):665–685
ical depth of the pile HC” and the Anderson AW, Blume JA, Degenkolb HJ, Hammill HB,
Knapik EM, Marchand HL, Powers HC, Rinne JE,
“unsupported length of the pile due to lique- Sedgwick GA, Sjoberg HO (1952) Lateral forces of
faction DL,” are estimated. Critical depth is a earthquake and wind. Trans ASCE 117:716–80
function of axial load acting on the pile (P), Bhattacharya S (2003) Pile instability during earthquake
flexural stiffness of the pile (EI), and the liquefaction. Ph.D Thesis, University of Cambridge,
U.K.
boundary condition of the pile above and Bhattacharya S (2006) Safety assessment of existing piled
below the liquefiable soil. On the other hand, foundations in liquefiable soils against buckling insta-
DL mainly depends on the earthquake charac- bility. ISET J Earthquake Technol 43:133–147
teristics, soil profile, and ground conditions. Bhattacharya S, Goda K (2013) Probabilistic buckling
analysis of axially loaded piles in liquefiable soils.
A case study is considered to illustrate an Soil Dyn Earthquake Eng 45:13–24
application of the methodology. Bhattacharya S, Madabhushi SPG (2008) A critical review
of methods for pile design in seismically liquefiable
soils. Bull Earthquake Eng 6:407–446
Bhattacharya S, Madabhushi SPG, Bolton MD (2004) An
alternative mechanism of pile failure in liquefiable
Cross-References deposits during earthquakes. Geotechnique
54:203–213
▶ Damage to Buildings: Modeling Bhattacharya S, Madabhushi SPG, Bolton MD
(2005) Reply to the two discussions on the paper “An
▶ Geotechnical Earthquake Engineering: Dam- alternative mechanism of pile failure in liquefiable
age Mechanism Observed deposits during earthquakes”. Geotechnique
▶ Seismic Collapse Assessment 55(3):259–263
SAR Images, Interpretation of 2427
Bhattacharya S, Adhikari S, Alexander NA liquefaction-induced lateral spreading in 1995

(2009) A simplified method for unified buckling and Hyogoken-Nambu earthquake. J Struct Constr Eng,
free vibration analysis of pile-supported structures in AIJ 495:95–100
seismically liquefiable soils. Soil Dyn Earthquake Eng Uzuoka R, Sento N, Yashima A, Zhang F (2002) 3-dimen-
29:1220–1235 sional effective stress analysis of a damaged group-
CEN (2004) Eurocode 8, design of structures for earth- pile foundation adjacent to a quay wall. J Jpn Assoc
quake resistance – Part 1: general rules, seismic Earthq Eng 2:1–14
actions and rules for buildings, EN 1998-1:2004. Zhao JX, Zhang J, Asano A, Ohno Y, Oouchi T, Takahashi
Comite Europeen de Normalisation, Brussels, T, Ogawa H, Irikura K, Thio HK, Somerville PG,
Belgium Fukushima Y, Fukushima Y (2006) Attenuation rela-
Cetin KO, Der Kiureghian A, Seed RB (2002) Probabilis- tions of strong ground motion in Japan using site
tic model for the initiation of seismic soil liquefaction. classification based on predominant period. Bull
Struct Safety 24:67–82 Seismol Soc Am 96:898–913
Cetin KO, Seed RB, Der Kiureghian A, Tokimatsu K,
Harder LF Jr, Kayen RE, Moss RES (2004) Standard
Penetration Test-based probabilistic and deterministic
assessment of seismic soil liquefaction potential.
J Geotech Geoenviron Eng 130:1314–1340
Dash SR, Bhattacharya S, Blakeborough A (2010) SAR Images, Interpretation of
Bending-buckling interaction as a failure mechanism
of piles in liquefiable soils. Soil Dyn Earthquake Eng
Antonio Iodice and Daniele Riccio
30:32–39
Goda K, Atkinson GM, Hunter JA, Crow C, Motazedian D Dipartimento di Ingegneria Elettrica e delle
(2011) Probabilistic liquefaction hazard analysis for Tecnologie dell’Informazione, Università degli
four Canadian cities. Bull Seismol Soc Am 101:190– Studi di Napoli Federico II, Napoli, Italy
201
IS 1893 (2001) Indian Standard for Seismic Design
JRA (1996) Japanese Road Association, Specification for
Highway Bridges, Part V, Seismic Design Introduction
Juang CH, Yang SH, Yuan H (2005) Model uncertainty of
shear wave velocity-based method for liquefaction
A synthetic aperture radar (SAR) is a remote
potential evaluation. J Geotech Geoenviron Eng
131:1274–1282 sensing microwave imaging system which con-
Lombardi D, Bhattacharya S (2014) Modal analysis of sists of a radar sensor mounted on a moving
pile-supported structures during seismic liquefaction. platform, such as an airplane or a satellite.
Earthquake Eng Struct Dyn 43(3):119–138.
As the platform flies along an approximately
doi:10.1002/eqe.2336
Mavroeidis GP, Papageorgiou AS (2003) A mathematical straight line, the radar emits microwave pulses
representation of near-fault ground motions. Bull at fixed rate (pulse repetition frequency (PRF))
Seismol Soc Am 93:1099–1131 and receives corresponding returns (echoes)
Moss RES, Seed RB, Kayen RE, Stewart JP, Der
backscattered by the illuminated scene. A SAR
Kiureghian A, Cetin KO (2006) CPT-based probabi-
listic and deterministic assessment of in situ seismic is able to distinguish points at different distances
soil liquefaction potential. J Geotech Geoenviron Eng from the line of flight based on different delays of S
132:1032–1051 their returns; in addition, points at different posi-
NEHRP (2000) National Earthquake Hazards Reduction
tions along a direction parallel to the line of flight
Program (NEHRP, 2000): commentary for Federal
Emergency Management Agency (FEMA, USA 369) are distinguished by forming a very long (and
on seismic regulations for new buildings and other therefore very directive, with a very narrow
structures beam) synthetic array. This is obtained by prop-
Public Works Research Institute (1995) Strong-motion
erly combining pulses received by the sensor at
acceleration records from Public Works in Japan (No.
21). Ministry of Construction, Tsukuba, Japan different positions along the line of flight, so that
Rankine WJM (1866) Useful rules and tables. London the synthetic array length is equal to the length of
Seed HB, Idriss IM (1971) Simplified procedure for eval- the portion of line of flight such that a given
uating soil liquefaction potential. J Soil Mech Found
ground point remains within the real antenna
Div 97:1249–1273
Tokimatsu K, Ohoka H, Shamoto Y, Asaka Y (1997) beamwidth. In this way, a two-dimensional
Failure and deformation modes of piles due to (2D) image is obtained, which is the projection
2428 SAR Images, Interpretation of
of the scene onto the plane containing the look sensor-to-scene distance, at variance with the
direction (range direction) and the line of flight one of optical sensors. In order to obtain a higher
(azimuth direction). This is at variance with opti- (better) resolution, although with a smaller illu-
cal images, which are the projection of the scene minated scene, a “spotlight” acquisition mode
onto the plane perpendicular to the look direction. can be used, in which the SAR antenna beam is
A SAR is an active sensor (i.e., it illuminates steered during the flight to constantly illuminate a
the scene), so that it can work during both the day given spot on the ground. In this way, a longer
and the night; in addition, it relies on micro- synthetic array can be obtained, this implying a
waves, which can propagate through clouds, so better azimuth resolution. The resolution of mod-
that it can image the Earth’s surface even in the ern spaceborne SAR systems spans from about
presence of a cloud cover. 10 m to fractions of meter.
An important parameter characterizing an A SAR image provides information on the
imaging sensor is its resolution, which is the imaged scene which is in some sense comple-
minimum distance between two points such that mentary with respect to that provided by an opti-
they can be distinguished by the sensor. For a cal image. In fact, while the intensity of a pixel in
SAR sensor, the resolution along the range direc- an optical image mainly depends on the chemical
tion is c/(2B), where c is the speed of light and properties of the surface of the imaged objects,
B is the pulse bandwidth, whereas the resolution the intensity of a pixel in a SAR image depends
along the azimuth dimension is lr/(2X), where l on electromagnetic properties (permittivity and
is the electromagnetic wavelength, r is the conductivity) of imaged objects and on their
sensor-to-ground distance, and X is the synthetic roughness at wavelength (i.e., centimetric)
array length. In its usual acquisition mode (“strip scale: smooth surfaces (calm water, concrete or
map” mode), the SAR radar antenna is constantly asphalt surfaces, etc.) appear as dark areas on the
pointed along a direction perpendicular to the line image, whereas surfaces with increasing rough-
of flight, forming a significant angle (“look ness appear as increasingly bright areas.
angle”) with the nadir direction (i.e., SAR is a Finally, another important peculiarity of SAR
“side looking” sensor); see Fig. 1. In this case, the sensors with respect to optical ones is their
synthetic antenna length X is lr/L, where L is the “coherent” nature: in fact, they are able to emit
real antenna effective azimuth length, so that the a coherent radiation and to measure not only the
azimuth resolution turns out to be L/2. Note intensity of the received signal but also its phase.
that the SAR resolution is independent of the This allows using interferometric techniques
(interferometric SAR (InSAR) and differential
interferometric SAR (DInSAR)) to obtain terrain
topography and to monitor small terrain move-
ments and tomographic techniques (SAR
Tomography) for the three-dimensional recon-
struction of imaged objects. In addition, using
two orthogonally polarized transmitting and/or
receiving antennas, information can be extracted
by observing how the polarization of the
backscattered wave is modified with respect to
the transmitted one (SAR Polarimetry). How-
ever, the coherent nature of SAR also causes the
appearance of the “speckle” noise, which gives a
“salt-and-pepper” look to SAR images: a macro-
scopically homogeneous area appears to be com-
SAR Images, Interpretation of, Fig. 1 Geometry of posed of pixels of randomly varying intensity.
SAR acquisition Speckle noise can be reduced, at the expense of
SAR Images,
Interpretation of,
Fig. 2 Geometric
distortions.
Foreshortening-
compression (A’-B’),
foreshortening-dilation
(B’-C’), layover (E’-D’),
shadow (D’-G’)
geometric resolution, by averaging over adjacent If the surface is tilted toward the sensor and its
pixels (multi-look image), or by proper filtering. slope is larger than the look angle, then there is an
By summarizing, SAR sensors have signifi- “inversion” of the SAR geometry: the positions
cant advantages with respect to optical ones along the range direction of peaks and bases of
(day-and-night, all-weather capabilities, coherent hills or mountains are exchanged, and the sides of
nature), but SAR images are more difficult to be hills or mountains are “folded” onto the valleys in
visually interpreted than optical ones, due to both front of them (so that a single pixel corresponds
geometrical (i.e., image projection plane includ- to both an area on the hill’s side and an area on the
ing the look direction) and radiometric (i.e., valley, and a very bright area on the image
involved dependence of image intensity on ter- appears) (see Figs. 2 and 3). This effect is called
rain electromagnetic and roughness properties, “layover.”
speckle) issues. Fundamentals on SAR systems Finally, if the surface is tilted away from the
and applications can be found, e.g., in Ulaby sensor and its slope is larger than 90 minus the
et al. (1986), Elachi (1988), and Curlander and look angle, then a portion of the surface is not
McDonough (1991). ▶ InSAR and ▶ SAR illuminated (see Fig. 2), and no return is present
Tomography are the subjects of other entries of in that portion of the SAR image: a “shadow”
this encyclopedia, whereas SAR Polarimetry is appears.
analyzed in detail in Lee and Pottier (2009). If SAR platform trajectory and terrain topog-
raphy are known, abovementioned geometric dis-
tortions can be mitigated by a post-processing
Foreshortening, Layover, Shadow, and step, called “geocoding,” that allows representing S
Geocoding the SAR images in a standard cartographic map
projection. In this way, the image can be, for
Due to the fact that a SAR image is the projection instance, easily integrated in a GIS. Note how-
of the scene onto a plane including the look ever that, although foreshortening can be
direction, geometric distortions on SAR images corrected by geocoding, layover and shadow
are very different from those experienced in opti- effects imply a loss of information that cannot
cal images. be recovered by this post-processing step.
If terrain slope is smaller than the look angle y, In addition, geocoding implies an interpolation
the resolution cell on the ground is compressed process that may alter the image information
with respect to the horizontal terrain case if the content: therefore, in some applications it may
surface is tilted toward the sensor; otherwise it is be preferable to extract the physical parameter of
dilated (see Fig. 2.). This effect is termed interest directly from the image in SAR native
“foreshortening.” geometry and then to geocode the obtained final
SAR Images, Interpretation of, Fig. 3 COSMO/ same image (right). The very bright area near the crater
SkyMed SAR image of the area of Mt. Vesuvius, Italy corresponds to a layover area
(left). Near range is on the left. Geocoded version of the
SAR Images,
Interpretation of,
Fig. 4 Elements on a
building facade forming
dihedral and trihedral
structures
map (Guida et al. 2008; Di Martino et al. 2012). very-high-resolution (VHR) SAR images of
An example of SAR image before and after urban areas have become routinely available.
geocoding is shown in Fig. 3. In particular, in the spotlight acquisition mode,
COSMO/SkyMed SAR sensors are able to obtain
a resolution even better than 1 m. Accordingly, in
Interpreting SAR Images of Buildings principle a lot of information on objects present
in the urban scenario can be extracted from such
Geometric distortions described in the previous images; however, due to the above-cited severe
section are particularly severe in SAR images of geometric distortions and due to the involved
urban areas, which are of prominent interest if, interaction between incident electromagnetic
for instance, a fast post-event earthquake damage wave and imaged scene, direct interpretation of
assessment is needed. With the launch of VHR SAR images is not straightforward. It is
COSMO/SkyMed and TerraSAR-X missions, easy to realize (see Figs. 4 and 5) that such
SAR Images, Interpretation of, Fig. 5 SAR image formation for a single building
images are dominated by the combination of sin- bounce paths have the same length,
gle scattering from terrain and buildings and mul- corresponding to twice the path length from the
tiple scattering from dihedral and trihedral sensor to the internal edge of the dihedral, and
structures. For moderate- and low-resolution sys- hence they reach the sensor simultaneously.
tems, they may be simultaneously present in a Accordingly, a very bright line appears on the
single resolution cell. Nonetheless, for very-high- image. Similarly, if we consider a trihedral struc-
resolution SAR systems, the resolution cell is so ture, for a wide range of structure orientations, all
small that dihedral and trihedral returns dominate triple-bounce paths have the same length,
with respect to the single scattering background. corresponding to twice the path length from the
Accordingly, a realistic description of VHR SAR sensor to the internal corner of the trihedral, and
amplitude images can be obtained by considering hence they reach the sensor simultaneously.
sparse brilliant points, or lines, over a dark back- Accordingly, a very bright point appears on the
ground: the positions of these brilliant points and image. Since balconies and windows usually are
lines can be considered randomly distributed, spatially distributed in an ordered way, they form
unless they belong to a building facade, in periodic patterns of brilliant lines and/or points
which case an ordered, periodic spatial distribu- within the layover area of the SAR image of the
tion is expected. building.
In fact, if we consider a single building, as By continuing to move from near to far range, S
illustrated in Fig. 5, by moving from near to far the end of the layover area is marked by a bright
range, we first find a layover area in which each line representing the double-bounce return from
pixel is the superposition of two or three contri- the large dihedral structure formed by the vertical
butions: one from the ground, one from the ver- wall and the ground. This line is particularly
tical wall, and possibly one from the roof. Ground evident if the wall is aligned with the sensor line
and roof are usually so smooth that corresponding of flight. Then, a dark area is present, including
contributions are weak, whereas often, on the the very weak triple-bounce return (wall-ground-
vertical walls, dihedral or trihedral structures of wall and/or ground-wall-ground) and, possibly,
balconies and windows are present (see Fig. 4) the weak return from the roof. And, finally, we
whose contributions to the backscattered signal find a very dark shadow area.
are significant. In fact, with regard to dihedral An example of VHR SAR image of an urban
structures, it is easy to realize that, if the structure area is reported in Fig. 6a. This is a 1-m resolution
is aligned with the SAR line of flight, all double- TerraSAR-X image of Naples, Italy. Near range
SAR Images, Interpretation of, Fig. 6 TerraSAR-X image of Naples, Italy (a). Near range is on the left. Excerpt of
the previous image showing the area of Piazza del Plebiscito (b). An optical image of the same area (c)
is on the left. In Fig. 6b an excerpt of the previous this building, by moving from left to right (i.e.,
image is reported, and in Fig. 6c an optical image from near to far range), we can distinguish three
of the same area is shown. The square at the bright lines, followed by a very bright line and by
center of this area is Piazza del Plebiscito, and a dark area. The three bright lines correspond
the building at its right side is the Palazzo Reale to an architectural structure at the roof edge
(Royal Palace). In agreement with our previous and to two lines of balconies, whereas the very
discussion, in correspondence of the façade of bright line corresponds to the wall-ground
double-bounce return. Finally, the dark area cor- HH (i.e., transmit a horizontally polarized elec-
responds to building roof and shadow. tromagnetic field and measure the horizontally
polarized component of the received field), VV
(i.e., transmit a vertically polarized electromag-
Model-Based Interpretation of SAR netic field and measure the vertically polarized
Images component of the received field), and HV or VH
(i.e., transmit a horizontally polarized electro-
Above presented description allows a qualitative magnetic field and measure the vertically polar-
interpretation of SAR images of natural and ized component of the received field, or vice
urban areas. Quantitative information on the versa). For a wide range of scattering surfaces
geometry (i.e., distances between objects, build- (reciprocal scatterers), HV and VH channel
ing size, floors’ height, etc.) can be also obtained returns are equal, so that only one of the two
from such an analysis if look angle and pixel can be actually used. A very useful combination
spacing are known. However, in order to obtain consists of HH + VV, HH-VV, and HV channels,
quantitative relations between image intensity and it is called a “Pauli decomposition” (Lee and
and scene properties, a deeper analysis is needed. Pottier 2009). In fact, it turns out that the sum of
First of all, the scene must be described in terms HH and VV returns is dominated by single scat-
of parameters of interest: for instance, for a nat- tering from rough surfaces (soil surfaces, sea
ural scenario, soil moisture and composition, ter- surfaces), the difference of HH and VV returns
rain roughness, vegetation biomass, etc. Then, a is dominated by double scattering (terrain-
direct electromagnetic scattering model must be building walls, or ground-tree trunks), and HV
used to express the backscattered field in terms of return is dominated by volumetric scattering or
such scene parameters. Inversion of this model extremely rough surface scattering (vegetation).
allows retrieving the scene parameters of interest Accordingly, an RGB color image can be
from SAR images of the scene. obtained by loading the HH + VV signal onto
This field is currently the subject of intense the blue channel, HH-VV onto the red channel,
research activity, and the description of scattering and HV onto the green one. Accordingly,
models and retrieval algorithms goes beyond the blue areas on the image will correspond to bare
scope of the present work. However, a review of or little vegetated soils, or sea; red areas to
scattering models can be found in some textbooks built-up areas or trees with little foliage; and
(e.g., Ulaby et al. 1986; Tsang et al. 2000; green areas to very vegetated soils or forests.
Franceschetti and Riccio 2007) and scientific Intermediate colors will correspond to pixels
papers (e.g., Fung et al. 1992; Franceschetti containing combinations of the previous targets.
et al. 2002). Examples of retrieval algorithms An example of false color SAR image obtained
can also be found in the scientific literature, by using the Pauli decomposition is reported in S
both for natural scenes (Iodice et al. 2011; Di Fig. 7.
Martino et al. 2012) and urban areas Another possibility is to load, onto the three
(Guida et al. 2008, 2010). different color channels, SAR images of the same
area acquired at different times. This allows to
easily identify areas subjected to changes
RGB Compositions between two different acquisitions. For instance,
in Fig. 8 a false color SAR image of an area in a
Visual interpretation of SAR images can be made semiarid region (Tougou basin, Burkina Faso) is
easier by combining different SAR images of the shown. An image acquired during the dry season
same area to create a false color image. is loaded onto the blue channel, another image
For instance, if a SAR polarimetric system is acquired in the wet season is loaded onto the
employed, a combination of different polarimet- green channel, and the interferometric coherence
ric channels can be used. Available channels are between the two acquisitions (see ▶ InSAR and
SAR Images, Interpretation of, Fig. 7 False color image obtained by performing a Pauli decomposition on the
polarimetric SAR image of an agricultural area
SAR Images, Interpretation of, Fig. 8 False color SAR image of the Tougou basin area (Burkina Faso)
A-InSAR: Theory) is loaded onto the red chan- Green areas: wet season vegetation (bright on wet
nel. Accordingly, different colors can be season image, intermediate on dry season
interpreted as follows (Amitrano et al. 2015): image, low coherence)
Red or white areas: man-made objects, village
Black areas: permanent basin water (dark on both (intermediate or bright on both acquisitions,
images, low coherence) high coherence)
Blue areas: wet season basin water (dark on wet Blue-green intermediate color, high intensity
season image, intermediate on dry season (cyan): trees (bright on both acquisitions, low
image, low coherence) coherence)
SAR Tomography for 3D Reconstruction and Monitoring 2435
Blue-green intermediate color, intermediate Curlander JC, McDonough RN (1991) Synthetic aperture
intensity (Prussian blue or dark green): bare radar: systems and signal processing. Wiley, New
York
soils (intermediate on both acquisitions, inter- Di Martino G, Riccio D, Zinno I (2012) SAR imaging of
mediate coherence) fractal surfaces. IEEE Trans Geosci Remote Sens
50:630–644
Note that the one described above is just an Elachi C (1988) Spaceborne radar remote sensing: appli-
cations and techniques. IEEE Press, New York
example: different RGB compositions can be Franceschetti G, Riccio D (2007) Scattering, natural sur-
employed, according to the considered applica- faces and fractals. Academic, Burlington
tion and to the scene characteristics that the user Franceschetti G, Iodice A, Riccio D (2002) A canonical
is interested to highlight. problem in electromagnetic backscattering from build-
ings. IEEE Trans Geosci Remote Sens 40:1787–1801
Fung AK, Li Z, Chen KS (1992) Backscattering from a
randomly rough dielectric surface. IEEE Trans Geosci
Summary Remote Sens 30:356–369
Guida R, Iodice A, Riccio D, Stilla U (2008) Model-based
interpretation of high-resolution SAR images of build-
SAR sensors have significant advantages with ings. IEEE J Sel Topics Appl Earth Observ 1:107–119
respect to optical ones: day-and-night, all-weather Guida R, Iodice A, Riccio D (2010) Height retrieval of
capabilities, and the possibility of measuring ter- isolated buildings from single high resolution SAR
rain topography and monitoring small terrain images. IEEE Trans Geosci Remote Sens
48:2967–2979
movements, due to its coherent nature. In addition, Iodice A, Natale A, Riccio D (2011) Retrieval of soil
SAR images of virtually any area of the Earth surface parameters via a polarimetric two-scale
surface are today routinely available, due to the model. IEEE Trans Geosci Remote Sens
different SAR satellite missions currently in orbit. 49:2531–2547
Lee JS, Pottier E (2009) Polarimetric radar imaging: from
Therefore, their use has a huge potential impact on basics to applications. CRC Press, Boca Raton
a number of applications, among which the fast Tsang L, Kong J, Ding K (2000) Scattering of electromag-
post-event earthquake damage assessment. netic waves – theory and applications. Wiley, New
However, visual interpretation of SAR images York
Ulaby FT, Moore RK, Fung AK (1986) Microwave
by a human operator requires that the latter is remote sensing, active and passive. Artech House,
properly trained to get used to the peculiar char- Norwood
acteristics of SAR images: geometric distortions,
involved dependence of image intensity on terrain
electromagnetic and roughness properties, and
speckle noise. Visual interpretation can be made
easier by properly combining different images to SAR Tomography for 3D
form a color image (RGB composition). Reconstruction and Monitoring
Finally, automatic quantitative interpretation S
of SAR images requires the availability or the D. Reale and G. Fornaro
development of electromagnetic scattering Institute for Electromagnetic Sensing of the
models and of corresponding retrieval algo- Environment, National Research Council,
rithms. This field is currently the subject of an Naples, Italy
intensive research activity.
Synonyms
References Cosmo-Skymed; Differential interferometry;

DInSAR; Displacement measurement; Infra-
Amitrano D, Di Martino G, Iodice A, Riccio D, Ruello G
structures monitoring; Multidimensional SAR
(2015) A new framework for SAR data RGB
multitemporal representation: rationale and products. imaging; SAR tomography; Synthetic-aperture
IEEE Trans Geosci Remote Sens 53:117–133 radar, SAR; TerraSAR-X
2436 SAR Tomography for 3D Reconstruction and Monitoring
Introduction imaging process returns only a 2D projection of

the 3D reality. The Shuttle Radar Topography
Synthetic-aperture radar (SAR) is one of the most Mission (SRTM) in the last decade has
important Earth remote sensing sensors whose represented the first case of an extensive use of
applications have grown dramatically in the InSAR for the generation of a worldwide (except
recent years. It provides images at microwaves for the poles) digital elevation model (DEM), that
with resolution comparable to that of optical sys- is, a digital topography map of the Earth surface
tems, but with the crucial advantage of all-time, (Van Zyl 2001). SRTM DEM (90 m spacing
day/night, and all-weather, imaging capability. DTED-1 standard) has been extensively used in
Similarly to classical surveillance radars, SAR many applications. Nowadays, the TerraSAR-X/
measures the distance (range) from sensor to tar- TanDEM-X mission (launched in 2007 and 2011)
get: resolutions of the order of meters are is providing a “refreshing” with higher resolution
achieved through the pulse compression of large (12 m spacing DTED 3 standard) of the Earth
bandwidth (frequency-modulated) signals. Very DEM on a global scale by exploiting the simul-
high resolution in the along-track direction is taneous acquisitions of two twin SAR sensors
achieved as well through the coherent combina- flying in a close formation (Krieger et al. 2007).
tion of target echoes received over the illumina- On the other hand, differential interferometry
tion interval, thus implementing a virtual (DInSAR) takes advantage of the very high pre-
(synthetic) array of antennas (i.e., a very large cision of radar systems in measuring phase to
antenna) by exploiting the movement of a very estimate differential displacements of the area
small antenna mounted on board airplanes or imaged at different time instants with an accuracy
satellites. The latter feature turns SAR systems of the order of a fraction of the used wavelength.
to be imaging radars. DInSAR is today routinely used to estimate dis-
SAR data are nowadays used in many areas of placements induced by large earthquakes as well
environmental risks monitoring situations such as as to monitor volcanic activities producing
flooding, glaciers, land cover, and forest monitor- ground movements, subsidence caused by water
ing (Curlander and McDonough 1991; Moreira and/or oil extraction, mining activities, and also
et al. 2013). Among all, one of the primary appli- slow moving landslides (Massonnet et al. 1993;
cations of SAR is the 3D reconstruction and Carnec et al. 1995; Fornaro and Franceschetti
monitoring of the Earth surface displacements 1999; Crosetto et al. 2005).
through the use of interferometric techniques. The availability of long-term data archives of
SAR interferometry (InSAR) exploits the former C-Band ESA ERS1/2 and ENVISAT SAR
coherent properties of the sensor, i.e., the capa- satellites has pushed the development of
bility to accurately control not only the envelope multipass interferometric processing techniques
but also the phase of the transmitted radiation. which coherently process large dataset of tens of
SAR images are in fact complex data character- images. In this way intrinsic limitations of clas-
ized by an amplitude measurement (envelope), sical, single pair, DInSAR, as the presence of
related to the backscattering properties of the atmospheric phase contribution and unwrapping
scene, as well as a phase signal related to both procedures needs, affecting the unambiguous
phase of the backscattering coefficient and to the estimation of the useful deformation signal, are
distance of the target from the sensor to an accu- overcome. These techniques are mainly catego-
racy of the order of the wavelength (centimeters rized based on the assumption on the scattering
at microwaves). Similarly to the human vision on the ground. From one hand, the approaches
system, acquiring images with a slight angular referred to as Persistent Scatterer Interferometry
diversity allows SAR to be sensitive to the 3D (PSI) prioritize the spatial resolution and uses all
scene properties, i.e., to estimate the topography available baselines for accurate monitoring of
of the observed scene. Topography is not acces- “strong” (i.e., persistent) scatterers, typically
sible in a single SAR acquisition because the located on anthropic structures, exhibiting
a temporal response stability over the whole general vertical surfaces, it happens that scat-
observation period (Ferretti et al. 2000, 2001). terers located at different heights may be sensed
On the other hand, multipass DInSAR techniques by the radar at the same distance and therefore
as the Small BAseline Subset (SBAS) limits the their returns imaged in the same pixel. A pictorial
processing to interferograms characterized by explanation of the layover induced on a building,
short temporal separation as well as reduced as well as its effect on a very high (1 m) resolution
angular diversity and makes use of a spatial TerraSAR-X amplitude image, is provided in
multilooking to enhance the signal quality at the Fig. 1: backscattering returns from targets located
expense of a spatial resolution loss. SBAS tech- on the left part of the roof are imaged at first, and
niques, known also as DInSAR stacking, are then the contributions from the facade and finally
devoted to the monitoring of wide areas including the base of the building are imaged in far range.
rural zones (Berardino et al. 2002; Ferretti Consequently, the building is tilted toward the
et al. 2011). sensor in the resulting SAR image: notice that
PSI and SBAS approaches have been used to because of the very high resolution, returns
investigate several aspects of risks for monitoring from the facades are spread over a large number
of coseismic and post-seismic deformation of pixels in which contribution from ground is
corresponding to several major earthquakes, vol- also expected. The vertical synthetic aperture
canoes deformation, landslides, etc., as well as exploited by SAR tomography allows improving
for the monitoring of buildings and infrastruc- resolution and tightening responses of the differ-
tures (Cascini et al. 2007; Arangio et al. 2013). ent interfering scatterers and then giving a chance
A recent advance for the technological view- to detect and localize separately each scattering
point in the application to building reconstruction mechanism (Fornaro and Serafino 2006). This
and monitoring is provided by SAR tomography capability has a major importance in the
which extends the SAR imaging concept to the processing of urban areas, where the presence of
third dimension of height: SAR tomography is buildings causes very frequent occurrence of lay-
also known as multidimensional SAR imaging over between building facades, surrounding
technique due to the capability of full 3D imaging structures and ground. It is worth to note that
plus time monitoring. As for PSI, SAR tomogra- this feature is own of the tomographic approach:
phy exploits full resolution data with angular none of interferometric approaches, including the
diversity but for the use of the whole complex PSI, can counteract this interference because of
measured data then introducing the virtual the leading assumption of only one scattering
antenna array along the height: the larger height mechanism per pixel and of the phase-only signal
antenna extent allows reducing again, as for the model which does not cope with an imaging
azimuth, the antenna beamwidth and viewpoint.
reconstructing with finer resolution the backscat- As a direct extension of DInSAR, differential S
tering along height (3D imaging) (Reigber and tomography (4D imaging) has been also pro-
Moreira 2000; Gini et al. 2002; Fornaro posed: it extends the 3D imaging to measure
et al. 2005). This latter allows, from one side, also the deformation parameters of scatterer in
estimating height parameters of scatterers with the focused 3D space (Lombardini 2005; Fornaro
better accuracy but, above all, detecting the preset al. 2009a; Zhu and Bamler 2010a). Along the
ence of possible multiple scattering mechanisms same lines of PSI, also time series explaining the
which may interfere within the same radar spatial temporal evolution of deformation can be
resolution cell. This interference, known as the extracted, even separately for each interfering
layover effect, is a direct consequence of the scatterers possibly exhibiting different deforma-
imaging principle of radar systems which dis- tion behaviors (Fornaro et al. 2009).
criminates scatterers in distance: in the typical SAR technology is also evolved to specifically
SAR side-looking geometry, in presence of accomplish requirements and improve monitor-
steep topography, as for buildings, walls, and in ing performances of multipass techniques.
SAR Tomography for 3D Reconstruction and Moni- TerraSAR-X image: closest targets located on the roof are
toring, Fig. 1 Pictorial illustration of the layover distor- imaged in near range respect to those on the base of the
tion on SAR images induced by the side-looking imaging building. Optical image courtesy of Google
principle of radar and evidence on a 1 m spatial resolution
Recent X-Band sensors as the TerraSAR-X and application: it is composed by four small-size
the Cosmo-Skymed constellation are providing satellites that ensure an average revisiting time
images with resolution never achieved in the of 4 days at the maximum acquisition rate, which
past, in the meter/submeter regime with reduced dramatically impacts in unexpected emergency
revisiting times. As far as the latter aspect is situations when immediate imaging of damaged
concerned, the Cosmo-Skymed mission is world- areas is required. High resolution allows captur-
wide the largest constellation for civilian ing much more information from the scene,
SAR Tomography for 3D

Reconstruction and
Monitoring, Fig. 2 3D
visualization of scatterers
detected by the SAR
tomography for the
building imaged in the
amplitude image of Fig. 1
(Reale et al. 2011a). Colors
are set according to the
estimated height
notably increasing the density of monitored scat- diversity in SAR imagery and their relationship
terers (Gernhardt et al. 2010). At the same time, with physical parameters of interest as topogra-
the layover becomes a major issue in the imaging phy and surface movements. SAR tomography is
of urban areas: the finer is the resolution, the introduced as an extension which turns the inter-
larger will be the number of pixels affected by ferometric processing into an imaging problem,
the layover induced by the buildings facades. allowing both improving performances in estima-
Multidimensional imaging have demonstrated to tion of parameters as well as allowing extracting
be an effective tool in resolving the distributed further information from the data with respect
layover on very high-resolution data, allowing to interferometry as the separation and detection
fully separating contribution from ground and of interfering scatterers in layover areas. Finally,
facade for the accurate 3D reconstruction of examples of the application of both multipass
buildings. An example of the capability of DInSAR (small scale), and tomographic (full
multidimensional imaging for 3D building recon- resolution) processing carried out on recent
struction and layover solution is provided in Cosmo-Skymed data are provided to point out
Fig. 2 which represents the 3D point cloud of the potentiality of the joint use of these technique S
the scatterers detected over the building imaged for a complete monitoring of risk situations at
in Fig. 1 by the SAR tomography on a dataset of different scales, from regional up to the level of
25 TerraSAR-X very high-resolution spotlight the single infrastructure, spatial scales.
images (Reale et al. 2011a). The precise recon-
struction proves the effectiveness of this
advanced processing which takes benefit of the The SAR Interferometry Background
detailed imagery provided by 1 m resolution
acquisition capability. SAR imaging allows discrimination of targets
This chapter is intended to introduce the prin- along the azimuth and range directions. Assum-
cipal concepts of the multidimensional imaging ing SAR sensors to fly locally rectilinear trajec-
approach. A brief introduction on classical inter- tories (airborne) or orbits (spaceborne), azimuth
ferometric approach is firstly provided to intro- x (directed along the sensor velocity vector) and
duce the concepts of angular and temporal range r (distance orthogonal from the flight track)
represent two coordinates of the natural radar 4p 4p

D’ ¼ dr ¼ ðdrz þ dr d Þ þ D’a þ D’o þ D’n
cylindrical reference system with the axis coinci- l l
dent with the flight track. By using large band- (1)
widths, reaching the order of hundreds of
megahertz, modern sensors distinguish targets in The first term drz = (b/r)s, with s being the slant
range with a resolution degree that reaches the height (orthogonal to the azimuth/range imaging
meter/submeter scale. In addition, a high azimuth plane corresponding to the master image) (s = z/
resolution capability of the final 2D images is sin(#), # is the look angle), accounts for the
achieved by synthesizing antennas in the order distance variation induced by the presence of
of kilometers through the exploitation of the topography on the ground which plays a role in
intrinsic motion of the platform along its orbit the presence of an imaging parallax measured by
(Curlander and McDonough 1991). spatial baseline b. Conversely, drd is the distance
The height information, which is not accessi- variation associated with possible deformation
ble in a single SAR image, can be estimated signal (LOS component) measured among the
through the interferometric concept. As in any two acquisition epochs. Subsequent terms play
electromagnetic coherent system, the phase a disturbance role. The term D’a is due to the
information is related to the traveled path and propagation delay variation between the two
therefore to the distance of the scene from the acquisitions induced by changes in atmosphere.
imaging radar. Classical SAR interferometry Slowing produces a time delay which is mapped
exploits the phase difference of, at least, two into a range variation. The atmospheric propaga-
images acquired with an angular diversity tion delay (APD) exhibits a spatial correlation
induced by a slight orbital offset (spatial base- over hundreds of meters: it is typically on the
line), to retrieve the 3D localization of ground same level of deformations which are also often
scatterers. Conversely, temporal separation spatially correlated; D’o and D’n are associated
(temporal baseline) is exploited by the differen- with orbital inaccuracies and noise. In applica-
tial SAR interferometry to measure possible dis- tions devoted to the estimation of the scene
placements along the radar’s line-of-sight (LOS) topography, simultaneous acquisitions are pref-
occurring at each acquisition epoch. The winning erable because both deformation and atmospheric
aspect of interferometry is the very high precision contributions are absent: this was the case of the
of radar in the estimation of the phase values, Shuttle Radar Topography Mission (SRTM)
whose accuracy is of the order of a fraction of which employed a dual-antenna system to recon-
the wavelength: this allows estimating move- struct the digital elevation model (DEM) of the
ments with sub-centimetric accuracy, using Earth and recently of the TerraSAR-X/TanDEM-
C- and recently X-Band radars from the space. X mission which employs a pair of twin satellites
Classical DInSAR has been extensively applied flying in close formation to produce a better
to measure large deformation caused mainly from resolved, 12 m spatial resolution, DEM. Differ-
earthquakes or volcanic activities. However, the ently, when deformation is of interest, as in dif-
presence of additional phase disturbing contribu- ferential interferometry, repeated passes over the
tions does not allow pushing accuracy to exactly same orbit should be required to avoid the
a millimeter accuracy through the exploitation impact of topographic phase contribution.
of just few images. As stated above, the interfer- Indeed, this requirement is problematic to
ometric phase difference D’ of a radar system enforce, and then DInSAR interferograms are
working with a wavelength l is related to the produced by subtracting from the original
difference dr in the traveled path of the signal interferogram an estimation of drz evaluated
forming the two images. This phase difference is from an external DEM (typically the SRTM
composed of multiple contributions associated DEM): such operation, referred to as zero base-
with different source (Fornaro and Franceschetti line steering (ZBS), aims at eliminating or
1999): at least mitigating the fringes corresponding to
the topography. The fringe pattern retrieved by It is worth pointing out that what is really
the external DEM is usually referred to as “syn- measured in interferometry is only a restricted
thetic interferogram.” Estimation of deformation version, corresponding to the modulus 2p of
is however affected by the presence of atmo- D’: one of the most critical steps of the interfer-
spheric contribution, and then classical two-pass ometric processing is therefore represented by the
DInSAR configuration is usually applied to esti- unwrapping procedure which is required to
mate predominant deformation caused, f.i., by retrieve the exact, unrestricted, absolute differen-
large earthquakes. tial phase from whom quantity of interest can be
The assumption leading to the differential estimated. Decorrelation effects, as well as large
phase in Eq. 1 is that the scene backscattering phase discontinuities, f.i., associated to steep
involved in the complex conjugate interferomet- height variations, affect the reliability of phase
ric products are the same in both the images. unwrapping algorithms (Ghiglia and Pritt 1998).
Changes lead to the presence of noise contribu-
tion in the interferometric phase D’ which is
referred to as decorrelation (Bamler and Hartl Multipass SAR Interferometry
1998; Fornaro and Franceschetti 1999). The
coherence measures the degree of decorrelation: Satellites regularly repeat orbits over the time.
it measures the modulus of the cross-correlation As a consequence, stacks of multipass acquisi-
index between the two images, which is tions, characterized by angular and temporal
a measure of the linear predictability of the two diversity, are available in remote sensing
random variables corresponding to the master archives for most of the Earth surface. In order
and slave SAR image values in the given pixel. to achieve a higher accuracy in the estimation of
The coherence is a product of several the deformation, modern advanced DInSAR
decorrelation effects, the most important being processing algorithms jointly process all the
associated with the change of the imaging geom- images in stacks of multitemporal acquisitions:
etry due to the spatial orbit offset, the temporal this processing allows in fact to provide
separation between the acquisitions and the ther- a discrimination between the atmospheric and
mal noise. Spatial decorrelation is caused by the deformation as well as to cancel possible residual
change of the imaging angle which determines topography components.
a change of the backscattering coefficient, Among all, interferometric techniques can be
because of the presence in each resolution cell distinguished in two main categories which are
of a large number of scattering sources. The tem- characterized by complementary assumptions
poral decorrelation is due to the change of the about the ground scattering: the multipass
scene backscattering properties over the time. DInSAR stacking techniques and the Persistent
This decorrelation source is of main importance Scatterers Interferometry (PSI). The class of the S
in repeat pass interferometry, especially with sys- DInSAR stacking methods is a direct extension of
tems operating at higher frequencies (f.i., the classical two-pass DInSAR technique.
X-Band). It is critical over the sea and in vege- It assumes the scattering to be spatially distrib-
tated areas where the growth and in general the uted over the resolution cell and is based on the
change of vegetation lead to strong variation of exploitation of both only small (temporal and
the backscattering coefficient. The last term, the spatial) baseline interferograms (hard baseline
thermal decorrelation, is due to the presence of thresholding) and of a spatial multilook in order
thermal noise in the receiving apparatus. It is par- to limit the effects of decorrelation and reduce the
ticularly evident in areas characterized by very low phase noise in the interferograms, as for the Small
scene backscattering. Other decorrelation sources BAseline Subset technique (Berardino
are associated with variations of the imaging et al. 2002). It is tailored to the monitoring of
aspect angle (Doppler centroid decorrelation) and wide areas and scattering mechanisms that
to processing artifacts. exhibit decorrelation including rural areas with
SAR Tomography for 3D Reconstruction and Moni- area close to L’Aquila obtained by processing a dataset of
toring, Fig. 3 Post-seismic deformation velocity map 33 Cosmo-Skymed images acquired between April 4 and
and co/post-seismic time series of a point in the Paganica October 13, 2009 (Reale et al 2011b)
slow temporal correlation losses. This technique (small scale). Additionally, through the analysis
relies on the inversion of the linear system relat- of multiple interferograms, the atmospheric
ing interferometric (differential) phase values phase contribution can be estimated and compen-
after phase unwrapping to the phase values on sated from the data. This latter, coupled with the
each acquisition. The separation of displacement removal of the background deformation signal
and atmospheric contribution is carried out by occurring on a small scale, are used to phase
exploiting their statistical characterization in calibrate the full resolution data for processing
terms of spatial and temporal correlation. at large scale, i.e., at the level of single building
DInSAR stacking methods analyze interfero- and infrastructure.
grams typically generated by pairing acquisitions A typical product of the processing through
characterized by small spatial baselines and tem- DInSAR stacking techniques is reported in Fig. 3:
poral separation, thus limiting the decorrelation. it represents the deformation mean velocity map,
They play a favorable role in the design of a superimposed to a Google Earth image,
two-scale processing: besides effectively corresponding to the slope of the deformation
counteracting decorrelation and phase noise, the time series for each pixel selected by looking at
spatial multilooking allows also to under-sample a quality index measuring the temporal consis-
the interferograms in the image spatial (azimuth tency of measurements after phase unwrapping.
and range) coordinates, thus reducing the amount The processing has been carried out through the
of data and consequently the computational effort application of the Enhanced Spatial Differences
for the analysis of large areas. Therefore, this (ESD) technique which extends the classical
class of algorithms is particularly suitable for SBAS approach by exploiting a model for the
the analysis of large areas at lower resolution phase differences between adjacent pixels to
counteract the effect of atmospheric contribution, and differently from SAR tomography, PSI uses
thus supporting the phase unwrapping step and only the phase information and assumes the pres-
increasing the coverage and accuracy of the ence of a single persistent scatterer per resolution
retrieved deformation measurements (Fornaro cell retaining correlation over the time (persistent
et al. 2009b). The result is relevant to the dataset scatterers). The use of the model however pre-
of Cosmo-Skymed acquisitions over the area of sumes the compensation of phase contributions
L’Aquila, in the central part of Italy, struck by the such as the atmospheric phase delay. Such
2009 earthquake. During the emergency acquisi- a compensation can be carried out either by ana-
tion plan, the Cosmo-Skymed constellation, at lyzing the phase on persistent scatterers
that time composed by three of the four satellites (PS) candidates, which are strong scatterers
nowadays operatively, intensified the acquisition where the phase is less affected by noise or by
to its highest possible rate (almost 1 acquisition using the coarse resolution product of previous
every 5 days on average) to acquire in only stacking techniques. In the latter case, a good
6 months datasets on ascending and descending practice is also to subtract the low-resolution,
orbits on different beams including a sufficient spatially correlated, deformation so to obtain
number of images for multipass differential also a zero deformation steering to carry out
interferometric processing. The result presented in high-resolution analysis on residual phase sig-
Fig. 3 is relevant to the processing of 33 H-image nals: the following model is assumed for the
ascending acquisitions (beam 09 corresponding to vector w collecting the compensated phase values
an incidence angle of about 40 ) taken between in the N available acquisitions at full resolution
April 4 and October 13 2009. Colors in the map
move from blue, associated to movements toward 4p b 4p
w¼ s þ dðs, tÞ þ wn (2)
the radar los (uplift), to red, associated to move- l r l
ments away from the sensor (subsidence). The where b is the N-dimensional vector of spatial
velocity map (evaluated only on the post-seismic baselines and d(s, t) is the vector collecting the
dates from April 12 to October 13) overlaid to displacements measured at the acquisition
a Google Earth image shows an area of subsidence instants collected in the vector t. Following the
with a rate exceeding the limit of 6 cm/year affect- compensation for low-resolution components,
ing the city of Paganica, in the eastern part of deformations are divided into the linear,
L’Aquila. In the same image, the plot of the time described by the average (mean) velocity v of
series of a point in the subsidence area shows the the pixel corresponding to the slope of the time
jump associated to the main shock (April 6) and series wrt to the epochs, and nonlinear dnl
the exponential decay of the subsidence with the addenda, i.e., d(s, t) = vt + dnl.
aftershocks. Measurements were shown to be in PSI assumes nonlinear terms having a small
full agreement with traditional leveling (Reale amplitude and carries out, for each pixel, S
et al. 2011b) and GPS measurements (D’Agostino a measure of the correlation of the measured
et al. 2012). signal with the model in Eq. 2, through the max-
The second class of multipass techniques, the imization of the scalar product in the following,
Persistent Scatterer Interferometry, as for SAR which returns also an estimation of the (s, v)
tomography, works at the highest spatial resolu- parameters:
tion to determine the deformation of single dom-
inant scatterers typically associated with 1 H 4p b

man-made structures (dihedral and trihedral of C ¼ max ejw ej l ð r sþvtÞ (3)
ðs, vÞ N
walls edges, poles, gratings, etc.) (Ferretti
et al. 2000, 2001). In this case, to achieve also Only pixels for which C is above a fixed threshold
high accuracy in the estimation of the localization are labeled as persistent scatterers and therefore
of scatterers, no limitations on the spatial baseline the algorithm provides the temporal series as
are introduced. Similarly to DInSAR stacking a product.
In the recent literature, the SqueeSAR 2005; Fornaro et al. 2009a; Zhu and Bamler
approach has been proposed as an extension of 2010a):
the PSI technique to handle the impact of target
ðð
decorrelation of distributed scatterers (Ferretti 4pbn s 4pvtn
gn ¼ gðs, vÞe j l r e j l dsdv þ wn (4)
et al. 2011). SqueeSAR performs a phase filtering I s Iv
of the interferograms by exploiting the correla-
tion matrix estimated from the data. SqueeSAR which shows that, but for the noise term wn, a 2D
extends PSI to partially correlated (i.e., Fourier transform (FT) relationship stands
decorrelating) scatterers: the algorithm looks for between the data gn and the backscattering distri-
a persistent scatterer (PS) mechanism equivalent bution g in the elevation/velocity (s, v) domain
to the real distributed scatterer, i.e., it assumes the with respect to the Fourier conjugate variables
response of the equivalent scatterer characterized xn = 2bn/(lr), associated with the elevation s,
only by phase variations with constant amplitude. and n = 2tn/l, associated with the velocity v.
Similarly to DInSAR stacking, SqueeSAR is tai- In particular, as a consequence of the deformation
lored to the analysis of rural areas, however it term Fourier expansion, g(s, v) plays in v the role
does not perform a hard threshold on the base- of the spectrum of the motion-related signal for
lines but rather it uses in a weighted way all the elevation s. For linear deformation, the spectral
interferograms. velocity coincides with the deformation rate, i.e.,
d(s, tn) = vtn, and then v is usually referred to as
deformation mean velocity, whereas for more
SAR Tomography System Model complex motion, v identifies the (velocity) har-
monic involved in the motion.
SAR tomography is a step forward, respect to The tomographic problem consists of the esti-
PSI: both are designed to monitor, at the full mation, in each image pixel, of the scene back-
available spatial resolution, the deformation scattering distribution g(s, v) starting from the
affecting ground scatterers. The key difference N samples gn and involves, in the most general
is however the different assumptions of the nature case, the inversion of Eq. 4, that is, a spectral
of the scattering which reflects into the interpre- analysis of the data. This analysis moves the inter-
tation of the received signal. As PSI processes ferometric processing toward an imaging problem
only the phase information interferograms and approach which extends the classical azimuth
therefore assumes the presence of a single scat- compression concepts widely known in the SAR
tering mechanism, SAR tomography removes 2D image focusing also for the third (elevation)
this latter hypothesis and considers the complex dimension. Large antenna spans are coherently
value of each image pixel measured at the generic processed to achieve narrow responses and
nth acquisition as the superposition of multiple improve height resolution to the order of meters,
elementary backscattering contributions distrib- allowing separating backscattering from source
uted along the elevation s (Reigber and Moreira which are located at different heights. The scatter-
2000; Gini et al. 2002; Fornaro et al. 2005). Dif- ing sources in the pixel can be spatially concen-
ferential tomography exploits the multitemporal trated (compact scatterers), as for the layover in
characteristics to allow tomography to also mon- urban area where scatterers are typically located
itoring deformation of scatterers. Particularly, on the roofs and facades of building and interfere
a Fourier expansion of the deformation term with those at lower heights, f.i., on the ground, or
d(s, t) is introduced and, assuming the atmo- can be distributed along the elevation as for appli-
spheric phase delay caused by the propagation cation in forest scenario where separation of
in atmosphere being compensated through ground level from canopy is of interest (Reigber
a preliminary multipass DInSAR processing and Moreira 2000; Cloude 2006; Tebaldini 2010).
(f.i., SBAS), the measured signal at the generic The 4D model in Eq. 4 represents the most
antenna is therefore modeled as (Lombardini general imaging model which can be
particularized under specific conditions: in case a discretized version of the model in Eq. 4: letting
of simultaneous acquisitions implying the g = [g(s0, v0), . . ., g(sM 1, vM 1)]T be the vector
absence of temporal diversity, as well as in case that collects the M = Ms Mv samples of g(s, v)
of absence of deformation, the signal model at the discrete points, hereafter called bins, (sm,
reduces to a 1D Fourier transform, and then the vm), with m = 0, . . ., M 1, belonging to the Ms
backscattering profile g(s) is of interest. This Mv elevation/velocity discretization grid
introduces to the 3D imaging framework. On (T defines the transposition operator), and
the other hand, whereas uniform motion mostly g = [g0, . . ., gN 1]T and w = [w0, . . ., wN 1]T
applies for classical risk situations associated to the vectors collecting, for each pixel, the mea-
slow, long-term deformation phenomena, more sured complex data and the noise contribution at
complex behavior can be also taken into account: each acquisition, respectively, the FT operator in
the analysis of latest X-Band SAR data points out Eq. 4 can be rewritten in the discrete case as
a higher sensitivity to small changes as those
caused by the thermal dilation of materials
g ¼ Ag þ w (5)
(Reale et al. 2011a; Zhu and Bamler 2011).
Moreover, since revisiting times are reduced
with respect to the former generation of SAR where A = [a0, . . ., aM 1]T is the N M system
sensors, the time requested to collect a sufficient matrix collecting the steering vectors associated
number of images for reliable application of with each discretization bin synthetically defined
tomographic processing reduces to the order of as am = a(sm, vm), whose generic element is
pffiffiffiffi
1 year, whereas typical C-Band ENVISAT and ½am n ¼ exp½j2pðxn sm þ n vm Þ= N .
ERS acquisitions spanned a temporal interval of Several techniques can be used to implement
observation spanning several years. The reduced the imaging, that is, the inversion of Eq. 5 that
observation times may imply possible correlation leads to the estimation of the backscattering dis-
with linear deformation behavior and impair the tribution in the elevation/velocity plane (s, v).
estimation of the deformation mean velocity Each is characterized by a different trade-off
(Reale et al. 2013). Even so, the sensitivity to between simplicity, computational efficiency,
thermal dilation can be exploited by SAR tomog- sidelobes reductions, and super-resolution
raphy as well. By extending the deformation capability.
model to account also for a second contribution The beamforming (BF) technique represents
linearly related to the average temperature Tn at the classical method to perform the inversion of
the acquisition instants, i.e., d(s, tn) = vtn + kTn, Eq. 5: it makes use of the conjugate operator AH,
the order of the tomographic imaging can be with H being the Hermitian (conjugate transpose)
extended (5D imaging) to estimate also operator, to profile the backscattering along the
a coefficient k which measures the expansion elevation bins (Fornaro et al. 2009a): S
along the line-of-sight for each degree of temper-
ature change (Zhu and Bamler 2011; Reale g^ ¼ AH g: (6)
et al. 2013). In application to the monitoring of
strategic infrastructures, this strategy allows esti- Once the backscattering in the (s, v) plane has
mating the stress induced by the temperature been estimated, scatterers are selected by looking
changes over the different segments of the struc- ^ In this context, a tool for the
for strong peaks in g.
tures (Fornaro et al. 2013). effective selection of reliable scatterers is
required: since in real data the useful information
is corrupted by noise, a detection stage is required
SAR Tomography Imaging Algorithms to control the false alarm rate, defined as the
probability to declare the presence of a scatterer,
Multidimensional SAR imaging algorithms pro- whereas the scatterer is not really present on the
posed in the literature typically work on ground. With reference to the case of a single
scatterer pixel, a test statistic based on the gener- ML estimation of the (s, v) parameters. The
alized likelihood ratio test has been proposed in GLRT in Eq. 7 allows also demonstrating the
the literature: it provides at the same time the effectiveness of an imaging approach, as in
maximum likelihood estimation of the (s, v) pair SAR tomography, with respect to the classical
and declares the presence (hypothesis ℋ1) or interferometric processing. The PSI technique,
absence (hypothesis ℋ0) of the scatterer for in fact, exploits a similar test but for the use of
a given probability of false alarm by exploiting only the phase information of each element of the
a test statistic which is strictly related to the BF vector g (Ferretti et al. 2000, 2001): the exploita-
reconstruction and is expressed as (De Maio tion of the whole complex, amplitude plus phase,
et al. 2009): data as in Eq. 7 provides a significant increase,
with respect to PSI, of the detection probability
ℋ1 for a fixed false alarm rate.
jgH aðs, vÞj
maxðs, vÞ ≷ T (7) Typical products achieved by the sequence of
kgkkaðs, vÞk
ℋ0 low-resolution DInSAR stacking techniques for
the derivation of coarse-scale deformation and
where T is the detection threshold, belonging to calibration of data for the subsequent full resolu-
the [0, 1] interval and set according to the desired tion processing is reported in Figs. 4 and 5. The
level of false alarm. It is worth noting that for processing involved a dataset of 29 Cosmo-
single scatterers, the test statistic represents the Skymed images acquired over the city of Rome,
highest peak of the normalized BF reconstruction Italy, from April 2011 to October 2012 in
and the argument of the maximization is just the H-image mode which provides images with 3 m
SAR Tomography for 3D Reconstruction and Moni- Colors are set on the estimated velocity and correspond to
toring, Fig. 4 Small-scale deformation mean velocity movements toward the sensor (uplift) in blue and away
map estimated by the ESD technique on a dataset of (subsidence) in red. Optical image courtesy of Google
29 Cosmo-Skymed images over the city of Rome, Italy.
SAR Tomography for 3D Reconstruction and Moni- the estimated deformation mean velocity. The area is
toring, Fig. 5 3D visualization of the scatterers, relevant to the Grotta Perfetta area in Rome, close to the
represented as dots, detected by the GLRT after 4D Tevere River. Optical image courtesy of Google
beamforming reconstruction. Colors are set according to
of spatial resolution. According to the SBAS hides some deformation signals occurring in the
approach, a total number of 80 interferograms central part of the city, which is of interest for the
has been generated by imposing maximum spa- application of full resolution tomographic analy-
tial and temporal baseline spans of 1,000 m and sis provided in a following section.
150 days, respectively. A multilook has also been Figure 5 provides a close view, in a Google
carried out through the use of a 16 16 pixels Earth framework, of the results of the application
moving average filter followed by a subsampling of 4D imaging and subsequent GLRT detection
of the same factor. In Fig. 4 it is shown the on the Cosmo-Skymed dataset over Rome intro-
deformation mean velocity map, superimposed duced before. It allows attesting the capabilities
to a Google Earth image, corresponding to the of imaging approaches in the single building
slope of the deformation time series for each monitoring. The area, in the Rome city center, is
pixel selected by looking at a quality index mea- affected by a severe subsidence induced by
suring the temporal consistency of measurements the consolidation of the alluvial sediments of S
after phase unwrapping. As for the L’Aquila the Tevere river. Each dot corresponds to
dataset, the low-resolution processing has been a detected scatterer after geocoding from the
carried out through the ESD technique. Colors in original radar geometry to the natural geographic
the map move again from blue (uplift) to red coordinates system: the correct estimation of
(subsidence). The results highlight the presence heights allows effectively positioning scatterers,
of a distributed deformation pattern in the west- as they overlap the 3D models provided by
ern part of the city which largely affects also the Google Earth. Furthermore, the colors corre-
Rome airport. The large number of detected scat- spond to deformation velocity rate estimated for
terers can be appreciated, although visualization each scatterer: several buildings in the area suffer
constraints of the Google Earth environment for large subsidence rates which cause extensive
impose a reduction of the total number of points, structural damages affecting their stability
in this case of a factor 5. The dynamic range also (Arangio et al. 2013).
The detection problem can be straightfor- presence of highly uneven baseline distributions
wardly extended to the case of multiple scatterers and at the same time achieve some super-
per resolution cell, which may occur in layover resolution capabilities, i.e., the possibility to
areas. In this framework, a detector which tests push the height resolution below the inherent
the presence of possibly a maximum of two scat- Rayleigh limit.
terers based on a sequential use of the GLRT The use of the singular value decomposition
discussed above has been also presented (SVD) of the operator A in Eq. 5 allows regular-
(Pauciullo et al. 2012). With reference to the izing the inversion by restricting the solution
separation of interfering scatterers within the space and benefiting of the inclusion of very
same image pixel, in application to a real acqui- limited a priori information on the expected
sition scenario, the acquisition geometry poses scene elevation extent. The regularization,
limitation on the imaging capabilities of BF: the obtained through the so-called Truncated SVD,
baseline distribution defines the unambiguous allows avoiding noise amplification and inver-
elevation range that is the maximum extension sion instabilities and hence generally provides
allowed in elevation direction to avoid aliasing a better sidelobe reduction and as well as slight
phenomena. Given Dbn be the spatial separation super-resolution than plain BF (Fornaro
between the successive antennas, the unambigu- et al. 2009a).
ous elevation interval will be Ds = lr/(2Db), Compressed sensing (CS) is a recent technique
where Db is the average value of Dbn in case of used in linear inversion problems for signal
nonuniform baseline distribution. Moreover, the recovery that takes benefit of the hypothesis that
final elevation resolution of BF cannot exceed the the signal to be reconstructed have (in some
Rayleigh resolution given by Ds = lr/(2B) with basis) a sparse representation, i.e., a small num-
B being the total baseline span. In the same way, ber of nonzero entries. Under certain assumptions
unambiguous velocity interval and velocity reso- of the measurement matrix, the signal can be
lution can be also defined as Dv = l/(2Dt) and reconstructed from a small number of measure-
Dv = l/(2T), where Dt and T are the average ments. SAR tomography in urban areas is
temporal separation and the total temporal span, a favorable application scenario for CS due to
respectively. The elevation resolution plays a key the fact that for typical operative frequencies,
role in the capability of distinguishing possible the scattering occurs only on some scattering
multiple scattering contributions within the same centers associated with ground, facades, and
pixel: it implies the minimum separation of scat- roofs of ground structures (Zhu and Bamler
terers as they can be distinguished as separate 2010b; Budillon et al. 2011).
scatterers. Latest satellite SAR missions as the
TerraSAR-X are characterized by a narrow radius
of the orbital tube, leading to small baseline Summary
excursions that result in final height resolution
of some tens of meters. In application to urban Available high-resolution synthetic-aperture
areas, such a resolution can be very restrictive for radar imaging sensors are capable of providing,
the solution of the interference due to the layover. in a systematic regular basis, images of single
Finally, in all the practical applications, baselines buildings and ground targets with very high spa-
are far from being uniformly distributed, thus BF tial details. Persistent Scatterers Interferometry
gives poor reconstruction performances in terms techniques have been already shown to largely
of sidelobes and leakage in the (s, v) point spread benefit of the use of very high-resolution data to
function. Alternative strategies can be adopted to monitor buildings. Despite the spatial resolution
improve both the resolution performances and increase, the steepness of the topography
also the quality of the reconstruction of g^ in the corresponding to vertical structured targets
generates critical distortion effect, the most crit- using compressive sampling. IEEE Trans Geosci
ical ones are shadow and layover. These are Remote Sens 49:488–499
Carnec C, King C, Massonnet D (1995) Measurement of
major impairing sources in the analysis of SAR land subsidence by means of differential SAR interfer-
images corresponding to urban areas. SAR ometry to sites of small extent. Geophys Res Lett
tomography represents a powerful technique 23:3579–3582
which allows the implementation of a radar scan- Cascini L, Ferlisi S, Peduto D, Fornaro G, Manunta
M (2007) Analysis of DInSAR Data via geotechnical
ner from the space with large 2D antennas to scan criteria. Rivista Italiana di Geotecnica 4:50–67
the details of building and overcoming problems Cloude SR (2006) Polarization coherence tomography.
of layover to generate dense point-cloud mea- Radio Sci 41:RS4017. doi:10.1029/2005RS003436
surements of buildings. Multidimensional SAR Crosetto M, Crippa B, Biescas E (2005) Early detection
and in-depth analysis of deformation phenomena by
imaging, based on the concept of SAR tomogra- radar interferometry. Eng Geol 79:81–91
phy, is a tool that represents the most advanced Curlander JC, McDonough R (1991) Synthetic aperture
method in the 3D reconstruction and monitoring radar – system and signal processing. Wiley, New
of buildings. PSI and SAR tomography with York
D’Agostino N, Cheloni D, Fornaro G, Giuliani R, Reale
very high-resolution sensors provide a unique D (2012) Space-time distribution of afterslip following
too for application of spaceborne microwave the 2009 L’Aquila earthquake. J Geophys Res 117:
radar imaging to urban areas, which is expected B02402. doi:10.1029/2011JB008523
in the near future to play a key role in vulnerabil- De Maio A, Fornaro G, Pauciullo A (2009) Detection of
single scatterers in multi-dimensional SAR imaging.
ity and damage assessment of buildings and IEEE Trans Geosci Remote Sens 47:2284–2297
infrastructures. Ferretti A, Prati C, Rocca F (2000) Nonlinear subsidence
rate estimation using permanent scatterers in differen-
tial SAR interferometry. IEEE Trans Geosci Remote
Sens 38:2202–2212
Cross-References Ferretti A, Prati C, Rocca F (2001) Permanent scatterers in
SAR interferometry. IEEE Trans Geosci Remote Sens
39:8–20
▶ Building Monitoring Using a Ground-Based Ferretti A, Fumagalli A, Novali F, Prati C, Rocca F, Rucci
Radar A (2011) A new algorithm for processing interfero-
▶ InSAR and A-InSAR: Theory metric data-stacks: squeeSAR. IEEE Trans Geosci
Remote Sens 49:3460–3470
▶ SAR Images, Interpretation of Fornaro G, Franceschetti G (1999) SAR interferometry,
▶ Urban Change Monitoring: Multi-temporal Chapter IV. In: Franceschetti G, Lanari R (eds) Syn-
SAR Images thetic aperture radar processing. CRC Press, Boca Raton
Fornaro G, Serafino F (2006) Imaging of single and double
scatterers in urban areas via SAR tomography. IEEE
Trans Geosci Remote Sens 44:3497–3505
References Fornaro G, Serafino F, Lombardini F (2005) 3D multipass
SAR focusing: experiments with long-term
Arangio S, Calò F, Di Mauro M, Bonano M, Marsella M, spaceborne data. IEEE Trans Geosci Remote Sens S
Manunta M (2013) An application of the SBAS- 43:702–712
DInSAR technique for the assessment of structural Fornaro G, Reale D, Serafino F (2009a) Four-dimensional
damage in the city of Rome. Struct Infrastruct Eng. SAR imaging for height estimation and monitoring of
doi: 10.1080/15732479.2013.833949 single and double scatterers. IEEE Trans Geosci
Bamler R, Hartl P (1998) Synthetic aperture radar inter- Remote Sens 47:224–237
ferometry. Inverse Probl 14:1–54 Fornaro G, Pauciullo A, Serafino F (2009b) Deformation
Berardino P, Fornaro G, Lanari R, Sansosti E (2002) monitoring over large areas with multipass differential
A new algorithm for surface deformation monitoring SAR interferometry: a new approach based on the use of
based on small baseline differential SAR interfero- spatial differences. Int J Remote Sens 30:1455–1478
grams. IEEE Trans Geosci Remote Sens Fornaro G, Reale D, Verde S (2013) Bridge thermal dila-
40:2375–2383 tion monitoring with millimeter sensitivity via
Budillon A, Evangelista A, Schirinzi G (2011) Three- multidimensional SAR imaging. IEEE Geosci Remote
dimensional SAR focusing from multipass signals Sens Lett 10:677–681
2450 School Seismic Safety and Risk Mitigation
Gernhardt S, Adam N, Eineder M, Bamler R (2010) Poten-

tial of very high resolution SAR for persistent scatterer School Seismic Safety and Risk
interferometry in urban areas. Ann GIS 16:103–111
Ghiglia DC, Pritt MD (1998) Two-dimensional phase Mitigation
unwrapping: theory, algorithms, and software. Wiley-
Interscience, New York Marla Petal1, Ben Wisner2,3, Ilan Kelman4,17,
Gini F, Lombardini F, Montanari M (2002) Layover solu- David Alexander2, Omar-Dario Cardona5,
tion in multibaseline SAR interferometry. IEEE Trans
Aerosp Electron Syst 38:1344–1356 Djillali Benouar6, Sanjaya Bhatia7, Jitendra
Krieger G, Moreira A, Fiedler H, Hajnsek I, Werner M, Kumar Bothara8,9, Amod Mani Dixit8, Rebekah
Younis M, Zink M (2007) TanDEM-X: a satellite for- Green10, Ram Chandra Kandel11, Tracy Monk12,
mation for high-resolution SAR interferometry. IEEE Bishnu Pandey13, Janise Rodgers14, Zeynep
Trans Geosci Remote Sens 45:3317–3341
Lombardini F (2005) Differential tomography: a new T€urkmen Sanduvaç15 and Rajib Shaw16
1
framework for SAR interferometry. IEEE Trans Risk RED (Risk Reduction Education for
Geosci Remote Sens 43:37–44 Disasters), Los Angeles, CA, USA
Massonnet D, Rossi M, Carmona C, Adragna F, Peltzer G, 2
University College London, London, England
Fiegl K, Rabaute T (1993) The displacement field of 3
the Landers earthquake mapped by radar interferome- Oberlin College, Oberlin, OH, USA
4
try. Nature 364:138–142 Institute for Risk and Disaster Reduction and
Moreira A, Prats-Iraola P, Younis M, Krieger G, Institute for Global Health, University College
Hajnsek I, Papathanassiou KP (2013) A tutorial on London, London, England
synthetic aperture radar. IEEE Geosci Remote Sens 5
Mag 1:6–43 Universidad Nacional de Colombia, Manizales,
Pauciullo A, Reale D, De Maio A, Fornaro G (2012) Colombia
6
Detection of double scatterers in SAR tomography. University of Bab Ezzouar, Algiers, Algeria
IEEE Trans Geosci Remote Sens 50:3567–3586 7
UNISDR Recovery Platform, Kobe, Japan
Reale D, Fornaro G, Pauciullo A, Zhu X, Bamler 8
R (2011a) Tomographic imaging and monitoring of NSET, Kathmandu, Nepal
9
buildings with very high resolution SAR data. IEEE Miyamoto Impact, Christchurch, New Zealand
10
Geosci Remote Sens Lett 8:661–665 University of Western Washington,
Reale D, Nitti DO, Peduto D, Nutricato R, Bovenga F, Bellingham, WA, USA
Fornaro G (2011b) Post-seismic deformation monitor- 11
ing with the COSMO/SKYMED constellation. IEEE Toronto, ON, Canada
12
Geosci Remote Sens Lett 8(4):696–700 Families for School Seismic Safety,
Reale D, Fornaro G, Pauciullo A (2013) Extension of 4-D Vancouver, Canada
SAR imaging to the monitoring of thermally dilating 13
University of British Columbia, Vancouver,
scatterers. IEEE Trans Geosci Remote Sens.
51:5296–5306 Canada
14
Reigber A, Moreira A (2000) First demonstration of air- GeoHazards International, Menlo Park, USA
15
borne SAR tomography using multibaseline L-band Risk RED, Istanbul, Turkey
data. IEEE Trans Geosci Remote Sens 38:2142–2152 16
University of Kyoto, Kyoto, Japan
Tebaldini S (2010) Single and multipolarimetric SAR 17
tomography of forested areas: a parametric approach. Norwegian Institute of International Affairs,
IEEE Trans Geosci Remote Sens 48(5):2375–2387 Oslo, Norway
Van Zyl J (2001) The Shuttle Radar Topography Mission
(SRTM): a breakthrough in remote sensing of topog- Synonyms
raphy. Acta Astronaut 48:559–565
Zhu X, Bamler R (2010a) Very high resolution spaceborne
SAR tomography in urban environment. IEEE Trans Comprehensive school safety; Critical infrastruc-
Geosci Remote Sens 48:4296–4308 ture; Safe school construction; Safe school facil-
Zhu X, Bamler R (2010b) Tomographic SAR inversion by ities; School construction
L1 norm regularization – the compressive sensing
approach. IEEE Trans Geosci Remote Sens
48:3839–3846 Introduction
Zhu X, Bamler R (2011) Let’s do the time warp:
multicomponent nonlinear motion estimation in dif- Access to education is a basic human right. It is
ferential SAR tomography. IEEE Geosci Remote Sens
Lett 8:735–739 enshrined in Convention on the Rights of the
School Seismic Safety and Risk Mitigation 2451
Child (1990), the World Declaration on Educa- of case studies on progress and struggles for
tion for All (in 1990), and the World Education school seismic safety which are updated in the
Forum (WEF 2000). It is one of the Millennium entry “▶ School Seismic Safety: Case Studies.”
Development Goals for the decade starting in They set out the magnitude of concern and main
2005 and continues to be part of the “post- arguments for advocacy in an unpublished arti-
2015” development agenda. Education is cle. At the time, the authors posited the gruesome
strongly associated with poverty reduction, and estimate of “4,800 school children killed by
there are strong global and national drives to earthquake-related school collapse or severe
implement it. The Global Partnership for Educa- damage over the next decade. . . It might be rea-
tion has 29 national partners, supporting the sonable and prudent to plan to avoid a loss of
implementation of universal, free, quality basic student life in earthquakes of somewhere
education in 57 partner developing countries. between 2,000 and 5,000 in a 10 year period.”
In GPE’s 2012–2015 Strategic Plan, the first of At the time it was written, this estimate seemed to
its four strategic goals is “All children have the authors to be somewhat alarmist.
access to a safe, adequately equipped space to The following year, shortly after the unprece-
receive an education with a skilled teacher.” dented destruction caused by the Indian Ocean
However, none of its monitoring indicators men- earthquake and tsunami, 168 countries agreed to
tion safety. Since 2004, the GPE has contributed the 2005–2015 Hyogo Framework for action.
to build, rehabilitate, and equip close to 53,000 Over the course of this 10 year period, this dire
classrooms (GPE 2014). However, up until at predication has been exceeded fourfold as the
least 2013, there was no systematic due diligence result of only two major earthquakes during
with respect to disaster-resilient construction. In school hours: the 8 October 2005 Kashmir earth-
the rush to fulfill the right to education, are chil- quake which killed more than 18,000 students, in
dren being put at risk? addition to staff, in schools, and the 12 May 2008
This entry assesses seismic threats to schools Sichuan earthquake which killed more than 5,300
and reviews incidents of children and teachers students, in addition to staff, in their schools
killed by structural failure of school buildings as (UNISDR 2008).
well as structural damage to schools and near In the powerful earthquake and massive tsu-
misses. It reviews progress, good practices, and nami on 11 March 2011 in northern Japan, schools
lessons learned based on these threats. The entry themselves were by and large structurally sound
goes on to overview school vulnerability, global and resisted earthquake damage, but tsunami-
experiences in seismic-resistant school construc- retaining walls were breached as the tsunami was
tion, and retrofit. A summary of progress in larger than expected and land subsidence had not
school seismic safety along with a recommended been factored in. Disaster drills and practice of
systematic all-hazards approach to comprehen- “tendenko” (automatic tsunami evacuation), by S
sive school safety set the stage to make the case many school children, saved many lives. Some
for continued advocacy for school seismic safety. schools provided vertical evacuation, and many
survived at evacuation and shelter centers. But
instances of confusion occurred and many school
The Threat pupils and teachers also died. Today, the students
now displaced by the resulting nuclear disaster
According to estimates made by the Center for recognize this neglected threat as the most cata-
International Earth Science Information Network strophic of all. The international community is
at Columbia University in 2006, there are more virtually silent on this threat.
than 100 million school-aged children exposed to In common with other infrastructure, school
significant seismic risk throughout the world buildings are subject to damage and collapse in
(New York Times 2008). In 2004, 10 of the earthquakes. Many of these have resulted in chil-
16 contributors to this article initiated recording dren killed while being educated (Table 1).
Pictorial evidence of historic earthquake dam- importance of ensuring the seismic safety of
age to schools is available in the NISEE, Earth- schools.
quake Engineering Online Archive.
There have also been many cases when an
earthquake destroyed school buildings when Making the Case for School Seismic
they were not in session, and thus deaths and Safety
injuries were narrowly avoided (Table 2).
However the severe impact on continuity of Many public buildings and different sorts of crit-
education and the potential magnitude of loss ical infrastructure are threatened by earthquakes.
of life in these events further highlight the The case can be made for giving priority to
School Seismic Safety and Risk Mitigation, Table 1 Children killed by structural failure of school buildings
Date/local
time (Source) Location/magnitudea Consequences/schools Consequences/children
12 Jan 2010 Port-au-Prince, Haiti MoE estimates 4,992 schools Deaths and injuries unknown.
16:53 M 7.0 affected (23 % of the nation’s Many children with disabling
(CNN 2014) schools) injuries. Some schools were
holding their third shifts. Est.
1.3 m children and youth affected
12 May 2008 Wenchuan, China 175 schools (7,000 classrooms) in >5,300 school children died in
14:28 M.7.9 Sichuan and Shaanxi provinces dozens of schools
(COGGS were destroyed In the Beichuan Middle school,
2008) 1,300 of 2,999 students and
teachers died
6 Mar 2007 Western Sumatra The wall of a primary school 4 primary school children died
11:00 M 6.4 collapsed. Fire followed. Up to
(COGGS 329 schools affected by several
2008) earthquakes (2005–2010)
8 Oct 2005 Kashmir, Pakistan, and More than 10,000 schools >18,000 school children died
St. 08:50 India collapsed >50,000 school children were
(UNISDR M 7.6 80 % of Mahesehra’s 2,749 66 % of seriously injured
2008) Batagram’s 678, and 37 % of
Abbottabad’s 1,829 public schools
were destroyed or seriously
damaged
1 May 2003 Bingöl, Turkey 4 school buildings collapsed. Only 84 students killed and
03:20 M 6.4 the dormitory was occupied 114 survived in the dormitory
(Rodgers
2012)
24 February Bachu, Xinjiang, 900 classrooms collapsed Students were outside in physical
2003 10:03 China education at the time of the
(COGGS M 6.4 earthquake. At least 20 students
2008) killed in one middle school
collapse
31 October San Giuliano di Puglia, San Giuliano infant school 26 children and 3 adults killed.
2002 11:40 Molise, Italy collapsed 35 children rescued alive from
(COGGS M 5.9 the building but some reports
2008) suggest that one child died later
26 January Gujarat, India 1,884 school buildings collapsed. 971 school children and
2001 Friday M 7.6 5,950 classrooms destroyed. 31 teachers were killed in school
08:16 36,584 unfit for instruction activities. 1,051 students and
Republic day 95 teachers seriously injured
holiday 32 children died at
(COGGS Swaminarayan School
2008)
(continued)
School Seismic Safety and Risk Mitigation, Table 1 (continued)

Date/local
time (Source) Location/magnitudea Consequences/schools Consequences/children
13 February El Salvador 85 schools damaged beyond repair. 50 % of fatalities were children
08:22 and M 6.6 In aftershock 22 preschoolers and
13 January their teacher were killed
2001
(COGGS
2008)
9 July 1997 Cariaco, Venezuela Two out of five school buildings 46 students killed
15:24 M 7.0 collapsed. Four reinforced concrete
(COGGS buildings had serious structural
2008) defects
10 May 1997 Ardekul, Iran Elementary school collapsed 110 young girls were killed
12:57 M 7.3
(COGGS
2008)
1992 Erzincan, Turkey 6-story medical school collapsed 62 students were killed.
(COGGS M 6.9
2008)
7 December Spitak, Armenia 380 children and youth institutions Likely thousands of school
1988 11:41 M 6.8 destroyed. 105 of 131 in Spitak and children killed. At least
(COGGS Leninakin destroyed 400 children died in the collapse
2008) of a Dzhrashen elementary
school
27 July 1976 Tangshan, China Most school buildings destroyed 2,000 students killed in the
03:42 M 7.8 dormitory of the College Mining
(COGGS Institute
2008)
13 April 1949 Olympia, Washington, 10 schools destroyed, 30 damage. 2 children in school were killed
11:58 USA Spring break
(COGGS M 7.1
2008)
31 October Helena, Montana, Newly built secondary school wing 2 students killed. Classes not in
1935 USA collapsed session, could have been much
(COGGS M 6.2 worse
2008)
10 March 1933 Long Beach, 70 schools destroyed. 120 with 2 children died in gymnasium
Long Beach California, USA major damage. Classes held in tents collapse. Spring break, classes
(COGGS M 6.4 for 2 years. First legislation for safe not in session, could have been
2008) school construction much worse
a
S
Magnitudes from USGS or Wikipedia
schools from three perspectives: Duty bearers consequences beyond the replacement cost of
have moral and legal obligations to fulfill chil- school buildings themselves. In most cases public
dren’s rights to both safety and survival and edu- discussion and debate on these issues tend to mix
cational continuity. In more affluent countries, these ethical and pragmatic arguments.
the cost benefits of investments in public safety,
the importance of safeguarding development
investments, and preventing educational disrup- Human Rights Argument
tion are undisputed. And, the uses of school
buildings as multipurpose community centers The human rights argument suggests that no soci-
and disaster shelters, even when children are not ety should tolerate a choice between the safety of
harmed, have cascading social and economic children’s lives and their education. The right to
School Seismic Safety and Risk Mitigation, Table 2 School structural damage from earthquakes
Date (Source) Location/magnitudea Structural and educational impacts
2 July 2013 Aceh, Indonesia 966 schools affected
14.37 M 6.2
(Pandey 2013)
4 April 2010 California, USA, and Significant structural damage to several schools in Mexico.
15:40 Baja, Mexico Significant nonstructural damage to several schools in the USA. Cost
(Rodgers 2012) M 7.2 of repair almost 20 % of annual budget for one school district
School was on spring break. Nonstructural damage would have
caused injuries and blocked egress. In California hazardous asbestos
from collapsed walkways and mercury from light fixtures closed
schools for extended periods
30 September Padang, West Sumatra, 2,164 severely damaged, 1,447 moderately damaged, 1,137 lightly
2009 17:16 Indonesia damaged
(Rodgers 2012) M 7.6 School was recently dismissed for the day. Temporary school
buildings of timber frame and corrugated steel
2 September West Java, Indonesia 716 schools affected
2009 14:55 M 7.0
(Pandey 2013)
21 September Mongar, Bhutan 91 schools affected: 6 destroyed, 17 required major repair,
2009 14:53 M.6.1 44 required partial repair, 24 required minor repair (cost 12.9 m USD)
(Rodgers 2012) plus damage to boarding schools, water, and sanitation
School was dismissed early for holiday. Temporary learning facilities
inadequate for weather
6 April 2009 L’Aquila, Italy 78 schools had extended closures and 12 partial closures
03:32 M 6.3
(Rodgers 2012)
12 September Bengkulu, Indonesia 240 schools affected (2005–2010)
2007 18:10 M 8.5
(Pandey 2013)
15 August 2007 Pisco, Peru 116 schools were severely damaged. 478 classrooms were needed to
23:40 M 8.0 restore school activities
(Rodgers 2012)
27 May 2006 Yogyakarta, Indonesia Yogyakarta: 2155 educational facilities damaged or destroyed;
05:53 M 6.3 Central Java: 752 damaged or destroyed. Damage and losses
(Rodgers 2012) estimated at 1.7 trillion Indonesian Rupiah
26 December Indonesia, Sri Lanka, School earthquake and tsunami damage combined:
2004 early am Maldives, Thailand Indonesia – 750 destroyed 2,135 damaged. Sri Lanka – 51 destroyed,
(UNISDR 2008) M 9.1–9.3 100 damaged. Maldives – 44 destroyed or damaged.
Thailand – 30 destroyed
26 December Bam, Iran 67 of 131 schools collapsed. The remaining 64 were heavily damaged
2003 5:26 M 6.6 and unusable
(COGGS 2008) 33,000 students were affected
22 September Puerto Plata, 50 public schools damaged, 140 classrooms impacted
2003 12:45 Dominican Republic 18,000 students were without classrooms
(UNISDR 2008) M 6.4
21 May 2003 Boumerdes, Algeria 130 schools damaged beyond repair. 753 schools extensively
19:48 M 6.8 damaged or destroyed
(COGGS 2008; The earthquake occurred out of normal school hours, so children were
OECD 2004) not at school. Cost of school rehabilitation $79 million+
24 February Xinjiang, China Dozens of schools collapsed
2003 The earthquake struck 27 minutes before thousands of children would
(COGGS 2008) have been in classrooms
25 April 2002 Tbilisi, Georgia Approximately $8 million US in school damage. No collapses;
22:41 M 4.5 1 school with very heavy damage; 35 with substantial damage;
(Rodgers 2012) 68 with moderate damage; 98 with negligible or slight damage
(continued)

21 September Chi-Chi, Taiwan 51 schools collapsed. 786 schools nationwide were damaged. 22 % of
1999 1:47 M 7.7 schools and 71 % of post-secondary institutes damaged
(COGGS 2008) The earthquake happened in the middle of the night, so no one was in
the building. Cost of repair and reconstruction $1.3 billion
June 23 2001 Arequipa, Peru 98 school buildings seriously damaged
15:33 School not in session on Saturday
(COGGS 2008)
17 August 1999 Kocaeli, Turkey 43 schools were damaged beyond repair. 381 minor to moderate
3:02 M 7.6 damage
(COGGS 2008) In Istanbul 60 km away 35 schools were unsafe and demolished
School was not in session but was suspended for 4 months. In Istanbul
131 schools were closed temporarily, for inspection
25 January 1999 Pereira and Armenia, 74 % of schools in Pereira and Armenia were damaged
13:19 Colombia School was not in session
(OECD 2004) M 6.2
20 August 1998 Udayapur, Eastern 1,200 schools heavily damaged or destroyed. 6,000 affected
(UNISDR 2008) Nepal
M 6.6
17 July 1998 Papua, New Guinea Schools destroyed
00:19 M 7.0
9 July 1998 5:19 Faial, Azores, Portugal Schools damaged
(COGSS 2008) M 6.2 School was not in session
20 May 1998 Afghanistan/Tajikistan Unknown
M 6.6
12 November Nazca, Peru 93 school buildings seriously damaged
1996 15:33
(COGSS 2008)
1996 Temouchent, Algeria 6 schools destroyed, 17 moderate damage, 36 light damage
(OECD 2004) M 5.6
17 January 1995 Hanshin-Awaji, Japan 54 buildings school damaged beyond repair. Extensive earthquake
5:46 M 6.9 and fire damage to 4,500 educational buildings. ¥94 billion. School
(COGSS 2008) was not in session
1994 Beni Chourgrane, 4 schools destroyed, 16 moderate damage, 30 light damage
(OECD 2004) Aleeria
M 5.6
17 January 1994 Northridge, California, 24 of 127 affected schools suffered significant structural damage.
04:31 USA Suspended lighting and ceiling systems were damaged in 1,500
(FEMA 2011) M 6.7 buildings
2, 204 schools were used as shelters. Had this occurred during the S
school day, significant injuries and lack of safe egress for thousands
would have resulted. The Los Angeles Unified School District,
amongst others, embarked on projects for nonstructural mitigation,
now the responsibility of school maintenance personnel
25 March 1993 Scott Mills, Oregon, Part of masonry school building collapsed
(COGSS 2008) USA Spring break, school was not in session
17 October 1989 Loma Prieta, California, 7 schools in three districts and one headquarters sustained severe
17:04 USA damage. 1,544 were schools surveyed. Total value of damage $81
(EERI 1990) M 6.9 million USD
10 October 1989 El Asnam, Algeria 70–85 schools suffered extensive damage or collapsed
(OECD 2004) M 7.3 The earthquake occurred out of normal school hours, so children were
not at school
20 August 1988 Bihar, India, and 950 school buildings were damaged in Bihar
4:39 Udaypur Nepal 6,000 schools destroyed in Nepal
(COGSS 2008) M 6.6 School was not in session
(continued)

8 November Yunan, China 1,300 schools destroyed in earthquake
1988
(UNISDR 2008)
19 September Mexico City, Mexico 137 school buildings collapsed, 1,687 school buildings were damaged
1985 :17 M 8.0 Schools were not yet open
(COGSS 2008)
2 May 1983 Coalinga, California, Extensive nonstructural damage noted
23:42 USA
(COGSS 2008)
10 October 1980 El Asnam, Algeria 70 schools totally destroyed, 25 moderately damaged
13:25 M 7.3 School was not in session
(OECD 2004)
9 February 1971 Sylmar, California, Only 4 of 1,544 buildings surveyed suffered severe damage. Nearly
06:01 USA all damage was nonstructural
(State of M 6.6 School was not in session
California 2009)
31 May 1970 Chimbote, Peru 6,730 classrooms collapsed and hundreds seriously damaged
4:23 M 7.9
(COGSS 2008)
27 March 1964 Alaska, USA Primary school destroyed by an earthquake-induced landslide. Half of
(COGSS 2008) Anchorage’s schools were significantly damaged
The earthquake struck on a holiday, Good Friday, so schools were
closed
1963 Skopje, Macedonia 44 schools (57 % of urban stock) were damaged
(COGSS 2008) 50,000 students affected. Sunday, school not in session
21 July 1952 Kern County, 20 schools damaged or destroyed (most built before 1933).
4:52 California, USA M 7.3 Significant nonstructural damage also noted
(COGSS 2008)
4 March 1952 Sapporo, Japan 400 schools collapsed in Sapporo
(USGS 2003)
10 March 1933 Long Beach, California, 70 schools collapsed
(COGSS 2008) USA The earthquake hit early in the evening after children had left for the
day which saved their lives. Five students were killed in a gymnasium
3 February 1931 North Island, New Several schools were severely damaged
(Dowrick and Zealand The earthquake happened at mid-morning during school playtime
Rhoades 2004) when the children were outdoors enjoying the summer weather. Some
students were killed, but the death toll could have been several
hundreds
17 June 1929 Murchison, New College tower and dormitory roofs collapsed
10:17 Zealand School was not in session
18 April 1906 San Francisco, 28 schools burned in fire. 41 schools damaged or destroyed
05:12 California USA Classes were not in session
a
Magnitudes from USGS
Bibliographic references on many structural impacts on schools are available on the internet (Rodgers 2012)
life and the right to education are both recognized Around the world, at least 100 million chil-
human rights, and both should be met. This argu- dren of school age do not attend school
ment takes on additional salience in view of the representing about 14 % of the world’s chil-
current international effort to increase school dren (UNESCO 2004). Providing facilities to
enrollment and attendance by girls, disabled chil- educate them requires construction of schools
dren, and children of the very poor and margin- and rapid expansion of building programs.
alized groups in society. The Education for All campaign originally
hoped to enroll 24 million of these children in Cost-Effectiveness Arguments

a decade. Millennium Development Goals
(MDGs) specifically aim to “[e]nsure that, by There are two forms that cost-effectiveness argu-
2015, children everywhere, boys and girls ments may take. One asserts that the authority
alike, will be able to complete a full course responsible for education incurs greater cost in
of primary schooling.” the long run to repair and replace schools dam-
In 2004 it was estimated that more than 7,500 aged by earthquakes than the cost of enforcing
new schools were needed within the next 3 years building codes and making sure that every new
solely in Afghanistan, a country with a significant school is a safe school (or even of retrofitting
seismic hazard. It would be ironic and tragic if in older or poorly built schools). In some cases,
the course of achieving one MDG, another is replacement of unsafe schools is more cost-
undermined. effective than repair (e.g., see entry “▶ School
Another MDG target is to reduce the under- Seismic Safety: Case Studies,” for examples
five mortality rate by two thirds, between 1990 from Algeria, Colombia, and Turkey). Notable
and 2015. On the one hand, the international studies of the benefits and costs of retrofitting
community is seeking to save the lives of under- schools in the USA, Italy, Mexico, and Peru
fives, only to put them at risk a few years later have been published in the decade between
when they go to school. 2004 and 2014.
Educational authorities changed with the con- A more ambitious and difficult case to make
struction and maintenance of schools are also the concerns the relative cost-effectiveness of invest-
ones tasked with many other functions: They ments in school seismic safety when compared to
develop curricula, hire teachers, and choose investing that money in other kinds of public
educational resources such as textbooks and health, safety, and welfare. In cases where child
computers. School safety issues have to find and infant mortality is high, longevity is shorter,
a place in the capital, maintenance, and basic vaccinations are not universal, or safe
operation budgets of school buildings and school domestic water and sanitation facilities are inad-
operation. equate, then the relative ranking of school safety
Retrofitting schools for seismic safety can be as a cost-effective public health intervention may
perceived to compete for funds with the rest of be low. Competition for public health funds could
the educational process. The question facing occur in trying to decide between clean water and
decision makers can actually appear to be: vaccinations for everyone versus school seismic
“What is more important: an up-to-date safety. In more affluent countries, the cost-
textbook and good laboratory facilities now or a effectiveness of saving lives in a future disaster
building that can withstand an extreme event usually has a high place among prioritized goals.
which might or might not occur with the next Of course, the physical safety of children, both S
few decades?” in schools and out in the world at large, goes well
Under most circumstances, young people do beyond school seismic safety. HIV/AIDS, mal-
not lobby for their own rights to health and safety. nutrition, sexual violence, malaria, labor prac-
Children cannot refuse to go to school because a tices, and forced military service are day-to-day
building is unsafe. By law, they must attend threats to the physical safety of many of the
school, though teachers, parents, and others may world’s children. The small potential for an earth-
advocate on their behalf. Faculty and support quake over the next century might appear to pale
staff in schools should also be concerned for beside other concerns which daily kill many more
their occupational safety and theoretically be nat- children.
ural advocates of safe school facilities. Yet there However, in places where school seismic
are no examples mentioned to date of teachers safety is a prominent issue – such as Tehran,
unions becoming involved in the issue of school Vancouver, Kathmandu, Bogotá, and
disaster vulnerability. Wellington – a significant earthquake has a high
probability of happening during the lifetime of temporary housing, while still fulfilling their
schools currently standing and, therefore, for the role as an education facility.
gradually changing cohort of children during the Thus schools have a value in the social fabric
life of the building. If earthquakes happen with of a community, providing adult education, pro-
equal probability around the clock, then approx- moting public health, building and maintaining
imately a 6–23 % chance exists of schoolchildren sustainable livelihoods, and protecting people.
being in the school during a damaging earth- The monetary value of those social gains defies
quake. Cost-benefit studies of seismic construc- estimation but clearly adds value and further jus-
tion estimate that it would add about 5 % to the tifies investment in safe school construction and
cost of building a school in the USA, and in other maintenance.
countries the highest estimates are about 15 %, We know from many disasters the important
making the expectation that “every new school be role that schools play in anchoring and speeding
a safe school” a realistic expectation. community recovery. Rapid school re-opening
When a population at risk is predominantly has tangible benefits in terms of children who
children, depending on the country, each death are safe, supervised, and progressing towards
represents 40–70 years of lost life and productiv- their educational goals. Intangible benefits of
ity, and each injury represents 40–70 years of schools functioning normally following a disaster
potentially expensive medical care, such as for include the psychosocial support in the face of
brain or spinal injuries. Fix schools and several loss and change. The importance of operational
generations of children are protected. Health eco- continuity of schools is linked to community
nomics and medical ethics agree that the greatest recovery.
social benefit comes from investment in the To take another example, retrofitting can
health and capacities of children. Aside from spread a message far beyond the school. When
saving lives, the cost of education interrupted, children see their school being seismically
and the serious potential for drop out adds another retrofitted, they may have and may be designed
cost factor that seismic safety could help avoid. to have ripple effects on safer residential con-
struction. However, this is by no means auto-
matic, and just how to maximize school
Argument from the School’s Multiple construction or retrofit experience into a wider
Functions learning opportunity is a promising line of pur-
suit. Schools certainly serve as community hubs
The symbolic, cultural, economic, and political for propagating the seismic safety messages.
significance of schools as a community hub gives School seismic safety can not only protect a
them an importance beyond merely being the site community’s children but also educate commu-
for educating children. Schools often play roles nities to protect themselves.
as central places for meetings and group activi-
ties, including literacy classes, religious services, Progress, Good Practices, and Lessons
political activities, and marriage ceremonies, par- Learned
ticularly in rural areas where the school might be Assessing School Safety from Disasters, A Global
the only location big enough to hold such an Baseline Report (UNISDR 2012) found several
event. Schools may also provide essential nutri- consistent threats to safe school facilities:
tion programs and serve as makeshift hospitals or
vaccination centers even in normal times. • Failure to assure every new school is a safe
Where schools are the safest buildings in a school: Neither donors, governments, nor
community, they often serve as temporary shelter NGO associations have unequivocally com-
from storms and floods. They may be staging mitted to providing evidence or assurances or
areas for first aid or rescue operation or other submitted to monitoring to assure that every
disaster response functions and even provide new school is a safe school. Many small-scale
donors are particularly unaccountable and potential harm they might cause to school
are not reached by the same accountability occupants.
mechanisms and efforts of UN agencies The expert review process that was part of the
and major international non-governmental Guidelines for Safer School Construction (INEE
agencies. 2010) yielded identification of a rich set of
• Multi-hazard awareness is often lacking: In enabling factors associated with successful and
the construction of school facilities, there are sustained programs for school structural safety
many examples of fulfilling resilience to one that all school safety advocates need to consider
hazard, while failing to mitigate against awareness, community ownership, partnership
others – sometimes resulting in schools being and dialogue, quality assurance, appropriate tech-
dangerous in spite of good intentions or lying nology, integrated education, cultivating innova-
unused. tion, encouraging leadership, and continuous
• Impact of construction on education and assessment and evaluation.
family life not well understood: School
remodeling, retrofit, and replacement all have
an impact on existing school programs and Overview of School Building
families. Planning these projects to minimize Vulnerability
adverse impacts continues to be a concern.
• Opportunity for construction and retrofit Rodgers (2012) reviewed earthquake damage
as an educational experience is untapped: assessment reports through 2009, for 32 earth-
School construction and retrofit provide ideal quakes globally and aggregated findings from
opportunities for students and communities to 31 school building vulnerability assessments.
learn the many principles of disaster-resilient Table 3 shows the most commonly cited sources
construction to be applied throughout their of vulnerability from both sources.
communities. This opportunity is typically The general lack of agreement between vul-
wasted as school sites are hidden from view nerability assessment and damage data likely
and the experience is not used as a learning reflects fragmentary and typically inadequate
opportunity. efforts to collect school damage data following
• Lifeline infrastructure failures threaten past earthquakes, as well as a tendency for vul-
school attendance: Vulnerabilities in roads, nerability assessments to identify common char-
bridges, and transportation systems must be acteristics (such as plan irregularities) that rarely
prioritized when school attendance is lead to the severe damage noted in post-
threatened. earthquake damage surveys and reconnaissance
• Failure to prioritize school re-opening jeop- reports. More complete earthquake damage data
ardizes community recovery: Schools play a would provide the best indicator of the S
critical role in disaster recovery and commu- vulnerability-creating characteristics more likely
nity resilience where adults cannot return to to cause severe damage, because many vulnera-
work (UNISDR 2012). bility assessments do not differentiate the sever-
ity of damage expected from observed
The same study found consensus around the deficiencies.
following core commitments required for safe The sources of and characteristics of structural
school facilities: (1) Every new school must be vulnerability can be summarized in terms of:
a safe school. (2) Legacy schools should be pri- configuration (large windows with partial height
oritized for replacement and retrofit. (3) Lifeline walls below create captive columns or narrow
infrastructure and nonstructural safety should be piers, large windows on one side, weak or soft
assessed locally and measures taken to mitigate stories, large rooms, buildings one bay wide often
[dangers]. (4) School furnishings and equipment with irregular plans), building type (vulnerable
should be designed and installed to minimize forms of vernacular and engineered construction,
School Seismic Safety and Risk Mitigation, Table 3 Characteristics found in damage and vulnerability assessments
Cited in 25 % or more Cited in 15–24 %
Damage Vulnerability Damage Vulnerability
Characteristics assessments assessments assessments assessments
Captive columns due to partial height masonry ✔ ✔
infill walls under windows
Non-ductile reinforced concrete frame ✔ ✔
construction
Generally poor construction quality ✔ ✔
Poor-quality engineered materials ✔
Soft or weak story ✔
General plan irregularity ✔
Exterior falling hazards ✔
Maintenance deferred or lacking ✔
Inadequate doors, windows, halls/corridors, or ✔
stairs
Vulnerable masonry construction ✔ ✔
Lack of seismic design understanding by ✔
engineers
Interior architectural and contents hazards ✔ ✔
Windows reducing solid wall area in masonry ✔
construction
Torsion ✔
General vertical irregularities ✔
Rodgers (2012), pp. 4–5
safer traditional construction forms and practices Overview of Global Experiences in

abandoned, standard building plans with seismic Seismic-Resistant School Construction
deficiencies, heavy roofs), location (sites suscep-
tible to ground failure, sites that amplify ground Some of the major policy and programmatic
motions), construction practices (poor quality, endeavors to assure seismic resilient construction
unskilled or low-skilled local labor, reducing of schools, worldwide, as of 2013, have involved
quality to save money or time), materials (poor- important steps such as providing risk maps for
quality engineered materials, weak local mate- safe school site selection, construction guide-
rials), lack of construction inspection, lack of lines, standards, and oversight and commitments
maintenance, subsequent modifications, falling to safe school construction in the context of both
hazards, and inadequate exit pathways (Rodgers post-disaster reconstruction and new school con-
2012). struction to meet Millennium Development
Underlying drivers create an environment Goals.
conducive to the vulnerability-creating character- The provision of risk maps for safe school
istics cited above. Published literature identifies site selection requires both national and
the following: unregulated community-based subnational coordination and often several differ-
construction, scarcity of resources, inadequate ent agencies reporting on the full spectrum of
building codes or zoning, lack of code enforce- geophysical and hydrometeorological risks and
ment, corruption of enforcement mechanisms, taking into account nuclear, biological, and
unskilled or unaware building professionals, chemical hazards. In Peru, a pool of trained con-
lack of accountability, lack of awareness, failure sultants based in universities around the country
to prioritize school safety, and urgent need for are now available to advise Regional Education
large numbers of new schools (Rodgers 2012). Offices on safe school site selection. They draw
from existing risk maps for 115 towns (UNISDR to international standards (UNISDR 2012).
2008). Following the devastating 2010 earthquake in
In the area of construction guidance and Haiti, many donors stated that the schools that
standards, California’s Field Act in 1933 stands they are supporting seismic, hurricane, and flood-
as the starting point of the movement. The Act resilient school reconstruction, though there is no
requited 15 % higher performance standards for program that monitors progress in this regard.
new school construction and introduced stringent In Indonesia in 2009, the Center for Disaster
supervision. Legacy school construction was Mitigation, Institute of Technology Bandung
raised as a policy issue as early as 1938 (CDM – ITB), and Save the Children Interna-
(Garrison Act) but was not enforced until 1968. tional published a handbook of typical school
The oversight system involves structural plans design and a manual on retrofitting of existing
prepared by engineers and approved by the Divi- vulnerable school buildings for the Aceh and
sion of the State Architect. There is recurring West Sumatra Earthquake Response programs.
on-site inspection and a final verification process. The guidelines take into account lessons learned
The more common approach is the develop- in safe school construction, weaknesses in over-
ment of technical guidance for planning, design, sight of local government construction, and the
construction, and local ongoing maintenance. need to incorporate design of dual-purpose multi-
There are numerous variations on this theme. hazard shelters. In 2014 they were considered
For example, in the Philippines, in 2007, the ready for an update.
Department of Education adopted the Principal- In the Philippines, following devastating
Led School Building Program approach where typhoons in 2006, 99 disaster-resilient schools
principals or school heads take charge of the and 26 day-care centers were constructed with
implementation of management of the repair the support of the Department of Education engi-
and/or construction. Assessment, design, and neers, school principals, and community mem-
inspection functions are provided by Department bers. The new buildings, with water and
of Education engineers who assist the principal sanitation facilities, can also serve as evacuation
during the procurement process. The Parent, centers with flexibility to accommodate large
Teacher, and Community Association and other numbers of people for emergency shelter
community stakeholders are responsible for (Global Education Cluster 2011).
auditing procurements (INEE 2010). Interest- Following the 1999 Kocaeli earthquake,
ingly, in Panama, it was the development and 820 of 1,651 schools that were 60 km away in
implementation of the maintenance guidance Istanbul, were found to have sustained some
tool that paved the way for new school construc- damage. Thirty-five schools were replaced,
tion standards (UNISDR 2012). 59 schools were strengthened, and 59 were
There have been several examples of post- repaired (COGSS 2008). S
disaster commitment to “building back better,” Clear warrants and commitments from donors
emerging from a general consensus following the IGOs or INGOs when it comes to safe school
2005 Indian Ocean earthquake and tsunami, on construction are still clearly much needed.
the need use humanitarian assistance and recon- There have also been too few and/or too quiet
struction financing more responsibly. However, commitments to safe school construction in the
in the area of school seismic safety, these good context of the Millennium Development Goals,
intentions have only translated vaguely to mea- in spite of the fact that the Global Partnership for
surable improvements in safe school construc- Education states as its first strategic goal the
tion. In Pakistan, 4 years after a devastating provision of a quality basic education in a safe
earthquake there, the National Education Policy environment. The most important and notable has
2009 section 5.5 addressed Education in Emer- been in Uttar Pradesh, India, where 23.5 million
gencies with several policy actions including children attend school in this moderate to severe
requirements for school construction according seismic risk zone; 21,000 new school buildings
(30/day) were to be built in a 2-year period. of Education develop the process and tools for a
In 2006–2007 the Elementary Education Depart- nationwide school vulnerability assessment pro-
ment proposed to integrate earthquake-resilient gram, which is currently underway.
design into all new school buildings. One primary More than a dozen countries have developed
school, two upper primary, and three additional approaches, conducted significant vulnerability
classroom designs were prepared with detailed assessments, and/or made commitments made to
construction manuals. Disaster-resilient mea- school retrofit since 2000. Several of these were
sures added 8 % to the construction costs. inspired by unacceptable levels of damage expe-
To cope with massive scale of the project, a rienced in recent large earthquakes. Many are
cascading approach prepared 4 master trainers instructive or inspiring, in terms of their scope,
for each of 70 districts. These individuals trained methods, and limitations. Looking at these
1,100 Junior Engineers and Education Officers. regionally allows an overview of both limited
Ten thousand masons were also trained. In Uttar scope and adequacy.
Pradesh every new school is now a safe school Middle East and North Africa: In Algeria,
(UNISDR 2008). vulnerability assessment was done on 526 build-
ings in 190 schools across 9 municipalities in
Algiers, using simple survey forms (Rodgers
Overview of Global Experiences in 2012). In Syria, UNDP is supporting earthquake
School Seismic Retrofit school safety program incorporated into 5-year
plan and institutions for disaster risk reduction
In Sichuan, China, Prior to the 2008 Sichuan are being consolidated (UNISDR 2012). The
earthquake, school principal Ye Zhiping pestered Arab League is currently considering a regional
local authorities until they consented to retrofit approach to disaster risk reduction, which will
the buildings of Sangzao Middle School to hopefully include a comprehensive approach to
improve their safety. He also initiated regular school safety.
evacuation drills. The result of his efforts was North America: In British Columbia, Can-
that during the devastating earthquake, this ada, Vancouver school buildings were surveyed
school provided life safety for all of its students in 1990 (Rodgers 2012). Responding to advocacy
and staff. efforts of the local “Families for School Seismic
The United Nations Center for Regional Safety,” in 2004, the provincial government com-
Development in Kobe began promoting school mitted $1.5 billion Canadian to ensure that BC
earthquake safety initiatives in 1999, in the pro- Schools meet acceptable seismic life safety stan-
cess of resilience-building following the dards by 2019.
Hanshin-Awaji earthquake. A multi-country In the USA, there has been detailed assessment
school seismic retrofit initiative (2005–2008) of 26 school buildings in Kodiak, Alaska, with
sought to make schools safer through self-help, recommendation for retrofit of four. In California,
cooperation, and education. The project engaged a desk assessment of 9.659 pre-1978 school build-
local communities, governments, and resource ings found 7,537 potentially vulnerable buildings.
institutions in demonstration vulnerability Twenty thousand uncertified projects have been
assessments and school retrofit projects in four mapped (California Watch 2011). The state of
to six schools each, in Fiji, India, Indonesia, and Oregon conducted collapse risk assessment of
Uzbekistan. In 2006, the state of Uttar Pradesh, in 2,185 K-12 school buildings using FEMA
India, undertook large-scale disaster-resilient 154 rapid visual screening (RVS) and produced
construction of new schools (Bhatia 2008). structural engineering reports for more than
GeoHazards International also conducted small- 300 buildings. South Carolina has completed a
scale screening and retrofit demonstration pro- prioritization exercise on all public schools; six
jects in vulnerable schools in Delhi, India have been retrofitted. In Tennessee 49 buildings
(Rodgers 2012), and helped Bhutan’s Ministry in 202 schools have been screened using ATC-21
plus local methods, and in Utah, RVS was used on schools were selected for pilot retrofits (Rodgers
a sample of 128 of 1,085 schools in the state 2012).
(Rodgers 2012). Europe and Central Asia: In Europe, discus-
Latin America and the Caribbean: The sion has been robust in Italy and Portugal, inno-
Organization of American States began a com- vations have been led by UNICEF and partners in
mitment to school safety in 1992. A coordinated Central Asia, and World Bank financing has
regional action plan was developed to benefit supported Turkey to make significant progress
Costa Rica, El Salvador, Guatemala, Honduras, in seismic safety (see entry “▶ School Seismic
Nicaragua, and Panama. Development assistance Safety: Case Studies” for case study on Turkey).
donors and local organizations contributed to In Yerevan, Armenia, full assessments have
strategies and capacity to carry out retrofitting been conducted by teams of dozens of people,
of educational facilities. School infrastructure mobilized from as many as seven different gov-
experts from each country received training. ernment agencies, over several days. Every year
In Bogotá, Colombia, in 1997, seismic 40 of Yerevan’s 200 schools are slated for special
microzonation studies paved the way for maintenance, upgrading, and retrofitting. It has
seismic-resistant building codes in 1998. In been noted that a 2-person expert team spending
2000 the Directorate of Prevention and Attention 2 h per conducting a rapid assessment would
of Emergencies in Bogotá found 434 of require 6 FTE years to assess Armenia’s 1,500
710 schools vulnerable to earthquake damage, schools. In Kyrgyzstan, a national school safety
3 in flood areas, and 20 in landslide-prone areas. assessment of over 3,000 learning facilities with
Two hundred and one were prioritized for retrofit support from USAID found that more than 80 %
or replacement. Between 2004 and 2008, an were vulnerable to earthquake damage. Public
investment of $460 m USD in school replace- access to this information is made possible
ment, retrofit, and risk management promotion through an online portal (UNICEF 2011).
provided structural reinforcement of 172 schools, In Uzbekistan 1,000 school buildings were
“nonstructural” risk reduction in 326 schools, and assessed, revealing that 51 % require demolition
the construction of 50 new mega-schools, com- and replacement, 26 % require capital repair and
pliant with earthquake-resistance requirements. reinforcement, and 27 % are life safe and require
Three hundred thousand children are safer as a no intervention (Khakimov et al. 2007). Eleven
result (see entry “▶ School Seismic Safety: Case design institutes participated in building code
Studies” for case study of Colombia). In Ecuador, revision for school building construction. Typical
initial screening of 340 high-occupancy school designs were created for new schools of different
buildings, and modified RVS of 60 most vulner- sizes. A database of typical construction and
able, detailed analysis for 20, and retrofit designs technical decisions seismic reinforcement were
for 15 have taken place (Rodgers 2012). In Lima, developed. UNCRD provided financial and tech- S
Peru, 28 schools in Barranco and 80 schools in nical support for demonstration projects on
Chorrillos were evaluated using ATC 21 RVS reinforced concrete frame, masonry, and frame
and EMS_98 estimation of damage potential panel buildings. The incremental cost of seismic
(Rodgers 2012). A retrofit solution was devel- reinforcement was shown to be between 3 % and
oped to mitigate potentially devastating struc- 14 % depending on intensity zone, type of con-
tural defect of “short columns.” And in struction, number of floors, capacity, and ground
Venezuela, 50-year-old schools were identified conditions (Khakimov et al. 2007).
as needing retrofitting in moderate and above In Italy, a substantial contribution comes in
seismic zones, whereas 20–30 year-old “box” the form of an overall risk management frame-
schools only require retrofit in higher-risk work developed for retrofit prioritization (Grant
zones. Practical retrofitting techniques were et al. 2007). Some schools have now been
developed. As of 2007, 28,000 schools were assessed in Emilia-Romagna (Rodgers 2012).
being surveyed in a national program. Twelve Portugal has demonstrated an important
innovation by incorporating school vulnerability Nepal has also made some strides in both
assessment and retrofit into its ongoing modern- vulnerability assessment and retrofit planning.
ization program. At least 330 public school build- There are an estimated six million children and
ings have been assessed and retrofits designed 140,000 teachers at risk of death and injury in
(Rodgers 2012; UNISDR 2012). schools. In the Kathmandu Valley, 643 schools
The Istanbul Seismic Risk Mitigation and (1,100 buildings) have been inventoried and
Emergency Preparedness Project (ISMEP) (with 378 (695 buildings) surveyed for vulnerability.
loans from World Bank and EIB) allowed for Seventy-five percent are expected to be damaged
retrofitting of 250 schools and reconstruction of beyond repair, in a scenario earthquake. A school
36 schools in 2007–2008 with 600 more under- day earthquake would kill 29,000 children and
going assessment and feasibility studies. In 2009 teachers and injure 43,000 (Dixit et al. 2013). The
the remaining 450 schools were slated for MoE has planned to retrofit 900 schools in the
retrofitting. Kathmandu Valley over 5 years (Dixit
South Asia: Bhutan has begun a nationwide et al. 2012). In Lamjung and Nawalparasi, vul-
vulnerability assessment of school buildings. The nerability screening has covered 745 and
first phase, covering 5 of Bhutan’s 20 districts, 636 buildings, respectively, some with detailed
began in 2013, with funding from UNICEF. assessments (Rodgers 2012).
In India there are several examples of large- In Pakistan, in 2008, the Aga Khan Planning
scale seismic vulnerability assessments: In Guja- and Building Services, Habitat Risk Management
rat a modified RVS was conducted for Program in Northern Pakistan, used retrofitting of
153 schools following the 2001 earthquake four schools to demonstrate structural and
(Rodgers 2012). In Shimla, SEEDS India took a nonstructural seismic retrofitting, to train
stepwise approach: Step one was low-cost mass builders and to train female village youth in map-
scale RVS of school buildings. From these, a ping, land-use planning, and disaster manage-
smaller number were selected for simplified vul- ment (INEE 2010).
nerability assessment using limited engineering Southeast Asia: There has been relatively
analysis. The highest-risk buildings were identi- sparse activity when it comes to seismic safety
fied for detailed vulnerability analysis (SEEDS of schools in Southeast Asia. It may be that the
2006). Retrofitting designs were drawn up for frequency of cyclones and flooding and even the
20 schools and implementation carried out in threat of tsunami precede thoughts of earthquake
ten schools. Guidelines were developed for retro- risks. It may also be that the rapid pace of devel-
fit and training of local masons and engineers and opment and the increasing numbers of new chil-
delivery of skill training. “Nonstructural mitiga- dren being brought into school have led to natural
tion plans” were carried out in 20 schools. An prioritization of safe new construction rather than
awareness campaign reached out to all retrofit. In the Philippines, local authorities are
750 schools, including nearly 100,000 students, responsible for school construction. However,
7,500 teachers and local builders, engineers, and assessment, design, and inspection functions are
officials (SEEDS 2006). The Government of provided by Department of Education engineers
India’s National School Safety Program plans to who assist the principal during the procurement
seismically retrofit more than 40 schools through- process. The Parent, Teacher, and Community
out the country as demonstration projects. The Association and other community stakeholders
National Center for Peoples’ Action in Disaster are responsible for auditing procurements.
Preparedness (NCPDP), GeoHazards Interna- Earthquake, typhoon, flood, and even volcanic
tional, and others also carried out school assess- ashfall resilience must often be factored in
ment and retrofit programs. (INEE 2010).
East Asia: School seismic safety has been on all-hazard and comprehensive approach so that
the agenda in Japan for many years, but it is only the solutions to seismic safety do not ignore
since 2005 that 125,000 public school buildings coexisting vulnerabilities to cyclones, floods,
nationwide have been assessed by the Ministry of and volcanic eruption nor conflated with the
Education (MEXT) (Rodgers 2012). Sixty-two broader approach that also addresses disaster
percent of these were constructed before 1981, management and education (Global Alliance for
when the current anti-seismic code enforcement DRRR in the Education Sector 2014).
began. About 25 % of schools are considered Overall, the threat of earthquake damage to
safe, but 48,000 older school buildings were school buildings has not been sufficiently well
found needing assessment or retrofitting. 10,000 appreciated. School safety issues have not fea-
of these were found to be at high risk of collapse tured in the major global campaign for increased
in expected earthquakes. The Ministry of Educa- school attendance (“Education for All” and the
tion raised subsidies for vulnerable school build- Millennium Development Goals). The full extent
ings from 50 % to 67 % in 2008 when 229 billion of the risk to school buildings and to students
JPY was allocated to meet the new goal of remains to be fully defined.
retrofitting of all highest-risk school buildings A global effort at mapping schools (by density
within 4 years. of occupancy and quality of construction) in rela-
Oceania: School seismic safety is also on the tion to seismic and other hazards has been pro-
agenda in New Zealand, where a walk through posed by the World Bank Global Facility for
survey of 21,000 buildings at 2,361 public Disaster Risk Reduction and Recovery, to begin
schools in 1998 triggered a follow-up investiga- in 2014. The full impacts of earthquakes on the
tion in 2000 (Rodgers 2012). A World Bank education sector cannot end with calculating the
GFDRR project demonstrated school retrofit in value of structural and nonstructural damage. The
six schools in two districts (2008–2009). impacts on children’s education are almost
entirely unmeasured. Research is needed to
understand how educational outcomes such as
Summary enrollment, attendance, and achievement are
impacted by earthquakes.
In the course of the past decade, an approach to There are strong arguments that support giv-
all hazards and all aspects of school safety has ing school seismic safety increased priority and
emerged in both the literature and practice of a higher profile. An initial step in raising the
global advocacy. The Global Alliance for Disas- visibility of this issue was the adoption of
ter Risk Reduction and Resilience in the Educa- school safety as one of the focal points for
tion Sector (led by UNESCO, UNICEF, advocacy in preparation for the Hyogo Frame-
UNISDR, IFRC, INEE, Save the Children, Plan work for Action 2005–2015 adopted at the S
International, World Vision) use the shared Com- World Conference on Disaster Reduction held
prehensive School Safety framework. The frame- in Kobe, Japan, in January 2005. The develop-
work takes a multi-hazard approach and ment of the Comprehensive School Safety
addresses the many different factors needed to framework in 2013 has begun to articulate
address safe school facilities, school disaster how school facilities safety can be understood
management, and disaster reduction education. within the wider context that includes school
While seismic vulnerability (and related second- disaster management as well as risk reduction
ary hazards) to school buildings are naturally of and resilience education. As a post-2015
concern to earthquake engineers and many agenda for both development and disaster risk
others, it is important to fit this into an reduction are currently under consideration, it
continues to be extremely important to raise In 2009, Guidance notes on safer school con-
the profile of school safety. In preparation for struction (INEE 2010) was published to synthe-
this, child-centered organizations have formed size and kick-start systematic guidance. An
a Global Alliance for Disaster Risk Reduction important global resource for documents to
and Resilience in the Education Sector. guide safe school construction was initiated by
Based in part on the case studies (see entry UNESCO IPRED, immediately after the Haiti
“▶ School Seismic Safety: Case Studies”), earthquake (UNESCO IPRED 2010). This
it seems evident that low-cost, accessible tech- resource database endeavors to compile both
nology and design exists with which to build building codes and the now numerous documents
new schools and to retrofit existing ones. produced by NGOs or at the national level with
A community-based approach holds great prom- standard designs for safe school construction, and
ise involving many stakeholders, including local in some cases with construction guidance.
buildings, masons, contractors, etc. Promising The past decade has seen several relevant sci-
demonstration and large-scale projects in Nepal, entific papers suggesting methods for vulnerabil-
India, Turkey (see entry “▶ School Seismic ity screening (e.g., in Italy, Grant et al. 2007), and
Safety: Case Studies”), Central Asia, and the detailing approaches to seismic retrofit. The chal-
Caribbean islands all provide strong experience lenge is whether or not the guidance and the
to build upon for case studies. science are put into practice. The written record
Case studies also make clear that child rights does not suggest that these approaches are yet
advocates, parents, and seismic safety experts systematic, are supported with training, are mon-
together, lobbying for school seismic safety, can itored, or are applied to both public and private
be extremely effective in achieving policy schools. Community-built and un-engineered
change, as case studies of British Columbia and construction has been addressed in far fewer pub-
Bogotá (see entry “▶ School Seismic Safety: lications and has not specifically addressed
Case Studies”) show. school construction.
School seismic safety has been the subject of There have been a small number of significant
both research and policy since the 1933 Long programmatic efforts to support seismic safety.
Beach earthquake that spurred California’s land- UNICEF’s regional office for Central and Eastern
mark Field Act, requiring that school construc- Europe and the Commonwealth of Independent
tion meet seismic safety standards. As both States, with support from the World Bank and
seismic risk assessment and building codes have DIPECHO, has partnered with national govern-
progressed, so too have expectations for selection ments in Central Asia and the South Caucasus to
of performance standards. However, globally, the address school safety. Part of that work has
application of these standards and codes falls included developing a broad regional framework
short in several major respects: community-built for assessing and ranking school facilities based
schools are frequently constructed using high- upon exposure and vulnerability to earthquakes
tech materials intended for engineered construc- and other natural hazards. Drawing upon INEE’s
tion, without the corresponding understanding, Guidance notes on safer school construction,
training, or supervision; where building codes UNICEF elaborated a list of 17 simple indicators
exist they are not known, understood, or consis- that local experts could use as part of a rapid
tently applied; and safe site selection is fre- visual assessment of school facilities in order to
quently skipped, and site-specific hazards are identify schools at risk of heavy damage in seis-
not factored in. Privately built schools are often mic events.
exempt from the same standards of construction In 2012, engineers in Kyrgyzstan localized
as public schools. The need for programs and this framework and carried out a national school
people that bridge the available engineering safety assessment of over 3,000 learning facilities
knowledge with scalable on-the-ground national with USAID funding. They reported to the
programs is significant. national government that over 80 % of learning
facilities were vulnerable to damage in seismic mapping of local hazards; enforcement of seis-
events and provided public access to the assessment mic building codes by national, provincial, and
through an online portal. Similar national assess- local governments; training of engineers and
ment strategies are being piloted in Kazakhstan, significant capacity-building efforts to train
Tajikistan, Armenia, and Azerbaijan. local masons and other builders; and invention
Similarly, UNCRD (UN Centre for Regional of more innovative models for funding rein-
Development) showcased community-based forcement of schools.
comprehensive school earthquake safety in It is important, however, not to fetishize the
selected countries of Asia Pacific. Under the pro- safety of school buildings and to take care not to
gram “Reducing Vulnerability of School Chil- separate the safety of the community of users
dren to Earthquakes,” school communities and educational continuity planning, which is
carried out seismic retrofitting of their school not limited to the buildings themselves. Neither
buildings with expert guidance from Bandung should the focus be solely on fatality preven-
Institute of Technology (ITB) in Indonesia. The tion. There is much similar work to be done to
retrofitting works in public schools were used for prevent disability and injury especially by
community awareness on earthquake safety securing the contents of the buildings and to
through community visits in the school premises assure educational continuity. All-school, par-
during construction time. Pilot school assessment ticipatory school disaster management plan-
and retrofitting in Fiji led to the National Disaster ning, local risk assessment and risk reduction,
Management Office (NDMO) adopting school mastery of emergency response skills, and reg-
safety program under regular government that ular drills to practice and improve readiness are
also developed seismic retrofit guidelines and important. A culture of safety is necessarily
mason’s training manual. Tashkent city govern- multifaceted, and activism in one area encour-
ment (Hokimiyat) in Uzbekistan appraised neigh- ages changes in consciousness, expectations,
borhood associations on schools retrofitting and demands.
programs and used school constructions for train- The enthusiasm for making education acces-
ing of engineers on seismic safety. sible to all does not absolve duty-bearers from
The United Nations International Strategy for assuring that school is safe from infrequent but
Disaster Reduction (UNISDR) launched the high-impact hazards such as earthquake and var-
2006–2007 biennial awareness campaign “Disas- ious secondary hazards. It would be an ironic and
ter Reduction Begins in Schools.” This was tragic result if the achievement of one Millen-
followed up in 2010 with the Resilient Cities nium Goal (increased school attendance) is
Global Campaign for One Million Safe Schools marred by increased death and injury of young
and Hospitals Campaign. The 10-point checklist people, thus setting back the achievement of
that 1,643 Mayors have signed on for, includes another Millennium Development Goal S
assessing and upgrading the safety of schools. (reduction of child mortality).
These successes deserve praise but should not
induce complacency. There is a long way to go
with respect to school seismic safety.
Initial programs and guidance for safe school Cross-References
facilities have been provided by OECD (2004),
UNCRD (2008), INEE/World Bank GFDRR/ ▶ “Build Back Better” Principles for
UNISDR (2010), and several other programs, Reconstruction
with modest support of donors and lenders. ▶ Building Codes and Standards
These approaches experiences are now ripe ▶ Earthquake Protection of Essential Facilities
for implementation at scale. These include ▶ Earthquake Risk Mitigation of Lifelines and
regional hazard mapping and revision (where Critical Facilities
necessary), the potential for crowd-sourced ▶ School Seismic Safety: Case Studies
References OECD (2004) Keeping schools safe in earthquakes.

Organisation for Economic Co-operation and Devel-
Bhatia S (2008) Earthquake-resistant school buildings in opment, Paris. http://www.oecd.org/document/36/
India. Urb Plan Des 161(4):147–149 0,2340,en_2649_201185_33630308_1_1_1_1,00.html.
California Watch (2011) K-12/public safety/spotlight/on Accessed 19 Feb 2014
shaky ground. 22 Dec 2011. Accessed 3 Feb 2012 Pandey B (2013) Political economy study on safe schools,
CNN (2014) Haiti earthquake fast facts. http://edition. World Bank, Global Facility for Disaster Reduction
cnn.com/2013/12/12/world/haiti-earthquake-fast-facts/ and Recovery
index.html. Accessed 6 Jun 2014 Rodgers JE (2012) Why schools are vulnerable to
Coalition for Global School Safety (2008) School disaster earthquakes. In: Proceedings, 15th world
reduction: think globally, act locally. ‘Mother’ slide conference on earthquake engineering, Lisbon, 24–28
presentation Sept 2012
Dixit A et al (2012) Institutionalization of school earth- SEEDS (2006) Earthquake Safety in Mountain Regions.
quake safety program in Nepal. In: Proceedings of the School Earthquake Safety Initiative. Shimla, Delhi
15th WCEE. http://www.iitk.ac.in/nicee/wcee/article/ State of California, Seismic Safety Commission
WCEE2012_5260.pdf. Accessed 16 Feb 2014 (2009) The field act and its relative effectiveness in
Dixit et. al. (2013) Public School Earthquake Safety Pro- reducing earthquake damage in California’s public
gram in Nepal, Geomatics, Natural Hazards and Risk, schools. http://www.seismic.ca.gov/pub/CSSC%
DOI: 10.1080/19475705.2013.806363 2009-02%20Field%20Act%20Report.pdf. Accessed
Dowrick D, Rhoades DA (2004) Risk of casualties in New 16 Feb 2014
Zealand earthquakes. Bull N Z Soc Earthq Eng http:// UNCRD (2008) Reducing school vulnerability to earth-
www.nzsee.org.nz/db/Bulletin/Archive/38(2)0053.pdf. quakes. United Nations Center for Regional Develop-
Accessed 5 Mar 2015 ment, Hyogo, Japan. http://www.hyogo.uncrd.or.jp/
EERI (1990) Earthquake Engineering Research Institute, publication/pdf/Report/2008SESI/UNCRD%20SESI%
Oakland. Loma Prieta Earthquake 17 Oct 1989. Prelim- 20publication%202008.pdf. Accessed 16 Feb 2014
inary reconnaissance report. https://www.eeri.org/wp- UNESCO (2004) ‘Education for all week 19–25 April 2004:
content/uploads/USA_CA_LomaPrieta_PrelimReport. the extent of the problem’, information from UNESCO’s
pdf. Accessed 16 Feb 2014 (United Nations Educational, Scientific and Cultural
FEMA (2011) Reducing the risks of non-structural Organization) http://portal.unesco.org/education/en/
safety – a practical guide, 4th edn. Washington, ev.php-URL_ID=28702&URL_DO=DO_TOPIC&
DC. http://www.fema.gov/media-library/assets/docu- URL_SECTION=201.html. Accessed 25 Jan 2014
ments/21405?id=4626. Accessed 19 Feb 2014 UNESCO IPRED (2010) Resource/database, disaster
Global Alliance for Disaster Risk Reduction and Resil- preparedness and mitigation. http://www.unesco.
ience in the Education Sector (2014) Comprehensive org/new/en/natural-sciences/special-themes/disaster-
school safety. http://preventionweb.net/go/31059. preparedness-and-mitigation/natural-hazards/earth-
Accessed 15 Jan 2014 quakes/ipred/unesco-ipred-database/. Accessed 25 Jan
Global Education Cluster (2011) Disaster risk reduction in 2014
education in emergencies: a guidance note for educa- UNICEF (2011) Toward safer schools – methodology for
tion clusters and sector coordination groups. http:// nationwide benchmarking of school safety http://www.
preventionweb.net/go/20366. Accessed 16 Feb 2014 unicef.org/ceecis/towards_safer_school_FINAL.pdf.
Global Partnership for Education (2014) Results and data. Accessed 5 Mar 2015
http://www.globalpartnership.org/results/. Accessed UNISDR (2008) Disaster prevention for
16 Feb 2014 schools – guidance for education sector decision-
Grant DN, Bommer JJ, Pinho R, Calvi GM, Goretti A, makers. Geneva. http://www.preventionweb.net/go/
Meroni F (2007) A prioritization scheme for seismic 7344. Accessed 25 Jan 2014
intervention in school buildings in Italy. Earthq Spec- UNISDR (2012) Assessing school safety from disasters, a
tra 23(2):291–314 global baseline report. http://preventionweb.net/go/
INEE/World Bank GFDRR/UNISDR (2010) Guidance 35274. Accessed 15 Jan 2014
notes on safer school construction. INEE (Inter-agency USGS (2003) ‘Magnitude 8.3 – Hokkaido, Japan region
network for education in emergencies), New York. 25 Sep 2003 19:50:06 UTC’. http://earthquake.usgs.
http://www.preventionweb.net/go/10478 gov/recenteqsww/Quakes/uszdap.htm. Accessed
Khakimov S, Nurtaev B, Tursunov K (2007) School earth- 25 Jan 2014
quake safety program in Uzbekistan, slide presenta- USGS (2015) Historic World Earthquakes. http://
tion. Tashkent earthquake.usgs.gov/earthquakes/world/historical.php.
New York Times (2008) School-age children in earth- Accessed 5 Mar 2015
quake zones, 27 May 2008. http://www.nytimes.com/ WEF (2000) UNESCO World Education Forum. http://
imagepages/2008/05/27/science/27school.graphic1. www.unesco.org/education/efa/wef_2000/. Accessed
ready.html. Accessed 16 Feb 2014 5 Mar 2015
School Seismic Safety: Case Studies 2469
Mitigation. They tell the stories of a variety of

School Seismic Safety: Case Studies national efforts to improve school seismic safety.
The contributors to the case studies are engineers
David Alexander1, Sanjaya Bhatia2, Djillali and parents, social workers, and international
Benouar3, Jitendra Kumar Bothara4,5, development specialists. They examine policy,
Omar-Dario Cardona6, Amod Mani Dixit4, advocacy, vulnerability, and solutions. They con-
Rebekah Green7, Manu Gupta8, Ram Chandra tain observations about stepwise progress, motiva-
Kandel9, Ilan Kelman10,17, Tracy Monk11, tion, political will, technical approaches,
Bishnu Pandey12, Marla Petal13, Zeynep innovations, moderate successes, and long roads
T€urkmen Sanduvaç14, Rajib Shaw15 and ahead. There are more stories to be added. The
Ben Wisner1,16 intention is to provide school seismic safety advo-
1
University College London, London, England cates with both elements of inspiration and way
2
UNISDR Recovery Platform, Kobe, Japan points on a road map with many options to
3
University of Bab Ezzouar, Algiers, Algeria consider.
4
NSET, Kathmandu, Nepal The case studies and their contributors are:
5
Miyamoto Impact, Christchurch, New Zealand
6
Universidad Nacional de Colombia, Manizales, Algeria – Djilali Benouar
Colombia Canada – Tracy Monk
7
University of Western Washington, Bellingham, China – Sanjaya Bhatia
WA, USA Colombia – Omar Dario Cardona
8
SEEDS of India, Delhi, India India (Delhi, Shimla) – Manu Gupta
9
Toronto, ON, Canada India (Uttar Pradesh) – Sanjaya Bhatia
10
Institute for Risk and Disaster Reduction and Italy – David Alexander
Institute for Global Health, University College Japan – Rajib Shaw
London, London, England Nepal – Amod Mani Dixit, Jitendra Kumar
11
Families for School Seismic Safety, Botendra, Ram Chandra Kandel, and Bishnu
Vancouver, Canada Pandey
12
University of British Columbia, Vancouver, Turkey – Marla Petal and Zeynep T€urkmen
Canada Sanduvaç
13
Risk RED (Risk Reduction Education for USA (California) – Marla Petal and Rebekah
Disasters), Los Angeles, CA, USA Green
14
Risk RED, Istanbul, Turkey
15
University of Kyoto, Kyoto, Japan It is important to note that school seismic
16
Oberlin College, Oberlin, OH, USA safety should not be isolated from school safety
17
Norwegian Institute of International Affairs, from disasters and climate change impacts in S
Oslo, Norway general. It cannot be considered responsible to
approach the rights of children from our narrow
professional silos. It is incumbent upon all advo-
Synonyms
cates for children to see the matter from their
point of view, that is, from the perspective of all
Education sector; Examples; Policies; Policy;
of the physical threats to their right to survival
Progress; Schools; Seismic safety
and safety and from all of the threats to their right
to education and development.
Introduction Similarly, key stakeholders in the seismic
safety of schools (engineers, architects, builders)
The case studies in this entry accompany the pre- must make the effort to think not primarily about
vious entry on School Seismic Safety and the structures, but rather about the people who
2470 School Seismic Safety: Case Studies
use the structures. The users must also have safe seismic risk; and (3) those built with technical
access and egress. For the users, infill walls that supervision after 1983 and the introduction of
fall out of plane and unsecured nonstructural Algeria’s seismic building code in 1981.
building elements or building contents are a far Schools in Algeria are all state owned and
greater threat than their “minor damage” desig- were built by the government. The government
nation suggests. If the building is going to be one adopted one typical structure for all schools that
of the strongest structures in the community, then could be duplicated easily across the country. The
it may also need to be planned to be as a cyclone standard architectural design of schools involves
shelter or to double as a shelter for people two fundamental elements: the classroom and the
displaced after an earthquake. In that case, circulatory corridors. Classrooms are 7 4 m and
water and supply storage and extra sanitation circulatory corridors are 2.5 m wide. These are
facilities have to be considered. And the job is far from those of an ideal seismically resistant
not complete when the key is handed over. Build- structure as recommended by Algeria’s own seis-
ings that do not come with a user’s manual and a mic codes. Other standard design elements also
school maintenance calendar stand little chance unfortunately reduce the resilience of these
of being safely maintained. Thus, the concern for school structures.
school seismic safety does not begin and end with Numerous reports show the deficiencies in
the structures themselves, but must take a holistic design, construction techniques, and materials
approach and, of course, include the user (poor quality of concrete) with respect to partic-
community. ular earthquakes at El Asnam (1980), Chenoua-
Tipaza (1989), Beni Chougrane-Mascara (1994),
Ain Temouchent (1999), and Boumerdes-Algiers
Algeria (2003) (Bendimerad 2004). The following typical
damage to school buildings was recorded in
Ninety percent of Algeria’s population of 30 mil- recent earthquakes:
lion is concentrated in a band about 60 km wide
and 1,200 km long along the Mediterranean coast • Rupture of staircases
on the African and Eurasian tectonic plate bound- • Destruction of joints
ary. This region has repeatedly experienced • Destruction of short columns
moderate-to-strong earthquakes. During the twen- • Damage in Masonry
tieth century, earthquakes claimed at least 10,000 • “Pancake” collapse due to weak columns,
lives, injured about 27,000, and made about half a overly strong beams, and heavy roofs com-
million homeless. In addition to building collapse posed of reinforced concrete slabs
due to earthquakes, considerable damage from
liquefaction and landslides was observed. Such damage causes enormous financial loss
School buildings have also suffered consider- to the government. For instance, after the
able damage in earthquakes, varying according to Boumerdes-Algiers earthquake disaster of 2003,
the period during which they were built: (1) those 100 primary schools had to be rebuilt completely
degraded through aging and lack of maintenance, for the sum of US$4.28 million and 253 rehabili-
built during the colonization era (1830–1962), tated for $10.65 million. In addition, 12 junior
account for about 30 % of the school building high schools were completely rebuilt for the sum
stock; (2) those built after independence, during of $10.28 million and 111 rehabilitated for
the 1970s, with rapidly growing population and $20.85. Also, 10 high schools were rebuilt for
democratization of educational opportunity the sum of $21.42 million, and 58 were rehabili-
(when primary school became free and compul- tated for $12 million.
sory), when school construction accounted for the So far these destructive earthquakes have
largest single item in Algeria’s budget, were occurred after school hours or on weekends, and
designed and built without taking into account thus, no loss of life or injuries have been recorded
at schools in Algeria. This good luck may have experiencing structural damage in an earthquake,
made the government and the civil society alike and 15 % were at moderate risk (Taylor 1989).
less aware of the high vulnerability of the schools Between 1990 and 2004, 11 schools have been
and partially explains why there has so far been upgraded, so if the current pace continues, this
no implementation of a Ministerial instruction work would be completed in 2064.
dating from 1989 that required application of While the seismic hazard facing schools in
“technical expertise and the eventual reinforce- greater Vancouver is similar to that in Seattle,
ment of all public buildings and in particular Washington, school safety has not been a priority
schools and universities.” Instead, the introduc- on the Canadian side of the border. Indeed, dif-
tion of new materials such as reinforced concrete ferences in seismic hazard mapping techniques
in the absence of proper seismic-resistant design, used by Canadian and US geoscientists may actu-
building codes, and enforceable regulations has ally underestimate the risk to Vancouver schools.
increased the risk to structures and their occupants. The current National Building Code of Can-
Relatively minor reinforcements could reduce the ada ranks buildings according to their priority as
potential for damage to these structures. critical infrastructure. The higher the number
assigned, the higher the priority. Average houses
are assigned an Importance Factor of 1.0, schools
Canada are designated 1.3, and hospitals, police stations,
and prisons are assigned 1.5. Schools – unless
British Columbia’s (BC) west coast is Canada’s they are designated as post-disaster
region of highest seismic hazard. Two-thirds of shelters – get a lower priority than hospitals,
the province’s 3.9 million people live within the police stations, and prisons. Vancouver City
zone of highest risk. The region has experienced Council is funding the seismic upgrading of com-
ten moderate-to-large earthquakes since 1870. munity centers so that they could be used as post-
In recent millennia, an earthquake on the order disaster receiving centers. Due to the differences
of the largest magnitude experienced globally in in funding sources, some community centers are
the last 100 years has struck approximately every being upgraded, while nearby schools, which
500 years. In addition to potential building col- might be less seismically resistant, are
lapse induced by ground shaking, significant sometimes not.
damage from liquefaction, tsunami, and land- In general in BC there is high awareness of
slides are expected. earthquake risk. For example, the City of Van-
Older BC schools were built from some of the couver has seismically upgraded its water supply
most seismically vulnerable materials – in the system and bridges, and the electric utility, BC
early 1900s, unreinforced masonry and then, in Hydro, is systematically upgrading its buildings
the mid-1900s, non-ductile concrete frame. Thus, and infrastructure, including dams. Even some S
in many communities, the school might be one of provincially run liquor outlets have been seismi-
the buildings at highest risk for earthquake dam- cally upgraded. With seismic safety so clearly on
age. A formal risk assessment of all BC school the policy agenda in many sectors in British
buildings was initiated in June 2004 with the full Columbia, why have public schools received so
report due in October 2004. Initial estimates from little attention?
the provincial government suggest that 800 of The answer is that in BC funding for public
BC’s schools might need some form of seismic school seismic upgrades has been part of the
upgrading. capital budget of the provincial Ministry of Edu-
In Vancouver, BC’s largest city with a core cation. Local school boards oversee this work and
population of about 560,000 and a metropolitan must proactively request provincial funding for
population of about two million, a 1989 rapid projects that they deem to be high priority. Advo-
seismic risk assessment found that 30 % of the cacy work by one of the authors on behalf of
city’s school buildings were at high risk of Families for School Seismic Safety British
Columbia (FSSS) identified and tackled concerns China

with this funding system.
First, there was no district-to-district standard- Following the 2008 Sichuan earthquake in which
ization of approach. Each local school district at least 15,000 children lost their lives in schools,
was individually approaching the issue of seismic the Ministry of Education and the Ministry of
safety for only their schools. FSSS pressed Construction and National Development and
the government to unify the approach taken by Reform Commission (NDRC) jointly released
centralizing expertise. The earthquake engineer- the Design Instructions for School Planning and
ing community, through its professional associa- Construction after the Sichuan earthquake. These
tion, the Association of Professional Engineers standards require that school sites are assessed
and Geoscientists of BC, is now advising the before the schools are built in accordance with
government on standardized, peer-reviewed national regulation, performance objectives are
methods for assessing and addressing seismic determined by the country-level government,
risk to schools. schools are built or retrofitted to meet perfor-
Second, local authorities were concerned that mance objectives, and schools’ furnishings and
informing parents of the seismic risks to local equipment are designed and installed to minimize
schools could cause panic and could be politi- potential harm they might cause to school occu-
cally damaging. FSSS’s and others’ work has pants. The quality-monitoring bureau leads mon-
ensured that Vancouver parents are now well itoring on the safety of equipment installation.
informed about the issue and are actively In Sichuan in the spirit of “building back bet-
involved in trying to solve it. This process did ter,” the investigation, design, construction,
not cause panic. Instead, bringing parents into the supervision, inspection, and acceptance of school
consortium has yielded an active and effective construction are conducted in line with relevant
lobbying group. national construction standards. In Sichuan, steps
Finally, there is the problem of setting priori- were also taken to make sure that there are mech-
ties. The primary concern of school anisms to ensure that schools’ maintenance is
boards – quite rightly – is the day-to-day educa- financed and executed. From 2000 to 2005 the
tion of children. Seismic safety of a school build- first and second session of school renovation and
ing does not lead to improved education, so maintenance was conducted. After 2006 a long-
school boards have sometimes had difficulty term mechanism for school building maintenance
making the issue a high priority. FSSS is trying was put into place. MOE and UNICEF collabo-
to help public officials see school seismic safety rated and prepared and revised the National
as an infrastructure, public health, and human Guidelines for Safe School Construction and
rights issue and to obtain new funding from out- Management and also collaborated to develop
side of the Ministry of Education, that is, from construction standards for kindergartens and
provincial and national, authorize with mandates preschools.
in those areas. The aim is that this work be seen as In 2009 the Ministry of Education (MoE) ini-
an infrastructure project for children and not per- tiated a 3-year national “School Construction
ceived as competing for funding with their Safety Programme” to upgrade the safety of pri-
day-to-day educational needs. mary and middle school buildings all over the
Ultimately, the two basic human rights of country with the aim of making schools the safest
children, to an education and to physical safety, places in China. The program has several key
should not be competing for the same funds. The elements:
expert community is now driving the initiative
and the government appears to be listening. Many • To screen and assess the quality of all school
positive steps have been taken in BC, but there is buildings across the country, to understand the
much work left to be done. qualities of buildings resistant to local disaster
risks, and to input the data and information of is the Department of Education’s effort to iden-
the assessment into a database tify school seismic risk and to reinforce schools.
• To understand disaster risks in the regions Much of the educational infrastructure in
where the schools are located, such as deter- Bogota is more than 50 years old and does not
mining whether local disaster risks come from meet minimal standards of safety. The Depart-
floods, landslides, earthquakes, or rainstorms ment of Education commissioned a systematic
threatening the safety of the school buildings review of schools that ran from 1997 to 2003
• To determine whether to repair, strengthen, or (Secretarı́a de Educación del Distrito Capital de
reconstruct school buildings that have not Santafé de Bogota 2000). This study covered
reached official standards based on the inten- approximately 2,800 buildings at 706 schools
sity of the identified disaster risks in the region serving roughly 54 % of the student population
where the schools are located in Bogotá. The other 46 % of the student popula-
• To allocate funds and to start the construction tion attends private schools and was not covered
work to upgrade the primary and middle in this review. The review found that 434 of the
school buildings that are at risk (International schools presented high risk to students. Some
Recovery Platform 2010) 772 buildings at these schools fell into this cate-
gory (16 %). The study also found that 60 schools
had buildings in immediate and urgent need of
Colombia reinforcement.
During the next city administration, from 2004
The capital city of Colombia, Bogotá, is the most to 2008, after a detailed technical explanation on
important political, administrative, economic, the need of schools’ retrofitting, the city mayor
and cultural center of the country and has one decided to implement a retrofitting program of
million school children. Bogotá’s population was the 200 most vulnerable schools. A special risk
estimated to be around 7.6 million in 2013. As a analysis was performed on each building that was
result of social investment over the past decade, identified and prioritized. From this analysis, the
more than 12 % live below the poverty line. structural reinforcement requirements were
Among the most common hazard events defined according to the seismic building code
affecting Bogotá are earthquakes and landslides. updated in 1998, with new special provisions for
Although there has not been a severe earthquake schools. The comprehensive improvement pro-
in Bogotá since 1917, there is certainly the poten- gram had to be adjusted. Taking into account
tial for one. Also, elsewhere in Colombia, 74 % of other technical, urban, economic, and environ-
the schools in the cities of Pereira and Armenia mental issues, many schools were not retrofitted:
suffered damage in the 1999 earthquake. Fortu- 67 schools were demolished and full restitution
nately this occurred during the lunch hour, when was made for a total of 107 new schools. This S
no children were in the school buildings. additional program was called the “50 Macro-
Several risk identification methods were put in schools plan” whose goal was to provide an edu-
place in the city prior to 2004. These include cational infrastructure of maximum specifica-
compilation of records of hazard events, generations and supplement the retrofitting and
tion of hazard maps, studies of physical and integrated improvement program. At the end,
social vulnerability, and studies of environmental due to the costs involved in reducing vulnerabil-
degradation. One of the means of reducing risk ity for 434 vulnerable schools, 201 schools were
from earthquakes and landslides in Bogotá is the considered in critical condition according to risk
assessment of seismic risk of bridges, hospitals, studies that were conducted to prioritize and rank
and schools. This has become a core part of the the schools. The reinforcement of these buildings
city’s economic and social development plan. had a cost about US$ 200 million and the total
Of these assessment programs, the best-known program including the new schools was about
US$ 430 million. Additionally, this program was population, and accelerating the rate of school
based on the implementation of a teaching strat- reinforcement to improve the safety of children
egy to incorporate risk management into the cul- and teachers. Bogota has had different gover-
ture. Both structural and nonstructural objectives nance problems during the last two administra-
were implemented to obtain a comfortable and tion periods. Unfortunately, two mayors have
safe school environment and a high-quality edu- been removed or suspended by the national gen-
cation service. eral attorney due to corruption and inefficiency.
Assessing and reducing the risk to schools in Due to these situations and perhaps changing
Bogotá took place in a more general planning and priorities in risk management, the school safety
management context. For example, zones of high retrofitting program was not continued.
risk of landslide, where no mitigation works are
possible, are declared to be protected land.
Human occupation is restricted in these areas as India: Delhi
well as those considered at high risk to floods. In
2003 it was estimated that some 185,000 people NGO partners SEEDS and GeoHazards Interna-
lived in informal settlements in a total of 34,230 tional (GHI), working with the Government of
informal housing units. In Bogotá there are Delhi, demonstrated earthquake nonstructural
173 illegal settlements that account for 14 % of risk reduction in a public school. The school
the total land area. The city administration has welfare committee comprised of faculty, staff,
developed a massive legalization program since and local community members learned to iden-
1995, thus reducing the number of informal set- tify the nonstructural building elements and
tlements from 1,451 to its current number, an building contents that could fall, slide, or collide
eightfold reduction in less than 10 years. during a likely Delhi earthquake, as well as fire
Nevertheless, as much as 60 % of the popula- and evacuation hazards. They were exposed to
tion of the city lives in informally constructed simple low-cost techniques for reducing these
dwellings. While most of these are located in risks (moving some items, fastening others) and
legal settlements, they still represent a challenge came up with innovative solutions of their own.
to seismic safety. The year 2000 land-use master The logic of regular fire and earthquake drills
plan for Bogotá contains hazard and risk maps became readily apparent to these new stake-
that determined land use, details of special treat- holders. A handbook for schools on nonstructural
ment for high-risk areas, and arrangements for risk reduction developed by the NGO partners
issuance of building permits, as well as protection and published by the Government of Delhi pro-
plans for utilities and services. The city also relies vides a new resource for generalizing these
on community-based networks to control illegal lessons.
land occupation and has developed a large-scale
relocation program for families living in high-
risk conditions. India: Shimla
The city of Bogotá has disaster risk reduction
at the center of its planning process, and in this A small-scale demonstration project for school
context school seismic safety ranks very high. retrofit was carried out by SEEDS of India and
Having diagnosed the scale and urgency of the GeoHazards International. Structural assessment
problem in Bogotá, steps were being taken to of school buildings was carried out using a filter-
reinforce the most hazardous school buildings. ing method: The first step was low-cost mass
The challenges the city still faces include com- scale Rapid Visual Assessment Survey of school
pleting the program, making the retrofitting of the buildings for potential seismic hazards. Based on
second priority group of vulnerable schools, these surveys a smaller number were selected for
extending its school safety program to private Simplified Vulnerability Assessment using lim-
schools that cover nearly half the school-aged ited engineering analysis. The highest-risk
buildings were identified for detailed vulnerabil- detailed construction manuals. The disaster-
ity analysis. Retrofitting designs were drawn up resilient measures added 8 % to the construction
for 20 schools and implementation of retrofit costs. To cope with the massive scale of the
carried out in 8 schools. Guidelines were devel- project, a cascading approach prepared 4 master
oped for retrofit and training of local masons and trainers for each of the 70 districts. These indi-
engineers and delivery of skill training. viduals in turn conducted trainings for 1,100 fel-
“Nonstructural mitigation plans” were carried low Junior Engineers and Education Officers.
out in 20 schools. An awareness campaign was Ten thousand masons were also trained. This
designed to reach all 750 schools in the region program ensures that every new school will be a
including nearly 100,000 students, 7,500 safe school. The problem of preexisting stock of
teachers, and local builders, engineers, and offi- 125,000 unsafe school buildings in need of retro-
cials (SEEDS 2006). fit remains to be tackled (Bhatia 2008).
India: Uttar Pradesh Italy
There are 23.5 million children attending school Of the 8,102 municipalities in Italy, all are
in this moderate-to-severe seismic risk zone. As regarded as “seismically active.” Up until 2003,
part of the Education for All campaign, the State 2,965 of them (representing 40 % of the land
Government of Uttar Pradesh constructed 82,000 surface and 45 % of the population) were consid-
additional elementary school classrooms and ered “highly seismic” category. As a result, new
7,000 buildings in 2006–2007. To ensure seismic construction in this category must observe strin-
resilience of the buildings, UNDP provided earth- gent anti-seismic building codes. Subsequently, a
quake engineers who examined the blueprints for more sophisticated classification was introduced,
the schools and modified the design to integrate based on a 50-year recurrence interval and local
seismic resilience. The marginal cost increase of estimates of peak ground acceleration (PGA).
8 % to assure seismic safety was funded by the This had the effect of increasing the areas classi-
government. fied as highly seismic. Whereas in previous clas-
For effective implementation of the new mod- sifications, some municipalities were effectively
ified designs, training and orientation programs regarded as aseismic, that is no longer the case and
were initiated by the government, supported by all 8,102 are now considered to be in a seismic
the local UNDP office, building the capacity of zone to a greater or lesser extent. Italy bases its
40 architects, over 200 engineers, and over seismic classification on historical records and
10,000 masons. To ensure transparent monitor- calculated return periods. Where these are a poor
ing, the designs were widely circulated to the reflection of seismic hazard, it can underestimate S
local communities where the schools were the earthquake threat. Hence, severe damage
constructed, so they could monitor the quality, occurred in northern Emilia and southern Lom-
along with support of departmental engineers. bardy in the earthquakes of May 2012, an area that
To complete the safety of the schools, school- had not had a major seismic disaster since 1574.
level safety committees were established, school Several hundred municipalities are faced with
emergency plans developed, and mock drills the highest seismic risk in the Strait of Messina
became a part of the school program. (including eastern Sicily) and southern Calabria
In 2006–2007 the Elementary Education (the Aspromonte) areas. The Apennine Moun-
Department proposed to integrate earthquake- tains, which form the “backbone” of the Italian
resilient design into all new school buildings. peninsula, are, in a tectonic sense, divided into
To prepare for this, one design of primary school blocks, which means that seismicity varies sig-
buildings and two upper primary and three addi- nificantly from one locality to another. However,
tional classroom designs were prepared with the vulnerability of buildings, including schools,
is almost universally high: modest local taxation practicing evacuation drills. Yet the area awaits a
revenues inhibiting retrofitting and maintenance. magnitude 6 earthquake, which it is predicted
The largest seismic event of the past century may kill up to 120 people and injure more than
remains that in Avezzano, in 1915. a thousand. How many of them will be school
An event that for Italians most encapsulates children? This situation is typical of the seismic
the seismic risk to schools was that of 31 October risk that affects highland Italy.
2002 at 11:40 a.m. where in the Apennine town of Elsewhere in Italy, much more progress has
San Giuliano di Puglia (population 1,195), the been made. The civil protection departments of
infants’ school collapsed onto a class of small several regional governments have introduced
children (Augenti et al. 2004). Twenty-six small comprehensive Scuola sicura (safe schools) pro-
children and three teachers were crushed to death, grams, notably in the northern regions of Lom-
and 35 children were rescued and lived. The bardy, Piedmont, and Emilia-Romagna and in the
building had been constructed with regional autonomous Region of Sicily. The programs
development funds in the early 1960s and had involve a combination of structural measures
had its roof renewed a year before the earthquake. and school disaster management efforts, such as
Evidently, the roof, of reinforced concrete with a evacuation drills and lessons in civil protection.
ring beam, was too rigid and too heavy for the In many of the major cities, fire brigades and
underlying structure, a concrete frame building volunteer civil protection services are heavily
with hollow-brick infill. There were signs that the involved in the programs, with public-private
quality of the cement was poor and the partnerships supporting attractive safety litera-
reinforcing steel was not used as it should have ture for school children.
been. Moreover, despite mounting evidence that Despite these developments, as in other seis-
the Molisan Apennines are significantly affected mic countries so in Italy, the building stock of
by periodic earthquake activity, the revision of schools continues to age and the civil protection
local building codes to take account of the new educators must fight against the indifference of
data on seismic risk cannot be applied as easily to teachers, principals, and administrators. In many
existing structures as it can to ones that are about respects, mass mortality in Italian schools during
to be built (Augenti et al. 2004, p. S258). recent seismic events has been avoided mostly by
The school at San Giuliano di Puglia the lack of major earthquakes during school hours,
succumbed because it contravened simple, well- a situation that will not prevail forever. Given an
known laws of dynamic response in structures overwhelming need to upgrade the seismic perfor-
affected by seismic acceleration. Inertial forces mance of schools in Italy, the response of the
applied to a heavy roof sitting upon a weak frame national government has been to rank the build-
structure amount to a recipe for tragedy. ings in terms of the deficit between design require-
Consider the schools of the Lunigiana, a sur- ments (a function of the rules that prevailed at the
prisingly remote mountainous area of northwest- time they were built) and the latest assessment of
ern Tuscany. The Lunigiana has a sparse and peak ground acceleration (PGA). Priority funding
dispersed population. Children attend elementary is given to those schools that have the greatest
schools in the villages and secondary schools in “PGA deficit” (Grant et al. 2007). However, the
major population centers, which they reach by problem of unsafe school stock is simply too
bus or car. Many of the school buildings were expensive to solve in the short to medium term.
constructed in the 1950s and 1960s to cater for
the postwar population boom, and, in a rural area
of relative economic stagnation, they have nei- Japan
ther been built to be fully anti-seismic nor
retrofitted. Indeed, in the minor population cen- In general, Japan is understood to be a leader in
ters, they are decidedly dilapidated, nor do their evaluating seismic risk and in implementing
staffs seem to have much interest in repeatedly building codes for seismic-resilient construction.
Two publications available in English are based on a scenario earthquake similar to the
MEXT’s school seismic retrofit handbook 1934 Bihar-Nepal earthquake, suggests that
(MEXT 2008a) and school nonstructural refer- more than six million children and 140,000
ence book (MEXT 2008b). In addition to high teachers are at risk in schools (Bothara
seismic performance standards for schools, fol- et al. 2002). A survey of 900 public schools in
lowing the 1995 Kobe earthquake, Japan also greater Kathmandu Valley estimated that more
began providing guidance for mitigation of haz- than 75 % of school buildings would suffer severe
ards due to building nonstructural elements and damage beyond repair (estimated at US$7 mil-
contents in schools. Nonetheless, the East Japan lion), and other 25 % would suffer repairable
earthquake and tsunami [of 11 March 2011, with damage. In the absence of intervention, an esti-
magnitude 9.0 earthquake off coast of East mated 29,000 children could be killed in their
Japan], destroyed 6,284 in the affected region schools. With intervention 24,000 of these could
with different damage levels. Most of these be saved and buildings protected (NSET 2000).
were affected by tsunami waves, rather than A more recent assessment of school
earthquake. This was due to the location of the buildings in other parts of the country shows
school buildings [proximity to coastal areas], that more than 9,000 school buildings, more
the layout and structure of the buildings, and the than 10 % of the total in Nepal, would suffer
subsidence of local tsunami retention walls. At partial to complete collapse, resulting in very
the immediate aftermath, some schools [with high casualties.
higher stories] were used as temporary evacua- Most Nepalese school buildings are commu-
tion sites, and later people were rescued by heli- nity built, by local craftsmen who have no formal
copters. In some schools, located in higher training and are often illiterate. Technically
ground, people took shelter, which lasted till trained people are not part of this process, unless
6 months in some cases, which caused serious it is funded by the government. Construction is
disruption of school education. characterized by the high degree of informality.
The disaster pointed out several dimensions of The local availability of the construction mate-
role of schools and disaster education: (1) Schools rials, such as fired or unfired bricks, stone in mud
can play an important public infrastructure of the mortar, and timber, controls the construction pro-
community; however, the structural safety of the cess. The use of modern materials such as
building needs to be linked to operational cement, concrete, and steel bars is limited by
[including supplies of emergency kit] and loca- affordability and accessibility and is confined to
tional issues. (2) School-community linkage is an urban areas and areas accessible by transport.
important element, and local communities played In Nepal, there is no mandatory policy to con-
an important role along with school teachers for trol school design and construction. While some
the management of the evacuation sites. schools are supported by international donor S
(3) While it is unavoidable to use the school as agencies and/or the government requires design/
shelter, the continuity of education in emergency drawing details, many are directly constructed by
is a crucial issue. (4) School-based community the communities without standard design criteria
recovery emerged as an effective concept, where or technical supervision. Likewise, site-specific
the reconstruction of school building was linked hazards are also not considered during the
to enhanced community cohesion (Shaw and design and construction. Some design details are
Takeuchi 2012). available, but they may not be entirely suitable
for specific sites/locations. At most local
levels, people lack the capacity to understand
Nepal and implement the earthquake-safe construction
method.
A risk estimate for Kathmandu Valley, the eco- Low budgets for most school construction and
nomic, political, and technological hub of Nepal, lack of awareness and knowledge on the part of
graduate engineers of traditional and informal Turkey

construction methodology result in most school
buildings lacking earthquake resilience. Turkey has more than eight million children
The National Society for Earthquake Technol- attending schools in 64 provinces in the first-
ogy in Nepal (NSET), a national NGO, conducted and second-degree seismic risk zones. The 1999
a program to strengthen existing school buildings Kocaeli (moment magnitude Mw 7.4) and Duzce
and promote structural as well as nonstructural (Mw 7.2) earthquakes with approximately 20,000
components of the school buildings for seismic fatalities raised awareness of the school safety
safety, leveraging the decentralized, traditional, question, and the 2002 (Mw 6.0) Afyon-
and informal approach to construction (Bothara Sultandagi and 2003 Bingöl (Mw 6.4) earth-
et al. 2004). This program involved craftsman quakes kept awareness high.
training, technology development and transfer, During the 1999 earthquake in Kocaeli,
and community awareness raising. Many local 43 schools were damaged beyond repair, and
masons became master masons. On-site master 381 sustained minor-to-moderate damage.
masons worked in residence, supported by School was suspended for 4 months causing
visiting engineers with far-reaching effects. major disruption to the lives of families and chil-
Shake-table demonstrations of typical versus dren. In Istanbul, 60 km away, there was damage
seismic-resilient construction impressed commu- at 820 of the 1,651 schools. Damage at 131 of
nities with the effectiveness and feasibility of these sites necessitated at least temporary school
seismic-resistant measures. By raising awareness closure. Thirteen were immediately demolished,
in schools, the entire community is reached and another 22 were later slated for demolition
because lessons trickle down to parents, relatives, when retrofitting proved too costly. Fifty-nine
and friends. schools were strengthened and 59 repaired.
The approach developed took into account In the Bingöl earthquake of 2003, out of
sociocultural and economic issues, with outreach 29 schools in the affected area, 4 school buildings
to all stakeholders – school staff, students, local collapsed completely, 10 were heavily damaged,
community, local clubs, and the local and central 12 slightly or moderately damaged, and
government. They have all been involved in the 3 undamaged.
process so that they become aware of the risk and Public schools in the Kocaeli earthquake
support the solution. School building construc- fared better than residential buildings and pri-
tion was taken as an opportunity to train masons vate schools. Had children been at school dur-
and to transfer simple but effective technology to ing the Kocaeli earthquake, far fewer would
others in the community, including house have lost their lives. The fatality rate in resi-
owners. dential buildings in the Kocaeli earthquake was
Following this approach, NSET retrofitted 1.5 % in heavily damaged buildings and
more than 40 schools, mostly of unreinforced 16.5 % in totally collapsed buildings (Petal
masonry buildings. The program was found suc- 2009). Similar damage in higher occupancy
cessful in transferring technology to local crafts- buildings of the same type would cause higher
men who were quite keen to learn about the fatality rates. In the single example of the
complete process and to adopt the technology. school dormitory in the Bingöl earthquake
These masons became the propagators of the where 84 children died, the fatality rate was
safety message in the vicinity of these schools 44 %. Average risks are theoretical and don’t
and the replication of earthquake-resilient con- occur. Instead, the reality is that either the
struction. The long-term sustainability of these school is not occupied and no one dies or it is
impacts has yet to be assessed, but NSET’s expe- occupied and the fatality rates are high, and the
rience shows that seismic retrofitting and tragedy is wholly unacceptable.
earthquake-resistant new construction are both There is much that is right with school con-
affordable and technically viable. struction in Turkey. As a result of an assigned
importance factor of 1.5, public schools are • Implementation of a comprehensive

designed to withstand a 50 % increase in earth- retrofitting and replacement for seismic risk
quake design loads. Schools have regular mitigation
symmetrical structural designs, and those that • Follow-through on an ambitious program of
are only one or two stories have fared well, for school expansion and construction initiated to
the most part meeting standards for life safety, if respond to the acute shortage of class space
not continuous occupancy. The lethality of occasioned by three additional years of com-
school buildings is almost entirely attributable pulsory education enacted in 1998
to shoddy construction and is particularly lethal
in taller buildings that may also have design After the 1999 earthquake, the responsibility
defects. for school construction was shifted to the Minis-
For decades all public construction was under try of Education’s Division of Investments and
the authority of the Ministry of Public Works and Facilities (DIF). In turn, DIF appointed consul-
Settlement. Earthquake building codes on the tants from the private sector to oversee the new
books since the 1930s were updated most facility design and construction. DIF also devel-
recently in 1976 and 1998, yet the existence of oped standard designs for the new facilities, and
these laws has not guaranteed the safety of con- new school construction was financed by a com-
struction. The reasons are numerous. bination of government funds and charitable con-
Beyond an undergraduate or graduate degree, tributions raised by not-for-profit foundations.
there have been no independent or nonacademic New construction and procurement laws also
professional qualifications, proficiency stan- went into effect; however, the cumulative impact
dards, continuing education requirements, or of these changes and pressures is not yet known
licensure for architects or engineers nor any qual- (G€ulkan 2004).
ifications for building contractors. There are also In Istanbul of all projects the highest priority is
no guidelines for reliable and systematic building given to regional boarding schools, then to
inspection during construction. Penalties for schools in the 12 highest-risk districts, and to
noncompliance with building codes are beset those in proximity to the Marmara seacoast. The
with bureaucratic and social impediments and overall mitigation and retrofit effort targets more
often are simply not applied. Legal liability in than 1,800 buildings that constitutes the 80 % of
some future event with low-frequency occur- stock predating the 1998 Building Code. This
rence can hardly be a deterrent with so many to ambitious program is budgeted for US$320 mil-
share blame. Public construction has also suf- lion (Y€uz€ug€ull€u et al. 2004).
fered from a standard (though not legally An additional problem in Turkey is that
required) preference for the lowest bid in public awareness of nonstructural hazards remains low.
tenders. The civil service employment system Classroom doors often open inward and shelving S
also lacks proficiency standards and qualifica- and laboratory equipment remains unfastened.
tions for professional staff; so at the local level However, concern that children advised to
there is a wide variety in the capacity for project “drop, cover, and hold” might be injured by
supervision and control. Wage and salary levels flimsy wooden desks led to production and dis-
are low, and there has been opportunity for both tribution of 80,000 steel desks to more than
favoritism and corruption. There are no ombuds- 500 schools in the most vulnerable areas.
man or advocacy services to support consumer Schools are a well-distributed means of public
whistle-blowers. education, and children can play a leading role in
Istanbul provides a dramatic example of three the dissemination of public safety messages.
overlapping tasks: Thus, Professor Isikara, former head of KOERI,
a major earthquake research institute in Istanbul,
• Immediate response to damages caused by the toured the country visiting schools, becoming
1999 earthquake known as “Grandpa Quake,” and produced the
first children’s books and popular educational Management Course and more than 50,000
and rap music cartoons for earthquake awareness. passed the final test. Almost 50,000 users have
Both the Istanbul Governor’s Office and completed the Individual and Household Disaster
KOERI’s newly established Istanbul Community Preparedness Course and more than 40,000 users
Impact Project (ICIP) produced handouts distrib- have passed the final test. There were a total of
uted to all school children. Schools also received 114,700 course completions, 92,800 final tests
books and CDs. At the national level an introduc- passed, and more than one million lessons were
tion to natural hazards was integrated into the successfully completed by users. Within the first
primary school curriculum in Environmental year, active users completed an average of almost
Studies in 2002. Annual school-wide earthquake 15 lessons each (Petal and T€urkmen 2012).
drills and preparedness and remembrance activi- In 2005, a loan to implement the Istanbul
ties were initiated on 11 November 2001, to coin- Seismic Risk Mitigation and Emergency Pre-
cide with the Duzce earthquake anniversary. paredness (ISMEP) was funded by the World
Between 2001 and 2003, a cascading model of Bank and European Investment Bank, to help
training and instruction called “ABCD Basic prepare Istanbul for a probable earthquake in
Disaster Awareness” was implemented by the Marmara Region. The project had broad
KOERI’s Istanbul Community Impact Project. aims to enhance the institutional and technical
A curriculum was developed to address specific capacity of emergency management-related insti-
assessment and planning activities, physical risk tutions, raise public awareness, assess priority
reduction, and response preparedness measures public buildings for retrofit or reconstruction,
to be taken prior to a disaster. This was a signif- and support building code enforcement (www.
icant reorientation from previous “awareness” ipkb.gov.tr). “Component B” addressed seismic
programs that began with what to do “during the risk mitigation for priority public buildings. This
shaking.” A single full day of instructor training included a feasibility study for retrofit of 1,128
for 3,600 teachers was provided in collaboration education buildings on 796 school sites. Of these,
with the Ministry of Education Provincial Direc- 506 were strengthened and 148 were
torates and outside donors. These teachers in turn reconstructed. At the time of completion of the
communicated with 121,000 school personnel project, further 34 were slated for retrofitting and
and through them with 1.68 million school chil- 27 for reconstruction.
dren as well as with 700,000 parents. The project
established an Internet-based monitoring system
to monitor dissemination. Based on the success of USA: California
this project, almost a decade later an even more
ambitious scale-up was attempted. In 2008, seismic safety advocates in California
In 2010 the Ministry of National Education is launched the first Great Southern California
committed to taking the program nationwide with ShakeOut. The now annual event has grown to
technical support from Risk RED, in a follow-up include five western states and several central US
project supported by the American Red Cross. states and has been conducted in cities in five
A distance-learning self-study curriculum was other countries. In its first year four million chil-
developed consisting of 1 course (10 lessons) in dren and adults participated through 207 school
household disaster preparedness and 1 course districts plus almost 750 individual schools (Risk
(9 lessons) in school disaster management, with RED 2009). In 2013 there were more than 9.6
the goal of reaching 25,000 school-based instruc- million participants in the Great California
tors (Petal and T€ urkmen 2012). During the first ShakeOut drill. Globally there were almost
year of deployment of the courseware in 2012, 25 million registered participants in similarly
more than 79,000 MoNE employees completed inspired Great ShakeOut Drills. Of these almost
one or more lessons. More than 65,600 users 75 % were school-based participants (Earthquake
completed the School Disaster and Emergency Country Alliance 2013). This regular public
awareness event has heightened interest and con- with the Field Act is incremental and minimal.
cern in safe school facilities. With 3.6 million (2) Timeliness, consistency, accuracy, and com-
children enrolled in 262 public school districts munication are being improved by the Division of
in seven counties in Southern California, a major the State Architect. (3) The exemplary perfor-
earthquake in the region could cause an unprece- mance of school buildings is directly attributable
dented catastrophe for schools, children, and to the stringent seismic design provisions, plan
teachers. review, field inspection, and testing required by
School seismic safety has been a policy and a the act and which go beyond the standard build-
community concern in California since the 1933 ing codes. (4) All public schools should be cov-
Long Beach earthquake, and school emergency ered. A 2009 study reported that in the four major
planning has been required statewide since 1984. earthquakes since the Field Act, there have been
With 75 years of public policy leadership to sup- no public school collapses. The construction to
port school safety, new school construction stan- Uniform Building Code, the special enforcement
dards are higher than those for regular buildings and quality control provisions, an oversight by
and come close to assuring life safety. An advice the Office of the State Architect, and the 2003
regarding non‐structural mitigation measures publication guiding mitigation of nonstructural
(fastening furnishings, etc.) has been in place hazards are all judged to be successful. Nonethe-
for 20 years and requirements for such mitigation less, the work is not yet complete.
have been in place for 10 years. In 1999 Assembly Bill 300 required desk
The 1933 Field Act implemented immediately assessment of 9,659 pre-1978 school buildings.
after the Long Beach earthquake that year The final report based on woefully incomplete
required that schools be built to 15 % higher records was released in 2002 and found 7,537
performance standards than normal construction. potentially vulnerable buildings requiring
In 1938 the Garrison Act required examination detailed seismic evaluation. The cost of retrofit
and improvements to pre-1933 construction, but was estimated at $4.5 billion (State of California
went unenforced until 1968. The Uniform Build- 2002). Due to fear of planning and financial
ing Code enacted in 1976 is now the current implications, details were not released to the
standard for safe school construction. The Field public. In 2005 an investigative reporting series
Act has been hailed as a high point in school by California Watch finally achieved this (Risk
seismic safety and California schools are consid- RED 2009). In 2011, 20,000 uncertified projects
ered the safest in the United States. Some school were released on an interactive map (California
facility managers feel that its requirements are Watch 2011).
too stringent and too costly, and many seismic A review of school seismic safety in Califor-
safety advocates feel that it does not go far nia identifies four remaining areas of concern:
enough. The Act requires that structural plans S
be prepared by licensed structural engineers and • There are still some 7,537 school buildings in
approved by an independent state agency (the California constructed before 1978 that are of
Division of the State Architect (DSA)). Schools questionable safety.
have continuous on‐site inspection (rather than • Portable classrooms, which may account for
periodic), by a DSA-approved project inspector. one-third of all classrooms in California, may
Project architect and engineers must perform be particularly hazardous if not properly
construction observation and administration, supported and fastened.
and a final verified report must be filed by the • Private schools are not currently required to
project architect, engineers, inspectors, testing meet these same construction standards as
labs, and the contractor (State of California, public schools.
DSA 2007, 2009). • Nonstructural mitigation measures continue to
In 2007 the California Seismic Safety Com- require consistent application to protect chil-
mission found that: (1) The cost of compliance dren and adults from both injury and death.
• Each school district and private school is Bendimerad F (2004) Earthquake vulnerability of school
strongly recommended to conduct its due dil- buildings in Algeria. In: Group meeting on earthquake
safety of schools. OECD, Paris, pp 35–44
igence and report on these issues transparently Bhatia S (2008) Earthquake-resistant school buildings in
to parents, staff, and students, so that collec- India. Urban Plan Des 161(4):147–149
tive action can be taken to address these seri- Bothara JK, Guragain R, Dixit A (2002) Protection of
ous vulnerabilities. Neither fear nor educational buildings against earthquake. National
Society for Earthquake Technology, Nepal
California’s persistent financial crisis in the Bothara JK, Pandey B, Guragain R (2004) Seismic
education sector should be acceptable excuses retrofitting of low strength unreinforced masonry
for inaction (Risk RED 2009). non-engineered school buildings. Bull N Z Soc Earthq
Eng 37(1):13–22
California Watch (2011) K-12/public safety/spotlight/on
Summary shaky ground. 22 Dec 2011. Accessed 3 Feb 2012
Earthquake Country Alliance (2013) Great shake out
earthquake drills. http://www.shakeout.org. Accessed
Most of these case studies have focused on the 30 March 2015
primary importance of safe school facilities, Grant DN, Bommer JJ, Pinho R, Calvi GM, Goretti A,
through both standards for new school construc- Meroni F (2007) A prioritization scheme for seismic
intervention in school buildings in Italy. Earthq Spec-
tion and strategies for school vulnerability assess-
tra 23(2):291–314
ment and planning for retrofit and replacement. G€ulkan P (2004) Seismic safety of school buildings in
While sound earthquake engineering expertise is Turkey: obstacles impeding the achievable? In:
fundamental to advocacy, communications, plan- OECD (ed) Keeping schools safe in earthquakes.
Organization for Economic Cooperation and Develop-
ning, and execution of these efforts, it is also
ment, Paris, pp 64–87
important to retain the perspective of the primary International Recovery Platform (2010) Case reports –
beneficiaries: children, teachers, and school com- China. United Nations International Strategy for
munities. This necessitates going beyond the Disaster Reduction. http://www.recoveryplatform.
org/outfile.php?id=548&href=http://www.recovery-
obvious: site selection, design, and construction.
platform.org/assets/recoverycases_reports/china/
As some of these case studies indicate, it is also CHINA_300910.pdf. Accessed 16 Feb 2014
important to consider nonstructural mitigation, MEXT (2008a) School seismic retrofit handbook
ongoing maintenance, safe access to school, http://www.nier.go.jp/shisetsu/pdf/e-taishinjirei.pdf.
Accessed 30 March 2015
school function as temporary emergency shelters,
MEXT (2008b) School non-structural reference book
and even structural awareness education and the http://www.nier.go.jp/shisetsu/pdf/e-jirei.pdf. Accessed
use of construction as an educational opportunity 30 March 2015
for children and communities. By taking this NSET (2000) Seismic vulnerability of the school build-
ings of Kathmandu valley and methods for reducing
wider (and multi-hazard) view, the focus on safe
it. Kathmandu Valley Earthquake Risk Management
school facilities overlaps with both ongoing Project implemented by National Society for Earth-
school disaster management and with risk reduc- quake Technology-Nepal, Kathmandu
tion education. Petal M (2009) Evidence-based public education for
disaster prevention: the causes of deaths and injuries
in the 1999 Kocaeli earthquake, VDM. Doctoral dis-
sertation, UCLA, 2004
Cross-References Petal M, T€ urkmen Z (2012) Dreams for Turkey: a case
study of scale and reach of distance learning self-study
for individual and household preparedness and school
▶ School Seismic Safety and Risk Mitigation disaster management. http://www.riskred.org/activi-
ties/dreamsforturkey.pdf. Accessed 30 March 2015
Risk RED (2009) School disaster readiness: lessons
References from the First Great Southern California ShakeOut.
http://www.riskred.org/schools/shakeout2009.pdf.
Augenti N, Cosenza E, Dolce M, Manfredi G, Masi A, Accessed 31 March 2015
Samela L (2004) Performance of school buildings Secretarı́a de Educación del Distrito Capital de Santafé de
during the 2002 Molise, Italy, earthquake. Earthq Bogota (2000) Análisis de vulnerabilidad sı́smica de
Spectra 20(S1):S257–S270 las edificaciones de la Secretarı́a de Educación del
Secondary Structures and Attachments Under Seismic Actions: Modeling and Analysis 2483
Distrito y diseños de rehabilitación de algunas de ellas. (2) foundation, (3) structural part of the construc-
Proyectos y Diseños, P&D, Ltda, Bogotá tion to be intended as the part in elevation, and
SEEDS of India (2006) Earthquake safety in mountain
regions. School Earthquake Safety Initiative, Shimla/ (4) nonstructural part to be intended as a part of
Delhi the construction with minor or no task to the
Shaw R, Takeuchi Y (2012) East Japan earthquake and structural capacity.
tsunami: evacuation, communication, education and From a conceptual point of view, it could be
voluntarism, Research Publisher
State of California Department of General Services easy to assert the following principle:
(2002) Seismic safety inventory of public schools the seismic evaluation of a construction has to be
(AB300 final report) performed based on 1) the definition of Structural
State of California, Seismic Safety Commission Resisting System (SRS) 2) a proper model imple-
(2007) The field act and public school construction: mentation of the SRS 3) a proper analysis of
a 2007 perspective. Accessed 30 March 2015 the SRS.
State of California, Seismic Safety Commission (2009)
The field act and its relative effectiveness in reducing Given that the previous principle could
earthquake damage in California’s public schools. resolve the problem, the SRS verification remains
Available online at http://www.seismic.ca.gov/
pub/CSSC%2009-02%20Field%20Act%20Report.pdf. one of the goals of the analysis process: it can be
Accessed 16 Feb 2014 pursued with a probabilistic approach (ATC
Taylor G (1989) Seismic assessment of Vancouver school 58 2012) (1) defining the required performance
buildings. TBG (Transit Bridge Group) Engineering (2) based on the predictable loss (direct and indi-
Firm, Vancouver
Y€uz€ug€ull€u Ö, Barbarosoglu G, Erdik M (2004) Seismic rect) consequent to a given seismic event.
risk mitigation practices of school buildings in Istan- A general approach, in which the strategy can
bul. In: OECD (ed) Keeping schools safe in earth- be framed, is the performance-based approach
quakes. Organization for Economic Cooperation and that (1) defines a given number of performance
Development, Paris, pp 176–183
levels (PLs), (2) chooses a seismic level for each
PL, and (3) requires a given performance for each
PL. Examples of PLs are operational, immediate
occupancy, life safety, and collapse prevention.
Secondary Structures and The performance-based approach delegates
Attachments Under Seismic Actions: the applicator of it the definition or selection of
Modeling and Analysis the most appropriate tools to be applied for the
(1) identification of the resisting system (RS),
Nicola Nisticò and Alessandro Proia (2) structural analysis (modeling included),
Dipartimento di Ingegneria Strutturale e (3) capacity definition, and (4) verification.
Geotecnica, Università La Sapienza, Rome, Italy The first step (RS identification) is not an easy
task: the RS includes the soil and the construction
that on the other hand can be split in structural S
Synonyms and nonstructural elements which are not sup-
posed to have any role in the global resistance
Nonstructural elements; Secondary of the construction with regard to neither of the
Nonstructural Elements; Secondary Structural so-called vertical loads or of the horizontal loads
Elements. such as those that schematize wind and seismic
actions.
The nonstructural elements, having no role
Introduction in the seismic capacity, are generally considered
as additional weight to be included in the
Seismic evaluation of a construction has to mass evaluation, neglecting the structural inter-
include, with a given level of accuracy, the action between them and the structural resisting
interaction phenomena between (1) soil, system.
2484 Secondary Structures and Attachments Under Seismic Actions: Modeling and Analysis
For many of the nonstructural elements, the base of the structure, including in it the effect of
absence of interaction can be considered realistic the propagation of the seismic action in the soil.
so that they can be considered as attached ele- Similarly, a spectra (FRS) is defined at the base of
ments (from which the name attachment derives) the attachment, including in it, with a cascade
that having a proper structure (mass, stiffness, procedure, the effect of the propagation of the
structural capacity) have to be verified with action at the soil (at first) and, subsequently, at
regard to the seismic action transferred to them the elevation structure.
(from the resisting system). The definition of a Response Spectra at the
The attachments (as previously introduced) base of the attachment solves the problem since
are objects with their morphology so that they the attachment can be analyzed with traditional
can be schematized with either continuous or methodologies that are, for example, Seismic
discrete models opportunely connected to the Modal Analyses or Time Histories Analyses
structure they are attached to. Examples of based on acceleration histories compatible with
attachments are (1) furnitures, (2) technical sys- the given Floor Response Spectrum.
tems, and (3) art objects of a museum. Usually the effect of the propagation of the
If the structure-attachment interaction is neg- seismic event (from the soil to the attachment) is
ligible as well as the soil-structure interaction, a performed considering a linear behavior of the
cascade procedure can be adopted evaluating the primary structure: this is supposed a realistic
seismic action (in terms of time histories or assumption for new conceived structure when
response spectrum) (1) at the base of the structure operational and immediate occupancy PLs are
in elevation and then (2) transferring it at the considered.
points to which the nonstructural element is The linear structural behavior of the princi-
attached. pal system could be considered as nonrealistic
The cascade procedure, not considering the in some cases, where the system’s nonlinearity
primary (SRS) and secondary (attachments) sys- could produce effects (on the attachment) more
tems as a whole entity (PS system), cannot be severe than those evaluable with a linear
adopted when the two systems are tuned; that behavior assumption (Chaudhri and Villaverde
means their periods are similar and the attach- 2008).
ment could be acting as a tuned mass damper The attachments, as discussed so far, are sec-
(TMD) for the primary system. ondary elements that do not give any contribution
The seismic analysis of the attachments can be to seismic resistance of the primary system and,
performed by means of different strategies in these terms, can be classified as secondary
(Villaverde 1997; Chen and Soong 1988) among nonstructural elements (NSEs) to be distin-
which linear and nonlinear analyses can be guished from the secondary structural elements
included: the PS system can be analyzed as a (SEs) that have a negligible role in supporting the
global system with an evident computational seismic action but can have a specific role in
effort. transferring the vertical load to the foundation
The need of efficient and accurate methods to system.
analyze the PS systems inspired methodologies Secondary elements, either attachment
(Igusa and Der Kiureghian 1985a, b) based on (secondary NSEs) or structural elements (SEs),
(1) modal synthesis, (2) perturbation theory, and are both systems having their structure, oppor-
(3) random vibrations. tunely linked to the primary structure: they have
The decoupling of the secondary system (from to be adequately modeled in order to be analyzed
the P system) allows the evaluation of the seismic with the strategies common to the seismic branch,
action in terms of Floor Response Spectra whose such as (1) static analyses (linear and nonlinear),
approach is similar to the approach that governs (2) modal response spectrum analyses, and
the decoupling between soil and structures: a (3) time domain analyses either linear or
Response Spectra is defined and applied at the nonlinear.
Morphological and Phenomenological cost of (1) structural elements, (2) secondary struc-
Aspects Versus Modeling and Analysis tural elements, and (3) nonstructural elements
such as the contents are. The cost of the
Depending on the structural resisting system nonstructural elements can be estimated to reach
typology and the construction usage, the NSE the 70 % of total construction costs if the hotel
typology can be wide (see Fig. 1, for the building buildings are concerned, while it is lower in office
case) and their taxonomy can be found in Taghavi buildings (62 %) and hospitals (48 %) where con-
and Miranda (2003) where a comprehensive data- tents (such as medical equipment) can be esti-
base of nonstructural components is presented mated to be 44 % of the total cost.
covering different aspects such as, among others, Economic loss due to seismic nonstructural
cost information from which is deducible that the damage can be relevant: during the 1994
structural cost of a building could be not relevant Northridge earthquake, the nonstructural damage
with respect to the global cost: the office buildings was (Kircher 2003) about 50 % of the global
structural costs, even if relevant, are only 18 % of building damage which was estimated to be
the total cost of construction that can be split in $18.5 billion.
Secondary Structures and Attachments Under Seismic Actions: Modeling and Analysis, Fig. 1 Typology of
building nonstructural elements. Reproduced from FEMA 74 (2005)
Impacts
other
equipment Connections
fail
Tall furniture Intermediate Squat furniture Equipment slides Support legs fail
will topple furniture will will slide
over slide and topple
Steel stops
Equipment
topples over
Bolts fail
EQUIPMENT ANCHOR BOLT Equipment jumps Force fails

TOPPLES OVER FAILURE over top of stops steel stops
Secondary Structures and Attachments Under Seismic Actions: Modeling and Analysis, Fig. 2 Principal rigid
body mechanism
Most of the NSEs have limited seismic perfor- reduce the seismic action on both the machinery
mance because they are not properly attached to and building.
the primary structures, so that, depending on their Referring (see Fig. 3) to the previously intro-
slenderness, they can (see Fig. 2) (a) topple over, duced example (where the equipment can be con-
(b) slide and topple, and (c) slide. The loss of sidered as an attachment of the primary
capacity of the NSEs or their connections can structure), the following indications can be
cause damage to other equipment (see Fig. 2) given for the seismic analysis:
and injury to people, so careful attention has to
be paid to the design of the connection (see 1. If the NSE is rigidly connected to the structure
Fig. 2). and its mass (MNS) is not negligible with
Unlike the old conceived NSEs, the respect to the building mass (MS), a global
new-generation elements can have a good seis- analysis of the PS system is required. In case
mic performance, thanks to the wide range of of modal spectra time history (TH) analysis,
connections that can be adopted to link the NSE some approximation in the damping definition
to the structure. Depending on the case at hand, is needed due to the different damping of the
the design can include (1) seismic joints oppor- attachment (steel structure) with respect to the
tunely designed to accommodate seismic dis- reinforced concrete structure. More appropri-
placements, (2) seismic isolators to reduce the ate step-by-step TH analyses can consider the
acceleration level, and (3) dissipative device to element damping, properly modeling it by
reduce the level of acceleration, velocity, and means of dashpots when nonclassical
displacement. damping is present.
So the new-generation NSEs cannot be con- 2. If MNS is negligible with respect to MS, but the
ceived without an adequate strategy for the con- NSE period (TNS) is close to structural period
nection design, an example of which is reported (TS), the so-called tuning happens and a global
in Fig. 3 where the case of a machinery (for analysis of the PS system is required, not
cement production) mounted on a steel structure excluding positive effect.
attached on a reinforced concrete building has 3. If MNS is negligible with respect to MS, and
been reported: dissipative devices have been TNS is not close to TS, the decoupling could be
introduced at the base (of the steel structure) to considered and the strategy for the analyses
a b
Equipment
Device
Industrial Building
Secondary Structures and Attachments Under Seismic Actions: Modeling and Analysis, Fig. 3 (a) Cement
industrial building. Machinery mounted upon an rc structure. (b) Cement Industrial Building Applications: Italy
could be oriented to the definition of the seis- given a specific soil and structure characterized
mic action at the level of the connection, con- by their own periods and dampings.
sidering the structural system as a stand-alone Differently from the secondary nonstructural
element subjected to a seismic action at the elements, the secondary structural elements
base of it. (SE) require different strategies of modeling and
analysis. It is useful to introduce them as reported
The following strategy can be considered: in CEN (2008): some structural elements (i.e.,
(a) time history analyses applied either to the beams and columns) can be designed as seismic
whole system or to the stand-alone attach- secondary elements, neglecting their contribution
ment, considering the acceleration histories to the global seismic resistance so that their stiff-
recorded at the level of the attachment con- ness and strength can be neglected. As further
nection and (b) spectrum-based analyses defin- specified in CEN (2008), the SE and the joint
ing an acceleration spectrum consistent with (that link them with structure) have to be designed
the time histories recorded at the points where considering (1) the vertical gravitational load,
the attachment is connected to the structures: (2) displacement consequent to the seismic action,
the generated spectrum is usually named Floor and (3) second-order effects that include the flex-
Response Spectrum (FRS) even if (as the case ural moments evaluated considering the deformed
reported in Fig. 4) the FRS has been evaluated element shape (P-Delta effect). S
where the attachment is linked and not at the Clearly the previous definition of the SE sup-
floor level. poses that they have a negligible influence in the
The FRS definition is an important task for the global structural behavior.
NSE analysis, being the referred analysis tool to Starting from the classification of the second-
be adopted, due to its recognized simplicity in ary elements in NSEs (attachments) and SEs, the
conjunction with good level of reliability following can be asserted: (1) given a construc-
(in those cases where the decoupling can be tion, it includes a principal (P) and a secondary
adopted). It is possible to affirm that the usually (S) structure; (2) if the S structure has a negligible
adopted acceleration spectrum is to structural influence on the P structure, the whole construc-
analysis as the FRS is to the analysis of tion (PS) structural behavior can be decoupled;
nonstructural elements, so that the FRS is a (3) the S structure can be classified in structural
period-dependent function that can be evaluated (SEs) and nonstructural (NSEs) elements; (4) the
for different value of the attachment damping SEs have to be designed for vertical gravitational
2.5
2.0
1.5
1.0
0.5
0.0
0.0 0.5 1.0
0.2
0.1
0
–0.1
–0.2
0.2
0.1
0
–0.1
–0.2
Secondary Structures and Attachments Under Seismic Actions: Modeling and Analysis, Fig. 4 FRS generation
load (transferred from the PS to them, including Secondary Structural Elements

the self-weight loads) considering the seismic- Typical case of SE is internal and external build-
induced displacement (P-Delta effect included); ing partitioning system (Glass Systems
(5) the NSEs have to be verified with regard to included): their seismic contribution is usually
self-weight loads and seismic action transmitted neglected in the seismic analysis (1) accepting
by the P structure; (6) the secondary structure (for severe earthquake) their damage and
elements and their supports (links) have to be (2) imposing that out-of-plane collapse (see
verified in order to avoid that their partial or Fig. 5) has to be prevented.
total failure can induce injury to people or impor- The infilled partitioning system can have a
tant objects; and (7) if the interaction between role in transferring the vertical load even if they
P and S systems is not negligible, a global PS can have a minor contribution in seismic
analysis is required. global capacity. If they have no role in the verti-
In the following, some of the principal char- cal load as well as in the seismic P structure
acteristic regarding secondary structural and capacity, their classification as NSE
nonstructural elements will be described. (attachments) is reasonable.
Secondary Structures and

Attachments Under
Seismic Actions:
Modeling and Analysis,
Fig. 5 Molise (Italy):
Seismic event occurred in
2002 (October 31 (M = 5.4)
and November 1 (M = 5.3).
Example of damaged
infilled frame: in-plane and
out-of-plane mechanisms
If no flexible joints are considered (in between Modeling and Analysis

the P and S systems they represent), the absence
of collaboration with the P system is not judicable NSEs are elements characterized by their mass
by means of qualitative considerations, but it can and stiffness, and independently of the seismic
require a structural analysis of the PS system action they are subjected to, they can be modeled
including them as structural elements. In this and analyzed based on FEM strategies consider-
case, the designer can follow some suggestions ing either their linear or nonlinear behavior.
such as those included in CEN (2008) that con- In general the NSE is a system composed by
sider a structural system as SE if its global stiff- subsystems with a structural complexity (see Fig. 6)
ness is lower than 15 % of the P system stiffness. that can require 3D complex models to be calibrated
The models concerning the infilled frames sys- by means of experimental tests (Fig. 6a) including
tems are well known. identification strategies: dynamic tests can be car-
The damage level in the partitioning system is ried out by means of shacking tables (Fig. 6c). S
usually controlled imposing a threshold to the The experimental tests in support of modeling
interstory drift (see Table 1) as function of the and analysis implementation are especially
performance level and construction usage (ASCE required either when the importance of NSE
2002). usage is considered strategic or when the cost of
it justifies the experimental activity. In some
Attachments cases, a qualification procedure can be required,
Typical cases of attachments are parapets, win- generally ruled by international standard (Gilani
dows, partitioning systems, antennas, electrical et al. 1999; IEEE 2005).
power systems, and furnitures. Depending on
their components, they can be sensitive to the Modeling
seismic acceleration or deformation (see Modeling has to take into account all the compo-
Table 2). nents that give stiffness and strength
Secondary Structures and Attachments Under Seismic performance given a defined seismic action.
Actions: Modeling and Analysis, Table 1 Drift control: Experimental tests can include shaking table
usually adopted values as function of usage and performance
level tests or static tests: this aspect is strictly linked
to the qualification process (IEEE 2005).
Damage
Performance description and
level downtime/loss Drift control Seismic Action Modeling and Structural
Immediate Negligible 0.3 % (stiff Analyses
occupancy: structural joint), 0.6 % Seismic action can be simulated according to the
usually required damage; (deformable usually adopted strategies that, for the case at
for construction essential systems joints)
which usage is operational; hand, include (1) time histories (usually in terms
considered minor overall of acceleration) and (2) response spectrum final-
strategic damage. ized either to modal analyses or to static linear or
Downtime/loss: nonlinear pushover analyses.
24 h
Seismic level will depend on the referred per-
Life safety Probable 0.5 % (stiff
structural and joint) formance level that (see Table 1) identifies the
nonstructural 1.0 % required performance associable to a seismic
damage; no (deformable event with a given return period, to be defined
collapse; joints) based on cost-benefit analysis.
minimal falling
hazards; General rules valid for secondary elements are
adequate the following:
emergency
egress. 1. Mass and stiffness uncertainties have to be
Downtime/loss:
possible total considered together with spatial distribution
loss of seismic effect in case of extended SE
Collapse Several Not required systems.
prevention structural and 2. Seismic effects on SE have to take into
nonstructural
account, in general, both horizontal and verti-
damage;
incipient cal components to be evaluated based on a
collapse; structural model of the principal system.
probable falling 3. If the SE behavior can be decoupled from the
hazards; possible
principal system, the datum method for the
restricted access.
Downtime/loss: evaluation of the peak acceleration at the SE
probable total is based on the Floor Response Spectrum
loss (FRS) that given an SE element, with a defined
structural period and damping, attached to a
given part of a structure, having its mechanical
contributions, including the connection elements properties, subjected to a given seismic event
that, if needed, have to be modeled as nonlinear (E), allows to define the peak acceleration to
elements. which the element will be subjected when the
In many cases, such as the bushing sketched in seismic event (E) is transferred at the base of
Fig. 6a, an accurate modeling requires informa- NS element.
tions about all the subcomponents (coil springs,
valves) in terms of mass stiffness and strength. Based on the knowledge of the FRS, one of the
The needed informations are not usually known following methods can be adopted: (a) static
and the element investigation has to be supported equivalent forces (including nonlinear pushover
by means of experimental tests devoted either to analysis), (b) modal analysis, and (c) time history
global information acquisition (frequencies, (linear or nonlinear) analyses based on
modal shapes) or to evaluation of the level of accelerograms compatible with the FRS.
Secondary Structures and Attachments Under Seismic Actions: Modeling and Analysis, Table 2 NSE classifica-
tion (ATC/BSSC 1997) and element sensitivity with regard to acceleration and deformation
Sensitivity Sensitivity
Component A D Component A D
A. Architectural B. Mechanical equipment
1. Exterior skin 1. Mechanical equipment
Adhered veneer S P Boilers and furnaces P
Anchored veneer S P General mfg. and process machinery P
Glass blocks S P HVAC equipment, vibration isolated P
Prefabricated panels S P HVAC equipment. Nonvibration isolated P
Glazing systems S P HVAC equipment, mounted in-line with P
ductwork
2. Partitions 2. Storage vessels and water heaters
Heavy S P Structurally supported vessels (category 1) P
Light S P Flat bottom vessels (category 2) P
3. Interior veneers 3. Pressure piping P S
Stone, including marble S P 4. Fire suppression piping P S
Ceramic tile S P 5. Fluid piping, not fire suppression
4. Ceilings Hazardous materials P S
(a) Directly allied to structure P Nonhazardous materials P S
(b) Dropped, furred, gypsum P 6. Ductwork P S
board
(c) Suspended lath and plaster S P
(d) Suspended integrated ceiling S P
5. Parapets and appendages P
6. Canopies and marquees P
7. Chimneys and stacks P
8. Stairs P S
A acceleration sensitive; D deformation sensitivity; P primary response; S secondary response
Floor Response Spectra-Based Evaluation low values of n (e.g., minor than 7), an envelope
Floor Response Spectra are functions that define spectrum has to be considered.
the response spectrum of a given response param- Usually the location of the attachment is not
eter (e.g., acceleration, velocity, displacement) as known in advance, so that the previous procedure
a function of period and damping of a given can be applied considering p points obtaining S
structure (attachment) localized at a given point p Response Spectra. For those points that are
of the construction. located at the same level (floor) of the
The generally adopted technique for the FRS P structure, a single spectrum can be evaluated
definition consists in (1) analyzing the P structure (enveloping the Response Spectra), naming it
(to which the S structure is attached) in the time Floor Response Spectrum.
domain, considering n time histories (e.g., accel- If the goal is the evaluation of conservative
eration TH), (2) evaluating (for each TH), at a FRSs, for each floor, a set of points has to be
given point of the structure, the TH of the accel- opportunely selected so that both translational
eration and the related response spectrum for a and rotational effects are captured: they usually
given damping value, and (3) defining one repre- include the floor centroids and one or more cor-
sentative spectrum (based on the n available ners for each floor.
FRSs) having a given overcoming probability It is worth mentioning that having defined the
(usually a 50 % probability is considered): for P system structural model, it is possible to

Attachments Under
Seismic Actions:
Fig. 6 (a) Morphology of a
bushing. (b) Bushing
experimental test carried
out at UC Berkeley (CA).
(c) Cabinet experimental
test carried out at UC
Berkeley (CA)
evaluate a transferring function Hp that (1) know- Go ðwÞ ¼ Gi ðwÞ jHp ðwÞj2 (1)
ing the Fourier Transform (Fi, input FT) of a
given accelerogram (2) allows the definition of Knowing Go(W) and the transferring function
the Fourier Transform of the TH acceleration at (HSDOF) of a single-degree-of-freedom system
given point (Fo, output FT) so that (3) the inverse (SDOF), the SDOF Spectra Power Density is
Fourier Transform of Fo gives the TH at the evaluable (see Eq. 2) and the related Response
considered point that is the required information Spectrum is the required FRS.
for the Response Spectrum evaluation.
Alternatively, if the Power Spectrum Gi of a GSDOF ðwÞ ¼ Go ðwÞ jHSDOF ðwÞj2 (2)
given earthquake or of a family of earthquakes
is known as well as the previous defined The previously presented approaches are not usu-
Input Transferring Function (Fi), the output ally adopted for conventional structures as they
Power Spectrum (Go) is evaluable according to are time consuming, so that predictive expres-
Eq. 1. sions are proposed in literature or enforced in
international recommendations: given the peak of 2–3), (2) a simplified expression for the eval-
ground acceleration, the floor peak acceleration uation of modal displacement Fi of a given floor
is evaluated by multiplying the PGA by an ana- supposed to be equal to z/H where z is the level of
lytical function, named Sa in the following. the considered floor and H is the total construc-
The usually proposed functions (Sa) are based tion height, and (3) a value of G between 1 and
on (1) simplified expressions for the evaluation of 1.5.
the floor acceleration and (2) simplified shape Based on the previous assumptions, Eq. 4 can
functions representative of the required FRS. be simplified as follows, having assumed
It is worth mentioning that given a structure, RS = 3, G = 1:
knowing of it (1) the prevalent modal shape (F),
(2) the prevalent period (Ts), and (3) the modal FRSi ¼ PGA G ½1 þ 2 ðz=HÞ RFRS ðTNS Þ
participation factor (G) if a seismic action is (5)
considered, the absolute structural acceleration
(€
ui) associated to the single modal coordinate In the following sections, some literature expres-
(Fi) can be evaluated according to Eq. 3 where sions (CEN 2008; FEMA 369 2001; AFPS 2007;
RS(TS) is the value of the normalized acceleration KTA 2012) will be given, expressing them in
spectrum, for a given value of the structural terms of the normalized FRS (Sa) that corre-
damping (zS) at the prevalent period of the sponds to FRSi evaluated for PGA = 1.
structure. It has to be specified that in order to show the
trend of the Sa functions, they will be plotted,
u€i ¼ PGA G f1 þ ½RS ðTS Þ 1 Fi g (3) contextualizing it to the simple structure
described in the Case Study section.
Assuming a given analytical function (RFRS),
being it dependent on the period (TNS) and the CEN (2008)
damping (zNS) of the nonstructural element, the The following Eq. 6, plotted in Fig. 7, is pro-
required FRS, associable to the Fi modal coordi- posed, supposing zS = zNS = 5 %.
nate, is equal to
z
FRSi ¼ PGA G f1 þ ½RS ðTS Þ 1 Fi g 1þ
Sa ¼ 3 H
RFRS ðTNS Þ (4) 2 0:5 1 (6)
1 þ 1 TTNSS
Usually, Eq. 4 is simplified adopting (1) a con-
stant value for RS(Ts) evaluated at the plateau of Equation 6 can be evaluated assuming
the acceleration spectrum (assumed in the range TNS = 0, obtaining (1) the value of the expression
S
Secondary Structures and 8
Attachments Under
7
Seismic Actions:
Modeling and Analysis, 6
Fig. 7 Amplification
factor Sa for different 5
values of z/H 4
3
2
1
0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
z/H = 0.24 z/H = 0.43 z/H = 0.62 z/H = 0.81 z/H = 1.00

Attachments Under
Seismic Actions: 7
factor (Sa) shape versus 5
TNS/TN 4
3
2
1
0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
z/H = 0.24 z/H = 0.43 z/H = 0.62 z/H = 0.81 z/H = 1.00
adopted for the evaluation of the normalized floor Secondary Structures and Attachments Under Seismic
acceleration (Eq. 7a) and (2) the value of the Actions: Modeling and Analysis, Table 3 Values of
adopted expression for the evaluation of RFRS RFRS as function of TNS/TN
(Eq. 7b). So that Eq. 6 can be rearranged as TNS/TN RFRS
reported in Eq. 7c. TNS/TN < 0.5 and TNS/TN > 1.0
It is possible to recognize that (1) a value of 2.0
2.5 has been assumed for the evaluation of Rs and 0.7 TNS/TN < 1.4 2.5
0.5 TNS/TN < 0.7 (7.5 TNS/TN)
(2) the expression adopted for the evaluation of
2.75
RFRS is supposed to be dependent on the normal- 1.4 TNS/TN < 2.0 6 – (2.5 TNS/TN)
ized floor height.
u€i n zo z
¼ 1 þ ½RS ðTS Þ 1 ¼ 1 þ 1:5
PGA H H
(7a) It is possible to recognize that (1) a value of 3.0
has been assumed for the evaluation of Rs and
z (2) the expression adopted for the evaluation of
1þ 3 0:5
¼ H
RFRS z 2 z RFRS is supposed to be independent on the nor-
1 þ 1:5 1 þ 1:5
H 1 þ 1 TS
TNS
H malized floor height.
(7b)
AFPS (2007)
z The following expression is proposed:
Sa ¼ 1 þ 1:5
8 H 9
z rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
z ffi
>
< 1þ >
H 3 0:5 = Sa ¼ 1 þ GS RS
2 2
2a RFRS (9)
z 2 z 1 H
>
:1 þ 1:5 1 þ 1 TNS 1 þ 1:5 >;
H TS H
(7c) where (1) a is a parameter to be calibrated in
order to minimize the difference between the
FEMA 369 (2001) effective modal displacements (Fi) and the pro-
The following Eq. 8, plotted in Fig. 8, is proposed posed simplified expression (z/H), (2) Rs is the
supposing zS = zNS = 5 %, where the values of value of the normalized structural acceleration
RFRS are reported in Table 3: evaluated at the fundamental structural period
(TS) for a considered value of the structural
z
Sa ¼ 1 þ 2 RFRS (8) damping, (3) the participation factor (GS),
H evaluable according to Eq. 10a, assumes a
maximum value of 1.6, if a = 1.5 is imposed, and at the lower limit value in the variation range of
(4) RFRS values are reported in Table 3 as a the system parameters, however, not lower than
function of parameter A (see Eq. 10b) that takes the rightmost corner frequency of the highest
into account the damping (zNS) of the plateau of the associated response spectrum;
nonstructural element. (2) highest decisive eigenfrequency of the prin-
cipal system for the upper limit value in the
2a þ 1 variation range of the component parameters,
GS ¼ (10a)
aþ1 however, not lower than the rightmost corner
frequency of the highest plateau of the associated
35 response spectrum; and (3) upper limit frequency
A¼ (10b)
2 þ xNS of the associated response spectrum.
The maximum value of the amplification fac-
It is worth mentioning that Eq. 9 derives from tor is reported in Fig. 10 (right), where D1 and D2
Eq. 4, with the following assumptions: (1) a uni- are respectively the damping ratios
tary participation factor is considered for the (in percentage of critical damping) of structural
ground acceleration, while the principal mode and nonstructural elements whose suggested
participation factor is considered according to values are reported in Table 5.
Eq. 10a. (2) The spectral acceleration Rs(Ts) is In order to compare the obtainable FRS with
considered for the evaluation of the floor relative those previously discussed, the amplification fac-
acceleration instead of the spectral relative nor- tor (Sa) is plotted in Fig. 11, having assumed (1) a
malized acceleration (Rs(Ts) 1). (3) The ground 5 % damping for both structural and nonstructural
acceleration and the relative structural accelera-
tion are combined through the SRSS (Square
Root of the Sum of the Squares) combination Secondary Structures and Attachments Under Seismic
Actions: Modeling and Analysis, Table 4 Values of
rule (Table 4 and Fig. 9). RFRS as function of TNS/TN
TNS/TN RFRS
KTA (2012)
TNS/TN < 0.5 and 1.0
The proposed expression does not give any infor- TNS/TN > 2.0
mation to evaluate the floor acceleration (ag), but 2/3 TN/TN < 3/2 A
it only defines the amplification shape (RFRS)
0.5 TNS/TN < 1.5 T
log32 TNS
reported in Fig. 10 (left), where the f-axis is the A ð A 1Þ
S
ðlog34Þ
component frequency axis (in logarithmic scale)
2 < TNS/TN T
log23 TNS
and f1, fn, and flimit, respectively, are (1) lowest K T ¼ A ðA 1Þ
S
decisive eigenfrequency of the principal system ðlog43Þ
S
Attachments Under
Seismic Actions:
20
factor (Sa) shape versus 15
TNS/TN (a = 1.5, GS = 1.6,
RS = 2.5) 10
0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
z/H = 0.24 z/H = 0.43 z/H = 0.62 z/H = 0.81 z/H = 1.00
Secondary Structures and Attachments Under Seismic Actions: Modeling and Analysis, Fig. 10 Amplification
factor shape (left) and maximum amplification factor (right)
Secondary Structures and Attachments Under Seismic elements, (2) the FEMA expression (see Eq. 11)
Actions: Modeling and Analysis, Table 5 Suggested for the evaluation of the normalized floor accel-
damping values (in percent of critical damping). Column
A: to be adopted for verifying the load-carrying capacity
eration, (3) f1 = 6.66 Hz that is the rightmost
and integrity and for determining the spectra. Column B: corner frequency of the acceleration plateau of
in the case of mechanically active components for which CEN (2008) type 1 Spectrum (A soil), and
the functional capability is verified by a deformation (4) fn = 11.1 Hz that is the highest decisive
analysis
eigenfrequency of the principal system described
Damping in the Case Study section.
ratios
Components A B z
Pipes 4 2 ag ¼ 1 þ 2 (11)
H
Steel with welded connections and welded 4 2
components (e.g., vessels, valves, pumps,
Verification
motors, ventilators)a
Steel with SL or SLP bolt connections 7 4
(SL – structural bolt connection with a As stated in KTA (2012), the verification process
borehole tolerance 2 mm; SLP – fitted has to regard (1) the load-carrying capacity in
bolt connection with a borehole terms of strength, stability, and secure position-
tolerance 0.3 mm)
ing (e.g., their protection against falling over,
Steel with SLV(P) or GV(P) bolt 4 2
connections (SLV(P) – preloaded fitted
dropping down, impermissible slipping); (2) the
bolt connection; GV(P) – fitted friction- integrity, that is, the capability of a component
grip bolt connection) above and beyond its load-carrying capacity to
Cable support structures 10b 7 meet the respective requirements regarding leak
Fluid media 0.5 0.5 tightness and deformation restrictions; and (3) the
a
If, on account of the design, deformations are possible functional capability, that is, the capacity of a
only in small regions of the structure (low structural system or component above and beyond its load
damping), the values as listed shall be halved (special
cases) carrying capacity to fulfill the designated tasks by
b
In well-substantiated cases, the damping ratio may be way of its respective mechanical or electrical
increased up to 15 % function.

Attachments Under
Seismic Actions: 40
Fig. 11 Normalized floor
30
acceleration based on
Eq. 11
20
10
0
0.1 1.0 10.0 100.0
z/H = 0.24 z/H = 0.43 z/H = 0.62 z/H = 0.81 z/H = 1.00
Depending on the importance of the element The verifications have to consider potential
to verify and the material (conventional or damage induced to other elements, which loss
nonconventional material), the verification proof capacity could induce either human or
cess could include experimental tests either for economic loss.
the validation of the numerical model or for the The verification is performed checking that
qualification of the element itself. The verifica- the element capacity will be greater than the
tion procedure can include a) analysis, b) physi- demand, defined in terms of different mechanical
cal experiments, and c) analogies and plausibility properties (stresses, forces, displacement)
considerations. depending on the adopted materials.
Based on the introduced classification that In order to define the design forces, the con-
distinguishes secondary element in structural sidered floor response acceleration spectra can be
and nonstructural, the following criteria can be reduced to take into account the nonstructural
defined: element ductility. If the FRSs have been numer-
ically evaluated, they have to be modified (see
1. Secondary structural elements have to be ver- Fig. 12) to take into account the structural stiff-
ified with regard to the vertical loads transmit- ness uncertainties: (1) an adequate plateau has to
ted from the P structure, opportunely be imposed in correspondence of the structural
combined with the other action considered to period, (2) the linear envelope has to be properly
be contemporary to the seismic action. The introduced, and (3) the ductility of the
connections have to be verified with regard nonstructural element can be considered, prop-
to the seismic-induced action, including erly reducing the FRS (see Fig. 12b, c). S
second-order effects such as those induced
by the axial load in the deformed configuration
(P-D effects). Case Study
2. Secondary nonstructural elements have to be
verified with regard to the self-weight loads The previously described procedures to deter-
opportunely combined with the other actions mine FRSs will be applied to a steel frame system
considered to be contemporary to the seismic hosting a set of equipments, whose characteris-
action. tics and localization are reported in Table 6 and
3. For both types of elements (S and NS ele- Fig. 13 (left), reproduced from KTA (2012).
ments), the action supposed to act contempo- The maximum acceleration of each equipment
rary with the seismic action can be consequent can be evaluated by means of (1) time histories
to different events such as those pertaining to considering the interaction between the principal
collisions, explosions, and fires. structure and the equipment or (2) FRSs

Attachments Under
Seismic Actions:
Fig. 12 (a) Design spectra
definition. Shape
modification: alternative
solution. (b) Design spectra
definition. Design spectra
based on ABCDEF
spectrum. (c) Design
spectra definition. Design
spectra based on
ABB’CDEF spectrum
Secondary Structures and Attachments Under Seismic equipments, (3) performing a dynamic modal TH
Actions: Modeling and Analysis, Table 6 Equipment analysis evaluating, for each accelerogram, the
mass (Ton) and period (sec)
maximum absolute value of a given quantity
Equipment Mass Floor Period (acceleration), and (4) averaging the maximum
1 20 1 0.051 values obtained (at step 3) for each analysis.
If the detailed model with interaction (WI) is
2 10 1 0.093
considered, the evaluated quantities are the mass
accelerations of the single equipment.
Regarding to the lumped mass system, (1) for
each considered accelerogram, an FRS has been
3,4,5 10 2 0.070 generated (Fig. 16a): the FRS is relative to the
acceleration of the top left floor corner
(no sensible variations in FRSs have been
observed if other floor points are considered);
6 10 3 0.034 (2) having generated (for each floor) seven
FRSs, the averaged FRS has been evaluated
(Fig. 16b); and (3) for each equipment, depending
7 20 3 0.060 on the floor it is attached to and its period (TNS),
the acceleration has been evaluated through the
resulting FRS.
8 30 4 0.033
The results of the performed evaluation are
reported in Table 7 where the acceleration of
9,10 15 5 0.036 each floor and the acceleration of each equipment
are reported, calculated with or without
interaction.
Comparing the maximum acceleration evalu-
ated considering the PS system with those evalu-
evaluated based on the previously described cas- ated through the cascade procedure, a significant
cade procedure or on the predictive expressions difference (if the Equipment 2 is concerned)
already presented. between the two approaches can be noticed: the
difference is aspectable since the equipment
Time History-Based Evaluation of Equipment period (0.093 s) is close to the period of the
Accelerations third modal shape (0.090 s) so that (see
A detailed model of the PS system could include Fig. 16c) in a very small period range
the secondary system modeled as reported in (in between 0.09 and 0.1 s), the acceleration S
Fig. 13 (right): PS principal modal shapes are ranges between 7.0 and 4.5 m/s – the already-
those reported in Fig. 14a. Alternatively equip- mentioned tuning effect causes the equipment
ments can be modeled by means of mass lumped acceleration reduction if the complete PS system
at pertinent position of the floor as reported in is analyzed in order to include the P-S interaction.
Fig. 13 (center): PS principal modal shapes are
those reported in Fig. 14b. Analytical FRS-Based Evaluation of
Both models have been analyzed by means of Equipment Accelerations
time histories, carried out (1) generating It has been already outlined that CEN proposal
7 accelerograms compatible with the acceleration (CEN 2008) and FEM proposal (FEMA
spectra (PGA = 0.15 g) suggested in CEN (2008) 369 2001) are based on a fixed value (5 %) of
for B soil and low-magnitude events (M < 5.5) structural and equipment damping. So that only
(see Fig. 15), (2) considering a constant damping the proposal reported in AFPS (2007) and KTA
value of 2 % for the structural model and for the (2012) will be considered in the following.
Beams Braces
HEB600 A = 0.02 m2
15 t 15 t
15 t 15 t
4.0
HEB 600
30 t
30 t
30 t
HEB 600
4.0
10 t
10 t 20 t
20 t 10 t 20 t
HEB 800
4.0
10 t 10 t 10 t 10 t 10 t 10 t
10 t 10 t 10 t
HEB 800
40
20 t
20 t 10 t 20 t 10 t 10 t
5.0
HEB 800
Secondary Structures and Attachments Under Seismic Actions: Modeling and Analysis, Fig. 13 Case study:
geometry (left) and models (center, right)
a b
Mode 1 Mode 2 Mode 3 Mode 1 Mode 2 Mode 3
Secondary Structures and Attachments Under Seismic 0.48, 0.158, 0.121. (b) Modal shape: equipment modeled
Actions: Modeling and Analysis, Fig. 14 (a) Modal as lumped masses. Periods (sec): 0.49, 0.159, 0.09
shape: direct modeling of the equipment. Periods (sec):
Secondary Structures and 0.6

Attachments Under
0.5
Seismic Actions:
Acceleration (g)
Modeling and Analysis, 0.4

Fig. 15 Target
acceleration spectrum 0.3
(PGA = 0.15 g) and spectra 0.2
of the generated
accelerograms 0.1
0
0 1 2 3 4
Period (sec)
Secondary Structures and a

Attachments Under 2.5
Seismic Actions:
2.0
1.5
Fig. 16 (a) Floor
1.0
Response Spectra
generations: scheme. (b) 0.5
Generated Floor Response 0.0
0.0 0.5 1.0
Spectra. (c) Generated
Floor Response Spectra: 1
Floor; Period range
0.07–0.1 s
0.2
0.1
0
–0.1
–0.2
0.2
0.1
0
–0.1
–0.2
b 60
0.49
50
Acceleration (m/sec2)
40
30 0.16
20
0.09
10
0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6
c 8 S
7
0
0.07 0.08 0.09 0.10
Period (sec)
Secondary Structures and Attachments Under Seismic It has to be specified that (1) concerning the
Actions: Modeling and Analysis, Table 7 Floor accel- KTA (2012) proposal, the floor acceleration eval-
eration (FA, m/sec2) and equipment acceleration evaluated
without (F_w/o) and with (F_w) interaction uated by means of TH analyses (see Table 7) has
been considered and (2) concerning the AFPS
NE NF Floor acceleration FRS W/0 FRS W
(2007) proposal, the FRSs have been evaluated
1 1 2.21 2.7 2.1
by means of Eq. 9, evaluating the spectral accel-
2 1 2.21 6.5 3.2
eration corresponding to the first structural period
3 2 3.26 4.20 4.0
4 2 3.26 4.20 4.0
(0.49 s) and a participation factor (GS) equal to
5 2 3.26 4.20 4.0 1.6 that correspond to a = 1.5 (see Eq. 10a).
6 3 3.65 3.70 3.8 It is clear (see Fig. 17a, b) that considering
7 3 3.65 5.10 3.8 equipment periods close to the lowest structural
8 4 4.28 4.30 4.3 periods, the KTA proposed expressions are more
9 5 5.25 5.3 5.3 conservative while the AFPS expressions are less
10 5 5.25 5.3 3.3 conservative (Tables 8 and 9).
Secondary Structures and a

Attachments Under 60
Seismic Actions:
Fig. 17 (a) Generated
FRSs. Acceleration versus

period: (1) numerical 40
simulation and (2) AFPS
predictive equations 30
evaluated based on Eq.
9 (a = 1.5, GS = 1.6, 20
RS = 2.4). (b) Generated
FRSs. Acceleration versus
frequency: numerical 10
simulation and KTA
predictive equations 0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Period (sec)
b
80
70
60
50
40
30
20
10
0
0 1 10 100
Frequency (Hz)
Secondary Structures and Attachments Under Seismic secondary element (with a given period and
Actions: Modeling and Analysis, Table 8 Normalized damping) will be subjected to when attached
floor displacement
(from which the alternative name attachments
Floor MD z/H derives) to a given part of the structures such as
1 0.21 0.24 a building floor (from which the name Floor
2 0.45 0.42 Response Spectra derives).
3 0.68 0.62
Given that FRS generation could require oner-
4 0.87 0.81
ous numerical analyses, simplified expressions
5 1.00 1.00
are proposed in literature and discussed in the
following together with the general methodolo-
gies tailored to secondary element modeling and
Secondary Structures and Attachments Under Seismic
Actions: Modeling and Analysis, Table 9 Equipment seismic analyses.
(Eq) acceleration
NE NF AFPS KTA
1 1 2.2 22.0
2 1 2.2 34.0 Cross-References
3,4,5 2 3.3 42.0
6 3 3.7 22.0 ▶ Behavior Factor and Ductility
7 3 3.7 42.0 ▶ Building Codes and Standards
8 4 4.3 25.0 ▶ Classically and Nonclassically Damped
9,10 5 5.3 35.0 Multi-degree of Freedom (MDOF)
Structural Systems, Dynamic Response
Characterization of
▶ Code-Based Design: Seismic Isolation of
Summary Buildings
▶ Earthquake Risk Mitigation of Lifelines and
The chapter deals with the methodologies Critical Facilities
focused on the seismic analyses of the so-called ▶ Equivalent Static Analysis of Structures
secondary (sometimes attachments) elements Subjected to Seismic Actions
that are part of a construction whose seismic ▶ European Structural Design Codes: Seismic
resistance is delegated to a primary resistant Actions
structure. ▶ Modal Analysis
Although secondary elements can be ▶ Nonlinear Dynamic Seismic Analysis
decontextualized from the primary resistant ▶ Performance-Based Design Procedure for
structures, they will be subjected to seismic Structures with Magneto-Rheological S
action as well and, having their own structures, Dampers
need to be modeled and analyzed by means of ▶ Response Spectrum Analysis of Structures
methods included in the general methodologies Subjected to Seismic Actions
proper of seismic branch. ▶ Seismic Fragility Analysis
Among the methodologies usually adopted for ▶ Soil-Structure Interaction
secondary element analyses, the Floor Response ▶ Spatial Variability of Ground Motion: Seismic
Spectra (FRS)-based analyses become popular Analysis
due to their recognized simplicity. ▶ Tuned Mass Dampers for Passive
FRSs provide acceleration (consequently Control of Structures Under Earthquake
velocity and displacement) to which the Excitations
2504 Seismic Accelerometers
References Taghavi S, Miranda E (2003) Response assessment of

nonstructural building elements. Pacific Earthquake
AFPS (2007) Guide methodologique pour la conception, Engineering Research Center, PEER 2003/05 Sept
l’installation et le diagnostic des equipment en zone Villaverde R (1997) Seismic design of secondary struc-
sismique. Association Francaise de Genie tures: state of the art. J Struct Eng 123(8):1011–1019
Parasismique, Paris
ASCE 7-02 (2002) Minimum design loads for buildings
and other structures. American Society of Civil Engi-
neers, Reston
ATC 58 (2012) Seismic performance assessment of build- Seismic Accelerometers
ings. Applied Technology Council, Redwood City
ATC/BSSC (1997) NEHRP commentary on the guide-
lines for the seismic rehabilitation of buildings. Report Gerardo Alguacil1 and Jens Havskov2
1
no FEMA 274, prepared by the Applied Technology Instituto Andaluz de Geofı́sica, University of
Council for the Building Seismic Safety Council, Granada, Granada, Spain
published by the Federal Emergency Management 2
Agency, Washington, DC
Department of Earth Science, University of
CEN (2008) Eurocode 8: design of structures for earth- Bergen, Bergen, Norway
quake resistance, part 1: general rules, seismic actions
and rules for buildings. Comité Européen de
Normalisation, Brussels
Chaudhri SR, Villaverde R (2008) Effect of building
Synonyms
nonlinearity on seismic response of nonstructural com-
ponents: a parametric study. J Struct Eng Broadband seismometers; Force-balance acceler-
134(4):661–670 ometers; MEMS accelerometers; Passive
Chen Y, Soong TT (1988) State-of-the-art review seismic
seismometers; Seismic recorders; Servo-acceler-
response of secondary systems. Eng Struct
10:218–228 ometers; Strong ground motion
FEMA 369 (2001) NEHRP recommended provisions for
seismic regulations for new buildings and other struc-
tures: part 2-commentary. Building Seismic Safety
Council, National Institute of Building Sciences,
Introduction
Washington, DC
FEMA 74 (2005) Earthquake hazard mitigation for Earthquake ground motion ranges in amplitude
nonstructural elements. Field Manual, Emergency from a few nanometers (e.g., for a distant earth-
Management Agency, Washington, DC
quake or a local microearthquake) to several
Gilani AS, Whittaker AS, Fenves GL, Fujisaki E (1999)
Seismic evaluation of 55 kV porcelain transformer. meters close to the fault causing a big quake.
Pacific Earthquake Engineering Research Center, It is very difficult to record that wide range of
Berkeley signals with a single type of instrument. Seis-
(IEEE) (2005) Recommended practice for seismic design
mometers are designed to be very sensitive for
of substations, Institute of Electrical and Electronics
Engineers, New York detecting weak signals of ground motion and
Igusa T, Der Kiureghian A (1985a) Dynamic characteri- have a response proportional to ground velocity
zation of two degree-of-freedom equipment-structure in a frequency band typically from 0.01 to
systems. J Eng Mech ASCE 111(1):1–19
100 Hz. On the other hand, strong ground motion
Igusa T, Der Kiureghian A (1985b) Dynamic response of
multiply supported secondary systems. J Eng Mech instruments are in general less sensitive and may
ASCE 111(1):20–41 manage on scale large ground motion amplitudes.
Kircher CA (2003) It makes dollars and sense to improve This strong ground motion may reach peak accel-
nonstructural system performance. In: Proceedings of
erations above 2 g (g is gravity acceleration) and
ATC 29–2 seminar on seismic d, performance, and
retrofit of nonstructural components in critical facili- peak velocities higher than 3 m/s (Anderson
ties, Newport Beach, 23–24 Oct 2003 2010). The motion experienced at specific points
KTA (2012) Safety standards of the nuclear safety stan- in a building or structure may be even higher.
dards commission: design of nuclear power plants
Seismic accelerometers were developed for
against seismic events; Part 4: components.
KTA-Geschaeftsstelle c/o BfS, Willy-Brandt-Str. recording on scale vibrations up to such range
5, 38226 Salzgitter of amplitude either on ground or in structures.
Seismic Accelerometers 2505
Due to instrumental and practical reasons, the

preferred motion measure for strong ground
motion is acceleration, since the inertial force
acting on a structure due to a seismic action is
proportional to ground acceleration. Neverthe-
less, some strong-motion velocimeters are com-
mercially available, but their use is much less
widespread.
Weak motion from local small earthquakes or
moderate to strong earthquakes at regional or
large distances are studied fundamentally on
seismograms, proportional to ground velocity in
a wide frequency band. However, even for strong
motion, ground velocity and displacement are Seismic Accelerometers, Fig. 1 Simplified schematics
of a seismoscope. A mass is suspended from the frame by
also of interest: Ground velocity is directly
a wire, and a stylus writes the relative motion on a spher-
related to important parameters such as shake ical cap surface
energy or soil liquefaction potential and is linked
to damage on intermediate period structures and
buried pipelines (e.g., Akkar and Bommer 2007). This instrument was built with a mass suspended
Ground displacement is significant for large on torsion wires, based on the Wood-Anderson
structures, where differential displacements may seismometer principle (known because it was
cause damage, and from a seismological view, it used by Richter to define his magnitude scale).
is proportional to seismic moment, a source Damping was achieved – like for the Wood-
parameter defining the earthquake size. There- Anderson seismometer – using a conductive cop-
fore, it is desirable that velocity and displacement per mass in which the field of a permanent
may be estimated from acceleration records. magnet induced parasitic currents. A more robust
pivoted suspension and a spring were later incor-
porated. Recording was made on photographic
Historical Overview paper on a drum, with a “starter” system built
with an independent undamped pendulum clos-
This section partly follows the review article by ing an electric contact when a strong motion
Trifunac (2009). See also “▶ Historical occurred. A clock mechanism interrupted the
Seismometer.” light beam to produce time marks on the record.
Earlier attempts to record strong ground The first strong ground motion accelerograms
motion were made with the so-called were obtained on three stations in 1933 from the S
seismoscopes (Fig. 1), most of which were lim- Long Beach earthquake (McComb and Ruge
ited to record the horizontal ground motion on a 1937).
(curved) surface without a time reference. These With analog records and the computational
instruments consisted of a pendulum, either nor- means available at that epoch, it was very difficult
mal or inverted, whose mass motion was drawn to estimate displacement from acceleration.
on a curved surface (usually a smoked glass) by a Therefore, some seismometers with unity ampli-
stylus. fication and photographic recording were used to
As a result of the program of strong-motion obtain displacement records (see “▶ Passive
earthquake investigation of the US Coast and Seismometers”).
Geodetic Survey and cooperative institutions A remarkable effort was made to test these
(McComb and Ruge 1937), the first instrument first instruments with a shaking table to assess
designed to record the strong ground motion ver- the independent motion of the components, zero
sus time was developed in the decade of 1930. shifts (instability of the mass rest position),
parasitic vibrations (instrument vibration modes

different from fundamental), the ability to calcu- y(t)
late velocity and displacement from acceleration
k
records, and other features (McComb and Ruge
d
1937, Ruge and McComb 1943). m
It was not until early 1960 that accelerographs
become commercially available. Around that
time the digital computation also made it possible z(t) u(t)
to calculate the velocity and displacement histo-
ries from the corresponding accelerograms.
Seismic Accelerometers, Fig. 2 A schematic single
In this decade, some seismic accelerometers degree of freedom oscillator. A spring makes a force
were still of torsion wire or inverted pendulum proportional and opposed to the relative displacement of
types, but the first servo-accelerometers using the the mass z(t), and a dashpot damps this motion with a force
concept of force balance appeared (Reasemberg proportional to the relative velocity. The ground (and the
frame) displacement is u(t)
1969; Eisenberg and McEvilly 1971). This con-
cept (see later) is to apply a feedback force on the
mass to keep it at rest relative to the ground. So
the applied force is the mass times the ground u(t) with respect to some inertial frame, the mass
acceleration. The main advantages are improved displacement is y(t) in this inertial frame, and
linearity and dynamic range (i.e., the relation z(t) = y(t) u(t) is the mass displacement rela-
between the maximal non-clipped signal ampli- tive to the instrument casing or the ground.
tude and the minimal resolvable amplitude; see A dashpot represents a damping device
later). Recording was analog on photographic (usually magnetic) that acts on the mass with a
paper, but some prototypes used digital recording force proportional to its relative velocity. This
(Reasemberg 1969). The first digitizers used, damping avoids the oscillation of the system
however, had a too low resolution to match the with its free period when it is excited. The two
dynamic range of servo-accelerometers. real forces acting on the mass are the spring force
Following the period of photographic record- and the damping force, so the dynamics equation
ing, some models adopted analog recording on is (Havskov and Alguacil 2010)
magnetic tape using frequency modulation, but
digital recording was soon the norm due to its k z d z_ ¼ m yðtÞ ¼ m½u€ðtÞ þ z€ðtÞ (1)
better characteristics. Presently all high-
performance seismic accelerometers use force- It is useful to write the equation as a function of
balance sensors and high-resolution digital the two coefficients experimentally measurable.
recording. The free or natural angular frequency o0 can be
defined as
rffiffiffiffi
Basic Principles of Accelerometers k 2p
o0 ¼ ¼ 2pf 0 ¼ (2)
m T0
Seismometers and accelerometers are both based
on the inertia principle. where T0 is the suspension-free period and its
In Fig. 2, a schematic view of the well-known inverse f0 is the natural frequency. The damping
passive damped oscillator is drawn. A mass m is coefficient or damping fraction h is
fixed to one end of a spring of elastic constant k.
The other end of the spring is fixed to a d
h¼ (3)
frame anchored to the ground. The motion is 2o0 m
mechanically limited to the horizontal axis
shown. When the ground moves a displacement Rearranging Eq. 1 then gives
z€ðtÞ þ 2ho0 z_ðtÞ þ o20 z ¼ €

uðtÞ (4) The explicit frequency response function is
obtained by substitution of s by io in Eq. 6,
Both o0 and h can be measured, e.g., from the that is,
system transient response or by exciting the sys-
1
tem with steady-state harmonics of frequencies in H a ðo Þ ¼ (7)
a band including f0. For low frequencies, the mass o2 þ 2ho0 io þ o20
velocity and acceleration terms in Eq. 4 are small,
For a sinusoidal motion of arbitrary frequency
so mass displacement z is then proportional to the
f = o/2p, the modulus of this complex function
ground acceleration u€.
represents the amplitude relation between the
The system response may be characterized in
mass relative motion z(t) and the ground acceler-
the Laplace transform domain. Let Z(s) and U(s)
ation u€(t), and its phase is the relative phase
be the Laplace transforms of z(t) and u(t). By
between them. Both are plotted in Fig. 3 as func-
assuming null velocities and displacements at
tions of frequency. The amplitude response is flat
initial time, the Laplace transform of Eq. 4
for acceleration up to the corner frequency f0 and
becomes
decays at higher frequencies as f 2. The frequen-
cies of interest are usually in the flat zone, so the
s2 Z ðsÞ þ 2ho0 sZ ðsÞ þ o20 ZðsÞ ¼ s2 U ðsÞ (5) instrumental correction for amplitude is simply a
constant factor. Observe, nevertheless, that the
The transfer function between the mass relative
phase response deviates from flatness even for
motion Z and the ground acceleration s2U is then
frequencies well under the natural frequency.
This response function is formally like the
Z ðsÞ 1
T a ðsÞ ¼ (6) response of a second-order low-pass filter with
s2 UðsÞ s2 þ 2ho0 s þ o20 cutoff frequency f0 = o0/2p.
Seismic Accelerometers, AMPLITUDE RESPONSE

Fig. 3 Top: amplitude 10−4
response of a mechanical
accelerometer for ground
acceleration. Bottom: phase
10−5
cm/(cm/s2)
response. The free period is

T0 = 0.02 s (f0 = 50 Hz)
and h = 0.70
10−6
10−7 S
0,1 1 10 100 1.000
PHASE RESPONSE
0
−30
−60
DEGREES
−90
−120
−150
−180
0,1 1 10 100 1.000

FREQUENCY Hz
Displacement Transducer difference, and the phase depends on the sign of

this difference. This may be demodulated with a
The mass motion has to be measured and phase-sensitive demodulator (PSD) circuit.
recorded by some device. In early accelerome- While this is the most used technique, other
ters, a light beam was reflected in a mirror which approaches are possible.
rotated with the mass motion. Presently, almost The output of the detector is proportional to
all seismic accelerometers use a capacitive trans- the capacitance difference. Let A be the
ducer, which gives a voltage output proportional overlapping area between plates and d the dis-
to the mass displacement. This type of transducer tance (gap) between them. If d is much smaller
is very sensitive – it can resolve displacements of than the plate dimensions, the capacitance of a
the order of pm (109 mm). parallel plane capacitor is
Two types of capacitive transducers may be
eA
used: variable gap or variable area. The most used C¼ (8)
arrangement, Fig. 4, is a pair of capacitors with a d
common central moving plate and two fixed where e is the dielectric permittivity (for air
plates or vice versa (variable gap). When the e = 8.85 103 pF/mm). In the variable gap
central plate moves with the mass relative to the transducer, for a displacement x of the central
fixed plates, one of the capacities increases and plate, it may be easily shown that the capacitance
the other decreases. The same happens with the difference is
variable area type. An identical sinusoidal or
square signal with frequency of several kHz is 2x
fed to each capacitor with opposite sign. The C1 C2 ¼ e A (9)
d 2 x2
signal amplitude at the common point of both
capacitors is proportional to the capacitance which is not linear with x.
a VARIABLE GAP b VARIABLE AREA

CAPACITIVE TRANSDUCER CAPACITIVE TRANSDUCER
MOVING
MOVING FIXED PLATE
PLATE PLATES
c +
C1
FIXED Vs
PLATES -
output
.
+
C2
−Vs
-
Seismic Accelerometers, Fig. 4 Schematics of variable increases). (b) Variable area. The effective area of each
capacitance displacement transducers. (a) Variable gap. capacitor (same colors as in a) varies in opposite sense
As the central plate (red) moves right, the gap with the left when the central plate (red) moves horizontally. The gap
fixed plate (blue) increases (capacitance of left capacitor, is not at scale, since in practice it is very narrow in relation
yellow, decreases), and the gap with the right plate (blue) to plate dimensions. (c) Circuit schematics, see text
decreases (capacitance of the right capacitor, light blue,
Seismic Accelerometers, CAPACITIVE TRANSDUCERS

Fig. 5 Normalized 150
sensitivity of capacitive
transducers of variable gap 100
NORMALIZED SENSITIVITY
(blue) and variable area
(red). A straight line is
plotted for reference 50
(dotted) with the slope of
the variable gap at the 0
origin
−50
VARIABLE GAP
−100 reference slope
VARIABLE AREA
−150
−0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5
DISPLACEMENT mm
R
FORCE
c
TRANSDUCER
D AMPLIFIER
-
MECHANICAL DISPLACEMENT
+
GROUND SYSTEM TRANSDUCER
OUTPUT
ACCELERATION
Seismic Accelerometers, Fig. 6 Block schematic of an FBA
In a variable area transducer with plate overlap Variable area transducers are used in some
dimensions a b, the capacitance difference is micro-electro-mechanical (MEM) accelerome-
linear with the displacement x: ters (see later) without feedback.
e
C1 C2 ¼ 2b x (10)
b The Force-Balance Accelerometer (FBA) S
In Fig. 5, these functions – excluding e – are Most present seismic accelerometers are of the
plotted for an overlapping area of force-balance type. The principle is to apply a
A = 100 mm2 and gap distance of d = 1 mm feedback force on the inertial mass opposed to
and 0.5 mm for variable gap and variable area, its motion, in such a way that this motion is
respectively. Two features are clear: variable gap reduced to a minimum. A simplified block sche-
type is more sensitive but nonlinear, and variable matic is shown in Fig. 6.
area type is less sensitive but linear. The ground acceleration produces an inertial
Variable gap transducers are often used in force on the mass, like in the basic passive accel-
feedback systems, like force-balance accelerom- erometer of Fig. 1. The relative mass motion z is
eters, since the mass displacement is held very sensed with a displacement transducer, whose
small – within the linear zone – by the servo output is amplified, with a total factor D (volt/
control. meter, V/m). The amplifier output vo is applied
101
100
AMPLITUDE RESPONSE
10−1
−2
10
a-mechanical accelerometer response mm/(m/s2)

b-voltage output without compensation V/(m/s2)
−3
10 c-voltage output with compensation V/(m/s2)
d-feedback acceleration
e-mass displacement micron/(m/s2)
−4
10
1 10 100 1.000
FREQUENCY Hz
Seismic Accelerometers, Fig. 7 Several responses relative to ground acceleration. See text. Note the different units
for each curve
through a resistor R to a force transducer This is not desirable, since a low damping will
(normally a coil-magnet system) with constant produce a ringing transient response or even an
c (newton/ampere, N/A). The feedback force unstable system. The frequency response (Fig. 7)
c vo/R = c A z /R is then applied to the has the same form as Eq. 7, but in this case it will
mass. Equation 1 has now to be modified to show a resonance peak.
Actually, the phase-sensitive demodulator
Dc (PSD) associated with the displacement trans-
k z d z_ z ¼ m½u€ðtÞ þ z€ðtÞ (11) ducer will often include a low-pass filter at fre-
R
quencies above the seismic band, thus
It is clear that the effect of feedback is to increase introducing an additional phase delay in
the effective stiffness of the system from k to this band.
k + D c/R. Arranging terms of Eq. 11 to get Some additional circuitry is then required to
the form of Eq. 4, it is found that with feedback, improve the transient response and prevent a
a new effective “free” angular frequency is of possible self-oscillation of the servo system. Sev-
eral techniques for this are possible. A simple one
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi consists of including a second feedback loop to
k þ D c=m Dc
of ¼ ¼ o20 þ (12) add a small force proportional to the time deriv-
m Rm
ative of the mass displacement, i.e., the mass
which is higher than the “open-loop” free angular velocity. This allows the feedback force to con-
frequency o0. trol the mass motion “in advance,” and the servo
On the other hand, the new damping coeffi- loop is stabilized.
cient with feedback hf is decreased to Figure 7 shows the responses relative to
ground acceleration of several systems and out-
o0
hf ¼ h (13) put points of the systems. An under-damped
of mechanical accelerometer (without feedback
force) is drawn in curve a, in this example with a limited bandwidth, but this will not affect the
free oscillation frequency of 10Hz. Curve band of interest. The feedback acceleration has
b corresponds to an FBA built with this system a little overshoot to counteract the resonance and
by adding a displacement transducer and a feed- keep the mass as stationary as possible. The mass
back force proportional to the mass displacement, displacement is also shown (with sign changed)
which increases the resonance frequency. The and has an amplitude about 0.6 mm, as predicted
damping is still too low and a strong resonance by the frequency response in Fig. 7.
peak at 200 Hz appears. Curve c shows the volt- The FBA principle is the basis not only of
age output response for the same FBA once sta- servo-accelerometers but also of ▶ broadband
bilized by the addition of a small feedback force seismometers (BB). These instruments have a
proportional to the mass velocity. Curve d shows flat frequency response for ground velocity in a
the feedback acceleration response (i.e., the feed- band from ~0.01 to several tens of Hz. The feed-
back force per unit mass), which is nearly equal to back loop keeps the mass almost at rest relative to
the ground acceleration (so the amplitude the ground, but the signal is integrated in the loop,
response is 1) up to the frequencies around the and the output is taken from a point where it is
resonance, where the feedback force needs to be a proportional to ground velocity in a wide band.
little higher to stabilize the motion. Curve e plots For more details, see the entry “▶ Principles of
the mass motion response measured in microns Broadband Seismometry.”
(mm) relative to ground acceleration in m/s2. It is
approximately 0.6 in the useful band, so a ground
acceleration of 1 g will produce a mass motion of Characteristic Parameters of
9.8 m s2 0.6 mm/( m s2)
6 mm, a very Accelerometers
small displacement, for which the capacitive
transducer will be quite linear. The performance of a seismic accelerometer may
It is also illustrative to view the transient be characterized by several parameters.
response of the FBA (Fig. 8). The input is a Sensitivity: The relation between the voltage
simulated ground acceleration pulse of 1 m/s2 output and the input acceleration in V/m/s2 or V/g
and with duration of 50 ms. It can be seen that (gravity normal acceleration = 9.807 m/s2).
the output voltage has only a small overshot, due Typical values are 0.5–5 V/g.
to the compensation circuit in the feedback. So, Input Full Scale (FS): The maximum ground
there is an unavoidable small distortion affecting acceleration before the instrument gets saturated.
the higher frequencies, like any instrument with a A large FS prevents saturation with strong ground
Seismic Accelerometers, 1.5

Fig. 8 Transient response
of the example FBA
output voltage V S
feedback acceleration m/s2
(Fig. 7) for a pulse in input acceleration m/s2
acceleration input of 1 mass motion micron
1 m s2
0.5
−0.5
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
TIME s
motion but at a cost of sensitivity. In general, for techniques of manufacturing integrated circuits
accelerometers to be installed in zones where to the production of very lightweight and
large ground motion is expected, FS should be low-power-sensitive accelerometers.
2–4 g, but in low-moderate seismicity areas, In recent years, the industry has found a large
0.5–1 g would be suitable. Nevertheless, it should number of consumer applications for this kind of
be kept in mind that some of the largest peak accelerometers. This has led to the development
ground motions ever recorded (up to 4.36 g for of micro-sized, light and low-power sensors, suit-
Iwate-Miyagi 2008 earthquake; Yamada able for mass production and thus inexpensive.
et al. 2009) have been produced by moderate- These are the micro-electro-mechanical system
size earthquakes (Anderson 2010). Some accel- (MEMS) devices. MEMS accelerometers are
erometers allow user setting its FS. used for automotive industry (e.g., for air-bag
Resolution: It is the minimum acceleration triggering), inertial navigation, medical applica-
amplitude that it may measure. In early acceler- tions, game consoles, cellular phones, tablets and
ometer digital recorders, it was limited by the laptops, etc. Most of these devices have a poor
digitizer (A/D converter) resolution. In modern performance for seismic recording, but a few
instruments, the digitizer is usually 24 bit, and the commercial models are suitable for
resolution is related to the self-noise level of both low-resolution strong ground motion recording
the accelerometer and the digitizer. So it is usu- and earthquake early warning (EEW) systems or
ally given as self-noise level, or it may be for triggering shutoff procedures on critical facil-
obtained from the dynamic range. Good-quality ities (trains, gas valves, lifts, critical machinery)
accelerometers have noise levels under 1 mm/s2 in case of a strong earthquake (e.g., Yanada et al.
rms (root mean square). 2002). Furthermore, some seismic instrumenta-
Dynamic Range: The relation (expressed in tion companies have developed MEMS acceler-
decibels, dB) between the maximum acceleration ometers for seismic-grade recorders with
amplitude (FS) and the resolution. A ratio comparable performance of conventional sen-
between two amplitudes a1 and a2 expressed in sors. The idea is to build the mass-spring system
dB is 20 log(a1/a2); for energy or power ratios, within the multilayer structure of an integrated
a factor 10 instead of 20 is used. The best circuit (Fig. 9), using the same techniques as for
accelerometers now have a dynamic range up to electronic components and circuits.
more than 150 dB, which means a ratio The internal noise of a spring-mass system is
10150/20 > 30 106, but most instruments due to the thermal-mechanical coupling by
achieve 120–140 dB. means of the damping dissipative forces. For
instance, the air molecules interchange kinetic
energy with the mass, which then “dances” ran-
MEMS Accelerometers domly with a Brownian motion.
It may be shown (e.g., Aki and Richards 2002;
The same principles described so far are valid Havskov and Alguacil 2010) that this noise is
regardless the size of the accelerometer, but proportional (Eq. 14) to the mechanical damping
accelerometer sensors of micrometric size have coefficient h (not included the feedback effect)
specific characteristics, so a brief description and inversely proportional to the suspended mass.
The power spectral density (PSD) Z€n of the mass
2
follows.
Originally, the aim of reducing the size of acceleration noise (the PSD integrated between
accelerometer sensors was due to the manufactur- two frequencies gives the noise variance in this
ing of integrated accelerometers for industry and band) is
navigation (airborne) applications. The space
exploration requirements, among other applica- 8kB T
Z€n ¼
2
ho0 (14)
tions, led to the adaptation of some of the m
Seismic Accelerometers, Fig. 9 Principal elements of a with the variance in the capacitance. For high-sensitive
MEMS (micro-electro-mechanical system) accelerometer applications, a feedback circuit is added which controls a
with capacitive transducer. The mass is the upper mobile restoring electrostatic force, thus making an FBA. The
capacitor plate which can rotate around the torsion bars. size of the sensor above is about 2 mm (Figure from
The displacement, proportional to acceleration, is sensed www.silicondesigns.com/tech.html)
where kB is the Boltzmann constant The two capacitive techniques described

(1.38 1023 J/K), T is Kelvin temperature, above are used: variable gap between plates and
m the suspended mass, h the mechanical damping variable area. The first is quite nonlinear and
coefficient, and o0 the free angular frequency. usually requires electrostatic feedback to linear-
This equation explains why a good part of the ize the transducer response. Recently, electro-
effort in improving MEMS accelerometer noise magnetic feedback has also been proposed for
is focused (Walmsley et al. 2009) in increasing MEMS accelerometers (Dwyer 2011). On the
the proof mass and reducing damping. other hand, the variable area technique is more
Thus, for a micro mass, the open-loop linear, and some devices using it operate in open-
damping should be kept as low as possible for loop mode (e.g., Homeijer et al. 2011). Figure 10
an acceptable noise level at ambient temperature. shows a possible arrangement for a MEMS
High-purity silicon spring elements are almost accelerometer.
perfectly elastic, and the spring is operating in MEMS sensors are now made with electro-
vacuum to avoid air damping, so the mechanical chemical sensors (Deng et al. 2013) with a noise
damping can be held quite low. PSD of 0.1 (mm/s2)/√Hz or 10 ng/√Hz at 1 Hz, so
Several strategies have been tested in proto- these sensors could be used at medium noisy sites.
types for sensing the mass motion. Most commer- All of these devices have two distinct parts S
cial devices use some kind of variable within the same package: the mechanical sensor
capacitance. Other sensing designs are based on and the associated electronics, typically
tunnel effect, optical diffraction and interferom- implemented by an ASIC (application-specific
etry, piezo-resistivity, piezoelectricity, reso- integrated circuit). Some commercial models
nance, or electrochemical phenomena. Very have a sigma-delta modulator (a device yielding
small capacitive transducers are not practical in a sequence of logical pulses whose average value
general, since the circuitry parasitic capacitances is proportional to its analog voltage input)
make the relative changes too small. Neverthe- included (see “▶ Recording Seismic Signals”),
less, these parasitic capacitances may be reduced and so its output is a pulse-density digital signal,
to a minimum in MEMS devices, since the asso- able to be directly interfaced to a microcontroller
ciated circuit is integrated within the same chip or computer. The integrated MEMS has to be
(see, for instance, Li et al. 2001).
Seismic Accelerometers,
FRAME
Fig. 10 A possible
arrangement of the mass,
spring, and capacitor plates CAPACITOR
in a MEMS accelerometer. PLATES
All elements are in a layer a
few microns thick
PROOF MASS
LEAF SPRING
mounted in a printed circuit board (PCB) with com/productfile.html?productid=10319). And

some more electronics, at least a power supply some MEMS manufacturers offer class B accel-
unit, and optionally assembled in a casing with erometers, e.g., Silicon Designs 1221 (www.
connectors, able to be fixed to another element. silicondesigns.com/pdfs/1221.pdf) has typical
This assembly has to be rigid enough not to noise PSD of 5 mg/√Hz – an rms noise of 50 mg
introduce parasitic resonances in the seismic over 100 Hz bandwidth. Recent designs
band. Including electronics and housing, the (Homeijer et al. 2011) report noise levels under
total weight of these sensors might be around 100 ng/√Hz or even 10 ng/√Hz above 1 Hz
0.1–0.5 kg. The sensor chip itself typically (Milligan et al. 2011) with new capacitance trans-
weights less than 1 g. ducers operating without feedback. These perfor-
The (US) Working Group on Instrumentation, mances make it even suitable for use as
Siting, Installation, and Site Metadata of the exploration sensors, substituting geophones.
Advanced National Seismic System (ANSS Class C devices are cheaper and mostly used
2008) defined four classes of strong-motion in consumer products. Those models with stable
stations – A, B, C, and D – in terms of perfor- response and enough bandwidth may be suitable
mance. Class A has the highest performance, with to be applied in seismic strong ground motion
resolution better than 7 mg and broadband monitoring, for instance, instrumental intensity
dynamic range > 111 dB (20bit). Class B has estimation (shake maps), structural response,
a resolution between 7 and 107 mg and a dynamic earthquake early warning, and shut off of critical
range 87–111db (or 16–20bit). Class C resolution facilities.
is 107–1,709 mg and dynamic range 63–87 dB As an example of this class of sensors, Fig. 11
(12–16bit). Class D has poorer performance than shows a MEMS triaxial accelerometer LIS
C. Currently (2014) there are no commercial 344 ALH (www.st.com/web/en/catalog/
MEMS accelerometers that fulfill the class sense_power/FM89/SC444) from STMicroe-
A specifications, which are only met by classical lectronics mounted on a small printed circuit
macroscopic FBA devices. One of the reasons is board, which is sold as an evaluation board, since
the difficulty of achieving low noise levels with a hand soldering the chip for prototyping is quite a
so small suspended mass. difficult task. This chip accelerometer has a user-
Several manufacturers use MEMSs for selectable full-scale 2 g or 6 g and comes with
accelerograph recorders with class B perfor- factory-trimmed sensitivity and offset. Its band-
mance, e.g., GeoSIG GMS-18 (www.geosig. width may be selected by an external capacitor.
Fig. 11 A MEMS triaxial
accelerometer LIS
344 ALH from
STMicroelectronics
(center) mounted in a small
printed circuit with the
minimal external
components
Seismic Accelerometers, 8
CRT 04 jan 2007
acceleration (gal)
Fig. 12 Upper,
4 mw=3.8; ML=3.4
accelerogram obtained with
a class B accelerograph CRT A N
0
from an earthquake of
moment magnitude R=16km
−4
mw = 3.8 at an epicentral
distance of 16 km. Lower, −8
the simulated record with 8
the MEMS sensor of Fig. 11
acceleration (gal)
by adding the recorded self- 4

noise (MEM sensor)
0
−4
−8
0 10 20 30
time (s)
Figure 12 plots a small local earthquake recorder in a case, which is called an accelerograph.
recorded by a standard accelerometer and the Some of them may be acquired separately, with an
simulated record with this sensor, using its real analog voltage output for each axis. For earthquake
noise. The MEMS sensor shows a higher noise, recording, the standard is a triaxial instrument. S
but a useful signal is still available. Practically all the recorders have GPS synchroni-
A comparative test of these sensors (Evans zation of internal clock. Some models have the
et al. 2014) shows that some of them could be possibility of Internet timing, for sites where GPS
very useful for low-cost seismic networks. Some signal cannot be received.
models have performances suitable for strong Most accelerograph manufacturers offer sev-
ground motion recording with acceptable SNR eral models with different performance and price.
to be used as class C seismic accelerometers. The models shown below are just examples and
may not be representative of the best instrument
offered by each company.
Accelerometer Examples Figure 13 shows one of the accelerometers
with highest dynamic range, the EpiSensor
Many seismic accelerometers commercially avail- from Kinemetrics (www.kinemetrics.com/p-87-
able are sold assembled with the corresponding EpiSensor-ES-T.aspx), a triaxial FBA with
Fig. 13 The EpiSensor
accelerometer from
Kinemetrics. Left, surface
sensor package. Right, the
case is removed to show the
components
Seismic Accelerometers, Fig. 14 The Etna

accelerograph from Kinemetrics with its lid removed. It
has an internal triaxial accelerometer (EpiSensor). The
hermetic case is made of fiberglass
full-scale selectable up to 4 g. Also available is Seismic Accelerometers, Fig. 15 CMG 5TD from
G€uralp Systems. This model can include a recorder and
a uniaxial and with different packages, including
an Ethernet interface. The photo shows a unit installed in
borehole. It may be mounted inside a compact an underground gallery with the mounting base anchored
accelerograph package like Etna model (Fig. 14). to the concrete ground
This robust accelerograph model has been in the
market for many years, so, in spite of the good
accelerometer inside, its performance as recorder with high dynamic range that can be supplied
(dynamic range, communication interface, data with built-in digitizer, recorder, and communica-
storage) does not fully match the sensor specifi- tion facilities. It is a class A accelerograph.
cations, and it may be considered a class A borehole version is also available.
B accelerograph. The same manufacturer offers Figure 16 shows Nanometrics Titan (www.
new recorder models with higher performance. nanometrics.ca/products/titansma). It is one of
G€uralp Systems (www.guralp.com/products/ the newest class A accelerographs, with internal
instruments/cmg-5) CMG5 (Fig. 15) is an FBA FBA and high dynamic range in a very compact
Seismic Accelerometers, Fig. 16 Nanometrics Titan

accelerograph. Its size is approximately 18 x 12 x 10 cm
Seismic Accelerometers, Fig. 18 The REFTEK 148–1
“QuakeRock” accelerograph, with MEMS sensor (Photo
from REFTEK (www.reftek.com/products/motion-
recorders-148-01.htm))
is a class C accelerograph with only event record-

ing (not continuous) and limited dynamic range
but may operate unattended for 2 years with two
“D” size batteries. The internal clock is free run-
ning (no GPS).
Accelerograph Installation
Two kinds of accelerometer or accelerograph

Seismic Accelerometers, Fig. 17 FBA with MEMS installations are possible: (a) free-field installa-
technology AC-43 from GeoSIG. It weights 2 kg and its tion and (b) structural-monitoring installation.
size is 19 x 11 x 9 cm (Figure from GeoSIG (www.geosig. The purpose of the first is to record the ground
com/AC-4x-id10357.html))
S
motion unaffected by man-made structures. The
acceleration records obtained from this kind of
stations may be considered the base-level excita-
package. It has a web interface and can store data tion of any building or structure in the zone, if the
in a removable SD memory card. building itself does not interact with the ground at
Finally, two accelerometers with MEMS tech- the station. The second type is an accelerograph
nology will be shown. In Fig. 17, the GeoSIG in a building or engineering structure at different
AC-43 (www.geosig.com/AC-4x-id10357.html) levels and positions to study and monitor the
is a triaxial FBA with class B performance when structure vibrations in response to ground seismic
mounted inside or connected to a recorder, such as motions.
the GSR-18, from the same company. The model The main consideration is that the installation
148–01 “QuakeRock” from REFTEK (Fig. 18) setup should not affect the accelerograph records.
For free-field stations, it is usual to build a small synchronized via Ethernet or the accelerometer
concrete pier on which the accelerometer is installed on the site, and the recorder is sepa-
(or accelerograph if the sensor is inside it) must rately installed near the open sky so the GPS
be firmly anchored so a strong motion cannot antenna can be placed outside.
move the instrument relative to ground! Most
commercial instruments include a suitable base
with anchoring holes or a similar system. Other Summary
materials for pier, like a table made with steel
bars, are not suitable since it may resonate Seismic accelerometers sense the ground or
with very low damping in the seismic band of structure seismic vibrations and, together with a
interest. suitable recorder, are called accelerographs. Most
Free-field accelerographs should not be close modern seismic accelerometers are of force-
to tall buildings that may modify the ground balance type (FBA), a servo system in which a
motion in their proximity. Actually some feedback force is applied to the suspended iner-
accelerograph stations are installed at the base- tial mass to keep its motion as small as possible.
ment of buildings and are considered as free-field This improves the instrument linearity and
but may not truly fulfill this condition, due to the dynamic range. Usually the mass motion is mea-
soil-structure interaction. sured by a sensitive capacitive transducer.
If the free-field installation is to be done on MEMS accelerometers are integrated
soil, the pier should have a suitable foundation to micromachined electromechanical devices
assure that it is well coupled to the ground, but widely used in industry that presently do not
not being so heavy as to modify the local soil match the classical FBA performance, but are
dynamic response substantially. Free-field useful for some seismic applications.
accelerographs must be protected with a cover Examples of commercial seismic accelerome-
or small hut from the weather. Provisions have ters and accelerographs and some brief guidelines
to be made for power supply, GPS antenna, and for the installation of free-field and structure-
communication, usually Ethernet via cable, sat- monitoring accelerographs are given.
ellite link, or cellular modem. And a fence around
the installation would protect it from animals,
human-made noises, and eventually vandalism.
Accelerographs are not very sensitive to weak Cross-References
motions, but modern high-resolution instruments
are capable of detecting human activity such as ▶ Passive Seismometers
traffic or machinery working at short distance, so ▶ Principles of Broadband Seismometry
this kind of noise should be avoided as far as ▶ Recording Seismic Signals
possible. ▶ Seismic Network and Data Quality
Structural-monitoring installations are usually
done under cover, and additional weather protec- References
tion is not required, but the instrument may have
to be protected from human activity or other Aki K, Richards PG (2002) Quantitative seismology.
disturbances. Normally a pier is not required, University Science Books, Sausalito CA, 700 pp
Akkar S, Bommer JJ (2007) Empirical prediction equa-
since the sensor may be anchored directly to a tions for peak ground velocity derived from strong-
structural element. GPS reception for time syn- motion records from Europe and the Middle East. Bull
chronism may be a problem if the station is far Seismol Soc Am 97:511–530. doi:10.1785/
from the open sky (e.g., in a dam gallery), but 0120060141
Anderson JG (2010) Source and site characteristics of
there exist technical solutions: e.g., a GPS earthquakes that have caused exceptional ground
receiver may be outside, and the signal is accelerations and velocities. Bull Seismol Soc Am
“repeated”; the accelerograph may be 100:1–36
Seismic Actions Due to Near-Fault Ground Motion 2519
ANSS Working Group on Instrumentation, Siting, Instal-

lation, and Site Metadata of the Advanced National Seismic Actions Due to Near-Fault
Seismic System Technical Integration Committee
(2008) Instrumentation guidelines for the advanced Ground Motion
national seismic system, U.S. Geological Survey.
Open-File Report 2008–1262, 41 pp George P. Mavroeidis
Deng T, Chen D, Wang J, Chen J, He W (2013) A MEMS Department of Civil and Environmental
based electrochemical vibration sensor for seismic
motion monitoring. J Microelectromech Syst Engineering and Earth Sciences, University of
99:920–923 Notre Dame, Notre Dame, IN, USA
Dwyer PD (2011) MEMS accelerometer. US Patent
8–065.915 B2, 29 Nov 2011
Eisenberg A, McEvilly TV (1971) Comparison of some
widely-used strong-motion earthquake accelerome- Synonyms
ters. Bull Seismol Soc Am 61:383–397
Evans JR, Allen RM, Chung AI, Cochran ES, Guye R, Damping coefficient; Forward directivity; Near
Hellweg M, Lawrence JF (2014) Performance of field; Near source; Permanent translation (fling);
several low-cost accelerometers. Seismological
Research Letters 85:147–158 Response spectrum; Seismic ground excitation;
Havskov J, Alguacil G (2010) Instrumentation in earth- Strength reduction factor; Time history
quake seismology. Springer, Dordrecht, 358 pp
Homeijer B, Lazaroff D, Milligan D, Alley R,
Wu J, Szepesi M, Bicknell B (2011) Hewlett
Packard’s seismic grade MEMS accelerometer. In: Introduction
Proceedings of the MEMS, Jan 2011, Cancún, Mexico,
pp 585–588
Near-fault seismic ground motions are frequently
Li B, Lu D, Wang W (2001) Micromachined accelerom-
eter with area-changed capacitance. Mechatronics characterized by intense velocity and displace-
11:811–819 ment pulses of relatively long duration that
McComb HE, Ruge AC (1937) Tests of earthquake accel- clearly distinguish them from typical far-field
erometers on a shaking table. Bull Seismol Soc Am
ground motion records. This observation, along
27:325–329
Milligan DJ, Homeijer BD, Walmsley RG (2011) An with its engineering significance, was first made
ultra-low noise MEMS accelerometer for seismic with respect to the C02 record (Fig. 1a) generated
imaging. In: Sensors 2011 IEEE, October 2011, Lim- by the 1966 Parkfield earthquake at a distance of
erick, DOI 10.1109/ICSENS.2011.6127185,
only 80 m from the fault break (Housner and
pp 1281–1284
Reasemberg P (1969) Non-linearity test of an accelerom- Trifunac 1967). The damage that the Olive
eter system designed for the seismic near-field mea- View Hospital sustained during the 1971 San
surement. Bull Seism Soc Am 59:1213–1225 Fernando earthquake was also attributed to the
Ruge AC, McComb HE (1943) Tests of earthquake accel-
effect of near-fault ground motions on flexible
erometers on a shaking table. Bull Seismol Soc Am
33:2–12 structures (Bertero et al. 1978). That was perhaps
Trifunac MD (2009) 75th anniversary of strong motion the first time that earthquake engineers linked the S
observation-a historical review. Soil Dyn Earthq Eng structural damage caused by an earthquake to the
29:591–606
impulsive character of near-fault ground motions
Walmsley RG, Kiyama LK, Milligan DM, Alley RL,
Erickson DL, Hartwell, PG (2009) Micro-G silicon (Fig. 1b). However, it was not until the 1994
accelerometer using surface electrodes. In: IEEE sen- Northridge and the 1995 Kobe earthquakes that
sors 2009 conference, October 2009, Christchurch, the majority of engineers recognized the destruc-
DOI 10.1109/ICSENS.2009.5398201, pp 971–974
tive potential of near-fault ground motions and
Yamada M, Mori J, Heaton T (2009) The slapdown phase
in high-acceleration records of large earthquakes. started considering methods to incorporate near-
Seismol Res Lett 80:559–564. doi:10.1785/ source effects into engineering design. Code pro-
gssrl.80.4.559 visions have historically been developed based
Yanada T, Furukawa H, Ichida S, Takubo K, Shimizu Y,
on recorded ground motions not sufficiently close
Koganemaru K, Suzuki T (2002) Earthquake sensor.
Sens Update 11:159–216. DOI:10.1002/seup. to the causative fault. Thus, the effect of near-
200211101 fault pulse-like ground motions on the dynamic
2520 Seismic Actions Due to Near-Fault Ground Motion
Seismic Actions Due to Near-Fault Ground Motion, Pacoima Dam (PCD) record from the 1971 San Fernando,
Fig. 1 Characteristic examples of near-fault pulse-like California, earthquake (Reprinted from Mavroeidis and
ground motion records: (a) Station No. 2 (C02) record Papageorgiou (2003). Copyright # 2003 Seismological
from the 1966 Parkfield, California, earthquake and (b) Society of America)
response of engineering structures has received damping coefficients for engineering analysis
much attention over the past two decades. and design in the near-fault region. The material
This entry focuses on the description of seis- presented in this entry is primarily based on pre-
mic actions due to near-fault ground motions with vious articles published by the author and Profes-
particular emphasis on the following topics: sor Apostolos S. Papageorgiou and is presented in
(1) characteristics of near-fault ground motions, a manner that provides established knowledge in
(2) effect of fault rupture parameters on near- the disciplines of engineering seismology and
fault seismic excitations, (3) synthesis of near- earthquake engineering to technically inclined
fault ground motion time histories for earthquake and informed readers. It should be emphasized
engineering applications, and (4) derivation of that this entry does not intend to be a specialized
response spectra, strength reduction factors and research article advancing the current state of
Seismic Actions Due to Near-Fault Ground Motion, (Reprinted from Mavroeidis and Papageorgiou (2002).
Fig. 2 Characteristic examples of forward and backward Copyright # 2002 Earthquake Engineering Research
directivity from the 1992 Landers, California, earthquake Institute)
knowledge or a review article summarizing the faults. Figure 2 illustrates a characteristic exam-
vast amount of archived research literature on the ple of forward directivity from the 1992 Landers
subject. earthquake. The fault rupture propagated to the
north along the indicated strike-slip fault. The
fault-normal velocity and displacement time his-
Characteristics of Near-Fault Ground tories recorded at Lucerne Valley (LUC) station
Motions (which is located in the forward direction with
respect to the propagation of rupture) are charac-
Not all ground motion time histories recorded at terized by intense pulse-like motions. In contrast,
stations in the vicinity of a fault exhibit intense the ground motion recorded at Joshua Tree (JSH)
velocity pulses. The existence of pulse-like station (which is located in the backward direc-
ground motions in near-fault records primarily tion with respect to the propagation of rupture) is
depends on the relative position of the station relatively weak.
that recorded the motion with respect to the direc- Permanent translation (fling) is a consequence
tion of propagation of rupture on the causative of permanent fault displacement due to an earth-
fault plane and on the magnitude and direction of quake; it appears in the form of step displacement
slip on that segment of the fault that is located in and one-sided velocity pulse in the strike-parallel
the vicinity of the station. Whenever these ground direction for strike-slip faults or in the strike- S
motion pulses do occur, they are typically caused normal direction for dip-slip faults. In the latter
by the forward directivity and/or permanent case, directivity and permanent translation
translation (fling) effects. effects “build up” in the same direction. Figure 3
Forward directivity occurs when the fault rup- illustrates characteristic examples of permanent
ture propagates toward a site with a rupture translation (fling) from the 1999 Izmit earth-
velocity that is approximately equal to the shear quake. The fault-parallel velocity and displace-
wave velocity. In this case, most of the energy ment time histories recorded at Yarimca (YPT)
arrives coherently in a single, intense, relatively and Sakarya (SKR) stations are affected by the
long-period pulse at the beginning of the record permanent displacement along the right-lateral
representing the cumulative effect of almost all strike-slip North Anatolian Fault.
the seismic radiation from the fault. Forward Even though emphasis has traditionally been
directivity pulses are polarized in the fault- given to the investigation of forward directivity
normal direction for both strike-slip and dip-slip and permanent translation (fling) effects, other
Seismic Actions Due to Near-Fault Ground Motion, (Reprinted from Mavroeidis and Papageorgiou (2002).
Fig. 3 Characteristic examples of permanent translation Copyright # 2002 Earthquake Engineering Research
(fling) from the 1999 Izmit, Turkey, earthquake Institute)
conditions may also give rise to near-fault pulse- Isochrones are frequently used in seismology to
like motions. A comprehensive review of the provide intuitive insight into factors that strongly
factors that influence near-fault ground motions, influence the generation of strong ground
along with a detailed list of references on the motions. By plotting the S-wave isochrones on
subject, has been presented by Mavroeidis and the fault plane of the investigated seismic events,
Papageorgiou (2002, 2003). Figure 4 illustrates the long-period velocity pulses of the near-fault
a large number of actual near-fault ground ground motions can be directly associated with
motion records with “distinct” velocity pulses. specific regions and characteristics of the fault
These records are part of the near-fault ground rupture.
motion database compiled by Mavroeidis and The results indicated that the seismic energy
Papageorgiou (2003). It is evident that the pulse radiated from the high-isochrone-velocity region
duration (or period), the pulse amplitude, as well of the fault arrives at the receiver within a time
as the number and phase of half cycles are the key interval that coincides with the time window of
parameters that define the waveform characteris- the long-period ground motion pulse recorded at
tics of the near-fault velocity pulses. the site. Furthermore, the near-fault ground
motion pulses are strongly correlated with large
slip on the fault plane locally driven by high
Effect of Fault Rupture Parameters on stress drop. In addition, the local rupture velocity
Near-Fault Ground Motions seems to be inversely correlated to the spatial
distribution of the strength excess over the fault
The effect of fault rupture characteristics on near- plane confirming findings of previous studies
fault ground excitations has been investigated by (e.g., Bouchon 1997). As an example, Fig. 5 illus-
Mavroeidis and Papageorgiou (2010) using trates time histories of near-fault ground motions
a kinematic modeling approach. In order to asso- and S-wave isochrones for selected stations of the
ciate fault rupture characteristics (such as slip, 1979 Imperial Valley earthquake. These stations
rupture velocity, and state of stress) with near- are located close to the ruptured fault where the
fault ground motions, four well-documented seis- effect of forward directivity was pronounced.
mic events (1979 Imperial Valley, 1985 The spatial distribution of the static slip offset
Michoacan, 1989 Loma Prieta, and 1999 Izmit) and rupture time inferred by Archuleta (1984)
were considered along with the concept of iso- and the spatial distribution of the static stress
chrones. An isochrone is the locus of all those drop and strength excess calculated using the
points on the fault plane, the radiation of which methodology proposed by Bouchon (1997) are
arrives at a certain observer at a specified time. also illustrated.
Seismic Actions Due to Near-Fault Ground Motion, Fig. 4 Strong motion records with “distinct” velocity pulses
(Reprinted from Mavroeidis and Papageorgiou (2003). Copyright # 2003 Seismological Society of America)
Finally, it was found that for various events, encompasses individual subevents, and, conse- S
the area of the fault that contributes to the forma- quently, crack-like slip functions (rather than dis-
tion of the near-fault pulse encompasses more location models) may be more appropriate for the
than one patch of significant moment release simulation of the near-fault ground motions. The
(subevent) (e.g., 1979 Imperial Valley, 1989 interested reader may find a detailed discussion in
Loma Prieta). This observation explains why Mavroeidis and Papageorgiou (2010).
a dislocation model with average properties
(i.e., slip, rise time, etc.) reproduces successfully
near-fault ground motions for strike-slip faults Time Histories of Near-Fault Ground
and for dip-slip faults with intermediate-to-large Motions
earthquake magnitudes. However, for very large
earthquakes, such as megathrust events on sub- The advent of performance-based earthquake
duction zones (e.g., 1985 Michoacan), the fault engineering, the growth of computer processing
region that contributes to the pulse formation power, and the associated increased availability
Seismic Actions Due to Near-Fault Ground Motion, slip offset, rupture time, static stress drop, and strength
Fig. 5 Recorded (black trace) and synthetic (gray trace) excess are also illustrated (Reprinted from Mavroeidis and
near-fault ground motion time histories and S-wave iso- Papageorgiou (2010). Copyright # 2010 Seismological
chrones for selected stations of the 1979 Imperial Valley, Society of America)
California, earthquake. Tomographic images of the static
of structural analysis software have made possi- histories, both in the far-field and near-fault
ble the performance of sophisticated nonlinear regions (Halldorsson et al. 2011).
structural analysis on a routine basis. However,
the overall seismic performance-based assess- Synthetic Near-Fault Ground Motions Using
ment of a given structure hinges on the use of Physical Models of the Seismic Source
realistic earthquake ground motions that reflect Strong motion seismologists have utilized vari-
the seismic hazard at the site of the structure, as ous schemes of deterministic and stochastic sim-
well as the local site conditions (Halldorsson ulation techniques to generate broadband ground
et al. 2011). motion time histories at specific locations in the
In general, earthquake engineers have the vicinity of the fault (see, e.g., Papageorgiou 1997
following options for selecting ground motion and references provided therein). These simula-
input when performing nonlinear structural tion methods are based on source mechanics prin-
analysis in the near-fault region: (1) use actual ciples and wave propagation theory. Site effects
records of near-fault ground motion, (2) gener- are also frequently taken into account. Regardless
ate synthetic records of near-fault ground of the degree of sophistication of the various
motion using physical models of the seismic ground motion simulation methods, the Earth
source, and (3) generate synthetic records of crustal structure and the seismic source should
near-fault ground motion using phenomeno- sufficiently be characterized and quantified.
logical models. For regions of intense seismic activity, the
crustal structure is frequently defined in terms of
Recorded Near-Fault Ground Motions one-dimensional velocity models. Detailed three-
The gradually increasing number of recorded dimensional crustal models have also become
near-fault ground motions has recently enabled available for specific regions in the benefit of
strong motion seismologists to compile these three-dimensional wave propagation codes that
records in publicly available ground motion data- may effectively take into account basin effects
bases (e.g., Pacific Earthquake Engineering and complex fault geometries at the cost of
Research Center Ground Motion Database, Cen- increased computational demands. The charac-
ter for Engineering Strong Motion Data, Euro- terization of the seismic source is a more compli-
pean Strong Motion Database, among others). cated issue. For kinematic descriptions of the
Even though the number of near-fault records is earthquake source, source parameters such as
still limited, they have served as an invaluable slip, rise time, rupture velocity, and slip function
resource to earthquake engineers. Researchers should properly be quantified and a priori
have also proposed methodologies for identifying defined. On the other hand, for dynamic descrip-
and extracting pulse-like motions from actual tions of the earthquake source, the source param-
near-fault records using wavelet analysis (e.g., eters may vary as long as the elastodynamics S
Baker 2007; Vassiliou and Makris 2011). equation with a prescribed fracture criterion on
However, the selection of strong motion a predetermined fault plane is satisfied. The
records for nonlinear structural analysis is not selected initial conditions and failure criterion
always a straightforward process. For instance, determine the time and space evolution of the
the available records may not reflect the appro- fault rupture in a dynamic source model.
priate earthquake magnitude, source mechanism, Once the seismic source and Earth crustal
site conditions, or source-site configuration. model have been adequately described, near-
While this could be a problem for far-field sites, fault ground motion simulations in the
it is even a greater challenge for sites in the low-frequency range (e.g., below 1 Hz) can be
immediate vicinity of the fault. For the above performed using deterministic modeling tech-
reasons, it is of paramount importance to earth- niques [e.g., discrete wavenumber method
quake engineers to have the ability to generate (DWN), finite difference method (FDM), finite
suites of realistic broadband ground motion time element method (FEM), boundary element
method (BEM), spectral element method (SEM), interpretation and scale, to the extent possible,
or hybrids of them] that involve calculations of with physical parameters of the faulting process.
synthetic Green’s functions. In order to generate
broadband synthetic ground motions, the Mathematical Representation of Near-Fault
low-frequency waveforms should be combined Ground Motion Pulses Proposed by Mavroeidis
with high-frequency ground motions (e.g., and Papageorgiou (2003)
above 1 Hz) simulated using: (1) the empirical The mathematical formulation for the represen-
or semiempirical Green’s function method or tation of the near-fault ground velocity pulses
(2) a stochastic modeling technique utilizing proposed by Mavroeidis and Papageorgiou
a source model that provides an unambiguous (2003) is the product of a harmonic oscillation
way to distribute the seismic moment of the sim- with a bell-shaped function. That is:
ulated event on the fault plane. This matter is of
8

great importance for near-fault ground motion >
> 1 2pf P
ðtt0 Þ cos ½2pf P ðtt0 Þþn,
<A 1þ cos
simulations due to the proximity of the point of 2 g
vðtÞ¼
>
> g g
observation to the source. It should be mentioned : t0 t t0 þ with g>1 0, otherwise
2f P 2f P
that high-frequency ground motion simulations
can be carried out using synthetic Green’s func- (1)
tions as well, excluding site effects and small-
where A controls the amplitude of the signal, fP is
scale heterogeneities.
the frequency of the amplitude-modulated har-
monic (or the prevailing frequency of the signal),
Synthetic Near-Fault Ground Motions Using
n is the phase of the amplitude-modulated har-
Phenomenological Models
monic, g is a parameter that defines the oscillatory
Ground motion simulation techniques based on
character of the signal, and t0 specifies the epoch
kinematic or dynamic source models are not
of the envelope’s peak. The pulse period (TP) is
always appealing to earthquake engineers
defined as the inverse of the prevailing frequency
because specialized seismological knowledge
(fP) of the signal, thus providing an “objective”
and, quite frequently, demanding computational
assessment of this important parameter. That is:
resources are required. Therefore, in practice,
earthquake engineers utilize actual near-fault
records of past earthquakes to investigate the 1
TP ¼ (2)
dynamic response of engineering structures to fP
pulse-like seismic excitations and rely on strong
motion seismologists only for generating site- The model input parameters have a clear physical
specific near-fault ground motions for the design meaning as they coincide with the key features
of special structures. that determine the waveform characteristics of
To overcome this deficiency, earthquake engi- the near-fault pulses (i.e., amplitude, duration,
neers have introduced idealized waveforms, phase and number of half cycles).
intending to represent typical ground motion The mathematical model proposed by
pulses observed in the near-fault region, in an Mavroeidis and Papageorgiou (2003) was cali-
effort to investigate the dynamic response of engi- brated using a large number of actual near-fault
neering structures to near-fault ground motions records. It successfully replicated a large set of
(e.g., Makris 1997; Sasani and Bertero 2000; displacement, velocity, and, in many cases, accel-
Alavi and Krawinkler 2001; Mavroeidis and eration time histories, as well as the corresponding
Papageorgiou 2003, among others). These ideal- elastic response spectra. A sample of the quality of
ized waveforms should successfully capture the fitting of the synthetic waveforms to actual near-
impulsive character of the near-fault records both fault records is illustrated in Fig. 6.
qualitatively and quantitatively. In addition, their The scaling characteristics of the model input
input parameters should have a clear physical parameters were also investigated through
Seismic Actions Due to Near-Fault Ground Motion, as well as the 5% damped elastic response spectra are
Fig. 6 Sample of synthetic waveforms (black trace) fitted illustrated (Reprinted from Mavroeidis et al. (2004).
to actual near-fault records (gray trace). Ground motion Copyright # 2004 John Wiley & Sons, Inc.)
time histories (displacement, velocity, and acceleration),
regression analysis, and simple empirical rela- motion records affected by forward directivity.
tionships were proposed. By performing Similar scaling equations have been proposed by
least-squares analysis (Fig. 7a), the following other investigators (e.g., Somerville 2003; Bray S
relationship was obtained between the pulse and Rodriguez-Marek 2004; Baker 2007, among
period and the earthquake magnitude: others). However, it should be pointed out that the
definition of the pulse period in these studies is not
logT P ¼ 2:2 þ 0:4 Mw (3a) the same as the definition provided by Mavroeidis
and Papageorgiou (2003), and therefore caution
If the empirical relationship is required to satisfy should be utilized when the mathematical model
the self-similarity condition, the following equa- of Eq. 1 is used in conjunction with scaling laws for
tion can be obtained: the pulse period proposed by other investigators.
Mavroeidis et al. (2004) also derived an equa-
logT P ¼ 2:9 þ 0:5 Mw (3b) tion that relates the pulse period (TP) with the rise
time (t) (i.e., the time it takes for a representative
Equation (3) was derived by Mavroeidis and point on the fault plane to reach its final displace-
Papageorgiou (2003) using near-fault ground ment). The rise time is an important physical
Seismic Actions Due to Near-Fault Ground Motion, rupture distance, according to Halldorsson et al. (2011)
Fig. 7 (a) Scaling of the pulse period with earthquake (Reprinted from Halldorsson et al. (2011). Copyright #
magnitude, according to Mavroeidis and Papageorgiou 2011 American Society of Civil Engineers)
(2003), and (b) attenuation of peak ground velocity with
parameter of the fault rupture process that greatly synthetic velocity pulse can be defined by con-
affects strong ground motion characteristics. In sidering that A ~ (0.85–1.00) PGV. Other inves-
fact, the rise time (and therefore the pulse period) tigators have also proposed attenuation
is related to one of the characteristic corner fre- relationships for PGV (e.g., Bray and
quencies of the source spectrum (i.e., the spec- Rodriguez-Marek 2004, among others).
trum of seismic waves radiated by the earthquake Finally, parameter g varies from a value slightly
source before these are modified by propagation larger than 1 up to a maximum value of 3, while
path and site effects). the phase angle (n) varies from 0 to 360 .
On the other hand, the peak value of the near- Halldorsson et al. (2011) have provided the prob-
fault velocity records appears to be a fairly stable ability density functions of g and n, assuming that
parameter. A value of 100 cm/s effectively repre- these two parameters are normally distributed.
sents peak ground velocities within a few kilome-
ters from the causative fault regardless of the Simplified Methodology for Generating
earthquake magnitude (Mavroeidis and Broadband Near-Fault Ground Motions Proposed
Papageorgiou 2003). This observation is in good by Mavroeidis and Papageorgiou (2003)
agreement with the typical slip velocity value of As explained by Mavroeidis and Papageorgiou
100 cm/s frequently considered by seismologists. (2003), the mathematical expression of Eq. 1 rep-
As indicated by Mavroeidis et al. (2004), there are licates accurately the intermediate-to-long-
solid physical reasons that explain the stability of period (“coherent”) features of the near-fault
the velocity amplitude close to the fault. More ground motions. The high-frequency components
recently, Halldorsson et al. (2011) proposed the that are “incoherent” cannot be simulated using
following attenuation relationship for peak ground simplified mathematical models (see, e.g., accel-
velocity (PGV) with rupture distance (R) (Fig. 7b): eration time histories and the short-period range
of response spectra in Fig. 6).
logPGV ¼ 2:040 0:032 R (4) A simplified methodology for generating real-
istic, broadband, near-fault ground motions that are
Once PGV has been determined using Eq. 4, adequate for engineering analysis and design was
parameter A that controls the amplitude of the proposed by Mavroeidis and Papageorgiou (2003).
Based on this technique, the coherent (long-period) Eq. 1 to the ground motion time histories and
ground motion component is simulated using the response spectra of the PCD record (see
mathematical model of Eq. 1, while the incoherent Mavroeidis and Papageorgiou 2003). For the syn-
(high-frequency) seismic radiation is synthesized thesis of the incoherent seismic radiation at the
using the specific barrier model (Papageorgiou and location of the Olive View Hospital, the specific
Aki 1983a, b) in the context of the stochastic barrier model of Papageorgiou and Aki (1983a) is
modeling approach. The specific barrier model is utilized. The selected parameters for the specific
a physical model of the seismic source that applies barrier model are consistent with the values
both to the “near-fault” and “far-field” regions, inferred by Papageorgiou and Aki (1983b) for
allowing for consistent ground motion simulations the 1971 San Fernando earthquake. The model
over the entire frequency range and for all distances consists of two subevents (Fig. 8c), in agreement
of engineering interest. The specific barrier model with the two distinct slip patches of similar size
has been calibrated to shallow crustal earthquakes inferred for this event by Heaton (1982) (Fig. 8b).
of three different tectonic regions: interplate, intra- The site characterization at the Olive View Hos-
plate, and extensional regimes (Halldorsson and pital is assumed to be NEHRP site class
Papageorgiou 2005). D consistent with available information.
This simplified methodology has been applied Figure 9a illustrates the synthetic ground
to hypothetical and actual earthquakes (e.g., motions (strike-normal component) at the Olive
Mavroeidis and Papageorgiou 2003; Halldorsson View Hospital. The top and middle panels dis-
et al. 2011) and is demonstrated in this entry play the incoherent and coherent ground motion
through the case study of the 1971 San Fernando components, while the bottom panels show the
earthquake (Mavroeidis 2004). According to superposition of the above two components. For
Heaton (1982), the 1971 San Fernando earth- comparison purposes, the ground motion
quake with MW 6.6 may have been a double seis- recorded at the nearby PCD station due to the
mic event that occurred on two subparallel thrust 1971 San Fernando earthquake is also shown in
faults, the Sierra Madre and San Fernando Faults, Fig. 9b. The overall agreement between the syn-
as indicated in Fig. 8a. The slip distribution on the thetic ground motions at the Olive View Hospital
causative faults, inferred by inversion of and the recorded ground motions at PCD station is
teleseismic, strong motion, and geodetic data, is very good. It is evident that acceleration amplitudes
illustrated in Fig. 8b. The damage that the Olive larger than those recorded at the PCD station char-
View Hospital sustained during the earthquake acterize the synthetic accelerogram at the Olive
has been attributed to the destructive potential View Hospital. On the other hand, the
of near-fault ground motions on flexible struc- corresponding velocity and displacement time his-
tures (Bertero et al. 1978). No strong motion tories are very similar. These differences in accel-
instruments were installed in the immediate eration amplitudes may be attributed to the different S
vicinity of the hospital building. However, there site conditions at the locations of the Olive View
are indications that the ground motion that the Hospital (NEHRP site class D) and Pacoima Dam
Olive View Hospital sustained was equivalent or (NEHRP site class B; rock good enough to serve as
greater than the ground motion recorded at the the foundation of a concrete dam).
nearby Pacoima Dam (PCD) station.
Broadband synthetic time histories are gener-
ated at the location of the Olive View Hospital for Response Spectra, Strength Reduction
the fault-station geometry of Fig. 8a. The mathe- Factors, and Damping Coefficients for
matical model of Eq. 1 is first employed to gen- Near-Fault Ground Motions
erate the coherent component of the ground
motion at the Olive View Hospital. The values In this section, the primary characteristics of
for the input parameters A, fP, g, and n are those near-fault ground motion response spectra are
inferred by fitting the mathematical model of discussed, and recommendations are made for
Seismic Actions Due to Near-Fault Ground Motion, subparallel thrust faults of Fig. 8a (Heaton 1982), and (c)
Fig. 8 (a) Cross-sectional view of the causative faults of subevents of the specific barrier model represented by two
the 1971 San Fernando, California, earthquake (Heaton circular cracks (o2–model) (Reprinted from Mavroeidis
1982), (b) slip distribution in meters for the two (2004). Copyright # 2004 G. P. Mavroeidis)
a Synthetic Ground Motions (High-Frequency)

2000 150
60 103
1500
Acceleration (cm/s2)
100 40
Displacement (cm)
1000 102
Velocity (cm/s)
50
PSV (cm/s)
500 20
1764 cm/s2 119.4 cm/s 65.8 cm
0 0 0 101
−500 −20
−50
−1000 −40 100
−100
−1500 −60
−2000 −150 10−1
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20 10−2 10
0
10
2
Time (s) Time (s) Time (s) T (s)
Synthetic Ground Motions (Long-Period)

2000 150
60 103
1500
100 40
Displacement (cm)
1000 102
Velocity (cm/s)
50
PSV (cm/s)
500 20
445.2 cm/s2 115.0 cm/s 33.2 cm
0 0 0 101
−500 −20
−50
−1000 −40 100
−100
−1500 −60
−2000 −150 10−1
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20 10−2 100 102
Synthetic Ground Motions (Combined)

2000 150
60 103
1500
100 40
Displacement (cm)
1000 102
Velocity (cm/s)
PSV (cm/s)
500 50 20 33.2 cm
1764 cm/s2 124.4 cm/s
0 0 0 101
−500 −20
−50
−1000 −40 100
−100
−1500 −60
−2000 −150 10−1
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20 10−2 100 102
b Recorded Ground Motions at PCD Station
2000 150 3
60 10
1500
100
40
Displacement (cm)
1000 102
Velocity (cm/s)
50
PSV (cm/s)
500 2
20
1266 cm/s 120.0 cm/s 31.1 cm
0 0 0 101
−500 −20
−50
−1000 −40 100
−100
−1500 −60
−2000 −150 10−1 S
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20 10−2 100 102
Seismic Actions Due to Near-Fault Ground Motion, response spectra are also shown. (b) Actual ground
Fig. 9 (a) Synthesis of near-fault ground motions at the motions recorded at the nearby Pacoima Dam (PCD)
location of the Olive View Hospital for the fault-station station (Reprinted from Mavroeidis (2004). Copyright #
geometry illustrated in Fig. 8a; the 5% damped elastic 2004 G. P. Mavroeidis)
design spectra, strength reduction factors and Response Spectra

damping coefficients for analysis, and design in Figure 10a displays the 5% damped equal-
the near-fault region. The interested reader may ductility pseudo-velocity response spectra of
find additional information in Mavroeidis elastic-perfectly plastic single-degree-of-
et al. (2004) and Hubbard and Mavroeidis (2011). freedom (SDOF) systems subjected to a large
a 103
b 103 c 102
102 102 101

PSV (cm/s)
PSV (cm/s)
101 101 100
PSV / A
μ= 1.0
100 100 10−1
10−1 10−1 10−2
10−2 10−2 −3 10−3

10−2 10−1 100 101 102 10 10−2 10−1 100 101 102 10−3 10−2 10−1 100 101 102
3 3 2
10 10 10
102 102 101

PSVy (cm/s)
PSVy (cm/s)
101 101 100
PSVy / A
μ= 4.0
100 100 10−1
10−1 10−1 10−2
10−2 10−2 −3 10−3

10−2 10−1 100 101 102 10 10−2 10−1 100 101 102 10−3 10−2 10−1 100 101 102
2
103 103 10
102 102 101

PSVy (cm/s)
PSVy (cm/s)
101 101 100

μ= 8.0
PSVy / A
100 100 10−1
10−1 10−1 10−2
10−2 10−2 −3 10−3

10−2 10−1 100 101 102 10 10−2 10−1 100 101 102 10−3 10−2 10−1 100 101 102
Tn (S) Tn / T P Tn / T P
Seismic Actions Due to Near-Fault Ground Motion, subjected to actual near-fault ground motion records: (a)
Fig. 10 Standard and normalized 5% damped equal- PSV versus Tn, (b) PSV versus Tn/TP, and (c) PSV/A versus
ductility (m = 1.0, 4.0, 8.0) pseudo-velocity response Tn/TP (Reprinted from Mavroeidis et al. (2004). Copyright
spectra of elastic-perfectly plastic SDOF systems # 2004 John Wiley & Sons, Inc.)
number of actual near-fault ground motion Mavroeidis and Papageorgiou (2003). This
records (Mavroeidis et al. 2004). Inspection of abscissa normalization yields response spectra
this figure reveals that peak spectral amplitudes characterized by peak spectral amplitudes that
of near-fault records vary significantly, espe- lie within a very narrow range of the normalized
cially for smaller values of the ductility factor period (i.e., Tpeak/TP
0.7–1.0 for elastic spec-
(m). Furthermore, the periods that correspond to tra). If the ordinates of the equal-ductility pseudo-
peak spectral amplitudes are characterized by velocity response spectra are further normalized
significant dispersion. with respect to A, the normalized response spec-
Figure 10b illustrates the equal-ductility tra of Fig. 10c are obtained; these spectra are
pseudo-velocity response spectra of Fig. 10a characterized by small dispersion of normalized
with the period axis normalized with respect to peak spectral amplitudes. In addition, they
the corresponding TP values estimated by exhibit smaller dispersion in the normalized
a 2
b 2
c 2
d 2
10 Mw 5.6 - 7.6
10 Mw 5.6 - 6.3
10 Mw 6.4 - 6.7
10 Mw 6.8 - 7.6
1 1 1 1
10 10 10 10
PSV /PGV
100 100 100 100

μ = 1.0
10−1 10−1 10−1 10−1
10−2 10−2 Mean

10−2 Mean
10−2 Mean
Mean
Mean+1SD Mean+1SD Mean+1SD Mean+1SD
−3
10 −2 0 2 10
−3
−2 0 2 10
−3
−2 0 2
10−3 −2 0 2
10 10 10 10 10 10 10 10 10 10 10 10
2 2 2 2
10 10 10 10
1 1 1 1
10 10 10 10
PSVy /PGV
0 0 0 0
μ = 4.0
10 10 10 10
−1 −1 −1 −1
10 10 10 10
−2 −2 −2 −2
10 10 10 10
−3 −3 −3 −3
10 −2 0 2 10 −2 0 2 10 −2 0 2 10 −2 0 2
10 10 10 10 10 10 10 10 10 10 10 10
2 2 2 2
10 10 10 10
1 1 1 1
10 10 10 10
PSVy /PGV
0 0 0 0
μ = 8.0
10 10 10 10
−1 −1 −1 −1
10 10 10 10
−2 −2 −2 −2
10 10 10 10
−3 −3 −3 −3
10 −2 0 2 10 −2 0 2 10 −2 0 2 10 −2 0 2
10 10 10 10 10 10 10 10 10 10 10 10
Tn / TP Tn / TP Tn / TP Tn / TP
Seismic Actions Due to Near-Fault Ground Motion, (MW 6.8–7.6). The solid and dashed lines represent the
Fig. 11 Normalized 5% damped equal-ductility pseudo- mean and mean-plus-one-standard-deviation pseudo-
velocity response spectra of elastic-perfectly plastic velocity response spectra, respectively. The gray region
SDOF systems: (a) all earthquakes (MW 5.6–7.6), (b) represents the range of variation of the spectral amplitudes
moderate earthquakes (MW 5.6–6.3), (c) moderate-to- (Reprinted from Mavroeidis et al. (2004). Copyright #
large earthquakes (MW 6.4–6.7), and (d) large earthquakes 2004 John Wiley & Sons, Inc.)
long-period range and larger dispersion in the In order to investigate the effect of the earth-
normalized high-frequency range (controlled by quake magnitude on the normalized response
the coherent and incoherent ground motion com- spectra of Fig. 11a, the seismic events have
ponents, respectively). been grouped into three categories: moderate S
Thus, it may be concluded that parameters (MW 5.6–6.3), moderate-to-large (MW 6.4–6.7),
A and TP can be used to normalize the response and large (MW 6.8–7.6) earthquakes. The normal-
spectra of actual near-fault records. This devel- ized equal-ductility pseudo-velocity response
opment facilitates the systematic investigation of spectra of these three categories are displayed in
the response spectrum characteristics of the Fig. 11b, c, d, respectively. It becomes evident
SDOF system subjected to near-fault ground that, for smaller values of Tn/TP, the normalized
motions. It should be noted that the normalization spectral amplitudes increase with earthquake
of the ordinates of the equal-ductility pseudo- magnitude. However, for larger Tn/TP values,
velocity response spectra of Fig. 10a with respect the normalized response spectra appear to exhibit
to PGV values yields normalized response spec- a uniform behavior regardless of the variation in
tra very similar to those illustrated in Fig. 10c. earthquake magnitude. As a consequence, the
This is anticipated because parameter normalized response spectra of large earthquakes
A effectively approximates PGV. exhibit flatter shapes around their peaks than the
Seismic Actions Due to Near-Fault Ground Motion, near-fault ground motion records (Reprinted from
Fig. 12 Schematic illustrations of idealized response Mavroeidis et al. (2004). Copyright # 2004 John Wiley
spectra in four-way logarithmic plots for far-field and & Sons, Inc.)
normalized response spectra of moderate [(Tn/TP)a, (Tn/TP)b, (Tn/TP)c, (Tn/TP)d, and

earthquakes. (Tn/TP)f] are provided in Mavroeidis et al. (2004).
The ensemble of the normalized elastic
response spectra illustrated in the first panel of Strength Reduction Factors
Fig. 11 can be utilized to derive normalized elastic The earliest and perhaps the simplest recommen-
design spectra for moderate, moderate-to-large, dation of a procedure to construct inelastic spec-
and large earthquakes, as well as for the entire tra from elastic spectra using ductility-dependent
set of seismic events considered by Mavroeidis strength reduction factors (Ry) is based on the
et al. (2004). The solid and dashed lines in the work of Veletsos and Newmark (1960). These
top panel of Fig. 11 represent the mean and results were further developed by Newmark and
mean-plus-one-standard-deviation 5% damped Hall (1982) based on a suite of far-field ground
normalized elastic response spectra. These aver- motion records. Mavroeidis et al. (2004) checked
age elastic response spectra can be used to derive the validity of the reduction factors proposed by
normalized elastic design spectra for two different Newmark and Hall (1982) for the response spec-
nonexceedance probability levels. tra of near-fault ground motion records by nor-
Figure 12a displays a sketch of the standard malizing the period intervals of the Ry design
idealized elastic design spectrum derived from equations as follows:
far-field ground motion records. This standard 8

>
> Tn Tn
design spectrum has been used for many decades >
> 1, <
>
> T T P a

since it was first introduced in engineering >
<
P
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi Tn Tn
practice. The acceleration-, velocity-, and Ry ¼ 2m 1, TP b < T < T
Tn
>
> P c0
displacement-sensitive regions of this design >
>
P
>
> T T
spectrum can readily be identified in Fig. 12a. >
: m,
n
>
n
On the other hand, the normalized response spec- TP TP c
tra of the near-fault ground motion records (see (5)
Fig. 11) can be approximated by the sequence of pffiffiffiffiffiffiffiffi
linear segments displayed in Fig. 12b. The values where TTPn 0 ¼ 2m1 m
Tn
T P . The characteristic
c c
of the characteristic normalized periods values of (Tn/TP)a, (Tn/TP)b, and (Tn/TP)c are
associated with the normalized elastic design The results indicated that damping coefficients
spectrum for near-fault ground motions (see proposed in design codes and previous studies,
Fig. 12b) proposed by Mavroeidis et al. (2004). based primarily on far-field ground motion
Figure 13 compares the computed values of Ry records, tend to not be conservative for near-
from the mean elastic and inelastic 5% damped fault pulse-like seismic excitations. Figure 14a
normalized response spectra of Fig. 11 with the illustrates the relationships between damping
Ry values obtained from Eq. 5. The agreement coefficient and period that were established for
between the two sets of curves is very good over viscous damping ratios in the range of 5–100%.
the entire period range, for all specified ductility These damping coefficient curves were generated
factors, and for all earthquake magnitude catego- using the definition of B factors and the median
ries. Figure 13 demonstrates that the Veletsos- pseudo-acceleration response spectra for various
Newmark-Hall design equations can be used for levels of damping.
near-fault ground motions as well, provided that In order to investigate the effect of earthquake
normalized response spectra are utilized and magnitude on damping coefficients using near-
appropriate values of (Tn/TP)a, (Tn/TP)b, and fault records, the seismic events were again
(Tn/TP)c are selected. grouped into three categories labeled as moderate
(MW 5.6–6.3), moderate-to-large (MW 6.4–6.7),
Damping Coefficients and large (MW 6.8–7.6) earthquakes. The varia-
Damping coefficients are frequently used in tion of damping coefficient with period for these
earthquake engineering as a simple way to adjust three categories is displayed in Figs. 14b, c, d for
the pseudo-acceleration or displacement viscous damping ratios in the range of 5–100%.
response spectra associated with a viscous While the B factors for all three earthquake mag-
damping ratio of 5% to the higher values of nitude categories attain approximately the same
viscous damping needed for design of structures peak values for a given damping ratio, the period
equipped with base isolation and/or supplemental range over which these peak values occur clearly
energy dissipation devices. Damping coefficients depends on earthquake magnitude. In addition,
are also frequently used for predicting the maxi- the damping coefficient curves of Fig. 14a
mum displacement demands of an inelastic struc- derived from the entire ground motion ensemble
ture from the maximum displacement demands of smooth out the effect of earthquake magnitude
its equivalent linear system characterized by and therefore do not capture the particular fea-
a longer natural period and a higher viscous tures of the damping coefficient plots illustrated
damping ratio. in Figs. 14b, c, d.
Damping coefficients (B) are defined as Figure 15 indicates that the normalization of
B(T,b) = PSA(T,b = 5%)/PSA(T,b), where T is the period axis of the B plots with respect to TP
the elastic period of vibration of the structure, b yields damping coefficient curves that show S
is the viscous damping ratio, and PSA are the a much stronger resemblance to each other.
ordinates of the pseudo-acceleration response More specifically, the normalized damping coef-
spectra for particular values of T and b. Damping ficient curves for all groups of records attain
coefficients (B) are also known as “damping comparable peak values for a given damping
adjustment factors.” The reciprocal of B is often ratio. These maximum values tend to be slightly
used in the literature and referred to as “damping closer than the peaks observed on the
correction factor,” “damping reduction factor,” non-normalized damping coefficient curves
“spectral scaling factor,” or “damping modifica- displayed in Fig. 14. In addition, the normalized
tion factor.” periods over which these peak values occur coin-
Hubbard and Mavroeidis (2011) calculated cide at a value slightly lower than 1.0 on the
damping coefficients for the SDOF system normalized period axis, a statistic that had previ-
subjected to near-fault pulse-like ground motions ously varied greatly by earthquake magnitude.
for a large range of periods and damping levels. The B curves illustrated in Fig. 15 may allow
2536
a b c d
1 1 1 1
10 Mean Mw 5.6 - 7.6 10 Mean Mw 5.6 - 6.3 10 Mean Mw 6.4 - 6.7 10 Mean Mw 6.8 - 7.6
Newmark -Hall Newmark -Hall Newmark -Hall Newmark -Hall
Ry
μ = 2.0
0 0 0 0
10 10 10 10
−3 −2 −1 0 1 2 −3 −2 −1 0 1 2 −3 −2 −1 0 1 2 −3 −2 −1 0 1 2
10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10
1 1 1 1
10 10 10 10
Ry
μ = 4.0
0 0 0 0
10 10 10 10
−3 −2 −1 0 1 2 −2 −1 0 1 2 −3 −2 −1 0 1 2 −2 −1 0 1 2
10 10 10 10 10 10 10−3 10 10 10 10 10 10 10 10 10 10 10 10−3 10 10 10 10 10
1 1 1 1
10 10 10 10
Ry
μ = 8.0
0 0 0 0
10 10 10 10
−3 −2 −1 0 1 2 −3 −2 −1 0 1 2 −3 −2 −1 0 1 2 −3 −2 −1 0 1 2
10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10
Tn / TP Tn / TP Tn / TP Tn / TP
Seismic Actions Due to Near-Fault Ground Motion, Fig. 13 Comparison of mean (c) moderate-to-large earthquakes (MW 6.4–6.7), and (d) large earthquakes (MW
values of the strength reduction factor with the Veletsos-Newmark-Hall design equa- 6.8–7.6); m = 2.0, 4.0, 8.0, and z = 5% (Reprinted from Mavroeidis et al. (2004).
tions for: (a) all earthquakes (MW 5.6–7.6), (b) moderate earthquakes (MW 5.6–6.3), Copyright # 2004 John Wiley & Sons, Inc.)
Seismic Actions Due to Near-Fault Ground Motion
Seismic Actions Due to Near-Fault Ground Motion, moderate-to-large earthquakes (MW 6.4–6.7), and (d)
Fig. 14 Calculated damping coefficients for near-fault large earthquakes (MW 6.8–7.6) (Reprinted from Hubbard
ground motion records: (a) all earthquakes (MW and Mavroeidis (2011). Copyright # 2011 Elsevier B.V.)
5.6–7.6), (b) moderate earthquakes (MW 5.6–6.3), (c)
S
for a single set of empirical equations to represent damping ratio. Equation 6 is designed to be rep-
near-fault damping coefficients that are normal- resentative of the damping coefficients at normal-
ized by TP. ized periods greater than ~0.83, a normalized
An empirical equation was developed by Hub- period where the peak in value seems to occur.
bard and Mavroeidis (2011) to fit the main behav- Below this characteristic normalized period, the
ior of damping coefficients observed in the set damping coefficients linearly reduce to one at
containing all records (Fig. 15a). In order to a normalized period of zero. Damping ratios are
effectively model these B curves, two equations represented as fractions instead of whole num-
were needed to describe different ranges of the bers within the context of Eq. 6:
Seismic Actions Due to Near-Fault Ground Motion, earthquakes (MW 5.6–6.3), (c) moderate-to-large earth-
Fig. 15 Damping coefficients for near-fault ground quakes (MW 6.4–6.7), and (d) large earthquakes (MW
motion records with the period axis normalized with TP: 6.8–7.6) (Reprinted from Hubbard and Mavroeidis
(a) all earthquakes (MW 5.6–7.6), (b) moderate (2011). Copyright # 2011 Elsevier B.V.)
b1:3 comparison to the computed damping coeffi-

B ¼ 3:4 1:3
þ 1 for 0:10 b 0:50 cients for the set containing all records. For the
ðT=T P Þ
considered ranges of damping ratio and normal-
(6a)
ized period, Eq. 6 provides a model that is con-
servative without exception. It also does an
ðb þ 0:3Þ1:5
B¼2 þ 1 for 0:50 < b 1:00 adequate job of capturing the shape of the curves,
ðT=T P Þ1:3 ensuring that there are no damping coefficients
(6b) that are greatly overconservative. In addition, the
model remains conservative and captures
The empirical expression that was developed the behavior of the damping coefficient plots of
through this method is shown in Fig. 16 as a direct the different earthquake magnitude groupings
Seismic Actions Due to Near-Fault Ground Motion, normalized with TP (Reprinted from Hubbard and
Fig. 16 A comparison between the proposed method and Mavroeidis (2011). Copyright # 2011 Elsevier B.V.)
the calculated damping coefficients with the period axis
of the dataset (Figs. 15b, c, d) as shown by Hub- The author is grateful to Professor Papageorgiou for pro-
bard and Mavroeidis (2011). viding him with the opportunity to work on this research,
as well as for his guidance and advice.
This material is based upon work supported by the
National Science Foundation (NSF) under Grant Nos.
Summary EEC-9701471, CMS-0077504, and CMMI-1032504; by
the Federal Highway Administration (FHWA) under
Contract DTFH61-98-C-00094; and by the US Geological
This entry focuses on the description of seismic Survey (USGS) under Award No. 04HQGR0029.
actions due to near-fault ground motions. Partic- Any opinions, findings, and conclusions or recommenda-
ular emphasis was given on synthesizing broad- tions expressed in this material are those of the author and
band near-fault ground motion time histories for do not necessarily reflect the views of the funding
agencies.
earthquake engineering applications, using
a simple mathematical model for the representa-
tion of the coherent ground motion component
and a physical model of the seismic source for the
Cross-References
description of the incoherent seismic radiation. S
In addition, recommendations on design spectra, ▶ Earthquake Response Spectra and Design
Spectra
strength reduction factors, and damping coeffi-
cients were made for engineering analysis and ▶ Engineering Characterization of Earthquake
design in the near-fault region. This included Ground Motions
▶ Physics-Based Ground-Motion Simulation
the normalization of the period axis with respect
to the period of the ground velocity pulses. The ▶ Response Spectrum Analysis of Structures
pulse period is controlled by the rise time on the Subjected to Seismic Actions
▶ Response-Spectrum-Compatible Ground
fault plane and scales directly with earthquake
magnitude. Motion Processes
▶ Selection of Ground Motions for Response
History Analysis
Acknowledgments The material presented in this entry
is primarily based on previous articles published by the ▶ Stochastic Ground Motion Simulation
author and Professor Apostolos S. Papageorgiou. ▶ Time History Seismic Analysis
2540 Seismic Analysis of Concrete Bridges: Numerical Modeling
References Newmark NM, Hall WJ (1982) Earthquake spectra and

design. Earthquake Engineering Research Institute,
Alavi P, Krawinkler H (2001) Effects of near-fault ground Berkeley
motion on building structures (Report). CUREE- Papageorgiou AS (1997) Engineering seismology. In:
Kajima Joint Research Program, Richmond Beskos DE, Anagnostopoulos SA (eds) Computer
Archuleta RJ (1984) A faulting model for the 1979 analysis and design of earthquake resistant structures.
Imperial Valley earthquake. J Geophys Res Computational Mechanics Publications, Southampton,
89:4559–4585 pp 153–200
Baker JW (2007) Quantitative classification of near-fault Papageorgiou AS, Aki K (1983a) A specific barrier model
ground motions using wavelet analysis. Bull Seismol for the quantitative description of inhomogeneous
Soc Am 97:1486–1501 faulting and the prediction of strong ground motion.
Bertero VV, Mahin SA, Herrera RA (1978) Aseismic I. Description of the model. Bull Seismol Soc Am
design implications of near fault San Fernando earth- 73:693–722
quake records. Earthq Eng Struct Dyn 6:31–42 Papageorgiou AS, Aki K (1983b) A specific barrier model
Bouchon M (1997) The state of stress on some faults of the for the quantitative description of inhomogeneous
San Andreas system as inferred from near-field strong faulting and the prediction of strong ground motion.
motion data. J Geophys Res 102:11731–11744 II. Application of the model. Bull Seismol Soc Am
Bray JD, Rodriguez-Marek A (2004) Characterization of 73:953–978
forward-directivity ground motions in the near-fault Sasani M, Bertero VV (2000) Importance of severe pulse-
region. Soil Dyn Earthq Eng 24:815–828 type ground motions in performance-based engineer-
Halldorsson B, Papageorgiou AS (2005) Calibration of the ing: historical and critical review. In: Proceedings of
specific barrier model to earthquakes of different tec- the twelfth world conference on earthquake engineer-
tonic regions. Bull Seismol Soc Am 95:1276–1300 ing, Auckland
Halldorsson B, Mavroeidis GP, Papageorgiou AS Somerville P (2003) Magnitude scaling of the near fault
(2011) Near-fault and far-field strong ground motion rupture directivity pulse. Phys Earth Planet In
simulation for earthquake engineering applications 137:201–212
using the specific barrier model. J Struct Eng ASCE Vassiliou MF, Makris N (2011) Estimating time scales
137:433–444 and length scales in pulse-like earthquake acceleration
Heaton TH (1982) The 1971 San Fernando earthquake: records with wavelet analysis. Bull Seismol Soc Am
a double event? Bull Seismol Soc Am 72:2037–2062 101:596–618
Housner GW, Trifunac MD (1967) Analysis of Veletsos AS, Newmark NM (1960) Effect of inelastic
accelerograms – Parkfield earthquake. Bull Seismol behavior on the response of simple systems to earth-
Soc Am 57:1193–1220 quake motions. In: Proceedings of the second world
Hubbard DT, Mavroeidis GP (2011) Damping coefficients conference on earthquake engineering, Tokyo
for near-fault ground motion response spectra. Soil
Dyn Earthq Eng 31:401–417
Makris N (1997) Rigidity-plasticity-viscosity: can
electrorheological dampers protect base-isolated Seismic Analysis of Concrete
structures from near-source ground motions? Earthq
Eng Struct Dyn 26:571–591
Bridges: Numerical Modeling
Mavroeidis GP (2004) Modeling and simulation of near-
fault strong ground motions for earthquake engineer- Andreas Kappos
ing applications. PhD dissertation, Department of Department of Civil Engineering, City
Civil, Structural and Environmental Engineering,
University London, London, UK
State University of New York at Buffalo, Buffalo
Mavroeidis GP, Papageorgiou AS (2002) Near-source
strong ground motion: characteristics and design
issues. In: Proceedings of the seventh U.S. national Synonyms
conference on earthquake engineering, Boston
Mavroeidis GP, Papageorgiou AS (2003) A mathematical
Concrete bridges; Modal analysis; Nonlinear
representation of near-fault ground motions. Bull
Seismol Soc Am 93:1099–1131 analysis; Seismic loading
Mavroeidis GP, Papageorgiou AS (2010) Effect of fault
rupture characteristics on near-fault strong ground
motions. Bull Seismol Soc Am 100:37–58 Introduction
Mavroeidis GP, Dong G, Papageorgiou AS (2004) Near-
fault ground motions, and the response of elastic and
inelastic single-degree-of-freedom (SDOF) systems. Bridges are deceptively simple systems, since
Earthq Eng Struct Dyn 33:1023–1049 they are typically single-storey structures
Seismic Analysis of Concrete Bridges: Numerical Modeling 2541
wherein the horizontal members (the deck) can intended plastic mechanism, which, contrary to
often be modeled either as a continuous beam or buildings, involves primarily yielding in the ver-
as a series of simply supported beams. In fact, the tical members, i.e., the piers, while seismic
continuous beam might be a valid approximation energy dissipation can also take place in the bear-
in seismic analysis of bridges if a “spine” model ings; in a small number of (important) bridges,
is adopted; in such a model the bridge deck is supplementary damping devices (such fluid or
simulated using 3D beam elements with 6 degrees friction dampers) are provided and have to be
of freedom (DOFs) at each node, located at the accounted for in the analysis. In practical design
centroid of the cross section. It is worth noting applications bridges are analyzed in the elastic
that in gravity load analysis the geometric com- range, and any inelasticity (material nonlinearity)
plexity of the deck is usually represented in the effects, wherever entering of some members in
computer model to greater detail, as compared to the inelastic range under the design earthquake is
that used for estimating the seismic response. On allowed, are accounted for by simply reducing
the other hand, bridges present peculiarities that the design response spectrum by a “behavior”
are not commonly encountered (or are far less (or force reduction) factor. However, special
important) in buildings, such as the modeling of types of bridges have been analyzed using
bearings, shear keys, and expansion joints, as advanced inelastic analysis tools (typically as a
well as the modeling of soil-structure interaction verification of an initial design based on the
at all bridge supports, including those at the abut- results of equivalent elastic analysis), and in gen-
ments which can play a major role in some cases. eral, inelastic analysis methods are gaining
In the sections that follow, modeling of the ground in recent years. Therefore, an attempt is
various bridge components is first addressed made herein to present some basic concepts and
(section “Modeling of Bridge Components”), models suitable for inelastic analysis, with
followed by an overview of methods currently emphasis on those that are better suited for prac-
used for seismic analysis of bridges (section tical application.
“Bridge Analysis Methods”), which is a topic
that is also addressed in other articles of the
encyclopedia; section “Bridge Analysis
Methods” also includes the presentation of a Modeling of Bridge Components
case study that illustrates the modeling and anal-
ysis procedures described in the previous sec- Deck
tions, while both sections “Modeling of Bridge In concrete bridges the deck can have various
Components” and “Bridge Analysis Methods” forms, i.e., solid slab, voided slab, beams
include specific modeling examples and selected (usually precast I-beams) with cast in situ top
results. Finally some concluding remarks are pro- slab, box girder (single-cell or multicell), and S
vided in section “Summary and Concluding other, less common, ones. The material used is
Remarks.” Methods of seismic analysis are typically prestressed concrete, except for some
presented in various parts of the encyclopedia short span bridges with slab-type deck, where
(e.g., article by Vayas and Iliopoulos (2014) ordinary reinforced concrete can be used.
focusing on modeling of steel and composite Very important, both for the seismic behavior
bridges). Hence, the focus herein is on analysis of the bridge and for its modeling, is the type of
of concrete bridges, wherein some specific issues pier to deck connection. There are three basic
arise as discussed in the following; of course, options in this respect:
several of the described models and techniques
are also applicable to bridges made of other • Monolithic connection: Very common in slab
materials. bridges and box-girder bridges, especially
Modeling concrete bridges for seismic design when the cantilever method of construction is
purposes should always take into account the used in the latter.
actual hinges can be introduced beforehand at

these locations.
In the usual case that the deck remains elastic
under the design seismic action, it can be
modeled with elastic elements of any appropriate
type: beam-column elements forming frames or
grillages or even shell elements. For seismic anal-
ysis the recommended approach is the use of 3D
beam columns (spine model), which is the sim-
plest option and generally an adequate one for the
purposes of this analysis. Relevant guidelines
Seismic Analysis of Concrete Bridges: Numerical
Modeling, Fig. 1 Pier region of a bridge with deck
like those of ACI 341 (2014) and FHWA (2006)
consisting of precast posttensioned beams and cast in recommend four to five elements per span, but
situ slab continuous over the piers usually more elements are used in practice, since
these are elastic members and do not noticeably
affect the computational demands. Besides,
• Bearing connection: The deck rests on the prestressed concrete members like beams and
piers through two or more bearings, which in box girders often have cross sections that vary
the case of modern concrete bridges are typi- along the span (thicker webs of beams and box
cally elastomeric bearings; this is the usual girders toward the piers, where shear forces are
type of connection in the case of beam-type maximum); hence, the 3D beam elements should
decks and also in box-girder bridges when the be arranged in such a way that they properly
incremental launching method of construction reproduce this gradual change in geometry.
is used. Moreover, since masses required for the dynamic
• Mixed type of connection: Some piers analysis of the bridge are typically lumped at the
(typically the taller ones) are monolithically nodes of the model (even when uniformly distrib-
connected to the deck, whereas others uted masses are automatically calculated from
(typically the squat ones) have bearing con- the geometry of the elements), use of a sufficient
nections. This is a fairly new solution, which number of elements leads to a more accurate
has advantages when bridge configurations representation of the mass distribution in the
with substantially unequal pier heights have bridge and hence of its dynamic characteristics.
to be used (e.g., in ravine bridges). In the case of monolithic connections, deck-to-
pier joints carry bending moments due to seismic
In the case of bearing connections (very com- loading, and in principle, parts of the deck may
mon in older concrete bridges), it is clear that the become inelastic. However, in most cases the
deck does not carry moments due to lateral strength of the deck that is governed by the sub-
(seismic) loading nor forms part of the energy stantial gravity loading on the bridge plus traffic
dissipation mechanism of the bridge. An excep- loads (very high in railway bridges) is clearly
tion is the case of continuity slabs, i.e., the parts higher than that of the pier that is typically
of a top slab in beam-type decks that continue governed by seismic moments in medium to high
over the piers (whereas beams terminate on each seismicity areas; hence, the deck remains in the
side of the pier), hence providing a continuous elastic range and the previous comments apply.
deck (Fig. 1). For these regions modern codes like In this case, accurate modeling of box girders
Eurocode 8-2 (CEN 2005a) allow formation of requires accounting for shear lag (nonuniform dis-
plastic hinges (in bending about the transverse tribution of the longitudinal stress across the flange
axis, see Fig. 1). When the focus of the analysis width due to the shear deformations within the
is on the response to the design earthquake and flange); the FHWA (2006) manual specifies that
inelastic behavior is allowed and anticipated, the flexural stiffness of the superstructure taken
about a transverse axis should be reduced near The results of seismic analysis in the trans-
piers when there is moment transfer between the verse direction of a bridge with box-girder deck
superstructure (deck) and the pier, without provid- will be influenced by the assumption made
ing values for this reduction. According to the regarding the torsional stiffness of the deck
ATC-32 report (ATC 1996), stiffness in these (which is substantial, as opposed to that of
regions is based on an effective width that should “open-type” orthotropic decks like beams with
be no greater than the width of the column plus top slab). It is recommended to assume 20 % of
twice the cap beam depth. If this width is practi- the uncracked value, based on the ratios (10
cally equal to the entire width of the superstruc- 30 %) of cracked-to-uncracked torsional stiffness
ture, no reduction in stiffness due to shear lag is estimated by Katsaras et al. (2009).
required in the model. In between the aforementioned simple (spine)
An example is shown in Fig. 2 depicting the and complex (shell) modeling approaches for the
spine model of an overpass bridge (Kappos deck is the grillage model, i.e., a horizontal pla-
et al. 2013a) with monolithic pier to deck con- nar system of longitudinal and transverse 3D
nections and free sliding connections at the abut- beams, as shown in the example of Fig. 3, refer-
ments. It is seen that, depending on the length of ring to a ravine bridge with a top slab on
the span, up to 11 3D beams have been used. The posttensioned I-beams supported through lami-
thickening of the webs of the box-girder section nated elastomeric bearings, studied by Ntotsios
toward the end of the spans is properly modeled et al. (2009). The longitudinal members of the
in the software used, SAP 2000 (CSI 2011). Fig- grillage have the properties of the I-beams and
ure 2 also conveys an idea of what the model the tributary part of the slab, the transverse ele-
would look like in case shell elements were ments above the piers have the properties of the
used. As shown in a study by Kappos actual transverse beams in the actual bridge,
et al. (2002), it is possible to achieve a good while the intermediate four transverse elements
match of the dynamic characteristics of a bridge of the grillage represent the coupling of the lon-
modeled using shell elements, using a simple gitudinal beams in the transverse direction due to
spine model; the match was particularly good the presence of the deck. Grillage models are
(differences in significant natural periods particularly suited for slabs or orthotropic decks
between 4 % and 11 %) when reduced flexural like that of Fig. 3 and provide a good balance
rigidity was specified for the 3D beam elements between accuracy and practicability. They are not
close to the monolithic connection. That study very easy to set up nor offer particular advantages
also confirmed that it is not necessary to use too over the spine model in the case of box-girder
many elements for the deck; in fact a model sections. An important issue here is the modeling
involving 488 3D beam elements predicted of the torsional stiffness of the deck which cannot
almost identical dynamic characteristics be estimated from the properties of the individual S
(periods and mode shapes) as another one using elements of the grillage but rather should be
only 77 elements (8 per span). derived for the entire box girder and then distrib-
The value of flexural rigidity (EI) to be assigned uted among the longitudinal members. In case the
to the elements used for modeling the deck depends grillage model is used for solid slab decks, it is
on the material used. In the usual case of recommended to add diagonal braces to account
prestressed concrete, it is allowable to assume neg- for the interaction between longitudinal and
ligible cracking and use the value for the gross transverse deck action due to Poisson effects,
cross section (EIg). In ordinary (non-prestressed) which cannot be neglected (as in the case of
reinforced concrete decks, EI should account for orthotropic decks with primary girders, see
cracking effects. Recommended values in Caltrans Vayas and Iliopoulos (2014) for orthotropic com-
(2013) are 50–75 % of EIg; the lower bound repre- posite decks). More details on the input parame-
sents lightly reinforced sections and the upper ters required for deck modeling with grillages
bound represents heavily reinforced sections. (flexural and torsional rigidities, shear areas)
F4SD
2544
F3SD
F3SD
z F1SD
F2SD
x F2SD
COLM2
F2SD
COLM1
27 m 45 m 27 m
F1SD F2SD F3SD F4SD
COLM1 COLM2
Parameter F1SD F2SD F3SD F4SD 3 3 rigid element rigid element
A (m2) 12.38 5.89 12.35 12.81
2 2 F4SD
Ix (m4) 12.28 8.41 12.63 13.54
Iy (m4) 4.47 3.34 4.74 4.97 rigid elements link element
Iz (m4) 56.37 40.73 56.99 57.72
gap
bearing bearing
Seismic Analysis of Concrete Bridges: Numerical Modeling, Fig. 2 Finite element modeling of an overpass bridge with a box-girder superstructure monolithically
connected to the piers (Kappos et al. 2013)
Seismic Analysis of Concrete Bridges: Numerical Modeling
Seismic Analysis of Concrete Bridges: Numerical Modeling, Fig. 3 Grillage model and detail of the connection
between the deck and the pier (above) for a ravine bridge with beam with top slab deck (below)
can be found in Chapter 2, of the recent book by Geometric Considerations

Kappos et al. (2012). In that chapter a brief dis- Single-column piers, whether solid or hollow, are
cussion of some advanced topics like the effects generally modeled as “sticks” using beam-
of skewness and curvature in plan and/or in ele- column elements; four to five elements are usu-
vation on the seismic response, and the verifica- ally enough, unless complex geometries (e.g.,
tion of deck deformation demands when this is flared columns) are involved. As shown in
necessary for seismic assessment, can also be Fig. 4, the top element of the column is connected
found. via a rigid link to the centroid G of the deck
section (which is the location of the
Piers and Their Foundations corresponding horizontal element); in most soft-
Piers commonly used in concrete bridges are of ware packages there is no need for specifying a
the following types: different element, but rather the top element of
the column extends up to G, and its end portion is
• Single columns with solid circular or (less specified as a rigid offset. In the common case of
frequently) rectangular section columns monolithically connected to the box S
• Single columns with hollow circular or rect- girder of the deck, there is no need for additional
angular section elements in the column, but when the box girder
• Multicolumn bents (frame-type piers, running is supported on bearings (which will be the case
in the transverse direction of the bridge) at the abutments), it is necessary to introduce
• Wall-type piers (usually of large dimensions, transverse elements extending from the first to
especially in the transverse direction) the last bearing and link these to the end node of
• Braced piers (usually V shaped) the deck model using rigid links (see bottom right
of Fig. 2).
It is noted that hollow rectangular piers are As shown in Fig. 4, the moment distribution
typically of large dimensions (could reach 7 m along the height of the pier is influenced by the
or more in the transverse direction), and their rotational restraint at the top, which in turn
structural behavior is closer to that of walls, depends on the torsional rigidity of the deck.
rather than of hollow circular columns. This is substantial in the case of box girders
M
G MTOP MTOP
M MTOP =0
hG
F F
Massless
Rigid Section
heq
heq
hclear
or or
heq
MBASE MBASE
hf
MBASE
a b c
Seismic Analysis of Concrete Bridges: Numerical Modeling, Fig. 4 Pier modeling and transverse response
accounting for the torsional stiffness of the deck
(as in Fig. 4) and low in the case of precast beams add the aforementioned horizontal rigid element
with top slab and other similar, “open-type,” at the top of the stick model. A more refined
sections. model could consist of a vertical grillage (see
Multicolumn bents are naturally modeled as notes on grillage models in section “Deck”).
2D frames, again with beam-column elements, The most refined model, feasible only in elastic
each of which has the properties of the analysis, is the use of shell elements; an example
corresponding member (column or cap beam). It (not necessarily a recommended one, unless for
is worth noting that although columns usually research purposes) is shown in Fig. 5; a mesh of
have circular sections, cap beams are rectangular. 18,885 shells was set up for the 3-span bridge,
Rigid offsets at the element ends properly capture keeping the aspect ratio of the shells rather low at
the effect of the finite size of the beam-column 1.2, since these elements perform best when their
joints that are quite massive members in bridges. shape is close to square. As noted in section
In case the cap beam is monolithically cast with “Deck,” the main normal modes of that bridge
the deck, the torsional resistance of the top of the were captured with reasonable accuracy using a
bent is substantially higher than that of the cap simple spine model with only 77 beam-column
beam alone; Aviram et al. (2008) recommend elements (Kappos et al. 2002).
multiplying the torsional resistance of the cap
beam by 102. Clearly this does not apply in the Stiffness Considerations
case that the deck is bearing-supported on the cap Even in elastic analysis of reinforced concrete
beam. V-shaped piers are modeled in a similar (R/C) piers for seismic loading, it is essential to
way, but of course, vertical elements are inclined account for the effect of cracking, to make sure
rather than upright. that displacements are not underpredicted. Prac-
There are several options available for model- tically all existing codes adopt approximate
ing wall-type piers and the aspect ratio is an values of the pier stiffness, corresponding to
important parameter in this case. The simpler yield conditions, and this stiffness is assumed as
model is clearly the stick one, previously men- known when design seismic actions (e.g., modal
tioned for the case of single-column piers. Decks forces) are estimated. These approximate values
are usually bearing-supported on wall-type piers are either very rough estimates, like the 0.5 EIg
(which are the preferred solution in seismic iso- (50 % of uncracked section rigidity) adopted by
lation designs wherein seismic energy dissipation both Eurocode 8-1 (CEN 2004) and AASHTO
takes place in the bearings and, whenever present, (2010), or slightly more sophisticated ones taking
the dampers); hence, it is essential in this case to into account the level of axial loading on the pier
Seismic Analysis of
Concrete Bridges:
Numerical Modeling,
Fig. 5 Finite element
(shell) mesh in the pier to
deck connection area of a
bridge with box-girder
superstructure
monolithically connected
to hollow rectangular piers
(Kappos et al. 2002)
Seismic Analysis of 0.80

ELASTIC STIFFNESS RATIO, Ie /Ig
Concrete Bridges:
Numerical Modeling, 0.70
Fig. 6 Effective stiffness Ast/Ag = .04
of cracked reinforced 0.60
concrete circular sections Ast/Ag = .03
(AASHTO 2010) 0.50 Ast/Ag = .02
0.40 Ast/Ag = .01
0.30
0.20
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
AXIAL LOAD RATIO P/fc′Ag
(which, in general, is not significantly affected by Obviously, Icr can only be estimated from Eq. 2
seismic actions) and/or the reinforcement ratio. when the pier has been designed, so that both
Eurocode 8-2 for Seismic Design of Bridges strength and yield curvature can be calculated; S
(CEN 2005a) in its (informative) Annex hence, use of these relationships is feasible only
C suggests the following relationship for the when iterative elastic analyses, or inelastic anal-
effective moment of inertia of R/C ductile ysis, are used.
columns: The Caltrans Seismic Design Criteria (2013)
adopt the same concept as EC8-2 (secant value at
Ieff ¼ 0:08 Ig þ Icr (1) yield), the only exception being that the 0.08 Ig
term (accounting for tension stiffening effects) is
where the cracked section inertia can be calcu-
not included in Eq. 1. As an alternative, the
lated as the secant value at yield (My is the yield
Caltrans Criteria allow the calculation of effec-
moment and jy the yield curvature, Ec the con-
tive stiffness as a function of the axial load ratio
crete modulus)
and the pier reinforcement ratio from graphs pro-

vided by Priestley et al. (1996); diagrams like that
Icr ¼ My = Ec :jy (2)
of Fig. 6 can be directly implemented for carrying
Seismic Analysis of Concrete Bridges: Numerical Modeling, Fig. 7 Lumped plasticity model (a) basic configu-
ration; (b) hysteresis law for springs
out elastic analysis, assuming a reasonable rein- Lumped plasticity models, also known as
forcement ratio (e.g., Ast/Ag = 0.01, which is the “point-hinge” models, are based on the simpli-
usual minimum reinforcement ratio), while, in fying assumption that all inelastic behavior
principle, analysis should be repeated if the takes place at the plastic hinge points that are
resulting reinforcement is substantially different. typically located at the member ends. This
concept can be materialized in different ways,
Nonlinear Models for Piers the most efficient one consisting in inserting
As mentioned in section “Introduction,” the most two nonlinear rotational springs at the element
common plastic mechanism on which seismic ends, as shown in Fig. 7a; more accurately, the
design is based is that involving inelastic springs are inserted at the ends of the rigid
response of the piers. Therefore, inelastic offsets located at the element ends to model
(material nonlinear) modeling of piers is impor- the finite width of joints (e.g., between the cap
tant, not only for research but also for practical beam and the column, in a multicolumn bent).
assessment purposes (which includes assessing All post-yield flexural deformation takes place
an existing, probably substandard, bridge as in these springs, whereas the remainder of the
well as a newly designed bridge that is important beam-column element remains elastic through-
enough to warrant this additional design effort). out; it is emphasized that the flexural rigidity
Space limitations do not allow a detailed treat- EI of the “elastic” element should account for
ment of this important issue, and only a brief cracking effects, as discussed in section “Stiff-
overview of the main available models will be ness Considerations.” The (local) stiffness
provided herein; more detailed information on matrix for this lumped plasticity element, relat-
nonlinear modeling of bridge piers and several ing chord rotations at the ends to the
case studies can be found i.a. in the recent book corresponding bending moments, can be read-
by Kappos et al. (2012). ily set up assuming a series connection
Nonlinear models for piers can be classified between the springs and the beam, hence
into three categories: adding the flexibility matrices of each compo-
nent, i.e.,
• Lumped plasticity models 2 3
L 1 L
• Distributed plasticity models þ
6 3EI KS i 6EI 7
• Continuum models ½F ¼ ½Fe þ ½FS ¼ 6
4
7
L L 1 5
þ
Several subcategories can be defined for each 6EI 3EI KS j
of the above, as briefly discussed in the following. (3)
where Ksi and Ksj are the stiffnesses of the springs that they can relatively easily account for the
at ends i and j, which are assumed for simplicity effects of shear, as well as of bond slip, and are
to be uncoupled (no off-diagonal terms Ksij) and easier to calibrate against experimental results
can be different, e.g., when one end has yielded than other, more sophisticated (and complex),
while the other is still in the pre-yield range. The models.
main advantage of this simple model is that Ks Distributed (or spread) plasticity models, still
values can be defined on the basis of any consti- of the beam-column type, drop the assumption of
tutive law, whether simple or complex. Figure 7b point hinges and directly account for the spread
shows a typical moment versus rotation hystere- of inelasticity along the bridge member, hence
sis law (with stiffness degradation) that can be leading, in principle, to more accurate results.
applied to either spring; note that prior to exceed- There are several different approaches in this
ing the yield moment the springs are rigid respect, i.e., inelastic response can be monitored
(Ks = 1), hence all elastic deformation takes at several predetermined sections of the element
place in the elastic member. The 2 2 flexural and the stiffness matrix be synthesized on the
stiffness matrix relating end moments to chord basis of the tangent stiffness of each such section,
rotations can be easily derived by inverting or variable length plastification zones be defined,
[F] and then transforming to the 6 6 matrix typically at the member ends, assuming the rest of
including the rigid body modes and the axial the element is quasi-elastic (as in lumped plastic-
deformations (axial stiffness EA is usually ity models). The latter option retains some of the
assumed to remain unaffected by flexural simplicity of point-hinge models while being
yielding). more rational and, in principle, accurate and
Bridge piers, in particular those of the wall will be described in the following.
type, develop significant shear deformations sub- A recent spread plasticity model (Mergos and
sequent to shear cracking, which may occur Kappos 2012) accounting for inelastic response
before or after flexural yielding. In the lumped in all mechanisms (flexure, shear, bond slip) is
plasticity context, shear deformations can be shown in Fig. 8. The length of the plastified zones
treated either in a simplistic way by using the at the ends (respective rigidities EIA and EIB) is
stiffness matrix of a Timoshenko beam defined on the basis of the moment diagram of the
(involving GA’ terms in addition to EI ones) element and the corresponding yield moments,
and a rough reduction factor for GA’, or, more e.g., the left zone has a length aAL, where
rigorously, by introducing additional springs at
the ends representing the relationship between MA MyA
aA ¼ 1 (4)
shear force and shear deformation; the issue of MA MB
inelastic shear is discussed later on in relation to
another model. Another important source of The flexural stiffness matrix can be set up using S
deformation in R/C piers, especially those with the principle of virtual work for the case of mem-
substandard detailing with respect to earthquake bers with variable cross section. The current
requirements, is bond slip, which can give rise to rigidities are calculated from the moment versus
substantial local rotations at the member ends curvature relationship at each member end.
(“fixed end rotations”). This effect can either be Models with more than three parts have been
modeled indirectly by decreasing the stiffness of proposed but are not deemed appropriate for
the M – y law (Fig. 7b) at the member ends, practical application.
something that requires proper calibration, or be A similar procedure can be followed in the
modeled directly by introducing additional rota- case of the shear sub-element (Fig. 8d), which
tional springs to the model of Fig. 7a. represents the hysteretic shear behavior of the
Despite their crudeness, lumped plasticity R/C member prior and subsequent to shear crack-
models (first develop in the 1960s) remain quite ing, flexural yielding, and yielding of the shear
popular due to their simplicity, as well as the fact reinforcement. In this case, the current shear
Seismic Analysis of
Concrete Bridges:
Numerical Modeling,
Fig. 8 Distributed
plasticity model: (a)
geometry of R/C member;
(b) beam-column finite
element with rigid arms; (c)
flexural sub-element; (d)
shear sub-element; (e)
anchorage slip sub-element
rigidities (GAA, GAB) are calculated from the In the previously presented models, the ele-
V – g (shear versus shear deformation) curves at ment stiffness matrix in the post-yield range is set
each end, details of which are given in Mergos up on the basis of EI values that are estimated
and Kappos (2012). After determining the distri- from predefined constitutive laws relating bend-
bution of GA along the R/C member at each step ing moment to either end rotation or end curva-
of the analysis and by applying the principle of ture; when axial (EA) and shear (GA) rigidities
virtual work, the coefficients of the flexibility are not taken as constant, their values are esti-
matrix of the shear sub-element are given by the mated from similar predefined laws (e.g., V – g).
following equation: Another option is the fiber model, wherein the
stiffness parameters are not estimated from
aAs 1 aAs aBs aBs
ij ¼
f sh þ þ ði,j ¼ A, BÞ predefined laws, but rather from
GAA L GAM L GAB L
moment – curvature analysis (and, far less often,
(5)
shear force versus deformation analysis) of a
Finally, rotations due to slip at the end anchor- number of “monitoring” or “control” sections,
ages are captured with the simple slip which are divided in a number of “fibers”
sub-element of Fig. 8e that consists of a rigid (in the general, biaxial, case, these are squares
bar with two uncoupled nonlinear springs at the or rectangles rather than horizontal fibers), using
ends. The M-yslip skeleton curve is derived Bernoulli’s principle and the stress-strain consti-
assuming uniform bond stress along different tutive laws of the pertinent materials, i.e., con-
segments of the anchored reinforcement bar fined and unconfined concrete (for the core and
(details in Mergos and Kappos 2012). the cover, respectively) and steel bars, as shown
Seismic Analysis of Concrete Bridges: Numerical Modeling, Fig. 9 Discretization of R/C member using the fiber
approach
in Fig. 9 (Kappos et al. 2012). The model has it is not necessarily more accurate than the latter,
been implemented in point-hinge models but has except in the case of R/C members with negligi-
found its main use in distributed plasticity ble effect of shear and bond-slip deformations,
models, wherein four (Fig. 9) or more sections which is not very common, even in well-designed
are “monitored” and the element stiffness matrix bridge members. For instance, a pier usually has
is set up assuming linear variation of stiffness such an aspect ratio that shear deformations can-
(or flexibility) between monitoring sections; the not be ignored. There are versions of the fiber
latter are typically taken as the end sections and model wherein shear deformations are included
the Gauss points used in the integration required (Kappos et al. 2012), but the computational cost
for deriving the element stiffness matrix (e.g., involved is particularly high. Finally, because
using the Gauss-Lobatto quadrature scheme). most of the available fiber model-based software
A rigorous application of the fiber model packages (like SeismoStruct) include constitutive
involves a number of difficulties, and different laws assuming that concrete is initially
approaches have been put forward, some of them uncracked, they overestimate the actual stiffness
based on the stiffness approach (and involving of real bridges (and other structures) that are S
displacement shape functions) and others on the cracked prior to being subjected to seismic load-
flexibility approach (involving force shape func- ing (due to shrinkage, traffic and ambient vibra-
tions, which do not change in the inelastic range); tions, and possible previous small earthquakes).
hybrid procedures have also been used. The flex- Nonlinear continuum models are typically
ibility approach is numerically more advanta- used for research purposes, whose aim is to
geous but computationally more demanding. study in detail the response of critical regions in
Details of all these procedures fall beyond the piers. Material nonlinearity is taken into account
scope of this entry and can be found in the liter- using either a “standard” plasticity model or a
ature (e.g., Fardis 1991; Kappos et al. 2012). combination of plasticity and damage models,
However, it has to be emphasized here that the latter being able to affect the elastic compo-
although the fiber approach is more rigorous nent of the deformation as (seismic) damage
than the “phenomenological” approaches based propagates. Geometric nonlinearity is less critical
on predefined force – deformation relationships, in R/C piers, except for very tall ones. The type of
80
MN
70
60
50
40
30 Dense mesh, z = 1.0 m
20 Dense mesh, z = 2.0 m
10 Coarse mesh, z = 1.0 m

e
0
0.000 0.005 0.010 0.015 0.020 0.025
Seismic Analysis of Concrete Bridges: Numerical Modeling, Fig. 10 Modeling of hollow circular piers with solid
elements using different heights and mesh densities
finite elements used can be quire “heavy,” i.e., capturing the effect of foundation compliance,
shell and 3D solid (“brick”) elements have been i.e., of the fact that pier foundations, especially
used for concrete piers. Figure 5 shows an exam- the shallow but also the deep ones, like piles, do
ple of using shell elements for both the deck and not provide full fixity to the base of the pier, but
the piers, but nonlinear behavior was not taken displace (horizontally), rotate, and even settle
into account in that model. (vertically), as the bridge is subjected to seismic
Figure 10 is taken from a study (Papanikolaou loading. Large-size foundations have also the
and Kappos 2009) focusing on the effect of con- effect of modifying the seismic input to the struc-
finement on the strength and ductility of solid and ture (“kinematic” interaction or “wave scatter-
hollow piers; solid elements were used for con- ing” effect), but this issue is not further
crete, whose behavior was governed by a sophis- addressed herein. The interaction of the founda-
ticated plasticity model accounting for tion ground with the bridge substructure (the
confinement effects, while line elements embed- piers and their foundations) does not only modify
ded to the solid elements were used for the rein- the fixity conditions but also increases the
forcement (transverse reinforcement consisted of damping (“radiation” damping at the ground-
spiral or hoop circular reinforcement, with or with- foundation interface); in simplified analysis this
out transverse links). It is seen that the axial load can be safely ignored, but there are ways to
versus axial deformation curves resulting from the explicitly include it, such as the addition of dash-
“coarse” mesh are not substantially different from pot elements at the base of the piers.
those from the dense mesh; it is worth pointing out A commonly adopted practical approach for
that an upper limit of about 6,000 solid elements calculating the pseudo-static interaction between
was found, beyond which the computational cost the bridge foundation and the soil is the Winkler
and volume of results were excessive, which is a spring model, wherein the soil reaction to the
good indicator of the type of models that can be foundation movement is represented by indepen-
analyzed in such a context. Further examples of dent (linear or nonlinear) unidirectional transla-
applications of continuum models to concrete pier tional spring elements. In the case of surface
components can be found in Kappos et al. (2012). foundations, the vertical springs are distributed
below the surface of the footing, while in pile
Soil-Structure Interaction Effects foundations horizontal springs are distributed
A key aspect in proper modeling of piers along the pile shaft. Although approximate,
(whether inelastic or equivalent elastic) is Winkler formulations are widely used not only
because their predictions are in good agreement for regions of moderate-to-high seismicity, G =
with results from more rigorous solutions but also 0.25 G0 is suggested (FHWA 2006).
because the variation of soil properties with depth In pile foundations the mechanical parameters
can be relatively easily incorporated. Moreover, for the springs are frequently obtained from
they are efficient in terms of computational time experimental results (leading to P-y curves for
required, thus allowing for easier numerical han- lateral and N-z curves for axial loading) as well
dling of the structural inelastic response, wher- as from very simplified models. A commonly
ever this is deemed necessary. used P-y curve is the lateral soil resistance versus
In the case of surface foundations, a simple deflection relationship proposed by the American
system of three translational (two horizontal, x Petroleum Institute (API):
and y, one vertical, z) and two (less often three)
rotational springs can be used at the base of the
kH
footing; the spring constants for the x, y, and z P ¼ 0:9pu tanh y (7)
0:9pu
springs can be estimated from relationships
(ASCE 2007) derived from the solution of the
problem of a rigid plate resting on the surface of where pu is the ultimate bearing capacity at depth
a homogeneous half-space: H, y is the lateral deflection, and k is the initial
modulus of subgrade reaction. The tip of the pile
"
# can either be modeled using a vertical spring
GB L 0:65
Kx ¼ 3:4 þ 1:2 (with a carefully selected axial stiffness) or
2n B
"
# assumed to be vertically fixed but free to rotate.
GB L 0:65 L Figure 11 shows an example (Kappos and
Ky ¼ 3:4 þ 0:4 þ 0:8 (6)
2n B B Sextos 2001) of modeling a pile group using
"
0:75 # Winkler springs with initial properties calculated
GB L
Kz ¼ 1:55 þ 0:8 from Eq. 7. This is a fully inelastic model wherein
2n B
both the piles and the pier (modeled using the
lumped plasticity approach, but with several ele-
where G is the shear modulus of the ground, n the ments for each pile) can yield; in practical appli-
Poisson ratio (0.35 for unsaturated soils and 0.5 cations equivalent linear properties (based on
for saturated soils), L the larger dimension of the secant stiffness at the estimated maximum dis-
(rectangular) footing, and the smaller one; sim- placement) are often used, especially for the soil.
ilar relationships are given in ASCE (2007) and The limitations of the P-y approached are
FHWA (2006) for the rotational springs. These discussed in detail in Kappos et al. (2012),
relationships are particularly convenient to use where an overview of more advanced soil-bridge
since they only include very fundamental prop- interaction (kinematic and inertial) models is also S
erties of the soil, which can always be estimated provided. It will only be mentioned here that
(e.g., the initial modulus G0 can be estimated despite the abundance of models and software,
from the shear wave velocity and the specific nonlinear analysis wherein both the bridge and
weight of the soil). However, an upper and the foundation ground are modeled with
lower bound approach to defining stiffness and nonlinear models is not only cumbersome but
(in nonlinear models) capacity is recommended often leads to convergence problems; hence,
because of the uncertainties in the soil properties explicit treatment of nonlinearity should be con-
and the static loads on the foundations of existing fined where the main interest lies, typically in the
bridges. The large-strain effective shear piers and perhaps in the piles (cf. Fig. 11).
modulus, G, can be roughly estimated on the
basis of the anticipated peak ground acceleration Abutments and Backfills
(PGA); for regions of low-to-moderate seismic- There are two common types of bridge
ity, a value of G = 0.5 G0 is recommended, while abutments:
54 21
53 20
ELASTIC 2D
52 19 BEAM (pier)
9m Half Stiffness
51 18 (symmetry)
17
106
105 4node
16 70
49 SHELL
35 64 rigid pile cap
15 69
48 102
34 65
14 68
47 101 INELASTIC
33 66 ROTATIONAL
13 67
46 100 SPRING
32 12 67 66
45 99
31 68 65 ELASTIC 2D
11
44 98 BEAM (pile)
30 10 69 64
43 97
NONLINEAR
29 9 70
63 TRANSLATIONAL
42 96
SPRING
28 71 62
8
41 95
27 7 72 61
15 m 40 94
26 6 73 60
39 93
25 5 74 59
38 92
25 4 75 58
37 91
23 3 76 57
36 90
22 77
2 56
1 55
M - θ curve for concrete Force - deflection curve for the soil
Seismic Analysis of Concrete Bridges: Numerical Modeling, Fig. 11 Modeling of soil-pile foundation-pier system
using Winkler springs
• Seat type (the deck is bearing-supported on the Integral- or diaphragm-type abutments are
horizontal seat of the abutment; see Fig. 2 always included in the model of the bridge, as a
bottom right) continuation of the deck model; in addition to the
• Integral or diaphragm type (abutment mono- body of the abutment, they also include the foun-
lithically connected to the end of the deck) dation, which typically consists of piles (often
relatively flexible ones). It is essential that the
The modeling of the abutment system, which flexibility of the abutment foundation be
also includes the wing walls and the foundation modeled; otherwise the displacements of the inte-
(footing or piles) can vary from very simple to gral bridge are seriously underestimated; besides
very complex, depending on the situation. Some problems with earthquake analysis, failure to
common cases are briefly discussed in the capture the flexible end supports of the bridge
following. also results in unrealistic stresses from
FX 12000
Pushover curve
.FY 10000
Stiff soil
8000
VA(KN)
Stiff soil–shear failure
6000 Soft soil
Soft soil–shear failure

4000
2000
0
0.00 0.10 0.20 0.30 0.40 0.50
ds(m)
Seismic Analysis of Concrete Bridges: Numerical Modeling, Fig. 12 Modeling of abutment system and resulting
pushover curves
temperature variations and shrinkage of concrete. can develop at the bridge ends; further movement
In addition to the foundation, the backfill and the (in either direction) can be captured by the model
embankment have to be accounted for in the only if the flexibility of the system, which
model, especially in short bridges with stiff includes both concrete members and the backfill
deck; a concept usually adopted in these cases is soil, is modeled.
the “effective” (or “critical”) length of the In abutments the important aspects of soil-
embankment, whose properties are calculated structure interaction (section “Soil-Structure
and then introduced as springs (discussed later Interaction Effects”) are modeled in practice-
in this section) at the end of the elements oriented applications through a system of linear
representing the integral abutment. or nonlinear springs at the ends of the bridge. The
Seat-type abutments can be modeled in a sim- properties of the springs can best be defined by an
ple way, as shown in Fig. 2, i.e., by just including analysis of the abutment-backfill system, prefer-
in the model the bearings (as springs, see section ably accounting for nonlinear effects directly, or
“Bearings” for property definition) of the abut- at least by proper selection of reduced properties
ment seat; so long as the end connection of the (e.g., estimates of G of the ground consistent with
bridge is dominated by the properties of the bear- the expected deformations). Figure 12 shows the
ings (in modern concrete bridges these are either modeling of an abutment system (Kappos and S
elastomeric or pot bearings that can slide in all Sextos 2009); the abutment wall is modeled
directions, see section “Bearings”), this simple with 2D shell elements, while the piles with
model is quite adequate. If pot bearings are frame elements supported on (depth dependent)
used, even a simple sliding connection (roller nonlinear horizontal springs. In the vertical direc-
support) can be defined in the model; this ignores tion, friction springs were used along the piles
friction forces at the pot bearings, which are small and an appropriate vertical stiffness was intro-
if monolithic connections or fixed bearings are duced with the use of a (compression only) spring
used elsewhere in the bridge. However, when a at the tip of the piles; a simpler model was also
longitudinal joint closes, or a shear key blocks the analyzed wherein infinite vertical stiffness was
transverse movement of the deck, or the gap assumed (tip displacements restrained). On the
between the deck and the abutment stem wall right of Fig. 12 are shown the pushover curves
closes (see Fig. 2 bottom right), the abutment- (i.e., seismic force versus monitoring point dis-
backfill system is activated and significant forces placement) derived for the transverse direction
(force Fy), wherein the behavior of the abutment one). The aforementioned Ki applies to well-
system is dominated by the nonlinear response of compacted backfills as required by Caltrans
the piles. For investigation purposes, the analysis (2013); otherwise it should be reduced by 50 %.
was performed both for soft soil conditions The ultimate abutment load was assumed to be
(which was the actual case in the specific abut- limited by a maximum static soil passive pressure
ment) and for the case of a significantly stiffer of 240 kPa; the latter is multiplied by the
supporting soil. These pushover curves reveal the corresponding surface, e.g., the product of the
sensitivity of the abutment response to the soil backwall width and height in seat-type abutments
conditions, as well as the importance of account- as well as the proportionality factor (h/1.7). The
ing for all possible failure modes in the analysis stiffness value for Eq. 8 applies when the elastic
(shear failure of the piles limits the deformation response of the bridge is dominated by the abut-
capacity, i.e., the ductility, of the entire system). ments; when this is not the case, Caltrans pre-
Equivalent linear or nonlinear springs based scribes reductions depending on the ratio of the
on the stiffness defined by curves like those in longitudinal displacement demand at the abut-
Fig. 12 can be used for modeling abutments to ment (from elastic analysis) to the effective lon-
which the deck forces are directly transferred gitudinal abutment displacement at idealized
(due to the previously mentioned reasons). yield (ratio of strength to stiffness).
A full-range model, covering all stages of the In the transverse direction, a nominal abut-
response, should include two (nonlinear) springs ment stiffness equal to 50 % of the elastic trans-
in series, one representing the stiffness of the verse stiffness of the adjacent bent can be used;
bearings and one the stiffness of the abutment this nominal stiffness has no direct correlation or
system; a gap element (with the width of the relevance to the actual residual stiffness (if any)
longitudinal joint of the bridge) should be added provided by the failed shear key but is meant to
if analysis is carried out in a single run; otherwise suppress unrealistic response modes associated
separate analyses with open and closed joint with a completely released end condition.
should be carried out (a usual practice in the Clearly, a full model for the abutment-backfill
USA). system should also include the backfill soil, as well
In lieu of carrying out a proper analysis of the as part of the embankment that is activated during
abutment, simplified procedures can be used for the seismic excitation of the bridge; as mentioned
estimating a reasonable stiffness (and, in previously, the latter is important in the case of
nonlinear analysis, strength) for the abutment. short bridges (like two-span overpasses) wherein
Arguably the most popular procedure is that pre- the embankment plays a key role in the seismic
scribed in Caltrans (2013), wherein the longitu- response of the bridge. In particular, consideration
dinal stiffness can be calculated from the initial of the abutment-soil system participating mass has
embankment fill stiffness Ki
28.7 kN/mm/ a critical effect on the mode shapes and conse-
(m width of the wall), and this has to be adjusted quently the dynamic response of the bridge. The
proportionally to the backwall (or diaphragm) critical length Lc of the embankment to be consid-
height (h): ered in the analysis can be estimated from the
relationship (Zhang and Makris 2002):
Kabut ¼ Ki w ðh=1:7Þ (8)
pffiffiffiffiffiffiffiffiffiffiffiffiffi
where w is the projected width of the backwall or Lc
0:7 S Bc H (9)
diaphragm, for seat and diaphragm abutments,
respectively, and (h/1.7) is a proportionality fac- where S is the slope of the embankment, H its
tor based on the 1.7 m height of the diaphragm height, and Bc its width at the crest. It should be
abutment specimen tested at UC Davis (the actual borne in mind that Lc actually changes with the
relationship is not linear, but so far there is no level of the seismic action, but this is difficult to
sufficient data to develop a more sophisticated capture in practical analysis.
Seismic Analysis of
Concrete Bridges:
Numerical Modeling,
Fig. 13 Modeling of
abutment and backfill
system using 3D finite
elements
An example of a “heavy” finite element model Among these, critical components of the
of the entire system is shown in Fig. 13 from bridge, particularly in an earthquake resistance
Kappos et al. (2012) wherein further examples context, are the bearings and the joints. Modeling
and details can be found. Results of analysis of a of these critical components for seismic analysis
number of typical abutment and backfills using is discussed in the remainder of this section,
sophisticated models such as that of Fig. 13, which also covers some other components, nor-
which considered the soil (backfill, embankment, mally located close to, or even within, the bear-
and foundation) as the nonlinear material mech- ings, i.e., shear keys and damping devices.
anism, have shown that both the stiffness and the
strength estimated according to the pre-2013 Bearings
Caltrans provisions (adopting Ki = 11.5 Bearings are mechanical systems which:
kN/mm/m, i.e., 60 % lower than the new value)
underestimated the values found from the 3D FE • Transmit loads from the superstructure (deck)
models. to the substructure (piers, abutments)
So long as the soil behind the abutment has • Accommodate relative displacements
been analyzed and its (macroscopic) stiffness between them
reduced to a spring constant, the entire bridge
can be modeled by combining in series the In the past, steel bearings of the pin, roller,
aforementioned translational springs (one for rocker, or sliding type have been used, and they
the backfill-embankment system and one for are still used in some bridges, in particular steel
the abutment and its foundation) in each direc- ones. In modern concrete bridge construction, S
tion of the bridge. More details on modeling bearings typically belong to one of the following
abutments and backfills using a system of categories:
nonlinear springs can be found in Aviram
et al. (2008). • Pot bearings
• Elastomeric (common) bearings
Bearings, Joints, and Shear Keys • Elastomeric (special) bearings
Bridge furnishings include a number of compo-
nents, i.e., Pot bearings (Fig. 14) allow sliding and rota-
tion and consist of a shallow steel cylinder
• Bearings (or “pot”) on a vertical axis with a neoprene
• Joints disk which is slightly thinner than the cylinder
• Parapets – rails and fitted tightly inside. A steel piston fits inside
• Waterproofing system the cylinder and bears on the neoprene, while flat
Seismic Analysis of
Concrete Bridges:
Numerical Modeling,
Fig. 14 Pot bearings
(www.agom.it): free-
sliding (left) and
transversely guided sliding
Seismic Analysis of
Concrete Bridges:
Numerical Modeling,
Fig. 15 Laminated
bearing with outer steel
plates: profiled (left) or
allowing fixing (EN1337,
CEN 2005b)
brass rings are used to seal the rubber between the The design of elastomeric bearings is carried
piston and the pot; the rubber behaves like a out (in European and some other countries)
viscous fluid, flowing as rotation occurs. Sliding according to the European Standard EN1337
can either take place in any direction or be guided (CEN 2005b). This standard prescribes maxi-
(through a groove and sliding bar system; see mum strains due to vertical load, rotations and
Fig. 14-right) in a specific direction. Use of pot horizontal actions, such as loads or displace-
bearings is very common, especially at the seats ments, and minimum thickness of the internal
of the abutments. and external steel plates. It also prescribes a num-
In general, pot bearings do not have to be ber of ultimate limit state verifications (limitation
explicitly modeled; it suffices to release the of distortion and rotation, tension in the steel
corresponding degrees of freedom at the support plates, bearing stability (buckling), and slip pre-
of the deck. vention). The procedure for designing a bridge so
Common elastomeric bearings are made of that the seismic action is resisted entirely by
elastomer, i.e., either natural or synthetic rubber elastomeric bearings on all supports (“seismic
(e.g., neoprene), which is flexible in shear (low isolation”) is prescribed in Chap. 7 of Eurocode
GA) but very stiff against volumetric change. To 8-3 (CEN 2005a).
avoid bulging (lateral expansion that adversely Proper modeling of elastomeric bearings is
affects the properties of the elastomer) different essential in the framework of seismic design.
types of reinforcement are used in the elastomer In all cases at least the horizontal shear stiff-
(fiberglass, cotton, steel). The most common and ness (Kh) should be captured, but in more
efficient (and also the most expensive) type of refined models the flexural (Kb) and the axial
reinforced elastomeric bearings is that reinforced (Kv) stiffness of the bearings are also intro-
with thin steel plates as shown in Fig. 15; these duced in the model. These three stiffness
are constructed by vulcanizing elastomer to these values can be calculated from the following
steel plates. relationships:
Fmax
Fy Kp
F0
Ke Keff
dy
ED dbd
Lead plug
Rubber sheets Steel plates
Seismic Analysis of Concrete Bridges: Numerical Modeling, Fig. 16 Lead rubber bearing (rectangular section)
and corresponding hysteresis law
Kh ¼ GA=tr (10a) material is due to its high hysteretic energy dis-

sipation (“fat” elastoplastic hysteresis loops
Kb ¼ 0:329 Ec I=tr (10b) under cyclic loading; see Fig. 16 right). In LRBs
the equivalent viscous damping ratio can reach
Kv ¼ Ec A=tr (10c) quite high values (in the range of 30 %). In
addition to damping, the lead core also provides
resistance to service lateral loads.
where tr is the thickness of the elastomer (not the
Modeling of special elastomeric bearings is
total height of the bearing) and the other symbols
similar to that of common bearings as far as
have their usual meaning. The shear modulus
stiffness is concerned (relationships 10), but the
G can be taken as 0.9 MPa for static loading and
additional damping has to be properly introduced
1.8 MPa for dynamic loading (rubber is a visco-
in the model. In all types of elastic analysis, a
elastic material, i.e., its resistance increases with
viscous damping ratio higher than that of
the loading rate).
reinforced concrete (5 %) is introduced, and the
Special elastomeric bearings are those that
corresponding spectra are used to derive the seis-
provide the high energy dissipation needed to
mic actions; in Eurocode 8 the ordinates of the
resist strong earthquakes. The most common
elastic spectrum for x 6¼ 5 % are estimated by
types used today are:
multiplying the reference spectrum by
• High damping elastomeric bearings qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi S
• Lead-rubber bearings (laminated bearings Z¼ 10
ð5þxÞ 0:40 (11)
with lead core)
The value of effective damping ratio to be used
Bearings of the first category can provide
can be estimated from (CEN 2005a)
fairly high values of effective damping ratio
(xeff
10–25 %). The main concern about them " #
is durability, as rubber properties are known to 1 SED, i
xeff ¼ (12)
deteriorate with time; of course, companies 2p Keff d2cd
manufacturing them claim design lives appropri-
ate for normal uses. Lead-rubber bearings (LRB) where SED,i is the sum of dissipated energies of
are laminated bearings with a cylindrical lead all special bearings i in a full deformation cycle at
core, as shown in Fig. 16; they are either circular the design displacement dcd and Keff ¼ SKeff , i ,
or rectangular. The selection of lead as the core i.e., the sum of the composite stiffnesses of the
isolator unit and the corresponding substructure example of an actual railway bridge wherein a
(pier) i. For the specific case of LRBs, the dissi- seismic isolation approach was adopted, involv-
pated energy ED,i is calculated from the pertinent ing a combination of lead rubber bearings and
elastoplastic hysteresis loop (Fig. 16 right). viscous fluid dampers; it is seen that the LRBs
An interesting type of special bearing, used located toward the end of the bridge are movable
exclusively for seismic isolation, is the friction horizontally, while free sliding pot bearings are
pendulum shown in Fig. 17, wherein the sliding used at the abutments.
surface of the bearing is concave; hence, the
restoring force is provided by the horizontal com- Supplemental Damping Devices
ponent of the structure itself. Sliding on the con- On several occasions involving large bridges, the
cave surface is resisted by friction of the contact amount of damping that can be provided by real-
material which is PTFE (polytetrafluoroethylene, istic arrangements of special bearings is not suf-
most common commercial name Teflon); the fric- ficient for limiting displacements to the required
tion coefficient is high initially (hence, no swaying limits (recall that long-span bridges are long-
of the superstructure takes place under normal period structures and when isolation is used
loading conditions) but substantially higher under their fundamental period can become very long,
high velocities induced by earthquake. more than 3 s). In such cases a more efficient
The articulation of the bridge, i.e., the solution can be the use of special damping
arrangement of the different types of bearings, devices (separate from the bearings) that will
is a critical aspect of the design of the bridge, in supplement the energy dissipation provided by
particular the seismic one. Figure 18 shows an the bearings (high damping or LRB); alterna-
tively, common elastomeric bearings or friction
pendulum bearings can be used in combination
with the damping devices. The most commonly
used devices are
• Viscous fluid dampers

• Steel yielding devices
Viscous fluid dampers are based on the con-

cept (long used in the automotive industry) of
forcing through a piston a viscous fluid (usually
silicon oil) through an orifice. Another, more
recent, alternative are shear panels containing
high-viscosity fluids. The constitutive law of
Seismic Analysis of Concrete Bridges: Numerical such dampers is not restricted to the well-known
Modeling, Fig. 17 Friction pendulum bearing linear dependence on velocity (through C, the
P8 P9 P10 P11 P12 P13 P14

P6 P7 P15
P4 P5 P16
P2 P3 P17
A1 P1 LRB 231-200 LRB 231-200 A2
LRB 231-200 LRB 231-200
LRB 231-200 LRB 231-200 LRB 286-250
LRB 231-200
+500/-350
LRB 231-200 LRB 231-200 LRB 231-200

LRB 231-200 LRB 231-200
LRB 231-200 LEGEND
LRB 286-250 LRB 231-200
LRB 231-200 +500/-350
+500/-350 Lead Rubber Bearing (LRB) Free Sliding POT Bearing
LRB te-Φ
L Plan Dimensions: 1200x1200 mm Capacity 11000kN, +500/-400 mm EDOMEN
POLIKASTRO te: Total Elastomer Thickness (mm)
LRB Movable Longitudinally
DISPLACEMENT CAPACITY I
Shear Key Movable Longitudinally (+) Towards the Bridge Center
Φ : Lead Core Diameter (mm) Plan Dimensions: 1200x1200 mm Capacity 4000kN, +550/-400 mm (-) Towards the Abutment
L
Single-Action Fluid Viscous Damper
Capacity 4500kN, +550/-400 mm
Seismic Analysis of Concrete Bridges: Numerical Modeling, Fig. 18 Arrangement of bearings in a seismically
isolated railway bridge (Kappos et al. 2014)
damping coefficient) but is often nonlinear, of the In a more involved, nonlinear, analysis, the
form viscous dampers can be directly introduced in
the model at the particular locations of the bridge
FD ¼ C ju_ ja sgnðuÞ
_ (13) wherein they are installed; such elements are
available in some software packages but are
where FD is the damping force and a is an expo- rarely used in practical design.
nent between 0 and 1; a = 1 corresponds to Steel yielding devices provide an almost
standard viscous damping, but in practical appli- elastoplastic hysteresis loop, and it is possible to
cations lower values are used since they lead to get a great variety of damping ratio values by
higher FD at lower velocities (e.g., in the bridge properly selecting the yield displacement and
shown in Fig. 18, a = 0.15 was selected for the post-yield stiffness ratio (hardening) of the isola-
viscous dampers, which had C = 5,440 kNs/m). tion system. They can be modeled either by
Modeling of viscous fluid dampers can be made expressing the hysteretic energy dissipation as
using two approaches: in the simpler approach, an equivalent damping (by equating the area of
appropriate for practice-oriented elastic analysis, the hysteresis loop to that of the ellipse
the dampers are substituted by an effective value representing viscous damping energy) or by
which is the sum of the basic damping x0 (typically directly including yielding elements at the perti-
5 %) and the contribution of the fluid dampers nent positions of the bridge (such elements are
(having exponent a and damping coefficients Ci). available in most programs).
Considering the fundamental mode of the bridge
(modal displacements ji at each mass mi) and Shear Keys
calculating the energy dissipated by the nonlinear Shear keys serve the purpose of preventing the
dampers in a cycle of sinusoidal motion, it can be displacement of the bridge deck in a certain direc-
shown (Hwang 2002) that tion and can be located at several positions in a
bridge; a typical one is at the abutments for
X
lCj f1þa
rj cos 1þa yj blocking the transverse movement of the deck.
j They can be either external as shown in Fig. 19
xeff ¼ x0 þ X (14)
2pA1a o2a mi f2i left (i.e., forming part of the seat) or as interior
i short cantilevers interlocking with corresponding
grooves in the deck (Fig. 19 right); exterior shear
where o is the circular frequency, yj is the angle keys are preferable because they are easy to
of inclination of the damping device j, l is a inspect and repair. Shear keys might directly
function of the exponent a (tabulated values of bear on the surrounding part of the deck, imme-
l are given in FEMA 1997), and ui = Aji (A is diately blocking its movement, or be located at a
the amplitude) are the actual displacements of the selected distance, forming a local joint (see sec- S
masses mi. tion “Joints” for modeling of joints).
Superstructure Superstructure
Exterior shear key
Abutment Interior shear

keys Abutment
Seismic Analysis of Concrete Bridges: Numerical Modeling, Fig. 19 Exterior and interior shear keys in bridges
(Kappos et al. 2012)
Modeling of shear keys should be consistent displacement capacity of around 100 mm or

with their design “philosophy,” i.e., whether the using more sophisticated multilinear constitutive
shear key is meant to remain essentially elastic laws with ascending and descending branches
for the entire range of seismic response of the (see more details in Kappos et al. 2012). In linear,
bridge considered in design, or a “sacrificial” “code-type,” analysis, one usual option in the
element whose capacity should be limited with USA is to carry out two analyses, one with and
a view to protecting other more critical and/or one without thee shear keys (essentially one
more difficult to repair components of the bridge; assuming displacement is blocked in the pertinent
a popular design concept in California is that the direction and one with the deck allowed to dis-
capacity of the shear keys should not exceed or be place freely).
close to the shear capacity of the piles (on which
the abutment is supported). Eurocode 8-2 (CEN Joints
2005a) vaguely specifies that the design actions Joints (also called expansion joints which is inap-
for the seismic links (one type of which is shear propriate when they are also designed for seismic
keys) should be derived as capacity design actions) are required to accommodate (with neg-
effects, with the horizontal resistance of the bear- ligible resistance and noise) movements of the
ings assumed equal to zero, without explaining deck due to:
how these effects should be derived (clearly one
possibility could be to relate the strength of shear • Thermal expansion/contraction
keys to that of piles, as mentioned previously). • Shrinkage and creep of concrete
The strength of shear keys can be derived from • Earthquake-induced horizontal movement
sophisticated models such as strut and tie ones or
simplified ones, usually based on the shear fric- There are several types of joints (compression
tion concept. Bozorgzadeh et al. (2006) proposed seal, strip seal, finger plate, sliding plate, modu-
the following relationship for the nominal capac- lar), each of them appropriate for up to a certain
ity of a shear key: design movement of the bridge. In bridges
designed for high seismic actions, the joint gap
mf cos a þ sin a might result as quite substantial (over 200 mm). It
Vn ¼ Auf f su (15)
1 mf tan b is noted that current seismic codes like Eurocode
8-2 require the joint gap to satisfy
where mf is a kinematic friction coefficient, b is
the angle of inclined face of the shear key (Fig. 19 dED ¼ dE þ dG þ c2 dT (16)
left), a is the angle of kinking of the vertical bars
with respect to the vertical axis (recommended where dG is the long-term horizontal displace-
value from test results 37 ), and fsu is the ment due to permanent and quasi-permanent
(ultimate) tensile strength of the vertical rein- actions (posttensioning, shrinkage, and
forcement that has an area Avf. Such models creep), dT the displacement due to thermal
strongly depend on the friction coefficient that actions ( c2 ¼ 0:5 for road bridges), and dE is
varies substantially depending on the detailing the seismic displacement calculated as
of the shear key; for sacrificial keys with smooth Zmd dEe where dEe is the displacement derived
finishing of the concrete interface, mf = 0.36 is from the analysis for the seismic loading combi-
recommended (Bozorgzadeh et al. 2006), but for nation (with the design spectrum reduced by the
properly detailed joints of adequate roughness, behavior factor q), md is the design ductility
much higher values apply (up to 1.4 for keys (md = q in the common case that the fundamental
monolithically cast with the abutment seat). period of the bridge T To ¼ 1:25Tc where Tc is
Having established a proper value for the the corner period of the design spectrum) and
shear key strength, the key can be modeled sim- is the damping correction factor for the design
ply as a rigid-plastic spring (or “link”) with a spectrum (Z ¼ 1 for x ¼ 5%).
Seismic Analysis of Concrete Bridges: Numerical Modeling, Fig. 20 Pushover curves for the longitudinal
direction of an overpass bridge (Kappos and Sextos 2009); “E” denotes the design earthquake level
When large movements (dEd > 100 mm) have joints of US (and New Zealand) bridges, with a
to be accommodated, modular joints are used, view to preventing unseating during earthquakes
wherein sealing elements and rail elements are stronger than the design one. In Europe this is
coupled; in this case intermediate elements seen as an option mainly in retrofitting of existing
(rails), supporting elements, and linkage ele- bridges with insufficient seating lengths.
ments (e.g., folding trellis) causing equal gap Modeling of joints in seismic analysis is
widths are needed, and the total movement essential, since the bridge boundary conditions
accommodated by the joint is the sum of the are drastically different when joints are open or
distances of the rails. closed. For instance, in the longitudinal direction
It is common practice in Europe to provide of the bridge, there are at least two joints at the
substantial joint gaps (as well as ample seat ends (over the seat-type abutments) that are
lengths) and make sure that the joints remain essential for accommodating “non-seismic” dis-
open during the design earthquake, without car- placements (dG and dT); when either of these
rying out specific verifications for the case that joints closes during the earthquake, substantial S
the gap closes. In California, “dual” analysis is forces are transferred to the abutment-backfill
the recommended (by Caltrans) practice, wherein system (see section “Abutments and Backfills”
the bridge is analyzed assuming either free move- and Fig. 20 later in this section), and the horizon-
ment or full restraint at the “compression end” of tal forces developed in the bridge-abutment-
the bridge (movement is always free at the “ten- backfill system can be much higher than those
sion end” where the deck moves away from the developed when longitudinal movement was
abutment) and taking the most unfavorable free.
response quantities from either set of analyses; Proper modeling of joints requires nonlinear
clearly this is conservative and might result in analysis using the gap (or other special link)
increased costs. Another interesting difference elements currently available in most software
in US and European practice is that restrainers packages. This element is often combined in
(either cables or bolt linkages, the bolts passing series with the spring (or link) elements modeling
through holes in the deck) are often used at the the bearings and/or shear keys at the abutment.
It is noted that although the gap element has an presented. These will be followed by a case
elastic behavior subsequent to gap closure, the study involving application of different analysis
analysis of the bridge is nonlinear, as the bound- method to an actual concrete bridge.
ary conditions change during the analysis. The
importance of capturing the effect of joint closure Code-Prescribed Analysis of Concrete Bridges
on the seismic behavior of a bridge can be seen in Among the important documents (codes-
Fig. 20, where pushover curves are given for the guidelines) for the analysis of concrete bridges
longitudinal direction of a typical overpass are those regularly published by Caltrans (2013,
(whose abutment system is shown in Fig. 12). latest version) and the ACI ( 2014, latest version)
Recall that two different soil conditions were in the USA and Eurocode 8-2 (CEN 2005a) in
studied; this has a noticeable effect on the initial Europe. Concrete bridges are the sole type cov-
stiffness of the bridge, but as soon as the longi- ered in the ACI Report 341.2, while the Caltrans
tudinal gap (120 mm in this case) closes, a dras- Criteria and Eurocode 8-2 also cover bridges
tic increase in both stiffness and strength is made of other materials (steel and composite).
noticed, as the abutment-backfill system is now For retrofitting of bridges a comprehensive doc-
activated. The final failure of the bridge is esti- ument is the FHWA (2006) Manual. Some key
mated to take place during this second stage of aspects of these documents specific to the analy-
the response (at a displacement of almost twice sis of concrete bridges are discussed in the fol-
the gap length) and is attributed to lowing. Other important sections relevant to the
unrecoverable damage to the soil behind the design of concrete bridges are those prescribing
abutment (50 % loss in strength), while the the procedures for resistance verification of
piers are still well within their rotational capac- reinforced concrete sections and the detailing of
ity (35–49 %). A different failure mechanism R/C members (piers, abutments, and retaining
(exceedance of available ductility of piers) walls), which fall beyond the scope of this entry
would have been predicted had the end support that focuses on modeling for structural analysis.
been simulated as longitudinal restraint (as per Of course, calculation of R/C member strength
the Caltrans simplified approach). and ductility (especially for piers) is essential for
nonlinear analysis of concrete bridges; aspects of
this issue are covered in the case study presented
Bridge Analysis Methods later in section “Comparative Case Study.”
Methods of analysis can be classified as: • All types of analysis are permitted for concrete
bridges, the equivalent static procedure being
• Elastic (equivalent) static subject to a number of limitations regarding
• Elastic dynamic (response spectrum) the effect of higher modes; this type of analy-
• Inelastic static (pushover) sis is usually suitable for the longitudinal
• Inelastic dynamic (response history) direction of straight bridges, which is domi-
nated by a single mode (that is often the fun-
The basics of all these methods are presented damental mode of the bridge). Three versions
in a series of articles in the encyclopedia and will of the method (Rigid Deck Model, Flexible
not be repeated herein. Instead, some aspects of Deck Model, Individual Pier Model) are pre-
analysis (other than member modeling that was scribed by Eurocode 8-2 (see basic aspects of
presented in section “Modeling of Bridge Com- these methods in Vayas-Iliopoulos (2014)).
ponents”) specific to concrete bridges will be The reference method in practical design is
briefly described, and the current trends in the (elastic) dynamic response spectrum anal-
nonlinear static (pushover) analysis of bridges ysis, while nonlinear methods are only used in
(not covered elsewhere here) will also be practice for the verification of the design of
some important bridges that have initially – The most important factor in reducing the
been designed using response spectrum anal- elastic response spectrum to the design one
ysis. Nonlinear methods are much more com- (which is, in fact, an inelastic spectrum
mon in the case of assessment of existing except when elastic response is foreseen
bridges. The FHWA (2006) Retrofit Manual for the design seismic action) is the force
provides sufficient guidance (especially with reduction factor, called behavior factor
respect to estimating strength and deformation (q) in Eurocode 8 and response modifica-
capacity) for the application of both the push- tion factor (R) in the US codes. Values for q
over and response-history analysis methods, for concrete bridges depend on whether the
which are covered more briefly in Eurocode bridge is designed as ductile or “limited
8-2; notably, assessment of existing bridges is ductile.” For ductile concrete bridges with
not currently covered by the Eurocode pack- vertical piers (working in prevailing bend-
age (it is one of the issues that will be added at ing) q ¼ 3:5lðas Þ, where lðas Þ ¼ 1:0 for
the next stage of development). ðas Þ 3 (as ¼ Ls =h is the shear span ratio
• Application of elastic methods to bridges is of the pier, where Ls is the distance from
the same as for other structures, but there are the plastic hinge to the point of zero
two aspects specific to concrete bridges that moment and h is the depth of the cross
have to be properly addressed: section in the direction of flexure of the
– In most concrete bridges both prestressed plastic hinge) and lðas Þ ¼ √ðas =3Þ for
and ordinary (non-prestressed) concrete are 3 > ðas Þ 1. For ductile concrete bridges
used, the former for the deck, the latter for with piers consisting of inclined struts (e.g.,
the piers and abutments. The damping ratio V shaped), q ¼ 2:1lðas Þ. For limited duc-
x is different for these two materials tile concrete bridges, the corresponding
(primarily due to the different degrees of values are 1.5 (vertical piers) and 1.2
cracking in each), i.e., 5 % for reinforced (inclined struts). For abutments rigidly
concrete and 2 % for prestressed concrete. connected to the deck (integral bridges)
In EC8-2 this is accounted for by consider- q = 1.5, except in “locked-in” structures,
ing the response spectrum for an equivalent i.e., bridge structures whose mass essen-
damping ratio: tially follows the horizontal seismic motion
Sxi Edi of the ground (hence, they do not experi-
xef ¼ (17) ence significant amplification of the hori-
SEdi
zontal ground acceleration), in which case
where Edi is the deformation energy induced q = 1. These values, as well as similar ones
in member i by the seismic action. This specified by the American Code (AASHTO
quantity is not a standard output of common 2010), are in many cases conservative. In a S
structural analysis programs, and in order to recent study evaluating the actual force
avoid ad hoc spreadsheet calculations, reduction factors for existing bridges in
designers often prefer to simply use an aver- Europe, Kappos et al. (2013) found that in
age value of 3.5 %. It is worth noting here all bridges studied the available q values
that Edi is not the energy dissipation through were higher than those used for design in
yielding mechanisms, but rather refers to the both the longitudinal and transverse direc-
pre-yield state; hence, it is not appropriate to tions. In fact, in many cases the code-
consider a spectrum for x = 5 % on the basis specified values (in particular those of
that only reinforced concrete members yield AASHTO for single-column bents) seem
and dissipate energy (prestressed concrete to significantly underestimate the actual
decks remain quasi-elastic as already men- energy dissipation capacity of concrete
tioned in section “Deck”). bridges. Seen from another perspective,
this is a clear indication that modern brid- significant mode of the bridge and construct
ges possess adequate margins of safety and the pushover curve (base shear Vbn
are able to withstand seismic actions that vs. displacement ucn of the “control” or “mon-
are often substantially higher than those itoring” point) for each mode; sn* are loading
used for their design. This high perfor- patterns; hence, the relative significance of
mance is due to their ductility, as well as each mode is not accounted for at this stage;
their overstrength; previous studies that this will be done at Step 5, through the target
have ignored the latter led to unrealistically displacement for each modal pushover analy-
low estimation of q-factor values. sis. Gravity loads are applied before each
MPA and P-D effects are included, if signifi-
Nonlinear Static Analysis of Concrete Bridges cant (e.g., in bridges with tall piers).
Interesting and useful work has been carried out Step 3: The pushover curve must be idealized as a
in the last decade on nonlinear (inelastic) static, bilinear curve so that a yield point and ductil-
also known as pushover, analysis of bridges; nev- ity factor can be defined and subsequently
ertheless this is clearly less than that for build- used to appropriately reduce the elastic
ings. A recent book presenting all available response spectra representing the seismic
methods for pushover analysis of bridges is that action considered for assessment. This ideali-
by Kappos et al. (2012), which also includes a zation can be done in a number of ways, some
substantial number of case studies involving the more involved than others; it is suggested to
comparative application of several methods. Due do this once using the full pushover curve (i.e.,
to space limitations, only one approach will be analysis up to “failure” of the structure,
presented herein which, in the writer’s opinion, defined by a drop in peak strength of about
combines sufficient accuracy with relatively lim- 20 %) and the equal energy absorption rule
ited effort and the possibility to be applied using (equal areas under the actual and the bilinear
available (commercial) software tools, with very curve). It is noted that the remaining steps of
limited need for additional spreadsheet calcula- the methodology can be applied even if a
tions; in fact, software for “single-run” applica- different method for producing a bilinear
tion of the method is currently at an advanced curve is used.
level of development. The method is usually Step 4: Several procedures are available (FEMA
referred to as (multi-)modal pushover analysis 1997; Chopra and Goel 2002; CEN 2004, all
(MPA); it was presented in a comprehensive referring to buildings) for defining the earth-
form for buildings by Chopra and Goel (2002) quake displacement demand associated with
and was extended to bridges by Paraskeva each of the pushover curves derived in Step
et al. (2006). The key idea is to perform multiple 3. Paraskeva et al. (2006) adopted the capacity
pushover analyses of the structure, one for each and demand spectra procedure based on
significant mode, and combine statistically the inelastic demand spectra (Fajfar 1999);
resulting displacements and rotations. The steps hence, Step 4 consists in converting the ideal-
involved in the latest version of the method ized Vbn ucn pushover curve of the multi-
(Paraskeva and Kappos 2010), which includes a degree-of freedom (MDOF) system to a
number of improvements, are summarized in the “capacity diagram” (Fig. 21). The base shear
following. forces and the corresponding displacements
in each pushover curve are converted to spec-
Step 1: Compute the natural periods, Tn, and tral acceleration (Sa) and spectral displace-
mode shapes, fn, for linearly elastic vibration ments (Sd), respectively, of an equivalent
of the structure. single-degree-of-freedom (SDOF) system,
Step 2: Carry out separate pushover analyses for using the relationships (Chopra and Goel
force distribution sn* = mfn for each 2002):
Vbn Sa
Demand
diagram
Idealized
curve
2
1
anKn
1 an 4π2
Vbny Tn
Actual
curve
4π2
Kn 2
1 1 Tn
Urny Ur n S dn Sd
Seismic Analysis of Concrete Bridges: Numerical Modeling, Fig. 21 Idealized pushover curve of the nth mode of
the MDOF system and corresponding capacity curve for the nth mode of the equivalent inelastic SDOF system
Vbn lead to equally good results as far as the final

Sa ¼ (18a)
M n response quantities are concerned. For practi-
cal purposes, a good selection is the deck point
above the most critical support (pier or abut-
ucn ment) of the bridge.
Sd ¼ (18b)
Gn fcn Step 6: In this step, a correction is made of the
displacement of the monitoring point of the
wherein fcn is the value of fn at the control bridge, which was calculated at the previous
(or “monitoring”) point, M n ¼ Ln Gn is the step. The correction is necessary only for cases
effective modal mass, with Ln ¼ fTn m 1 , that significant inelasticity develops in the
Gn ¼ Ln =Mn is a mass participation factor, structure. If the structure remains elastic or
and Mn ¼ fTn m fn is the generalized mass, close to the yield point, the MPA procedure
for the nth natural mode. For inelastic behav- suggested by Paraskeva et al. (2006) is used to
ior, estimation of the displacement demand at estimate seismic demands for the bridge. The
the monitoring point is made with the aid of response displacements of the structure are
inelastic spectra. evaluated by extracting from the database of
Step 5: Since the displacement demand calcu- the individual pushover analyses the values of
lated in Step 4 (for each mode) refers to the desired responses at which the displace-
SDOF systems with periods equal to those of ment at the control point is equal to ucn (see S
the corresponding modes, the next step is to Eq. 18b). These displacements are then
correlate these displacements to those of the applied to derive a new vector fn0 , which is
actual bridge. Hence, Step 5 consists in the deformed shape (affected by inelastic
converting the displacement demand of the effects) of the bridge subjected to the given
nth mode inelastic SDOF system to the peak modal load pattern. The target displacement at
displacement of the monitoring point, ucn of the monitoring point for each pushover analy-
the bridge, using Eq. 18b. The selection of this sis is calculated again with the use of fn0 ,
point is a critical issue for MPA of bridges, solving Eq. 18b for ucn’, and recalculated Gn
and as discussed by Paraskeva et al. (2006), using fn0 .
several choices of monitoring point are Step 7: The response quantities of interest
acceptable as long as the derived pushover (displacements, plastic hinge rotations, forces
curve has a reasonable shape, but they do not in the piers) are evaluated by extracting from
the database of the individual pushover ana- at the bottom of Fig. 2 is closed. The Greek
lyses the values of the desired responses rn, Seismic Code design spectrum (similar to that
due to the combined effects of gravity and of EC8) scaled to a PGA of 0.16 g was used for
lateral loads for the analysis step at which the seismic design. The design spectrum
displacement at the control point is equal to corresponded to ground category “B” (close to
ucn (or ucn’). ground “C” in the final version of EC8 (CEN
Step 8: Steps 3–7 are repeated for as many modes 2004)). The bridge was designed as a ductile
as required for sufficient accuracy; there is structure (plastic hinges expected in the piers)
little merit in adding modes whose participa- for a behavior factor q = 2.4.
tion factor is very low (say less than 1 %), and The bridge was analyzed applying a number of
application of the method to a number of brid- popular assessment procedures, i.e.:
ges shows that it is not necessary to assure that
the considered modes contribute to 90 % of the • Modal analysis
total mass. • “Standard” pushover analysis (SPA) (first
Step 9: The total value for any desired response mode loading)
quantity (and each level of earthquake inten- • Pushover analysis for a “uniform” loading
sity considered) can be determined by com- pattern (as required by Eurocode 8 (CEN
bining the peak “modal” responses rno using 2004) and by the ASCE Standard 41–06
an appropriate modal combination rule, e.g., (ASCE/SEI 2007))
SRSS or CQC. This simple procedure is used • Modal pushover analysis (MPA) as proposed
for both displacements and plastic hinge rota- in Paraskeva et al. (2006)
tions, which are the main quantities commonly • Improved modal pushover analysis as pro-
used for seismic assessment of bridges. If posed by Paraskeva and Kappos (2010)
member forces (e.g., pier shears) have to be • Nonlinear response-history analysis (NRHA),
determined accurately, a more involved pro- for artificial records closely matching the
cedure of combining modal responses should demand spectrum (see Paraskeva et al. 2006)
be used, consisting in correcting the bending
moments at member ends (whenever yield All inelastic analyses were carried out using
values were exceeded) on the basis of the the SAP2000 software package (CSI 2011). Plas-
relevant moment versus rotation (M – y) dia- tic hinging in the piers had to be modeled slightly
gram and the value of the calculated plastic differently in the NRHA and the pushover analy-
hinge rotation; this procedure blends well with sis, due to limitations of the software used. More
the capabilities of currently available specifically, nonlinear rotational spring elements
software. were used in the finite element models used in
NRHA, while the built-in beam hinge feature of
Comparative Case Study SAP2000 was implemented in the models set up
The overpass shown in Fig. 2 (some aspects of its for pushover analysis. In both cases, though, the
modeling were discussed in section “Modeling of same moment versus rotation (M-y) relationship
Bridge Components”) has three spans and total was used (i.e., bilinear with 2–6 % hardening,
length equal to 100 m, typical in modern motor- depending on the calculated ultimate moment),
way construction. Piers have a cylindrical cross with input parameters defined from fiber analysis
section, while the pier heights are 8 m and 10 m. performed for each pier section, utilizing the
The deck is monolithically connected to the piers, in-house developed computer program
while it rests on its two abutments through elas- RCCOLA-90.
tomeric bearings; movement in both the longitu-
dinal and transverse directions is initially allowed Nonlinear Static Analysis
at the abutments, but transverse displacements The dynamic characteristics of the bridge were
are restrained whenever the 150 mm gap shown determined using standard modal (eigenvalue)
mode1: T1 =0.77s, M1*/Mtot =63.4% mode2: T2 =0.65s, M2 */Mtot =31.6%

(a) (b, d) (c) (c) (b) (d) (a)
mode3:T3 =0.41s, M3 */Mtot =2.3%
(d) (b) (a, c)
mode1 mode2 mode3

xcentre mass /(total length) (a) 0.50 0.50 0.50
xSDOF/(total length) (b) 0.73 0.08 0.44
xmax(total length) (c) 1.00 0.00 0.50
xcritical pier /(total length) (d) 0.73 0.27 0.27
Seismic Analysis of Concrete Bridges: Numerical Modeling, Fig. 22 Modal force distribution, location of the
equivalent SDOF systems, and modal parameters for the main transverse modes of the overpass bridge
analysis. Figure 22 illustrates the first three trans- structure, are not identical but rather the esti-
verse mode shapes of the overpass bridge, mated deformed shape of the bridge depends on
together with the corresponding participation fac- the monitoring point selected for drawing the
tors and mass ratios, as well as the locations of pushover curve for each mode.
alternative monitoring points for each mode. For inelastic behavior, it appears that the esti-
Consideration of the modes shown in Fig. 22 mated values of ucn are different not only because
assures that more than 90 % of the total mass in of the deviation of the elastic mode shape fn
the transverse direction is considered. For MPA, from the actual deformed shape of the structure
applying the modal load pattern of the nth mode but also due to the fact that the spectral displace-
in the transverse direction of the bridge, the ment Sd is dependent on the selection of monitor-
corresponding pushover curve was constructed ing point if the structure exhibits inelastic
and then idealized as a bilinear curve (Fig. 21). behavior (due to the bilinearization of the capac-
As noted under Step 4 of the MPA procedure (see ity curve). An improved target displacement of
section “Nonlinear Static Analysis of Concrete the monitoring point is calculated (from Eq. 18b)
Bridges”), the inelastic demand spectra method using fn0 , the actual deformed shape of the struc-
was used for defining the displacement demand ture (see Fig. 23), while the spectral displacement
for a given earthquake intensity. remains the same. The response quantities of S
Figure 23 illustrates the deck displacements of interest are evaluated by extracting from the
the bridge derived using pushover analysis for “database” the values of the desired responses,
each mode independently, as well as the MPA rn, for the analysis step at which the displacement
procedure initially proposed in Paraskeva at the control point is equal to ucn0 (the improved
et al. (2006). If the structure remains elastic for estimate of ucn derived on the basis of fn0 ).
the given earthquake intensity, both spectral dis- Figure 24 illustrates the deck displacements of
placement Sd and the product Gnfn will be inde- the overpass bridge, calculated from MPA using
pendent of the selection of the control ucn0 as target displacement for each mode. It is
(monitoring) point; this means that deck displace- noted that, due to the approximations involved in
ments are independent of the location of the mon- the capacity and demand spectra procedure, deck
itoring point. On the contrary, it was found that displacements derived with respect to different
deck displacements derived with respect to dif- control points are not the same, but differences
ferent control points, for inelastic behavior of the are significantly reduced and results are deemed

Concrete Bridges: 0.07 mode 1 control point:M2 (critical pier,SDF)
Numerical Modeling, control point:A2 (max displ)
0.06
Fig. 23 “Modal” deck
displacements derived with 0.05
u (m)
respect to different control 0.04
points – inelastic behavior 0.03
of the overpass bridge 0.02
(Ag = 0.16 g)
0.01
0
0 10 20 30 40 50 60 70 80 90 100
–0.01
A1 M1 M2 A2
0.07
0.06 mode 2
control point:M1(critical pier)
0.05
0.04 control point:A1(max displ)
0.03
u (m)
0.02
0.01
0
0 10 20 30 40 50 60 70 80 90 100
–0.01
–0.02 A1 M1 M2 A2

Concrete Bridges:
0.07 mode 1
Numerical Modeling,
Fig. 24 “Modal” deck 0.06 controlpoint: M2 (critical pier)
displacements derived with 0.05 controlpoint: A2 (max displ)
respect to different control
points using urn0 as target 0.04
u (m)
displacement according to 0.03

the improved MPA
procedure – overpass 0.02
bridge (Ag = 0.16 g) 0.01
0
0 10 20 30 40 50 60 70 80 90 100
–0.01
0.07
0.06 mode 2
0.05
control point:M1(critical pier)
0.04 control point:A1(max displ)
0.03
u (m)
0.02
0.01
0
0 10 20 30 40 50 60 70 80 90 100
–0.01
–0.02 A1 M1 M2 A2

Concrete Bridges: RHA
Numerical Modeling, 0.07 MPA 0.16g
Fig. 25 Response to the 0.06 SPA(mode1)
design earthquake SPA(uniform load pattern)
(Ag = 0.16 g) and to twice 0.05
the design earthquake
0.04
u (m)
(Ag = 0.32 g) calculated
from SPA, MPA, and 0.03
NRHA: deck
displacements of the 0.02
overpass bridge 0.01
0.00
0 10 20 30 40 50 60 70 80 90 100
–0.01
A1 M1 M2 A2
0.12
RHA
0.32g
0.10 MPA
SPA(mode1)
0.08 SPA(uniform load pattern)
0.06
u (m)
0.04
0.02
0.00
0 10 20 30 40 50 60 80 70 90 100
–0.02
A1 M1 M2 A2
acceptable for all practical purposes. Paraskeva analysis, the latter considered as the most rigor-
and Kappos (2010) also studied other bridges ous procedure to estimate seismic demand.
with different configurations and noted that the To this effect, a series of NRHAs was performed
differences between deck displacements derived using five artificial records compatible with the
with respect to different control points, as well as design elastic spectrum. The Newmark g = 1/2,
the improvement in the prediction of deck b = 1/4 integration method was used, with time
displacements using the procedure described step Dt = 0.0025 s and a total of 10,000 steps S
in section “Nonlinear Static Analysis of Concrete (25 s of input). A uniform damping value of 5 %
Bridges,” are more significant in bridges was assumed for all modes of vibration, while
longer than the overpass of Fig. 2, especially hysteretic damping was accounted for through
if the increased length is combined with signifi- the elastoplastic behavior of the structural
cant curvature in plan, which amplifies the members.
complexity of its dynamic behavior and results The displacements determined by the SPA and
in more significant contribution of the higher MPA procedures were checked against those
modes. from NRHA for increasing levels of earthquake
excitation, as shown in Fig. 25. It is noted that the
Reliability of Static Analysis Procedures deck displacements shown in the figures as the
Results of the standard and modal pushover NRHA case are the average of the peak displace-
approaches were evaluated by comparing them ments recorded in the structure during the five
with those from nonlinear response-history response-history analyses. In this study the
displacement demand was estimated indepen- deemed better than that resulting from using a
dently in static and dynamic (response-history) single modal load pattern.
inelastic analysis, whereas in some previous stud- Additional case studies reported by Paraskeva
ies comparisons of displacement profiles were and Kappos (2010) confirm that SPA predicts
made assuming the same maximum displacement well (i.e., matches closely the values from the
in both cases; the choice adopted here is deemed NRHA approach) the maximum transverse dis-
as more relevant for practical applications, as it placement, when applied to bridges of regular
permits an evaluation of all aspects of the pro- configuration, where the higher mode contribu-
posed procedure (including the uncertainty in tion is not significant. In such cases the improve-
estimating displacement demand in pushover ment of the displacements derived by the MPA
analysis). procedure is not significant even for high levels of
As shown in Fig. 25, the MPA procedure pre- earthquake excitation.
dicts well (i.e., matches closely the values from
the NRHA approach) the maximum transverse
displacement. On the other hand, the SPA proce- Summary and Concluding Remarks
dure underestimates the displacements of the
deck at the location of the abutment A1 and the It is clear that today the bridge engineer has at
first pier of the bridge, compared to the more his/her disposal a set of powerful analysis tools
refined NRHA approach. This is not surprising that can be used for the seismic design or assess-
if one notes the differences between the first two ment of any bridge type. The potential of these
mode shapes in the transverse direction (Fig. 22), tools, when properly utilized, was revealed by
which are strongly affected by torsion (they con- their success in predicting the response of bridges
tribute more than 90 % of the torsional response, tested under high levels of earthquake actions
as well as over 90 % of the transverse response of that caused substantial amounts of inelasticity;
the bridge) due to the unrestrained transverse an example was presented in section “Compara-
displacement at the abutments (until the tive Case Study.” The information on bridge
150 mm gap closes), combined with the different modeling presented in section “Modeling of
stiffness of the two piers caused by their different Bridge Components” offers to researchers and
height. What is essentially achieved by the MPA designers the necessary information regarding
is the combination of these first two modes (the the available models for the various parts of the
3rd transverse mode is not important in this par- bridge (deck, bearings and shear keys, isolation
ticular bridge), each of which dominates the and energy dissipation devices, piers, foundation
response in the region of the corresponding abut- members), as well as tools for modeling the
ment. In the case of applying ground motions dynamic interaction between piers, foundation,
with twice the design earthquake intensity (also and soil, as well as the abutment-embankment-
shown in Fig. 25), where the structure enters superstructure system. It also provides informa-
deeper into the inelastic range and higher mode tion on important parameters that help ensuring
contributions become more significant (without that inelastic analysis of bridge earthquake
substantial alteration of the mode shapes), it is response is conducted properly. It has to be
noted that the displacement profile derived by the emphasized in this respect that the power and
MPA method tends to match that obtained by the versatility of the analysis tools also makes the
NRHA, whereas SPA predictions remain poor. results particularly sensitive to improper
Note that, regardless of earthquake intensity, the application.
uniform loading pattern (also shown in Fig. 25) Special emphasis was given to modeling of
fails to capture the increased displacements piers (section “Piers and Their Foundations”), as
toward the abutments; nevertheless its overall these members are both the ones wherein energy
prediction of the displacement profile could be dissipation through plastic hinging is intended to
occur (unless a seismic isolation system is used) bridges, these depend primarily on the configura-
and those whose inelastic response is relatively tion of the bridge analyzed. As a rule, the longi-
easier to model in inelastic analysis (compared, tudinal direction of the bridge is the easier one to
for instance, to the abutment-backfill system or analyze, and even simple, equivalent static elastic
some foundation types or, indeed, some types of methods can lead to a reasonable design. In most
joints). Having said this, it is also clear from the other cases dynamic analysis is required to prop-
material presented in this entry that proper erly capture the higher mode effects that are
modeling of the other components of the bridge, important, especially in the areas close to the
even those that are typically assumed to remain abutments. When the expected plastic mecha-
elastic during the seismic excitation (such as nism does not involve more-or-less uniform
prestressed concrete decks), is also important, yielding in the energy dissipation zones of the
since, through their stiffness characteristics, bridge (this is the case of irregular pier configu-
they affect the dynamic characteristics of the rations), inelastic analysis is strongly
bridge and the way seismic actions are transferred recommended for verifying the design initially
to the dissipating zones. Of great importance is carried out using standard modal (response spec-
also the modeling of the various connections in trum) analysis. Nonlinear analysis is clearly the
the bridge system, i.e., those between piers and preferred choice in the case of assessing existing
deck, abutments and deck, and, in the common bridges not properly detailed for seismic perfor-
case (especially in the transverse direction) that mance; for such bridges the simplest choice is
the movement of the deck is restrained at the standard pushover analysis, but whenever more
location of the abutment, the proper modeling of than one mode affects the response (this is very
the response of the abutment-backfill system. As often the case in the transverse direction of the
noted in section “Abutments and Backfills,” in a bridge) more sophisticated tools like the
practical context and when the main objective of multimodal pushover method presented in sec-
the analysis is the response of the bridge itself tions “Nonlinear Static Analysis of Concrete
(rather than that of the surrounding ground), the Bridges” and “Comparative Case Study” have
recommended solution is to carry out an indepen- to be used. The use of such analytical tools is
dent analysis of the abutment-embankment sys- expected to increase when the software required
tem, determine its resistance curves (in all for applying them in a single run becomes widely
relevant directions), and use them to describe available.
the nonlinear response of the equivalent springs
to which the bridge model will be connected. If
such an analysis cannot be afforded, the Cross-References
properties of these springs can be defined on the
basis of simplified guidelines from the literature. ▶ Buildings and Bridges Equipped with Passive S
For pier-foundation-soil interaction, the existing Dampers Under Seismic Actions: Modeling
literature is more mature and often it is not nec- and Analysis
essary to carry out separate analysis of the system ▶ Seismic Analysis of Steel and Composite
to derive the nonlinear properties of the soil- Bridges: Numerical Modeling
foundation dynamic impedance to be introduced
in the bridge model, especially when surface
foundations are used; in these cases information References
from the literature can be used to account approx-
imately for the interaction with the surrounding AASHTO (American Association of State Highway and
Transportation Officials) (2010) LRFD bridge
ground. design specifications, 5th edn. American Association
Regarding the feasibility and reliability of dif- of State Highway and Transportation Officials,
ferent methods used for the analysis of concrete Washington, DC
ACI (American Concrete Institute) Committee 341 (2014) Kappos AJ, Sextos AG (2001) Effect of foundation type
Seismic analysis and design of concrete bridge sys- and compliance on the lateral load response of R/C
tems (ACI341.2R-14). American Concrete Institute, bridges. J Bridg Eng ASCE 6(2):120–130
Detroit Kappos AJ, Sextos AG (2009) Seismic assessment of
ASCE (American Society of Civil Engineers) (2007) Seis- bridges accounting for nonlinear material and soil
mic rehabilitation of existing buildings – ASCE stan- response, and varying boundary conditions. In:
dard 41-06. American Society of Civil Engineers, Coupled site and soil-structure interaction effects
Reston with application to seismic risk mitigation, NATO
ATC (Applied Technology Council, USA) (1996) Improved science for peace and security, series-C. Springer,
seismic design criteria for California bridges: provisional Dordrecht, pp 195–208
recommendations. ATC report no ATC-32. Applied Kappos AJ, Manolis GD, Moschonas IF (2002) Seismic
Technology Council, Redwood City assessment and design of R/C bridges with irregular
Aviram A, Mackie KR, Stojadinović B (2008) Guidelines configuration, including SSI effects. Eng Struct
for nonlinear analysis of bridge structures in Califor- 24(10):1337–1348
nia. PEER report 2008-03. University of California, Kappos AJ, Saiidi M, Aydinoglu N, Isakovic T (2012)
Berkeley Seismic design and assessment of bridges: inelastic
Bozorgzadeh A, Megally S, Restrepo JI, Ashford SA methods of analysis and case studies. Springer,
(2006) Capacity evaluation of exterior sacrificial Dordrecht
shear keys of bridge abutments. J Bridg Eng ASCE Kappos AJ, Gkatzogias KI, Gidaris I (2013a) Extension of
11(5):555–565 direct displacement-based design methodology for
Caltrans (California Department of Transportation) (2013) bridges to account for higher mode effects. Earthq
Seismic design criteria version 1.7. Caltrans Division of Eng Struct Dyn 42(4):581–602
Engineering Services, Sacramento, California Kappos AJ, Paraskeva TS, Moschonas IF (2013b)
CEN (2005a) Eurocode 8: design of structures for earth- Response modification factors for concrete bridges in
quake resistance – part 2: bridges (EN 1998-2). CEN, Europe. J Bridg Eng ASCE 18(12):1328–1335
Brussels Kappos AJ, Papadopoulos I, Tokatlidis A (2014) Design
CEN (2005b) Structural bearings – part 3: elastomeric of a seismically isolated railway viaduct over axios
bearings (EN 1337-3). CEN, Brussels river in Northern Greece. In: Railways 2014: the sec-
CEN (Comité Européen de Normalisation) ond international conference on railway technology,
(2004) Eurocode 8: design of structures for earthquake Ajaccio, Corsica, 8–11 Apr, Civil-Comp Press, paper
resistance – part 1: general rules, seismic actions and no 80
rules for buildings (EN1998-1-1). CEN, Brussels; Katsaras CP, Panagiotakos TB, Kolias B (2009)
Corrigenda: July 2009 and Jan 2011 Effect of torsional stiffness of prestressed concrete
Chopra AK, Goel RK (2002) A modal pushover analysis box girders and uplift of abutment bearings on
procedure for estimating seismic demands for build- seismic performance of bridges. Bull Earthq Eng
ings. Earthq Eng Struct Dyn 31(3):561–582 7:363–375
Computers & Structures Inc. (CSI) (2011) Mergos PE, Kappos AJ (2012) A gradual spread inelastic-
SAP2000 – version 15.0.1: linear and non linear static ity model for R/C beam-columns, accounting for flex-
and dynamic analysis and design of three-dimensional ure, shear and anchorage slip. Eng Struct 44:94–106
structures. Computers & Structures Inc. (CSI), Ntotsios E, Karakostas C, Lekidis V, Panetsos P,
Berkeley Nikolaou I, Papadimitriou C, Salonikios T (2009)
Fajfar P (1999) Capacity spectrum method based on Structural identification of Egnatia Odos bridges
inelastic demand spectra. Earthq Eng Struct Dyn based on ambient and earthquake induced vibrations.
28(9):979–993 Bull Earthq Eng 7(2):485–501
Fardis MN (1991) Member-type models for the non-linear Papanikolaou VK, Kappos AJ (2009) Numerical study of
seismic response analysis of reinforced concrete struc- confinement effectiveness in solid and hollow
tures. In: Donea J, Jones PM (eds) Experimental and reinforced concrete bridge piers: Methodology.
numerical methods in earthquake engineering. CEC, Comput Struct 87(21–22):1427–1439
JRC, Ispra Paraskeva TS, Kappos AJ (2010) Further development of
FEMA (1997) NEHRP guidelines for the seismic a multimodal pushover analysis procedure for seismic
rehabilitation of buildings, FEMA-273. FEMA, assessment of bridges. Earthq Eng Struct Dyn
Washington, DC 39(2):211–222
FHWA (Federal Highway Administration, USA) Paraskeva TS, Kappos AJ, Sextos AG (2006) Extension of
(2006) Seismic retrofitting manual for highway brid- modal pushover analysis to seismic assessment of
ges part 1 – bridges (FHWA- HRT-06-032). Turner- bridges. Earthq Eng Struct Dyn 35(10):1269–1293
Fairbank Highway Research Center, McLean Priestley MJN, Seible F, Calvi GM (1996) Seismic design
Hwang JS (2002) Seismic design of structures with vis- and retrofit of bridges. Wiley, New York
cous dampers. In: International training program for Vayas I, Iliopoulos A (2014) Numerical modeling guide-
seismic design of building structures. NCREE, Taiwan lines for seismic analysis of steel and composite
Seismic Analysis of Masonry Buildings: Numerical Modeling 2575
bridges. In: Encyclopedia of earthquake engineering. administration (UNIDO 1983); furthermore,

Springer, Berlin being historically one of the earliest forms of
Zhang J, Makris N (2002) Kinematic response functions
and dynamic stiffnesses of bridge embankments. construction, they also house the people’s cul-
Earthq Eng Struct Dyn 31(11):1933–1966 tural heritage, most often being such by them-
selves (Figs. 1, 2, 3, and 4).
Due to their geographic spread, masonry
structures are exposed to different levels of seis-
Seismic Analysis of Masonry mic hazard; in fact, a significant portion of cul-
Buildings: Numerical Modeling tural heritage structures are located in earthquake
prone-areas of Europe, Asia, and South America.
Christos A. Zeris and Elizabeth Vintzileou Recent devastating earthquakes (e.g., in New
Department of Structural Engineering, Faculty of Zealand, Italy, Chile, India, Pakistan, and else-
Civil Engineering, National Technical University where) have shown that existing masonry struc-
of Athens, Athens, Greece tures are quite vulnerable to seismic actions, as
also recorded in many seismic damage reconnais-
sance reports published after major earthquakes
Synonyms (e.g., among others, Hughes et al. 1990; Rossetto
et al. 2009; DesRoches and Comerio 2011).
Adobe; Confined masonry; Finite elements; As a consequence of this fact, social and eco-
Load-bearing masonry; Masonry building nomic requirements for human safety and opera-
reinforced concrete; Reinforced masonry; Seis- bility as well as the need for preservation of
mic analysis; Unreinforced masonry cultural heritage require more and more often
that these structures be analyzed for seismic
actions, in order to be designed (new construc-
Introduction tion), or their seismic vulnerability can be
assessed and evaluated (existing construction)
Masonry building construction encompasses the for the purpose of repair, rehabilitation, and/or
large inventory of structures built worldwide strengthening to current seismic standards. The
solely or partially of individually laid masonry seismic performance assessment of existing
units bonded or not together with some type of masonry buildings also follows performance-
mortar, possibly with the incorporation of rein- based design (PBD) and analysis procedures,
forcement; depending on the constituent mate- similar to other types of structures (e.g., concrete
rials, stacking, and bonding techniques, and steel buildings): hence, in the design of new
different technologies are adopted for building construction, normative regulations, practices,
the masonry elements that comprise the entire and experience of good seismic performance S
building; thus, materials and techniques adopted have been encompassed (e.g., EC6 2005 for grav-
worldwide vary with local customs, socioeco- ity load design and EC8 2004 for seismic design).
nomic conditions, and available technology. On the other hand, codes and guidelines are being
A comprehensive continuously updated descrip- drafted for the assessment and retrofit of existing
tion of the different materials and technologies of masonry buildings (e.g., FEMA-356 2000) or for
construction around the world is given in the cultural heritage structures (Moro 2007;
World Housing Encyclopedia (EERI/IAEE). ICOMOS; ISCARSAH).
Because of their widespread use, masonry Compared to the more recently evolved types
structures house not only the population but also of construction like steel and reinforced concrete,
important social and economic operations such as however, masonry buildings have certain inher-
hospitals and schools, high congregation areas, ent idiosyncrasies, which their modeling for seis-
business, small to medium industry, and civil mic analysis should account for:
2576 Seismic Analysis of Masonry Buildings: Numerical Modeling
Seismic Analysis of Masonry Buildings: Numerical

Seismic Analysis of Masonry Buildings: Numerical Modeling, Fig. 3 The Alcantara bridge (104 AD), Spain
Modeling, Fig. 1 The stone masonry castle of Mycenae, (Source: http://en.wikipedia.org/wiki/Roman_bridge)
Peloponnese, Greece, 2nd millennium BC (Source http://
en.wikipedia.org/wiki/Mycenae)
Seismic Analysis of Masonry Buildings: Numerical Seismic Analysis of Masonry Buildings: Numerical
Modeling, Fig. 2 The Byzantine church of Hagia Sophia Modeling, Fig. 4 The Aoos river bridge (eighteenth
(360 AD), Istanbul, Turkey century), Hepirus, Greece
(i) Masonry buildings are both non-engineered, ASCE 2013; EC6 2005). In addition to the
namely, structures built with traditional possible lack of proper design, they are also
means and experience, and engineered, characterized by the usual problems in the
namely, structures designed and constructed load-bearing system similar to other struc-
following engineering principles and, more tures, such as: irregularities in plan and/or
recently, code regulations (e.g., EC6 2005; elevation, improper foundation conditions,
EC8 2004). In fact, earlier engineered a history of (possibly undocumented) mod-
masonry buildings have been designed for ifications in plan and elevation, and the
gravity loads only; only some types of decay of the material properties under envi-
recent masonry construction, following the ronmental exposure.
evolution of seismic regulations, have also (ii) Depending on the prevailing socioeconomic
been designed for earthquake (IAEE 2008; conditions, masonry buildings are and have
Seismic Analysis of Masonry Buildings: Numerical Modeling, Fig. 5 Residential masonry buildings in Greece,
built in the nineteenth and early twentieth century, reflecting different levels of socioeconomic status
been constructed using a variety of locally poorly connected to the masonry for lateral
available technologies and materials (Fig. 5). load transfer, the relatively high flexibility
Often, currently acceptable levels of quality of the floor diaphragms, and the way the
control and supervision were not enforced. masonry wall elements are constructed and
New masonry buildings in developed coun- tied together through the thickness and/or at
tries may use units manufactured industrially their intersections. Seismic analysis model-
and with quality control; this is not the case for ing techniques should not overlook these
existing buildings in all countries or new particularities, if the analysis results are to S
masonry construction in underdeveloped be reliable.
countries (see typical examples offered in the
World Housing Encyclopedia, EERI/IAEE). Depending therefore on the problem at hand,
(iii) Masonry, unless suitably reinforced or con- reliable modeling for seismic analysis of masonry
fined, as in the case of new construction, structures will have to account of these building
cracks and fails in a brittle manner. Seismic characteristics, in order to reliably predict the
analysis methods for existing or historical damages expected and to identify the methods
unreinforced masonry buildings need to and extent of intervention required for strength-
take this fact into account. ening these structures. Following a brief review
(iv) Masonry buildings suffer from structural of masonry building characteristics, the methods
inadequacies in the load-bearing system and limitations of different methodologies
inherent in this type of construction, such adopted for the seismic analysis of masonry con-
as: the presence of diaphragms which are struction are subsequently considered.
Modeling Techniques for Seismic They assist masonry in taking shear and bending
Analysis of Masonry Buildings: (in and out of plane) while they also contribute to
Classification and Definitions the confinement of plain masonry, thus enhanc-
ing the ductility of the system.
Masonry Materials Used In cultural heritage buildings in Europe and
According to the type of masonry units elsewhere, most frequent construction types of
employed, masonry is classified as stone masonry stone masonry are double-leaf masonry, made of
(Fig. 5), industrially manufactured clay brick two leaves, either independent or connected
(solid, hollow), site-produced mud brick or plinth between them with sporadic header stones
(also known as adobe masonry), or industrially (Fig. 6), and three-leaf masonry, made of two
manufactured hollow or solid concrete masonry independent leaves, with the space between
units (CMU) in fully, partially grouted, or them being filled with a more or less loose
ungrouted construction. Other industrially material of poor mechanical properties. Other
manufactured types of block are also used world- types of masonry, like cavity masonry or timber
wide such as lightweight concrete, cinder, fly ash, reinforced one, are also quite frequent in historic
or autoclaved concrete, among others. Depending structures.
on the type of construction, reinforcement is also Finally, the material that bonds the masonry
used to enhance the bearing capacity of masonry units together may be none (also called dry con-
elements and to provide ductility to the structure. struction), mud, or different types of mortar such
Another construction system (covered by modern as lime, lime–pozzolan, lime mortar reinforced
codes, e.g., EC6 2005) is that of confined with animal hair, or, in case of modern masonry
masonry: horizontal and vertical reinforced con- construction, cement or lime–cement.
crete elements are provided, at distances
depending on the dimensions of the building Load-Bearing Function of the Masonry
and on the seismicity of the region. The ties, The type of load path of the vertical and trans-
constructed during the construction of the verse loads down to the foundation defines two
masonry, function as linear tensioned members. basic types of masonry:
Seismic Analysis of Masonry Buildings: Numerical Argos, -Peloponnese, thirteenth to fourteenth century
Modeling, Fig. 6 Unreinforced masonry construction BC). (b) Double-leaf masonry wall. (c) Three-leaf rubble
technologies. (a) Dry masonry construction (Tiryns, stone masonry. (d) Cavity wall (Van, Turkey)
(i) Load-bearing (LB) masonry, bearing all Because of this contact mechanism, practical
the gravity and lateral loads from floors and seismic analysis of this type of masonry con-
roof to the foundation. Thus, the masonry struction involves macromodeling using axial
elements themselves are responsible to pro- load-bearing struts, as briefly discussed in sec-
vide overall lateral stability and to transfer tion “Modeling of NLBM Infilled Frame Build-
the seismic inertia loads under in-plane and ings for Seismic Analysis.”
out-of-plane actions, down to the foundation,
through combined flexural, axial, shear, or Masonry Wall Construction Types and
rigid body rocking mechanisms. Seismic Resisting Mechanisms
analysis of this type of masonry involves Masonry walls, whether LBM or NLBM, range
adequate modeling of the entire load-bearing from single-leaf or multiple-leaf walls (Fig. 6)
structural system response at the individual with interior cavities among the leaves, which are
element level, including the diaphragms and either void (e.g., the common double-leaf walls with
the foundation, accounting for both in-plane space for sliding window frames) or are filled with
and out-of-plane response, as discussed in rubble (typical for old masonry buildings, cultural
the next section. As a special case of LB heritage buildings included) or concrete (in modern
masonry construction, masonry arches are masonry construction of reinforced cavity walls
constructed with keystone elements at the according to EC6 2005). Multiple-leaf walls may
apex, in order to provide the path of the line be tied to each other in modern construction (e.g.,
of thrust to the foundation. modern veneer brick masonry walls) or, as often the
(ii) Non-load-bearing (NLB) masonry, in case in existing structures, they are untied.
which the bearing function is provided by The laying of the masonry units varies
a structural skeleton made of other structural according to the local techniques, the material,
materials such as steel and timber of and the production form of the unit. Industrial
reinforced concrete (RC). Such is the case units are laid in courses (use of header courses or
of the widely adopted masonry panel infilled random laying construction and keystones at the
RC building construction, a form of confined intersections) and the provision of collar joints
masonry, whose modeling is covered in sec- filled with mortar (and possibly steel) and/or the
tion “Modeling of NLBM Infilled Frame inclusion of bond beam elements. The vertical
Buildings for Seismic Analysis.” elements are traditionally built on a stone or
brick or concrete footings, on which they rest or
If anything else, NLBM infill panels have to are tied to with reinforcement. Often, interior
support with adequate resistance capacity and walls are in fact lighter construction partitions of
in a stable manner their own inertia forces as brick or wood not tied to the load-bearing system,
these are materialized at the building elevation or the wall intersections were poorly connected. S
relative to ground in which they are supported. Irrespective of material and load-bearing type,
Furthermore, through the deformation compati- masonry is classified according to its resisting
bility between the infill and the confining mechanisms as unreinforced masonry, reinforced
frame structure, they are forced to resist seis- masonry (prestressing possibly included), and
mic load through a friction contact separation confined masonry, depending on whether rein-
mechanism with the frame under lateral load forcement or additional confining elements
response. Therefore, unless specifically isolated within the masonry are used. The use of each
by adequate details from the bearing frame type in new constructions in seismic regions
(over the entire lateral inelastic deformation depends on the seismicity of the region, on the
expected), NLBM panels are seismically bear- number of stories, etc. (see, e.g., EC8 2004).
ing, and consequently, even NLBM elements
need to be included in seismic analysis of the (i) Unreinforced masonry (URM), namely,
entire building in which they are constructed. masonry without any or very small amounts
Seismic Analysis of Masonry Buildings: Numerical wikipedia.org/wiki/Masonry_structure). (b) Reinforced

Modeling, Fig. 7 Reinforced masonry construction masonry construction detail using clay bricks with
technology and detail. (a) Modern CMU reinforced improved thermal insulation properties
masonry residential building (Source: http://en.
of reinforcement included and without any and other special shaped proprietary ties,
additional confining members within the placed vertically and horizontally. Other
masonry body. This form of masonry con- reinforcing materials include prestressing
struction is the most frequently encountered strands (prestressed masonry) and polymers,
type of construction in existing or historical used for external strengthening existing
buildings and is associated with the largest masonry. In addition to steel reinforcement
amount of modeling problems, due to the used in modern construction, timber ele-
variety of techniques and materials adopted ments have traditionally been used in histor-
and the variability of their properties. Seis- ical masonry structures and are still used in
mic analysis methods for URM buildings new buildings constructed in seismic regions
need to account for the mechanical charac- in the developing world, with very good
teristics of the material, obtained from in seismic performance characteristics (also
situ evaluation using flat jacks or through called timber-laced buildings, Figs. 8 and 9).
testing (e.g., Clough et al. 1979; Magenes Generally, reinforcement serves to con-
et al. 2008; Vintzileou and Miltiadou- nect the leaves of the masonry wall, to allow
Fezans 2008; Ruiz-Garcia and Negrete for uniform distribution of vertical loads at
2009; and the TCCMaR test series, Kings- floors and roof levels, to act as shear and
ley 1994); furthermore, modeling should flexural reinforcement for in-plane and
account for the brittle nature of its response. out-of-plane seismic load transfer, and also
Only elastic methods of analysis are there- to provide ductility, when reinforcement
fore meaningful in their seismic perfor- yielding precedes brittle failure of the
mance assessment, whereas a low behavior masonry; therefore, its beneficial role
factor (1.50) is applicable for evaluation should be included in seismic analysis; fur-
of the seismic forces. thermore, for new construction, when duc-
(ii) Reinforced masonry (RM) is masonry tility is enforced by design, both elastic and
with horizontal reinforcement in mortar inelastic methods of analysis are meaning-
bed joints and vertical reinforcement posi- ful in seismic performance assessment,
tioned in a cavity or in holes of the vertically with a suitable response reduction coeffi-
perforated masonry units or in grooves cient applied in the expected seismic loads;
between adjacent blocks (Fig. 7a, b). in this context, behavior factors up to 3.0,
In modern construction, reinforcement comparable to RC wall construction, can
comprises steel reinforcing in the form of be used in modern RM building design
bars, trusses or cut wire mesh, dovetails, (EC6 2005).
Seismic Analysis of Masonry Buildings: Numerical Modeling, Fig. 8 Traditional timber-laced stone masonry
buildings in earthquake-prone areas. (a) Antalya, Turkey. (b) Kastoria, Greece
Seismic Analysis of Masonry Buildings: Numerical Seismic Analysis of Masonry Buildings: Numerical
Modeling, Fig. 9 Bhatar construction: traditional Modeling, Fig. 10 Model of the Gaiola pombalina
unreinforced load-bearing timber-laced masonry structure masonry confinement with timber, developed in Lisbon
in Pakistan in modern construction (Source: http://www. after the 1755 earthquake (http://en.wikipedia.org/wiki/
holcimfoundation.org/Projects/advocacy-of-traditional- Gaiola_(construction))
earthquake-resistant-construction-north)
A special form of URM and LBM construction
originating in older (historic) buildings (a well-
(iii) Confined masonry (CM) is masonry which known case being the Pombalinas, constructed S
encompasses during construction horizontal during the rebuilding of Lisbon following the
and/or vertical confining RC or RM ele- devastation of the city on 1755) and still adopted
ments, monolithically bonded into the in less developed countries (e.g., the dhajji in
masonry structure (e.g., RC lintel beams Pakistan) makes use of timber reinforcing ele-
and pilasters). In fact, modern clay brick ments in the form of tension–compression braces;
walls in seismic regions make use of such they are built within the masonry walls at the
horizontal and vertical RC elements. In the floor levels or between floors, in order to confine
case of RC confining elements, similar the masonry infilling the voids (Fig. 10), thereby
modeling conventions for seismic analysis enhancing its in-plane shear resistance and pro-
are adopted as for RM buildings, using how- viding damping under seismic response. This
ever higher seismic forces; in this case, type of LBM construction is also behaving as
a behavior factor of 2.0 is adopted in CM and partially RM, due to the load-carrying
EC8 (2004). mechanism provided by the timber elements and
the in-plane stabilization and stiffening role pro- exhibiting clearly separated orthogonal modes
vided by the masonry infill, confined in turn by of vibration. Linear elastic modeling neglects
the timber structure; seismic modeling of such cracking of the masonry elements and its influ-
systems follows the macromodel approach (see ence in the kinematics and the redistribution of
section “FE Modeling of LBM Structures for forces. Even if cracking is neglected, however,
Seismic Analysis”), whereby the confining tim- linear elastic models provide useful information
ber elements are included in the model together about the structure and the model adopted since:
with the masonry and are verified accordingly. (i) they give an indication of the areas of
increased tensile or compressive stress in the
masonry, which potentially need to be strength-
Seismic Analysis Methods of Masonry ened or rehabilitated; (ii) they can be easily com-
Buildings pared with a visual or in situ measurement of
cracking and stresses in the masonry, thus
Depending on the problem at hand, both linear allowing to verify the reliability of the model
and nonlinear analysis methods, as also and the reasons for the existing condition of the
employed for seismic analysis of RC and steel structure; and iii) possible interacting factors of
buildings, are being applied to LBM buildings overstress can be established (e.g., earthquake
according to the limit state objective of the veri- following a long-term preexisting foundation
fication and the ductility capacity of the masonry settlement).
elements (if any).
Static and Modal Elastic Analyses
Linear Elastic Analyses Linear elastic analysis includes both equivalent
For serviceability limit state (cracking, service static and modal analysis:
load deformation) verifications, linear analysis
methods are adequate, up to the onset of cracking (i) Static analysis can be used in buildings in
of the masonry. The use of linear elastic analysis which higher modes are not dominant in the
for the ultimate limit state (strength) design ver- response, the building is orthogonal in plan,
ification of URM buildings under seismic load and it does not exhibit major irregularities:
combinations provides meaningful verification two different lateral load distributions with
results only under seismic load levels obtained height should be considered, representative
using a response reduction coefficient equal to of different modes of lateral deformation,
1.0; for the design of new RM buildings, it is namely, (a) triangular distribution and
possible to use linear elastic methods with higher (b) uniform distribution. In the case of
behavior factors (up to 3.0), due to the inherent micromodels (see section “FE Modeling of
ductility capacity of the masonry. It should be LBM Structures for Seismic Analysis”), pos-
noted, however, that in case of historic buildings, sible refinements on the linear elastic model
lower seismic actions than those prescribed by can be considered through local modification
modern codes are frequently adopted. Actually, of the finite element (FE) stiffness character-
the application of the requirements of current istics in order to account for cracking of the
codes may lead to invasive interventions that masonry, whether such is predicted from an
are against the internationally accepted charters elastic analysis or it is obtained from field
for the preservation of the built cultural heritage inspection of the condition of the structure.
(e.g., ICOMOS; ISCARSAH). (ii) Modal analyses, when used, should include
Linear elastic analysis methods can also be all the modes with a modal mass that is
adopted for ultimate limit state verifications of greater than 5 % of the total mass of the
URM buildings, provided the structure does not building and enough modes whose modal
have excessive irregularities (EC6 2005) and masses sum to at least 75 % of the total
lacks significant torsional effects, thereby building mass.
For the evaluation of the modal response of the building, taking into account the actual
characteristics – deformations and internal inelastic characteristics of the elements,
forces – using macromodels (see section “FE brittle or ductile; evaluation of this capacity
Modeling of LBM Structures for Seismic Analy- curve yields the expected target deformation
sis”), the square root of the sum of squares com- demands, under different seismic excitation
bination rule of modal quantities can be used if levels (performance levels), at which point the
the modes differ with each other by as much as onset, distribution, and extent of structural dam-
90 %, or, better, using the complete quadratic ages are obtained and compared to code damage
combination rule, giving accurate maxima for levels (FEMA-356 2000), either in terms of
closely coupled mode combinations. For seismic element resistance (brittle element response) or
modal analyses using micromodels (e.g., shell in terms of inelastic deformations (ductile
finite elements), the peak response characteristics element response). For the evaluation of these
(deformations and internal stresses) Emax should generalized force–deformation characteristics,
be evaluated following: different failure mechanisms – accounting also
for out-of-plane effects – can be considered, and
( )
X
n the weakest governing mechanism should be
Emax ¼ maxt Ej ðtÞ (1) adopted as governing the failure response.
j¼1
where Ej(t) is the time history of the Modeling of LBM Buildings for Seismic
corresponding parameter due to response in the Analysis
jth eigenmode.
The primary load-bearing elements of LBM
Inelastic Analyses construction are the vertical load-supporting
In currently accepted methodologies for the elements, namely, the floor and roof structure,
assessment of existing LBM buildings (both as well as the perimeter and interior bearing
RM or URM) in the context of PBD, nonlinear masonry walls. All these elements carry the
static analysis methods and corresponding vertical loads (including self-weight) and the
modeling conventions are employed with differ- lateral forces to the foundation. Secondary
ent levels of modeling detail, depending on elements, not part of the lateral resisting sys-
whether cracking, post-ultimate, and cyclic hys- tem (such as light partitions), are not included
teretic characteristics are included in the model in the model as earthquake resisting elements.
(Fig. 11). The purpose of these methodologies is Only their self-weight alone is taken into
to evaluate the inelastic lateral load deformation account in analysis.
S
a b
Generalized Generalized
force force
Ry Ru
Rres Rcr
Rres
du
dy Generalized dcr dRu du Generalized

deformation deformation
Seismic Analysis of Masonry Buildings: Numerical Modeling, Fig. 11 Inelastic analysis constitutive models of
masonry building elements for in-plane only action (bending or shear). (a) Simplified. (b) Refined
Floor and Roof Elements and Diaphragmatic the supporting walls, failure or collapse of flexi-
Action ble diaphragms takes the form of failure (usually
Floor structures and the roof structure transfer the pullout) of the diaphragm-to-wall connection,
vertical loads and in-plane inertia seismic loads due to inertia force transfer or excessive out-of-
to the rest of the bearing elements. While in plane deformation of the wall and loss of support
modern masonry buildings floor elements are of the timber elements to the wall. This type of
typically stiff RC diaphragms, in typical existing behavior depends mostly on the vertical support
or historical masonry construction, light wooden system of the diaphragm and the detailing of the
floors on wood or steel purlins have been used connections, which all need to be part of the
traditionally. Brick or stone masonry vaults are building model (see, e.g., Vintzileou et al. 2007,
also typical in the lower story of several historic for a description of the system used in the tradi-
structural systems; other variants include arched tional masonry structures in Lefkada, Greece).
brick constructions between joists or two-way
Zoellner diaphragms made of brick infilled Seismic Load-Bearing (Primary) Vertical
voids. Purlins usually span one way and are sim- Elements
ply resting or encased in the bearing walls at each The primary lateral load-bearing elements of
end. A peculiarity of masonry buildings is that LBM buildings are the perimeter and interior
masses are not concentrated at floor levels; they bearing walls, which are typically perforated
are distributed over the height of the building. with openings, forming wall elements, spandrel
Unlike modern construction concrete slabs beams (also arched lintel beams), and pilasters
which provide diaphragmatic action that distrib- (Fig. 12b). The in-plane lateral load transfer of
utes the inertia loads in plan and tie all the vertical the wall elements depends on their aspect ratio
masonry elements at the floor level, existing (the height to width ratio): walls with relatively
masonry structures were constructed with flexi- longer width compared to the element height
ble diaphragms which deform in plane and oper- (height to width ratio less than 2–3) tend after
ate differently during the earthquake (Fig. 12a). initial cracking to transfer the lateral force to the
Therefore, the presence or absence of diaphrag- lower level directly through an inclined strut
matic action and the way the diaphragm is tied to (including also the vertical load), whose horizon-
the vertical elements are two important aspects to tal component equals the lateral load; more slen-
consider in seismic modeling, since the dia- der walls or multiple wall elements (wallets)
phragm stiffness will affect both the dynamic created in a wall with openings may opt for
characteristics of the masonry building and the a more flexure-dominated behavior (similar con-
transfer of forces among the stiff vertical wall siderations apply also to the horizontal spandrel
elements. beams between openings), while failure of these
Along the same context, the function of the elements is brittle.
diaphragm (one-way flexible or two-way stiff Since the walls are the primary lateral load
action) shall also define the distribution of the resisting elements, their distribution in plan and
floor plan masses to the walls: one-way joist their stiffness (namely, geometric size and per-
diaphragms will only distribute inertia reaction centage of opening area) determine the eccentric-
loads across the walls at which their wooden ity between the center of application of the floor
joists are inframed, making the usual uniform inertia forces (the center of mass for stiff dia-
mass distribution assumption in the building phragms) and the center of rigidity of the building
model, namely, a lumped rotational/translational in plan. Consequently, irregularities that may
mass at the center of mass, incorrect. arise in plan and also in elevation due to the
In addition to load distribution, failure of the wall distribution and geometry will influence
diaphragm is also important to model: apart from the distribution of the lateral loads among the
failure of arch-supported diaphragms that tend to different resisting elements, something which is
be sensitive to relative in-plane deformations of crucial to model in seismic analysis. It should
a Flexible diaphragm action transfers inertia

loads to the two façades only.
diaphragm
beam vertical loads
Diaphragm action restrains building

deformations at the floor levels.
wall
wall
element
Seismic Analysis of Masonry Buildings: Numerical through diaphragmatic action (where this exists). (b) Con-
Modeling, Fig. 12 Good practices in the seismic analy- tributions of the three-dimensional analysis to the in-plane
sis of masonry buildings. (a) Contributions of the seismic and out-of-plane actions on the facades and to the corner
forces to the masonry façade in a typical URM building element forces
further be overemphasized that these structural what has been assumed in an elastic analysis,
characteristics are dependent on the intensity of redistributing forces as well as increasing the
the earthquake, since masonry (like concrete) lateral deformation demands during the an earth-
cracks in tension or softens under extreme com- quake. The analysis sophistication therefore and
pression stresses, and therefore the relative stiff- the modeling capabilities should reflect the level
ness of the bearing elements changes beyond of response prediction.
In-plane response failure of the walls or other Secondary Elements

structural elements (where present) will take Such elements are typically the interior masonry
place due to inadequate resistance or excessive (or other) column elements that support the dia-
interstory drift. Masonry element failure includes phragm or narrow, slender elements on the façade
wall pier, pilaster, or spandrel beam failure under that function as columns. The failure of these
in-plane actions. Depending on the aspect ratio of elements does not alter significantly the response
these elements and the existence of diaphrag- of the building, and therefore, they are ignored in
matic action, failure of the wall elements affects the seismic analysis model. However, their abil-
individual masonry elements and will jeopardize ity to bear vertical loads under the seismic defor-
the entire vertical load path to the lower floors, mation of the building should still be verified.
the building dynamic characteristics, as well as
the redistribution of forces in and out of plane. Foundation
Element failure takes the following forms: The foundation of the masonry walls is not neces-
sarily responding as rigid and non-deforming, as
(i) Flexure-dominated failure including crack- typically assumed in seismic modeling. Founda-
ing and/or rocking of the wall, compressive tions may settle in the long term under the weight
toe failure of the wall of the walls, inducing tensile cracks in the building
(ii) Shear failure in plan, including sliding at that change the distribution of forces and the way
mortar bed joints, diagonal cracking or diag- the vertical elements respond (rocking rather than
onal crushing of the wall between the cracks flexing). Furthermore, due to the lack of reinforce-
(iii) Entire wall movement ment, contact type of connection develops
(iv) Apart from these in-plane response failure between the URM footing, the masonry wall, and
modes, walls may also fail under concurrent the soil, with possible uplift and rocking under
in-plane and out-of-plane action. In this case tensile or rocking response. These mechanisms
failure is closely associated with the exis- should be captured in the seismic model, if the
tence or not of adequate diaphragmatic vertical and lateral loads are such as to allow for
action and a suitable tensile diaphragm con- this kind of motion. Consequently, if preliminary
nection to the façade walls that will mobilize analyses indicate this to be the case, the soil resis-
the entire building façade from the footing tance to the footing stresses should be modeled
line (below grade) to the roof (Fig. 12a). using an elastic but tensionless type of behavior
Special cases of out-of-plane bucking failure (e.g., a Winkler model with uplift), in order to
under in-plane compressive action are also obtain the proper footing flexibility as is the case
included in this combined failure mode, pri- in situ. An in situ geotechnical study and adequate
marily for slender wall elements though, e.g., knowledge of the foundation shall provide the soil
unsupported multiple-leaf walls. constitutive characteristics and the foundation
(v) Failure under combined biaxial effects conditions (e.g., the possible existence of a well
(combined axial load) characterizes the cor- or a septic tank near the footing).
ner walls and pilasters, due to the three-
dimensional nature of the building response. Principle Modeling Requirements for Seismic
Bidirectional rocking of the building Analysis of LBM Buildings
(Fig. 12b) induces axial loads that are con- In summary, the building seismic model used
siderably higher than uniaxial predictions, should identify the basic force transfer mecha-
together with biaxial bending and shear nisms of the structural system, irrespective of the
forces. In this case, a three-dimensional method of analysis used:
model of the building is needed to predict
such overstress in the corners, with the (a) Plane or three-dimensional model. Plane
results of plane analyses only being frames are often used as an approximation
unconservative. for modeling the building. It is important to
note that the proper idealization of seismic idealization is not justified in masonry model-
behavior is through modeling of the three- ing, and, instead, the model should incorpo-
dimensional response. Two-dimensional rate distributed masses; due to the large
models are unable to monitor the three- masonry element size (compared to the nor-
dimensional response of the entire structure; mal operational loads of the building) and
consequently their use will not predict the often the lack of a heavy concrete diaphragm,
spatial response effects, namely, (i) the loads are not distributed according to the
simultaneous action of seismic axial forces lateral stiffness of the elements. For this rea-
and biaxial bending effects in the corner son, three-dimensional modal dynamic anal-
piers, (ii) the corner element vertical defor- ysis should at best be adopted, reflecting
mation compatibility under concurrent more accurately the system deformation and
actions in the two orthogonal directions, (iii) the load path of the inertia forces from the
the influence of out-of-plane bending in the diaphragm to the foundation. Similarly,
resistance of in-plane actions of the vertical out-of-plane effects need also to be taken
elements, and (iv) the possible in plan tor- into account in a “distributed with height”
sional effects which will enter into the sense for assessing the in- and out-of-plane
response in the case of a relatively rigid dia- interaction of forces to the walls, the dynamic
phragm and eccentric distributions of the connection forces at the wall intersection, and
mass (e.g., an opening) or the stiffness (e.g., the seismic deformations and forces at the
asymmetric facade opening distributions roof-to-wall connections.
and/or interior masonry walls). (d) Modeling of the building foundation.
(b.1) Modeling of the diaphragm. The presence Masonry building response and past defor-
or lack of a diaphragm in the model should mation history are affected by the deforming
be in accordance with the function of the foundation at the base of the building. Con-
diaphragm in the structural system at hand. sequently, full foundation fixity is often an
The in-plane rigidity of the diaphragm is an unrealistic assumption, given that the stresses
important consideration to account for in under the masonry wall may be relatively
the model, particularly because of the fact high. Evidence of cracking around openings
that masonry buildings tend to be very stiff or unsymmetrical distribution of cracks in an
and the relative diaphragm-to-wall element existing masonry building is often the effect
in-plane stiffness will define the distribu- of differential settlements due to variable
tion of inertia loads from the floor to the ground conditions in plan, such as the pres-
vertical seismic load resisting elements. ence of an abandoned well or a septic tank at
(b.2) Modeling of the diaphragm connection one end, improper ground preparation at the
with the walls. Furthermore, as far as the time of construction, or a partial plan base- S
modeling of the diaphragm is concerned, it ment. For modeling the foundation, an
is important to consider in the model acceptable practical modeling approach
whether its connections with the vertical makes use of tensionless elastic springs, pro-
masonry elements justify the use of viding vertical, lateral, and bending restraint
a deformation fully compatible FE nodal in the embedded footings.
connection, transferring load. The same is
true for the roof structure, which is nor-
mally simply supported on the masonry FE Modeling of LBM Structures for
walls, and sliding of the roof trusses on Seismic Analysis
the walls is often possible under differential
lateral seismic movements across walls. From the analysis of the complexity of the
(c) Modeling of the mass. The conventional response, the failure modes, and the fact that
frame analysis assumption of a lumped mass different technologies and materials comprise
masonry construction, practical modeling of primary issue in this case as well as in certain
masonry buildings for seismic analysis relies on aspects, the actual ability to model the material
the use of FE models and follows two different behavior under cyclic loading conditions.
techniques (Fig. 13):
Macromodels for Entire Building Analysis
(i) The less refined “global” (phenomenological) Since the use of micromodels is too expensive
macromodels, suitable for the analysis of and complicated for entire building analysis
entire LBM buildings. In this case, the walls, under seismic excitations and due to the fact
piers, and spandrel beams of the masonry that available software capabilities are often lim-
structure are modeled using one-dimensional ited in handling such FE micromodels, equivalent
line elements with nonlinear characteristics. FE macromodels are being used for practical
These macromodel elements (Fig. 13d) seismic analysis of LB or NLB masonry
are characterized by equivalent axial load, construction.
bending, and shear interaction response char- For load-bearing masonry, the usual modeling
acteristics. Such models have been used both conventions adopted in conventional frame
for elastic and inelastic three-dimensional building analysis are also adopted for masonry
seismic analysis for design, assessment of buildings as well: roof and diaphragm elements
seismic vulnerability, and fragility studies of are modeled using line FE, or plain diaphrag-
masonry buildings under monotonic or cyclic matic action is enforced (if it exists). Similarly,
loading. wall piers, columns (pilasters), and spandrels are
(ii) The refined “local” FE micromodels where modeled at their centerline using equivalent or
the entire masonry building or a plane por- actual property line FE with linear or nonlinear
tion of the building, such as a masonry characteristics, while the element joint regions
façade under planar response, is modeled are modeled using infinitely stiff elements
using two- or three-dimensional FE approx- and/or rigid zone transformation models
imations, with associated material and load- depending on the analysis software conventions.
ing description (Fig. 13c). In terms of For inelastic analysis, phenomenological
geometric representation, thick shell or axial, shear, and flexural constitutive relations
across the length and through thickness should be specified, possibly with cracking and
brick FE approximations are adopted, possi- post-failure modeling capabilities and, if possi-
bly coupled with beam FE for the reinforce- ble, interaction of axial/flexural and shear
ment, if any. Even more refined micromodel stresses. Where this is not possible and given
approximations have been adopted in that the variation of axial loads is not high for
research studies, in which the actual masonry low-rise buildings, uncoupled values may be
unit and mortar have been separately assumed based on initial state vertical load levels.
modeled using brick and plate elements, Masonry building macromodels for inelastic
respectively. analysis evolved from: (i) the simplified weak
spandrel strong pier model, whereby spandrels
In terms of material approximation, phenom- crack early and are neglected (therefore the
ena such as cracking and compression piers are considered to act as uncoupled cantile-
nonlinearity or the presence of steel are smeared vers joint by hinged rigid link beams at the floor
within the element integration area through levels) (Fig. 14b); (ii) the strong spandrel weak
equivalent stiffness and resistance modifications pier model (the shear frame analogy), whereby
at the FE integration point. Models of this type the piers crack first and are therefore assumed to
have been used in the parametric investigation of deform with their inelastic characteristics, the
conventional or historical masonry construction spandrels remaining relatively rigid (Fig. 14c);
and for the validation of test results. Complexity and (iii) the equivalent frame model, where the
and computational cost and resources are the masonry structure is modeled as an assembly of
Seismic Analysis of Masonry Buildings: Numerical a stone masonry ground floor. (a) Photo). (b) Bearing
Modeling, Fig. 13 Modeling example of a historical structure. (c) Micromodel. (d) Macromodel (Vintzileou
masonry building in Lefkada, Greece, built of timber et al. 2007). Note that several load-bearing elements in
confined plinth masonry walls at the upper floor on the model are omitted for clarity
vertical pier and horizontal spandrel line ele- by several investigators for modeling both the
ments interconnected by rigid joint regions in-plane shear and the flexural response of S
(Fig. 14d): the geometry of the joints is obtained masonry wall piers, incorporating the different
from the geometry of openings and an equivalent failure mechanisms of these elements (Fig. 15).
pier height, which is defined by the extent of Magenes and Calvi (1997) proposed that the
cracking observed in the vertical elements fol- peak shear resistance Ru (Fig. 15) of a rocking
lowing an earthquake or, if uncracked, assuming masonry wall pier bearing an axial load P is
a crack inclination at about 30 that extends from given by:
the opening toward the joint (Lagomarsino

et al. 2013). D2 t p p
Ru ¼ 1 (2)
For the constitutive modeling of inelastic FE H0 2 kf u
macromodels under static seismic-type load, the
multilinear (simplified or more refined) shear where D and t are the length and thickness of the
force–interstory drift or bending moment–pier pier; H0 is the effective height equal to the shear
rotation diagrams of Fig. 11 have been proposed span, namely, the height to zero moment, taken as
Seismic Analysis of a b
Masonry Buildings:
Numerical Modeling,
Fig. 14 Equivalent frame
model for load-bearing
masonry construction. (a)
The masonry facade to be
modeled. (b) Weak
spandrel and strong
cantilever pier model. (c)
The weak pier and rigid
spandrel shear frame
model. (d) The equivalent
frame model using c d
spandrels, piers, and rigid
joint regions (Lagomarsino
et al. 2013)
Seismic Analysis of P
V Shear
Masonry Buildings:
force V R
Numerical Modeling, u
Fig. 15 Model of the wall 2H0 Fcr
pier and lateral load
deformation constitutive e Fres
relation under axial force 0.85fu
V
D P dcr du Shear
P dRu
M = Pe = VH 0 deformation
being equal to the pier height, for cantilever piers, where, in addition to the parameters defined
and half the pier height, for piers in contraflexure; above, tu is the average masonry shear strength;
p is the average vertical load pressure, equal m is the Coulomb friction coefficient; and c the
to the axial load P divided by the wall area cohesion of the wall (namely, the average fric-
( p = P/Dt); fu is the compressive strength of the tional resistance at P = 0). Both of the latter
masonry; and k is a coefficient that depends parameters are global material constants for the
on the stress distribution at the toe of the wall wall and are obtained from testing of masonry
(k = 0.85 for an equivalent rectangular stress elements. Following micromodel analysis at fail-
block; see Fig. 15). ure and comparison with test results, their model
For shear failure mechanisms of brick masonry accounted for two types of shear wall failure,
walls, when mortar bed and head stone failure is namely:
involved, they proposed to adopt a Mohr–Coulomb
friction model for the wall, namely: (i) Failure at the cracked wall ends, with the

peak shear strength tu being resisted in the
P compression area of the cracked section D0
Ru ¼ D t tu ¼ Dtðc þ mpÞ ¼ Dt c þ m (3)
Dt by t, where D0 (the compressed portion of
rffiffiffiffiffiffiffiffiffiffiffiffiffiffi
the wall length D) is obtained by equilib- f p
rium of normal force and moment (Eq. 4a): Ru ¼ D t tb ¼ Dt
bt 1þ
H0 f bt
2:3 1 þ
D

0 3V H 0 (6)
D ¼ bD ¼ 1:5 D (4a)
P D
and the minimum of the values (Eqs. 5

(ii) Failure at the mid-height of the wall, with
and 6) used for resistance. In terms of
the maximum shear strength tu being
deformation, the wall pier deformation
resisted by the entire wall thickness (area
corresponding to the ultimate strength (dRu,
D by t) decreasing in an inverse linear man-
Fig. 15) was found to be close to 0.5 % of the
ner with the wall shear span ratio H0/D:
wall height H in most test results they
0 1 performed or evaluated (Magenes and
Calvi 1997).
tu B c þ mp C
Ru ¼ Dt ¼ Dt@
H0 A
(4b)
H0
1þ 1þ Detailed Micromodels for Seismic Analysis
D D
In addition to the macromodels above,
micromodels have been proposed and are
The shear strength of the wall in diagonal
employed for equivalent static linear or nonlinear
tensile failure is therefore given by the min-
seismic analysis of masonry buildings and
imum strength of the two resisting
(primarily) historical structures; furthermore,
mechanisms – failure modes:
micromodels have also been used to calibrate
80 macromodel topology and the masonry wall fail-
10 19
>
> >
<B 1:5c þ m C c þ mp >= ure and constitutive response. Micromodels ide-
B C
Ru ¼ Dttu , tu ¼ min B@
C, @ A alize masonry in detail using:
>
> 3cH 0 A H0 >
>
: 1þ 1þ ;
pD D (i) Two-dimensional thick plate and shell ele-
(5) ments that account for both in-plane and
out-of-plane stiffness and resistance character-
It is further noted in the above expres- istics (for out-of-plane bending effects of the
sions that the coefficients of friction and walls). The use of shell element elastic models
cohesion m and c may be modified so as to is quite common in the seismic analysis of
obtain effective values m and c, respectively, masonry structures, since they do not require
corrected for the geometry of the masonry excessive computational resources and pro-
unit, in accordance with the fact that the vide the basic load path and demand concen- S
expressions above sometimes trations within the building, accounting for
overestimated the experimental value of complex geometries and multitude of mate-
strength, due to the influence of the masonry rials, such as timber, masonry, etc. (Fig. 13c).
headjoints. These are therefore suitable to use in entire
(iii) In addition to shear failure by Coulomb building seismic analysis models.
friction at the joints, shear failure due to (ii) Three-dimensional (brick) FE models of the
in-plane cracking of the bricks was also entire masonry structure, taking into account
experimentally observed for weak brick in the model both complex geometric ideal-
and strong mortar, in the presence of high izations of the structure and the foundation
axial stresses. It was proposed that the shear and the material complexities associated
strength be estimated in this case in terms of with the presence and interaction of several
the tensile strength of the bricks fbt, follow- different materials with complex constitu-
ing Eq. 6: tive characteristics, such as: stress-strain
nonlinearity due to cracking or crushing; tri- space), using suitable contact friction
axial capacity interaction and volumetric elements.
dilatancy under loading for the mortar, con- • Models based on the discrete element ideali-
crete, stone, or brick; yielding or pullout zation that fall into this category have been
phenomena of the reinforcement where it promoted for modeling primarily historical
exists; interface failure between brick, mor- monuments; in this case, the brick units
tar, or steel (where it exists); and inelasticity (or stone building blocks) are modeled as indi-
of timber. vidual deformable or undeformable volume
elements, and their interface is described
Depending also on the capabilities of the soft- with Coulomb frictional contact-separation
ware adopted for the seismic evaluation, characteristics.
micromodels adopted for modeling of inelastic
seismic response of LBM buildings include:
Modeling of NLBM Infilled Frame
• The smeared representation models. These Buildings for Seismic Analysis
model nonlinearity as spatially averaged, by
considering the distributed cracking of Due to the abundant use of CM infilled panels in
masonry in the vicinity of the FE integration steel and RC frame structures, the seismic analysis
points and/or the average constitutive modeling of entire frame buildings with NLBM
response of masonry over the entire FE inte- infills is also briefly examined herein for complete-
gration volume (area) using suitable two- or ness. Only the modeling of the infills is considered
three-dimensional inelastic constitutive herein, since the modeling of the entire frame is
behavior and a strength interaction surface. beyond the scope of this text. It should be noted,
Plane stress or three-dimensional yield surface however, that in the design of RC frame buildings,
characteristics have been proposed for CMU, the infills are (and have been) neglected in the
brick masonry, or stone, possible candidates structural model assuming that these contribute
being, among others, the models proposed by only to the inertia mass. Only recently modern
Gambarotta and Lagomarsino (1997) and seismic codes (EC8 2004) provide structural
Stavridis and Shing (2010). forming and detailing guidelines for taking into
• The discrete representation models. These account possible adverse response effects due to
model nonlinearity discretely through detailed the presence of the infill; furthermore, PBD meth-
FE modeling of all different material regions odologies for the assessment and retrofit of existing
involved, namely, individual modeling of the RC frames require that these be fully accounted for
mortar as a brick, shell, or zero-length contact in the seismic model, in their as-built configura-
FE, the masonry units and the concrete (where tions and properties (FEMA-356 2000).
present) as a brick or plain stress shell element Infill panels provide a large increase in the
(primarily for industrially manufactured units lateral stiffness of the confining frame with
of constant geometry), and the steel or timber a disproportionate increase in its mass; conse-
reinforcement (where present) as truss or quently, masonry infilled frames exhibit short
beam elements (for CM or RM buildings). fundamental periods compared to the bare frame
• As a special case of these are the discrete crack structure and therefore attract higher inertia
representation models, which further monitor forces at shorter drifts. Properly engineered infill
crack formation using fracture energy criteria panels, constructed of good quality modules and
and the evolution of cracking within the mortar and adequately wedged into the panel,
masonry element through mesh redefinition without excessive openings and a regular in plan
(average material representation) and/or and in height configuration, will provide the bare
through or along predefined mortar beds frame structure with considerable overstrength
(where these exist in physical and model and stiffness enhancement.
Seismic Analysis of
Masonry Buildings:
Numerical Modeling,
Fig. 16 Damages to
infilled RC frames
following the 1985 Mexico
City earthquake. (a)
Damage of the confining
frame elements. (b) Out-of-
plane failure of the infills.
Note that the panels were
confined by RC lintels and
pilasters tied to the RC
frame
Masonry infills, like other masonry elements compression only in the direction of the lateral
already discussed, tend to fail in plane or out of force. The strut has a thickness tinf equal to the
plane (Fig. 16) following similar mechanisms panel thickness and an equivalent width Weff
as other masonry elements (diagonal thrust or based on the frame to panel contact length; pro-
tension, corner compression, sliding at joints posed strut widths adopt contact lengths that
through brick or mortar, out-of-plane collapse, account for the relative stiffness characteristics
etc.). Such a full or partial failure of the of the confining frame and the infill panel in terms
infills will lead to local failures of the confin- of lh,L (Stafford Smith and Carter 1969), where
ing elements due to the formation of lh,L expresses the relative stiffness of the infill
unintentional short column effects or due to and the concrete elements assuming elastic con-
a shear failure of the top of the column or the tact (Fig. 17a):
beam–column joint. For these reasons, in the where Em and Ec are the Young’s moduli of
PBD assessment approach of existing struc- masonry and concrete; Icol is the uncracked
tures, the infills need to be accounted for in moment of inertia of the confining column ele-
seismic analysis. ments; y is the geometric inclination of the infill S
In modeling infilled frame buildings in practi- strut (Fig. 17a); Linf and Hinf are the clear length
cal seismic analysis of entire structures, and height of the infill; and Hc is the centerline
macromodels are used for the panels, as shown distance between the beams. For seismic applica-
in Fig. 17. Micromodels have also been adopted tions cross strut configurations are used with
in the literature, however primarily for research compression only properties (Fig. 17b). In order
applications and in order to calibrate the to model the effect of short column formation in
macromodel parameters. The use of the infilled frame response, multiple strut config-
macromodels stems from the observation that, urations as shown in Fig. 17b have also been
due to the partial contact separation of the panel proposed (Crisafulli et al. 2000). Their use is
from the surrounding frame elements under lat- recommended in the case of asymmetric infill
eral response of the infilled frame, the function of configurations within a floor, such as infilled
the infill can be modeled as an equivalent axial frame bays next to open bays or end bays of the
load-bearing diagonal strut element resisting infilled frame.
a aL
sw 4 E m ⋅ tinf ⋅ sin (2θ )

λh =
ah
4 ⋅ Ec ⋅ Icol ⋅Hinf
4 Em ⋅ tinf ⋅ sin (2θ )

λL =
4 ⋅ Ec ⋅ Ib ⋅ Linf
Hinf
Hc
Weff
π π
αh = ,α L =
2λh λL
Weff = α h2 + α L2
q
Linf
Lb
Seismic Analysis of Masonry Buildings: Numerical Model proposed by Stafford Smith and Carter (1969). (b)
Modeling, Fig. 17 Equivalent strut macromodel of Single and multiple strut macromodels of infilled RC
infilled RC frames and effective strut width equation. (a) frames for static lateral load and cyclic dynamic analysis
Summary Several modeling conventions have been

proposed so far and are still under rigorous
The seismic response and forms of failure for research investigations, for the seismic analy-
different types of masonry construction have sis of masonry buildings. Overall, these range,
been presented and discussed. It is commonly depending on the capabilities of the FE code
believed that masonry buildings, being the most at hand and also on the reliability of the
common form of past construction in all seismic structural information available, from simple
affected areas, as well as a financially viable strut-and-tie models (the equivalent frame
alternative for new low- to mid-rise construction, analysis models, which are also referred to
they need to be evaluated for seismic loading, due as macromodels) to the refined micromodels,
to the seismic vulnerability of such structures as it using plane or three-dimensional FE analysis
has been proven in numerous seismic events, past tools, both linear and nonlinear material-wise
and recent. and geometry-wise.
Not all methods are suitable for all cases, and ▶ Seismic Vulnerability Assessment: Masonry
often analysis “overkill” for a problem that is Structures
quite complex to model will give a false sense of ▶ Seismic Vulnerability Assessment: Reinforced
security: due to the complexity of the problem, Concrete Structures
involving different masonry materials (brick, mor- ▶ Strengthening Techniques: Masonry and
tar, stone, rubble, and also RC or steel and timber), Heritage structures
different types of construction (LB or NLB, CM),
and different structural topologies (low or mid
rise, with or without diaphragms, irregular in
References
plan or elevation, with openings, with flexible
foundation), equally complex methods should be ASCE (2013) Building code requirements and specifica-
used, justifying the accuracy of the input information for masonry structures. Standards ASCE/SEI 5-
tion, namely, the material properties for all mate- 13, 6-13. American Society of Civil Engineers, Reston
Clough R, Mayes R, G€ ulkan P (1979) Shaking table study
rials involved in the construction and an adequate
of single-story masonry houses, vol 3, Summary, con-
knowledge of their interaction. clusions and recommendations, UCB/EERC-79/25.
For entire structural models, macromodels or Earthquake Engineering Research Center, University
area micromodels with averaged properties over of California, Berkeley
Crisafulli F, Carr A, Park R (2000) Analytical modeling of
the FE region are adequate to capture global quan-
infilled frame structures; a general review. Bull N Z
tities (forces and deformations) and seismic per- Soc Earthq Eng 33(1):30–47
formance. For detailed damage prediction DesRoches R, Comerio M (2011) The 2010 Haiti Earth-
(cracking, crushing, region disintegration under quake. Earthquake Spectra, vol 27, S1. Earthquake
Engineering Research Institute, Oakland
overload), three-dimensional FE models are
EC6 (2005) Eurocode No. 6, Design of masonry
adopted, using either deformable elements with structures – Part 1-1: General rules for reinforced and
smeared or discrete cracking representation or dis- unreinforced masonry structures, EN-1996-1. Euro-
crete FE models with contact friction interfaces, pean Committee for Standardization, Brussels
EC8 (2004) Eurocode No. 8, Design of structures for
where crack spreading is not feasible (e.g., URM
earthquake resistance – Part 1: General rules, seismic
construction and historic building analyses). These actions and rules for buildings, EN-1998-1. European
models, although they provide realistic damage Committee for Standardization, Brussels
predictions compared to observation and test, are EERI/IAEE. World Housing Encyclopedia. http://www.
world-housing.net/
too detailed and complex to be applied to ordinary
FEMA-356 (2000) Prestandard and commentary for
construction, and they need to be substantiated by the seismic rehabilitation of buildings. FEMA,
adequate testing in order to establish the material Washington, DC
behavior as input to the FE model. Gambarotta L, Lagomarsino S (1997) Damage models for
the seismic response of brick masonry shear walls. Part
II: The continuum model and its applications. Earthq
Eng Struct Dyn 26:441–462 S
Cross-References Hughes R, Nichols R, Paul T, Riding V (1990) The
Augusta, Sicily earthquake of 13 December 1990,
a field report. EEFIT, Institution of Structural Engi-
▶ Ancient Monuments Under Seismic Actions:
neers, London
Modeling and Analysis IAEE (2008) Regulations for seismic design: a world list.
▶ Assessment of Existing Structures Using http://www.iaee.or.jp/worldlist.html
Inelastic Static Analysis ICOMOS. International Council on Monuments and Sites.
http://www.icomos.org/en/
▶ Equivalent Static Analysis of Structures
ISCARSAH. The International Scientific Committee on
Subjected to Seismic Actions the Analysis and Restoration of Structures of Archi-
▶ Masonry Modeling tectural Heritage. http://iscarsah.icomos.org/
▶ Masonry Structures: Overview Kingsley G (1994) The U.S.-TCCMAR full-scale five-
story masonry research building test, Part 2, Report
▶ Nonlinear Finite Element Analysis
SSRP-94/02. University of California, San Diego
▶ Numerical Modeling of Masonry Infilled Lagomarsino S, Penna A, Galasco A, Cattari S (2013)
Reinforced Concrete Frame Buildings TREMURI program: an equivalent frame model for
2596 Seismic Analysis of Steel and Composite Bridges: Numerical Modeling
the nonlinear seismic analysis of masonry buildings. Introduction

Eng Struct 56:1787–1799
Magenes G, Calvi GM (1997) In-plane seismic response
of brick masonry walls. Earthq Eng Struct Dyn Reinforced concrete slabs rigidly connected with
26(11):1091–1112 steel girders have been used to form the basic
Magenes G, Morandi P, Penna A (2008) Test results on the superstructure of large numbers of deck bridges
behaviour of masonry under static cyclic in plane for many decades. This is due to the fact that the
lateral loads. ESECMaSE project, report RS-01/08.
Department of Structural Mechanics, University of composite construction method offers the bridge
Pavia, Pavia engineers a great variety of solutions for different
Moro L (coord. editor) (2007) Guidelines for evaluation types of problems. A typical composite cross
and mitigation of seismic risk to cultural heritage. section of a highway bridge is shown in Fig. 1.
Gangemi Editore, Rome
Rossetto T, Peiris N, Alarcon J, So E, Sargeant S, A series of parallel steel girders are rigidly
Libberton C, Verrucci E, Del Re D, Free M (2009) connected with a reinforced concrete slab
The L’Aquila (Italy) Earthquake of 6th April 2009, through shear connectors. The shear connectors
a field report. EEFIT, Institution of Structural Engi- installed are mostly welded studs allowing use of
neers, London
Ruiz-Garcia J, Negrete M (2009) Drift-based fragility the deck as part of the top flange (deck plate
assessment of confined masonry walls in seismic girders). The longitudinal bending of the com-
zones. Eng Struct 31:170–181 posite T-girders, at sagging bending areas, results
Stafford Smith B, Carter C (1969) A method of analysis in tension in steel and compression in concrete.
for infill frames. Proc Inst Civil Eng 44:31–48
Stavridis A, Shing PB (2010) Finite element modelling of The simultaneous operation of both of these
nonlinear behaviour of masonry-infilled RC frames. materials generates the composite action which
J Struct Eng 136(3):285–296 is the most important feature for the formation of
UNIDO (1983) Repair and strengthening of reinforced con- stiff and high-strength cross sections. At hogging
crete, stone and brick masonry buildings. Building con-
struction under seismic conditions in the Balkan region, moment areas, concrete is considered to be fully
vol 5. UNDP/UNIDO Project RER/79/015. United cracked, and only the slab reinforcement, but not
Nations Industrial Development Organization, Vienna concrete, contributes to bending resistance.
Vintzileou E, Miltiadou-Fezans A (2008) Mechanical In pure steel bridges the reinforced concrete
properties of three-leaf stone masonry grouted with
ternary or hydraulic lime based grouts. Eng Struct slab is replaced by an orthotropic steel deck. This
30(8):2265–2276 is composed of the steel deck plate; the longitu-
Vintzileou E, Zagkotsis A, Repapis C, Zeris C (2007) dinal, mostly trapezoidal, stiffeners; and the
Seismic behaviour of the historical structural system crossbeams (Fig. 2). Orthotropic are lighter than
of the Island of Lefkada, Greece. Construct Build
Mater 21(1):225–236 concrete decks. However, they require high fab-
rication costs due to intensive welding operations
and are susceptible to fatigue. Therefore, pure
steel bridges (see EN 1993-2 2006) are nowadays
Seismic Analysis of Steel and mostly limited to cases where it is essential to
Composite Bridges: Numerical limit the deck weight, e.g., for very large spans,
Modeling movable bridges, etc. The present article
refers mainly to composite bridges that constitute
Ioannis Vayas1 and Aristidis Iliopoulos2 the vast majority in modern steel bridge
1
Traffic Engineering Laboratory, National construction.
Technical University of Athens (N.T.U.A.), Modeling for analysis is required in order to
Athens, Greece determine internal forces and moments, deforma-
2
Peikko Greece SA, Marousi, Athens, Greece tions, and vibrations of bridge decks including
bearings, piers, abutments, piles, etc. In addition
Synonyms models should include the foundation when soil-
structure interaction is accounted for. A bridge
Bearings; Bridge modeling; Composite bridge; analysis model should be based on the following
Seismic design; Seismic isolation criteria:
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling 2597
Precast planks Bracing at piers

Bearing
Intermediate Cast in place stiffeners
bracings concrete slab
Bearing
Steel girder
At mid-span At pier
SECTION
Expansion
After deck concreting Before deck concreting joint Superstructure Abutments
a hc
a
Piers Bearings
hc
hc Piles
Section a-a
Earthquake
PLAN VIEW
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 1 Layout of a typical composite
deck bridge
Seismic Analysis of Steel Deck plate Longitudinal ribs

and Composite Bridges:
Numerical Modeling,
Fig. 2 Steel bridge with
orthotropic deck
Cross beam
Main girder
plate
Deck Long
itudin
al rib
s
S
m
s bea
Main Cros
girde
r
• It should reflect the structural response in etc.) and their possibly eccentric
terms of deformation, strength, and local and connections.
global stability. • It should also include bearings and piers indi-
• It should include as many as possible vidually, piles, etc.
structural elements and parts of the • It should cover all construction stages and
bridge deck (cross frames, stiffeners, loading cases.
• Loads should be easily introduced. Kdir. Mdir includes the entire mass of the deck
• It should allow the performance of dynamic without the mass of the piers. The global spring
analysis and include the most important stiffness Kdir represents the combined stiffness of
modes of vibration. bearings and piers and the foundation of soil and
• The resulting output should be such that it is calculated from
enables easily the execution of the code-
prescribed verifications. 1 1 1
¼X þX
• It should be supported by commercial analysis Kdir Kbearings Kpiers
and design software. 1
þX where dir ¼ direction X or Y
Kfoundation
Models for Seismic Analysis (1)
There exist several possible models for bridge seis- The fundamental period for this single mass
mic analysis that could be employed depending on oscillator is calculated from (see Fig. 3a)
the bridge configuration, the bearing types, the sffiffiffiffiffiffiffiffi
connection between superstructure (bridge deck) Mdir X
and piers or abutments, the type of foundation Tdir ¼2p , Mdir ¼ m (2)
Kdir
with possible soil-structure interaction, etc.
Figure 3 shows possible modeling levels for
The seismic forces acting on the entire deck
seismic analysis of steel and composite bridges,
FEA, dir, tot are determined from the relevant
starting from the simple to the comprehensive
response spectrum. The seismic forces of bear-
ones. Simple models, like the first three in
ings at the top of one pier i may be determined
Fig. 3, can be used for preliminary seismic anal-
from
ysis. They may be employed for the derivation of
seismic forces and displacements on bearings, X
piers, foundations, or the soil. Comprehensive Kbearing, i þ Kpier, i þ Kfoundation, i
Fbearing, i ¼
models, like the last three of Fig. 3, are mostly Kdir
used at the main design phase since they also FAE, dir, tot
cover analysis for construction stages and service (3)
conditions where other loads due to traffic, wind,
temperature, time-dependent concrete effects,
This model may be applied for the longitudinal
etc. must be taken into account.
direction of straight bridges with continuous
In the following the various analysis models for
decks, when the mass of the piers is less than
steel and composite steel-concrete bridges are
20 % of the tributary mass of the deck. The
presented. Models for superstructures (decks) are
model may be also applied for the transverse
linear since decks are expected to remain elastic in
direction provided that all conditions (a) to
the seismic situation. Any inelastic activity should
(c) referred below apply (see EN 1998-2 2005):
be restricted to bearings, piers, piles, soil, etc. For
bearings and piers, nonlinear characteristics are
(a) L=b 4:0
provided. For other elements (piles, abutments,
where
soil), reference is made to other chapters.
L is the total bridge length and
b is the width of the deck.
(b) Dd=da 0:2
Rigid Deck Model where Dd and da are, respectively, the maxi-
mum difference and the average value of
In this model the superstructure is represented as the displacements in the transverse direc-
a single mass, Mdir, acting on a spring of stiffness tion of all pier tops under FAE,Y.
a Rigid deck models

FAE,m,X FAE,m,Y
Seismic forces on
mass element (m) m
FAE,m,Z
X
KY m KY
Y Z KX KZ KZ KX Spring elements representing

Mass
element bearings, piers and
Rigid element foundation, see Eq. (1)
representing the deck
b Individual pier models

Reaction forces
R1 R2 R3 R4 R5 Σ Ri
FAE,Y FAE,Y
Mpier,i = f(Ri, i=1, 2, ...)
Seismic
force
Beam element
Kϕ representing pier’s rigidity
KY
KZ Springs representing
foundation’s rigidity
c Spine models
Mass element
m
Beam elements
m representing deck’s
m stiffness
Seismic Beam elements
forces representing piers
Springs representing
foundation’s rigidity
d
Grillage models
Beam element
Mass element
(Main girders)
S
Spring element
(bearing) Intermediate
cross girders Transverse beams
(Slab)
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 3 (continued)
e
3D beam Models Longitudinal beams
(Slab)
Mass element
Transverse beams
(Slab)
Beam element
(top flange +
part of the web)
Post beam
Truss element
(part of the web) Beam element
(bottom flange +
part of the web)
Spring element
Truss element (bearing)
(part of the web)
f
Comprehensive FE Models Deck slab
(shell or volume elements)
Top flange
(beam elements)
Internal bracing
(truss elements)
Cross-Section
Stiffeners Bearings
(beam elements) Plated diaphragm (spring elements)
(shell elements)
Seismic Analysis of Steel and Composite Bridges: models. (b) Individual pier models. (c) Spine models. (d)
Numerical Modeling, Fig. 3 Analysis models for seis- Grillage models. (e) 3D beam models. (f) Comprehensive
mic design of steel and composite bridges. (a) Rigid deck models
(c) The theoretical eccentricity between the mass stiffness of the bearing device and is deter-
center of the deck and the stiffness center of mined iteratively to correspond with the bear-
the supporting members does not exceed 5 % ing’s displacements. In addition higher
of the deck’s length. damping values are achieved, so that
a reduction factor must be employed to the
The application of the rigid deck model may resulting seismic forces. More detailed infor-
be extended to cases where bearings with mation is given in sections “High-Damping
damping properties are employed. The bearing Reinforced Elastomeric Bearings” and “Lead
stiffness Kbearing is expressing the secant Rubber Bearings (LRB).”
Individual Pier Model Spine Models
This model may be used for seismic analysis of Spine models are appropriate for performing
bridges in the transverse direction. Each pier and multimodal dynamic analysis on bridges and let
the associated part of the superstructure is con- the seismic forces resisted by each pier deter-
sidered separately and represented as a single mined in a “natural” way in accordance with
mass oscillator. The mass of the oscillator is their relative stiffness. They may be employed
Mpier, i, and includes the mass of the deck for “normal” bridges, normal meaning more or
between half distances of piers, while Kpier, i less straight, low skew, narrow deck, and cross
represents the pier stiffness. section with limited distortion, i.e., with rigid
The fundamental period for pier i is calculated closely spaced transverse frames or crossbeams.
from In spine models the bridge deck is represented
by beam elements that are positioned at the cen-
sffiffiffiffiffiffiffiffiffiffiffiffiffi troid of the cross section and have six (6) degrees
Mpier, i
Tpier, i ¼ 2 p (4) of freedom (DOFs) at end nodes (Fig. 4).
Kpier, i The degrees of freedom are the translations
(u, v, w) along the principal axes coordinate sys-
Based on the fundamental period, the seismic tem (x, y, z) and the corresponding rotations
forces acting on top of the pier FAE,Y are deter- (yx, yy, yz). The resulting internal forces
mined from the relevant response spectrum. The (N, My, Mz) act at the gravity center, while the
individual pier model is not appropriate for shear forces and torsion moments (Vy, Vz, Mx)
curved or skew bridges, bridges with varying at the shear center. Cross-sectional warping
spans, varying pier lengths, etc. It may be used may be taken into account by introduction of an
for long bridges where each pier is able to act additional seventh DOF per node as independent
independent of the rest of the bridge. These variable (yx0 = dyx/dx), which results in the
requirements are met when following conditions bimoment Mw as additional internal moment
apply: (see Kindmann and Kraus 2011).
The initial beam cross section is the gross
(a) The seismic action in transverse direction is cross section. However, for wide flanges, either
mainly resisted by piers. of concrete or steel, effective widths must be
(b) There is no significant interaction between introduced to consider shear lag effects (Fig. 5).
adjacent piers. Although effective widths vary along the length
Tpier, i
(c) 0:90 Tpier , iþ1 1:10. of the bridge, they are smaller at internal supports
DEGREES OF FREEDOM S
INTERNAL FORCES ACTING ON INTERNAL FORCES ACTING
GRAVITY CENTER ON SHEAR CENTER
(N, My, Mz) (Vy, Vz, Mx, Mw)
θyj Gravity center Nj Mw(j)
node j
uj elem
ent Mx(j)
θzj Δz My(j) i node j+1 Mw(j+1)
Mz(j) node j
Vy(j) elem
My(j+1) Nj+1 ent
vj Shear i node j+1
Mz(j+1) Vz(j)
θxj center Vy(j+1)
Mx(j+1)
wj dθx Vz(j+1)
+ 7th DOF
dx
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 4 Representation of the bridge deck
by 7 DOF beam elements
Seismic Analysis of Steel Bending stress distribution in concrete flange

and Composite Bridges: beff,c beff,c
Numerical Modeling,
Fig. 5 Gross and effective Bending
cross section for wide stresses
concrete or steel flanges [s x ]
Effective part
Effective part
beff,a beff,a
Bending stress distribution in bottom steel flange
Uncracked zone Cracked zone Uncracked zone
0.85·L1 0.15·L1 0.15·L2 0.85·L2
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 6 Determination of the cracked
regions for composite bridges
than at spans, fixed values, equal to those at approximation of the true behavior (EN 1994-2
spans, are considered for global analysis. It 2005), however, for the purpose of seismic analy-
should be mentioned that effective widths are sis is considered as adequately precise.
calculated differently for concrete flanges than As an alternative to the fully cracked section,
for steel flanges (see EN 1993-2 2006 and EN the tension stiffening effect of concrete may be
1994-2 2005). taken into account in the cross-sectional proper-
For composite bridges the flexural stiffness, ties by increasing the area of steel reinforcement
denoted as E I1, is calculated for the uncracked by the factor 1/[1 (0.5 fctm)/(rs, tot fsk)] where
section in regions where concrete is in compres- rs, tot is the total reinforcement ratio, fctm the
sion. For seismic analysis, the short-term modulus mean tensile strength of concrete, and fsk the
of elasticity of concrete is considered. At hogging characteristic yield strength of the reinforcing
moment areas, concrete is in tension for beam- steel (see Vayas and Iliopoulos 2013).
type bridges. Cracking of concrete in those regions The torsional stiffness G It for composite box
is considered by introducing the flexural stiffness girders may be calculated from mechanics, where
E I2 of the “cracked” section in which the contri- the shear modulus of concrete is taken into
bution of the concrete slab is neglected. For con- account by introducing 0.2 or 0 as the Poisson’s
tinuous bridges the cracked region may be ratio for uncracked and cracked regions
considered to be adjacent to the internal supports correspondingly.
in a length equal to 15 % of the corresponding span Although the superstructure is represented by
length (Fig. 6). The 15 % rule constitutes a rough a single beam in spine models, bearings and piers
Seismic Analysis of Steel a

and Composite Bridges:
Numerical Modeling,
Fig. 7 Two-span
continuous bridge. (a)
Physical model and (b) At mid-span At pier
numerical spine model SECTION
Rigid link
Mass element Beam element
(deck)
Beam element
Spring elements (pier)
(bearing, abutments)
Spring element
(soil, foundation)
appear individually in the model. Figure 7a variety of bridge decks. In this model, the struc-
shows for a two-span continuous composite ture is idealized by means of a series of longitu-
girder bridge that is supported by two pile bent dinal and transverse beam elements rigidly
abutments and one two-column bent. The cross interconnected at nodes. Each element is given
section is a composite section consisting of two an equivalent bending and torsion inertia to rep-
I-girders and a concrete deck. The girders rest resent the relevant portion of the deck.
individually on bearings. Figure 7b shows the
spine model for this bridge. The cross section of Plate Girder Bridges
the superstructure is represented by a single beam Figure 8 illustrates a grillage representation of
element, while each bearing by two horizontal a simply supported composite bridge with four
springs. The beam and the springs are coupled main girders that may similarly be applied to
by a rigid link that represents the crossbeam. continuous systems or different number of main
Below the crossbeam are the two piers that are girders. Longitudinal grillage members are
represented by beam elements that rest on trans- arranged to represent the main girders with the
lational and rotation springs representing the pile inertia properties of the composite section (steel
foundation and the soil. Similar conditions apply section with a part of the slab corresponding to
to the abutments, where each bearing is the effective width). Transverse members repre-
represented by two horizontal springs, coupled sent the deck slab with thickness hc equal to the S
in series with horizontal abutment/foundation thickness of the slab and width b equal to the
springs. distance between transverse beams; it is conve-
nient to select b equal to the distance of the axle
loads. A non-cracked flexural rigidity for the slab
Grillage Models elements is usually applied. The torsional rigidity
of the transverse slab elements can be set to zero.
General The total in-plane second moment of area of the
The most popular computer-aided modeling slab is equally shared to the two extreme main
method for the analysis of composite bridges is girders (A and D), while the intermediate girders
the simulation by means of a plane grillage sys- (B and C) are given IZ = 0. This is because wind
tem. This is due to the fact that this system is easy loads mainly act on the edge girders of the bridge.
to apply and comprehend as well as that it has In case of intermediate cross girders whose stiff-
been proved to be sufficiently accurate for a wide ness may influence the transverse distribution of
beff
X
Y Mass element hc
D Main girders
Z C For A and D: IZ = IZ,tot / 2
B For B and C: IZ = 0
A
Gravity
center
Rigid
Spring elements b b
b h
(bearings)
hc
Rbear
Transverse beams Truss elements
(diagonals) u
(slab)
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 8 Grillage model of a plate girder
bridge
the vertical loads, these girders are taken into At piers truss elements are used for the rep-
account by appropriate modification of the trans- resentation of the cross braces. Due to the height
verse member properties. In case of long dis- h of the rigid elements, the geometry of the
tances between adjacent transverse beams, bracings in model may not follow the exact
X-bracing concrete elements may need to be set geometry of the bracings in real structure.
in the deck’s plane in order to simulate the dia- A height adjustment for the rigid elements may
phragmatic behavior of the deck slab. then be necessary. This should be done only for
Reinforced elastomeric bearings, usually the purpose of estimating the forces of the brac-
implemented in bridges, are represented by ing members due to horizontal loadings, i.e.,
three axial springs of equivalent stiffness wind or earthquake. It has to be stated that in
corresponding to the relevant stiffness properties most bridges the gravity center of the composite
in horizontal and vertical directions; the calcula- cross sections is located near the top flange. For
tion of the bearings’ stiffness is presented in such cases a height adjustment has little influ-
section “Reinforced Elastomeric Bearings.” The ence on final results.
axes of the main beams coincide with the center
of gravity of their cross sections. However, the Skew Bridges
bearings are positioned beneath the lower flange. In skew bridges the support abutments or piers
Accordingly, rotations of the main girders result are placed at angles other than 90 from the
in horizontal deformations u of the bearings and longitudinal centerlines of the girders (Fig. 9).
additional support reactions Rbear. The support The skew angle is usually defined as the angle
nodes are therefore put at a lower level from the between the longitudinal axis of the bridge and
grillage members and are connected to the longi- a line square to the supports. The presence of
tudinal beams by rigid vertical bars whose height skew affects the geometry and the behavior of
h is equal to the distance between the center of the structure. Special phenomena, like twisting
gravity of the main composite beams and the and out-of-plane rotation of the main girders dur-
bottom flange; for better accuracy, the shear ing concreting, uplifting forces at bearings, and
center of the cross section should be used fatigue problems due to out-of-plane web distor-
which is assumed to be the “real” center of rotation, make the analysis and design of skewed
tion. In case of intermediate cross girders whose bridges intricate. The transverse elements
stiffness may influence the transverse distribution representing the slab are usually oriented perpen-
of the vertical loads, these girders are taken into dicular to the main girders (orthogonal mesh);
account with beam elements of an appropriate this is the most usual grillage model used by the
stiffness. designers. Alternatively, the transverse members
Intermediate bracing
r ts
po
Skewed bracing
sup
ed
θ
ew
Skew angle
Sk
ORTHOGONAL MESH SKEWED MESH

Cross girders
ORTHOGONAL MESH CURVED BRIDGE
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 9 Grillage models for skew and
curved plate girder bridges
Transverse beams
X (internal braces)
Y
Fictitious girder Girders A and B
Z
A B
Mass element
rigid
Support
diaphragms bf
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 10 Grillage model of a simply
supported box girder bridge
can be placed parallel to the line of supports Box Girder Bridges

(skewed mesh). Generally the skewed mesh is A grillage model can be implemented also for
convenient for low skew angles (y < 20 ) or single-box girder bridges (Fig. 10). The box
when the intermediate bracing is not arranged girder is divided in two opened composite cross S
square to the main girders. sections in which the shear lag effect in the deck
slab and the lower flange is considered through
Curved Bridges the effective widths. The grillage is thus com-
Curved decks pose no particular problem for gril- posed of two main composite girders A and
lage modeling (Fig. 9). A curved bridge deck can B transversely connected with beams
be represented by a grillage of curved members representing the internal braces or diaphragms,
or of straight members. Some computer programs not the slab. The torsional rigidity of the compos-
support curved members but others do not. Gen- ite box girder is represented in the model by
erally, a grillage of straight beams with a very a fictitious girder located between the main com-
fine mesh is for small values of curvature suffi- posite girders. The central girder comprises also
ciently accurate. For highly curved bridges, the whole bending (IZ, tot) – and shear stiffness of
3D – or FE – models should be used (Adamakos the deck slab (AY
slab area). The flanges for
et al. 2011). girders A and B extend over their effective
a L b beff
+P −P +P1 −P1
T v
Rtop
H
30·ε·tw
T A0
ϕ T
Rbot
30·ε·tf
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 11 (a) Shear flow of the box section
due to St. Venant torsion. (b) Deformation of cross frames
widths. Care should be given that these widths are braces. This torsion is split into antisymmetric
different for concrete and for steel flanges and loading P = Mt/ L (see Fig. 11). The St. Venant
also different for analysis and for design. shear flow is given by
Longitudinal stiffeners in the bottom flange
are considered by “smearing” them over the Mt PL
T¼ ¼ (5)
flange width. The bottom flange is then consid- 2 A0 2 A0
ered having a total thickness of ttot = t + tadd,
where t is the thickness of the panel and tadd = where A0 is the shaded area in Fig. 11a.
Astiffeners/ bf. The same procedure is followed Therefore, the forces resisted by the cross ele-
for bridges with orthotropic steel deck, where the ments are equal to
top flange panel is taken into account with
increased thickness due to “smearing” of the lon- P1 ¼ P T Hinclined sin ’ ¼ P T H (6)
gitudinal stiffeners. However, the increase in
thickness does not apply to the web panel due to where P = Mt /L. It may be seen that the cross
the fact that existing longitudinal stiffeners are elements are resisting part of the global torsion. If
usually placed only to enhance its buckling resis- the webs are not inclined (j = 900), the force
tance and are not necessarily continuous. resisted by the cross elements is equal to:
Transverse beams in the global model repre-
sent the flexibility of the cross frames or of cross PL
P1 ¼ P H
braces. Since a single beam shall represent an 2HL
entire frame, the definition of its properties P
¼ , i:e:, 50 % of the acting forces
requires some preliminary analysis. The beams 2
are considered as rigid in bending but flexible in (7)
shear deformations. Accordingly they are
assigned an infinite in-plane second moment of Subsequently the vertical deformation v of the
area and a shear area As. Cross frames or cross cross frame due to antisymmetric loading P1 as
braces, and accordingly the transverse beam, determined before is numerically calculated
resist part of the torsion while they are not (Fig. 12). Due to symmetry, only half of the
strained due to global bending. A global torsion cross section is analyzed and hinges are placed in
moment Mt is partly resisted by the St. Venant the middle of the flanges. The two flanges and the
action and partly by the cross frames or cross web are represented by beam elements with cross
sections composed of the transverse stiffeners or girders are represented by beam elements, the
transverse girders and an associated flange width. former with steel sections, the latter with compos-
The effective width for the web and p theffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
bottomffi ite section including the effective slab width.
flange panels is equal to 30 e t, e ¼ 235 f y , Lateral torsional buckling phenomena of the
where fy = yield stress in MPa and t = thickness of top compression flange cannot be captured by
the web or bottom flange. this grillage model. 3D models are then
Orthotropic steel decks are represented in recommended.
a similar way. For composite bridges with con-
crete decks, the cross section of the relevant beam
element is composed of the transverse girder and 3D Models
the effective width of the concrete slab.
The deformation of a transverse beam is equal General
to v = P L/(As G) where G is the shear modulus The structural representation of bridge decks with
(see Fig. 12). By setting this deformation equal to truss girders or I-shaped plate girders may be
the deformation of the cross frame calculated done by means of 3D models as proposed from
previously, the shear area of this beam is defined Vayas et al. (2010, 2011) and Vayas and
as equal to Iliopoulos (2013). Truss girders are represented
by their chord and bracing members, while plate
PL girders are transformed to equivalent trusses.
As ¼ (8)
vG Such models have been proven to be advanta-
geous for modeling orthogonal, skewed, and
where v is the deformation of the system of curved bridges. Unlike grillage models, they are
Fig. 11 due to the load P1 given before. able to consider:
Half-Through Bridges • Eccentricities among the structural elements

Half-through bridges may be also represented by of a bridge and therefore additional internal
plane grillage models (Fig. 13). Main and cross forces and possible load distributions
−P
• The transversal variation in the level of the
neutral axis
• Torsion and distortional warping effects
+P
v • The dispersed structural behavior of the deck
slab, in which bending takes place in two
directions
As
• Buckling phenomena of the steel girders dur-
L ing erection stages S
Seismic Analysis of Steel and Composite Bridges: • Diaphragms, bracing systems, and
Numerical Modeling, Fig. 12 Deformation of the trans- stiffeners – possible overload or fatigue
verse beam of the global model effects are taken into account
Main Girder
der
Cross girder
er d
er gir
er gir
Trimm
Trimm
Cross girder
Main Girder
Main Girder Main Girder
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 13 Grillage model of a half-
through bridge
Girder Representation and Cfo. The nodes of the elements that represent
Steel and composite I cross sections are modeled the slab are the same nodes of those representing
by a “hybrid” truss as shown in Fig. 14. For the the upper flange of the truss. It is recommended
steel girder, the flanges of the truss are beam that a fine mesh is used for the beam elements of
elements with a cross section composed of the the concrete slab and the top flange of the steel
flange and a part of the web of the steel girder. girder so that a full shear connection is achieved.
Comparative analyses showed that one third of Without a fine mesh the beam elements of the
the web height may be associated to the flange. slab may deflect differently than those of the top
Therefore, the flanges of the truss are T-sections flange.
consisting of the flange of the steel girder and one
third of the web and are positioned at the center of Slab Representation
gravity of the T-section. The webs are Slabs are structurally continuous in both direc-
represented by diagonal truss elements with tions X and Y, and they resist applied loads by
width equal to one third of the web height and shear forces, bending moments and torques
thickness equal to the web thickness. It has been which are coupled with each other. For this rea-
also shown that the cross-sectional area Ad = son it was previously mentioned that the trans-
hw tw/ 3 for the diagonals adequately corre- verse slab elements of the grillage models should
sponds to the shear stiffness of the web. The not be used for the final design of the slab.
post-beams are located at a spacing s
5 % of A grillage model which considers the dis-
the span of the bridge. This distance is generally persed bending and torsion stiffness of a solid
acceptable for small and medium span bridges slab is illustrated in Fig. 15. The grillage mesh
because the angle between the diagonals and the should be sufficiently fine so that the grillage
flange elements usually remains between 35 and deflects in a smooth surface in a similar way as
45 . Post-beams represent both the in-plane and a real slab. A smooth deflected surface is equiv-
out-of-plane stiffness of the web. alent to the requirement that the twist @ 2w / @x@y
For a composite section the same procedure is the same in orthogonal directions and that myx
is followed, with the concrete slab represented = mxy. The spacing of the beams shouldn’t be
by another beam element connected with the less than 2.5 times the slab depth. Transverse
upper flange of the truss through the appropriate beams should have spacing similar to that of the
offset: offset = distance between the centroids Cc longitudinal beams. It is also recommended that
Longitudinal Mass element

Upper flange + 1/3·web loading
slab beam beff
(beam element) Fine mesh
Cc hc
Offset
Post beam hw
(beam element) Cfo 3
A= s · tw tw hw
Iy = s3· tw / 12 Cw
3
Iz = s · tw3 / 12 y
tw Cfu hw
z s 3
Lower flange + 1/3·web
1/3·web
(beam element) (truss element)
C : Centre of gravity
Ad = (hw/3)·tw
s ≈ 5%·L L
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 14 Truss idealization for a steel-
concrete composite girder
Y
GRILLAGE WITH FINE MESH
Z X
Slab (beam) element
A = b·hc
y h3
hc Iy = b· c
z 12
x b3
b Iz = hc·
12 h3
IT IT = Ix = b· c
6
bs
b = bs (for transverse elements)

≤2
b = bl (for longitudunal elements)

.5·
hc
Poisson’s ratio vc = 0
Gc = Ecm / 2
bl ≤ 2.5·hc
[M] Grillage output
True moments
Shear flow at the hc
edge of the slab
0.3·hc
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 15 Grillage model for an isotropic
solid slab
the row of longitudinal beams at each edge of the necessary structural elements, i.e., cross-
grillage should be located in a distance of 0.3 hc bracings, bearings, etc. Imperfections,
from the edge of the slab, where hc is the slab precambering, and girders with variable cross
depth (see Hambly 1990). This is where the resul- sections can also be implemented in the model.
tant of the shear flows is located. The width of the Therefore, structural phenomena which may be
edge member for the calculation of IT should be difficult or impossible to investigate with plane
therefore reduced to b-0.3 hc. grillages are included in the outputs of the 3D
model, e.g., arch effects in integral bridges with
The 3D Model Implementation longitudinally variable cross sections.
The grillage model for the slab’s representation
in Fig. 15 can be combined with the truss model
which is shown in Fig. 14. Figure 16 illustrates Modeling of Bearings
a 3D model which is recommended for the struc- S
tural analysis both of simple and continuous com- General
posite bridges. Attention must be paid so that Bridge decks rest usually on bearings that are
the grillage has its longitudinal members coinci- important elements for seismic analysis and
dent with the center lines of the steel sections. design. Most of the bearings, especially modern
At sagging moment areas, longitudinal slab bearings, provide seismic isolation to the bridge,
elements are used with their uncracked properties. reducing the forces on piers, foundations, and the
At hogging moment areas, concrete is considered soil. Seismic isolation is provided due to the low
as fully cracked and the total reinforcement is stiffness of such bearings that results in a shift to
positioned due to simplicity at center of the slab. longer fundamental periods and lower spectral
Transverse slab elements can be considered with values (see Chopra 1995). An appropriate repre-
their uncracked properties. sentation of bearings in the structural model is
One can see that the model can be set up in essential for seismic bridge design and will be
a detailed way by taking into account all the presented in the following. Bearings may deform
Transverse
slab beams
b
Longitudinal slab
beams hc
Longitudinal slab beams (uncracked)
(fully cracked) bl
Reinforcement bl hc
As,tot = Aso + Asu Aso
AT SAGGING MOMENT AREAS
Iy ≈ 0 y As,tot
Iz ≈ 0 z Asu
IT ≈ 0
AT HOGGING MOMENT AREAS
Beam element
(bottom flange +
1/3 of the web)
Post beam
Truss element
Spring elements
web
(1/3 of the web)

(bearings)
Beam element
X - Bracing (top flange + 1/3 of the web)
Bearing stiffener (truss elements)
(beam element)
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 16 3D model of a two-span
continuous bridge
during seismic action in the nonlinear range. Te is the nominal thickness of the elastomer
Accordingly linear and nonlinear characteristics layers.
are provided. dir = global direction X or Y.
Reinforced Elastomeric Bearings The shear modulus Gb considered has an

Reinforced elastomeric bearings (Fig. 17) consist increased value compared to G = 0.9 MPa of
of uniformly spaced layers of elastomer elastomers to account for the speed of loading in
(natural or synthetic rubber) and reinforcing the seismic situation so that (see EN 1998-2
steel plates that obtain their bonding through the 2005):
process of vulcanization. The equivalent viscous
damping ratio for common elastomers x is less Gb ¼ 1:1 G (10)
than 6 %.
Reinforced elastomeric bearings are intro- Beyond this, upper and lower values of the
duced as linear springs in global analysis. The shear modulus are introduced in the seismic com-
spring stiffness in each unrestrained horizontal bination with recommended values as follows:
direction may be obtained from
Gb, max ¼ 1:5 Gb ¼ 1:65 G (11a)
A Gb
Kdir ¼ (9)
Te Gb, min ¼ 1:0 Gb ¼ 1:1 G (11b)
where: Upper values are supposed to result in maxi-

mum forces, minimum values maximum dis-
A = a b or p D2/4 is the plan area of the placements. However, this might not be always
bearings. the case, depending on the resulting periods and
a elastomer
b FAE
M
steel
plates superstructure Ktot
elastomer
KX, KY, KZ = ∞
M
Natural period: T1 = 2 ⋅π⋅
Ktot
Spectral
acceleration Seismic force: FAE = M ⋅ Sad(T1)
Sad(T)
steel plate
steel plate Sad(T1)
T
T1
steel plate
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 17 (a) Composition of reinforced
elastomeric bearings. (b) Single DOF seismic analysis of a simply supported composite bridge with a rigid deck model
Bolts
Bottom flange of
steel girder Pinned Support
Sole plate
PTFE sliding layer Concave spherical segment
In case of
Convex spherical guidance
High strength, non- segment
Roller Support
shrinkable mortar
Concrete of sub-structure
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 18 Layout and representation of
spherical bearings
the spectral values. In vertical direction the bear- is taken into account by application of the reduc-
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ing is practically incompressible so that the ver- tion factor Z ¼ 10=ð5 þ xÞ 0:55 on the S
tical displacement is considered to be restrained forces determined from a response spectrum
without use of springs. with 5 % damping.
High-Damping Reinforced Elastomeric Spherical Bearings

Bearings Spherical bearings consist of a sole plate that
These are bearings in which the common elasto- transfers loads from the superstructure,
mer is substituted by high-damping elastomer. a concave spherical segment and a convex spher-
Their equivalent viscous damping ratio x reaches ical segment that provides a mating surface for
values between 10 % and 20 %, while common the concave segment and transfers load to the
elastomeric bearings have damping ratios x substructure (Fig. 18). PTFE (Teflon) sliding
below 6 %. These bearings are modeled in global layers are provided between the three parts to
analysis as linear springs like common elasto- allow horizontal displacements and rotation in
meric bearings. The reduction in seismic forces all directions. Spherical bearings provide
Bolts Bottom flange of

steel girder
Sole plate Pinned Support
cap Sealing ring
PTFE sliding layer Elastomer pad
steel pot
Roller Support
High strength, non-
shrinkable mortar
By provision of
Concrete of sub-structure sliding material
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 19 Layout and representation of
pot bearings
a b F
Bottom flange KP
of steel girder Fmax
Elastomer Upper plate
Lead core Steel plates Fy
F0
Lower plate
High strength, non- d
shrinkable mortar Keff
K dy dbd
Concrete of e
sub-structure
ED
Studs or
anchor bolts
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 20 (a) Layout of lead rubber
bearings. (b) Force-displacement behavior of lead rubber bearings
therefore only vertical support. They are accord- Lead Rubber Bearings (LRB)
ingly represented by hinges allowing also hori- These are common reinforced elastomeric bear-
zontal displacements. Spherical bearings may be ings with low-damping elastomer and
also guided to restrain displacements in one or a cylindrical lead core that may reach damping
both horizontal directions. In this case displace- values up to 40 % (Fig. 20a).
ments in the relevant horizontal direction are During cyclic loading the lead core is yielding
restrained. and strain hardening so that the hysteretic
response of the bearings, as illustrated in
Pot Bearings Fig. 20b, is typical for a yielding and strain
Pot bearings are based on the incompressibility of hardening material. In analysis the bearings
natural rubber when placed in a closed steel pot may be accordingly represented by a bilinear
where natural rubber behaves like a fluid horizontal springs with elastic stiffness Ke
(Fig. 19). Pot bearings are able to transfer high for displacements up to the yield displacement
compression forces in a small surface and allow and post-elastic tangent stiffness Kp for larger
rotations around all axes. They provide horizon- displacements.
tal and vertical support while rotations are free. The elastic stiffness may be determined from
Provision of a sliding material allows horizontal
displacements in which case the bearings provide
only vertical support. Ke ¼ KR þ KL (12)
where: where all symbols are indicated in Fig. 20b. For

linear analysis the value of xeff is updated in each
KR is the shear stiffness of the elastomers. iteration step in order to determine the reduction
KL is the shear stiffness of the lead core. factor for increased damping on the resulting
seismic forces. When the rigid deck model is
The yield force is determined from employed, all bearings are represented by
a single spring with an equivalent stiffness and

KR damping ratio determined from
Fy ¼ FLy 1 þ (13)
KL
X
Keff , tot ¼ Keff , i (19)
where FLy is the yield force of the lead core.
The force at zero displacement is equal to X
ED, tot ¼ ED , i (20)
F0 ¼ Fy Kp dy (14)
where the summation refers to all bearings i.
The post-elastic stiffness is attributed to the The equivalent damping ratio for all bearings
lead core only and is calculated from is calculated from
Fmax Fy
KP ¼ ¼ KR (15) ED, tot
dbd dy xeff , tot ¼
2 p Keff d2bd
However, the above bilinear representation is 1 ED, tot
appropriate for nonlinear analyses. For linear ¼ X (21)
2 p dbd
2
Keff , i
analysis the spring properties are linearized by
introduction of the secant stiffness. This effective
stiffness is written as The steps of the iterative procedure for the
rigid deck model are as follows:
Fmax
Keff ¼ (16) 1. Selection of the bearings
dbd
2. Initial assumption for the displacement dbd
The analysis must be iterated in this case by 3. Calculation of Keff, tot, xeff, tot and ED, tot as
introduction of an initial value for Keff and update a function of dbd
it consecutively as a function of the resulting 4. Determination of the fundamental pperiod of theffi
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
maximum displacement of the previous step. equivalent system Teff ¼ 2 p Mtot =Keff , tot
Figure 20b indicates that lead rubber bearings where Mtot is the total mass S
dissipate energy when displaced beyond the yield 5. Determination of the seismic forces and the
displacement. The energy dissipation per cycle new displacement from Table 1
equals to the shaded area of Fig. 20b and may be 6. Repetition of steps 2–5 until converge of the
determined from displacements is achieved

ED ¼ 4 Fy dbd Fmax dy (17) Friction Pendulum Bearings (FPS)
Friction pendulum bearings constitute a variation
The energy dissipation may be transformed to of the spherical bearings, where the plane sliding
increased damping with an effective damping top surface is replaced by a curved one (Fig. 21a,
ratio xeff determined from b). They exhibit, unlike spherical bearings, a
re-centering capability and are not subjected to
ED ED
xeff ¼ ¼ (18) rotations. The force at zero displacement is pro-
Fmax dbd 2 p Keff d2bd vided by the friction and is given by
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Table 1 Seismic forces and design
displacements (EN 1998-2 2005)
Fundamental
period Seismic forces Design displacement
TC Teff, tot FAE ¼ b Teff, tot n Ag W
TC
d ¼ eff, tot d
bd
T
C
TC
TD
TD < Teff, tot FAE ¼ b TT2C TD n Ag W dbd ¼ TTDC dC
eff , tot
Spectra Acceleration spectrum Displacement spectrum

Spectral Displacement
acceleration spectrum
Sad(Teff) dbd(Teff)
Teff Teff
TB TC T TD TB TC TD
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

Notation Ag peak ground acceleration, W total weight, and Z ¼ 10= 5 þ xeff , tot 0:55
a dbd NEd
c
F
Rb F0 Kp = NEd / Rb
Fmax
F0
d
dbd
b
R2 ED
R1
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 21 Layout of (a) single- and (b)
double-friction pendulum bearings. (c) Force-displacement behavior of friction pendulum bearings
F0 ¼ m N (22) good approximation when d / R 0.25 and con-

sidering the developing friction between sur-
The kinematics of the bearings leads to fol- faces, the following relation may be derived for
lowing relations: the restoring force:
sin j ¼ d=R (23a) dN

F¼ _
þ m N sign dÞ (24)
R
tan j ¼ F=N (23b)
where:
Equating the above equations for small angles
sinj
tanj that may be considered as valid with N is the vertical force.
d is the horizontal displacement. and for the overall system are equal to
R is the radius of the spherical surface. xeff, tot = xeff, i.
m is the friction coefficient. The steps of the iterative procedure for the

sign dÞ_ is the sign of the velocity. rigid deck model are as follows:
It may be seen that the restoring force 1. Selection of the bearings, specifically the fric-
increases linearly with the displacement and the tion coefficient m and the radius R
support reaction, i.e., with the mass. For two 2. Initial assumption for the displacement d
spherical surfaces, similar relations apply, 3. Calculation of Keff, tot and xeff, tot as a function
where R = R1 + R2 (Fig. 21b). of d
Figure 21c displays the force-displacement 4. Determination of the fundamental period of
characteristic of friction pendulum bearings that the equivalentp system
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
indicates a symmetrical response. The effective Teff ¼ 2 p Mtot =Keff , tot, where Mtot is the
linear spring stiffness is proportional to the com- total mass
pression force and is determined from 5. Determination of the seismic forces and the
new displacement from Table 1

6. Repetition of steps 2–5 until converge of the
F 1 m
Keff ¼ ¼N þ (25) displacements is achieved
d R d
The energy dissipation in a cycle is given by The equivalent spring stiffness and energy
dissipation may then be calculated for each indi-
ED ¼ 4 m N d (26) vidual bearing.
and the equivalent effective damping ratio by Fixed Bearings

Until the 1950s, steel bearings were used that
ED 2m consisted of four types: pins, rollers, rockers,
xeff ¼ ¼ (27)
2 p F d p ðd=R þ mÞ and metal sliding bearings. Pins are fixed bear-
ings allowing rotations. Rollers and rockers allow
When the rigid deck model is employed, the translation and rotation, while sliding bearings
equivalent stiffness of all bearings is determined utilize one plane metal plate sliding against
from another, with PTFE as intermediate lubricant
material, to accommodate translations. These
X

1 m older bearing types are modeled as hinges (see
Keff , tot ¼ Keff , i ¼Wtot þ (28)
R d Fig. 22). Fixed bearings may also be represented
by springs, their flexibility resulting from the S
where the summation refers to all bearings i and flexibility of the bearing bars and the elongation
Wtot is the total weight of the bridge in the seis- of the anchor bolts.
mic situation. The stiffness in longitudinal direction of fixed
The individual spring stiffness for each bear- bearings may be determined from
ing i is then
1 h3 h lb
¼ þ X4 (30)
Ni K 3EI EA li
Keff , i ¼ Keff , tot (29) 1
Wtot
where:
where
Ni is the compression force of bearing i. E is the modulus of elasticity for steel.
If bearings with identical radius R are I is the second moment of area of the bearing bar
selected, the damping ratios for each bearing (I = b3 h/12).
l4
b l3
Bearing bar
h
X a a
h
Longitudinal lb
direction
Anchor bolts
l1
l2 Section a-a
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 22 Fixed bearing
h is the height of the bearing bar. a b

A is the area of the anchor bolts. m m7
lb is the length of the anchor bolts.
m m6
li are the distances of the anchor bolts from the tip
of the base plate. m m5
m m4
The stiffness in transverse direction is given
by a similar equation, neglecting the first term m m3
due to the high stiffness of the bearing bar in this
m m2
direction and changing the direction of the
lengths li. If a bearing set is going to be m m1
represented by one translational and one rota-
tional spring, the relevant spring stiffness may
Seismic Analysis of Steel and Composite Bridges:
be expressed by Numerical Modeling, Fig. 23 Representation of piers
Xn through discrete beam mass models – pier with (a) con-
Khor ¼ K
1 i
(31a) stant and (b) variable dimensions
and
Xn
Krot ¼ Ki l2bi (31b) subdivided in more elements, especially if
1
its cross section varies along the height.
where: The cross section of each element corresponds
to the average pier section within its length
Ki is the stiffness of each bearing as determined (Fig. 23).
before. In frame bridges the superstructure runs
n is the number of bearings in the set. continuously and is rigidly connected to the
lbi is the distance of bearing i to the centerline of piers so that bearings and expansion joints are
the bridge deck. avoided. Such bridges are allowed to behave
nonlinear in the seismic situation. Unlike in
moment-resisting frame buildings where plastic
Modeling of Piers hinges are developing in the beams, plastic
hinges are expected to develop in bridges in the
Piers are generally represented by means of piers so that the superstructure behaves elastic
6 DOF beam elements. The pier is usually (Fig. 24).
Seismic Analysis of Steel Earthquake

and Composite Bridges: force
Numerical Modeling,
weak
Fig. 24 Plastic hinges in
weak
frame bridges Plastic hinge (ductile)
Strong piers
X
a b Composite girder
Y (beam element)
Z
Soil spring Seismic forces Mass element
Endscreen wall
Gravity center FAE,X

Approach Rigid link FAE,Y
slab FAE,Z
Temporary bearing Pile cap
(beam element)
Pile cap
Soil springs
Backfill
Pile
(beam element)
Pile
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 25 (a) End connection of super-
structure to abutments in integral bridges. (b) Numerical grillage model of the end region of integral bridges
Accordingly for nonlinear analyses, the A structural model that includes nonlinearity
nonlinear properties of the piers shall be in a distributed fashion, using finite fiber
modeled. Two types of models may be used: elements, is able to characterize in higher detail
the reinforced concrete elements and thought
(a) Plastic hinge models: The nonlinear behav- to capture more accurately response effects
ior of the pier elements is located in rotational on such elements. Geometrical and
springs at the ends of the elastic behavior part material properties are the only required ones as
of the element. The characterization of input. S
a plastic hinge requires a moment-curvature
diagram to be defined. This is obtained from
the cross-sectional response to monotonic Modeling of Integral Bridges
loading that is derived from a finite element
analysis. Integral bridges are those where the steel girders
(b) Plastic zone models: This accounts for are encased in the abutment walls so that bearings
spread-of-plasticity effects in sections and and expansion joints are avoided (Fig. 25a). Steel
along the beam-column element. Beam- girders are connected to the abutment through
column finite fiber elements may be used that reinforcement that passes through their web and
are able to better characterize reinforced con- by shear studs in the flanges or the web. Move-
crete elements. Consequently, higher accuracy ments of the bridge due to thermal actions or
in the structural damage estimate is attained, time-dependent deformations of concrete are
even for the case of high inelasticity levels. accommodated by the flexibility of piles that
Seismic Analysis of Steel

and Composite Bridges: Road layout
Numerical Modeling, Concrete slab
Fig. 26 Finite element (shell elements)
model of a filler deck
bridge
Steel girders
(beam elements)
Cross section
Spring elements
(bearings)
support the abutments. To increase the support Horizontal springs in the longitudinal direction
flexibility, steel piles are oriented with their represent the soil-passive pressure behind the
flanges parallel to the girders so that bending abutment.
occurs around their weak axis. Integral bridges When nonlinear methods of analysis are
are allowed for straight and skewed bridges but employed, the soil springs and the beam elements
not for curved bridges. Settlements of the backfill of the piles are assigned nonlinear properties. In
are accommodated by an approach slab. Integral an extension of the model, the connection
bridges exhibit a good seismic performance due between girders and abutment is considered flex-
to increased redundancy, smaller displacements, ible so that compression-only springs are intro-
and larger damping due to nonlinear soil-pile- duced at the contact points between girders and
structure interaction. concrete material.
A grillage model for the end region of integral
bridges is illustrated in Fig. 25b. Steel girders and
piles are represented by 6 DOF beam elements. Modeling of Filler Deck Bridges
Beam elements representing the girders are
assigned the properties of the cracked composite Filler beam where small distanced steel beams
cross section due to the fact that negative are encased in concrete are small span bridges
moments develop in abutment so that the con- and are used both as simply supported (max. span
crete slab is in tension. This section includes the
15 m) and continuous systems (max. span
section of the steel girder and the reinforcement 30 m). Shear connection is ensured by transverse
within the effective width. Due to the small dis- reinforcement that passes through holes at the
tance between piles, the pile cap is supposed to be webs of the steel beams. Figure 26 shows such
rigid so that piles are connected with horizontal a bridge that is composed of closely spaced lon-
rigid elements and not with beam elements. The gitudinal IPE girders resting on two transverse
connection of the embedded girder and the abut- steel H-shaped (HEA section) girders that are
ment is supposed to be rigid. Accordingly the supported by reinforced elastomeric bearings.
beam elements of the girders are connected by The bridge is orthogonal but the road is skew.
vertical rigid elements with the pile cap. Finally, The longitudinal elements are represented by
horizontal rigid elements are connecting the beam elements their section being composed
beam elements of the girders. The vertical pile from the steel girder and the associated concrete
support is supposed to be hinged. Springs in two embedment. The transverse H girders are also
horizontal directions represent the soil that sup- represented by beam elements with an HEA
ports the piles, with stiffness varying in depth. cross section. Shell finite elements represent the
Ag,Y [m/sec2]
Accelerograms Max=0.92 m/sec2
Malibu – Point Dume, 17/1/1994
Ag,X[m/sec2]
0.46
1.27 Max=1.27 m/sec2
Cross girder’s internal forces
0.00
Vz[kN]
Max = 94.29 kN Cross girder
0.64 -0.46 96.0
t [sec] 48.0 Vz
14.995 29.990 44.985 59.98

0.00 0.0
-48.0
-96 t [sec]
-0.64 Min = -96.04 kN
14.995 29.990 44.985 59.98
My [kNm]
-1.27 t [sec] Max = 86.58 kNm Cross girder
85.0
14.995 29.990 44.985 59.98
My
X 42.5
Y
Y X
3D – Beam Model
Z
0.0
34
Z m
-42.5
KX = 2.8 MN/m -85.0 t [sec]
KY = 2.8 MN/m Min = -85.01 kNm
14.995 29.990 44.985 59.98
u [cm] Support translations
2.10 Max=1.8 cm 34
m
1.05 KX = 7.92 MN/m
KY = 7.92 MN/m
0.00
−1.05
Support Reactions Cross-girder
−2.10 t [sec]
1000+300 mm
Min = -2.1 cm RX [kN]
HEB 300
14.995 29.990 44.985 59.98 150.0 Max = 146.04 kN
½ HEA 400
v [cm]
Max = 4.4 cm 75.0
4.40
0.0
2.20 RZ [kN]
-75.0 Max = 234.56 kN
0.00 235.0
-150.0 t [sec]
−2.20 Min = -164.61 kN 117.5
14.995 29.990
−4.40 t [sec] RY [kN] 0.0
Min = -4.4 cm Max = 355.50 kN
360.0
14.995 29.990 44.985 59.98 -117.5
180.0
-235.0 t [sec]
0.0 Min = -238.92 kN
14.995 29.990 44.985
-180.0 S
-360.0 t [sec]
Min = -362.07 kN
14.995 29.990 44.985 59.98
Seismic Analysis of Steel and Composite Bridges: Numerical Modeling, Fig. 27 Time history analysis of
2 34 m long span plate girder bridge with a 3D beam model
concrete slab that serve for an easy application of low modulus of elasticity. However, concrete
traffic loads, distribute vertical and horizontal strength is introduced with its design value in
loading between longitudinal beams, and ensure order to allow correct design of the composite
diaphragm action. In order to correctly represent action. Deflections are calculated as average
the stiffness of the deck, the concrete of the values between uncracked and cracked analysis,
longitudinal beam elements is assigned a very the latter being simulated by considering the
thickness of the concrete shell elements not as the recommendations for plate, filler deck, and box
total slab thickness but as the thickness of the girder bridges are provided. The implementation
compression zone. Bearings are represented by of structural bearings in the models is also
linear springs. The above-described model is discussed, and guidance on the calculation of
a combination of a grillage and a FEM shell the bearings’ stiffness and the seismic forces is
element model. given. The entry ends with an example of a time
history analysis of a continuous composite beam
and the calculation of reaction forces and
Application: Time History Analysis with displacements.
a 3D Beam Model
A time history analysis of a two-span composite Cross-References

bridge with six main girders is indicatively dem-
onstrated in Fig. 27. The bridge has been modeled ▶ Buildings and Bridges Equipped with Passive
according to the recommendations of the section Dampers Under Seismic Actions: Modeling
“3D Models.” The superstructure rests on and Analysis
low-damping elastomeric bearings whose stiff- ▶ Seismic Analysis of Concrete Bridges:
ness has been estimated through Eqs. 8, 9, 10, Numerical Modeling
and 11a and 11b. For the analysis a damping
factor equal to 2 % for the three first natural
modes has been implemented, value taken from References
the fabricator of the bearings. The mass of the
superstructure includes all the permanent loads Adamakos T, Vayas I, Petridis S, Iliopoulos A (2011)
and 20 % of the traffic loads. One can see that Modeling of curved composite I-girder bridges using
spatial systems of beam elements. J Construct Steel
with the 3D beam model, the time variation of the
Res 67(3):462–470
vertical reaction forces RZ can be calculated; Chopra AK (1995) Dynamics of structures. Prentice Hall,
such a calculation with a grillage model would Upper Saddle River
not be feasible. Moreover, the internal forces of EN 1993-2 (2006) CEN European Committee for Stan-
dardization. Design of steel structures. Part 2: Steel
the cross girders are calculated. Not only the
bridges
grillage model but a detailed FE model as well EN 1994-2 (2005) CEN European Committee for
would not give the possibility of such Standardization. Design of composite steel and con-
a calculation since shell or volume elements crete structures. Part 2: General rules and rules for
bridges
offer stresses and not forces or bending moments
EN 1998-2 (2005) CEN European Committee for Stan-
as final results. With the 3D beam model, the dardization. Design of structures for earthquake resis-
frame action of the cross girders (HEB 300) and tance. Part 2: Bridges.
the transverse stiffeners (1/2 HEA 400) could be Hambly EC (1990) Bridge deck behaviour, 2nd edn. E. F.
& N. Spon, London
adequately investigated.
Kindmann R, Kraus M (2011) Steel structures – design
using FEM. Ernst & Sohn, Berlin
Vayas I, Iliopoulos A (2013) Design of steel concrete
Summary composite bridges to Eurocodes. CPC Press, Taylor
and Francis Group, Boca Raton
Vayas I, Iliopoulos A, Adamakos T (2010) Spatial
Different types of models are used for the design systems for modelling steel-concrete composite
and seismic analysis of steel and steel-concrete bridges – comparison of grillage systems and FE
composite bridges. In this entry 1D, grillage, models. Steel Construct Des Res 3:100–11
Vayas I, Adamakos T, Iliopoulos A (2011) Three dimen-
spine, 3D beam, and finite element models are sional modeling for steel-concrete composite bridges
presented. The advantages and drawbacks of using systems of bar elements – modeling of skewed
each model are debated, and modeling bridges. Int J Steel Struct 11(2):157–169
Seismic Analysis of Steel Buildings: Numerical Modeling 2621
the layout of structural elements, as will be

Seismic Analysis of Steel Buildings: described in the following section. The prelimi-
Numerical Modeling nary architectural drawings include proposed col-
umn locations, indicated by their I-sections in the
Charis J. Gantes plan views of Figs. 1 and 2.
Institute of Steel Structures, School of Civil
Engineering, National Technical University of
Athens, Zografou Campus, Athens, Greece Conceptual Design
The initial stage of the structural design process

Synonyms consists of the so-called conceptual design, in
which the different structural elements that are
Numerical simulation of steel buildings; Struc- needed to support the structure are selected and
tural design of steel buildings in seismic regions are geometrically defined, taking into account the
architectural requirements. In parallel, all restric-
tions imposed by the building’s use, as well as its
Introduction location and associated environmental actions
have to be satisfied.
The aim of this contribution is to present the state To that effect, close cooperation and interac-
of the art in numerical modeling of steel build- tion of the structural engineer with other involved
ings, aiming at their structural analysis and engineering disciplines are necessary. Such dis-
design. To that effect, the successive stages of ciplines include, at a minimum:
this process will be presented, including the con-
ceptual design, numerical modeling, application • The architectural engineering team, to deter-
of support conditions and loads, structural analy- mine the overall geometry; positions of open-
sis, and design checks. The two latter issues, ings; types of materials for floors, cladding,
structural analysis and design checks, will be etc.; and associated dead loads and, in general,
discussed briefly, to the extent that they interact to ensure that structural choices do not inter-
with the modeling process, as they are also cov- fere with the building’s desired function and
ered in detail in other parts of this encyclopedia. aesthetics
The interaction with other engineering disci- • The mechanical/electrical/HVAC engineering
plines throughout the design process will be dem- team, to determine loads that are associated
onstrated. For the purpose of this presentation, a with these functions of the building and to
prototype building will be used, which is prevent conflicts with required shafts and
described in the next section. other openings for passage of lifelines S
• The geotechnical engineering team, to obtain
data about the soil and its mechanical proper-
Description of Prototype Building ties and to determine the type of the building’s
foundation
A simple single-story laboratory building, part of
an industrial project, will be used as prototype In conceptual design, all basic decisions,
building for this presentation. The building’s plan regarding material properties, structural sys-
view is 28.00 m by 15.00 m (Fig. 1), and there is a tem, loads, and codes to be applied, are
small loft in part of the building (Fig. 2). The roof taken. All structural engineering experts agree
is double pitched with minimum required clear- that this is the most important design stage and
ance equal to 4.50 m (Fig. 3). The locations of that these decisions affect significantly the pro-
exterior doors are fixed, imposing restrictions on ject’s design and construction schedule and
1 2 3 4 5 6
A
2622
28000
4000 4 BAYS @ 6000 = 24000
DUST EXTRACTION
FILTRATION UNIT
A
STORAGE F.E.
UNIT METAL TOPPED TABLE PULVERISING PULVERISING
CRUSHER MILL MILL
w/ EXTRACTION HOOD
MECHANICAL
ROOM
FREEZER FRIDGE
15000
B
Seismic Analysis of Steel Buildings: Numerical Modeling
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 1 Architectural plan view of ground floor of prototype building (out of scale)
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 2 Architectural plan view of loft of prototype
building (out of scale)
cost, sometimes even its feasibility. This is a tools, both for preliminary design of the dif-
process mostly dictated by the available expe- ferent building’s facilities but also for commu-
rience of all participating design teams but can nication between involved teams (building
nowadays be assisted by modern software information modeling, BIM). However, a
2624 Seismic Analysis of Steel Buildings: Numerical Modeling

Buildings: Numerical
Modeling,
Fig. 3 Transverse section
A-A of prototype building
(out of scale)
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 4 3D view of prototype building’s 3D model
more detailed treatment of this design stage is The building’s structural system consists of:
beyond the scope of this presentation.
Several conceptual design choices for the pro- • Two-span main frames spaced at 6.00 m (with
totype building described in section “Description the exception of the first span, which is equal
of Prototype Building” are illustrated in Figs. 4, to 4.00 m) in accordance with the architectural
5, 6, 7, 8, 9, 10, and 11. The figures have proposal. Columns and girders have I-section,
been obtained from a 3D model of the building’s oriented so that the strong axis acts for
structure, made with a specialized modeling in-plane loading. It is noted that a 6.00 m
software, which enables integration of architec- distance between main frames is optimum for
tural and structural engineering and other solu- avoiding loss of material in purlins and longi-
tions and production planning systems, working tudinal beams, taking into account that steel
for both detailing and producing steel structures members are commonly fabricated in 12.00 m
of any type. lengths. Haunches are provided between
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 5 Plan view of prototype building’s 3D model
above foundation level
S
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 6 Plan view of prototype building’s 3D model
above loft level
columns and girders at internal frames, in • Longitudinal beams connecting the main
order to achieve reduction of girder cross sec- frames at the top and at mid-height of col-
tion and to accommodate the required number umns. On one side of the building, the
of bolts for a moment connection. External mid-height head beam is eliminated in the
frames are equipped with vertical X-bracing, last span, to accommodate a required door at
so that truss action dominates rather than that location. A beam over the door is used
frame action. As a result, smaller column and instead. It is noted that this is not an optimum
girder sections are possible, and there is no structural configuration, but it is an
need for haunches between columns and acceptable compromise to respect functional
girders. requirements.
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 7 Roof plan view of prototype building’s 3D model
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 8 Side view of prototype building’s 3D model
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 9 Front view of prototype building’s 3D model
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 10 Transverse section of prototype building’s 3D
model at spans with loft
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 11 Transverse section of prototype building’s 3D
model at spans without loft
• Vertical X-bracing at two spans to provide at the roof’s peripheral panels, to ensure suffi-
lateral stability in the longitudinal direction. cient diaphragm action.
The type of bracing depends on panel geome- • Roof and side purlins to support cladding.
try, aiming at inclination that is as close as
possible to 45 . A peripheral reinforced concrete wall is pro-
• Horizontal roof bracing at the same spans vided, to ensure water tightness and to protect the
where vertical bracing is provided, as well as steel columns from eventual vehicle collisions.
Taking also into account the available geotechni- Numerical Modeling

cal information for this site, foundation beams are
provided at the periphery of the building, along For the simulation, analysis and design of super-
the axis of intermediate columns and below the structure and foundation use is made of an
auxiliary columns supporting the loft. A steel appropriate finite element software. Views of
staircase is provided to connect the ground floor the finite element model are shown in Figs. 12,
with the loft. 13, 14, and 15.

Modeling, Fig. 12 3D
view of prototype
building’s finite element
model
1 2 3 4 5 6
4 6 6 6 6
C
7.31
B
7.31
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 13 Foundation plan view from prototype build-
ing’s finite element model
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 14 Loft’s plan view from prototype building’s
finite element model
1 2 3 4 5 6
CB
S
A
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 15 Roof’s plan view from prototype building’s
finite element model
Some recommendations of good practice columns and girders are continuous but do not
when setting up the finite element model are the transmit moments to each other, would require
following: rigid links for correct modeling, which is pos-
sible in most structural programs, but would
• It is recommended not to include roof and side render the model unnecessarily complex. Only
purlins and other secondary members in the those members that have an active role in the
numerical model. The type of connection building’s spatial response to either vertical or
between purlins and main frame, which is a horizontal actions need to be modeled. Other
simple support so that both purlins and frame members can be modeled, analyzed, and

Modeling,
Fig. 16 Column base
connections of prototype
building, modeled as hinges
dimensioned independently, and the loads flanges are commonly modeled as hinged, as
they transfer to the frames can be applied is the case with the supports of all columns in
directly upon the frame columns and girders the prototype building (Fig. 16). In case bolts
in the 3D finite element model. are provided outside the flanges, rigid moment
• Haunches can be modeled by means of uni- connections are considered in the model, as is
form beam section over the entire length of the the case with the column to girder connections
haunch, usually the one at haunch mid-length. in the prototype building (Fig. 17). Welded
This provides a satisfactory representation of connections are usually modeled as rigid,
increased girder stiffness in the area of the while connections between or at the end of
haunch and enables safe prediction of strength bracing members (Fig. 18) are modeled as
for the design phase. hinged.
• Particular attention is needed for the appropri- For moment connections, no intervention is
ate modeling of supports and connections, by required in the model, as this is the default
releasing the appropriate member degrees of option for frame structures. Hinged connec-
freedom depending on the connection detail. tions are modeled by means of releases of
Even though real connections between steel appropriate rotational degrees of freedom at
members are neither perfect hinges nor perfect the members’ ends. In Figs. 19 and 20,
moment connections, it is common practice to moment releases at the ends of roof and verti-
model all connections as either hinges or cal bracing as well as secondary beams are
moment connections, at least initially. Modern denoted by black dots.
codes provide tools for eventually classifying • If there is a diaphragm in the structure, as, for
connections as semirigid (e.g., European example, is the case with the composite slab of
Committee for Standardisation 2003c), in the loft in the prototype building, it is
which case either the connections should be recommended to avoid the more complex
stiffened to become rigid or a rotational spring option of modeling it with shell elements and
should be inserted in the model to represent instead create a “master” node that will be
the partially restricted connection between the connected with all nodes of the diaphragm
members in question. The decision about the (Fig. 21). In this case, in order to account for
type of joint to be included in the model uncertainties in the location of masses and in
depends mainly on the lever arm between the spatial variation of the seismic motion, the
bolts in the connection. Connections of calculated center of mass at each floor shall be
I-beams where all bolts are between the considered as being displaced from its
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 17 Column to girder connections of prototype
building, modeled as rigid moment connections
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 18 Connections between or at the end of bracing
members, modeled as hinges
1 2 3 4 5 6
C
S
B
A
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 19 Roof’s plan view from numerical model
illustrating with dots the joint releases on roof’s bracing and on secondary beams
A A A A A A
1 2 3 4 5 6
STORY2
STORY1
BASE
X
Seismic Analysis of Steel Buildings: Numerical bases, on vertical bracing, and on secondary beams
Modeling, Fig. 20 Building’s side view from numerical (reinforced concrete pedestals on which steel columns
model illustrating with dots the joint releases on column are supported are also shown)

Modeling,
Fig. 21 Diaphragm
definition of loft slab
nominal location in each direction by an acci- and environmental loads such as those due to
dental eccentricity, commonly taken equal to snow, wind, and temperature, as well as seismic
e = 0.05∙L, where L is the corresponding span actions in case the building is located in a seismic
length. region. Characteristic nominal values of live,
snow, and wind loads and seismic actions are
prescribed by pertinent codes (e.g., European
Loads Committee for Standardisation 2001b, 2003a, b,
d, 2004a).
The next phase of the design process consists of For the prototype building presented here,
setting up appropriate loading situations includ- Excel spreadsheets for the calculation of
ing basic loads and load combinations. Basic snow, wind, and seismic loads are illustrated in
loads commonly include permanent loads, live Figs. 22, 23, 24, and 25. Issues that may require
loads depending on the building’s intended use, special attention include snow accumulation due
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 22 Excel spreadsheet for snow load calculation for
the prototype building according to Eurocode 1 – Part 1–3
to roof geometry, increased snow density in the values of behavior factor, for which recom-
regions of low temperatures, or unusual building mendations are provided in the pertinent seismic
shapes not covered by the code guidelines for codes, depending also on the type of structural
wind loads. Choices that differentiate the design system.
spectrum of steel structures from those of struc- In case part of the structure is buried (e.g., a
tures made of different materials are the values of basement), it is recommended to define accord-
damping, commonly taken equal to 2 % for ingly the level above which the structure vibrates
welded and 4 % for bolted steel structures, and independently of the ground.
Total building height h (m) = 7.90

Plan dimensions b (m)= 15.00 d=28.00m
d (m)= 28.00
e0 (m)= 15.00 e90 (m)= 15.80
θ=0°
FROM TO D(+)
HEIGHT
0.00 7.90 ze1 –1.240 1.667 –1.240
–1.801 1.106 –1.801
Zone we wi Total A(–)
3.16
A –1.801 –0.561 –1.240 (+)
B –1.200 –0.561 –0.640
11.84
B(–)
b=15.00m
C –0.750 –0.561 –0.189 –0.640 (+) (+) –0.640
– 0.561(wi)
D 1.106 –0.561 1.667 –1.200 –1.200
E –0.561 –0.561 0.000
0.00
–0.189 (+) –0.189
–0.750 C(–) –0.750
(+):Pressure directed towards the surface E(–)
(–):Suction pressure (directed away from surface) 0.000
–0.561
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 23 Excel spreadsheet for wind load calculation on
the side walls of the prototype building according to Eurocode 1 – Part 1–4
WIND DIRECTION O=0∞

Parapet curved corner gradient corner
Building height h (m)= 7.90 hp= 0.00
Parapet height hp (m)= 0.00 α
hp/h= 0.00 r
h= 7.90
Radius corner r (m)= 0.00 h
r/h= 0.00
Corner angle α (ο)= 0.00
ze=h
Building dimensions b (m)= 15.00 e=min(d,2h)
d (m)= 28.00
e0 (m)= 15.00 e90 (m)= 15.80 e/4= 3.95 e/4= 3.95
d= 28.00
qp(h)= 1.500 kN/m2
e/10= 1.58 F G F
b= 15.00
Position cpe we wi Total
H
F –1.300 –1.951 –0.561 –1.390
G –1.000 –1.500 –0.561 –0.940
e/10= 1.58 J
H –0.450 –0.675 –0.561 –0.114
I –0.500 –0.750 –0.561 –0.189
I
J –0.800 –1.200 –0.561 –0.640
(+):Pressure directed towards the surface

(–):Suction pressure (directed away from surface)
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 24 Excel spreadsheet for wind load calculation on
the roof of the prototype building according to Eurocode 1 – Part 1–4
Following load calculation, basic loads are for ordinary steel buildings are provided in
then applied on the model, as shown schemati- Table 1.
cally in Figs. 26, 27, 28, and 29. In Table 1:
The next step is to describe pertinent load
combinations in accordance with the relevant • Live load corresponds to all relevant types of
codes (e.g., European Committee for live load, such as service load or load due
Standardisation 2001a), including combinations to snow.
in the ultimate (ULS), serviceability (SLS), and • Wind load corresponds to four basic individ-
seismic limit states. Common load combinations ual load cases for each direction, including

Modeling,
Fig. 25 Horizontal
component of elastic
response spectrum of the
prototype building
according to Eurocode 8

Modeling,
Fig. 26 Application of
dead loads on the finite
element model of the
prototype building
uplift or downward pressure on the roof and A variety of linear and nonlinear (geometrically
positive or negative pressures on the walls. and/or material) and static or dynamic analysis
• Earthquake load corresponds to three basic indi- algorithms are commonly available in commer-
vidual load cases, one for each global direction. cial structural software (e.g., Gantes and
• Temperature load corresponds to two basic Fragkopoulos 2010). However, in the vast major-
individual loads (increase, decrease). ity of common steel buildings, linear static anal-
ysis is used in design practice for all loads and
Structural Analysis load combinations, with the exception of seismic
actions, for which frequently both equivalent
A decision that has to be taken next concerns the static as well as response spectrum dynamic anal-
type of structural analysis to be performed. ysis are carried out. The principle of

Modeling,
live loads on the finite
prototype building

Modeling,
wind loads on the finite
prototype building
superposition is routinely employed to obtain maximum displacements, forces, and stresses are
action effects for load combinations. computed throughout the structure for each vibra-
In the equivalent static method, first the base tion mode. These modal values for a given
shear is calculated, which is then distributed over response quantity (displacements, forces, or
the height of the building as specified by the per- stresses) are then combined appropriately to pro-
tinent seismic code. In the response spectrum anal- duce a single, positive result for the given direc-
ysis, for a given direction of acceleration, the tion of acceleration, using, for example, the SRSS
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 29 Application of temperature increase on the
finite element model of the prototype building
or the CQC method. It is noted that most modern used for design. Indicative results for the proto-
codes contain a requirement of modal participat- type building are presented in the following fig-
ing mass ratios at least equal to 90 %. In case this is ures, including mode shapes (Figs. 30 and 31),
not satisfied, the response spectrum analysis must seismic deformed shapes (Figs. 32 and 33), and
be repeated with higher number of participating internal force diagrams for individual load cases
modes. (Fig. 34) and load combination envelopes S
The analysis results include mode shapes and (Figs. 35, 36, 37, and 38).
natural periods of vibration as well as deformed
shapes and internal force diagrams for all indi-
vidual loads and load combinations. A qualitative Structural Design Checks
evaluation of analysis results is always highly
recommended in order to detect possible model- Following the qualitative evaluation of analysis
ing errors. For that purpose, it is a good practice results on the basis of mode shapes and
to start with mode shapes and natural periods of corresponding vibration periods, deformed
vibration, proceed with deformed shapes for indi- shapes and internal force diagrams, and provided
vidual load cases, continue with internal force that confidence is gained for the correctness of
diagrams for individual load cases, and finally, these results, design checks are performed,
conclude with internal force diagrams for load consisting of general checks, member checks,
combinations and envelopes, which are also and connection checks. Foundation checks are
Seismic Analysis of Steel Buildings: Numerical Modeling, Table 1 Common load combinations for ordinary steel
buildings in the ultimate (ULS), serviceability (SLS), and seismic limit states according to the Eurocodes
Load factors
gunfavorable; gfavourable
Load combination Dead Live Wind Temperature Earthquake
ULS COMB1 1.35 1.50
COMB2 1.35 1.50 0.90
COMB3 1.35 1.50 0.90
COMB4 1.35 1.50 0.90 0.90
COMB5 1.35;1.00 1.50
COMB6 1.35;1.00 0.90 1.50
COMB7 1.35;1.00 1.50 0.90
COMB8 1.35;1.00 0.90 1.50 0.90
COMB9 1.35 1.50
COMB10 1.35 0.90 1.50
COMB11 1.35 0.90 1.50
COMB12 1.35 0.90 0.90 1.50
SLS COMB13 1.00 1.00
COMB14 1.00 1.00 0.60
COMB15 1.00 1.00 0.60
COMB16 1.00 1.00 0.60 0.60
COMB17 1.00 1.00
COMB18 1.00 0.60 1.00
COMB19 1.00 1.00 0.60
COMB20 1.00 0.60 1.00 0.60
COMB21 1.00 1.00
COMB22 1.00 0.60 1.00
COMB23 1.00 0.60 1.00
COMB24 1.00 0.60 0.60 1.00
SEISMIC COMB25 1.00 0.30 1.00

Modeling,
Fig. 30 Dominant
vibration mode in
transverse direction of
prototype building

Modeling,
Fig. 31 Dominant
vibration mode in
longitudinal direction of
prototype building

Modeling,
Fig. 32 Deformed shape
of prototype building from
equivalent static
method – transverse
direction
also part of this process, which are presented in are acceptable for all load combinations. This
other parts of this encyclopedia. This is an itera- includes interstory drifts and overall building
tive process, and frequently, there is a need to drift for all load combinations with predominantly
return to the model, modify it, and run a new set horizontal components, such as wind and seismic
of analyses and checks, as schematically illus- combinations. Depending on the use of the build-
trated in Fig. 39. ing, general checks may also include restrictions
General checks (e.g., European Committee for about the vertical vibration frequencies, associated
Standardisation 2004b) consist mainly of with a sensation of unease of the users, as is the
confirming that overall structure deformations case in stadium grandstands. In case such checks

Modeling,
Fig. 33 Deformed shape
of prototype building from
equivalent static
method – longitudinal
direction

Modeling,
Fig. 34 Moment M3
diagrams of prototype
building from equivalent
static method – seismic
action in transverse
direction
are violated by far, a modification of the structural maximum deflections to allowable upper
system may be the only solution, while smaller bounds in the serviceability limit state. As
violations may be addressed by increasing mem- flexural and lateral-torsional buckling are in
ber cross sections. most cases critical for steel members, it is
Member checks (e.g., European Committee highly recommended that the engineer
for Standardisation 2004b) consist of compari- reviews and modifies as needed the buckling
son between actions and resistances in the lengths initially proposed by the software
ultimate limit state and comparison of for the corresponding checks. It is also

Modeling,
Fig. 35 Envelope of
moment M3 diagrams of
prototype building for ULS
combinations

Modeling,
Fig. 36 Envelope of
moment M2 diagrams of
combinations
noted that most seismic design codes require Connection checks (e.g., European Commit-
capacity design checks, leading, for exam- tee for Standardisation 2003c) include strength
ple, to the necessity of larger bracing sec- checks, to ensure safe transfer of internal
tions to satisfy minimum slenderness actions between members, as well as stiffness
requirements or larger column sections, to checks, so that the behavior of actual connec-
adhere to the weak beam-strong column tions (Fig. 40) is in accordance with the
design approach. hinged/semirigid/rigid assumptions adopted in

Modeling,
Fig. 37 Envelope of shear
V2 diagrams of prototype
building for ULS
combinations

Modeling,
Fig. 38 Envelope of axial
force N diagrams of
combinations
the numerical simulation. As there is so far Foundation checks depend on the type of
limited experience with the behavior of semi- foundation (mat foundation, foundation
rigid connections under cyclic loading, it is beams, spread footings, pile foundation, etc.)
recommended to avoid their use in seismic and consist of general stability checks
regions. Instead, it is proposed to stiffen the (overturning, sliding), soil bearing capacity
connection so that it can be classified as rigid. checks, comparison of absolute and differen-
Strength calculation and stiffness classification tial settlements to allowable values, and cal-
of steel connections are commonly performed culation of reinforcement for the reinforced
by means of dedicated software. concrete elements.
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 39 Flowchart of structural design process for steel
buildings
Seismic Analysis of Steel Buildings: Numerical Modeling, Fig. 40 (a) Column base, (b) girder to column, (c)
frame ridge connections of prototype building
Summary Conceptual design, numerical modeling, struc-

tural analysis, and design checks have
The state of the art in numerical modeling of been discussed, but emphasis has been directed
steel buildings has been presented, from the toward modeling, as the other design phases are
point of view of the practicing structural engineer covered in detail in other parts of this
designing such structures in seismic regions. encyclopedia.
2644 Seismic Analysis of Steel–Concrete Composite Buildings: Numerical Modeling
Cross-References European Committee for Standardisation (2003b)

Eurocode 1: actions on structures – part 1–5: general
actions – thermal actions
▶ Assessment of Existing Structures Using European Committee for Standardisation (2003c) Eurocode
Response History Analysis 3: design of steel structures, part 1–8: design of joints
▶ Behavior Factor and Ductility European Committee for Standardisation (2003d)
▶ Buckling-Restrained Braces and Their Imple- Eurocode 8: design of structures for earthquake
resistance – part 1: general rules, seismic actions and
mentation in Structural Design of Steel rules for buildings
Buildings European Committee for Standardisation (2004a) Eurocode
▶ Earthquake Response Spectra and Design 1: actions on structures – part 1–4: wind actions
Spectra European Committee for Standardisation (2004b)
Eurocode 3: design of steel structures, part 1–1: gen-
▶ Earthquake Return Period and Its Incorpora- eral rules and rules for buildings
tion into Seismic Actions Gantes CJ, Fragkopoulos KA (2010) Strategy for numer-
▶ Equivalent Static Analysis of Structures ical verification of steel structures at the ultimate limit
Subjected to Seismic Actions state. Struct Infrastruct Eng 6(1–2):225–255
▶ European Structural Design Codes: Seismic
Actions
▶ Mixed In-Height Concrete-Steel Buildings
Under Seismic Actions: Modeling and Seismic Analysis of Steel–Concrete
Analysis Composite Buildings: Numerical
▶ Modal Analysis Modeling
▶ Plastic Hinge and Plastic Zone Seismic Anal-
ysis of Frames Konstantinos Daniel Tsavdaridis
▶ Response Spectrum Analysis of Structures School of Civil Engineering, University of Leeds,
Subjected to Seismic Actions Leeds, UK
▶ Seismic Analysis of Masonry Buildings:
Numerical Modeling
▶ Seismic Analysis of Steel and Composite Brid- Synonyms
ges: Numerical Modeling
▶ Seismic Analysis of Steel–Concrete Compos- Concrete-encased steel sections; Connections;
ite Buildings: Numerical Modeling Constitutive models; Cyclic behavior; Damping;
▶ Soil-Structure Interaction Frames; Hysteretic rules; Nonlinear; Numerical
▶ Steel Structures modeling; Panel zones; Seismic analysis;
▶ Strengthening Techniques: Code-Deficient Steel–concrete composite; T-stub components
Steel Buildings
▶ Structural Design Codes of Australia and New
Zealand: Seismic Actions Introduction
▶ Time History Seismic Analysis
Steel–Concrete Composite (SCC) Systems
Composite construction includes a wide range of
References structural systems, e.g., framed structures
employing all steel–concrete composite (SCC)
European Committee for Standardisation (2001a)
Eurocode – basis of structural design
members and components (e.g., composite
European Committee for Standardisation (2001b) beam-to-columns and connections) and
Eurocode 1: actions on structures – part 1–1: general sub-assemblages of steel and/or reinforced con-
actions – densities, self-weight, imposed loads for crete (RC) elements. Such components and ele-
buildings
European Committee for Standardisation (2003a)
ments are employed to optimize the resistance
Eurocode 1: actions on structures – part 1–3: general and deformation capacity (Uchida and Tohki
actions – snow loads 1997). SCC structures have been used
Seismic Analysis of Steel–Concrete Composite Buildings: Numerical Modeling 2645
extensively in recent years because of benefits in constitutive models which emphasize on the
combining the two construction materials. SCC description of post-peak material characteristics
structures are also known for their excellent such as strain hardening and softening, tension stiff-
earthquake performance owing to their high ening, shear-retention ability, etc. (Cotsovos and
strength, high ductility, and large energy absorp- Kotsovos 2011). The derivation of such constitutive
tion. Their good structural damping properties models has been based on a variety of theories and
arising from the friction between the their combination. However, the application of FE
steel–concrete interfaces make them an even packages in practical structural analysis has shown
more attractive alternative for seismic resistance. that the constitutive relationships are case-study
Consequent effects of combining the two dependent, since the solutions obtained are realistic
materials are the enhanced lateral strength and only for particular problems. Therefore, the applica-
stiffness of the frame, with apparent effects of bility of packages to a different set of problems
the alteration of the structural natural period requires modifications of the constitutive relation-
of vibration and the complex local behavior of ships. This is entirely dependent on the interpreta-
beam-to-column connections. Furthermore, SCC tion of the observed material behavior as well as the
beams subjected to lateral loading show complex use of the experimental data to validate the consti-
behavior due to several factors, including the slip tutive relationships.
between the concrete slab and the steel beam, the To this end, the aim of the present chapter is to
variation of longitudinal stress across the width of provide an indication of the concepts which are
the slab, and the overall configuration of the widely used for modeling the steel–concrete
numerous different types of models, while the composite behavior and to develop numerical
steel and concrete parts can be subjected to dif- guidelines for the nonlinear analysis of such
ferent actions in every case. For the above rea- structures and their components, considering
sons, the calculation of the seismic response of the seismic actions under earthquake events. The
composite structures is not a straightforward task numerical analyses presented herein model the
due to the interaction of local and global effects behavior of SCC structures/components using
and hence the unexpected failure modes that macro-models (i.e., the use of line elements and
might incur. Consequently, it is very important spring connections) rather than micro-models
for the analysis of such structures to account for (continuum FE models) due to their simplicity
the local interactions (e.g., interface behavior and accuracy in nonlinear analysis. Different
between steel and concrete) as well as the local aspects of modeling including the geometry, mate-
behavior of structural systems (e.g., beam-to- rial nonlinearity (through the constitutive laws
column and base-to-column response). All these adopted), hysteretic behavior, and geometrical
factors make the analysis of SCC structures and nonlinearity as well as other parameters important
their individual components an intriguing but for seismic analysis are also presented in this S
challenging task. chapter.
Although experimental procedures can be
performed in order to enhance the understanding Chapter Synthesis
of the behavior of SCC structures under earth- The modeling of SCC elements and frames is
quake loading, they are typically expensive and based on three approaches. The first one is the
time-consuming and do not cover a broad range simplified modeling approach presented in this
of SCC structures and elements. As a result, chapter, engaging the use of springs and line
numerical modeling procedures have been devel- elements for the elementary simulation of the
oped and tested in order to facilitate the analysis behavior of each component to the entire frame
of such structures. assembly. The scope is to initiate numerical
Most finite element (FE) packages (e.g., guidelines based on the simplified approach and
ANSYS, ABAQUS, ADINA, DIANA, to present modeling examples of SCC beam
LS-DYNA, MIDAS FEA, etc.) rely on the use of cross sections, flooring systems, fully composite
Numerical modeling of SCC components & systems

based on simplified approaches
Modeling of typical SCC beams Modeling of

& beams with deformable shear frames using
connections beam elements
Modeling of partial Modeling of panel

strength beam-to- zone in semirigid
column connections SCC connections
Modeling of conventional SCC

members & frames comprising
concrete-filled steel sections
Seismic Analysis of Steel–Concrete Composite Buildings: Numerical Modeling, Chart 1 Structure of the
Chapter
members, beam-to-column connections, as well confinement should be carefully studied. On the

as holistic approaches modeling a frame. The intermediate level, the ductility of the member in
breakdown of this chapter is given in Chart 1. terms of rotation/displacement should be
Based on the bottom-up approach (i.e., com- established. Additionally, second-order effects
bining different structural components starting (P-D) on forces and deformations should be
from the most fundamental of a system and taken into account through the provisions for
giving rise to grander systems), the engineer large displacement analysis. Modeling the beam-
will gain a decent understanding on the parame- to-column connection is also essential when the
ters to be considered during the computational fully rigid assumption is not suitable. On the
modeling procedure. Modeling of these structural global level, the overall ductility and strength of
components will enable the computation of the structure should be established through
their response to different load histories and force–displacement relations. The progressive
moreover will enable the engineer to carry out yielding and the hinge formation at the structural
the state determination of a member from a frame frame should also be established through
assembly. moment–rotation relationships. The complete list
of requirements for the collapse analysis and the
Requirements for Collapse Analysis of modeling of SCC structures subjected to earth-
Composite Structures quake actions is presented below:
Numerical modeling procedures should aim to
address a number of issues regarding the local, • Stress–strain relationships for the steel mate-
intermediate, and global level of SCC structural rial including strain hardening and softening
design. On the local level, aspects such as the • Provision for the effects of local buckling in
cyclic behavior of the steel and the concrete mem- the steel section
bers (including the softening and hardening of the • Stress–strain relationships for the concrete
material), the local buckling of steel flanges, the material including cycling loading regimes
load carrying capacity, the curvature ductility of and the effect of the confinement on the peak
the components, as well as the effects of stress and corresponding strain
• Explicit representation of the slip boundary Model Geometry of a Typical SCC Beam
conditions of the shear connection both at To model the geometry of the macromodel for
local and global levels a typical SCC beam, the following assemblies
• Provision for second-order effects on forces should be utilized:
and deformations
• Effective beam-to-column connection 1. Four-node-layered shell elements representing
models, including panel distortion the concrete slab. Reinforcement layers com-
• Iterative and advanced dynamic analysis tech- prising steel material properties should be used
niques for analyzing the structural response to simulate the steel reinforcement located at
near collapse state the top and bottom of the concrete slab.
2. Two-node fiber beam-to-column elements for
modeling the steel beam. The reference sur-
Modeling of Steel–Concrete Composite face of the slab will be located at the centroid
(SCC) Beams of the steel beam cross section.
3. Dummy nodes at the same locations as the
A variety of different models have been devel- beam-to-column element nodes simulating
oped by researchers in order to capture the behav- the connection between the nodes of the steel
ior of SCC beams, based on either concentrated beam and the shell elements.
or distributed plasticity. In concentrated plastic- 4. Rigid beam elements connecting the dummy
ity models, all the inelasticity is concentrated at nodes and the corresponding ones of the shell
the ends of the member; therefore, it deals with elements, located on the same x- and z-
material nonlinearity in an approximate but effi- coordinates.
cient manner. On the contrary, distributed plas- 5. Discrete spring elements with only translation
ticity models simulate the inelastic behavior in the z-direction connecting both the
along the length of the member. This approach dummy nodes and the beam-to-column
is more accurate but at the same time is more element nodes in order to control the interface
computationally demanding. Most of the formu- shear–slip surface along the length of
lations for both approaches are rather complex the beam.
and not amenable to generic and routine applica-
tion in structural engineering design. Model Geometry of Beam with Deformable
The present subchapter presents a simplified Shear Connection
(new) modeling approach based on the work of Modeling of two-dimensional beams with
Zhao et al. (2012) for the nonlinear analysis of deformable shear connection is based on the
SCC beams and composite frames with deformable Newmark et al. (1951) model, in which (i) the
shear connections (based on the distributed plastic- Euler–Bernoulli beam theory applies to S
ity approach) using line elements to simulate the both components of the SCC beam and (ii) the
structural beam and column members, layered fiber deformable shear connection is represented
section to simulate the reinforced concrete slabs, by an interface model with distributed bond
and nonlinear spring elements for the simulation of allowing interlayer slip as well as enforcing
the interface between the structural steel beams and contact between the steel and concrete
the reinforced concrete slab. Vertical interactions components.
between the slab and steel beams are not expected A local coordinate system should be
to be significant, therefore are not accounted into established to enhance the understanding of kine-
the analysis. The geometry of the model, along matics of Newmark’s model. With reference to
with a simple set of details, is outlined below. Fig. 2, Z axis is parallel to the beam axis and the
The assembled model is shown in Fig. 1. vertical plane YZ is the plane of geometrical and
Seismic Analysis of Steel–Concrete Composite Buildings: Numerical Modeling, Fig. 1 Assembled macromodel
representing a typical steel-concrete composite beam
material symmetry of the cross section. Loads are eza ðy, zÞ ¼ w0a ðzÞ
also assumed symmetric with respect to the YZ þ ðya yÞv00 ðzÞ on Aa ða ¼ 1, 2Þ
plane. The displacement field u of a material (2)
point of the beam is given by:
0
dðzÞ ¼ w2 ðzÞ w1 ðzÞ þ hv ðzÞ (3)
uðy, zÞ ¼ vðzÞj
h i where h = y2 y1 is the distance between the
0
þ wa þ ðya yÞv ðzÞ k on Aa ða ¼ 1, 2Þ reference points (G1 and G2 in Fig. 2) of the two
(1) components. At the locations of the longitudinal
reinforcement, Eq. 2 also provides the strain in
the reinforcement, due to the assumption of per-
where wa is the axial displacement of the refer- fect bond between the steel and the concrete.
ence point of domain Aa, the ordinate of which is
ya(a = 1 : concrete slab, a = 2 : steel beam); v is FE Formulations
the vertical displacement of the cross section; A simple two-dimensional 10 degree-of-freedom
and j and k denote the unit vectors along the (DOF) SCC frame element with deformable
Y and Z axes, respectively. It is observed that shear connection is presented herein, similarly
the transverse displacements and rotations of to Zona et al. (2008). With reference to Fig. 3,
the slab and of the steel beam are equal due 8 of the 10DOFs are external (4 DOFs per end
to the enforced contact between the two compo- node) allowing for the axial displacement, the
nents. The only nonzero strain components are transverse displacement, and the rotation of the
the axial strain eza and the interface slip d: steel beam and 1DOF for the axial displacement
k
Z
w1k
j
w2k
v j
Y
v'
i
x
y1 vj
G1
v'
h y2
δ
G2
Seismic Analysis of Steel–Concrete Composite Buildings: Numerical Modeling, Fig. 2 Kinematics of

two-dimensional composite beam model and reference system
Seismic Analysis of
Steel–Concrete a
Composite Buildings:
Numerical Modeling,
Fig. 3 10DOF SCC beam
element used (a) references
defined at the beam and
slab centroids and (b)
references defined at the
slab–beam interface S
b
of the concrete slab. The remaining internal elastic and plastic strength, and strain hardening)
2DOFs allow for axial displacement of the steel can be obtained from the uniaxial stress–strain
beam and the concrete slab (Fig. 3b). curves derived from coupon tests and then
applied to the corresponding fibers across the
Modeling of Inertia and Damping Properties composite cross section. In order to accurately
Modeling the inertia properties of the frame ele- simulate the behavior of SCC beams under
ments can be achieved using lumped masses at earthquake conditions, robust material models
the DOFs of the external nodes. Consequently, capable of simulating the material nonlinearity
the inertia properties of the FE model are inde- as well as other damaging effects under
pendent of the type of finite elements employed dynamic or cyclic loading (i.e., softening/harden-
(i.e., the structure’s mass matrix can be obtained ing) need to be employed. Several models have
using force, displacement, or mixed-based for- been developed to achieve the aforementioned
mulation frame elements). scope, some of which are presented in the follow-
Even though the friction between steel beams ing sections.
and concrete slabs in SCC frames may be a strong
source of structural damping, quantitative infor- Constitutive Law for Concrete Parts (Based on the
mation about this energy dissipating mechanism Kent–Park Model)
usually referred to as structural damping is lim- The proposed constitutive law modeling the con-
ited owing to the partial availability of experi- crete in monotonic compression for the cases of
mental dynamic data. Consequently, the well- confined and unconfined concrete is the
known and widely used Rayleigh damping Kent–Park model as described in Park and Paulay
model can be used by the practicing engineer. In (1975). As it is shown in Fig. 4, the material
this model, the damping matrix can be obtained follows a parabolic stress–strain curve up to
using the classical Rayleigh damping relationship a maximum stress equal to the cylinder’s
(Eq. 4), where the damping matrix is proportional strength, after which it decays linearly with strain
to the mass matrix and the initial stiffness matrix: until the residual strength is reached. In tension,
the model assumes a linear stress–strain behavior
½C ¼ m½M þ l½K (4) up until the tensile limit of the material is
reached, and then the stiffness and strength
where decays with increasing strain (Fig. 5).
The cyclic behavior of the concrete can be
m = mass proportional Rayleigh damping described by the Blakeley–Park model also
coefficient presented in Park and Paulay (1975). The
l = stiffness proportional Rayleigh damping stress–strain response lies within the Kent–Park
coefficient envelope; however, the effect of concrete con-
M = system structural mass matrix finement is not taken into account. The model
K = system structural stiffness matrix assumes that unloading and reloading takes
place along a line without energy dissipation or
Note: The proposed model presented in the stiffness deterioration for strains smaller or
above sections considers only rigid beam-to- equal to the strain corresponding to peak stress
column connections. Nevertheless, semirigid (e < ec). Beyond this point, the stiffness deterio-
connections can be considered in the same ration is taken into account through the introduc-
numerical procedure by introducing special joint tion of reduction factors, given by Blakeley and
elements with prescribed constitutive behavior. Park. Along the first unloading branch, the stress
is reduced approximately 50 % without any
Constitutive Stress–Strain Relationships reduction in strain. The reloading branch with
For modeling purposes, the material properties slope equal to fcE extends back to the envelope
(such as the Young’s modulus, Poisson’s ratio, (Fig. 5).
fc
kfc'
Confined
fc'
Unconfined
0.2fc'
εo Kεo ε20C ε
Seismic Analysis of Steel–Concrete Composite Buildings: Numerical Modeling, Fig. 4 Monotonic stress–strain
Kent and Park model for unconfined and confined concrete
Seismic Analysis of Stress σ

Steel–Concrete
Composite Buildings: σc
Numerical Modeling,
Fig. 5 Cyclic stress–strain
Blakeley–Park model for Compressive
concrete
σcu
Strain ε
0 εc εcu
Tensile
Constitutive Law for Concrete (Based on the Constitutive Law for Steel
Popovics–Saenz Law) Figure 7a, b describes the elastoplastic response
The constitutive law for concrete is a uniaxial of the steel under monotonic and cyclic loading
cyclic law with monotonic envelope defined by respectively. For monotonic loading, the charac-
the Popovics–Saenz law (Balan et al. 1997). teristic yield plateau in the stress–strain model S
Linear unloading–reloading branches with pro- followed by a region of increased strength
gressively degrading stiffness characterize the owing to strain hardening of the material. The
cyclic behavior of the material. The response of unloading from the yielded condition is elastic;
concrete under cyclic loading is shown in thereafter, the Bauschinger effect can be
Fig. 6. According to the same figure, after represented by a Ramberg–Osgood relationship
each unloading–reloading, the monotonic (Eq. 5) until the yield stress is reached. This
envelope is reached again when the absolute model uses a single nonlinear equation to char-
value of the largest compressive strain attained acterize the observed curvilinear response of steel
so far is surpassed. The tensile behavior subjected to monotonic loading:
of concrete is characterized by the same
loading–unloading–reloading rules with the s Z
same initial stiffness and appropriate values s
e¼ þ K (5)
for the other parameters. Ε Ε
Seismic Analysis of Strain

Steel–Concrete −0.006 −0.005 −0.004 −0.003 −0.002 −0.001 0.000
Composite Buildings: 0
D B O
Numerical Modeling,
Fig. 6 Hysteretic concrete H −5
material model under I
compression
−10
Stress (MPa)
G −15
F −20
−25
−25
E
−30
A
C
−35
a b
Stress σ
Stress
σult
Yield
σy
Stabilized cycles
Strain
Strain ε Cyclic stress-

Yield
εsh εult strain curve
Seismic Analysis of Steel–Concrete Composite Buildings: Numerical Modeling, Fig. 7 (a) Stress–strain model
under monotonic loading. (b) Steady-state cyclic response of mild steel
where constant strain amplitude cycles is described by

strain hardening for large amplitudes and strain
s
Ε= equal to the elastic part of the strain softening for small amplitudes. For the accurate
sZ
K Ε = accounts for the plastic part of the strain simulation of the steel material response under an
K and Z = parameters that describe the hardening arbitrary load, the constitutive model needs to
behavior of the material account for all the monotonic response, the
steady-state cyclic behavior, as well as the tran-
When the steel material is subjected to con- sient behavior involving softening and hardening.
stant strain amplitude under cyclic loading, it This can be achieved using an efficient simplistic
exhibits a response that converges to computationally bilinear model.
a stabilized saturation loop which depends only
on the cycling amplitude (Fig. 7b). As it is shown Bilinear Stress–Strain Steel Model In this
in Fig. 8, the response of the steel material under bilinear model, the elastic range remains
Seismic Analysis of Stress

Steel–Concrete
Numerical Modeling, Yield
Fig. 8 Cyclic response of
steel under constant strain Stabilized cycle
amplitude cycles
Strain
Yield
Softening under small strain amplitudes
Stress
Yield
Strain
Stabilized cycle
Yield
Hardening under large strain amplitudes
Seismic Analysis of σ
Steel–Concrete
Numerical Modeling, μE
Fig. 9 Loading and
unloading paths of bilinear S
kinematic model
E 2 σy
2 σy ε
μE
constant throughout the various loading (Fig. 9). The calculation of the current stress
stages. The kinematic hardening rule for the state is expressed mathematically using
yield surface is assumed to be a linear func- Eqs. 6, 7, and 8, and it is presented graphi-
tion of the increment of plastic strain cally in Fig. 10:
σ σ
(ε,σ)
σy
(εo,σo) σy (εo,σo)
α
αo αo
ε α ε
σy
σy
(ε,σ)
Positive strain increment Negative strain increment
Seismic Analysis of Steel–Concrete Composite Buildings: Numerical Modeling, Fig. 10 Stress determination
with the bilinear kinematic model

a ¼ ao Constitutive Law for Steel (Based on the
s ¼ so þ Εðe eo Þ Menegotto–Pinto Model)
o (6) The constitutive law describing the behavior of
a sy so a þ sy so
If Ε ð e e oÞ
Ε the steel material is the uniaxial Menegotto–Pinto

9 model (1973). This computationally efficient
a o þ s y so = nonlinear law is capable to model both kinematic
a ¼ a þ mΕ e e
o o
Ε ; and isotropic hardening as well as the
s ¼ a þ sy Bauschinger effect, allowing for accurate simu-
o
a þsy so
If ðe eo Þ > Ε
lation and reproduction of experimental results.
The response of the steel material is defined by
(7) the following nonlinear equation:

) ð1 bÞe
a o s y so s¼ beþ
a ¼ a þ mΕ e e
o o 1 (9)
Ε ð 1 þ eR Þ R
s ¼ a sy
o
a sy so
If ðe eo Þ < Ε
where the effective strain and stress (e, s) are
a function of the unload–reload interval, b is the
(8)
ratio of the initial to final tangent stiffness, and
R defines the shape of the unloading–reloading
where according to Figs. 9 and 10: curves. Figure 11 presents a typical stress–strain
response based on the Menegotto–Pinto model.
Ε = Young’s modulus The model assumes a symmetric response for
m = strain hardening parameter loading in compression and tension.
sy = initial yield surface
e = current strain Interaction of Material Surfaces: Evaluation of
s = current stress Spring Properties
a = current center of elastic range The degree of composite action and interaction
Subscript “0” denotes values at the start of an between the steel beam and the concrete slab is
increment a fundamental mechanism that needs to be
Seismic Analysis of
Steel–Concrete
Numerical Modeling,
Fig. 11 Menegotto–Pinto
material constitutive model
for structural steel; typical
cyclic stress–strain
response
considered by the engineer during the modeling Nv

¼ ð1 ens Þm (10)
procedure of SCC beams and structural systems Nvu
owing to the implications on the serviceability
and ultimate limit states, the energy dissipation where
under cyclic loading, and the local stress
distribution. Nvu = connection (ultimate) strength
Two different modeling approaches can be Nv = shear load
used for the description of partial bond in SCC s = slip between two components of the compos-
structural systems. The concentrated bond ite beam (interface slip)
approach is based on the use of concentrated n, m = empirical parameters defining the shape of
springs for the modeling of the connection. The the curve calibrated from experimental data
springs are attached at the location of each con-
nector, modeling either the action of the shear In Fig. 12b, the monotonic envelope is
stud connectors between the steel and the con- presented by the definition of an ultimate slip,
crete slab or the friction in concrete-filled hollow Sult. When ultimate slip, the shear force–slip
sections and partially encased steel sections. The behavior follows zero stiffness with constant
second approach is based on the distributed bond shear force Nv = tfr, where tfr is the residual
model, which assumes a continuous bond stress shear force.
and bond slip along the contact surface. For both S
approaches, the uplift is typically neglected;
therefore, it is considered that the concrete slab Modeling of Steel–Concrete Composite
and the steel beam have the same vertical dis- Beam-To-Column Partial-Strength
placement and curvature. Semirigid Connections
Shear–slip relationships are widely available
providing information regarding the behavior of Compared to traditional bare steel structures,
the connectors. Figure 12a presents a simplified SCC frames can achieve more effective beam-
bilinear shear–slip relationship based on a widely to-column connections through the contribution
used shear–slip model proposed by Ollgaard of the concrete slab in resisting bending moments
et al. (1971). The Ollgaard model is described under gravitational and lateral loads. Addition-
by the following exponential function (Eq. 10) ally, these structures comprising partial-strength
representing an experimentally observed large partially restrained beam-to-column joints
reduction of stiffness with increasing slip: designed in such a way to exhibit ductile seismic
Seismic Analysis of a
Steel–Concrete 1.0
Numerical Modeling,
Fig. 12 Shear–slip 0.8
relationships
0.6
Nv/Nvu
Ollgaard
0.4
Johnson
0.2 Aribert
Bilinear model
0.0
0.0 0.5 1.5
1.0 2.0 2.5 3.0
s (mm)
Equivalent bi-linear model for shear-slip of each connector
b Force
Fu
Fy
fr
Sy Su Slip
fr
Fy
Fu
Cyclic response with slip
response through the plastic deformation of their nonlinear spring elements as shown in
components can achieve the formation of Fig. 13. With reference to the same figure,
a desirable beam hinging global frame mecha- the elements comprising the model should
nism, with large hysteretic energy dissipation account for the response of the:
capacity and reduced force demand on the
columns. 1. Concrete in compression
This section of the chapter presents 2. Column web panel in shear
a simplified approach based on the work 3. Upper T-stub in compression (+Ve moment)
done by Braconi et al. (2007) for the 4. Lower T-stub in tension (+Ve moment)
nonlinear analysis of a partial-strength beam- 5. Concrete slab in tension
to-column connection using a component 6. The shear studs (+Ve moment)
model. The behavior of partial-strength 7. Upper T-stub in tension (Ve moment)
beam-to-column connections under the appli- 8. Upper T-stub in compression (Ve moment)
cation of seismic load is described using 9. Shear studs (Ve moment)
a 5 b Unconfined Concrete
u4
u3
1 9 hCLS,nc
6 8 hCLS
θ2 θ1 hCLS,nc
γ
3 γ
2 α
u3
Confined Concrete
7 u2
u1 4 γ
c d u3 γ
concrete in tension 6 1
θ1 θ2 hSR hCLS 3 2
θ1
u3 u4 γ
reinforcing bars u1 4
Seismic Analysis of Steel–Concrete Composite concrete slab in compression (component 1), (c) kinemat-
Buildings: Numerical Modeling, Fig. 13 Component ics of the concrete slab in tension (component 5), and (d)
model: (a) overall joint model, (b) kinematics of the overall exterior joint model
Kinematics comprising the model are linked to the seven

Equilibrium must be maintained between the degrees of freedom through the following
force acting in the nine components and the inter- equations:
nal and external forces. The response of the
assembled model is therefore defined by a set of Dnc
eight equations related to translational equilib- 1, i ¼ u3
dnc ð2i 1Þ tan ðy1 Þ
2

rium between components in the same beam-to- Dnc
column connection, equilibrium between the þ hcs ð2i 1Þ tan ðgÞ (11)
2
shear studs and the beam steel profile, rotational
equilibrium between the internal forces in the

beam-to-column connection and the bending Dc
moment of the beam framing in it, as well as dc1, i ¼ u3 ð2i 1Þ þ hcls, nc tan ðy1 Þ
2
rotational and translation equilibrium acting on
Dc
the column web panel. The set of eight equations þ hcls ð2i 1Þ tan ðgÞ
can be then solved using a numerical procedure 2
S
(i.e., Newton–Raphson) considering the storey (12)
deformation (drift, d) as such an external action tbf
in the format of imposed deformations. d3 ¼ u1 hcs þ tan ðy1 Þ (13)
On the basis of the small displacement theory, 2
the local kinematics could be described using tbf
a total of 7DOFs: the horizontal displacement of d4 ¼ u1 hcs þ hb tan ðy1 Þ (14)
2
the bottom surface of the slab on both sides of the

column, u1 and u2; the horizontal displacement of Dct
the top surface of the slab on both sides of the d5, i ¼ u4 hcs ð2i 1Þ tan ðy2 Þ
2
column, u3 and u4; the relative rotation of the two
Dct
beams with respect to the column faces, y1 and u1 ð2i 1Þ tan ðy1 Þ
y2; and the column panel zone shear distortion, g. 2
The deformations, d, of the nine components (15)
Seismic Analysis of δ
Steel–Concrete
Numerical Modeling,
Fig. 14 Global kinematics Elc
of the sub-assemblage joint
h1
specimen
ϕ
Elb,sagg γ +θ2+ϕ
hsp htot
γ +θ1+ϕ ϕ Elb,hogg
h2
Elc
L L
tbf
d5, SR ¼ u4 hcs þ hb hSR tan ðy2 Þ thickness of each confined and unconfined con-
2 crete layers, respectively. Similarly, Dct is the
u1 hSR tan ðy1 Þ thickness of the concrete layers in tension.
(16) Figure 14 presents the deformation of the entire
sub-assemblage. The storey drift, d, is obtained
d6 ¼ u1 u3 (17) through Eq. 21:
d7 ¼ u2 (18) d ¼ ghSP þ jhtot þ del, column (21)
d8 ¼ u2 ðhb tbf Þ tan ðy2 Þ (19) where the first term represents the storey drift owing
to the web panel over the height of the joint, hSP, the
d9 ¼ u2 u4 (20) second term corresponds to the rotation due to
flexural deformations of the beam at beam
In the above equations, hb is the beam depth mid-depth, and the final term corresponds to the
and hcs is the overall slab thickness including the elastic flexural deformation of the column.
depth of the steel deck, but the model assumes Figure 14 shows the global kinematics of the model.
that the interaction between the slab and the col-
umn only occurs over the slab thickness above the Component Modeling
steel deck, hcls, to represent actual test conditions. Necessary work preceding the assembling of
The parameter, tbf, is the beam flange thickness. a SCC connection model is the verification of
For components 1 and 5, the deformation varies each individual component model. In
among the series of n parallel springs used over a semirigid SCC connection, the major compo-
the slab thickness. For the concrete in compres- nents to be considered are outlined below.
sion (Component 1), a distinction is also made
between unconfined (dnc c
1,i) and confined (d1,i) con- Concrete Slab
crete fibers. Unconfined concrete condition is The nonuniform stress distribution over the slab
assumed above the slab reinforcing steel thickness is taken into account in the model with
(of thickness hcls,nc in Fig. 13b), and therefore, the use of fibers as shown in Fig. 13b, c for the
the parameters Dc and Dnc correspond to the cases of compression and tension, respectively.
a 60 b 5.0
Confined Concrete
50 4.0
Stress ( MPa)
Stress ( MPa)
40
3.0
30
2.0
20
Unconfined Concrete
1.0
10 ε PL
ct Unloading-Reloading Path
1000
0 0.0
0 0.002 0.004 0.006 0.008 0 0.0005 0.001 0.0015 0.002 0.0025 0.003
Deformation Deformation
c 700 d 1000
600
Total Shear Force (kN)

800
Stress ( MPa)
500
400 600
300 400
200
Unloading-Reloading Path 200
100
0 0
0 0.02 0.04 0.06 0.08 0 2 4 6 8
Deformation Relative Displacement (mm)
e 1200 f 800
Interior Point 700
Shear Forse -VSP (kN)
1000
600
Force -F (kN)
800 500
(4)
600 400
Exterior Point Experiment Model
300
400
Experiment Model 200
200 100
0 0
0 10 20 30 40 50 60 0 2 4 6 8
Angular deformation - γ (mrad) Displacement - δ (mm)
Seismic Analysis of Steel–Concrete Composite (b) concrete in tension, (c) reinforcing steel, (d) shear
Buildings: Numerical Modeling, Fig. 15 Component stud, (e) panel zone in shear, and (f) lower T-stub in
constitutive relationships: (a) concrete in compression, tension
S
Concrete slab in compression can be modeled conditions. In tension (Fig. 15b), the behavior of
using stress–strain relationships proposed in concrete can be represented using a linear
EC2 or any of the stress–strain relationships response until cracking, followed by a softening
presented in sections “Constitutive Law for Con- branch where the tensile resistance reduces expo-
crete Parts (Based on the Kent–Park Model)” and nentially as proposed by Stevens et al. (1991).
“Constitutive Law for Concrete (Based on the Linear unloading–reloading branches can be
Popovics–Saenz Law)” for both unconfined adopted intersecting the deformation axis at
(concrete above reinforcement level) and con- a residual plastic deformation epl ct given by
fined (concrete below reinforcement level) con- Eq. 22 and also shown in Fig. 15b:
ditions. A typical example of the stress–strain
relationship that can be used in analysis is max 2
presented in Fig. 15a for both concrete ct ¼ 146 ect
ePL þ 0:523 emax
ct (22)
Steel Reinforcement GS ðhc tcf Þtcw

Bilinear stress–strain relationships as shown on Ks, wp ¼ (25)
aH z
Fig. 15c can be adopted for representing the
behavior of steel reinforcement.
where
Shear Connectors
GS = shear modulus of steel
The slip between the concrete slab and the beam
Avc = shear area of the column section
owing to the flexibility of the shear stud connec-
z = centerline vertical distance between the col-
tors can be modeled using force–deformation
umn stiffeners
relationships similar to those presented earlier in
bcf = column flange width
section “Interaction of Material Surfaces: Evalu-
tcf = column flange thickness
ation of Spring Properties” (“simulation of
hc = column depth
composite action”). The recommended
tcw = column web thickness
force–deformation relationship for this particular
aH = hardening coefficient related to the thick-
model is based on the model proposed by Aribert
ness of the column web panel and the column
and Lachal (2000) (and is tailored to the guide-
flanges
lines of EC4 and EC8 for the calculation of the
ultimate shear stud resistance), and it is presented
Using the elastic and post-elastic stiffness,
in Fig. 15d and Eq. 23.
shear forces in the panel zone can be obtained.
C2 The elastic stiffness, Kel,wp, is obtained from EC3
F ¼ Qu 1 eC1 ðu3 u4 Þ (23) and is applicable until the shear force, Vwp,
reaches the yield limit, Vel,w, which is also spec-
where ified in EC3 (Eq. 26). The shear forces corre-
spond to the post-elastic branches, described by
Qu = ultimate shear stud resistance calculated Krawinkler’s model (Krawinkler 1978), in which
according to EC4 and EC8 the shear force is obtained through Eqs. 27 and
C1, C2 =coefficients suggested by Aribert and Al 28. The three different branches of the shear
Bitar (1989) depending on the height of studs force–deformation relationship are presented in
and type of steel profile Fig. 15e (dashed line):
Panel Zone in Shear

0:9 f y, cw Avc
The behavior of the panel zone plays a significant Vel, wp ¼ pffiffiffi (26)
role in determining the overall stiffness and 3
capacity of the frame. In terms of seismic design,
the panel zone can have a significant influence on 0:9 f y, cw Avc bcf t2cf
the distribution of plasticity and energy dissipa- Vt, wp ¼ pffiffiffi þ1:04 GS 3 gy
3 z
tion on the overall performance of the structure.
(27)
A multi-linear shear force to shear deformation
response is retained for the web panel zone. The
following equations describe this multi-linear f u, cw Avc f u, cf bcf t2cf
relationship through the gradients of elastic, Vs, wp ¼ pffiffiffi þ (28)
3 4z
post-elastic, and strain hardening stiffness:
Avc where
Kel, wp ¼ GS (23)
z
gy = panel distortion at yield = Vel,wp/Kel,wp
bcf t2cf fu,cw, fu,cf = ultimate tensile stress for column
Kt, wp ¼ 1:04 GS (24)
z web and flanges
Seismic Analysis of Steel–Concrete Composite Buildings: Numerical Modeling, Fig. 16 Cross-sectional model
under negative bending
T-Stubs Components column. The modeling engineer can use

The geometry of the equivalent T-stub compo- more advanced modeling procedures (i.e., as
nents can be determined using the effective width proposed by Fabbrocino et al. 2002) when
concept presented in EC3 for stiffened columns continuous SCC beams are considered for the
and end-plates. The force–deformation relation- assessment of the behavior of the connection
ship of T-stub elements required for component capacity, in terms of global quantities such as
modeling can be obtained from the model pro- the rotations and deflections as well as local
posed by Piluso et al. (2001) (dashed line) quantities such as the slip and the curvature of
(Fig. 15f). members, the interaction forces, and the rebar
strain.
Simplifications and Assumptions 4. Column web buckling and beam flange buck-
1. The difference in column web shear stiffness ling under compression are not considered in
between the concrete-encased and such modeling procedure.
non-encased segments of the columns should
be taken into account in modeling through the Fabbrocino et al. 2002
assumption of infinite stiffness of the upper 1. Modeling the cross section of composite
encased portion of the column and a flexible beams is achieved through a modification of
diagonal spring in the bare steel column web the well-known Newmark’s kinematic model,
over the beam depth. as shown in Fig. 16. This approach requires S
2. Bolt pretension effects need to be considered the definition of the slab effective width
in modeling through modification of the stiff- depending on the type of loading (hogging or
ness of the equivalent T-stub springs in the sagging) and on the connection detailing at the
elastic range. beam end. A linear strain pattern is then
3. There is an interaction between the connec- applied to each component of the cross sec-
tions on both column sides due to the continu- tion. Under the assumption that the curvature
ity of the slab and the slab longitudinal steel and the rotation for each of the components are
reinforcement between the two beams. Hence, the same (e.g., for both the concrete slab and
including the anchorage steel bars, the nega- the steel profile), the uplift is neglected. Using
tive moment capacity of one side is dependent analytical procedures, the tensile stresses
on the concrete capacity of the opposite side developed in the concrete slab; the slip
while essentially transferring the tensile forces between the different components of the
acting from the reinforcing steel to the cross section; the interaction force, F; the
global bending moments in the steel profile, particular model proposed by Castro et al. (2005)
Ms; and the concrete slab, Mc, as well as the is based on a realistic stress distribution at the
moment–curvature relationship of the cross edges of the panel, aiming to account for the loca-
section can be obtained. tion of the neutral axis. This methodology enables
2. Modeling the continuous composite beam is assessment of the shear stress distribution through
based on a combination of the main behavioral the panel depth representing the distribution of the
aspects on the different regions of the beam plasticity in the vicinity of that region. Both shear
(i.e., Newmark’s model is used for sagging and bending deformations are considered in the
moments, whereas its modified version is elastic and post-elastic stages. The additional resis-
used when cracked zones of the beam are tance of the panel zone owing to the contribution
considered). The moment–curvature relation- of the column flanges is also taken into account
ship in each section of the beam can be then by considering the column depth and flange
defined through an iterative process. Once the thickness (Fig. 17).
generalized moment–curvature relationship is
established, rotations and displacements can Procedures and Details
be obtained by integration of the curvature With reference to Fig. 18, the physical dimen-
distribution. The numerical procedure for the sions of the panel zone, dc and db, are taken into
solution of a simple structural system of account. The part of the column being in contact
a beam characterized by geometrical and with the slab is modeled through an assemblage
mechanical symmetry is based on the compat- of links on top of the panel. This modeling
ibility method; therefore, the support bending approach is essentially a determination of the
moment is the main unknown and the beam is spring properties of both the “panel zone” and
statically determined. The reader is referred to the “top panel” which are derived analytically
Fabbrocino et al. (2002) for a step-by-step allowing for implementation in frame analysis
guidance for the solution of the composite software. The procedure establishes an analogy
section and beam. between the analytical (consisting of the actual
connection) and the corresponding numerical
model for frame analysis, as shown in Fig. 18.
Modeling of the Panel Zone in Semirigid The location of neutral axis is calculated con-
Steel–Concrete Composite (SCC) sidering a linear stress distribution, based on the
Connections assumption that the SCC beam behaves elasti-
cally until the panel yields.
When moment-resisting frames are subjected to An assessment of the effective width of the
horizontal loading such as earthquake excitation, slab in the vicinity of the connection is also
unbalanced moments occur at the beam-to- required. For positive moment (sagging), it is
column connections resulting to shear deforma- assumed that the slab is limited to the contact
tions in the panel zones of the columns. There- width of the column flange width, bc. For nega-
fore, the behavior of the panel zone plays tive moment (hogging), the slab is not consid-
a significant role in determining the overall stiff- ered, under the assumption that the reinforcement
ness and capacity of the frame. In addition to this, is not anchored to the column.
in terms of seismic design, the panel zone can
have a significant influence on the distribution of Notion of Calculations
plasticity and energy dissipation mechanisms and The main aspect of the proposed model is the
in turn significant effects on the overall perfor- determination of the spring properties for both
mance of the structure. the panel zone and the top panel. For a given
This section deals with the modeling of the moment carried by the SCC beam,
panel zone region within the beam-to-column a corresponding equivalent shear is applied to
joints of SCC moment-resisting frames. This the panel in the numerical model. The stiffness
tslab
Ktop
ds
K
db
dc
Seismic Analysis of Steel–Concrete Composite Buildings: Numerical Modeling, Fig. 17 Numerical and analyt-
ical representation of joint models
a Vcol b Vcol
tslab −
Nc
ds
x Mc ~= 0
B −
x
y+
db M
yG db NS
+
MS
A
x α
σ
Vcol Vcol
dc S
dc
Seismic Analysis of Steel–Concrete Composite Buildings: Numerical Modeling, Fig. 18 Analytical and numer-
ical representation of joint models
that should be used in the numerical model can be for both elastic and post-elastic ranges is outlined
then determined from the distortion caused to the below.
analytical model owing to the application of This section serves as a numerical guideline;
moment. Furthermore, the difference between the reader is referred to J. M Castro et al. (2005)
load level corresponding to first and full yielding to exploit the complete set of equations.
of the panel can be derived given that the shear
stress distribution through the panel is known. Elastic Range
The procedure needed to be followed by the engi- 1. The neutral axis location of the steel beam and
neer for derivation of these important parameters in turn its second moment of area can be
Seismic Analysis of 1 db/2

Steel–Concrete
Composite Buildings: B
Numerical Modeling,
Fig. 19 The virtual system
used to find panel db
deformations
A 1 1 db/2
V M
calculated under the assumption that the steel to be used in the numerical model is
beam behaves in an elastic manner up to yield- calculated by
ing of the panel zone.
2. The ratio of axial force (Ns) and bending Veq
Kel ¼ (31)
moment (Ms) carried by the beam can be jDshear j þ Dbending
then obtained using analytical procedures.
3. The total moment (M) acting to the connection 8. Using the calculated elastic stiffness, the rela-
can be then calculated in the numerical model tive drift of the panel (Dy,el) at the onset of
using Eq. 29. The moment (Mc) developed in yielding can be obtained analytically.
the slab is considered to have insignificant 9. Finally, using the calculated elastic stiffness
influence and therefore it is ignored: and the relative drift of the panel zone, the
elastic stiffness and the relative drift of
db tslab the spring can be obtained using the Eqs. 32
M ¼ Ms þ Ns þ ds þ (29)
2 2 and 33:
4. The equivalent moment carried by the panel Kel

can be then obtained from Eq. 30: Kel, spring ¼ (32)
cos2 a
Ms Ns Dy, el
Veq ¼ þ Vcol (30) Del-y, spring ¼ (33)
db 2 cos a
where Vcol ¼ M hs and hs is the storey height. where a is the angle of the spring as indicated
5. From the analytical model, knowing the nor- in Fig. 18.
mal stress distribution of the composite sec-
tion, the shear force and bending moment Post-Elastic Range
distributions can be obtained. 1. Beyond the yielding of the panel, the shear
6. Using the principle of virtual work (Fig. 19) stiffness of the column web is assumed to
and the calculated shear force and bending drop to the strain hardening stiffness of the
moment distributions, the relative horizontal material. Consequently, the post-elastic stiff-
displacement can be obtained. Having the ness of the panel is provided by the strain
application of opposite unit forces in the vir- hardening of the column web by the flanges
tual system, the internal virtual forces (Vint, and a portion of the column web delimiting the
Mint) can also be obtained. panel zone, as shown in Fig. 20. Following the
7. Using the calculated equivalent shear force same assumption for the beam remaining
(Veq) and the relative horizontal displace- largely elastic, and by following similar pro-
ments (|Dshear|, |Dbending|), the elastic stiffness cedures to those described for the elastic
0.9 tcf + 0.05 dc tcf Kpl

Kpl, spring ¼ (36)
cos2 a
Dy, pl
Dply, spring ¼ (37)
cos a
bc
4. Finally, the stiffness provided by the panel
zone owing to strain hardening in shear (the
first term of Eq. 34) is given by:
Veq
dc
KS:Hardening ¼ m (38)
Dshear
Seismic Analysis of Steel–Concrete Composite Build-
Therefore, the strain hardening stiffness of
ings: Numerical Modeling, Fig. 20 The cross-sectional
definition for the post-elastic range the diagonal spring can be determined from
KS:Hardening
KS:HSpring ¼ (39)
range, the post-elastic stiffness for the panel cos2 a
can be obtained from Eq. 34:
Moment–Rotation Relationship of Panel Zone
Veq Veq 2It The rotational springs in the panel zone are
Kpl ¼ m þ (34)
Dshear Dbending Icol modeled using trilinear moment–rotation rela-
tionships while using the stiffness derived from
where the above expressions for the elastic and post-
m = strain hardening parameter elastic range (later in Fig. 22). According to that
It = second moment of area of the T-section figure, the panel zone is expected to behave in an
obtained analytically asymmetric manner in tension and compression
Icol = second moment of area of the column owing to the presence of concrete slab which
obtained analytically influences the stiffness, yield moments, and
2. The relative deformation, Dy,pl, of the panel cyclic behavior of the panel zone. Under negative
zone in the post-elastic range can be obtained moment, the concrete slab contribution is ignored
using the principle of virtual work and can be owing to cracking of the concrete. Yield points
readily derived as: (M+py, y+py) and (M
py, ypy) are controlled by the
steel yielding, and the ultimate points (M+pu, y+pu)
f y db 2 and (M
pu, ypu) are controlled by the ultimate S
Dy, pl ¼ Dy, el þ (35)
6E dCG strength of the concrete. Under cycling loading,
unloading occurs in a straight line with the same
where slope as the initial stiffness Ke. The reloading is
dCG = distance from the centroid of the directed towards the previous peak, thereby con-
T-section to the external fiber of the column sidering some strength and stiffness degradation.
flange
f y = reserve stress in the same fiber after
shear yielding of the panel zone Modeling of Frames Using Beam
3. Having known the calculated post-elastic Elements
stiffness and relative drift of the panel zone,
the post-elastic stiffness and relative drift of This section presents an analytical approach for
the spring can be obtained using the Eqs. 36 the assessment of SCC frames under earthquake
and 37: excitation using two-dimensional SCC beam
Seismic Analysis of s1 V1
Steel–Concrete s3
Numerical Modeling, J
y
Fig. 21 Element relative v3
forces and deformations in θ1 θJ
local coordinate system V2 VJ
s2 R6 R4
x
V1 J
I R5
L UJ
R1 U1
R3 R2
elements (Kim and Engelhardt 2005). In order to behavior which was observed from experiments.
model the behavior of SCC beams under earth- Each complete element is characterized by two
quake excitation, factors such as the beam-to- external and two internal nodes. The internal
column connection details, the local crushing of nodes are located between the connection of the
concrete, the loading pattern on beam, as well as linear elastic beam element and the hinges, while
the bond behavior between the reinforcing steel the external nodes connect the entire structure.
and the concrete need to be considered into the Each of the external nodes has 3DOFs, 2-
modeling. It is widely accepted that three- translations, and 1-rotation in the local coordinate
dimensional FE models can accurately predict system as presented in Fig. 21.
the behavior of SCC beams at the expense of In the local coordinate system, the element can
time and computational efficiency. On the other be considered as a simply supported beam given
hand, these simpler but reasonably accurate that the rigid body motions are removed. Based
two-dimensional SCC beam models can provide on equilibrium, using the values of relative forces
an alternative tool for frame response assessment. (s1, s2, s3), all the components of local nodal
forces (R1 to R6) can be calculated. The transfor-
Beam Elements mation of forces is defined using the following
The beam elements are described as a relationship:
one-component series hinge-type model combin-
R¼As (40)
ing analytical formulations calibrated against to
experimental data and to other data from “sophis- where
ticated” described models, while intended to rep-
resent the clear span of beams in moment frames A = force transformation matrix (this is well
(i.e., the length of the beams between column known and can be found in literature, i.e.,
flanges). The two-dimensional SCC beam ele- Przemieniecki (1968))
ments are described by a linear elastic beam
with a nonlinear zero-length hinge at each end; From the geometry, the transformation from
the resulting element is referred as “complete the local displacements, r, to the relative defor-
element.” Each of the hinges is described using mations (v1, v2, v3) is performed by:
nonlinear rigid-plastic moment–rotation relation-
ships in order to simulate the real structural v ¼ AT r (41)
where element” expressed in terms of degrees of free-

dom, v, as follows:
rT ¼ fUI , VI , yI , UJ , VJ , yJ g (42)
dv ¼ ds þ dwp ¼ Ft ds (45)
Element Stiffness
The initial stiffness of the aforementioned “com- The hinge flexibility coefficients, fp, can be
plete element” is that of the linear elastic beam. simply added to the appropriate coefficients of
As gradual yielding occurs at the hinges owing to the elastic element flexibility matrix, f, in order to
increased moments at the element ends, the stiff- obtain the tangent flexibility matrix, Ft, for the
ness of the “complete element” reduces accord- “complete element,” as shown in Eq. 46. Once
ingly. In order to obtain the reduced stiffness at the 3 3 tangent flexibility matrix is obtained, it
any load step after yielding, the instantaneous is then inverted to obtain the 3 3 tangent stiff-
tangent flexibility of the nonlinear rigid-plastic ness matrix, Kt:
force–deformation relationship for a hinge is
combined with the flexibility of the elastic beam
L
element. 0 0
EA
A flexibility matrix, f, is first formed for the
0 L
þ
1
þf i

L
þ
1
elastic element including the effects of elastic Ft ¼ F F
EAþ EAþ
ii ij
GAs L p GAs L

shear deformation through the following 0 L F þ 1 L 1
þf jp
þ ij þ Fjj þ
relationship: EA GAs L EA GAs L
(46)
dq ¼ f ds (43)
where where
dq = (dq1, dq2, dq3) = elastic deformation incre- EA+ = flexural rigidity of the composite beam
ment at the internal nodes EA = axial rigidity of the composite beam
ds = action increment in which GAs = effective shear rigidity of the composite
dsT = {dF, dMI, dMJ} = {ds1, ds2, ds3} beam
Fii = Fij = 1/3 and Fjj = 1/6 for a uniform
For hinges at nodes I and J, the incremental member
action–deformation relationship is expressed by fpi, fpj = flexibility of hinges at nodes I and J
8 9 8 9
< 0 = < dv1 dq1 = Hysteretic Rules
dwp ¼ dyIP ¼ dv2 dq2 ¼ f p ds (44) Due to the cross-sectional asymmetry of the SCC
: J; : ;
dyP dv3 dq3 beam, the response will be different for positive S
and negative moments. As a result, a hinge must
where discern the load paths to model the hysteretic
behavior of SCC beam for an arbitrary cyclic
dwp = vector of plastic hinge deformations at loading. Apart of the cross-sectional asymmetry,
nodes I and J the hysteretic rules employed for the complete
dyIP, dyJP = incremental plastic rotation at nodes element need to take into account factors such
I and J as the strength deterioration and the stiffness
fp = hinge or plastic flexibility matrix in which degradation.
nonzero terms are the second and third ele- In this part of the chapter, the hysteretic rules
ments in the diagonal are determined from the modification of Lee’s
model (1987) in order to better fit the curves
Using Eqs. 43 and 44, the action–deformation and the nonlinear behavior of experimental test
relationship can be obtained for the “complete specimens. This model employs a specified
Seismic Analysis of a Monotonic envelope curve

Steel–Concrete M
+ + A
Numerical Modeling, My K sh
Fig. 22 Proposed cyclic
moment–rotation model of
a SCC beam; (a) monotonic
Positive bound line
model and stiffness + +
degradation for negative − Ke Ke
Ksh
moment; (b) stiffness
degradation and pinching Neutral line
for positive moment B
−
− αM y θ
Ke
− −
Ksh −
Ke
Ke C −
K sh
D − Negative bound line

My
−
K sh Negative inelastic curve
b M βΔθp
+ A +
K sh A' K sh
Δθp = plastic rotation
+
accumulated My
for path A - D +
Ke
+ Linear crack closing
Ke
F
−
Ke
Positive inelastic curve
B
γ θA' C θA' θ
E
−
Ke x
− δin
Ke
− D
Ksh δx
multi-linear moment–rotation relationship based The basic parameters describing the

on two bilinear curves. moment–rotation relation is the moment at
For the monotonic loading, two bilinear the yielding point and the elastic and inelastic
moment–rotation relationships are employed to stiffness for both positive and negative bend-
consider the asymmetrical cross section and the ing moments. These parameters can be readily
early cracking of the concrete slab under negative obtained analytically. The value of the strain
moments (Fig. 22a). For the cyclic loading, the hardening stiffness is expressed as a fraction
moment–rotation relationships of the steel beam of the respective elastic stiffness, i.e., the
are modified in order to account for the effect of 2.5 % of elastic stiffness in the positive bend-
the concrete slab (i.e., crack closing and opening) ing and 5 % of elastic stiffness in the negative
(Fig. 22b). bending.
bcf
1.3 fc'
1.3 fc' bcf tc
tc
Ar Fyr
yn Fyr
yr ysc
Asc Fy
P.N.A. Fy
ys
(As - Asc) Fy
Fy
Seismic Analysis of Steel–Concrete Composite Buildings: Numerical Modeling, Fig. 23 Plastic stress distribu-
tion for composite beam
The engineer is then required to calculate the is the plastic moment of both the steel beam
effective width of the concrete slab on each side section and the reinforcing steel bars within the
of the beam centerline for computing the positive effective width.
elastic stiffness. This will enable the calculation The contribution of the concrete slab to the
of the second moment of inertia of the ultimate moment at the connection is determined
transformed SCC section. The effects of slip using the column width and a concrete compres-
between the concrete and the steel on the positive sive bearing stress of 1.3f’c, due to concrete con-
elastic stiffness are taken into account through finement near the face of the column.
the use of a fraction of the second moment
of inertia of the transformed SCC section.
The calculation of the moment of inertia for the Modeling of SCC Frames with
negative elastic stiffness needs to be performed, Concrete-Filled Steel Columns
taking into account the steel beam section
and the reinforcing steel bars within the effective The advantages of concrete-filled steel (CFS)
slab width. The varying moment of inertia of structures in terms of high strength, high ductil- S
SCC and cracked sections along the length ity, and large energy absorption led to their exten-
of the beam is assumed to be equivalent sive use in high-rise structures in earthquake-
to a uniform moment of inertia of a cracked prone regions.
section with reinforcing bars within the effective This section presents a numerical procedure
width. for the nonlinear inelastic analysis of CFS frames
based on a fiber beam-to-column element. The
Calculation of Moment at the Yielding and nonlinear response of SCC frames is captured
Ultimate Point Using a plastic stress distribu- through the inelasticity of materials or due to
tion for SCC beams (Fig. 23), the ultimate changes in the frame geometry. Global geometric
moment (Mmax) at the connection can be nonlinearities (P-d effects) are taken into account
obtained. The positive yield moment (My+) is by the use of stability functions derived from the
assumed to be a fraction of the calculated ulti- exact stability solution of a beam-to-column ele-
mate moment. The negative yield moment (My) ment subjected to axial forces and bending
Seismic Analysis of y
Steel–Concrete
Fiber i zi Steel fiber
Numerical Modeling,
Fig. 24 Fiber hinge
concept
yi
Z
Concrete fiber
y
A B
X
Z X
L
moments. The spread of plasticity over the cross models, the fiber strains are used to calculate the
section and along the member length is captured fiber stresses, which are in turn integrated over
by tracing the uniaxial stress–strain relationships the cross-sectional area to obtain stress resultants
of each fiber on the cross sections located at the (i.e., forces and moments).
selected integration points along the member Using the fiber model, a number of assump-
length. The nonlinear equilibrium equations can tions have been made as follows:
be then solved using an incremental iterative
scheme, based on the generalized displacement 1. Sections remain plane after bending.
control method. 2. Due to the latter assumption, cracking is con-
sidered to be smeared and normal to the
Fiber Beam-to-Column Element and Material member axis.
Nonlinearity 3. Torsional and shear effects are ignored.
The gradual plastification of a composite cross 4. Multi-axial stress states (due to the confine-
section can be described using the concept of ment effects) can be included in the model by
fiber section model, similarly to the modeling of increasing the concrete strength and modify-
the concrete slabs in section “Introduction.” The ing its post-peak response.
fiber model is presented in Fig. 24. The concept 5. Local buckling effects or initial stress arising
behind this model is rather simple; the cross- from thermal effects or erection loads are typ-
sectional area of the SCC element is subdivided ically not included.
into fibers represented by their area, Ai, and
coordinate location (yi, zi – with origin the cen- Confinement of Concrete-Encased Steel
troid of the section). Different material properties Sections
(e.g., concrete confined and unconfined, steel, In order to utilize the material constitutive models
reinforced steel) can be assigned to each of the described in the previous sections, the concrete
fibers. Based on the relevant constitutive material confinement zones need to be identified (Fig. 25).
Seismic Analysis of y
Monitoring area: Am
Steel–Concrete zm
Unconfined Monitoring point 'm'
Numerical Modeling,
Fig. 25 Confinement
zones and distribution of Partially confined
monitoring points of SCC ym
sections z
Fully confined
Partially encased section
y Monitoring area: Am
zm
Unconfined Monitoring point 'm'
Partially confined
ym
z
Fully confined
Fully encased section
Several methods for the identification of the con- Local Flange Buckling
fined zones have been suggested for partially and The flange buckling is a phenomenon that largely
fully encased sections, and confinement factors, depends on the width-to-thickness ratio, the
k, have been developed for use in numerical boundary (i.e., restraint) conditions, and the
modeling. According to EC8, the confinement material properties of the components compris-
factor, k, is given by: ing the section. The effect of the local flange
buckling is the reduction of the ultimate strength
of the section and/or the diminishing of its rota- S
k ¼ 1:0 þ 5:0 ac f1 =f co for f1 =f co < 0:05
tional capacity in the inelastic range. Addition-
(47) ally, the ductility of encased SCC members is
adversely affected by local buckling, and this
k ¼ 1:125 þ 2:5 ac f1 =f co for f1 =f co 0:05 needs to be considered when estimating the rota-
(48) tional capacity.
To account for the local buckling of bare and
where encased steel sections, simple methods have been
developed and can be readily utilized in a frame
ac = confinement effectiveness coefficient (area analysis software. One of the most popular
of concrete/area of confined concrete) methods has been developed by Ballio
f1 = lateral confinement pressure from transverse et al. (1987). In this approach, the cross section
reinforcement is divided into a finite number, i, of strips with
fco = unconfined concrete compressive strength each strip having an assigned area. If the
compressive strain in any strip exceeds the criti- (i.e., structures with low natural period and small
cal strain, ecr, the area of the strip reduces to zero lateral displacement response) subjected to earth-
for the subsequent load cases. In order to quake excitation, the second-order effects are
extend the applicability of this method beyond insignificant and therefore neglected. As steel
the elastic limit into the inelastic range, the structures become taller nowadays, the P–Delta
elastic critical stress is divided by the yield effects are amplified due to the corresponding
strain, as it is represented in the following increase of lateral displacement.
relationship: The effect of axial force acting through the
relative transverse displacement of the member
ecr k p2 Ε ends known as P–d effect can be taken into
¼ (49) account in the modeling by using the geometric
ey 12 sy ð1 n2 Þ wi 2
t stiffness matrix, [Kg], as:
where
½K s ½Ks T
Kg ¼ (50)
wi = distance between the centroid of the strip ½Ks T ½Ks
and the plate connection
t = thickness of the plate where
k = confinement factor
2 3
Ε = Modulus of elasticity of steel 0 a b 0 0 0
v = Poisson’s ratio 6 a c 0 0 0 07
6 7
6 b 0 0 0 0 07
½Ks ¼ 6
6 0
7 (51)
Geometric Nonlinear P–d Effect 6 0 0 0 0 077
Geometric nonlinearities can be classified in two 4 0 0 0 0 0 05
categories. The first category is related to the 0 0 0 0 0 0
global geometric nonlinearities, usually referred
to as P–d effects. The second category is related and
to local geometric nonlinearities (i.e., local buck-
ling), which are generally neglected in frame MzA þ MzB MyA þ MyB P
analysis (while they are carefully considered in a¼ b¼ c¼
L2 L2 L
advanced finite element analyses with discretized (52)
models). The global geometric nonlinearities can
be incorporated in the models following basic
procedures used in nonlinear frame analysis. MzA, MzB, MyA, MyB = end moments with
One of these procedures employs the updated respect to z and y axes, respectively
Lagrangian formulation in order to account for P = axial force
geometric nonlinearities such as large displace- L = length of the element
ments and rotations.
In most of the analyses of multi-storey struc- The tangent stiffness matrix of a beam-to-
tures subjected to earthquake excitation, the column element is then obtained by the following
effects of the combination of gravitational forces relationship:
and lateral displacement are ignored. Such effects
are often referred as second-order effects. The ½K 1212 ¼ ½ T T612 ½ Ke 66 ½ T 612
reason behind overlooking the second-order þ Kg 1212 (53)
effects can be explained by the fact that tradition-
ally, in low-rising reinforced concrete structures where
2 3
1 0 0 0 0 0 1 0 0 0 0 0
6 0 0 1=L 0 1 0 0 0 1=L 0 0 0 7
6 7
6 b 0 0 0 0 0 0 0 1=L 0 1 0 7
½T 612 ¼ 6
6 0
7 (54)
6 0 0 0 0 0 0 1=L 0 0 0 0 7
7
4 0 0 0 0 0 0 0 1=L 0 0 0 1 5
0 0 0 0 0 0 0 0 0 1 0 0
and this chapter is considered as a step towards these

directions.
½Ke 66 ¼ element stiffness matrix given by fDFg The modeling of SCC members in this chapter
¼ ½Ke fDdg (i) serves primarily the computation of the
response of such members when they are
The reader is referred to Thai and Kim (2011) subjected to seismic actions and (ii) acts as
for details on how to derive the element stiffness a vehicle for carrying out the state determination
matrix. of the section (or integration point) to a frame
element and ultimately to the whole frame assem-
Constitutive Models bly. The outcome of the former application is
Any of the constitutive models presented in the typically the moment–curvature response under
previous sections for steel, concrete, and steel a constant axial load. The latter application typi-
reinforcement can be applied for modeling the cally returns section forces that correspond to
material characteristics of the concrete-encased given section deformations (in uniaxial bending
steel sections of the SCC frames. axial strain, and curvature).
In earthquake engineering, the stiffness of the
column members is one of the most important
Summary parameters of the entire structural systems since
it governs the lateral resistance of the frame. The
The validity of the above case studies has been natural period of vibration of the frame
verified by comparing the numerical predictions decreases with increasing stiffness but also
with experimental data obtained from a wide increases with increasing mass. Therefore, the
range of structural systems subjected to static members comprising the frame need to be accu-
and hysteretic loadings. Full details of these com- rately simulated in order to derive the stiffness
parative studies are presented in the literature. and mass matrices to be accounted for the frame
However, such constitutive laws and models are analysis. According to the typical response spec-
usually dependent on parameters which are tra, the acceleration response of a SCC frame S
evaluated through the particular use of the reduces with increased natural period. This
experimental data, and it is in the designers’ implies that a composite structure will have to
discretion to chose and interpret these when resist lower base shear; therefore, the earthquake
data are used for specific purposes. In this chap- effects will be less significant. The displacement
ter, it has been attempted to generalize the con- response of the structure also increases propor-
stitutive models for a number of applications. On tionally with increasing natural period. For the
the other hand, the lack of generality and objec- case of increased lateral displacements, second-
tivity that characterizes most FE packages can order effects (P–d effects) could be developed
only be balanced through the use of material which will determine the design and amplify the
models which are compatible with valid experi- demand on the structure. The engineer is
mental information. In fact, the work presented in required to reduce the influence of the second-
order effects by controlling the lateral displace- de planchers mixtes réalisés avec un bac de tôle mince
ment of the frame providing ductility in the nervurée (Optimization of the dimensioning in the case
of partial connection of beams of composite floors
beams, columns, and connections. As a result, carried out with a thin ribbed sheet vat). Construction
the accurate analysis of each individual compo- Métallique 4 (1989) 12, pp 4–33
nent plays an important role in the aseismic Aribert J-M, Lachal A (2000) Cyclic behaviour of the
design. In the holistic frame assembly, global shear connection component in composite joints. In:
Behaviour of steel structures in seismic area. Proceed-
geometric nonlinearities can be incorporated in ings of STESSA 2000 conference, Montreal, Canada.
the models following basic procedures used in Balkema, Rotterdam, pp 105–112
the nonlinear frame analysis (i.e., modification Balan TA, Filippou FC, Popov EP (1997) Constitutive
of the stiffness matrix). model for 3D cyclic analysis of concrete structures.
J Eng Mech 123(2):143–153
Using nonlinear static analysis, the engineer Ballio G, Calado L, Iori I, Mirabella Roberti G (1987)
can obtain information on the global I problemi delle grandi costruzioni in zona sismica.
ductility and strength of the structure through Associazione Italiana Cemento Armato
force–displacement relations. At each point on e Precompresso, Roma, pp 31–44
Braconi A, Salvatore W, Tremblay R, Bursi OS
the force–displacement curve, the engineer can (2007) Behaviour and modeling of partial-strength
check the member behavior and see whether the beam-to-column composite joints for seismic applica-
limit states are fulfilled. Weak areas and progres- tions. Earthq Eng Struct Dyn 36:142–161
sive hinge formation on the structural frame are Castro JM, Elghazouli AY, Izzuddin BA (2005) Modelling
of the panel zone in steel and composite moment
revealed during the analysis. frames. Eng Struct 27(1):129–144
Cotsovos DM, Kotsovos MD (2011) Constitutive model-
ling of concrete behaviour. In: Papadrakakis M,
Fragiadakis M, Lagaros ND (eds) Computational
Cross-References methods in earthquake engineering, Computational
methods in applied sciences. Springer, Netherlands,
pp 147–175
▶ Equivalent Static Analysis of Structures Fabbrocino G, Manfredi G, Cosenza E (2002) Modelling
Subjected to Seismic Actions of continuous steel-concrete composite beams:
▶ Nonlinear Analysis and Collapse Simulation computational aspects. Comput Struct 80(27–30):
2241–2251
Using Serial Computation Kim DK, Engelhardt MD (2005) Composite beam ele-
▶ Nonlinear Dynamic Seismic Analysis ment for nonlinear seismic analysis of steel frames.
▶ Nonlinear Finite Element Analysis J Struct Eng 131:715–724
▶ Performance-Based Design Procedure for Krawinkler H (1978) Shear in beam-column
joints in seismic design of steel frames. Eng
Structures with Magneto-Rheological J 15(3):82–91
Dampers Lee SJ (1987) Seismic behavior of steel building struc-
▶ Response Spectrum Analysis of Structures tures with composite slabs. PhD thesis, Dept. of Civil
Subjected to Seismic Actions Engineering, Lehigh Univ., Bethlehem
Menegotto M, Pinto PE (1973) Method of analysis for
▶ Seismic Analysis of Steel and Composite cyclically loaded R.C. plane elements including
Bridges: Numerical Modeling changes in geometry and non-elastic behaviour of ele-
▶ Seismic Analysis of Steel Buildings: ments under combined normal force and bending. In:
Numerical Modeling IABSE, Report of the working commissions, Band
13, symposium on resistance and ultimate
▶ Steel Structures deformability of structures acted on by well defined
▶ Seismic Vulnerability Assessment: Reinforced repeated loads, Lisbon
Concrete Structures Newmark NM, Siess CP, Viest IM (1951) Tests and ana-
lyses of composite beams with incomplete interaction.
Proc Soc Exp Stress Anal 9(1):75–92
Ollgaard JG, Slutter RG, Fisher JW (1971) Shear strength
References of stud connectors in lightweight and normal-weight
concrete. AISC Eng J 8(2):55–64
Aribert JM, Al Bitar A (1989) Optimisation du Park R, Paulay T (1975) Reinforced concrete structures.
dimensionnement en connection partielle de poutres Wiley, New York
Seismic Analysis of Wind Energy Converters 2675
Piluso V, Faella C, Rizzano G (2001) Ultimate behavior of

bolted T-stubs I: theoretical model. J Struct Eng
127(6):686–693
Przemieniecki JS (1968) Theory of matrix structural anal-
ysis. McGraw-Hill, New York
Stevens NJ, Uzumeri SM, Collins MP, Will GT
(1991) Constitutive model for reinforced concrete
finite element analysis. ACI Struct J 88(1):49–59
Thai HT, Kim SE (2011) Nonlinear inelastic analysis of
concrete-filled steel tubular frames. J Construct Steel
Res 67:1797–1805
Uchida N, Tohki H (1997) Design of high-rise building
using round tubular steel composite columns. In:
IABSE international conference on composite con-
struction conventional and innovative, Innsbruck,
16–18 Sept 1997
Zhao H, Yuan Y, Ye Y (2012) Simplified nonlinear sim-
ulation of steel-concrete composite beams. J Construct
Steel Res 71:83–91
Zona A, Barbato M, Conte JP (2008) Nonlinear seismic
response analysis of steel-concrete composite frames.
ASCE J Struct Eng 123(6):986
Seismic Analysis of Wind Energy

Converters Seismic Analysis of Wind Energy Converters,
Fig. 1 Horizonal axis wind turbine (HAWT) in Egeln,
Germany (photo by Hadhuey)
Giuseppe Failla
Dipartimento di Ingegneria Civile, dell’Energia,
dell’Ambiente e dei Materiali, (DICEAM), enough to provide about 4 % of the global elec-
University of Reggio Calabria, Reggio Calabria, tricity demand (GWEC 2014).
Italy While many different design solutions have
been considered in the early stages, for commer-
cial use the modern wind industry has now stabi-
Synonyms lized on horizontal axis wind turbines (HAWT).
A typical example is shown in Fig. 1: a land-
Aerodynamic damping; Response spectrum; based tower with a nacelle mounted on the top,
Seismic risk; Soil structure interaction; Wind containing the generator, a gearbox, and the rotor. S
energy converter Typically, three-bladed upwind rotors are used.
With the continuous increase of wind power
production, the search for optimal design is fac-
Introduction ing new and challenging tasks. The design of
land-based HAWTs has been traditionally driven
Wind power is one of the fastest-growing renew- by high wind speed conditions. However, follow-
able energy segments on a percentage basis. In ing the introduction of new technologies such as
2013, over 35 GW of new wind capacity was variable pitch and active control in larger, lighter,
installed all over the world, bringing the total and cost-effective HAWTs, in some cases
wind capacity to 318 GW at the end of 2013 the design-driving considerations have been
(GWEC 2014). The total installed wind capacity changed, with fatigue and turbulence being con-
is expected to reach 365 GW by the end of 2014, sidered in addition to high wind speed conditions.
2676 Seismic Analysis of Wind Energy Converters
For these lighter HAWTs, especially when presented. Throughout the entry, land-based
installed in seismically active areas, a question HAWTs will be referred to.
has soon arisen as to whether seismic loads shall
be considered among design loads. On the other
hand, the need to investigate the potential impor- Seismic Assessment of HAWTs
tance of seismic loads has been corroborated by
the damage that occurred to land-based HAWTs, Load Combinations
following the 1986 North Palm Springs Earth- The selection of appropriate load combinations
quake, USA, and the 2011 Kashima City Earth- for seismic assessment is a relevant issue
quake, Japan. Post-earthquake surveys in the addressed by ISs and CGs. In general, they are
wind farms nearest the epicenter of North Palm recommended based on the observations that
Springs Earthquake documented that 48 out of follow.
65 HAWTs were damaged, generally due to At sites with a significant seismic hazard, there
buckling in the walls of the supporting tower is a reasonable likelihood that an earthquake
(photographs are available in the report by Swan occurs while the HAWT is in an operational
and Hadjian (1988)). Earthquake-induced failure state, i.e., while the rotor is spinning; in this
may occur also at the foundation level, as for the case, the HAWT is subjected to simultaneous
case of the footing of a HAWT in the Kashima earthquake loads and operational wind loads. It
wind farm (photographs are available in the paper shall be considered, also, the possibility that the
by Umar and Ishihara (2012)). In this context, the earthquake triggers a shutdown and that, as a
seismic assessment of HAWTs has drawn an result, the HAWT is subjected to simultaneous
increasing attention in the last years, and as a earthquake loads and emergency stop loads.
result, seismic loading has been progressively Another possible scenario is that the earthquake
included in International Standards (ISs) and Cer- strikes when the turbine is parked, i.e., not oper-
tification Guidelines (CGs) (DNV/Risø 2002; GL ating due to wind speeds exceeding the cutoff
2010; IEC 2005; AWEA 2011). wind speed of the turbine; specifically, blades
The key points in the seismic assessment of may be locked against motion (fixed pitch tur-
HAWTs can be briefly summarized as: bines) or feathered such that no sufficient torque
is generated for the rotor to spin (active pitch
• Selection of the load combinations turbines). In recognition of these observations,
• Use of a specific analysis method the load combinations generally suggested by
• Definition of the structural model ISs and CGs for the seismic assessment of
HAWTs are:
On these points, sufficient information is gen-
erally available in existing ISs and CGs. How- • Earthquake loads and operational wind loads
ever, because a certain flexibility is allowed, • Earthquake loads and emergency stop loads
especially in the definition of the structural • Earthquake loads and wind loads in a parked
model and the selection of an appropriate analy- state
sis method, it is important that engineers be aware
of the potential options available and how they Both earthquake loads and wind loads are
may affect the reliability of the results. In an stochastic processes. The wind process is gener-
attempt to respond to these needs, this entry will ally treated as a stationary process. Samples can
provide, first, a preliminary introduction to the be generated from well-established power spec-
relevant issues involved in the seismic assess- tral densities (PSDs) in the literature (e.g., Von
ment of HAWTs. Hence, detailed prescriptions Karman PSD or Kaimal PSD, see Manwell
of existing ISs and CGs will be reported and, et al. (2010)), with parameters to be set
finally, examples of the possible options for the depending on site conditions. Wind acts on the
implementation of the seismic assessment will be blades of the rotor and along the tower.
Obviously, wind loading on the blades varies approximate since, as explained earlier, the
significantly depending on whether the rotor is actual wind loads on the blades depend on the
spinning or not; to generate wind loading on a oscillations of the tower top, to which contributes
spinning rotor, concepts of classical aerodynam- also the earthquake ground motion at the tower
ics are used, for instance, those of Blade Element base. Nevertheless, numerical comparisons with
Momentum (BEM) theory and subsequent mod- benchmark results obtained by fully coupled
ifications (Manwell et al. 2010). The earthquake time-domain simulations have shown that
process is inherently nonstationary. Spectrum- decoupled analyses can yield accurate results,
compatible samples may be synthetized from provided that the separate response to earthquake
site-dependent response spectra or site-specific loading is computed using an appropriate level of
historical records may be used, according to the damping. In particular, it has been found that:
prescriptions of the adopted ISs and CGs.
• A percentage equal to 5 % of critical damping
Structural Analysis Method is appropriate when the separate response to
The computation of the HAWT response to the earthquake loading is to be combined with the
different load combinations is a crucial step of the response to operational wind loading, i.e., for
seismic assessment. In general, two approaches the load combination = earthquake loads +
can be pursued: operational wind loads (Witcher 2005;
Prowell and Veers 2009; Prowell 2011).
• A fully coupled time-domain simulation com- • In contrast, a percentage varying between
puting the response to simultaneously acting 0.5 % and 2 % of critical damping is appropri-
wind loading and seismic loading ate when the separate response to earthquake
• A decoupled analysis where the responses to loading is to be combined with the response to
wind loading and seismic loading are com- wind loading in a parked state, i.e., for the load
puted separately and then superposed combination = earthquake loads + wind loads
in a parked state (Prowell and Veers 2009;
A fully coupled time-domain simulation is the Prowell 2011; Stamatopoulos 2013).
most desirable approach. The reason is that it
allows the actual wind loads on the blades to be Such a variability of the damping level, to be
evaluated correctly, taking into account that the considered in a decoupled analysis, can be
oscillations of the tower top, induced by the explained as follows:
earthquake ground motion, affect the rotor aero-
dynamics (in particular, the relative wind speed at • The low damping ratios (0.5–2 %), which are
the blades, depending on which lift and drag required for the load combination = earth-
forces are calculated). However, for the imple- quake loads + wind loads in a parked state, S
mentation of fully coupled time-domain simula- are motivated by the fact that when the turbine
tions, dedicated software packages are required, is parked, the only damping source is the
capable of solving the nonlinear motion equa- structural damping of the tower that, as cus-
tions of the structural system under simultaneous tomary in steel structures, is generally low.
wind and seismic excitations. • The higher damping ratio (5 %), which is
When performing a decoupled analysis, required for the load combination = earth-
instead, the responses to wind loading and seis- quake loads + operational wind loads, reflects
mic loading are built separately. This means that the fact that the earthquake loading signifi-
wind loads are evaluated as no earthquake ground cantly affects the aerodynamics of a spinning
motion is acting at the tower base. Correspond- rotor. In particular, a motion of the tower top
ingly, the response to the earthquake ground due to the earthquake loading, against or in the
motion is computed as no wind loading is acting wind direction, causes, respectively, an
on the rotor. It is evident that this approach is increase or a decrease of the instantaneous
thrust force, with respect to that computed Structural Model

assuming no earthquake ground motion at the A fundamental assumption of existing ISs and
tower base (such increase or decrease of the CGs, with regard to the structural model, is mate-
instantaneous thrust force mirrors an increased rial linearity. This assumption is essentially jus-
or decreased relative wind speed at the tified by the fact that the primary intent is to
blades). Since, in both situations, this alterna- ensure power production for the design life of
tion of the thrust force is oriented opposite to the HAWT, usually 20 years, and that nonlinear
the tower top motion, it can be understood that deformation (damage) to the turbine would inter-
its effects are to be modeled by introducing rupt reliable operation. Material linearity means
additional damping with respect to the struc- low operational stresses, and this provides some
tural damping, when computing the separate safety margins against failure (Bazeos
response of the HAWT to earthquake loading. et al. 2002). Therefore material linearity will be,
The same observation holds true when a in general, a prerequisite of ISs and CGs also
decoupled analysis is performed to compute when assessing the response to seismic
the response of HAWTs to combined wind excitations.
loading and wave loading (Kuhn 2001). Starting from the assumption of material line-
arity, in general, two types of structural modeling
The difference between damping for the are feasible:
parked state and operational state is generally
referred to, in the literature, as aerodynamic • Simplified models which model the tower and
damping, to mean that its source is essentially consider the rotor-nacelle assembly (RNA) as
the rotor aerodynamics. a lumped mass at the tower top
Decoupled analyses may be performed in time • Full system models which describe the whole
and frequency domain. Especially frequency- turbine, including the nacelle and rotor with a
domain formulations have been awarded a con- certain level of detail
siderable attention, because in this case the sepa-
rate response to earthquake loading can be built Simplified models are appealing since the
by coded response spectra, a concept most engi- complexities involved in modeling the rotor are
neers are familiar with. However, in light of the avoided. Full system models include the rotor
earlier observations on the appropriate level of blades and, in general, turbine components such
damping, it is evident that particular care shall be as power transmission inside the nacelle and pitch
taken when following this approach. In fact, the and speed control devices, with a different degree
typical 5 % damped response spectra for building of accuracy depending on the specific modeling
structures (ICC 2012) will be suitable only for the adopted, for instance, a finite element (FE) or a
load combination = earthquake loads + opera- rigid multi-body modeling.
tional wind loads, while 0.5–2 % damped Simplified or full system models can be used
response spectra shall be used for the load depending on the selected structural analysis
combination = earthquake loads + wind loads in method. In particular:
a parked state. It is evident that selecting the
correct damping ratio is crucial: if the response • Fully coupled time-domain simulations
to earthquake loading was obtained from 5 % involve only full system models as they
damped response spectra, nonconservative require modeling the rotor aerodynamics,
results would be certainly obtained when the with the earthquake ground motion simulta-
turbine is parked. Some ISs and CGs take account neously acting at the tower base.
of these issues (IEC 2005; AWEA 2011), but no • Decoupled analyses may be implemented
explicit indications on the damping ratio to be using either a full system model or a simplified
adopted are given in the others (DNV/Risø 2002; model. If a simplified model is adopted, seis-
GL 2010). mic loads are built considering the mass of
RNA lumped at the tower top, while wind location of the HAWT relative to the alignment
loads are obtained by a dedicated software of faults, the epicentral and focal distances, the
package, capable of modeling the rotor aero- source mechanism for energy release, and the
dynamics, with no earthquake ground motion source-to-site attenuation characteristics. In this
at the tower base, since the analysis is case, the evaluation should aim to estimate both
decoupled. the design earthquake and the maximum expect-
able earthquake, taking into account also the
potential influence of local soil conditions on
International Standards and the ground motion.
Certification Guidelines No specific recommendations are given on the
earthquake-wind load combinations to be consid-
Guidance for seismic loading on HAWTs can be ered. However, since it is prescribed that in seis-
found in the following ISs and CGs: mically active areas the HAWT should be
designed so as to withstand earthquake loads, it
• DNV/Risø: Guidelines for design of wind tur- is implicit that the three, typical load combina-
bines (DNV/Risø 2002). Released by Det tions described earlier (i.e., earthquake loads and
Norske Veritas (DNV) and Risø National operational wind loads, earthquake loads and
Laboratory emergency stop loads, earthquake loads occur-
• GL 2010: Guideline for the certification of ring in a parked state) shall be referred to.
wind turbines (GL 2010). Released by As for what concerns the method of analysis,
Germanischer Lloyd (GL) DNV/Risø provides explicit suggestions only for
• IEC 61400-1: Wind turbine generator sys- the response spectrum method, as used in a
tems. Part 1: Safety requirements (IEC decoupled analysis. In particular, the use of
2005). Released by International a single degree-of-freedom (SDOF) system with
Electrotechnical Commission (IEC) a lumped mass on top of a vertical rod is
• ASCE/AWEA RP2011: Recommended prac- suggested, with the rod length equal to the
tice for compliance of large land-based wind tower height and the lumped mass including the
turbine support structures (ASCE/AWEA mass of the rotor-nacelle assembly (RNA) and ¼
2011). Released by American Society of of the mass of the tower. It is prescribed that the
Civil Engineers (ASCE) and American Wind fundamental period of the SDOF system is used
Energy Association (AWEA) in conjunction with a design acceleration
response spectrum to determine the loads set up
DNV/Risø Guidelines by the ground motion, by analogy with the sim-
DNV/Risø Guidelines are meant to provide a plified procedures used in building codes. Ana-
basic introduction to the most relevant subjects lyses shall be performed for horizontal and S
in wind turbine engineering (DNV/Risø 2002). vertical earthquake-induced accelerations. How-
Consistently with this general purpose, quite gen- ever, no explicit recommendations are given on
eral suggestions are given to deal with seismic the criterion to translate the resulting spectral
loading. response acceleration into design seismic loads,
It is prescribed that earthquake effects should as well as on the damping ratio to be used. Since
be considered for HAWTs located in areas in the absence of specific guidance on this matter,
that are considered seismically active based a most intuitive choice of engineers could be
on previous records of earthquake activity using the typical procedures of the International
(Section 3.2.8). For those areas known to be seis- Building Code (ICC 2012), it has to be remarked
mically active but with no sufficient information that the 5 % damping ratio, embedded in the
available for a detailed characterization of seis- standard design response spectrum, is appropriate
micity, an evaluation of the regional and local only for seismic loading acting during an opera-
geology is recommended to determine the tional state, but overestimates considerably the
actual damping in a parked state. This aspect code should be applied or, in the absence of
should be well kept in mind when referring to locally applicable regulations, those of either
DNV/Risø for seismic assessment of HAWTs. Eurocode 8 (2004) or American Petroleum Insti-
Regarding the structural model, attention is tute (API 2000).
drawn to the need of including the actual stiffness Regarding the method of analysis, GL 2010
of the structural component of the foundation specifies that fully coupled or decoupled analyses
and an appropriate model of the supporting or are possible, with at least three modes in both
surrounding soil, the latter through a proper cases. Time-domain simulations shall be carried
soil structure interaction (SSI) modeling out considering at least six simulations per load
(Section 8.4). Although, for this purpose, case. As with DNV/Risø, no guidance is provided
nonlinear and frequency-dependent models are on the damping ratio to be adopted when using
recommended in principle, appropriate linearized the design response spectrum in a decoupled
models are allowed, depending on the expected analysis. Again, because of the lack of guidance
strain level in the soil (typically, it may be up to on this matter, it shall be kept in mind that the 5 %
101 for earthquake loading and considerably damping ratio is appropriate only in the opera-
larger than for other loading conditions). The tional state and that lower damping ratios shall be
linearized models consist of translational and considered in the parked state.
rotational springs for circular footings and piles. GL 2010 gives no particular prescriptions on
the structural model to be adopted. However,
GL Guidelines because at least three modes have to be
GL 2010 guidelines aim to set a number of included in the vibration response, the use of a
requirements for the certification of wind turbines multi-degree-of-freedom (MDOF) structural
(GL 2010). For this reason, they are quite pre- model is implicitly suggested. In general, a
scriptive and provide detailed information on linear elastic behavior shall be assumed. A duc-
some particular aspects of seismic risk. tile response can be considered only when the
In agreement with DNV/Risø, GL 2010 pre- support structure has a sufficient static redun-
scribes that seismic loading shall be taken into dancy, such as a lattice tower. However, if
account in seismically active areas (Section ductile behavior is assumed, the structure shall
4.2.4.2.3). Earthquake loading is included in a be mandatorily inspected after occurrence of an
group of design load cases (Table 4.3.2) classified earthquake.
as load cases accounting for “extended” design
situations, including special applications and site IEC Standards
conditions. These design load cases are not man- IEC 61400-1 Standards aim to specify essential
datory for certification purposes, but may be cho- design requirements to ensure structural integrity
sen for the verification of the HAWT to of wind turbines (IEC 2005). They have the status
complement the applicability in specific design of national standards in all European countries
situations. The response to seismic loading is to whose national electrotechnical committees are
be assessed both in the operational state and the CENELEC members (CENELEC = European
parked state (Table 4.3.2) under normal wind Committee for Electrotechnical Standardization).
loading. For the operational state it is also IEC 61400-1 recommends that, in seismically
suggested to consider the activation of the emer- active areas, the integrity of the HAWT is dem-
gency shutdown triggered by the earthquake. The onstrated for the specific site conditions
safety factor for all the loads to be combined with (Section 11.6), while no seismic assessment is
seismic loading is equal to 1.0 (Section 4.3.5.4). required for sites already excluded by the local
A return period of 475 years is prescribed as the building code, due to weak seismic actions. The
earthquake design level. To model the seismic seismic loading shall be combined with other
loading, recommendations of the local building significant, frequently occurring operational
loads. In particular, IEC 61400-1 prescribes that Annex C of IEC 61400-1 presents a simplified,
the seismic loading shall be superposed with conservative method for the calculation of seis-
operational loads, to be selected as the higher of: mic loads. This procedure is meant to be used
when the most significant seismic loads can rea-
(a) Loads during normal power production, sonably be predicted on the tower, and shall not
by averaging over the lifetime be used if it is likely that the earthquake ground
(b) Loads during emergency shutdown, for a motion may cause significant loading on the rotor
wind speed selected so that the loads prior blades or the structural components of the foun-
to the shutdown are equal to those dation. The principal simplifications in Annex
obtained with (a) C are ignoring the modes higher than the first
tower bending mode and the assumption that the
No explicit reference is made, however, to the whole structure is subjected to the same acceler-
load case of an earthquake loading striking in a ation. Upon evaluating or estimating the site and
parked state. soil conditions required by the local building
The safety factor for all load components to be code, or adopting conservative assumptions
combined with seismic loading shall be set equal while detailed site data are not available, the
to 1.0. The ground acceleration shall be evaluated simplified method can be applied as follows:
for a 475-year recurrence period based on ground
acceleration and response spectrum requirements • The acceleration at the first tower bending
as defined in local building codes. If a local natural frequency is set using a normalized
building code is not available or does not provide design response spectrum and a seismic
ground acceleration and response spectrum, an hazard-zoning factor. For this, a 1 % damping
appropriate evaluation of these parameters shall ratio is assumed.
be carried out. • Earthquake-induced shear and bending
Regarding the method of analysis, fully moments at the tower base are calculated by
coupled or decoupled analyses are possible applying, at the tower top, a force equal to the
(11.6). In time-domain analyses, sufficient simu- total mass of the RNA + ½ the mass of the
lations shall be undertaken to ensure that the tower times the design acceleration response.
operational load is statistically representative. It • The corresponding base shear and bending
is prescribed that the number of tower modes moments are added to the characteristic loads
used in either of the above methods shall be calculated for an emergency stop at rated wind
selected in accordance with a recognized build- speed, i.e., the speed at which the limit of the
ing code. In the absence of a locally applicable generator output is reached.
building code, consecutive modes with a total • The results are compared with those obtained
modal mass of 85 % of the total mass shall against the design loads or the design resis- S
be used. tance for the HAWT. If the tower can sustain
IEC 61400-1 gives no particular indications the resulting combined loading, no further
on the structural model for seismic analysis. In investigation is needed. Otherwise, a thorough
agreement with GL 2010, however, it is implicit investigation shall be carried out on a MDOF
that the structure shall be modeled as a MDOF structural model.
system, since the use of consecutive modes with a
total modal mass equal to at least 85 % of the total With regard to such a simplified method,
mass is recommended. In general, the response described in Annex C, it shall be pointed out
should be linearly elastic, while a ductile that ignoring the second tower mode is a signifi-
response with energy dissipation is allowed only cant nonconservative simplification (e.g., see
for specific structures, in particular for lattice Zhao and Maisser (2006) on the role of the second
structures with bolted joints. tower mode in the seismic response of HAWTs).
This is somehow compensated for by incorporat- Unlike the alternative ISs and CGs, ASCE/
ing ½ of the tower mass with the tower head mass AWEA RP2011 provides quite detailed prescrip-
and prescribing superposition with the character- tions on a “best practice” load combination
istic loads calculated for an emergency stop at including seismic loads plus operational loads:
rated wind speed, which represent quite conser-
vative aerodynamic loads. U ¼ ð1:2 þ 0:2SDS ÞD þ 0:75ðrQE þ 1:0MÞ
ASCE/AWEA Recommended Practice (1)
The general purpose of ASCE/AWEA RP2011 is
to clearly identify specific US national recom- U ¼ ð0:9 0:2SDS ÞD þ 0:75ðrQE þ 1:0MÞ
mendations for wind turbine design, which are (2)
compatible with IEC 61400-1 but may provide
proper recommendations for those cases in which where:
US practice and IEC 61400-1 differ. As for what U = factored load effect
concerns seismic assessment, ASCE/AWEA D = dead load
RP2011 makes a quite comprehensive effort to M = operational loading equal to the greater
harmonize some relevant specific prescriptions of of (1) loads during normal power production at
certification agencies with the traditional per- the rated wind speed or (2) characteristic loads
spectives of US standards ASCE/SEI 7-05, calculated for an emergency stop at rated wind
which sets the minimum design loads for build- speed
ings and structures in general (ASCE 2006). For QE = effect of horizontal seismic (earthquake-
the level of detailed information provided, induced) forces
ASCE/AWEA RP2011 can be considered a very SDS = design spectral response acceleration
useful and comprehensive reference tool for the parameter at short periods
seismic assessment of HAWTs. r = 1.0, redundancy factor (for nonbuilding
ASCE/AWEA RP2011 points out that structures not similar to buildings r = 1.0,
although standard HAWT classes shall be gener- according to Chapter 12.3.4.1 of ASCE/SEI 7-05)
ally designed for normal wind conditions, ASCE/AWEA RP2011 suggests Equations
extreme wind, conditions and other environmen- 15.4-1 and 15.4-2 of ASCE/SEI 7-05 for the
tal conditions including temperature and air den- seismic response coefficient Cs (nonbuilding
sity, specific prescriptions on the criteria for the structures), if Equation 12.8-1 of ASCE/SEI 7-
design of HAWTs subjected to earthquake 05 is used to compute the seismic base shear.
ground motions are necessary, in recognition of Specifically, in Eq. 15.4-2 of ASCE/SEI 7-05
the fact that earthquake events are common in for Cs a response modification factor R = 1.5 is
many US jurisdictions. According to ASCE/ recommended by ASCE/AWEA RP2011. The
AWEA RP2011, it is of critical importance to use of R = 1.5 does not necessarily imply that a
recognize that seismic loads plus operational ductile response or material overstrength is
loads may in some cases govern tower and foun- expected but accounts for a certain conservatism
dation design. For these reasons, load combina- in the seismic response coefficient Cs prescribed
tions involving earthquake occurring in an for nonbuilding structures.
operational state and earthquake triggering emer- In Eqs. 1 and 2, the use of a load factor of 0.75
gency stop loads, an earthquake occurring in a on both seismic loads and operational loads is
parked state should be considered (Section 5.4.4). supported by results of time-domain analyses on
Seismic ground motion values should be deter- HAWTs ranging from 65 to 5 MW, subjected to
mined based on the acceleration response spec- 100 earthquake ground motion records, for vary-
trum or site-specific ground motion procedures as ing orientation of wind and earthquake loads
prescribed by ASCE/SEI 7-05 (see Section 11.4 (Prowell 2011). It is observed that when the seis-
and Chapter 21 in ASCE/SEI 7-05). mic hazard at a particular site is dominated by
known faults, consideration of site-specific As for what concerns decoupled analyses,

prevailing wind direction and maximum earth- ASCE/AWEA RP2011 suggests that in those
quake component direction may be appropriate. cases when only the peak seismic loads and
In this case no load factor may be applicable, if peak operational loads are available, the pro-
wind and wave propagation directions are posed combination method for seismic loads
expected to coincide. and operational loads may be overly conserva-
ASCE/AWEA RP2011 recommends that for tive, especially in recognition of the fact that the
load combinations not including operational respective peak loads do not occur at the same
loads, the spectral response acceleration parame- instant of time and in the same loading direction.
ter should be based on a 1 % damping ratio, which Therefore, to reduce potential design conserva-
reflects the low inherent damping of typical steel tism and obtain a more accurate prediction of the
support structures for HAWTs. The multiplica- response, fully coupled time-domain analyses are
tive spectral adjustment factor, B, to adjust spec- suggested, considering earthquake ground accel-
tral response acceleration, Sa, from 5 % (standard eration in combination with operational or emer-
IBC value for determining Sa) to 1 % damped gency stop loads. Seismic analysis should comply
values is equal to 1.40 (Table 5-6). For load with the requirements of ASCE/SEI 7-05,
combinations that include operational loads, the Chapter 16, concerning, for instance, the mini-
spectral response acceleration parameter should mum number of simulated earthquake ground
be based on 5 % damped values. ASCE/AWEA motions. It is recommended that time-domain
RP2011 points out that this increase in damping is analyses be conducted with analysis software
based on the aerodynamic damping inherent to an capable of simulating the structural response
operating HAWT as verified by experimental and and global turbine dynamics, including the aero-
numerical results showing that a damping level of dynamic interaction.
1 % produces overly conservative results ASCE/AWEA RP2011 gives some interesting
(Prowell 2011). points of view on the applicability of typical pre-
Regarding the method of analysis, according scriptions of building codes to HAWTs design.
to ASCE/AWEA RP2011 a fully coupled time- For instance, according to ASCE/AWEA
domain analysis and decoupled analyses based on RP2011, enhanced performance objectives may
equivalent lateral force method or modal be established to meet specific owner require-
response spectrum method are acceptable, as per- ments and to improve expected behavior during
mitted by the local building code. For the specific and after an earthquake. For this purpose the use
implementation of each method of analysis, the of a performance factor, similar to an importance
local building code or ASCE/SEI 7-05 is referred factor of 1.5 for essential facilities, is suggested.
to. In particular, if the equivalent lateral force This performance factor shall be agreed with the
procedure is used, the vertical distribution of wind turbine manufacturer to establish accelera- S
seismic forces should be calculated based on the tion thresholds for turbine components that will
procedure given in ASCE/SEI 7-05, ensure operational performance. Also, according
Chapter 12.8.3, with some modifications: the to ASCE/AWEA RP2011, no specific drifts or
seismic forces corresponding to the seismic displacement need to be defined. This is moti-
weight of the RNA should be located at the tur- vated essentially by the fact that thorough analy-
bine’s center of gravity, and those corresponding sis and design considerations of the ultimate and
to the seismic weight of the tower structure fatigue limit states implicitly limit the displace-
(including ladders, platforms, railings, etc.) ments of the tower. In addition, the wind turbine
should be distributed to nodes distributed along controls monitor and limit the possible tower top
the tower height. No further prescriptions are accelerations, to prevent exceeding the design
given on a specific structural model to be loading.
adopted, when implementing a fully coupled or Another distinctive feature of the ASCE/
a decoupled analysis. AWEA RP2011 is the emphasis put on
consideration of seismic forces in the foundation component loads instead of only tower loads,
design, for areas with historical earthquake activ- which cannot be estimated in a simple tower-
ity. Evaluation of earthquake effects should be based model. In this regard, it is remarked that
performed in accordance with the requirements higher modes involving the rotor dynamics may
of the local building code or IEC 61400-1. In any play an important role, as they may fall in the
case, geotechnical evaluation of earthquake region of maximum spectral response accelera-
effects should include ground shaking, liquefaction (Prowell et al. 2010).
tion, slope instability, surface fault rupture, seis- Regardless of the structural model adopted,
mically induced settlement/cyclic densification, another important issue in the seismic assessment
lateral spreading, cyclic mobility, and soil of HAWTs is SSI modeling. Modeling the base as
strength loss. In areas susceptible to earthquake fixed, with no consideration of the SSI, could be
effects, appropriate mitigation should be pro- justified in the case of overdesigned foundations
vided for foundations. For projects located near and stiff soil conditions. However, HAWTs may
active faults, the characteristics of the fault be installed on relatively soft soils or loose soils
including type, seismic setting, subsurface condi- containing alluvial deposits, and under these cir-
tions, ground motion attenuation, and maximum cumstances the SSI modeling, particularly for
earthquake magnitude should be considered. dynamic loads, could become a major concern
At any rate, HAWTs should be located with ade- in the design of the foundation and, consequently,
quate setbacks from fault zones. Where relatively of the entire support structure of the HAWT
loose unsaturated cohesionless soils are present at (Bazeos et al. 2002; Zhao and Maisser 2006).
the project site, the effect of ground shaking from A proper SSI modeling may play an important
a design level earthquake should be taken into role also in consideration of the ground motion
account. Also, potential settlement due to cyclic amplification effects on soft and loose soils.
densification of the site soils should be evaluated. In the following, simple and full system
models will be described briefly as used in the
recent literature, along with relevant information
Implementation of Seismic Assessment on SSI modeling.
Despite the prescriptions given by existing ISs Simple Models

and CGs (DNV/Risø 2002; GL 2010; IEC 2005; One of the first studies on the seismic response of
ASCE/AWEA 2011), engineers dealing with the HAWTs has been carried out by Bazeos and
seismic assessment of HAWTs may face a few coworkers (Bazeos et al. 2002). They have inves-
issues, which are only partially addressed by ISs tigated a 38 m high HAWT resting on a concrete
and CGs. This section is meant to provide some block, located in a site with 0.12 g peak ground
insights into these aspects, illustrating the most acceleration and semi-rock soil conditions. Seis-
relevant studies for this purpose. Although inves- mic analyses have been conducted on two differ-
tigations are not fully accomplished, it is worth ent models, a FE model of the tower with shell
mentioning these studies, as they may provide elements and a simplified FE model with lumped
engineers with very useful data for a correct seis- masses along the tower height and 3D beam ele-
mic assessment of HAWTs. ments approximately mapping mechanical and
As mentioned earlier, an important step is the geometrical properties of the tower. In both
construction of the structural model. Simplified models, a top mass has been added to model the
models, which avoid the complexities involved in RNA; SSI has been modeled by a set of discrete
modeling the rotor, are frequently used, espe- springs and dashpots and, adding to the mass of
cially for a preliminary design. In contrast, for a the concrete foundation block (modeled as rigid),
comprehensive investigation of all factors rele- a virtual soil mass moving in phase (Mulliken and
vant to seismic risk, full system models are gen- Karabalis 1998). Parked conditions only have
erally considered. They allow prediction of been considered, with no aerodynamic loads
along the tower. Seismic analysis has been car- five modal responses obtained by the specified
ried out in the time domain using ground motions design spectrum, match very well the
compatible with the elastic response spectrum, as corresponding values obtained with time series
prescribed by the Greek Aseismic Code with analyses. Results have been provided for HAWTs
0.5 % damping. For the relatively low ground with different size.
acceleration under consideration (0.12 g), low Stamatopoulos (2013) has addressed the
stress levels have been found due to seismic exci- response of HAWTs to near-fault ground
tations, with respect to the stress levels due to motions. He has investigated a 53.95 m tall tur-
wind in either operational or emergency states. bine resting on a circular footing. A FE model of
Results obtained by the time-domain simulations the tower and the circular footing has been used,
on the two FE models have been validated by a with a lumped mass at the tower top modeling the
response spectrum analysis on a SDOF system RNA; SSI has been modeled by uncoupled
with a mass set equal to the total mass of the nonlinear springs distributed below the footing.
system and a stiffness computed from the first The response to near-fault ground motion has
natural period of the FE model with shell ele- been investigated by three methods: a response
ments. Interestingly, Bazeos and coworkers spectrum method based on the elastic accelera-
(2002) have showed that higher tower modes tion spectrum provided by the Greek Aseismic
can be significantly affected by SSI modeling. Code, suitably increased by 25 % to account for
This result is important in consideration of the proximity to a seismic fault; a response spectrum
fact that the natural frequencies of the higher method involving an elastic local acceleration
tower modes may fall within the region of max- spectrum built based on actual records for the
imum spectral acceleration (Haenler et al. 2006). project site; and a time history analysis using
Umar and Ishihara (2012) have focused on the synthetized ground motions compatible with the
construction of a response spectrum for HAWTs elastic local acceleration spectrum. The two
under seismic excitations only, i.e., in a parked response spectrum analyses have been carried
state. The need for a specific response spectrum is out on a FE model with the tower grounded by a
motivated by the observation that the support linearly elastic rotational spring, with stiffness
structures for HAWTs exhibit, unlike buildings, computed as the ratio of the bottom bending
long period, heavy top, and different mass distri- moment to the bottom rotation. An iterative pro-
bution along the height. Besides this general cedure has been implemented, since the bottom
observation, by carrying out numerical simula- bending moment and rotation depend on the seis-
tions using a database of strong earthquake mic loading computed from the spectral acceler-
ground motions, they have shown that the very ation; the latter depends on the first natural
low damping levels in parked conditions deter- period, which depends in turn on the stiffness of
mine excessive fluctuations in the response spec- the grounded rotational spring. Stamatopoulos S
trum, and such uncertainty cannot be captured by (2013) has shown that the acceleration spectrum
existing damping correction factors in Eurocode provided by the Greek Aseismic Code, although
8 (2004) and Japanese Building Standard Law appropriately increased to account for proximity
(BSL 2004). Umar and Ishihara (2012) have of seismic fault, significantly underestimates the
modeled the HAWT as a MDOF system with a shear and bending moment demand at the tower
lumped mass at the top, and a sway-rocking base by 55 %.
model to take into account SSI effects. They Nuta and coworkers (2011) have proposed a
have proposed a modified correction factor for methodology to assess the probability of failure
the damping ratio of the BSL response spectrum of a HAWT under seismic excitations in a parked
used in Japan (BSL 2004), depending on the state. Despite material linearity is generally pre-
natural period and the targeted reliability. They scribed for design, in fact, nonlinear behavior has
have shown that the maximum seismic loads, as to be taken into account when examining the
obtained by a complete quadratic combination of potential failure mechanisms induced by seismic
excitations. For a HAWT with 1.65 MW rated with typical load cases. A combined multi-body
power and 80 m hub height, they have built log- dynamics and modal formulation is adopted also
normally distributed fragility curves to estimate by FAST(Jonkman and Buhl 2005), with flexible
the probability of reaching a defined damage components modeled based on user-provided
state. Parameters of the lognormal distributions mode shapes. A fully coupled time-domain sim-
have been obtained from nonlinear incremental ulation can be implemented, with seismic loading
dynamic analyses, assuming the magnification generated as a user-defined loading imposed at
factor with respect to the design earthquake as the tower base. Like GH BLADED, FAST has
intensity measure, and peak displacement, peak been validated by GL.
rotation, residual displacement, and peak stress as Using BLADED, Witcher (2005) has com-
damage measures. For the specific sites of this pared the results from a response spectrum
study, two in Canada and one in the USA, no method and time-domain simulations as applied
significant probability of failure has been found to a 60 m tall 2 MW turbine subjected to earth-
for the 1.65 MW HAWT under consideration quake ground motion in both operational and
(Nuta et al. 2011). parked cases. He has found that the elastic design
spectrum with 5 % damping ratio yields a maxi-
Full Models mum bending moment at the tower base in a very
In the last few years, a significant effort has been good agreement with that computed by time-
spent on developing advanced tools that may domain simulations, thus inferring that aerody-
allow a full system modeling of HAWTs. Due namic damping experienced by an operating tur-
to computational complexity involved in a FE bine is quite close to 5 %. However, the results of
modeling of all components of a HAWT, models the response spectrum method and time-domain
with a limited number of degrees of freedom have analyses were very different in the parked case,
been built, and, in general, a modal approach, a with the first significantly underestimating the
multi-body approach, or a combination of the two maximum tower base bending moment. This
has been used. Many high quality full-modeling result has confirmed that in the parked case no
software packages are now available for the wind aerodynamic damping is generated and that using
industry, such as GH BLADED (Bossanyi 2000) the response spectrum method with a 5 %
and FAST (Jonkman and Buhl 2005), developed damping ratio does lead to nonconservative
at the United States National Renewable Energy results. Although, in the specific case under
Laboratory (NREL). A few comments on these examination, the bending moment demand due
packages are in order, especially with regard to to earthquake loading in the parked case was
the options available for seismic analysis. lower than that due to earthquake loading in the
GH BLADED uses a multi-body dynamics operational case, Witcher (2005) has drawn the
approach in conjunction with a modal represen- attention to the fact that, in some cases, the driv-
tation of the flexible components like tower and ing load can be that corresponding to earthquake
blades (Bossanyi 2000). A fully coupled time- loading in the parked case and has recommended
domain simulation is feasible, with wind and further investigations on this issue.
seismic loadings simultaneously generated. Two Using FAST, one of the most comprehensive
methods are available for simulating seismic and fruitful studies on the seismic assessment of
loading. The first method allows recorded accel- HAWTs has been carried out by Prowell and
eration time histories to be used, while the second coworkers (2013). They have run a multiyear
method uses an iterative procedure to synthetize research program including extensive numerical
acceleration time histories providing an elastic simulations with FAST on HAWTs featuring dif-
response spectrum that closely matches a speci- ferent sizes and rated power (65 kW; 900 kW;
fied design response spectrum. SSI can be also 1,5 MW and 5 MW) and experimental tests on a
modeled. GH BLADED has been validated by HAWT (65 kW rated power, 22.6 m hub height,
GL for calculating operational loads associated and a 16 m rotor diameter) mounted on the
Seismic Analysis of Wind

a b
Energy Converters,
Fig. 2 HAWT on the
outdoor shake table at the
University of San Diego,
CA. The arrows indicate
the direction of shaking
(From Prowell et al. 2013)
Configuration 1 Configuration 2
(FA shaking) (SS shaking)
outdoor shake table at the University of Califor- fore-aft) and another perpendicular (SS = side-
nia, San Diego (Fig. 2). Prowell and coworkers to-side) to the rotation axis of the rotor. Structural
have collected a considerable amount of data, response characteristics have been recorded for
which has also served as a basis for ASCE/ motions imparted in both configurations and both
AWEA RP2011 prescriptions. Using a set of operational states (Prowell et al. 2013).
99 ground motions with different magnitude and The results have shown that for shaking
source-to-recording distance, they have run imparted in the SS direction, no appreciable dif-
numerical simulations showing that the consid- ferences are encountered between operational
ered earthquakes may produce, in the 5 MW and parked states, in terms of bending moment
HAWT, a bending moment demand at the tower envelopes (Fig. 3b). In contrast, for shaking
base well above that from extreme wind events. imparted in the FA direction, the bending
This result has been found for parked, opera- moment demand at the tower base in the opera-
tional, and emergency shutdown simulations tional state was reduced by approximately S
and confirmed that seismic loads may be design 15–33 % from that while parked (Fig. 3a). This
driving for large turbines in regions of high seis- reduction of demand confirmed that in the oper-
mic hazard. ational state, aerodynamic damping has to be
As for what concerns the experimental tests on accounted for when performing a decoupled anal-
the 65 kW HAWT, only a few key results are ysis, with separately generated wind loading and
reported here, but interested readers can find seismic loading (Prowell et al. 2013). However,
detailed information on both numerical and due to the influence of many factors such as wind
experimental findings in the study by Prowell speed, earthquake magnitude, wind and earth-
and Veers (2009), Prowell et al. (2010, 2013), quake relative directions, and SSI modeling, mul-
Prowell (2011), and references therein. Experi- tiple simulations including likely distributions of
mental tests were carried out in operational and wind speed and earthquake shaking have been
parked states. In each state, shaking has been recommended for an accurate quantification of
imparted in two directions, one parallel (FA = the aerodynamic damping (Prowell et al. 2013).

Energy Converters,
Fig. 3 Experimental
bending moment envelope
for three earthquake ground
motions (From Prowell
et al. 2013)

Energy Converters,
Fig. 4 Experimental
acceleration envelope for
three earthquake ground
motions (From Prowell
et al. 2013)
The same simulations should serve for a proba- absolute acceleration envelopes in the FA and SS
bilistic description of the influence of seismic direction have shown that, in addition to the first
excitations on extreme loads. mode, the second mode contributes significantly,
A further important result obtained by Prowell as indicated by the high acceleration values at
and coworkers (2013) concerns the relative con- two thirds of the tower height (Fig. 4). This result
tributions of various tower modes. The maximum has confirmed what predicted by other

Energy Converters,
Fig. 5 SSI modeling for
seismic assessment of
HAWTs (From Zhao and
Maisser 2006)
researchers (Haenler et al. 2006) on the impor- The MBS has been used by Zhao and Maisser
tance of the second tower mode for estimation of (2006) to assess the seismic response of a 65 m
seismic loads on large turbines. However, on the high HAWT. SSI has been modeled approxi-
basis of the response PSDs, it has been found that mately by a frequency-independent discrete
amplification of energy imparted near the fre- parameter model, as a 3D set of uncoupled
quency of the second modes was not significantly spring-damper devices, including translations
influenced by operational state, thus implying and rotations (Fig. 5).
that aerodynamic damping shall preferably be Seismic analysis has been carried out in oper-
accounted for only in the first mode response ational conditions with a three-component weak
(Prowell et al. 2013). real earthquake record. Numerical results have
Besides the software packages used by the showed that while the top displacement is domi-
wind industry, there exist also a few full models nated by the wind thrust, force and bending
of HAWTs, which have not been translated in moment in the longitudinal direction at the
software packages for the wind industry yet, but tower base are affected considerably by earth-
have provided very interesting results on the seis- quake loads. In addition, in the lateral direction
mic response of HAWTs. (where there are no wind loads), the force and
For instance, Zhao and coworkers (Zhao and bending moment are essentially decided by the S
Maisser 2006; Zhao et al. 2007) have developed a earthquake loads and amplify several times.
hybrid multi-body system (MBS) for full model- Another interesting result concerns the lateral
ing of HAWTs. By this approach, the elastic reaction force at the main bearing that is found to
tower is discretized into a series of rigid bodies be significantly increased with respect to the case
coupled elastically by constraint joints and of no earthquake loading. This increase of the
springs. The wind rotor is treated as a rigid disk; lateral reaction force is attributed to the gyro-
nacelle and machine carrier are coupled by an scopic forces that arise because of the change of
in-plane joint and treated as a rigid ensemble the wind rotor direction, due to the earthquake
connected to the tower top through a revolute loading. A further result of the study by Zhao and
joint. The governing equations are derived using Maisser (2006) is that SSI modeling affects sig-
Lagrange’s equations. This approach, though nificantly the higher tower modes, especially the
more mathematically rigorous, does not require second lateral bending mode. In this respect, this
external calculation of component mode shapes. result confirms the results obtained by Bazeos
et al. (2002) using a simplified model of the contributions due to higher modes, obtained on
HAWT and shall be taken into account in consid- the basis of the FE analysis. Results have been
eration of the fact that especially the higher tower provided for two HAWTs, one with 400 kW rated
modes may fall within the region of maximum power, 36 m hub height, and 31 m rotor diameter
spectral acceleration (Haenler et al. 2006). and the other one with 2 MW rated power,
Using a multi-body system with flexible parts 67 m hub height, and 80 m rotor diameter.
(tower, blades) described by a variable number of By comparison with the results obtained from
modes and including SSI modeling, Haenler and time-domain simulations, it has been shown that
coworkers (2006) have investigated the seismic the BSL 5 % damped response spectrum provides
response of a HAWT with an 80 m rotor diameter nonconservative seismic load profiles for both the
and 60 m hub height, operating under a 13 m/s 400 kW HAWT and the 2 MW one, thus
wind speed and subjected to an earthquake confirming that a 5 % damping ratio is not appro-
ground motion with a 0.3 g peak ground acceler- priate for HAWTs in parked conditions, which
ation (PGA). They have showed that the full experience much lower damping level. It is worth
system model predicts modes at frequencies in remarking that the time-domain simulations car-
the region of maximum spectral response accel- ried out by Ishihara and Sarwar (2008)
eration for typical design response spectra. An highlighted that contribution of higher modes
important contribution of this study regards the may become significant for large HAWTs
relative increase in higher mode response. It has (2 MW) under earthquake excitations, confirming
been found that, for normal wind loading, 80 % of similar results obtained by other researchers
the tower energy is associated with the first mode, (Haenler et al. 2006).
while, during the considered earthquake, the
energy in the first mode is reduced to 54 % per-
cent only of the tower energy, thus concluding Summary
that higher tower modes are more important for
earthquake loading than for typical wind loading. The importance of seismic loading in the design
Another full model of HAWT has been proof HAWTs is recognized in existing ISs and CGs,
posed by Ishihara and Sarwar (2008). Starting and in seismically active areas attention shall be
from the observation that, unlike wind loads, paid to the possibility that design is driven by
seismic waves may excite a wide range of fre- load combinations involving seismic excitations.
quencies including those of higher modes, they Existing methods for seismic assessment of
have developed a nonlinear FEM code (CAST) HAWTs are fully coupled time-domain simula-
for a full FE modeling of the HAWT and its tions, computing the response to simultaneously
components. Beam elements with a linear mate- acting wind loading and seismic loading, and
rial have been used to model tower and blades. decoupled analyses, where the responses to
Analyses performed by the FE code on HAWTs, wind loading and seismic loading are built sepa-
in a parked state, have been used in conjunction rately and then superposed. Fully coupled time-
with the Japanese BSL response spectrum (BSL domain simulations, although complex and time
2004) to derive design formula for the prediction consuming, remain the indispensable benchmark
of seismic loads on two turbines. In particular, tool for an accurate seismic assessment of
following a semi-theoretical codified method HAWTs. However, decoupled analyses are also
provided by the Japan Society of Civil Engineer- important, especially in the early stages of
ing (JSCE 2007), a profile of seismic loads (shear design.
and moment demand) acting on a HAWT in For the implementation of fully coupled time-
parked conditions was estimated from a base domain simulations, a full system model must be
shear built as the sum of a shear force due the employed. Simplified models, where the rotor is
first mode, obtained by using the BSL accelera- not modeled, are allowed to build separate seis-
tion response spectrum, and the shear force mic loads in decoupled analyses.
Despite the prescriptions of existing ISs and Eurocode 8 (2004) Design of structures for earthquake
CGs, a few important aspects are actively being resistance. Part 1: general rules, seismic actions and
rules for building. EN 1998-1. European Committee
investigated, such as the influence of SSI model- for Standardization, Brussels
ing, the importance of higher modes in the seis- GL (2010) Guideline for the certification of wind turbines.
mic response, the potential failure mechanisms Germanischer Lloyd, Hamburg
under earthquake loading, and the estimation of GWEC (2014) Global wind report: Annual market update
2013. Global Wind Energy Council, Brussels
aerodynamic damping for decoupled analysis. Haenler M, Ritschel U, Warnke I (2006) Systematic
A full understanding of these aspects is desirable modelling of wind turbine dynamics and earthquake
for a correct seismic assessment of HAWTs. loads on wind turbines. In: Proceedings of the Euro-
pean Wind Energy Conference & Exhibition (EWEC),
27 Feb–2 Mar 2006. Athens
ICC (2012) International building code. International
Cross-References Code Council, Country Club Hills
IEC (2005) Wind turbine generator systems. Part 1: safety
requirements, 3rd edn, IEC 61400-1. International
▶ Earthquake Response Spectra and Design Electrotechnical Commission, Geneva
Spectra Ishihara T, Sarwar MW (2008) Numerical and theoretical
▶ European Structural Design Codes: Seismic study on seismic response of wind turbines. In: Pro-
ceedings of the European Wind Energy Conference &
Actions
Exhibition (EWEC), 31 Mar–3 Apr 2008. Brussels
▶ Modal Analysis Jonkman JM, Buhl ML (2005) FAST user’s guide, NREL/
▶ Response Spectrum Analysis of Structures EL-500-38230. NREL, Golden
Subjected to Seismic Actions JSCE (2007) Guidelines for design of wind turbine sup-
port structures and foundations, Japan Society of Civil
▶ Soil-Structure Interaction
Engineers, Tokyo (in Japanese)
▶ Stochastic Ground Motion Simulation Kuhn M (2001) Dynamics and design optimisation of
▶ Time History Seismic Analysis offshore wind energy conversion systems. Report
no. 2001.002, Delft University Wind Energy Research
Institute (DUWIND), Delft
Manwell JF, McGowan JG, Rogers AL (2010) Wind
References energy explained: theory, design and application,
2nd edn. Wiley, Chichester
API (2000) Recommended practice for planning, design- Mulliken JS, Karabalis DL (1998) Discrete models for
ing and constructing fixed offshore platforms, API through-soil coupling of foundations and structures.
RP-2A. American Petroleum Institute, Washington, Earthq Eng Struct Dyn 27:687–710
DC Nuta E, Christopoulos C, Parker JA (2011) Methodology
ASCE (2006) Minimum design loads for buildings and for seismic risk assessment for tubular steel wind tur-
other structures, ASCE/SEI 7-05. American Society of bine towers: application to Canadian seismic environ-
Civil Engineers, Reston ment. Can J Civil Eng 38:293–304
ASCE/AWEA (2011) Recommended practice for compli- Prowell I (2011) An experimental and numerical study of
ance of large land-based wind turbine support struc- wind turbine seismic behavior. PhD dissertation, Uni-
tures, ASCE/AWEA RP2011. American Society of versity of California, San Diego S
Civil Engineers/American Wind Energy Association, Prowell I, Veers P (2009) Assessment of wind turbine
Reston/Washington, DC seismic risk: existing literature and simple study of
Bazeos N, Hatzigeorgiou GD, Hondros ID, tower moment demand, SAND2009-1100 report.
Karamaneas H, Karabalis DL, Beskos DE Sandia National Laboratories, Albuquerque
(2002) Static, seismic and stability analyses of a pro- Prowell I, Elgamal A, Uang C, Jonkman J (2010) Estima-
totype wind turbine steel tower. Eng Struct tion of seismic load demand for a wind turbine in the
24:1015–1025 time domain. In: Proceedings of the European Wind
Bossanyi EA (2000) Bladed for windows user manual. Energy Conference (EWEC), 20–23 Apr 2010, Warsaw
Garrad Hassan and Partners, Bristol Prowell I, Elgamal A, Uang C, Luco JE, Romanowitz H,
BSL (2004) The building standard law of Japan. The Duggan E (2013) Shake table testing and numerical
Building Centre of Japan, Tokyo (in English and simulation of a utility-scale wind turbine including
Japanese) operational effects. Wind Energy. doi:10.1002/
DNV/Risø (2002) Guidelines for design of wind turbines. we.1615
Det Norske Veritas/Wind Energy Department, Risø Stamatopoulos GN (2013) Response of a wind turbine
National Laboratory, Copenhagen/Roskilde subjected to near-fault excitation and comparison
2692 Seismic Anisotropy in Volcanic Regions
with the Greek Aseismic Code provisions. Soil Dyn forecasting (see “▶ Volcanic Eruptions, Real-
Earthq Eng 46:77–84 Time Forecasting of”). Another example of geo-
Swan S, Hadjian AH (1988) The 1986 North Palm Springs
earthquake: effects on power facilities, NP-5607 physical precursors to eruptions is surface defor-
research project 2848. Electric Power Research Insti- mation from inflation or deflation of a volcano
tute (EPRI), Palo Alto due to magma movement. However, some volca-
Umar AB, Ishihara T (2012) Seismic load evaluation of noes do not display these clues, and there remains
wind turbine support structures considering low struc-
tural damping and soil structure interaction. In: Pro- a need for techniques that are sensitive to other
ceedings of the European Wind Energy Association physical attributes that might change in conjunc-
Conference (EWEA), 16–19 Apr 2012. Copenhagen tion with the eruption process. Any overpressured
Witcher D (2005) Seismic analysis of wind turbines in the magma storage reservoir, be it a system of dikes,
time domain. Wind Energy 8(1):81–91
Zhao X, Maisser P (2006) Seismic response analysis of sills, conduits, spherical chamber, or a combina-
wind turbine towers including soil-structure interaction. tion of these, will exert a stress on the surround-
Proc Inst Mech Eng K J Multi-Body Dyn 220(1):53–61 ing country rock that may or may not be manifest
Zhao X, Maisser P, Wu J (2007) A new multibody model- as observable strain. Detecting and understanding
ling methodology for wind turbine structures using a
cardanic joint beam element. Renew Energy this stress may be a key to predicting if and when
32:532–546 a volcano will erupt.
Definition
Seismic anisotropy is the variation of seismic
Seismic Anisotropy in Volcanic wave speed with direction. It is an indicator of
Regions geometric ordering in a material, where features
smaller than the seismic wavelength (e.g., crys-
Jessica H. Johnson tals, cracks, pores, layers, or inclusions) have
School of Earth Sciences, University of Bristol a dominant alignment. This alignment leads to
Wills Memorial Building, Bristol, UK a directional variation of elastic wave speed.
School of Environmental Sciences, University of Seismic anisotropy can be measured using many
East Anglia, Norwich, UK seismological techniques but is most frequently
observed using shear wave splitting, which
requires clear shear wave energy recorded on
Synonyms three-component seismometers. Measuring the
effects of anisotropy in seismic data can provide
Seismic anisotropy/shear wave splitting at important information about processes in the
(active) volcanoes; Shear wave splitting in vol- Earth such as stress conditions, material flow,
canic regions; Using seismic anisotropy/shear and the structure of the subsurface and deep
wave splitting to monitor (active) volcanoes; Earth.
Using seismic anisotropy/shear wave splitting to
track/monitor/detect volcanic/magmatic activity Historical Background
Hess (1964) first made a significant observation
of large-scale anisotropy when seismic refraction
Introduction measurements in oceans showed that the P-wave
velocity of the upper mantle (Pn) was consis-
Modern geophysical techniques enable changes tently higher for profiles recorded perpendicular
to be observed at some volcanoes before mag- to an oceanic spreading center (i.e., parallel to the
matic eruptions: detection of seismicity (see direction of spreading or plate movement) than
“▶ Seismic Monitoring of Volcanoes”) from for profiles recorded parallel to the spreading
magma pushing through cold country rock is center. These measurements were attributed to
one of the most common and successful monitor- the alignment of olivine crystals in the mantle
ing techniques and can lead to short-term lithosphere because of flow during the formation
Seismic Anisotropy in Volcanic Regions 2693
Fast direction timescales, and temporal changes are traditionally

φ interpreted as stemming from variations in the
stress field due to large earthquakes or magmatic
delay time intrusions. There is mounting evidence, however,
δt
that the dominant mechanism for seismic anisot-
ropy can switch between a static condition, such as
aligned fractures in fault zones, and a dynamic
process, such as stress causing aligned
microcracks to dilate. In areas where there are
SHmax strong changes in SHmax direction and magnitude
on observable timescales, such as at active volca-
noes, seismic anisotropy analysis has proven
a valuable tool when combined with ground defor-
mation or other seismological observations for
interpretation of volcanic processes such as
magma migration (e.g., Gerst and Savage 2004;
Bianco and Zaccarelli 2009; Unglert et al. 2011).
Seismic Anisotropy
Seismic Anisotropy in Volcanic Regions, Fig. 1 Shear
wave splitting in an anisotropic medium. Anisotropy is Shear Wave Splitting
caused by preferentially aligned cracks due to a maximum Shear wave splitting occurs when a shear wave
horizontal compressive stress (SHmax). A vertically propa-
travels through a seismically anisotropic
gating shear wave that is arbitrarily polarized gets split
into a fast wave with polarization (f) parallel to crack medium, i.e., one in which seismic waves travel
alignment and a slow wave, which is polarized at 90 to f. faster in one direction or with one polarization
The waves are separated with delay time dt than another (Fig. 1). In the Earth’s upper crust,
anisotropy is most likely to be caused either by
of the oceanic plate at the ridge. Since the 1970s, stress conditions preferentially aligning
improvements in computing power and memory microcracks parallel to the maximum compres-
and in seismic field observation have led to sive stress or by pervasive structural features. For
a greater understanding of the seismic anisotropy a near-vertical propagation direction, the shear
of the Earth at all levels and scales (Savage wave with the displacement in the plane of the
1999). Measurements of seismic anisotropy open cracks will travel faster than that crossing
have been used to detect fabric and stress in the the plane of the cracks, and so a fast shear wave S
Earth’s crust, flow in the upper mantle, topogra- with orientation f and a slow shear wave orthog-
phy of the core–mantle boundary, and differential onal to f, separated by a delay time dt, will be
rotation of the inner core. observed (Babuška and Cara 1991). Crack-
The measurement of seismic anisotropy has induced anisotropy has in some studies been con-
been found to be a proxy for determining the sidered a direct indicator of present-day stress,
direction of maximum horizontal compressive with f providing information about the orienta-
stress (SHmax) in the crust; applied stress can tion of SHmax and dt giving information about the
cause microcracks to preferentially open parallel strength of anisotropy and the amount of time that
to the maximum compressive stress, creating an the wave spent traversing the anisotropic
anisotropic medium with the fast direction parallel medium. Studies that combine independent stress
to SHmax (Fig. 1). The mechanism of aligned estimation methods with shear wave splitting
microcracks is thought to be the only one that results have found that strong geological fabric
allows seismic anisotropy to vary on observable or aligned structures rather than the maximum
stress can govern the observed f in some situa- Miller and Savage (2001) measured shear wave
tions. Both stress-induced and structure-related splitting from earthquakes in 1994 and 1998 and
anisotropies have been observed in the same observed a change in the dominant f spanning
regions and, in some cases, have been detected the time of the last major magmatic eruption in
at the same station (e.g., Zinke and Zoback 2000; 1995/1996. That study was extended by Gerst and
Johnson et al. 2011). Savage (2004), who used the same techniques
The subject of temporally varying anisotropy and an additional deployment of three-
is a hotly debated topic (e.g., Crampin and Pea- component seismometers in 2002 to observe fur-
cock 2008). Temporal variations in shear wave ther changes in f (Fig. 2). The results of both
splitting can take the form of a rotation in the fast studies were interpreted as being caused by
direction of anisotropy, an increase or decrease in a dike-shaped magma reservoir, or system of
the delay time, or a combination of both. The dikes, trending NE–SW. According to this
changes are thought to stem from perturbations model, the magma reservoir was pressurized
of the elastic properties of the crust due to crack before the eruption, producing a local stress
opening either by local concentration of shear field different from the regional stress field. This
stress or by a change in pore-fluid pressure. Mon- interpretation is favored above one in which dikes
itoring these changes can therefore provide cru- are intruded and solidify, causing a new structural
cial information about the state of stress and pore anisotropy, because of the lack of detectable
content around active volcanoes. However, spa- deformation and seismicity associated with dike
tial variations of anisotropy masquerading as intrusion. The model suggests that following the
temporal variations are one of the main caveats eruption the reservoir was less full and corre-
associated with the interpretation (Johnson spondingly less pressurized, meaning that the
et al. 2011). This can occur when the location of local stress returned to that of the surrounding
the source earthquakes changes or migrates region. The Gerst and Savage (2004) study
throughout the duration of the observation period suggested that the later changes in f were due
leading to the seismic waves experiencing differ- to repressurizing of the reservoir in response to an
ent anisotropic conditions due to heterogeneity in increase of magma in the system because f from
the anisotropic structure. Temporal variations in deep earthquakes displayed the regional trend,
anisotropy associated with major magmatic erup- while f from shallow earthquakes was oriented
tions, such as the 1995/1996 eruptions of Mount to the pre-eruption direction. The return of anisot-
Ruapehu, have been subject to doubts about the ropy to the pre-eruptive state also supports the
possibility of shear wave splitting variations interpretation of stress-controlled anisotropy as
being due to changing source earthquake loca- the intrusion of dikes would be a permanent
tion. Therefore, the methods must be fully under- change.
stood before changes in anisotropy with time can Johnson et al. (2011) investigated the spatial
truly be used as an eruption forecasting tool. To variations in anisotropy in more detail using
do this, one must first explore the spatial varia- a dense seismometer deployment in 2008, to
tions in anisotropy at volcanoes before examining compare future changes in anisotropy and to
the temporal changes. identify the regions and causes of past changes
Mt. Ruapehu volcano in the North Island of in anisotropy with more confidence. Johnson
New Zealand has been the subject of some of the et al. (2011) divided the mapped anisotropy into
most comprehensive shear wave splitting studies regions in which the fast polarizations agreed
of any volcano on Earth. Therefore, to illustrate with stress estimates from focal mechanism
the spatial versus temporal variation issue, inversions, suggesting stress-induced anisotropy,
anisotropy studies at Mt. Ruapehu will be and those in which the fast polarizations were
explored. Temporal variation of shear wave split- aligned with structural features such as faults or
ting at Mt. Ruapehu was investigated by Miller metamorphic fabric, suggesting structural anisot-
and Savage (2001) and Gerst and Savage (2004). ropy. Using this benchmark of anisotropy,
N N N
1994 Eruption 1998 2002
Seismic Anisotropy in Volcanic Regions, the dike system refilled, and the stress field in the anom-
Fig. 2 Schematic stress and anisotropy model after alous region was dominated by the dike again. The align-
Gerst and Savage (2004). In 1994, a pressurized dike ment of cracks was not as strong as in 1994, so the
system created a local stress field. In 1998, after the anisotropy in the anomalous region was not strong enough
eruption, when the dike system was depressurized, stress to affect fast directions from deep events
directions partially returned to the regional trend. In 2002,
Johnson and Savage (2012) examined temporal geometry, pore content, or stress regime. Ideally,
changes in shear wave splitting from 1994 to the procedure will be able to accurately process
2010. They observed a region of strong anisot- large quantities of three-component data in an
ropy centered on Mt. Ruapehu in 1995, the time efficient, unbiased, and objective way, without
of a major magmatic eruption, agreeing with operator intervention, while providing quantita-
Miller and Savage (2001) and Gerst and Savage tive evaluation of the uncertainties for each mea-
(2004), which was interpreted to be due to an surement of shear wave splitting.
increase in fluid-filled fractures during the erup- A manual method for analyzing shear wave
tion. They also observed strong anisotropy and splitting is that of direct inspection of the 3D S
a change in fast direction (~80 ) at Mt. Tongariro particle motion projected onto a horizontal
in 2008, which was initially interpreted to be due plane (hodograms). This method relies heavily
to a change in the geothermal system but was on observer judgment and therefore can produce
later the location of a small eruption in 2012. biased results and is inefficient for large datasets.
Other methods employ partially automated pro-
Measuring Shear Wave Splitting grams, where the user must pick a window
Many methods have been developed to measure around the shear wave arrival and/or evaluate
shear wave splitting in the Earth’s crust (e.g., the results. There are several approaches com-
Crampin and Gao 2006). The goal is to identify monly used to semiautomatically determine
the orientation of the fast split shear wave and the shear wave splitting parameters: the covariance
delay time between the fast and slow split shear matrix technique searches for nonzero eigen-
waves. These parameters can then be used to infer values to “unsplit” the shear wave; the cross-
rock properties such as crack distribution and correlation technique searches for the rotation
and time delay that yields the highest cross- strongly on the thickness and strength of
correlation between orthogonal components; the anisotropy of the layers, on the relative fast
aspect ratio technique searches for the rotation directions, and on the wavelength of the wave.
and time delay that yields the most linear particle If the shear wave has been sufficiently split in
motion with the maximum aspect ratio; and the the first layer that the fast and slow waves are
vector amplitude technique uses the maximum separated, then when it enters the second
amplitude in a time window to identify the split layer, which has a fast direction 2070 dif-
shear waves. Any degree of user interaction may ferent, both of the quasi-shear waves will be
introduce some subjectivity and is usually time- split again. In this case both waves will now
consuming, although many practices of auto- have the fast and slow directions of the second
mated quality control will result in loss of signif- layer. However, when the splitting from the
icant amounts of data. Completely automated first layer is weak so that the two quasi-shear
shear wave splitting analysis has proven elusive, waves are not more than one wavelength apart,
as the step that is difficult to automate is the both waves are still resplit, but the result is
picking of the shear wave arrival. Several a complex waveform that is difficult to inter-
methods boast full automation other than this pret but can still be meaningful. In general,
problematic step. In general, since phase arrivals shear wave splitting fast direction is represen-
are often picked for previous analyses such as tative of the last layer that the wave passed
hypocenter location, these automated methods through and the delay time can be approxi-
are preferable to the ones with a lot of user inter- mated as accumulating along the path.
action; more data can be evaluated, reducing the 3. That the anisotropic medium is localized
effect of loss of data through the quality control beneath the receiver: As seen in the previous
steps. point, in general, the anisotropy parameters
are representative of the last anisotropic
Interpreting Shear Wave Splitting: Some medium that the wave travelled through.
Assumptions Therefore, the assumption that the medium is
Most shear wave splitting observations are local to the station is usually justified. How-
interpreted under some assumptions, which are ever, in regions with heterogeneous aniso-
rarely completely true in the Earth: tropic structure, data at a single station will
not be consistent and the assumption will not
1. That the medium possesses hexagonal sym- be appropriate.
metry with a horizontal axis: The majority
of anisotropic rocks in the Earth have, or can Further discussions of these points can be
be approximated to have, hexagonal symme- found in Silver and Savage (1994), Savage
try. This is because the most common symme- (1999), and Johnson et al. (2011).
tries have patterns that do not differ
significantly for horizontal fast axis alignment Other Estimates of Seismic Anisotropy
and near-vertical incidence angles. The sim- Shear wave splitting analysis is becoming a very
plest models used to explain variations in two popular method for determining seismic anisot-
orthogonal directions are hexagonally sym- ropy in the crust. This is in part due to the abun-
metric models. Therefore, shear wave splitting dance of data and methods available and also to
is usually interpreted in terms of transverse the relative insensitivity to the source–receiver
anisotropy with a horizontal symmetry axis. geometry (other than deep enough earthquakes
2. That the anisotropic medium is in a single, for the rays to arrive within the shear wave win-
homogeneous layer: When a shear wave dow) and independence from the need for dense
passes through multiple anisotropic layers, networks of seismometers (see “▶ Seismic Net-
the observed splitting parameters depend work and Data Quality”). Another benefit of
shear wave splitting analysis is that, even though Variations in Shear Wave Splitting
the results can be averaged over multiple mea-
surements, they represent a snapshot of anisot- Spatial Variation of Shear Wave Splitting
ropy. There are, however, other methods for Most shear wave splitting results are plotted as
determining seismic anisotropy. Most of the rose diagrams (circular histograms) of fast direc-
other methods involve calculating anisotropic tion at the station at which the measurements
velocities using tomographic techniques (see were made (e.g., Fig. 2). This implicitly assumes
“▶ Seismic Tomography of Volcanoes”) with that the anisotropy is localized beneath the sta-
body or surface waves from active or passive tion. In many regions this may be appropriate;
sources, or using ambient seismic noise (see however, when there is lateral heterogeneity, the
“▶ Noise-Based Seismic Imaging and Monitor- rose diagrams will become scattered or multi-
ing of Volcanoes” and “▶ Seismic Noise”). modal. Johnson et al. (2011) detected distinct
Tomographic methods (see “▶ Seismic splitting parameters for measurements using
Tomography of Volcanoes”) require larger earthquakes from several different clusters in
amounts of data with relatively even coverage the region of Mt. Ruapehu. This backazimuthal
of sources and sensors, which (usually) take dependence implies that the f value obtained
longer to acquire, thereby rendering the methods from shear wave splitting analysis is highly
less practical for time-lapse investigations than dependent upon the path that the ray takes,
shear wave splitting investigations. Tomo- which has also been found in other regions. Fur-
graphic methods that are used to calculate seis- thermore, this suggests that the anisotropy
mic anisotropy are different to techniques that changes over shorter distances in the crust than
conduct tomographic inversions on shear wave is often observed for mantle anisotropy and that
splitting data. Anisotropy parameters are averaging f over the whole region may not be
derived from the inversions in the former case, appropriate. If the causes of the different regions
while the latter is an inversion of the anisotropy of anisotropy are known, it is easier to identify
parameters. and map the differences using shear wave split-
Inversions of body wave arrival times for ting analysis. If the crustal stresses or fabrics are
three-dimensional velocity structures are com- more complex or unknown, then it is more diffi-
mon practice at volcanoes using teleseismic cult to map and interpret heterogeneous anisot-
waves, local earthquake sources, and active seis- ropy, although a denser array of seismic stations
mic sources. These inversions can also account and a broad range of backazimuths increase
for 3D Vp azimuthal anisotropy, which is param- the likelihood of identifying the source of
eterized with a percent anisotropy and an orien- heterogeneity.
tation of the fast axis. Seismic anisotropy can be For time-lapse studies, it is important to miti-
detected using inversions of Love and Rayleigh gate the chance of spatial variations of shear S
surface waves from large earthquakes in the same wave splitting being erroneously interpreted as
manner. temporal variations. One way to do this would be
Surface waves constructed from cross- to map spatial variations in detail at a time when
correlations of ambient seismic noise can be there are no hypothesized temporal changes. This
inverted for 3D seismic velocity structure. Seis- becomes difficult, but essential, when the anisot-
mic anisotropy from ambient noise tomography ropy is very heterogeneous and the seismograph
can be calculated. These calculations are differ- array is less dense than the spatial changes. This
ent from the time-lapse studies that detect tem- section outlines some examples of shear wave
poral variations in isotropic seismic velocities splitting tomography studies, which attempt to
using ambient noise interferometry (see solve the problem of spatial variation.
“▶ Tracking Changes in Volcanic Systems with Shear wave splitting tomography is difficult
Seismic Interferometry”). because of the nonlinear effect of multiple layers
strength of anisotropy (s/km)

−39°00⬘00⬙ 0.025
−39°10⬘00⬙ 0.020
−39°20⬘00⬙ 0.015
0.010
−39°30⬘00⬙
0.005
−39°40⬘00⬙
Stations
km
0 5 10
0.000
175°00⬘00⬙ 175°10⬘00⬙ 175°20⬘00⬙ 175°30⬘00⬙ 175°40⬘00⬙ 175°50⬘00⬙ 176°00⬘00⬙
Seismic Anisotropy in Volcanic Regions, shaded area shows the limit of statistical significance
Fig. 3 Delay time tomography from the inversion of calculated from the model variance matrix, white inverted
shear wave splitting data at Mount Ruapehu Volcano, triangles are seismic stations (After Johnson et al. (2011))
New Zealand. Warm colors indicate strong anisotropy,
of anisotropy on waveforms. This is different to constructed with nodes regularly spaced between
travel time tomography, in which the travel time each earthquake and station. This grid was then
of a wave is often linearly related to the velocity treated as a new dataset, and f for each node
structure of the media it has passed through. within a polygon, or within a box of a regular
Because of this difficulty, many techniques treat lattice, was averaged. This created average f
f independently from dt. Three-dimensional values at regular intervals that could be denser
tomographic inversions can be carried out on than the station spacing, hereby identifying spa-
the delay time data to characterize fracture dential trends in fast directions, but not accounting
sity distribution. This tomography uses only dt for delay times. Johnson et al. (2011) adapted
from local earthquakes to investigate anisotropy a combination of the above methods to map the
strength in the crust. In this way, regions of high heterogeneous seismic anisotropy field around
anisotropy can be identified, but information Mt. Ruapehu in New Zealand (Fig. 3). To con-
regarding fast directions is not accounted for. strain the locations of high anisotropy, they
Audoine et al. (2004) presented a simple conducted a two-dimensional tomographic inver-
method of 2D spatial averaging to examine het- sion on the delay time estimates. They then used
erogeneous anisotropy in the crust. A grid was a spatial averaging technique similar to that of
Audoine et al. (2004), but with the fast polariza- crust, which would not change with changing
tions weighted according to the strength of anisot- stress. However, it is also clear that anisotropy
ropy calculated from the tomography. The method due to stress-aligned fluid-saturated microcracks
uses a quad-tree gridding system to enable higher can change with time, as has been demonstrated
resolution where the data permit and couples the in industry when small changes associated with
two shear wave splitting parameters, even though injection and removal of fluids from reservoirs
they are not used in a joint inversion. were examined. The use of similar earthquakes
Abt and Fischer (2008) carried out full 3D (i.e., those that have the same source mechanism
shear wave splitting tomography for the mantle. and location) for shear wave splitting analysis
The method parameterized the mantle as a 3D helps to eliminate some of the discrepancies
block model of crystallographic orientations. with interpretation. Johnson et al. (2010) used
Nonlinear properties of shear wave splitting shear wave splitting analysis, a multiplet of
were accounted for by applying the inversion 25 similar earthquakes and double-difference
iteratively and recalculating partial derivatives relocation to examine temporal variations in seis-
after each iteration. Using this method, Abt and mic properties prior to and accompanying mag-
Fischer (2008) modeled an idealized subduction matic activity associated with the 2008 eruption
zone with uniform stations and sources. When of Okmok Volcano. They found a general change
applied to real data, they found that the geometry in f but could not rule out dependence on
of stations and observed seismicity in the backazimuth and no significant change using the
Nicaragua–Costa Rica subduction zone yielded multiplet. Using earthquakes originating from the
partial to good resolution. This method also has subducted slab in order to reduce the effect from
the potential to be applied to crustal studies such changing paths, several modes of f were identi-
as active volcanoes. fied, relating to the anisotropy of the mantle
wedge, regional stress direction, and local stress
Temporal Variation of Shear Wave induced by the pressurization and depressuriza-
Splitting in the Crust tion of the magma reservoir. These modes were
The temporal variation of shear wave splitting found to have different prominence at different
and its interpretation is highly controversial. times throughout the eruptive cycle.
The main point of dispute is whether the accumu- Once again returning to the example of
lation of stress before significant events such as Mt. Ruapehu, Miller and Savage (2001) and
eruptions or large earthquakes allows the time Gerst and Savage (2004) observed general
and magnitude of impending events to be stress changes in seismic anisotropy throughout the
forecast through shear wave splitting monitoring. eruptive cycle. Johnson et al. (2011), however,
The effects have been reported with hindsight showed that seismic anisotropy around
before at least 15 earthquakes. Crampin and Pea- Mt. Ruapehu is heterogeneous and that averaging S
cock (2008) present a summary of observations the whole region is not appropriate. Therefore,
of temporal variations in shear wave splitting Keats et al. (2011) used seismicity generated
attributed to stress-aligned fluid-saturated from a consistently active area of seismicity
microcracks. However, alternative interpreta- about 20 km to the west of Mt. Ruapehu. Shear
tions of these observations of temporal changes wave splitting results revealed a decrease in delay
suggest there may be observer bias in data selec- time in the 2006–2007 eruption period and
tion, unsound statistical analyses, misinterpreta- a significant variation in the fast shear wave
tion of spatial variation, and lack of correlation polarization in the same time period (Fig. 4).
with other stress-determining factors/correlation These changes were attributed to an increase in
with structural evidence. pore-fluid pressure in the region due to fluid
Clear evidence has been obtained that crustal movement, and it was suggested that this fluid
shear wave splitting can vary over short distances movement may be associated with the eruptions
and can be caused by structural features in the in 2006 and 2007.
0.40 2.00
0.30 1.50
b−value
δt (s)
0.20 1.00
0.10 0.50
0.00 0.00
1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
90
1 t 2 t 3 t 4
60
30
φ(°)
−30
−60
−90
Seismic Anisotropy in Volcanic Regions, by the red horizontal bars with 90 % confidence interval
Fig. 4 Moving average plot of fast polarization (f) and (dashed red lines). The times of the two phreatomagmatic
delay time (dt) using earthquakes within the Erua swarm eruptions that occurred are also marked with grey bars.
at station FWVZ at Mount Ruapehu, New Zealand. Indi- Rose diagrams of f are displayed in their respective time
vidual measurements for f and dt are displayed in light periods. The b-value for the Erua swarm catalogue is also
blue and 10-point moving averages are displayed in dark plotted against time in black at the top using a window of
blue. The error bars indicate 95 % confidence intervals. 40 events and an eight-event overlap. Dashed black lines
The four time periods, marked by the numbers 1–4, and indicate 95 % confidence interval (After Keats
three transition zones, marked by a t, are indicated with et al. (2011))
vertical red lines and the mean for each period are shown
Temporal variation of shear wave splitting wave splitting around volcanoes (Fig. 5). This is
parameters as a stress indicator has been due, in part, to the generally noisy waveforms and
employed at several other volcanoes. While complicated interpretation of such observations
most studies concentrate on either stress-induced when taking into account heterogeneity and com-
anisotropy or structurally dominated anisotropy, plex stress regimes. It should be noted that there
several studies have found that the dominant is significant literature about shear wave splitting
mechanism can change between the two (Keats in the mantle beneath active volcanoes because
et al. 2011; Johnson and Poland 2013) and that these regions are invariably of interest tectoni-
more than one mechanism can be dominant over cally, but this entry focuses on crustal studies
short distances, potentially creating an apparent only.
temporal change that is actually an artifact of
changing earthquake location (Zinke and Zoback
2000). Relating Shear Wave Splitting to Other
Even though we have seen that shear wave Observations
splitting analysis can be used as an indicator of
stress and of fluid saturation in the crust, surpris- Shear wave splitting analysis has been proven to
ingly few studies have been conducted on shear be a useful indicator of stress and of fluid
Seismic Anisotropy in Volcanic Regions, Fig. 5 Map indicates that temporal variations were observed, red indi-
showing volcanoes at which shear wave splitting investi- cates that no temporal variations were observed
gations have been carried out. Green volcano symbol
saturation in the crust but can also be caused by volcano monitoring. For this reason, many shear
structures such as macroscopic fractures or min- wave splitting studies have been designed to
eral alignment. At volcanoes, the generally noisy investigate temporal changes in anisotropy dur-
waveforms, complicated heterogeneity, and coming a volcanic crisis (e.g., Savage et al. 2010;
plex stress regimes can make interpretation of Johnson et al. 2010; Roman and Gardine 2013).
shear wave splitting difficult. For this reason, Bianco and Zaccarelli (2009) calculated
shear wave splitting investigations are often a background seismic anisotropy at Mount Vesu-
coupled with other stress or strain indicators and vius of 4 %, but at times of seismic crisis, the
structural geology data to minimize the ambigu- average anisotropy was 8 % and the fast direc-
ity in the interpretation. tions tended to flip by 90 .
As noted above, however, heterogeneity in the
Volcanic Seismicity, Fault Plane Solutions, anisotropic media must be carefully considered,
and b-Value especially when there is a migration of seismicity
Studies of shear wave splitting in areas of devel- associated with magma movement. These S
oped industry may use direct indicators of the migrating earthquakes are used as sources of
stress field in the crust, such as borehole break- shear wave energy, but temporal variations
outs, for comparison with shear wave splitting observed in the anisotropy measurements could
results. However, few active volcanoes have be due to spatial variations.
these clues, so other stress monitoring techniques VT earthquakes contain additional informa-
must be employed. tion about the state of stress in the crust. Fault
The occurrence of volcano-tectonic (VT) (see plane solutions (FPS) provide information about
“▶ Volcano-Tectonic Seismicity of Soufriere the orientation of the fault plane that slipped, as
Hills Volcano, Montserrat”) earthquakes is an well as the slip vector for each earthquake. From
immediate indicator of differential stress in the the FPSs, the pressure and tension axes can be
crust around a volcano. Changes in the stress field identified, which give information about the
may trigger more VTs, the detection of which is stress regime. The FPS P-axis azimuth, which
one of the most successful tools of seismic represents the orientation of the principal
compressional axis of the moment tensor, may in comes from the Gutenberg–Richter law and is
reality differ significantly from the orientation of a frequency magnitude (see “▶ Frequency-
maximum compressive stress, depending on the Magnitude Distribution of Seismicity in Volcanic
orientation of preexisting planes of weakness Regions”) relation. For crustal earthquakes,
(faults). This is sometimes countered by calculat- b-values are typically ~ 1 for tectonic earth-
ing large numbers of FPS P-axis azimuths to quakes, though they tend to be higher in volcanic
highlight the major trends that most likely repre- regions (see “▶ Frequency-Magnitude Distribu-
sent the actual orientations of SHmax (e.g., Roman tion of Seismicity in Volcanic Regions”). The
and Cashman 2006; Roman and Gardine 2013). b-value has been related to physical properties
Another step can be taken by inverting for the full such as stress, material homogeneity, and pore
stress tensor using the FPS results (e.g., Johnson pressure and can therefore be useful in compari-
et al. 2011). The true direction of SHmax can then son with seismic anisotropy measurements to
be compared with shear wave splitting parame- determine changes in physical properties around
ters to jointly interpret the results. volcanoes. Keats et al. (2011) used seismicity
The orientations of FPS of VT earthquakes from a discrete cluster of earthquakes near
have been observed to display systematic Mt. Ruapehu in New Zealand to compare shear
changes related to episodes of magmatic activity wave splitting and b-values over 12 years (Fig. 4).
at several volcanoes (Roman and Cashman The joint analysis allowed them to attribute tem-
2006). The pressure axes of FPS have been poral changes to an increase in pore-fluid pres-
observed to rotate orthogonal to the dominant sure due to fluid movement associated with
regional stress orientation in some cases, indicat- eruptive activity.
ing that these earthquakes may result from dike
inflation in the direction of minimum compres- Attenuation, Vp / Vs Ratio, and Noise Cross-
sive stress (Roman and Cashman 2006). Roman Correlations
and Gardine (2013) investigated shear wave split- Perturbations of the elastic properties of the crust
ting and earthquake fault plane solutions at around volcanoes are often linked to crack open-
Redoubt volcano around the time of the March ing either by local concentration of shear stress or
2009 eruption. They found that an approxi- by an increase of pore-fluid pressure. Shear wave
mately ~ 90 change in the polarization of shear splitting analysis is one way to monitor these
wave splitting fast polarization accompanied the properties, but there are other techniques that
earliest signs of seismic unrest in 2008 and con- can provide additional information about the
tinued through the eruption before diminishing in state of stress or pore content. Earthquake coda
2009. A similar change in the orientation of fault (the part of a high-frequency seismogram follow-
plane solutions occurred 18–48 h prior to the ing the P and S waves) is generated by random
eruption onset on March 23, 2009, but almost scattering processes in the crust. The coda decay
2 months after a strong increase in the rate of parameter (or Q 1
c ) is often used to parameterize
shallow VT earthquakes. The shear wave split- characteristics of the medium, and temporal
ting and FPS results suggested a protracted period changes in Q c
1
have been linked to crack open-
of slow magma ascent followed by a short period ing (Del Pezzo et al. 2004).
of rapidly increasing magma pressurization Increased gas content in pores and cracks has
beneath the volcano. Roman and Gardine (2013) the effect of lowering P-wave speed due to the
found that after the eruption, both shear higher fluid compressibility, but not significantly
wave splitting fast polarization and fault plane affecting the shear modulus and hence S-wave
solution pressure axes had a direction more con- speed. The ratio between P-wave velocity (Vp)
sistent with the regional stress than before the and S-wave velocity (Vs), Vp/Vs, is therefore use-
eruption. ful for characterizing pore-fluid content.
The b-value of a cluster of earthquakes is often Gas-enriched pore space has been reported to
used to describe the size distribution. The b-value affect Vp/Vs above magmatic intrusions and has
been combined with shear wave splitting studies controlled. An anticorrelation between geodetic
to characterize pore content at volcanoes measurements and shear wave splitting results
(Unglert et al. 2011; Johnson and Poland 2013). was observed at Kīlauea Volcano, implying that
The perturbations of the elastic properties of some volcanic process was affecting both mea-
the crust around volcanoes caused by changes in surements but that the mechanism was different
stress or pore fluid can be monitored continuously to the traditional interpretation of a pressurizing
using cross-correlations of ambient seismic noise magma reservoir. Johnson and Poland (2013)
(see “▶ Noise-Based Seismic Imaging and Mon- used these observations, combined with Vp/Vs
itoring of Volcanoes” and “▶ Seismic Noise”). ratio to interpret these changes as being due to
Monitoring of Q 1
c , Vp/Vs ratio, and isotropic increased gas flux.
velocity from seismic noise cross-correlations
therefore naturally complements shear wave Geology
splitting investigations as the mechanisms of Numerous investigations have found that in some
change are so similar (Del Pezzo et al. 2004). regions, the fast direction of anisotropy is oblique
to the direction of maximum horizontal compres-
Deformation sive stress. In these cases, other clues for the
Geodetic techniques that determine changes in cause of anisotropy are sought and the anisotropy
strain, such as GPS and InSAR, can provide is frequently found to align preferentially with
some of the stress tensor elements and can there- macroscopic structures at the surface. Macro-
fore be useful to compare with shear wave split- scopic fractures are often aligned because they
ting investigations. Studies such as those by are caused by faulting in a regional stress field.
Savage et al. (2010) and Unglert et al. (2011) These macroscopic structures also cause anisot-
use strain data to determine that the anisotropy ropy and shear wave splitting that often aligns
is due to stress-oriented microcracks in the upper with SHmax. Exceptions arise when strike-slip
crust. At Mount Asama, Savage et al. (2010) cor- faults initially align at 45 to the direction of
related GPS baseline length measurements with SHmax, faults are a product of a paleostress, or
shear wave splitting measurements to analyze the rock has been deformed since the faulting.
stress changes accompanying the eruption in Velocity anisotropy is also strongly dependent on
2004. They found that the best model from the rock fabrics and metamorphic rocks with distinct
GPS analysis of a vertical dike and conduit also foliations can have anisotropies of up to 20 %
fit the shear wave splitting measurements, as did even in the absence of cracks (Babuška and Cara
the temporal variations. From this, a crack aspect 1991, and references therein). This is mainly due
ratio of 2.6 105 was calculated and to the preferred alignment of intrinsically aniso-
a differential horizontal stress of 6 MPa at 3 km tropic minerals such as biotite and hornblende.
depth was inferred. Unglert et al. (2011) Geological textures can be identified through S
performed shear wave splitting analyses on analysis of in situ rocks or oriented drill cores.
local earthquakes around Aso Volcano between Shear wave splitting that is caused by structural
2001 and 2008 and compared the results to strain features is unlikely to change over time periods of
from GPS measurements in the area. They investigation. However, due to changing condi-
observed, using clusters with relatively stable tions in stress or pore content, the dominant
epicenters, that two stations showed mechanism of anisotropy can change, rendering
a significant change in f in 2004–2005. Models a temporal variation in shear wave splitting
from seismic tomography and receiver functions parameters. Johnson and Poland (2013) investi-
were found to fit both the anisotropy and strain gated shear wave splitting changes at Kīlauea
measurements. Other studies have used the obser- Volcano associated with the onset of the summit
vations of no correlation between geodetic mea- eruption in 2008. They found that the orientation
surements and shear wave splitting results to infer of fast shear waves at Kīlauea was usually con-
that the mechanism of anisotropy is structurally trolled by structure, but in 2008 showed changes
with increased SO2 emissions preceding the start Vargas-Bracamontes and Neuberg (2012) found
of the summit eruption. that the stress patterns gradually approach those
corresponding to the absence of a regional stress
Tectonic Versus Magmatic Stress-Controlled field. Therefore, in cases with extremely high
Anisotropy magma pressures, such as before large explosions
Studies that have recorded rotations in the fast at Soufriere Hills Volcano, Montserrat, the
direction of anisotropy due to a localized pertur- regional stress field may be omitted in numerical
bation in stress often relate the background stress models.
regime to the regional tectonics of the area. Rota-
tions of f by 90 are commonly observed, but
several mechanisms for the rotation have been The Future of Seismic Anisotropy in
proposed. The 90 rotations should be treated Volcanic Regions
carefully because cycle skipping, when the
match of the fast and slow waveforms has Using seismic anisotropy to measure stress at
a factor of T/2 ambiguity where T is the dominant active volcanoes holds enormous potential as
period, can lead to an error of 90 in the recorded a monitoring and eruption forecasting tool.
fast direction and a false delay time. When the Changes in seismic anisotropy associated with
rotations are real, Crampin et al. (2002) suggest volcanic activity have already been detected;
that they are “flips” caused by wave propagation however, interpretation of seismic anisotropy
through cracks containing fluids at high pore- observations is usually qualitative, and
fluid pressures. However, a 90 rotation would researchers struggle to use the results to quantify
also be expected due to a dike intrusion (Gerst the magnitude of stress variations and the cause.
and Savage 2004). A dike will exert pressure on Eruption forecasting is increasingly evolving
the surrounding rock, generating a local stress from empirical pattern recognition to forecasting
field that is superimposed on the regional stress based on models of the underlying dynamics. For
field. The stresses of such an elongated structure shear wave splitting monitoring to be beneficial,
are mainly oriented perpendicular to the strike physical models must be developed that can
axis. When the pressure in the dike system is explain not only changes in general trends but
high enough, the generated stress field locally also the scatter in the data and anomalous obser-
reorients the principal stresses as well as the vations. Central to this are the links between
local crack alignment. The pattern of stress per- stress changes and fracture or crack compliance
turbation around an intrusion can be complex, in the country rock and the role that fluids in
however, and the interaction with the regional, cracks play, particularly the effects of hydrother-
or tectonic, stress has been the subject of several mal circulation, pore pressure, and gas flux.
studies (e.g., Roman and Cashman 2006; Vargas- Another limitation is that the majority of current
Bracamontes and Neuberg 2012). Vargas- research seeks to interpret only f or only dt
Bracamontes and Neuberg (2012) found that in observations, but the two parameters are inher-
the presence of a dominant regional stress field, ently linked. Models used to explain shear wave
the stress perturbation from an intrusion will be splitting observations must predict both
negligible. As the pressure from the intrusion parameters.
increases, both the regional and the local stress To quantify the response of seismic anisotropy
field will coexist. This phenomenon was to pre-, co-, and post-eruption subsurface magma
observed at Okmok Volcano, Alaska, when the movement, shear wave splitting data should first
shear wave splitting results displayed a mode be used to map the areas affected by changing
corresponding with the regional stress direction stress and identify the mechanism of seismic
and one corresponding with the pressurization anisotropy in areas that do not have stress-
of the magma reservoir (Johnson et al. 2010). controlled anisotropy. Methods have recently
As the pressure is progressively increased, been developed for the inversion of
geomechanical parameters such as crack size, The changes are thought to stem from perturba-
geometry, density, and content (e.g., Wuestefeld tions of the elastic properties of the crust due to
et al. 2012). Monitoring of these properties over crack opening either by local concentration of
time will not only indicate the occurrence of shear stress or by a change in pore-fluid pressure.
changes but will elucidate the nature of the sub- There is mounting evidence, however, that the
surface changes. dominant mechanism for seismic anisotropy can
Increasingly realistic numerical and experi- switch between a static condition, such as aligned
mental models of the fluid dynamics and fractures in fault zones, and a dynamic process,
elastodynamics are becoming possible. Integra- such as compressive stress causing aligned
tion of multidisciplinary data within these models microcracks to dilate. In areas where there are
will enable greater understanding of the underly- strong changes in maximum compressive stress
ing mechanisms and may be used to calculate the direction and magnitude on observable time-
shear wave splitting and other important param- scales, such as at active volcanoes, seismic
eters for different volcanic scenarios. Ultimately, anisotropy analysis has proven a valuable tool
it may be possible to use information contained in when combined with ground deformation or
seismic anisotropy to monitor subsurface magma other seismological observations for interpreta-
movement and forecast changes in a volcano’s tion of volcanic processes such as magma migra-
behavior by establishing the characteristic stress tion. By inverting shear wave splitting data for
field response for a given volcano, or through geomechanical parameters and integrating the
a deeper understanding of the complex relation- results with numerical and experimental models,
ships between seismic anisotropy, local crustal it may be possible to monitor subsurface magma
stresses, and the physical mechanisms of movement.
magma migration.
References
Summary
Abt DL, Fischer KM (2008) Resolving three-dimensional
Seismic anisotropy is the variation of seismic anisotropic structure with shear wave splitting tomog-
wave speed with direction. Shear wave splitting raphy. Geophys J Int 173(3):859–886. doi:10.1111/
j.1365-246X.2008.03757.x
occurs when a shear wave travels through
Audoine E, Savage MK, Gledhill K (2004) Anisotropic
a seismically anisotropic medium. Temporal var- structure under a back arc spreading region, the Taupo
iations in shear wave splitting can take the form Volcanic Zone, New Zealand. J Geophys Res
of a rotation in the fast direction of anisotropy, an 109(B11):305. doi:10.1029/2003JB002932
Babuška V, Cara M (1991) Seismic anisotropy in the
increase or decrease in the delay time, or
Earth, vol 10. Klewer Academic, Dordrecht
a combination of both. The measurement of seis- Bianco F, Zaccarelli L (2009) A reappraisal of shear wave S
mic anisotropy has been found to be a proxy for splitting parameters from Italian active volcanic areas
determining the direction of maximum horizontal through a semiautomatic algorithm. J Seismol
13(2):253–266. doi:10.1007/s10950-008-9125-z
compressive stress in the crust; applied stress can
Crampin S, Gao Y (2006) A review of techniques for
cause microcracks to preferentially open parallel measuring shear-wave splitting above small earth-
to the maximum compressive stress, creating an quakes. Phys Earth Planet In 159(1–2):1–14.
anisotropic medium with the fast direction paral- doi:10.1016/j.pepi.2006.06.002
Crampin S, Peacock S (2008) A review of the current
lel to the maximum horizontal compressive understanding of seismic shear-wave splitting in the
stress. The mechanism of aligned microcracks is Earth’s crust and common fallacies in interpretation.
thought to be the only one that allows seismic Wave Motion 45(6):675–722. doi:10.1016/j.
anisotropy to vary on observable timescales, and wavemoti.2008.01.003
Crampin S, Volti T, Chastin S, Gudmundsson A, Stefans-
temporal changes are traditionally interpreted as
son R (2002) Indication of high pore-fluid pressures in
stemming from variations in the stress field due a seismically-active fault zone. Geophys J Int 151(2):
to large earthquakes or magmatic intrusions. F1–F5. doi:10.1046/j.1365-246X.2002.01830.x
2706 Seismic Behavior of Ancient Monuments: From Collapse Observation to Permanent Monitoring
Del Pezzo E, Bianco F, Petrosino S, Saccorotti G (2004) Unglert K, Savage MK, Fournier N, Ohkura T, Abe Y (2011)
Changes in the coda decay rate and shear-wave split- Shear wave splitting, vP/vS, and GPS during a time of
ting parameters associated with seismic swarms at enhanced activity at Aso caldera, Kyushu. J Geophys
Mt. Vesuvius, Italy. Bull Seismol Soc Am Res-Sol Ea 116(B11):203. doi:10.1029/2011JB008520
94(2):439–452. doi:10.1785/0120030141 Vargas-Bracamontes D, Neuberg J (2012) Interaction
Gerst A, Savage MK (2004) Seismic anisotropy beneath between regional and magma-induced stresses and
Ruapehu Volcano: a possible eruption forecasting tool. their impact on volcano-tectonic seismicity.
Science 306(5701):1543–1547. doi:10.1126/ J Volcanol Geotherm Res 243–244:91–96.
science.1103445 doi:10.1016/j.jvolgeores.2012.06.025
Hess HH (1964) Seismic anisotropy of the uppermost Wuestefeld A, Verdon JP, Kendall J-M, Rutledge J,
mantle under oceans. Nature 203(494):629. Clarke H, Wookey J (2012) Inferring rock fracture
doi:10.1038/203629a0 evolution during reservoir stimulation from seismic
Johnson JH, Poland MP (2013) Seismic detection of anisotropy. Geophysics 76(6):WC157–WC166.
increased degassing before Kīlauea’s 2008 summit doi:10.1190/GEO2011-0057.1
explosion. Nat Commun 4:1668. doi:10.1038/ Zinke JC, Zoback MD (2000) Structure-related and stress-
ncomms2703 induced shear-wave velocity anisotropy: observations
Johnson JH, Savage MK (2012) Tracking volcanic and from microearthquakes near the Calaveras Fault in
geothermal activity in the Tongariro Volcanic Centre, central California. Bull Seismol Soc Am
New Zealand, with shear wave splitting tomography. 90(5):1305–1312. doi:10.1785/0119990099
J Volcanol Geotherm Res 223–224:1–10. doi:10.1016/
j.jvolgeores.2012.01.017
Johnson JH, Prejean S, Savage MK, Townend J (2010)
Anisotropy, repeating earthquakes, and seismicity
associated with the 2008 eruption of Okmok volcano, Seismic Behavior of Ancient
Alaska. J Geophys Res 115, B00B04, doi:10.1029/ Monuments: From Collapse
2009JB006991. (Figure 1)
Johnson JH, Savage MK, Townend J (2011)
Observation to Permanent
Distinguishing between stress-controlled and struc- Monitoring
tural shear wave anisotropy at Mount Ruapehu vol-
cano, New Zealand. J Geophys Res-Sol Ea 116:B12. Vincenzo Gattulli and Francesco Potenza
doi:10.1029/2011JB008308
DICEAA - Dipartimento di Ingegneria Civile,
Keats BS, Johnson JH, Savage MK (2011) The Erua
earthquake cluster and seismic anisotropy in the Edile-Architettura, Ambientale, CERFIS -
Ruapehu region, New Zealand. Geophys Res Lett Centro di Ricerca e Formazione in Ingegneria
38(L16):315. doi:10.1029/2011GL049014 Sismica, University of L’Aquila, L’Aquila, Italy
Miller V, Savage M (2001) Changes in seismic anisotropy
after volcanic eruptions: evidence from Mount
Ruapehu. Science 293(5538):2231–2233.
doi:10.1126/science.1063463 Synonyms
Roman DC, Cashman KV (2006) The origin of volcano-
tectonic earthquake swarms. Geology 34(6):457.
Masonry failure; Monumental structures; Seis-
doi:10.1130/G22269.1
Roman DC, Gardine MD (2013) Seismological evidence mic damage; Seismic vulnerability; Structural
for long-term and rapidly accelerating magma pressur- analysis; Structural health monitoring
ization preceding the 2009 eruption of Redoubt Vol-
cano, Alaska. Earth Planet Sci Lett 371–372:226–234.
doi:10.1016/j.epsl.2013.03.040
Savage MK (1999) Seismic anisotropy and mantle defor- Introduction
mation: what have we learned from shear wave splitting?
Rev Geophys 37(1):65–106. doi:10.1029/98RG02075 The earthquake occurred on April 6, 2009, in
Savage MK, Ohminato T, Aoki Y, Tsuji H, Greve SM
Italy at L’Aquila has been a catastrophic event
(2010) Stress magnitude and its temporal variation at
Mt. Asama Volcano, Japan, from seismic anisotropy for both the city and the University of L’Aquila
and GPS. Earth Planet Sci Lett 290(3–4):403–414. (Ceci et al. 2010). Since then, numerous scientific
doi:10.1016/j.epsl.2009.12.037 activities have accompanied both the immediate
Silver PG, Savage MK (1994) The interpretation of shear-
recovery and the long reconstruction program in
wave splitting parameters in the presence of two aniso-
tropic layers. Geophys J Int 119(3):949–963. different fields of earthquake science and engi-
doi:10.1111/j.1365-246X.1994.tb04027.x neering. In particular, the development of new
Seismic Behavior of Ancient Monuments: From Collapse Observation to Permanent Monitoring 2707
systems and technologies for both understanding sequence showed variability in ground motion
and enhancing the structural behavior of signifi- amplification throughout the city. General con-
cant historical palaces and churches, densely clusions highlight that the building stock of the
populating the L’Aquila city center, has attracted city suffered different levels of damage that can
the attention of specialists and researchers com- be partially explained by the combination of
ing from a broadband spectrum of scientific dis- building vulnerability and surface geology. In
ciplines. This rich multidisciplinary approach has particular, the observation of damages in RC
permitted the development of a new paradigm in buildings highlighted that the few collapse cases
the preparatory works necessary for planning the (<1 % of the whole RC buildings stock) are
retrofitting and the reconstruction of ancient mainly the consequence of structural deficiency,
buildings. Therefore, the impact of new technol- inadequate structural layout, and maybe site
ogies in the area of observation, survey, testing, effects (Decanini et al. 2012). The majority of
modeling, restoration, retrofitting, and monitor- RC construction survived the earthquake with
ing of historic constructions and monuments, dif- minor damages or with damages to nonstructural
fusely used at L’Aquila, merits to be discussed elements. The extensive observed damage to
and briefly reported, here, as starting point for infill walls that can be associated to a significant
future novel findings and process optimization. amount of energy dissipation prevented the
In this respect, it appears reasonable to men- development of plastic hinges in RC columns,
tion the geophysical and geotechnical research thus allowing the maintenance of structural integ-
efforts conducted to characterize various aspects rity. Concerning the seismic demand, the study
of what has been observed and measured during highlighted the noticeable difference between the
foreshock, mainshock, and aftershock in the elastic and inelastic demand, indicating that the
L’Aquila valley. Starting from a historical seis- displacement demand is moderate, associable to a
mological study regarding the earthquakes that non-extraordinary event, even if, in several cases,
took place in the area of L’Aquila (central Italy) it overcomes the threshold of serviceability limit,
from the ancient Roman period to the late Middle producing economic losses and potential injuries
Ages, the persistence and magnitude of earth- and deaths (Ceci et al. 2013).
quakes demonstrated to have a strong bearing Much more complex than the damage scenario
on the economy and culture of the communities affecting the RC buildings is the case of technical
(Guidoboni et al. 2012). Furthermore, under- observations made on which occurred specifi-
standing the rupture slip distribution for the cally in historic constructions and monuments at
2009 L’Aquila mainshock has allowed for the L’Aquila (D’Ayala and Paganoni 2011; Da Porto
development of a complex high-resolution 3D et al. 2012; Indirli et al. 2013) and in the small
FE model incorporating surface topography and historical centers of the valley (D’Ayala and
rheological heterogeneities, deduced from real Paganoni 2011; Carocci 2012; Indirli S
tomography (Volpe et al. 2012). The 3D et al. 2013). These studies show that within the
approach provides a more concentrated and local- historic city center of L’Aquila, the number of
ized slip distribution on the rupture plane, total collapses observed was minor, while the
evidencing a single area of high-slip release SE proportions of partial collapses of the upper
of the hypocenter. Moving from the underground stories were perhaps greater than would have
level to the ground level, scientific studies have been expected given the level of shaking
tried to correlate the distribution and the severity (D’Ayala and Paganoni 2011). In a minority of
of the damage with the geological setting of the cases, total collapse of the facade was observed,
area, taking into account the characteristics of the and these were usually associated to substantial
building stock through time (Tertulliani alteration of the original structure. Although the
et al. 2012). Strong-motion records and ambient number of undamaged masonry buildings is very
noise measurements taken soon after the modest, the majority suffered either minor
mainshock and during the entire aftershock damage or repairable structural damage.
This situation allowed on one hand to identify of the disaggregated masonry, made by arbitrary
and recognize collapse mechanisms and on the materials, unable to resist horizontal forces. Sev-
other hand to better correlate ultimate capacity to eral damage/collapse types have been encoun-
strong-motion characteristics and hence lends tered, classified, and associated with the
itself to a more detailed in situ survey and ana- simplified mechanisms foreseen by the Italian
lytical assessment. In the specific case of MEDEA procedure (Indirli et al. 2013).
churches, a statistical analysis has defined the The needs of simplified performance-based
percentages of possible mechanisms related to assessment procedures in order to support the
local construction practices and architectural fea- interpretation of observed damage is evidenced
tures of the region. For each possible mechanism, in all recent earthquakes, such as those occurred
the relative percentage of activation and calcu- in L’Aquila (2009), Christchurch (2010–2011),
lated average damage allowed some conclusions and Emilia Romagna (2012), which have caused
to be drawn on the seismic vulnerability of the not only a significant death toll and huge eco-
various structural and nonstructural elements nomic losses but also heavy damage to the world-
making up a church. These conclusions may be wide cultural heritage (Parisi and Augenti 2013).
regarded as widely valid and, if further corrobo- Earthquake damage to monumental constructions
rated by other post-earthquake surveys, may be has been discussed and critical issues affecting
used to define hierarchies of interventions, also the seismic response of historic masonry struc-
on non-damaged structures (Da Porto et al. 2012). tures are identified such as masonry quality, con-
The study showed that it would be useful to nections among structural elements, diaphragm
introduce weights to damage mechanisms when flexibility, out-of-plane resistance of masonry
calculating damage indices evaluated through walls, structural irregularities, wrong retrofit
survey form compilation, which is still a power- interventions, and earthquake ground motion
ful tool in the management of emergency and characteristics.
post-emergency phases. Furthermore, it has Moreover, since 2009, the reconstruction idea
been pointed out that the recorded high-peak of the city of L’Aquila and its surrounding was a
ground accelerations reached high values, both main goal, primarily of the citizens and, with
in the horizontal and vertical directions. The them, of the Italian Government. However, dur-
simultaneous combination of these acceleration ing every catastrophic earthquake, Italian citizens
components, has produced a reduction of the and institutions seem to face a series of typical
masonry shear capacity, especially in the upper problems due to bureaucratic burdens, corrup-
floors. Consequently, the widespread vulnerabil- tion, excess legislation, political clientelism,
ity of unreinforced masonry buildings resulted in poor performance, and weak public control, and
extensive damage, collapse, and victims in a they are the combined effects of cultural, institu-
medium-size city. The underestimation of the tional, and political factors at the governance
earthquake actions is a considerable drawback level (Özerdem and Rufini 2013).
that can be overcome only by integrating the Notwithstanding several difficulties, numer-
study of earthquake scenarios with more accurate ous applied research activities have been
methodologies. Moreover, with regard to the conducted in the area of new and sophisticated
structural vulnerability, most of the unreinforced methods for the assessment of masonry complex
masonry structures were built (and structurally structures, which need retrofitting interventions.
modified with subsequent interventions) before Most of the works conducted at L’Aquila in this
the adoption of the 2002 updated requirements; area are characterized by a series of correlated
thus, the percentage of unreinforced masonry activities ranging from historic search, survey,
buildings heavily damaged or collapsed in the and analysis of the structure with regard to pro-
epicenter area is relatively high (approximately posals for repair and seismic strengthening. In
45 %). The first cause of widespread overall particular, several problems are generally
disruption originates from the very poor quality encountered in the various levels of survey and
damage interpretation and those of modeling and traditional residential buildings characterized by
analysis of the structural behavior of complex and shear-type behavior. Further, the activation of
clustered masonry buildings (Da Porto many local modes also calls for retrofit interven-
et al. 2013). The in-depth comparative studies tions, which should “tie up” the building, thus
between the surveyed damages and the results avoiding the local failure modes that are often
of structural assessment through different observed (Brandonisio et al. 2013).
approaches have produced a great enhancement Detailed 3D nonlinear numerical analyses
in the understanding of the complex behavior of have been performed on several palaces and
masonry structures, evidencing that use of the churches and also on the macroelements compos-
“Natural Laboratory of Earth” is the best ing them such as towers and facades. Linear and
approach to evaluate the seismic behavior and nonlinear structural analyses conducted by means
the performance of structural systems as well as of a full 3D finite element model have been used
the failures occurring in reality and is a to partially explain the collapse of the transept
non-substitutable step for the advancement of area of the Basilica of S. Maria di Collemaggio.
knowledge in the field of seismic engineering The transept structural system formed by arches,
(Brandonisio et al. 2013). For example, the obser- pendetives, and barrel vaults transferring the ver-
vation made on large historical palace, such as the tical loads onto the multilobed pillars has been
case of Palazzo Centi, has shown that, during the demonstrated to be vulnerable with respect to
2009 earthquake, differently from other historical both transversal and longitudinal actions. In par-
masonry buildings, it responded “reasonably” ticular, even in the case of longitudinal actions,
well in the main structural parts also due to recent the nonlinear static analysis evidences the pres-
retrofit interventions carried out in 2003. These ence of high tensile stresses in the arches
mechanical interventions proved to be the reason connecting the pillars to the apse walls. This
of the good seismic behavior of the building and mechanism, observed in the numerical simula-
that “saved” it from more serious damages dem- tions and amplified by the strong vertical accel-
onstrating the effectiveness of the traditional eration component registered close to the site,
techniques in designing and retrofitting masonry together with the minimal resistance of the mate-
buildings (Lucibello et al. 2013). rial inside the core of the collapsed pillars, is the
The structural assessment and retrofitting of most probable explanation of the implosion of the
important churches in L’Aquila, some of them transept structures in the church (Gattulli
having a monumental value, has been elaborately et al. 2013).
studied. Results concerning the seismic behavior The bell tower of the San Pietro di Coppito
of masonry churches have shown that the Church in L’Aquila suddenly collapsed during
dynamic excitation due to the seismic ground the devastating earthquake, which occurred in
motion activates many vibration modes of the April 2009. A series of analyses were performed S
building structure, though all of them are charac- to provide an ex post evaluation of the causes at
terized by small participation factors. Therefore, the base of the collapse and give operative design
churches are not behaving through a superposi- information for the reconstruction (Milani
tion of global modes but more as dynamic inter- et al. 2012). For this aim nonlinear static and
action of localized modes. Consequently, in limit analysis has been performed evidencing
many examined case studies, the ratio between the role played by the actual geometry and by
the total base shear and the church total weight the masonry mechanical characteristics of the
ranged between 20 % and 30 %, evidencing a tower through reconstructing the failure mecha-
reduction with respect to the plateau value of nisms by both modeling approaches. More
the spectral acceleration provided by the Italian recently, in a similar modeling approach, an
code. Therefore, appropriate choices of the force attempt has been made to take into account the
reduction factor should be adopted for these mon- effects of fiber reinforcement polymers posi-
umental buildings different from the case of tioned on a large facade of Palazzo Camponeschi,
as example of engineering analyses on the effec- with different characteristics and sensitivities.
tiveness of specific retrofitting intervention with Instead, more complex monitoring systems are
new material on ancient masonry structure in used in monumental churches and buildings
order to augment their seismic capacity (Gattulli where accelerometers are accompanied by
et al. 2014). crackmeters, inclinometers, temperature mea-
Finally, the dynamic behavior of masonry surement devices, etc. (Casarin et al. 2012;
structures under three-dimensional shaking is Russo 2013; Gattulli et al. 2014).
not completely understood, and it needs specific Within this broad range of activities, in the
investigations. In this sense, the direct observa- present chapter are summarized specific findings
tions and measuring of such behavior through the of the researchers of the University of L’Aquila
use of novel structural permanent monitoring constantly involved to define sustainable recon-
systems may help to reach such a goal. struction interventions, in the framework of
Before the 2009 L’Aquila earthquake, very broad multidisciplinary collaboration, open to
few structures were equipped by permanent the use of innovative technologies and accompa-
structural monitoring systems managed by the nied by valuable training of young scholars.
Department of Civil Protection (DPC). However,
the response of the Pizzoli Town Hall during the Seismic Behavior of Ancient Monuments:
mainshock has been recorded and analyzed by From Collapse Observation to Permanent
DPC, giving special insights on the potentiality Monitoring
of these systems for immediate evaluation of the The historic center of L’Aquila, which was
damages occurring during an earthquake. Succes- widely extended before the earthquake, was
sively, the large amount of installed, temporary inhabited by about 7,000 residents and 8,000
or permanent, different types of devices students, constituting a large and widespread uni-
(accelerometers, smart wireless devices, dis- versity campus. In this situation, dense of monu-
placement and velocity transducers, inclinome- ments of great value, characterized by a particular
ters, etc.) reach a number of around three hundred urban fabric, having an orthogonal system due to
(300) or more, evidencing a large impact of this the Angioina division into lots even if modified
technology in the post-earthquake emergency by an accelerated stratification, made by previous
phase, especially during the earthquake swarms earthquake, are placed more than 800 businesses,
(Gattulli 2013). In particular several monitoring several professional offices, and headquarters of
systems have been installed in the emergency the administrations of various public institutions.
phase, to understand the occurred behavior in This complicated system is developed for about
damaged building (Cimellaro et al. 2012; Foti 10 km, including historical centers, consolidated
et al. 2013), or during the construction of tempo- tissues, parts of the city under construction, and
rary scaffolding, in order to verify the efficacy of the natural park of Gran Sasso and Sirente-
the added structural system especially in the case Velino. The earthquake has seriously
of monumental building (Russo 2013). Because compromised the system and this area corre-
of this scope, in many cases, the permanent mon- sponds, quite closely, to the so-called seismic
itoring has worked ranging from a limited num- crater in which the most important damage
ber of hours up to several months. In other cases, occurred. In the revitalization of this large zone
the monitoring system is permanently installed affected by the earthquake, a central role is
on the structure for years; it can be used also to played by the recovery of the L’Aquila historical
determine the change that will occur in the struc- center accompanied by the similar ones in the
tural behavior during the reconstruction phase neighboring small villages. Hence, the character-
(Federici et al. 2012; Gattulli 2013). ization of this system, using a large scale, starting
In several cases, the structural monitoring sys- from an urban point of view up to geomorpho-
tem uses only accelerometers, starting from very logical and geotechnical vision, is an element of
few measurements to a larger number of devices strong relevance also to understand what happens
to the smaller scale of the individual monument, numerical checks that not always ensure a com-
integrating the specific observations with an plete reliability. Moreover, the physical nature of
overview of the problem. Indeed, the numerous masonry structures, product of improvements
campaigns for the characterization of the seismic obtained following successive attempts, and
behavior of the L’Aquila’s valley made by a empirical rules lead to the need to examine each
microzoning mesh of the historical center indi- case as a special case, making prohibitive to have
cated a number of site effects which are very a general characterization also due to the historic
important for better understanding the L’Aquila analyses which produces a series of unique cases.
earthquake and its effects. Recent studies, char- However, the entire reconstruction process must
acterizing the underground of the historical cen- also take into account other problems such as
ter, point out a sudden transition between the energy efficiency, architectural restoration, and
L’Aquila breccia and silty clay loam through a refunctionalization. All these actions have to
highly detailed reconstruction. make considering permanent monitoring of the
The problem of the reconstruction of historical monument so as to better schedule maintenance
buildings and monuments, especially within the and following the target of long-term operation of
walls of the historic center, brings to investigate the object.
the main open issues concerning the interventions A series of research groups of the University
that ensure a real improvement. The structural of L’Aquila have conducted numerous activities
behavior of the monuments, comprising a related to earthquake engineering immediately
masonry structure, still today open to a full under- after the earthquake to support the city and the
statement, depends on factors such as the entire community strongly hit by the catastrophic
mechanical properties of the different materials event. Here, we attempt to report this huge effort
composing the masonry walls, the features of the motivated by the willingness to recover as soon as
horizontal structure and their connection with the possible to the original state. This entails a path
vertical ones, and the used construction methods. that goes from collapse observation to permanent
Therefore, it is important to realize an optimal monitoring, thus characterizing a broadband
use of the wealth in the technical and scientific research activity and picking, according to the
knowledge, stored over the years, to evaluate authors’ opinion, the most significant results.
properly the physical consistency and safety of Therefore, the presentation is subdivided in sev-
historical buildings. The answer, of course, can- eral specific fields.
not be unique but should result in a critical and
competent attitude. Some ideas may be received Collapse and Damage Observation and Survey
from the reading of a book dealing with the tech- A series of partial or complete collapses have
nical problem of preservation. For example, in a occurred in historical buildings at the center of
classical book on this issue is reported “. . .In this L’Aquila, some of them having monumental S
type of intervention is of particular importance value. Among them, one of the most studied is
the respect of the typological, structural and func- the collapse of the transept of the Basilica di
tional quality of the system, avoiding those trans- S. Maria di Collemaggio (Gattulli et al. 2013).
formations that alter the character.” The matter is The major damage suffered by the Basilica in
still very complicated to be solved. Indeed, the 2009, indeed, was probably induced by a mecha-
properties of fragility, heterogeneity, and anisot- nism of implosion of all the structures that com-
ropy of masonry and their different typological pose the transept: the great multilobed pillars, the
varieties make it difficult to well describe unitar- triumphal arch and the wall above, the barrel
ily their mechanical behavior. Recent technical vaults, the dome, and the roofing structures
codes take into account these uncertainties, (Fig. 1). In particular, the failure occurred almost
allowing to use simplified numerical model, simultaneously in the two large pillars and the
implemented “ad hoc,” to evaluate the consis- arches connecting the nave walls with the chan-
tency of masonry structures, avoiding the use of cel, closed at the end by the apse, and it was
Seismic Behavior of Ancient Monuments: From Collapse Observation to Permanent Monitoring,

Fig. 1 Basilica of S. Maria di Collemaggio transept collapse: (a) aerial view and (b) internal view
Seismic Behavior of Ancient Monuments: From Col- (a) shaping of the pillars in a multilobed medieval form,
lapse Observation to Permanent Monitoring, (b) material pillar core evidenced by the collapse, and (c)
Fig. 2 Pillars and columns of S. Maria di Collemaggio: column damages
followed by the overtopping dome collapse. century plan, but only the works on the nave had
It should be pointed out that the transept is real- been completed, leaving the transept with its orig-
ized through a structural system in which the arch inal Baroque decoration. At the same time, he
and vault thrusts, induced by permanent vertical shaped the pillars in a multilobed form, removing
loads, play an important stabilizing role. During material along the height starting from the original
the simultaneous vertical and horizontal actions multilobed base (Fig. 2). In performing the resto-
of the 2009 earthquake, the equilibrium state in ration, the surface of the pillars was realized using
the masonry element was significantly altered. In blocks of freestone placed around the masonry
particular, due to the proximity to the epicenter, core of the Baroque pillars having only an aes-
the vertical and horizontal acceleration compo- thetic function with no resistance increase. Small
nents had similar amplitudes. stones placed randomly, and probably coming
Moretti’s restoration intervention in the early from previous earthquake debris, were used for
1970s was intended to remove the Baroque deco- the core material.
rative features added to the Basilica and to unearth Moreover, a subhorizontal fracture positioned
the “powerful original structure” of the thirteenth- at the height of around 2 m is present in the north
external wall, hosting the Holy Door, apparently elements, a partial detachment, and the presence
created at the interface between the lower of permanent relative displacements, which
masonry part subjected to grout injection occurred in one of the ribs. The lateral chapels
performed during the restoration works suffered more serious damage. However, the
conducted in 1999 and the upper masonry part study of the mechanical behavior of the monu-
preserving the original workmanship. ment cannot avoid to consider the low quality of
Besides that, several of the threaded rods of the masonry, even present in peculiar elements as
the bracing system under the central nave roof arches and vaults, exposed clearly in evidence by
were broken, almost all in the vicinity of the large the collapse (Fig. 3a) and the peculiar realization
crack on the north side wall of the church. In the of the masonry walls at L’Aquila that along the
north aisle, slippage of bolts designed to anchor time due to the occurrence of several earthquakes
the wooden trusses was noticed. Both of these is the product of different craftsmanship
cases of damage show the presence of transverse (Fig. 3b).
deformations of the longitudinal walls mainly
located in the north, which may occur even after Laser Scanning Measurements
the transept collapse. Besides the evident collapses easily visible, the
The damage pattern which affects the columns relief of the damage, caused by the earthquake on
of the nave walls was developed in a complex April 6, 2009, has produced continuous and stim-
manner; there were deep cracks due to heavy ulating activities in damage observation and sur-
compression, initially appearing only in the cen- vey, producing a positive and critical approach
tral columns of the nave (Fig. 2c) and then pro- about different aspects of the problem. Indeed, in
gressively emerging, with the aftershocks, in particular regarding this specific area, the cata-
almost all the others as well. The progression of strophic event was an opportunity to apply the
damage was very similar to that found during the most advanced available technologies in the field.
intervention in the last restoration of the 1970s, For the first time, the laser scanner technique was
which consisted of inserting some missing frag- extensively used for the surveying of a great
ments and grout with cement mortar in the portion of L’Aquila territory and in particular
opened joints. The only column that initially for the historical buildings and monuments posi-
exhibited minor damage was the first one towards tioned within the old L’Aquila walls. The survey
the transept, which had been completely rebuilt, of the portion of the city hit by an earthquake has
although simply with brick facing of limited been considered, in several cases, a useful sup-
thickness and with an inner core done in small port for other structural analysis and conservation
stones immersed in cement mortar. and restoration studies. Laser scanning has been
In all the other columns, large cracks were used just after the earthquake in L’Aquila and
visible. These columns, in distinction from the during several summer schools that took place S
first ones, were made using stones shaped like every year since the earthquake, to survey the
truncated pyramids; these stones were juxta- facades of numerous buildings. The point cloud
posed, in an almost dry manner, in order to form has been treated and filtered, in order to quantify
the octagonal section. In most cases, cracks and verify exactly the deformations due to the
occurred in the stone blocks, despite the appar- earthquake, compared with the original shape
ently high resistance of the material. The area of and geometry of the building. Laser scanning
the apse was also characterized by severe dam- surveys allow the acquisition of the shape com-
age. This area was affected by clear cracking in plexity of the building, with its irregularity, not
the mortar joints of the hewn stone blocks of the only in some essential lines, as it is in topography.
main apse, where there were also expelled blocks The whole information package is more complex
forming a wedge shape above the mullioned win- than what is obtained with traditional techniques
dow. Observation of the interior reveals cracks in (topography and photogrammetry), but it is
the apse vault, which is one of the most loaded extremely useful for survey and measurements
Seismic Behavior of Ancient Monuments: From Col- evidenced by the collapse and (b) masonry walls realized
lapse Observation to Permanent Monitoring, with material belonging to different periods and with
Fig. 3 Masonry features at L’Aquila: (a) masonry quality different craftsmanship
of structural deformations and cracking. More- the earthquake and to predict the retrofitting
over, the data acquisition system permits to improvements (Gattulli et al. 2014).
acquire information on difficult geometry from
a certain distance, avoiding difficulties related to Applications of Infrared Thermography for the
traditional methods of measurement. Moreover, Investigation of Historic Structures
the University of L’Aquila placed important Infrared thermography is used as a nondestruc-
resources in the field, through several labs tive technique to identify the most significant
(CERFIS, CSE, etc.), renewing the technical inhomogeneity of materials and properties in the
equipment and using specialist teams, which surface and structures that are difficult to appre-
operate for the geometric and damage relief. ciate by simple visual inspection but are instead
Data processing, made by Leica instrumentation detectable by the relief of anomalies associated to
(total station HDS2600 and software Leica thermal phenomena. A portable instrumentation,
Cyclone 7.2) and by a scanner velocity max of which consists in thermal imagers, enables the
1,016,727 point/s and flow rate in the range of observation of the materials constituting the
ambiguity of 79 m, permits a return of three- main and secondary walls (e.g., infill, detach-
dimensional model (see Fig. 4) from which ments, fills, and so on) and naturally the discon-
plants, prospects, and sections can be obtained. tinuities. The infrared thermography is a
The opportunity of having this great volume of noninvasive technique that uses infrared energy
data has stimulated to develop interactions emitted spontaneously by any body or object:
between different disciplines, and so they consti- electromagnetic radiation whose wavelength
tute a new wealth of information. In particular the depends on the temperature of the same body
detailed reconstruction of crack path in the main (Wien’s law). Because the temperature of the
facade of historical buildings (like Camponeschi bodies in ambient condition have a wavelength
Palace) allowed a verification of the numerical that belongs to the range of infrared spectrum
models to reproduce the observed behavior under (l = 0.3 0.78 mm), the technique is called
Seismic Behavior of Ancient Monuments: From Col- using laser scanner techniques: (a) Margherita Palace,
lapse Observation to Permanent Monitoring, headquarter of L’Aquila Municipality, and (b) De Amicis
Fig. 4 Reconstruction of tridimensional geometrical school building
models of the structures damaged by the earthquake
infrared thermography. The value of radiant resolution IR, 640 480 pixel; and spectral field,
energy issued change when the objective of the from 7.5 to 13 m. An extensive campaign of
camera passes from a point at a higher tempera- infrared thermography has been conducted to
ture to one with a lower temperature, and so also characterize the masonry of Camponeschi Pal-
the wavelength of the radiation assumes two dif- ace, where the Letter and Philosophy Faculty is
ferent values (Boltzmann’s law). Where there is a located. Figure 5 shows a series of findings.
thermal gradient, the scanner captures the energy Moreover, this technique can be used without
variation that is linked with the body temperature any interruption of the current activities and
through Stefan’s law, and so it can determinate works. The fast data processing allows for the
the value of emissivity’s coefficient. Using a ded- recognition of peculiar masonry zones with
icated scanner optical system that investigates doors or windows under the plaster that are par-
point by point all the recorded fields, the camera tially or completely closed, as well as with the S
reproduces an electrical signal representative of presence of chimneys and underground utilities,
the image’s thermal information and proportional masonry wall textures, warping of the floor,
to its intensity, which will be shown in a kine- nonhomogeneity of the materials, structural dam-
scope. In particular for each range of temperature ages, delamination of plaster as well as
variation, there will be a corresponding range of malfunctions of electrical and plumbing systems,
radiant intensity variation to which the instru- infiltration of moisture, leaks, and thermal
ment assigns a particular shade of color. The bridges.
main characteristics of the instrumentation used
are the following: minimum focusing distance, Ultrasonic Testing
0.3 m; temperature range, from 20 C to In ultrasonic testing (UT), very short ultrasonic
+120 C; field of view (FOV), 24 18 ; thermal pulse waves with center frequencies ranging from
sensitivity (NETD mK), <40 mK @ +30 C 0.1 to 15 MHz and occasionally up to 50 MHz are
(+86 F); accuracy, 1 C o 1 % of readings; transmitted into materials to detect internal flaws
Seismic Behavior of Ancient Monuments: From Col- identification of concrete elements, (d–g) reading of tex-
lapse Observation to Permanent Monitoring, tures walls under plaster, (h, i) reading of weaving arched
Fig. 5 Infrared thermography results obtained in lintels, (l, m) presence of openings in the internal masonry
Camponeschi Palace at L’Aquila: (a) evidence of opening walls transversal to the perimeter walls
closed with arched lintel, (b) failure of pipes, (c)
or to characterize materials. An ultrasound trans- medium. Imperfections or other conditions in

ducer connected to a diagnostic machine is the space between the transmitter and receiver
passed over the object being inspected. reduce the amount of sound transmitted, thus
There are two methods of receiving the ultra- revealing their presence.
sound waveform: reflection and attenuation. Ultrasonic testing, used on masonry, permits
In reflection (or pulse-echo) mode, the transducer to obtain information about the homogeneity and
produces and receives pulsed waves when the composition of the masonry wall of the analyzed
“sound” is reflected back to the device. Reflected building. This technique has the benefit to be a
ultrasound comes from an interface, such as the nondestructive test and can be done in a very
back wall of the object or from an imperfection rapid and simple way in wide region of structural
within the object. The diagnostic machine dis- parts, but on the contrary, this large volume of
plays these results in the form of a signal with data needs a specific interpretation. For this rea-
amplitudes representing the intensity of the son, it is important to combine the sonic tests with
reflection and the distance and registering the destructive ones, performed in a limited and sig-
arrival time of the reflection. In attenuation nificant structural zone, to have a well-specific
(or through-transmission) mode, a transmitter and useful reference which defines the right rela-
sends ultrasound through one surface, and a sep- tion between the speed of sound and the mechan-
arate receiver detects the transmitted signal on ical characteristics of the investigated masonry.
another surface after traveling through the Therefore, it is useful to highlight the main
technical properties of the instrumentation used 48 mm, maximum frequency for pulse emis-
in a lot of campaign tests at L’Aquila by the sion: 1 per second, minimum measurement
University: pitch: 10 mm
• Acquisition, measuring range from 100 mv to In Fig. 3 images obtained from sonic tests are
20 v reported. From the comparison of the graphs
• Time bases from 20 ns to 81.9 s reported in Fig. 3a, b, the distribution of pulse
• Sample resolution 8 bit wave speed in the same portion of masonry
• Samples for event 8,192 for contact measure- before and after fluid mortar injection grouting
ments, 640 for log, bandwidth 50 mhz, filter is highlighted. In the first case the maximum
for ultrasounds: central frequency 50 kHz, speed is 250 m/s, while in the second case it
measurement channel 1 grows up to 1,709 m/s. The different velocity
• Probes through contact and sonic test with transmission indicates clearly the increase of
hammer: resonant frequency 53 kHz, diameter stiffness due to the grouting (Fig. 6).
a 50 b 50
45 45
40 40
35 35
30 30
25 25
250
1800
20 240 20
1600
15 230 15
1400
10 220 10 1200
5 210 5 1000
0 200 0 800
−5 190 −5 600
−10 180 −10 400
−10 −5 0 5 10 15 20 25 30 35 40 45 50 −10 −5 0 5 10 15 20 25 30 35 40 45 50
Seismic Behavior of Ancient Monuments: From Col- distribution of the velocity of propagation (a) before and
lapse Observation to Permanent Monitoring, (b) after the fluid mortar injection grouting and (c) loca-
Fig. 6 Sonic tests on a wall of Camponeschi Palace: tions of injection pipes
Georadar (Ground Penetrating Radar: GPR) Fig. 7b in terms of QRI distribution, evidencing
Tests a relevant variability of QRI ranging from 1 to
The GPR methodology that employs electromag- 3.25.
netic waves permits the boundary definition of
mechanical characteristics of the investigated Removal of Plaster and Endoscopic Tests
body by the identification of the interface Among minimal invasive tests, the removal of
between layers that have different resistivity and plaster rendered a zone of inhomogeneity, even-
dielectric constant. Structural degradation is clas- tually pointed out by infrared thermography, vis-
sified (four classes of QRI “Quality Radar ibly permitting to check the true nature of the
Index”) by a qualitative analysis of the power walls and of the composing elements. Succes-
radar signal behavior with the depth by both sively, endoscopic tests have been usually made
radar anomalies in the section (presence/absence, in several historical masonry walls at L’Aquila.
point 0 and 1) and energy recovery, intense and Endoscopy is simply an extension of the essential
localized (also called “pick”) in the graphs (point visual survey into areas inaccessible to the naked
1 for the pick up to 6 dB, 2 point up to 12 dB, and eye. The equipment ranges from relatively simple
3 point over 13 dB). The total QRI is the sum of borescopes, consisting of a light source, a small-
the two scores described above. The method is diameter rigid tube with built-in optics, and an
calibrated using correlation between radar data eyepiece, to complex controllable systems with
and a direct destructive test made on hundred numerous specialized attachments. By drilling a
scans acquired on masonry tester. This method hole (normally less than 12 mm) and inserting the
permits a direct or relative classification tube, it is possible to inspect voids under floors or
depending on whether there are direct data on behind paneling, for example. Any hidden prob-
each structure or not. The relative classification lems such as fungal growth can, in theory, be
is divided in four classes: QRI 4, class of low identified. The more sophisticated and expensive
quality; QRI 3, class of medium-low quality; QRI equipment is fully flexible and can be steered by
2, class of medium quality; and QRI 1, class of wires built into the casing. Systems are available
medium-high quality. In Table 1 the main results down to 6 mm in diameter, and more specialized
of the georadar tests are reported and the obtained systems may reach less than 2 mm. It is possible
classification, in particular, is highlighted in the to attach still or video cameras to the eyepiece to
last column. record the findings. The theory is fairly simple,
In Fig. 7a a grid in the scanned surface of a but in practice it can be very difficult to retain a
portion of a masonry wall of the Camponeschi sense of scale of the observed image and keep
Palace is presented while the obtained results track of the location and orientation of the tip.
through the georadar test are described in The focal range, depth of field, and strength of

Table 1 Main information deduced by GPR tests
GPR graphics GPR sections
Scansions Max Presence/absence
Codex Length db Max height of the picks equalized Point Localized anomalies QRI index
LEA10000 0.6 5 5.0 1 0 1
LEA10001 0.6 2 2.0 1 0 1
LEA10002 0.6 5 5.0 1 1 2
LEA10003 0.6 7 7.0 2 1 3
TEA10000 0.6 9 9.0 2 0 2
TEA10001 0.6 8 8.0 2 1 3
TEA10002 0.6 5 5.0 1 0 1
TEA10003 0.6 7 7.0 2 0 2

Fig. 7 Georadar tests on masonry of Camponeschi Palace: (a) grid scans and (b) QRI distribution
light are greatly reduced in the smaller diameter taking into account explicitly the inherent uncer-
systems. It is not unknown for insulation lagging tainties of the problem. A more reliable descrip-
to be misidentified as dry-rot. However, reliable tion of the seismic behavior of the masonry
reconstruction of the stratification in the main mechanical characteristic will realize a well-
masonry walls has been obtained, to know the balanced design of retrofitting interventions.
construction technique and to assess qualitatively Figure 8 summarizes the results obtained
the physical conditions. The employed instru- through double flat-jack tests realized at different
mentation has permitted to register short movies locations of the Camponeschi Palace. The three
of what is visible inside the cavities, producing a performed tests evidence the variability in both
large quantity of data in digital form to be elastic modulus and measured strength through
processed. the adopted testing procedure. Indeed, in the ana-
lyzed case, the average strength value is 2.47 N/
Flat-Jack Tests mm2, while the variance is 0.736 N/mm2,
Several tests of single and double flat jacks were evidencing the large scatter in the material prop-
carried in different historic buildings in the center erties belonging to the same historical building. S
of L’Aquila including Camponeschi Palace, De
Amicis school, the headquarter of the Carispaq Computational Mechanics and Masonry
Bank and Margherita Palace, and others. More- Structural Modeling
over, diagonal tests have also been performed to Structural analysis and numerical modeling of the
characterize the shear strength in selected seismic behavior of masonry structures is still an
masonry panel individuated in existing walls in open research topic, especially in Italy and in
order to characterize the mechanical properties of particular after the L’Aquila earthquake. Rele-
the main typology of historical masonry walls at vant issues pertain to the inherent uncertainties
L’Aquila. A probabilistic characterization of the related to the evaluation of the seismic adequacy
available data on the mechanical features of the of complex masonry buildings through reliable
historical masonry at L’Aquila is going on in numerical models. Starting from the limit analy-
order to have the possibility to investigate the sis in which the simplified assumption of describ-
structural assessment of historical buildings, ing the incipient collapse behavior through chains
2720
Medium vertical value Medium vertical value Medium vertical value

2.5 4 1.6
2
3 1.2
1.5
2 0.8
1
σ [N/mm2]
σ [N/mm2]
σ [N/mm2]
1 0.4
0.5
0 0 0
0 0.001 0.002 0.003 0.004 0 0.001 0.002 0.003 0.004 0 0.001 0.002 0.003 0.004 0.005
ε ε ε
Elastic Elastic Elastic

Range of Range of Range of
Reference secant Reference secant Reference secant
tension tension tension
test modulus test modulus test modulus
N / mm2 N / mm2 N / mm2
N / mm2 N / mm2 N / mm2
0 ÷ 1.5 1307 0 ÷ 2.0 1891 0 ÷ 0.8 265

MP2 MP3 MP4
0 ÷ 2.3 697 0 ÷ 3.2 913 0 ÷ 1.2 244
Seismic Behavior of Ancient Monuments: From Collapse Observation to Permanent Monitoring, Fig. 8 Results of double flat-jack tests on Camponeschi Palace
Seismic Behavior of Ancient Monuments: From Collapse Observation to Permanent Monitoring
Seismic Behavior of Ancient Monuments: From Col- transept: model A, (a) maximum P1 and (b) minimum
lapse Observation to Permanent Monitoring, P3; model B, (c) maximum P1 and (d) minimum P3
Fig. 9 Principal stresses due to static loads in the
of rigid masonry blocks, a series of different used to model the main walls in which a prevalent
modeling approaches have been experimented bidimensional behavior has been recognized
within the activities of the computational (Fig. 9a, b) (see also Gattulli et al. 2013); in a
Mechanics Lab of the University of L’Aquila second case (model B), a more refined model has
directed by the first author. On this respect, the been considered in which all the masonry ele-
experience gained at L’Aquila benefits from the ments have been modeled using eight-node
large amount of data available for possible com- isoparametric solid elements (Fig. 9c, d). It is S
parison with the numerical simulations. There- evident as the stress flux is more clearly defined
fore, potentiality and drawbacks of several in the arches and in the pillars by the more refined
modeling assumptions have been evidenced model, even if the level of stress amplitude is
along the path, which have shown to be particu- comparable in the two cases. At the same time,
larly useful also to enrich the tools for retrofitting the analysis evidences that in the pillars is present
design. The structural modeling of the Basilica of an important flexural deformation induced by the
S. Maria di Collemaggio constitutes a relevant thrust differences in the longitudinal direction
case study in this field. Figure 9 reports, synthet- between the last arch belonging to the nave
ically, the results of two stress analyses induced walls and the arch of the transept on which is
by live loads conducted by means of two different laying the dome and the vault. Even if the results
finite element models of the entire Basilica: in the are obtained under the hypothesis of modeling the
first case (model A), four-node quadrilateral masonry material as an elastic and homogeneous
isoparametric plate-shell elements have been solid, they evidence the role played by the
three-dimensional geometry and the consequent with a 3 3 Gauss integration scheme

internal action distribution. Figure 9c evidences were used. Figure 10b shows the adopted mesh.
as the maximum tension flux in the column is The incremental/iterative solution procedure
prevalently radial, while the minimum one in employed is a modified Newton–Raphson
Fig. 9d is vertical. approach. A constitutive law, comprising a para-
A different activity has been carried out on the bolic hardening rule and a parabolic exponential
Camponeschi Palace which represents a good softening branch, modeled the compression
example of how damage observation and exper- behavior of the masonry; a linear hardening branch
imental survey, performed using the techniques followed by a linear softening branch character-
described above, can be a valuable aid in ized the tension behavior. The FRP reinforcement
supporting the modeling choices for both damage was modeled using truss elements directly
interpretation and simulating the effect of connected to the nodes of the mesh of the panels
retrofitting interventions. without using interface elements. Thus, perfect
The Camponeschi Palace is an important his- adhesion between the nodes of trusses and
torical building located in the historical city cen- the corresponding nodes of the mesh was consid-
ter of L’Aquila. The palace was built near the end ered. The strips were considered not able to carry
of the sixteenth century by the Jesuit order that compression loads. The tensile behavior of the
established and hosted in the structure an institu- FRP was represented with a stress–strain relation-
tion of higher education. Several modifications to ship, characterized by a linear elastic behavior in
the original configuration and structural interven- tension, followed by exponential degradation.
tions have been made during the years. The pal- Nonlinear incremental static (pushover) ana-
ace sets upon a rather sloping site and it is lyses were carried out on unreinforced and
characterized by an L-shaped plan with two reinforced facade, separately for the positive
arms of equal length delimiting an internal court- (Fig. 10c) and negative x direction (the horizontal
yard. The building was hosting the Faculty of one) (Fig. 10d) and for both the unreinforced and
Philosophy and Letters of University of L’Aquila FRP-reinforced masonry (Gattulli et al. 2014).
at the time of the 2009 earthquake. In general, a correct and reliable implementa-
Due to the relevant structural damage, which tion of the model allows performing in-depth
occurred, the Department of Civil Protection analyses of the expected structural behavior
classified the palace structurally unsafe. The under seismic loads enabling robust design
crack pattern of the facade due to the 2009 choices.
L’Aquila mainshock and successive earthquake
swarms was detected through automated laser- Structural Monitoring for Historic Structures
based surveys. The crack pattern distribution of Structural health monitoring is an emerging tool
the main facade is shown in Fig. 10a. The major for reliable structural assessment. Efficient mon-
damage was concentrated on the first story. itoring programs may help in the characterization
On the left side of the facade, subvertical and of the progressive decay of structural perfor-
diagonal cracks developed at the end sections of mance, both in the short and in the long term,
the spandrels above and below the openings. providing useful information for optimized main-
Single and double flat-jack tests were carried tenance and safety. Generally, electronic equip-
out on site in order to evaluate the compressive ment deployed with the purpose of structural
strength and the deformability properties of the health monitoring is implemented via wired sys-
masonry for the entire building. tems. In this case, the development of the system
A numerical model was built specifically to can be limited by a series of factors such as the
simulate the nonlinear static behavior of the high cost of measurement equipment, difficulty
facade and of a retrofitting system using fiber in installation due to large instruments size, and
reinforcement polymers for the masonry panels. need of a wired communication and power infra-
Eight-node isoparametric plane stress elements structure. In recent years, great attention has been
c 2,0
1,8
a
1,6
1,4
1,2
1,0
V/W
0,8
0,6
Unreinforced facade
0,4 FRP-reinforced facade (case 1)
0,2 FRP-reinforced and increased fm (case 2)
0,0
S 0,0 5,0 10,0 15,0 20,0 25,0 30,0 35,0 40,0
Horizontal displacement [mm]
d 2,0
b
1,8
1,6
1,4
1,2
1,0
V/W
0,8
0,6
Unreinforced facade
0,4 FRP-reinforced facade (case 1)
0,2 FRP-reinforced and increased fm (case 2)
0,0
0,0 5,0 10,0 15,0 20,0 25,0 30,0 35,0 40,0
Horizontal displacement [mm]
Seismic Behavior of Ancient Monuments: From Collapse Observation to Permanent Monitoring
Seismic Behavior of Ancient Monuments: From Collapse Observation to Permanent Monitoring, Fig. 10 Camponeschi Palace: (a) damage pattern on the facade (b)
mesh of the FEM model with the FRP layout; numerical pushover curves for horizontal loads (c) +x direction and (d) x direction
2723
S
paid to the development of wireless sensor net- reinforcements, the mechanical parameters
works that are able to overcome several problems characterizing the local behavior of the rein-
linked to the development of the entire layout but forcement either in the reinforcing fibers and in
evidencing other issues related to the new tech- their support may be opportunely monitored.
nology. One of which is connected to the collec- Subsequently, through information data fusion
tion and transmission of a great quantity of data. made available by arrays of sensitized reinforce-
In fact relevant information can be obtained by ments distributed on a masonry macroelement,
the registrations of the vibration data induced by mechanical parameters influencing the global
ambient sources, wind, earthquakes with low behavior may be identified. Possible develop-
intensity, operation of machines, and traffic. The ments concern the use of measurements at dif-
sensors are placed in specific points of the struc- ferent levels of excitation amplitudes, which are
ture and appropriately selected and the choice of available in seismic areas. They may be used to
which is based, for example, on numerical model- define a methodology for identifying nonlinear
ing. Using the acquired data, it is possible to behavior, distinguishing between scenarios
identify dynamical models, which predict the where damage occurs in the fiber, in the support,
observed behavior on the basis of error minimi- or in the masonry structure.
zation. Differences in the identified systems pro- The implementation of a full SHM system in
duced by data acquired in different periods historical buildings is still an open problem espe-
separated by a long interval of time may permit cially in aspects relating to the correct fusion of
to identify relevant changes in the behavior of the vibration measurements, integrated with other
systems which can be linked to degradation of type of measurements. For example, in concrete
the structural performance or incurred damages. structures it has proven useful to follow the trend
For example, in largely damped, like masonry, of humidity and temperature in order to evaluate
structures, it is convenient to use high sampling deterioration and updated material performances.
times for acquisition as opposed to frame struc- In this manner early assessment and warning of
tures. Although the interesting structural modes incipient problems may be used to drive quick
can be placed within an interval 0–10 Hz, the interventions or the planning and scheduling of
structural response may be highly damped and the maintenance programs. To the contrary, in
attenuated with a reasonable amplitude level monumental structures often numerous cracks
contained in few seconds. The acquired time his- are present which can be the objective of perma-
tories under seismic action will be composed of a nent static monitoring. Indeed, following the
few samples that make difficult the reconstruc- slow opening and closing of the crack may be
tion of the frequency spectrum. Some important used as an important correlation measurement for
benefits are that smart sensors include micropro- validating information provided for the identifi-
cessors that are able to run some basic algorith- cation of other parameters. The opening and clos-
mic tasks and then store some condensed ing may depend on the cyclic trend of the average
information to be exchanged between the other temperature corresponding to a certain time inter-
sensors (nodes) that comprise the wireless net- val (day, month, season). Similarly to this trend,
work (WN). it may possible to observe a reduction or increase
The long-term activity at the University of of stiffness or a change of configuration that may
L’Aquila aims to develop innovative automated induce also a cyclical trend in the identification of
techniques, which permit to automatically natural frequencies.
update the appropriately selected models, Of course, this set of actions permits both a
which in turn may address questions regarding rationalization of the cost and an improvement of
the seismic safety of the monitored structures. In the service-life predictions, potentially also for
particular, following a parametric identification monumental structures, especially if they have
approach, using the distributed available mea- been recently restored and retrofitted with a con-
surements from sensors embedded in the sequent important investment.
In general there are no particular rules for the response amplitude in the acquired time history
design of a fully operating SHM system, but each (one of this shown in Fig. 11f) determines the
case is unique and it requires particular care espe- possibility to get information regarding the main
cially with respect to the goals to be achieved. modal features.
The observation and the investigation on the Nowadays, in order to derive the processed
damage reported by the structure after a seismic information, from experimental data, different
event can lead to a customized system able to modal identification procedures can be used
provide specific information for continuously which are generally classified based on whether
assessing structural condition. Correlations they operate in the time domain or in the fre-
among obtained results should be sought for and quency domain. However, the amount of
only after analyzing all available information; the contained information is relatively independent
responsible team, formed by professionals with of the analysis domain.
different skills (structural, geotechnical, architec- Among several procedures probably, the two
tural, electronic, and informatics), on the basis of most popular algorithms are the Enhanced Fre-
their experience and sensibility, can formulate a quency Domain Decomposition (EFDD) and the
final diagnosis and define the actions to be Stochastic Subspace Identification (SSI) (Peeters
performed. and De Roeck 2001), and the first operates in the
Within such a context, in the first phase of the frequency domain and the second in the time
SHM system deployment, at the Basilica di Santa domain.
Maria di Collemaggio, different actions have The two procedures assume that the system
been performed including design, implementa- input (loads) is white noise that can be thought
tion, installation, and test of a network monitor- as a representative model of environmental noise.
ing accelerations. The main goal was an accurate However, even if the adoption of the white noise
measurement of the basic structural response, model in the case of a seismic event may be
both to environmental actions and to questionable, the straightforward applications of
low-amplitude seismic events, which occasion- the abovementioned procedures to response data
ally continue to occur at the site. A wireless sen- acquired during small earthquake still allow to
sor network was used as a platform combining a derive valuable information especially if special
MEMS tri-axial accelerometer and also sensors care is taken, such as in the case of the studied
to measure humidity, temperature, and luminos- system at the Basilica (Gattulli et al. 2015a, b).
ity (Gattulli et al. 2014). The proper positioning Indeed, here, using the SSI procedure, for exam-
of the sensors has been designed through the ple, the evaluation of the identified stable poles
study of different numerical models accounting may include dominant components of the input,
for the interaction between the main wall element producing possible errors or misinterpretation of
and the temporary scaffolding (Fig. 11a). In par- the dynamic behavior. However, the direct com- S
ticular, the dynamic behavior can be observed parison with results obtained by numerical
looking at the principal modal characteristics models driven by engineering judgment permits
(frequency, modal shape, and percentage of par- to extract realistic structural modal data. More-
ticipating mass) of the linear dynamic response over, the robustness of this data can be pursued
(Fig. 11b, c). The analyses suggest, for example, through the use of several different registrations
that, by placing the sensors along the internal generated by earthquakes with different charac-
longitudinal facade, it is possible to survey the teristics dependent on the source location. Fol-
transversal displacements especially during a lowing the reliable identification of the modal
seismic event. Until today, after almost 3 years characteristics, the second step is the updating
of permanent monitoring, the system is still able of the numerical model parameters in order to
to capture relevant data during the occurrence of minimize the error between measured and numer-
several seismic events coming from far and near ical modal features. In the error index formula-
fields. Generally, the registered maximum tion, natural frequencies and/or modal shapes
Seismic Behavior of Ancient Monuments: From Col- element and the temporary scaffolding; (b) first
lapse Observation to Permanent Monitoring, (1.09 Hz) and (c) second (1.97 Hz) mode shape; layout
Fig. 11 Structural monitoring at the Basilica of of the multifunction wireless sensor network: (d) plant, (e)
S. Maria di Collemaggio: (a) finite element model sensor, (g) section, and (f) acquired registration during a
accounting for the interaction between the main wall seismic event
a 80
70
60
model order
50
40
30
20
10
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

frequency [Hz]
b c
Seismic Behavior of Ancient Monuments: From Col- (a) stabilization diagram of the SSI procedure; (b) first
lapse Observation to Permanent Monitoring, (0.90 Hz) and (c) second (1.45 Hz) identified mode shape
Fig. 12 Identified modal model by structural monitoring:
S
may be considered. In the first case, a simple the Hankel matrix considered in the identification
difference in percentage is generally considered, process, while the abscissa is reporting the fre-
while the Modal Assurance Criteria (MAC) is quency; the circles with a cross inside indicate
used to compare measured and numerical modal stable modes, while the others, colored in gray,
shapes. The MAC criterion indicator permits to are unstable. At least one stable mode is present
evaluate the differences between similar modes at different frequencies, but in the case of two
but also the cross-orthogonality between differ- specific frequencies, stable modes seem to be
ent numerical and experimental modal shapes. present at almost all different model orders con-
As an example, Fig. 12a illustrates the stabili- sidered in the analysis. These stable poles are
zation diagram obtained from the time histories present at 0.90 Hz and in the range between
acquired by the wireless sensor network using 1.45 and 2 Hz. In the latter case, a comparison
SSI during a seismic event. The ordinates include with numerical models permits possible interpre-
the model order, which simply refer to the size of tation of the modal identification. The first two
identified modes are illustrated in Fig. 12b, c ▶ Masonry Modeling

illustrating the deformed shape of the portion of ▶ Retrofitting and Strengthening Masonries of
the Basilica in which the response used in the Heritage Structures: Materials Used
identification procedure has been registered. ▶ Seismic Vulnerability Assessment: Masonry
The second phase of the monitoring system Structures
development involves the deployment of ▶ Stochastic Structural Identification from
crackmeter and inclinometer sensors at the Basil- Vibrational and Environmental Data
ica. Figure 11d summarizes the employed instru-
mentation, where the red circles indicate the
wireless sensor network relative to the acceler- References
ometer; the blue squares and the orange triangles
correspond to the crackmeters and the inclinom- Brandonisio G, Lucibello G, Mele E, Luca AD
eters, respectively. Moreover, the green circles (2013) Damage and performance evaluation of
masonry churches in the 2009 L’Aquila earthquake.
refer to an accelerometer placed at the base of a
Eng Fail Anal 34:693–714
pillar necessary for acquiring the acceleration of Carocci CF (2012) Small centres damaged by 2009
the seismic input, while the star indicates the L’Aquila earthquake: on site analyses of historical
positioning of the node gateway that collects the masonry aggregates. Bull Earthq Eng 10(1):45–71
Casarin F, Benetta MD, Modena C, da Porto F, Valluzzi
data of each sensor and sends them to the server.
MR, Cantini L, Tedeschi C, Condoleo P (2012) Struc-
The data fusion of the two sets of information is tural monitoring and investigation campaign in the
currently under investigation. church of St. Giuseppe dei minimi in L’Aquila, after
the 6th of April 2009 earthquake. RILEM book series,
Springer, Berlin, vol 6. pp 1187–1193
Ceci AM, Contento A, Fanale L, Galeota D, Gattulli V,
Summary Lepidi M, Potenza F (2010) Structural performance of
the historic and modern buildings of the University of
The earthquake, which occurred on April 6, 2009, L’Aquila during the seismic events of April 2009. Eng
Struct 32(7):1899–1924
has been a catastrophic event for both the city and
Ceci AM, Gattulli V, Potenza F (2013) Serviceability and
the University of L’Aquila. Nevertheless, the damage scenario in RC irregular structures: post-
disaster has transformed itself into a tremendous earthquake observations and modelling predictions.
opportunity to revitalize the area, providing the J Perform Constr Facil 27(1):98–115
Cimellaro GP, Piantà S, De Stefano A (2012) Output-only
potential for the national and international scien-
modal identification of ancient L’Aquila city hall and
tific community to test the effectiveness of new civic tower. J Struct Eng 138(4):481–491
systems and technologies and consequently to D’Ayala DF, Paganoni S (2011) Assessment and analysis
establish main directives for the monitoring of of damage in L’Aquila historic city centre after 6th
April 2009. Bull Earthq Eng 9(1):81–104
monumental structures. The applicability of new
Da Porto F, Silva B, Costa C, Modena C (2012) Macro-
technologies in the areas of observation, survey- scale analysis of damage to churches after earthquake
ing, testing, modeling, restoration, retrofitting, in Abruzzo (Italy) on April 6, 2009. J Earthq Eng
and monitoring of ancient monuments, of which 16(6):739–758
Da Porto F, Munari M, Prota A, Modena C (2013) Anal-
the city center is densely populated, is exten-
ysis and repair of clustered buildings: case study of a
sively documented herein, through an organized block in the historic city centre of L’Aquila (Central
outline and a series of case studies. Italy). Construct Build Mater 38:1221–1237
Decanini LD, Liberatore L, Mollaioli F (2012) Damage
potential of the 2009 L’Aquila, Italy, earthquake.
J Earthq Tsunami 6(3):art. no 1250032, 32 pp
Cross-References Federici F, Graziosi F, Faccio M, Gattulli V, Lepidi M,
Potenza F (2012) An integrated approach to the
▶ Ambient Vibration Testing of Cultural design of wireless sensor networks for structural health
monitoring. Int J Distrib Sens Netw Art. No. 594842,
Heritage Structures
16 pp
▶ Damage to Ancient Buildings from Foti D, Gattulli V, Potenza F (2013) Output-only identifi-
Earthquakes cation and model updating by dynamic testing in
Seismic Collapse Assessment 2729
unfavorable conditions of a seismically damaged Tertulliani A, Leschiutta I, Bordoni P, Milana G (2012)

building. Comput Aided Civ Infrastruct Eng 29 Damage distribution in L’Aquila city (central Italy)
(9):659–675 during the 6 April 2009 earthquake. Bull Seismol Soc
Gattulli V (2013) Advanced applications in the field of Am 102(4):1543–1553
structural control and health monitoring after the 2009 Volpe M, Piersanti A, Melini D (2012) Complex 3-D finite
L’Aquila earthquake. In: Sebastiano D’Amico element modelling of the 2009 April 6 L’Aquila earth-
(ed) Engineering seismology, geotechnical and struc- quake by inverse analysis of static deformation.
tural earthquake engineering. InTech, Croatia, Geophys J Int 188(3):1339–1358
pp 208–235. ISBN: 978-953-51-1038-5, doi:10.5772/
55438
Gattulli V, Antonacci E, Vestroni F (2013) Field observa-
tions and failure analysis of the Basilica S. Maria di
Collemaggio after the 2009 L’Aquila earthquake. Eng Seismic Collapse Assessment
Fail Anal 34:715–734
Gattulli V, Lampis G, Marcari G, Paolone A (2014) Sim- Christoph Adam1 and Luis F. Ibarra2
ple modeling of FRP reinforcement: from masonry 1
Department of Engineering Science, Unit of
panels to a historic palace facade. Eng Struct
75:604–618 Applied Mechanics, University of Innsbruck,
Gattulli V, Potenza F, Federici F, Graziosi F, Colarieti A, Innsbruck, Austria
Faccio M (2015a) Distributed structural monitoring 2
Department of Civil and Environmental
for a smart city in a seismic area. Key Eng Mater
Engineering, University of Utah, Salt Lake City,
628:123–135
Gattulli V, Potenza F, Lepidi M, Federici F, Graziosi F, USA
Colarieti A (2015b) Seismic structural monitoring by
wireless sensor networks of the Basilica S. Maria di
Collemaggio after the transept collapse. Earthq Eng
Struct Dyn (in preparation)
Synonyms
Guidoboni E, Comastri A, Mariotti D, Ciuccarelli C,
Bianchi MG (2012) Ancient and medieval earthquakes Collapse criterion; Collapse variability; Compo-
in the area of L’Aquila (northwestern Abruzzo, central nent deterioration; Nonlinear dynamic analysis;
Italy), A.D. 1–1500: a critical revision of the historical
P-delta effect; Structural modeling strategies
and archaeological data. Bull Seismol Soc Am
102(4):1600–1617
Indirli M, Kouris LAS, Formisano A, Borg RP, Mazzolani
FM (2013) Seismic damage assessment of Introduction
unreinforced masonry structures after the Abruzzo
2009 earthquake: the case study of the historical cen-
ters of L’Aquila and Castelvecchio Subequo. Int In earthquake engineering, structural collapse is
J Archit Herit 7(5):536–578 defined as the local or global failure of a system
Lucibello G, Brandonisio G, Mele E, Luca AD that occurs due to the loss of vertical load-
(2013) Seismic damage and performance of Palazzo
carrying capacity in the presence of seismic
Centi after L’Aquila earthquake: a paradigmatic case
study of effectiveness of mechanical steel ties. Eng events. The two primary modes of global collapse
Fail Anal 34:407–430 are sidesway and vertical collapse. Sidesway col- S
Milani G, Russo S, Pizzolato M, Tralli A (2012) Seismic lapse is the global failure of the system caused by
behavior of the San Pietro di Coppito Church bell
a reduction of the lateral load-bearing capacity
tower in L’Aquila, Italy. Open Civ Eng J 6(1):131–147
Özerdem A, Rufini G (2013) L’Aquila’s reconstruction due to large horizontal displacements, whereas
challenges: has Italy learned from its previous earth- vertical collapse is caused by a direct loss of the
quake disasters? Disasters 37(1):119–143 gravity load-bearing capacity in one or several
Parisi F, Augenti N (2013) Earthquake damages to cultural
structural components (Krawinkler et al. 2009).
heritage constructions and simplified assessment of
artworks. Eng Fail Anal 34:735–760 Vertical collapse is a type of progressive col-
Peeters B, De Roeck G (2001) Stochastic system identifi- lapse, which is the total or disproportionate fail-
cation for operational modal analysis: a review. J Dyn ure of the system triggered by an initial local
Sys Meas Control 123(4):659–667
failure that spreads out from element to element.
Russo S (2013) On the monitoring of historic Anime Sante
church damaged by earthquake in L’Aquila. Struct Commonly, in ductile frame structures,
Control Health Monit 20(9):1226–1239 sidesway collapse is the predominant mode of
2730 Seismic Collapse Assessment
collapse during catastrophic earthquake events. UCSD test PWD east wall
These systems are affected by component mate-
monotonic
rial deterioration and/or P-delta effects. Material
deterioration reduces the component’s strength
under large inelastic displacements. Moreover,
load
the strength and stiffness of the structural com-
ponents deteriorates in each load cycle, reducing cyclic
the load-bearing capacity. The importance of
material deterioration decreases for long-period
structures, where the P-delta effects (also referred displacement
to as second-order effects) may lead to a negative
Seismic Collapse Assessment, Fig. 1 Force–
post-yield stiffness that in combination with large displacement relations from a monotonic and a cyclic
inelastic displacements accelerates the onset of test (Modified from Gatto and Uang 2002; With permis-
collapse. sion from Prof. Uang, UC San Diego)
In most non-ductile frames, however, columns
lose the capacity to sustain gravity loads due to captures these failure modes, and modeling
shear and subsequent axial-load failures, and as a uncertainties are quantified. Evaluation of seis-
result, the collapse mechanism is controlled mic collapse should be based on structural ana-
by the loss in vertical load-bearing capacity. lyses that incorporate strength and stiffness
As outlined in Baradaran Shoraka (2013), this material deterioration, as well as geometric non-
type of collapse of older concrete buildings may linearities (P-delta effects). Collapse of single-
be precipitated by axial-load failure of columns, degree-of-freedom (SDOF) systems was first
punching shear failure of slab-column connec- studied by including only P-delta effects in seis-
tions, failure of slab-diaphragm connections, or mic response, which may cause a negative inelas-
axial-load failure beam/column joints. tic tangent stiffness that eventually leads to
Prediction of earthquake-induced collapse system collapse (Jennings and Husid 1968;
requires the identification of all possible modes Bernal 1987; MacRae 1994). The development
of collapse, appropriate structural modeling of of hysteretic models that include strength and
building components under cyclic loading, and stiffness deterioration (Kunnath et al. 1990;
suitable representation of earthquake excitation Sivaselvan and Reinhorn 2000; Song and
including the consideration of epistemic and ale- Pincheira 2000; Ibarra and Krawinkler 2005)
atory uncertainties. Krawinkler et al. (2009), improved the assessment of collapse capacity.
Haselton et al. (2009), and Baradaran Shoraka As observed in Fig. 1, collapse assessment
(2013) provide insights into the phenomenon of requires models that can predict strength reduc-
dynamic collapse of earthquake-excited struction under large displacements, as well as
tures. For basic concepts in prediction of collapse strength and stiffness deterioration due to cyclic
and its implementation to performance-based loading.
earthquake engineering, the reader is referred to This article focuses on sidesway collapse of
Zareian et al. (2010). Villaverde (2007) presents a frames, which has been traditionally estimated by
comprehensive literature that provides a path means of non-degrading systems, in which judg-
through which seismic collapse assessment has mental limits are assigned to response quantities
evolved in the last decades. such as roof drift, story drift, ductility, and hys-
The required accuracy and the available teretic energy. These parameters are usually
resources determine the selection of a seismic referred to as engineering demand parameters
collapse assessment procedure. In a first step, (EDPs) (Moehle and Deierlein 2004). The devel-
the governing seismic modes of collapse such as opment of degrading models permits tracking the
vertical collapse or sidesway collapse are identi- collapse limit state, but EDPs become very sen-
fied. Then a structural model is created that sitive when the system is close to collapse, and
Seismic Collapse intensity of the ground motion

Assessment, Fig. 2 IDA Sa(T )
curve for a single ground g strength parameter Vy / W
motion up to collapse:
intensity measure (IM)

elastic, inelastic, and CC
EDP-sensitive domain; collapse
collapse (With permission
relative
from Adam 2014)
EDP sensitive domain
inelastic domain
engineering demand parameter (EDP)

elastic domain
small perturbations in input parameters lead to needs to be evaluated in a probabilistic frame-

large variations in the response. For this reason, work that includes uncertainties in the frequency
global collapse methodologies are commonly content of the ground motions (e.g., record-to-
based on an intensity measure (IM) instead of record (RTR) variability) and in the input param-
an EDP. A relative IM is the ratio of ground eters of the system (e.g., epistemic uncertainty).
motion intensity (demand) to a structure strength Then, nonlinear response history analyses usually
parameter (capacity). For a given structure and include a set of ground motion records that rep-
ground motion, collapse evaluation consists of a resents the RTR variability at the site of the
series of response history analysis starting with a structure. The effect of uncertainty in nonlinear
relative intensity that produces an elastic deterioration parameters on the variance of col-
response of the system. Then the relative inten- lapse capacity can be quantified by means of
sity is increased to cover the full range of interest approximate methods (e.g., FOSM method) or
or until collapse takes place. This series of carrying out Monte Carlo simulations (Ibarra
response history analysis is called incremental and Krawinkler 2011; Ugurhan et al. 2014).
dynamic analysis (IDA) (Vamvatsikos and Cor- The variance in collapse capacity can be used
nell 2002). As an example, Fig. 2 shows an IDA to generate collapse fragility functions that pro-
curve that results from increasing the IM up to vide the probability of collapse given that an
collapse. Collapse is imminent when a very small event with a certain IM occurred. Shome and
increment in relative intensity causes a very large Cornell (1999) summarize the arguments used
increment in the EDP of the system, a condition to consider the collapse fragility as a S
that indicates dynamic instability. In Fig. 2 log-normally distributed function. For simplified
dynamic instability occurs when the slope of the procedures, only the median, 16th, and 84th per-
IDA curve approaches zero. centiles of the individual collapse capacities are
The global seismic collapse capacity of a sys- required to find an approximation of the collapse
tem is defined as the largest earthquake intensity fragility. A final product of collapse assessment is
at which the structure still maintains dynamic the mean annual frequency of collapse, which is
stability (Krawinkler et al. 2009), denoted as CC obtained by combining fragility curves with the
in Fig. 2. This definition is associated with a hazard information at the site.
specific structure under a given ground motion. In this contribution some of the issues
Dynamic analysis requires the specification of involved in seismic sidesway collapse assess-
system properties and input ground motions, but ment of moment-resisting frames are addressed
variations in these quantities may produce large in more detail, including nonlinear dynamic anal-
dispersion in the response. Therefore, collapse ysis, structural modeling, strength and stiffness
material deterioration, and P-delta effect. selecting ground motions is avoided (Baker
The probabilistic framework used to evaluate 2013). These records need to be scaled to the
collapse and the effect of epistemic uncertainties same IM, and since there is not a unique defini-
and aleatory variability on the collapse capacity tion, several IMs have been proposed: (a) elastic
variance are discussed, although ground motion ground motion-based scalar IMs such as peak
selection is not part of this article. The last sec- ground acceleration (PGA), peak ground velocity
tions deal with the development of simplified (PGV), and peak ground displacement (PGD);
methods to evaluate collapse capacity. (b) elastic and inelastic spectral-based IMs such
as spectral acceleration and spectral displace-
ment at the fundamental period of the structure,
Nonlinear Dynamic Analysis as well as spectral values related to higher mode
effects or period elongation; and (c) vector
In the last two decades, significant progress has valued IMs (e.g., Baker and Cornell 2005).
been achieved in the nonlinear dynamic analysis Currently, the most widely used IM is the 5 %
of structures. It is now established as the most damped spectral pseudo-acceleration Sa(Τ1,
reliable tool to predict realistically the seismic z = 0.05) at the fundamental period T1 of the
collapse capacity of a building, and several pro- structure. For example, if Sa(Τ1,z = 0.05) is nor-
visions such as FEMA P-695 (2009), ATC-58 malized by the product of the acceleration of
(2012), and ATC-78 (2013) propose collapse gravity g and the base shear coefficient g, the
assessment methodologies based on nonlinear relative seismic collapse capacity of the building
dynamic analysis. subjected to the ground motion “i” reads as

Incremental Dynamic Analysis Sa, i
CCi ¼ (1)
If sidesway collapse is the governing failure gg collapse
mechanism, the IDA procedure can be used to
determine the largest earthquake intensity at The base shear coefficient g is defined as the base
which the structure still maintains stability. In shear Vy at onset of yielding divided by the total
this approach nonlinear dynamic response history weight W.
analyses are performed repeatedly, increasing in The collapse fragility function is obtained by
each subsequent run the ground motion intensity. counting the fraction of collapse capacities CCi
The IDA curve of the structure for the considered (i = 1,2,. . .,n) that correspond to the n ground
earthquake record is obtained by plotting an motions of the considered set. When all ground
appropriate record intensity measure against its motions lead to collapse, the collapse capacity
corresponding EDP (Vamvatsikos and Cornell distribution is usually log-normal distributed
2002). The analysis can be stopped when the (Ibarra and Krawinkler 2005). As an example,
EDP satisfies a certain failure criterion that may Fig. 3 shows the IDA curves of a structure
correspond to structural sidesway collapse. In this subjected to the 44 ground motions of the
case, the corresponding intensity of the ground FEMA P-695 far-field ground motion suite and
motion is referred to as collapse capacity of the the corresponding collapse fragility function.
building for this specific ground motion record.
IDA is commonly performed for a representa- Multiple Stripe Analysis
tive set of ground motions to account for RTR The IDA approach cannot consider certain site-
variability. Examples of such sets are the LMSR specific ground motion conditions. As outlined
ground motions (Medina and Krawinkler 2003) by Baker (2013), some of the ground motion
and FEMA P-695 sets for far-field and near-field properties relevant for collapse assessment are
ground motions (FEMA P-695 2009). In this the spectral shape of the elastic response spec-
sense, the ground motions are used as standard- trum, the duration, and the effects of near-fault
ized excitation protocols, and the effort of ground motions and velocity pulses. The
a b
Seismic Collapse Assessment, Fig. 3 (a) IDA curves fragility function according to a log-normal distribution
of a structure for all records of a ground motion set. (b) (With permission from Adam 2014)
Counted collapse fragility and corresponding fitted
expected spectral shape of a ground motion does The counted fraction of observed individual col-
not only vary with location but it also depends lapses at each IM level leads then to the collapse
strongly on the ground motion amplitude (Baker fragility curve.
and Cornell 2006). For instance, at different seis-
mic intensity levels, the spectral shape of the Collapse Criteria
expected earthquake is dissimilar. These differ- The numerical sidesway collapse prediction
ences in the spectral shapes can be ascribed to the depends on the model capabilities. Historically,
soil condition, magnitude, distance, and parame- the first nonlinear numerical models did not
ter e, which quantifies the difference between the include deteriorating characteristics, and damage
selected ground motion’s spectral acceleration at indexes were developed to predict collapse based
a specific period and the median of the ground on maximum displacements and/or dissipated
motion spectral acceleration obtained from atten- energy. The most widespread cumulative damage
uation relationships (Baker 2013). Because the model was the Park–Ang model (Park and Ang
same set of records are often used for different 1985), which was developed specifically for
locations, however, IDA yields the same risk of reinforced concrete components. The damage
sidesway collapse for two equal buildings located measure (DM) consisted of a linear combination
in different locations with similar seismic hazard, of displacement and energy demands:
although the anticipated ground motion spectral S
ð
shapes are different. dmax b
DM ¼ þ dHE (2)
To address some of the above issues, Jalayer dult Fy dult
and Cornell (2009) developed a multiple stripe
analysis (MSA) that accounts for specific site where dmax is the maximum displacement of the
conditions and for ground motion properties that system, dult denotes the monotonic displacement
change with increasing magnitude. In this capacity of the system, b represents the structural
approach, the ground motions are reselected at performance parameter, and HE is the hysteretic
each IM level according to the changing spectral energy. In concept, the parameters dult and b of
shape. For each record the peak response of the this cumulative damage model are equivalent to
EDP derived from nonlinear response history the deteriorating parameters of recent hysteretic
analysis and the number of events that leads to models that include cumulative deterioration.
sidesway collapse is plotted as a function of the However, damage index models were based on
corresponding IM, thus leading to “stripes.” non-deteriorating models, leading to unrealistic
Seismic Collapse unconservative collapse prediction

Assessment, Fig. 4 Two IM
IDA curves for the same
non-degrading model
building and ground motion
record based on two degrading model
different structural models. CCi(n)
Definition of the collapse CCi(u)
capacity: CC(u)
i collapse CCi(d )
capacity associated with "near collapse" limit state
infinite deformation. CC(d)
i ,
CC(n)
i : collapse capacity predefined EDP threshod collapse limit state
predictions based on a
predefined EDP limit (With
permission from Adam
2014)
EDP
displacement–force relationships for large inelas- criterion such as the exceedance of a predefined
tic excursions. EDP threshold. In particular non-ductile collapse
In recent practical applications, when modes such as column shear failure and subse-
non-degrading models are used to predict col- quent vertical collapse are frequently considered
lapse, this limit state is associated with the defor- as non-simulated collapse modes (e.g., Liel and
mation of a single structural component that Deierlein 2013).
exceeds a predefined EDP threshold, often The IDA curves of Fig. 4 illustrate sidesway
represented by a plastic rotation demand collapse prediction for degrading and non-
(Haselton et al. 2009). Eurocode 8 (CEN 2005) degrading structural models. The intensity CC(u) i
provisions, for instance, are based on capacity is the ultimate collapse capacity associated with
thresholds for the deformation of RC members infinite EDP deformations for the degrading
at yielding and failure developed by Fardis and model, and it corresponds to the “exact” collapse
collaborators (e.g., Panagiotakos and Fardis capacity. However, if a predefined deformation
2001; Fardis and Biskinis 2003). As a further threshold is used to define collapse, the use of the
example, ASCE/SEI 41-13 (2014) procedure degrading model always results in a conservative
compares the component demands of an existing near-collapse prediction. That is, the intensity at
building with component acceptance criteria that this EDP threshold, CC(d) (u)
i , is smaller than CCi .
define collapse prevention. In such an approach, On the other hand, the use of a non-degrading
the component that fails first governs the state of model may underestimate or overestimate
the entire structure. That is to say, component the actual collapse capacity, depending on the
capacity thresholds do not capture the global preestablished EDP threshold. In Fig. 4 the
behavior and the capability of the structure to corresponding intensity denoted as CC(n) i results
redistribute forces after failure of one component. in a nonconservative prediction because the
A structure may exhibit several modes of predefined EDP threshold is rather large.
seismic-induced collapse such as sidesway col- Baradaran Shoraka (2013) developed a robust
lapse or vertical collapse, depending on the his- collapse assessment framework to predict verti-
tory of spread of inelastic deformations, load cal collapse, such as column shear failure and
distribution, etc. If the corresponding degrading subsequent axial failure in non-ductile reinforced
structural model captures only a specific type of concrete frames. In this approach, progression of
collapse such as sidesway collapse, the other damage is tracked throughout the numerical anal-
collapse mechanisms that might occur are ysis by comparing stepwise floor level gravity
referred to non-simulated modes of collapse. demands with corresponding capacities. Explicit
That is, they only can be predicted by an artificial modeling of progressive collapse involves
element (commonly column) removal until the these phenomena may govern the deterioration
structure attains a state that cannot resist the process, strain “artificial” limits need to be spec-
vertical load demand. Baradaran Shoraka ified in fiber models (Krawinkler et al. 2009).
et al. (2014) also compared the results of explicit In a lumped plasticity model, locations within
gravity load collapse modeling and those the structure that might undergo inelastic defor-
obtained from nonlinear acceptance collapse pre- mations in a severe seismic event are identified
vention criteria specified in ASCE/SEI 41-13 up front. Inelastic nonlinear zero-length springs
(2014) for retrofit of non-ductile frames. are assigned to these locations, while the rest of
the model components remain elastic. These sim-
ple phenomenological component models are
Structural Modeling Strategies for calibrated with experimental outcomes to repre-
Assessing Sidesway Collapse sent appropriate cyclic response behavior.
For instance, in a moment-resisting frame,
Structural Modeling on the System Level rotational springs are placed at the end of beams
The structural model may be two-dimensional or and columns. Frames are designed according to
three-dimensional, but when the building the strong column-weak beam philosophy, in
exhibits a coupled bending-torsional response, a which beams and bottom end of base columns
three-dimensional system better predicts the are intended to exhibit inelastic deformations at
building seismic behavior. Such a system their ends when subjected to strong ground
requires sophisticated modeling of the nonlinear motions. However, many columns may undergo
constitutive behavior at the component level that inelastic deformations at the story level when
is not well understood. Thus, three-dimensional approaching the collapse limit state. Thus,
models are avoided for collapse analysis as long nonlinear spring elements need to be placed at
as it is feasible, and a more generic these locations to accurately predict the seismic
two-dimensional structural model is utilized. collapse capacity.
As outlined by Krawinkler et al. (2009), there A concentrated plasticity model includes
are two practical modeling strategies for frame beam/column elements of different degree of
structure evaluation: fiber models and lumped sophistication. In the simplest case, all elastic,
plasticity models. In a fiber model, an axial inelastic, and deteriorating properties of the ele-
stress–strain relation is separately assigned to ment are assigned to rotational springs at both
each fiber of the cross section. For example, dif- ends of the element, whereas the element itself
ferent constitutive laws can be used in reinforced is considered to be rigid; see Fig. 5a. This model
concrete sections to represent confined and may capture the development of plastic hinges.
unconfined concrete and the longitudinal steel As a disadvantage the spring stiffness must be
bars. In this manner, the spread of inelastic defor- defined a priori, which is a constant regardless of S
mations within the structural component is cap- the potential changes in the moment gradient of
tured. In many situations this approach can the element. To avoid this drawback, Ibarra and
accurately predict earlier limit states such as Krawinkler (2005) suggest to keep the beam/col-
yielding and crack initiation, among others. How- umn element elastic and to add rotational springs
ever, several problems have been observed in at the ends, whose rotational stiffness is several
fiber models. Since these models usually are times stiffer than the rotational stiffness of the
based on axial stress–strain relations, shear defor- elastic beam/column element. In this model, an
mations cannot be modeled appropriately. elastic beam/column element is connected to two
According to Krawinkler et al. (2009), further springs at the ends, and thus, the properties of the
phenomena that are difficult and often impossible parameters can be calibrated according to the
to capture are, for example, bond slip, rebar frac- component “real” behavior (Fig. 5b). Jäger and
tion and rebar buckling, fracture of concrete, and Adam (2012) showed that the latter element can
local and lateral torsional buckling in steel. Since describe column and beam behavior sufficiently
beam/column element model constitutive behavior of constitutive

the element between the springs spring behavior
a M M
Ksp Ksp My
rigid
y
b M M
mod mod
Ksp Ksp
J mod My
mod
y,sp
Seismic Collapse Assessment, Fig. 5 Different beam/ rotational springs. (b) Modified elastic element with two
column elements and their constitutive behavior to mono- nonlinear rotational springs (Modified from Jäger and
tonic loading. (a) Rigid element with two nonlinear Adam 2012)
accurate for collapse assessment. The rigid ele- nonlinear dynamic analysis. Several studies
ment with two nonlinear springs at its ends should have revealed that inappropriate consideration
be used for beams only because when it is used in of viscous damping leads to spurious damping
columns, the elastic axial-bending loading inter- forces that distort the nonlinear dynamic response
action cannot be taken into account. prediction. Particularly affected are lumped plas-
A shortcoming of lumped plasticity models is ticity models with elements that exhibit large
that the nonlinear column spring strength remains initial stiffness (Charney 2008). According to
constant during the response history analysis, the ATC-72-1 report (ATC-72-1 2010), modal
independently of the moment - axial force damping is the preferred choice for modeling of
interaction. damping in inelastic structures, but in many soft-
Figure 6a shows a concentrated plasticity ware packages, this is not possible. In such cases
model for a multistory three-bay steel moment- a combination of mass and stiffness proportional
resisting frame, whose plan configuration is damping (e.g., Rayleigh damping) should be
shown in Fig. 6b. At both ends of each beam used. Modeling guidelines for damping in inelas-
and column, a nonlinear rotational spring is tic structures are provided in ATC-72-1 (2010).
arranged with an offset from the node to account
for the member’s cross section. In contrast, Structural Modeling of Cyclic Component
generic frame models used for research purposes Behavior
are often “centerline models,” i.e., the lumped Sidesway collapse assessment based on lumped
plasticity elements are located at the nodes with- plasticity structural models requires hysteretic
out any offset from the nodal point. models that include all the important modes of
The structural model of Fig. 6a includes a deterioration observed in experimental tests. As
leaning column with zero flexural stiffness to illustrated in Fig. 7, monotonic tests on structural
account for the P-delta effect resulting from grav- components show that strength of the backbone
ity loads that are not directly applied to the frame curve is “capped” and is followed by a negative
(Haselton et al. 2009). This leaning column is tangent stiffness. In addition, the cyclic response
loaded with a vertical load at each story level indicates that strength and stiffness deteriorate
representing half of the total system gravity load with the number and amplitude of cycles. As a
that is not directly tributary to the columns of the result, the strength predicted by the monotonic
frame (NIST GCR 10-917-8 2010), as shown in backbone curve is usually not reached during
Fig. 6b. cyclic loading. Several hysteretic models have
Another issue that needs to be properly been developed to represent this behavior (e.g.,
addressed is modeling of viscous damping for Song and Pincheira 2000). However, few models
a
xN
N
leaning
lumped column
mass
concentrated
nonlinear
spring elements
tributary area for gravity loads

applied to the leaning column
tributary area for gravity loads
applied to the frame columns
moment-resisting frame
Seismic Collapse Assessment, Fig. 6 Example of resisting steel frame. (b) Rectangular plane configuration
structural modeling of a frame building. (a) Overall of the frame building (Modified from NIST GCR
lumped plasticity structural model of the moment- 10-917-8 2010)
S
integrate strength deterioration in the backbone bilinear model (Fig. 8a) is based on the standard
curve, as well as cyclic deterioration of strength hysteretic bilinear rules with kinematic strain
and stiffness. This section describes exemplarily hardening, but a “strength limit” is introduced
the deteriorating models developed by Ibarra when the backbone curve includes a branch with
et al. (2005) for bilinear, peak-oriented, and negative slope. The limit for strength is the
pinched hysteretic systems. smallest strength reported on the post-capping
Figure 8 presents bilinear, peak-oriented, and branch in previous excursions. If this condition
pinching deteriorating models based on a back- were not established, the strength in the loading
bone curve that includes a post-capping negative path could increase in later stages of deteriora-
slope, and sometimes a residual strength, as tion. Note that the original peak-oriented and
shown in Fig. 7. The deteriorating models pre- pinching models exhibit stiffness deterioration
serve most of the rules of the original in the reloading path. However, in this article
non-deteriorating models. For instance, the the term “deterioration” is used for hysteretic
Seismic Collapse moment M

experimental
Assessment, monotonic loading curve
capping strength Mc
Fig. 7 Initial backbone yield strength My initial backbone curve
curve for the modified
Ibarra–Krawinkler model
(Modified from Krawinkler residual strength Mr
et al. 2009)
y c r u chord rotation
ultimate
deformation
a moment M b moment M
2 2,8 9
M y+ 1
M y+
1 10
strength limit 6⬘
3
6
0 7 0 3 11
− + − +
c0 c0 cord rotation c0 c0 cord rotation
4
My 6 My
4
5 5
moment M Mmax
c 2
My+ 1
"break point"
8
8' 3 Mp2 = κ f Mmax

7 0
Mp1 = κ f My cord rotation
4 4' d per1
per1
My
5
Mmin
6
Seismic Collapse Assessment, Fig. 8 Basic deteriorating hysteretic models: (a) bilinear, (b) peak oriented, and (c)
pinching (Modified from Ibarra and Krawinkler 2005)
models that possess a post-capping stiffness oriented one, but the reloading consists of two
branch in the backbone curve. These models parts. First the reloading path is directed toward a
may also be subjected to cyclic deterioration. “break point,” which is a function of the maxi-
The peak-oriented model (Fig. 8b) keeps the mum permanent deformation and the maximum
basic hysteretic rules proposed by Clough and load experienced in the direction of loading. The
Johnston (1966), but the backbone curve is mod- break point is defined by the parameter кf that
ified to include strength capping and residual modifies the maximum “pinched” strength
strength. The negative post-capping stiffness (points 4 and 8 of Fig. 8c) and the parameter кd
does not modify any basic rules of the model. that defines the displacement of the break point
The pinching model is similar to the peak- (points 40 and 80 ).
Component Deterioration Based on Hysteretic negative excursions, and Et,m denotes the hyster-
Energy Dissipation etic energy dissipation capacity for the cyclic
Consideration of cyclic component deterioration deterioration mode “m.” The original hysteretic
is important for collapse prediction, especially rules defined Et,m as a function of twice the elastic
for short-period highly ductile systems. The strain energy, Et, m ¼ gm My ’y (Ibarra and
Ibarra–Krawinkler hysteretic model includes Krawinkler 2005). The modified
four cyclic deterioration modes based on energy Ibarra–Krawinkler model defines Et,m as a func-
dissipation that are activated once the yielding tion of the plastic chord rotation ’p,
point is surpassed. As observed in Fig. 9, basic Et, m ¼ lm My ’p , or Et, m ¼ Lm ’p (Lignos and
strength and post-capping strength deterioration Krawinkler 2012). The parameters gm, lm, and
effects translate the strain-hardening and post- Lm are calibrated from experimental tests, and
capping branch toward the origin, unloading reasonable results are obtained if all cyclic dete-
stiffness deterioration decreases the unloading rioration parameters have the same value. Expo-
stiffness, and reloading (accelerated) stiffness nent cm is the rate of deterioration parameter. cm
deterioration increases the target maximum = 1 implies a constant rate of deterioration.
displacement. The parameter bm,i modifies the strength and
The amount of deterioration depends on the stiffness properties of the previous cycle. For
parameter bm,i, which is based on the hysteretic basic strength deterioration, for instance, the
energy dissipated when the component is strain-hardening branch moves toward the origin
subjected to cyclic loading (Ibarra and by an amount equivalent to reducing the yield
Krawinkler 2005), strength as follows (for this mode index m = s):
0 1 cm þ
Mþ
y, i ¼ 1 bs, i M y, i1 (4)
B C
B Ei C
bm, i ¼ B
B
C
C (3) where Mþ y, i is the deteriorated yield strength after
@ Xi
A
Et, m Ej excursion i and Mþ y, i1 is the deteriorated yield
j¼1 strength before excursion i. Except for the
unloading stiffness mode, a positive and a nega-
where Ei is the
Xhysteretic energy dissipated in tive value is defined for each deterioration param-
excursion i, Ej represents the hysteretic eter because the algorithm deteriorates the
energy dissipated in all previous positive and strength independently in both directions.
Seismic Collapse moment M

Assessment, Fig. 9 Four ϕp ϕpc
modes of component S
deterioration in a pinching reloading stiffness
component (Modified from deterioration
Lignos and Krawinkler
2013; With permission
from American Society of Mr+ = My+
Civil Engineers) unloading stiffness
deterioration
ϕr ϕu chord rotation ϕ
M r− = M y−
post-capping
strength deterioration
strength deterioration
Seismic Collapse
Assessment,
Fig. 10 Inelastic structural plastic drift
response history with
displacement
ratcheting effect due to
P-delta (With permission
from Adam 2014) time
collapse
That is to say, M
y, i is updated after every positive However, this consideration not only depends
inelastic excursion, and Mþ y, i is updated after on the structure characteristics but also on the
every negative inelastic excursion. For example, ground motion applied to the structure. For exam-
bs,i is calculated each time the inelastic path ple, if a strong-motion earthquake causes struc-
crosses the horizontal axis. These parameters tural collapse after a few load cycles, component
are calibrated from experiments. For instance, degradation is of minor significance compared to
Lignos and Krawinkler (2012, 2013) conducted a situation where a long-duration earthquake
deterioration modeling of steel components exhibits many cycles before the structure fails.
based on an experimental database. In an inelastic SDOF system, the gravity load
generates a uniform shear deformation of its hys-
teretic force–displacement relationship, as shown
Global Seismic Sidesway Collapse and exemplarily in Fig. 11, for the backbone curve of
the P-Delta Effect an inelastic SDOF system. Characteristic displace-
ments (such as the yield displacement) of this
In a vibrating structure, gravity loads acting relationship remain unchanged, whereas the char-
through large horizontal deflections induce addi- acteristic forces (such as the strength) are reduced.
tional bending moments, which in turn increase As a result, the slope of the elastic and post-elastic
the lateral deflections. In structural analysis, this branch rotates. The magnitude of this reduction
destabilizing effect of gravity loads (i.e., global can be expressed by means of the stability coeffi-
P-delta effect) is usually interpreted as a reduc- cient y (MacRae 1994). The parameter y is a
tion of the lateral structural stiffness by the function of the gravity load, geometry, and stiff-
so-called geometric stiffness. Generally, for a ness. For instance, for the bilinear SDOF system
real building within its elastic range of deforma- shown in Fig. 12a, the stability coefficient reads as
tion, this stiffness decrease is of minor impor-
tance because its magnitude is small compared Ph
y¼ (5)
to the first-order elastic stiffness. During severe kr
seismic excitations, however, inelastic deforma-
tions combined with gravity may cause a struc- where P is the gravity load, h the length of the
ture to approach a state of dynamic instability if rod, and kr the initial stiffness of the elastoplastic
the post-yield tangent stiffness becomes nega- rotational spring at the base. In the example of
tive. In such a condition, the displacement Fig. 12, the post-yield stiffness is negative,
response tends to amplify in a single direction because the stability coefficient y is larger than
due to the ratcheting effect, as shown in Fig. 10. the hardening ratio a. In a nonmaterial degrading
Very flexible structures are especially affected by SDOF system, a negative post-tangential stiff-
seismic-induced sidesway collapse in the P-delta ness ratio, i.e., y a > 0, is a necessary condition
mode, and in many of these cases, cyclic compo- for potential structural collapse under severe
nent material deterioration can be disregarded. earthquake excitations.
Seismic Collapse M
Assessment,
Fig. 11 P-delta effect on
the backbone curve of an
inelastic SDOF system
(Modified from Ibarra and no P-delta
Krawinkler 2005)
s ke
with P-delta
( s ) ke
ke c ke
(1 θ ) ke
( c ) ke
θ ke
a x(t) P b
f no P-delta
m a
1
q
a-q
(t ) rigid, h
1 μ
1
kr,rr
with P-delta
xg (t )
S
Seismic Collapse Assessment, Fig. 12 (a) Model of an SDOF system vulnerable to P-delta. (b) Normalized bilinear
cyclic structural behavior of an SDOF system with and without P-delta (With permission from Adam and Jäger 2012a)
Adam and Jäger (2012a) showed that the col- and Adam (2013) conducted a rigorous paramet-
lapse capacity of a P-delta vulnerable inelastic ric IDA study to reveal the impact of character-
SDOF system with non-deteriorating hysteretic istic parameters (i.e., period T, negative post-
behavior and bilinear backbone curve is a func- yield stiffness ratio y a, damping z, and hys-
tion of the negative slope of the post-tangential teretic loop) on the collapse of those systems,
stiffness ratio y a, the elastic structural period presenting the outcomes in the form of “collapse
of vibration T, the viscous damping coefficient z, capacity spectra.”
and the shape of the hysteretic loop such as bilin- In a collapse capacity spectrum, the P-delta
ear, peak oriented, or pinching. Based on this vulnerable SDOF system collapse capacity is
observation, Adam and Jäger (2012a) and Jäger represented as a function of the initial period of
a 10 b 10
FEMA P-695 far-field set FEMA P-695 far-field set
bilinear hysteretic loop bilinear hysteretic loop
8 θ − α = 0.10 8
ζ = 0.02
ζ = 0.02
median
6 T = 1.2 s 6
S a / g /γ
CC
84th percentile
4 4
median
θ − α = 0.10
2 2
16th percentile
0 0
0 2 4 6 8 10 0 1 2 3 4 5
max|x|/xy period T [s]
Seismic Collapse Assessment, Fig. 13 (a) IDA curves set and statistically evaluated results. (b) Corresponding
of an SDOF system with parameters defined in the figure median “instability collapse” capacity spectrum (With
for all records of the FEMA P-695 far-field ground motion permission from Adam 2014)
Seismic Collapse 10
Assessment,
Fig. 14 Median collapse FEMA P-695 far-field set
capacity spectra (blue lines) bilinear hysteretic loop
8
for a set of negative post-
ζ = 0.02
yield stiffness ratios and
corresponding smooth median 0.04 θ −α
collapse capacity spectra 6
(With permission from
CC d
Adam 2014) 0.06

4
0.10
CCd θa − α
2 0.20
0.80
0
0 1 2 3 Ta 4 5
period T [s]
vibration T for a fixed set of characteristic param- 84th percentile collapse capacity spectra
eters. For example, Fig. 13a shows the individual (Fig. 13b). Based on a collapse capacity spectrum
IDA curves of an SDOF system for the 44 ground database, Adam and Jäger (2012a) and Jäger and
motions of FEMA P-695 far-field set (in Adam Adam (2013) formulated analytical expressions
and Jäger (2012a) referred to as ATC63-FF set) of median, 16th, and 84th percentile collapse
up to “instability collapse,” as well as median, capacity spectra with a “smooth” shape via
16th, and 84th percentile curves. Collapse capac- nonlinear regression analyses, as shown in
ity spectra are obtained by repeating this proce- Fig. 14.
dure for different vibrational periods within a In multistory frames, gravity loads may sub-
certain interval. As before, collapse capacity stantially impair the complete structure or only a
spectra can be computed for a specific suite of subset of stories (Gupta and Krawinkler 2000).
earthquake records, leading to median, 16th, and The local P-delta effect may induce collapse
xN
N
V
bilinear idealization
no P-delta effect
aSKS
Vy
1
VPy 1
qi KS
1 (aS − qi )KS
qeKS
(1−qe)KS with P-delta effect
1
V xNy xN
Seismic Collapse Assessment, Fig. 15 Global pushover curves disregarding P-delta (black graphs) and considering
P-delta (red graphs) (With permission from Adam 2014)
of a structural element, without necessarily disregards gravity loads in the nonlinear static
affecting the stability of the complete structure. analysis.
An indicator of the severity of the local P-delta Global pushover curves of multi-degree-of-
effect is the story stability coefficient proposed in freedom (MDOF) structures do not exhibit a uni-
the Eurocode 8 (CEN 2005). However, a consis- form stability coefficient in the entire range of
tent relationship between the local P-delta effect elastic and inelastic deformations, unlike SDOF
and the global P-delta effect, which characterizes systems. As observed in Fig. 15, two stability
the overall impact of gravity loads on the coefficients can be identified in a bilinear approx-
structure, cannot be established due to dynamic imation of pushover curves with and without
interaction between adjacent stories in a P-delta effect, that is, a stability coefficient in
multistory frame structure (Gupta and the elastic range of deformation (ye) and a stabil- S
Krawinkler 2000). ity coefficient in the post-yield range of deforma-
Strong evidence for the vulnerability of a reg- tion (yi). According to Medina and Krawinkler
ular frame structure to P-delta-induced global (2003), yi can be much larger than ye: yi > (>)ye.
seismic collapse is obtained from a first-mode
pushover analysis (Adam and Jäger 2012b). For
instance, the red curve of Fig. 15 illustrates the Variance of Collapse
global pushover curve of a very flexible structure
that exhibits a negative post-yield stiffness due to Earlier studies on quantification of epistemic
effect of gravity loads. When severe seismic exci- uncertainties on the seismic response concluded
tation drives the structure into its inelastic branch, that the contribution of epistemic uncertainties to
a state of dynamic instability may be approached, the variance of the system’s seismic performance
and the global collapse capacity is attained at a is in general much smaller than that from aleatory
rapid rate. The pushover curve plotted in black uncertainties, and thus it can be omitted
Seismic Collapse 1.0

Assessment, P(CC| total,FOSM )
Fig. 16 Collapse fragility 84th perc.
curves considering the RTR
0.8 counted
uncertainties only (red line,
red markers) and combined collapse fragility
probability of collapse
RTR and epistemic P(CC|total,LHS)
uncertainties (black lines) 0.6
median
0.4
0.2 16th perc. P(CC| RTR )
0.0
0 2 4 6 8 10 12
collapse capacity
(e.g., Esteva and Ruiz 1989). In contrast, recent For deteriorating models, however, epistemic
studies indicate that the epistemic uncertainty uncertainty can be relevant for collapse capacity
caused by degrading nonlinear system parame- evaluation because of the large uncertainty on
ters is very relevant when assessing the total degrading nonlinear parameters, such as plastic
variance of the collapse limit state (Ibarra and rotation and post-capping stiffness (Fig. 17)
Krawinkler 2005; Zareian et al. 2010). For exam- (Ibarra and Krawinkler 2011; Ugurhan et al. 2014).
ple, Haselton et al. (2011) evaluated the total The first-order second-moment (FOSM)
variance of collapse capacity of a four-story method (e.g., Melchers 1999) can be used to
reinforced concrete special moment frame, compute the additional variance of collapse
including RTR variability and modeling uncer- capacity due to uncertainty in the system param-
tainties. He estimated the logarithmic standard eters. The total variance of the collapse capacity,
deviation associated with modeling uncertainties s2ln CC(TOT), based on FOSM is (Ibarra and
as 0.45, a value comparable to the standard devi- Krawinkler 2011):
ation of the logarithm of RTR variability, which
n

varies in general between 0.25 and 0.45 (Liel Xn X
@g X @g X
et al. 2009; Adam and Jäger 2012a). As an exam- s2lnCCðTOTÞ ffi
i¼1 j¼1
@xi @xj X¼mx
ple, Fig. 16 shows the collapse fragility of a
P-delta vulnerable SDOF structure as shown in rxi , xj sxi sxj þ s2lnCCðRTRÞ
Fig. 12a, considering the RTR uncertainty only, (6)
and the total variability considering RTR and
epistemic uncertainties. The first term on the right hand side of Eq. 6
The effect of RTR variability on the variance of represents the contribution to the variance of
collapse is directly obtained by performing IDAs collapse capacity due to uncertainty of the system
for a set of ground motions on a deterministic parameters, whereas s2ln CC(RTR) is the contribu-
system. The variance of collapse capacity of tion to the variance of collapse capacity due to
nonmaterial deteriorating systems is also affected RTR variability, and X represents the set of
by uncertainty in parameters, such as the yield system random parameters. The function g X
moment and post-yield hardening ratio, but their is the collapse capacity as a function of the vari-
effect on the variance of collapse capacity is small ation on system parameters, and the variance is
compared to that originated by RTR variability. computed from the gradient of g X , which is
Seismic Collapse 16th 50th 84th

Assessment,
uncertainty in
Fig. 17 Uncertain
parameters of the backbone plastic rotation c
curve (Modified from moment M
Ibarra and Krawinkler
2005) Mc
My
uncertainty in
post-capping stiffness, c
y c chord rotation
linearized using a Taylor series expansion

about the first moments of a function are calculated in
the mean X . The derivative @g X =@xi can be terms of the first moments of the random vari-
approximated by evaluating the performance ables (Rosenblueth 1981). The total variance in
function at two system parameter values. The collapse capacity can be more accurately
derivative is estimated as the slope of the straight predicted using Monte Carlo simulation, but this
line that connects the pairs [xi, g(xi)]. The stan- method is computationally expensive. An alter-
dard deviation of the system parameters, sxi , and native is to use the Latin hypercube sampling
the correlation coefficients, rxi , xj , need to be (LHS) method to reduce Monte Carlo realiza-
estimated in advance. tions. LHS is a stratified sampling that allows
Ibarra and Krawinkler (2005, 2011) used the efficient estimation of the quantity of interest by
FOSM method to evaluate the effect of uncer- reducing the variance of Monte Carlo simulations
tainty in system parameters on the collapse (Vamvatsikos and Fragiadakis 2010). In the
capacity of SDOF systems. They found that example of Fig. 16, the FOSM method and the
uncertainty in the displacement at the peak (cap) LHS method deliver almost the same prediction
strength and the post-capping stiffness contrib- of the collapse fragility function.
utes significantly to the variance of collapse
capacity. The uncertainty in deterioration param-
eters on the variance of collapse capacity may be Simplified Methods
comparable to that caused by RTR variability.
Although the FOSM method is relatively sim- IDA and MSA procedures require the numerical
ple to implement, it cannot predict the shift in the solution of the governing equations of motions of S
median caused by system uncertainties. the structural model together with constitutive
Vamvatsikos and Fragiadakis (2010) concluded equations in each time step for each response
that this shortcoming should not be considered history analysis. As a consequence, it is time-
important for practical applications because for consuming and computationally expensive.
more cases the shift in the median is less than Thus, there is a need for simple to apply, but yet
10 %. Liel et al. (2009) found shifts in the median sufficient accurate methods to predict the global
closer to 20 % and proposed a simplified method, collapse of MDOF structures under seismic exci-
termed ASOSM (approximate second order sec- tations (Lignos and Krawinkler 2012).
ond moment), that uses FOSM to predict the
increase in fragility’s logarithmic standard devi- Methods Based on Nonlinear Static Analysis
ation and the shift in the median of the limit state One of the simplest approaches is to conduct a
fragility. The total variance can also be approxi- nonlinear static analysis to estimate the seismic
mated using the point estimate method, in which collapse capacity. During this so-called pushover
V may be reasonable applied as an ingredient for

collapse
the collapse assessment of regularly shaped
low-rise buildings whose seismic response is
dominated by the first mode. Even if the pushover
procedure is not appropriate for a full collapse
assessment, it can be used to identify weaknesses
xN
in the structural design, to debug a structural
model used for collapse assessment, and to obtain
Seismic Collapse Assessment, Fig. 18 Global push- a better understanding of the strength and defor-
over curve up to global collapse (With permission from mation demands of the structure (Deierlein
Adam 2014) et al. 2010). In contrast, Villaverde (2007) states
that it is doubtful that nonlinear static methods
analysis, gravity loads are applied to the struc- can be used reliably to predict the seismic col-
tural model that considers explicitly the inelastic lapse capacity of structures and to estimate their
constitutive behavior of its elements. Subse- margin of safety against a global collapse. These
quently the model is subjected to lateral forces methods, however, are commonly utilized in
with a predefined invariant load pattern that is practical procedures. For example, Shafei
amplified incrementally in a displacement- et al. (2011) proposed a simplified methodology
controlled procedure. A first-mode load pattern for collapse assessment of degrading moment-
may be used. Alternatively, the load pattern resisting frames and shear walls utilizing
recommended in ASCE 7-10 (2010) can be uti- closed-form equations, given that the global
lized. As a result the global pushover curve of the pushover curve is provided. These equations cor-
structure is obtained, at which the base shear is relate median and dispersion of collapse fragility
plotted against the displacement of the control curves and were generated through multivariate
node (usually the roof displacement). The global regression analysis from a comprehensive data-
pushover curve represents the global capacity of base of collapse fragilities and pushover curves.
the building against horizontal loads, and it is
assumed that it reflects the global or the local Methods Based on Equivalent Single-Degree-
mechanism involved when the structure of-Freedom Systems
approaches dynamic instability. In the perspec- Several studies use equivalent single-degree-of-
tive of a “near-collapse” limit state, collapse is freedom (ESDOF) systems to predict the global
assumed if one of the selected EDPs, such as seismic collapse capacity (e.g., Bernal 1998;
story drifts, plastic hinge rotations, etc., exceeds Fajfar 1999; Adam et al. 2004). Application of
a certain threshold. When the base shear drops to ESDOF systems implies that the first mode dom-
zero (Fig. 18), global “instability collapse” is inates the dynamic response and the collapse
assumed to be attained. mode. A global collapse capacity assessment of
Pushover analysis was originally developed low- to medium-rise buildings by means of an
for first-mode dominated structures and later ESDOF system is straightforward when the story
refined to account also for higher mode effects drifts remain rather uniformly distributed over
(e.g., Chopra and Goel 2002). The basic assump- the height, regardless of the extent of inelastic
tion of these procedures is that the nonlinear deformation. Adam and Jäger (2012b) showed
static response can be related to the nonlinear that the application of these simplified systems
dynamic response. This is, however, not the yields reasonable collapse capacity predictions
case for many structures because the dynamic for P-delta-sensitive regular high-rise buildings,
response is strongly path dependent, and effects, because P-delta-induced collapse is primarily
such as cyclic deterioration, damping, and dura- controlled by the first mode. However, if a partial
tion of an earthquake, among others, cannot be mechanism develops, the global collapse capac-
captured. Nonetheless, nonlinear static analysis ity is greatly affected by the change of the
deflected shape, and it will be amplified in those of the displacements of the structure in its
stories in which the drift becomes large (Bernal ESDOF approximation (Fajfar 2002). Note that
1998). f should be affine to the horizontal load pattern
of the pushover analysis.
Capacity Spectrum Method (N2 Method) Dividing f* by the equivalent mass L* renders
In the last decade, the capacity spectrum method- its ordinate in the “acceleration” dimension; see
ology became a popular tool for assessing the Fig. 19. This transformation permits the compar-
seismic performance of regular first-mode domi- ison of the capacity curve with the seismic
nated structures. It represents a compromise demand represented by the response spectrum in
between nonlinear dynamic analyses with com- the ADSR format of the actual site. In this format
plex modeling strategies and simplified linear for an inelastic SDOF system with a target duc-
static analysis methods. In Eurocode 8 (CEN tility m, the spectral acceleration Sav at yield is
2005), a version of this method, i.e., the N2 plotted against the corresponding peak spectral
method (Fajfar 1999), is implemented. displacement Sd (Fajfar 1999), as shown in
In this approach a global lateral Fig. 20a. Note that the ductility defines the ratio
load–displacement relation, which is the outcome of the maximum imposed (inelastic) deformation
of a pushover analysis, represents the seismic to the deformation at onset of yield.
capacity of the structure and is referred to as In the subsequent step, the intersection point
capacity curve. The base shear V and the roof between a bilinear approximation of the capacity
displacement xN of a bilinear idealization of this curve and the response spectrum is obtained.
curve are transformed into the domain of the If this point is in the inelastic branch of deforma-
corresponding ESDOF system according to the tion, the elastic spectrum must be reduced such
following Eurocode 8 equations (CEN 2005) (see that the ductility of the capacity curve at this
Fig. 19): point and of the corresponding inelastic spectrum
coincide, Fig. 20b. For periods larger than the
f ¼ V=G, D ¼ xN =G, corner period Tc, this intersection point, which
XN X
N
is referred to as performance point, is found by
G ¼ L =m , L ¼ mi fi , m ¼ mi f2i
i¼1 i¼1 application of the equal displacement rule.
(7) According to Eurocode 8 (CEN 2005), struc-
tural stability under seismic excitation cannot be
Here, f* is the equivalent spring force, and verified if the displacement demand at the perfor-
D the displacement of the ESDOF system. The mance point is larger than two-thirds of the ulti-
equation includes the story masses mi, i = 1,. . ., mate deformation capacity of the structure or if
N, of the ductile N-story structure, and the no performance point can be found. Conse-
N components fi of the shape vector f (with quently, collapse is indicated, as presented in S
fΝ = 1), which prescribes the vertical distribution Fig. 20b.
Seismic Collapse pushover curve

Assessment,
bilinear idealisation
Fig. 19 Capacity curve
With permission from
V f * / L*
Adam (2014; Modified Vy f *y / L*
from CEN 2005)
Em
xNy Dy xN D
a Say response spectrum in ADRS format b Say

Tc performance point
μ = 1.5 μ =1 μ
assumed collapse
elastic f */ L*
μ=4
μ=2
μ=8
Sd D y Dt ( = μ Dy ) 2 Du Sd , D
D
3 u
Seismic Collapse Assessment, Fig. 20 (a) Elastic and corresponding constant ductility response spectra in the ADRS
format. (b) Performance point With permission from Adam (2014; Modified from Fajfar 1999)
Collapse Capacity Spectrum Method yi ye as

ya ¼ (8)
The collapse capacity spectrum method is 1 ye þ yi as
another example of a simplified ESDOF system-
based method. This method predicts the global yields the backbone curve considering P-delta in
collapse capacity of regular P-delta-sensitive analogy to a real SDOF system; see Fig. 21. The
moment-resisting frame structures, where cyclic parameter aS is the strain-hardening coefficient of
component deterioration can be omitted (Adam the global pushover curve without P-delta. The
and Jäger 2012b). In a first step, the vulnerability discussed backbone curves of the ESDOF system
of the structure to P-delta-induced collapse is are depicted in Fig. 21: The bilinear curve with
evaluated based on global first-mode pushover the largest strength represents the auxiliary back-
curve of the structure with applied gravity loads. bone curve, and the graph with the smallest
When this curve exhibits a negative post-yield strength is the backbone curve with P-delta. Sub-
stiffness, as shown in Fig. 15, the building may sequently the negative post-yield stiffness ratio
collapse under severe earthquake excitation in ya aS of the auxiliary ESDOF system is evalu-
the P-delta mode. In this situation the structure ated. Since the initial stiffness of the auxiliary
is transformed into an ESDOF system. backbone curve is larger than the original one,
Collapse capacity spectra (Adam and Jäger the period of the ESDOF system with assigned
2012a) are the further ingredient of this method. auxiliary backbone curve is (Adam et al. 2004)
However, application of collapse capacity spec- vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
tra is not straightforward for an MDOF structure, rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffirffiffiffiffiffiffiffiu N
1 aS xN y u t
X
because the backbone curve of the ESDOF sys- T a ¼ 2p f mi
tem is derived from the global first-mode push- 1 ye þ yi aS V y i¼1 i
over curves with and without considering P-delta. (9)
As discussed before, bilinear idealization of these
curves does not exhibit a uniform stability coef- In Eq. 9 xNy and Vy denote the roof displacement
ficient as in a real SDOF system, but an elastic ye and the base shear, respectively, at the onset of
and an inelastic stability coefficient yi that is yield of the global pushover curve without
always larger, in some cases even much larger P-delta; see Fig. 15.
than ye: yi > (>)ye. Assigning a specific auxiliary Now the appropriate collapse capacity spec-
backbone curve (Ibarra and Krawinkler 2005) to trum with respect to the underlying ground
the ESDOF system solves this problem because motion set, viscous damping z, the hysteretic
its rotation by the uniform auxiliary stability loop, and the negative post-yield stiffness ya
coefficient ya, aS is consulted to obtain the median (CCd), 16th
Seismic Collapse auxiliary backbone curve

Assessment, f*
Fig. 21 Backbone curves
*
with and without P-delta S ka
no P-delta effect
and auxiliary backbone *
fay
curve (Modified from *
Adam and Jäger 2012b) *
f0y S k0
*
i k0
ka* with P-delta effect
ka*
( i) k0 =
a *
S
k0* *
e k0
( S ) ka*
a
Dy D
(CCp16 p84
d ), and 84th percentile (CCd ) collapse and real moment-resisting frame structures
capacities of a SDOF system at the period Ta of showed that the collapse capacity can be esti-
the auxiliary ESDOF system. For instance, in mated quick, but yet accurate, without
Fig. 14 CCd is read at a period Ta = 3.2 s for conducting time-consuming dynamic analyses
ya aS = 0.20. These collapse capacity quanti- (Adam and Jäger 2012b). The collapse capacity
ties are transformed into the domain of the spectrum method is particularly useful in engi-
ESDOF system: neering practice because the structure can be
evaluated with respect to its seismic collapse
CCd capacity in the initial design process without a
CCESDOF ¼ ,
lMDOF detailed dynamic analysis.
!2 !
X
N X
N X
N
lMDOF ¼ fi mi = mi f2i mi
i¼1 i¼1 i¼1
Summary
(10)
This article summarizes procedures currently
lMDOF is the transformation coefficient that
used for assessing global collapse of structures
relates the actual structure with the corresponding
induced by strong-motion earthquakes. Through
SDOF system. For details it is referred to Fajfar
intensive research in the last two decades, sub-
(2002). CCESDOF is an estimate of the median
stantial progress was achieved in this field, and
collapse capacity of the MDOF system,
the outcomes of these studies are compiled in S
CCMDOF
CCESDOF . The same applies to the
numerous publications. Thus, in the present con-
percentiles: CCp16 p16
MDOF
CCESDOF and
p84 p84 tribution, only some of the issues involved in
CCMDOF
CCESDOF . Since the collapse capaci- seismic collapse assessment are elaborated in
ties follow generally a log-normal distribution, an detail. The selection of the presented material is
approximation of the collapse fragility function is naturally biased by research and experience of
derived from the authors.
Seismic collapse assessment is a branch of
lnN ðm, s2 Þ, m ¼ lnq
ðCC ESDOF Þ,
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi earthquake engineering that combines multidis-
s ¼ ln CCp16 p84
ESDOF CCESDOF ciplinary fields of research such as seismology,
(11) structural dynamics, materials science, applied
mathematics, and computational mechanics. It
Application of this method for assessing the includes adequate prediction of the seismic haz-
seismic collapse fragility of a series of generic ard, ground motion selection, identification of
possible modes of collapse, modeling of cyclic from equivalent SDOF systems. In: Proceedings of
component deterioration, appropriate consider- the 13th world conference on earthquake engineering
(13 WCEE), Vancouver, 1–6 Aug 2004, digital paper,
ation of hysteretic and viscous damping, quanti- 15 pp
fication of modeling and parameter uncertainties, ASCE/SEI 7-10 (2010) Minimum design loads for build-
and nonlinear dynamic analyses based on stable ings and other structures. American Society of Civil
numerical algorithms. Engineers, Reston
ASCE/SEI 41-13 (2014) Seismic evaluation and retrofit of
For accurate assessment of seismic structural existing buildings. American Society of Civil Engi-
collapse, however, there are still challenges that neers, Reston
need to be addressed. For example, understand- ATC-58 (2012) Seismic performance assessments of
ing of nonlinear cyclic component behavior in a buildings, 100% draft. Applied Technology Council,
Redwood City
condition of oblique bending requires experimen- ATC-72-1 (2010) Modeling and acceptance criteria for
tal tests and advances in constitutive modeling. seismic design and analysis of tall buildings. Applied
Another issue is the quantification of the contri- Technology Council, Redwood City
bution of nonstructural elements on the seismic ATC-78 (2013) Identification and mitigation of seismi-
cally hazardous older concrete buildings: interim
collapse capacity. On the other hand, from engi- methodology evaluation. Applied Technology Coun-
neering practice, there is a strong demand on cil, Redwood City
simplified assessment strategies that are simple Baker JW (2013) Trade-offs in ground motion selection
to apply and at the same time reasonably techniques for collapse assessment of structures. In:
Adam C, Heuer R, Lenhardt W, Schranz C (eds) Pro-
accurate. ceedings of the Vienna Congress on recent advances in
earthquake engineering and structural dynamics 2013
(VEESD 2013), Vienna, 28–30 Aug 2013, digital
paper, paper no. 123, 10 pp
Cross-References Baker JW, Cornell CA (2005) A vector-valued ground
motion intensity measure consisting of spectral accel-
▶ Analytic Fragility and Limit States [P(EDP| eration and epsilon. Earthq Eng Struct Dyn
34:1193–1217
IM)]: Nonlinear Dynamic Procedures Baker JW, Cornell CA (2006) Spectral shape, epsilon
▶ Analytic Fragility and Limit States [P(EDP| and record selection. Earthq Eng Struct Dyn
IM)]: Nonlinear Static Procedures 35:1077–1095
▶ Conditional Spectra Baradaran Shoraka MB (2013) Collapse assessment of
concrete buildings: an application to non-ductile
▶ Performance-Based Design Procedure for reinforced concrete moment frames. PhD thesis, The
Structures with Magneto-Rheological University of British Columbia, Vancouver
Dampers Baradaran Shoraka MB, Elwood KJ, Yang TY, Liel AB
▶ Seismic Risk Assessment, Cascading Effects (2014) Collapse assessment of non-ductile, retrofitted
and ductile reinforced concrete frames. In: Elwood KJ,
▶ Site Response for Seismic Hazard Assessment Kim InSung, Dragovich J (eds) Seismic assessment of
existing reinforced concrete buildings: new develop-
ments, ACI special publication ACI SP 297, SP297-8,
References 19 pp., American Concrete Institute, Farmington Hills,
MI
Adam C (2014) Structural dynamics and earthquake engi- Bernal D (1987) Amplification factors for inelastic
neering (in German). Internal report. University of dynamic P-D effects in earthquake analysis. Earthq
Innsbruck Eng Struct Dyn 15:635–651
Adam C, Jäger C (2012a) Seismic collapse capacity of Bernal D (1998) Instability of buildings during seismic
basic inelastic structures vulnerable to the P-delta response. Eng Struct 20:496–502
effect. Earthq Eng Struct Dyn 41:775–793 CEN (2005) Eurocode 8: design of structures for earth-
Adam C, Jäger C (2012b) Simplified collapse capacity quake resistance. Comité Européen de Normalisation
assessment of earthquake excited regular frame struc- (CEN), Brussels
tures vulnerable to P-delta. Eng Struct 44:159–173 Charney FA (2008) Unintended consequences of
Adam C, Ibarra LF, Krawinkler H (2004) Evaluation of modeling damping in structures. J Struct Eng
P-delta effects in non-deteriorating MDOF structures 134:581–592
Chopra AK, Goel RK (2002) A modal pushover analysis Ibarra LF, Krawinkler H (2005) Global collapse of frame
procedure for estimating seismic demands for build- structures under seismic excitations. Report No. PEER
ings. Earthq Eng Struct Dyn 31:561–582 2005/06, Pacific Earthquake Engineering Research
Clough RW, Johnston SB (1966) Effects of stiffness deg- Center, University of California, Berkeley
radation on earthquake ductility requirements. In: Pro- Ibarra LF, Krawinkler H (2011) Variance of collapse
ceedings of the 2nd Japan earthquake engineering capacity of SDOF systems under earthquake excita-
symposium, Tokyo, October 1966, pp 227–232 tions. Earthq Eng Struct Dyn 40:1299–1314
Deierlein GG, Reinhorn AM, Willford MR Ibarra LF, Medina RA, Krawinkler H (2005) Hysteretic
(2010) Nonlinear structural analysis for seismic models that incorporate strength and stiffness deterio-
design. A guide for practicing engineers. NEHRP Seis- ration. Earthq Eng Struct Dyn 34:1489–1511
mic Design technical brief no. 4, produced by the Jäger C, Adam C (2012) Seismic collapse assessment of
NEHRP Consultants Joint Venture, a partnership of non-deteriorating frames with irregular structural
the Applied Technology Council and the Consortium properties vulnerable to P-delta. In: Proceedings of
of Universities for Research in Earthquake Engineer- the 15th world conference on earthquake engineering
ing, for the National Institute of Standards and Tech- (15 WCEE), Lisbon, 24–28 Sept 2012, digital paper,
nology, Gaithersburg, NIST GCR 10-917-5 paper no. 0245, 10 pp
Esteva L, Ruiz S (1989) Seismic failure rates of multistory Jäger C, Adam C (2013) Influence of collapse definition
frames. J Struct Eng 115:268–284 and near-field effects on collapse capacity spectra.
Fajfar P (1999) Capacity spectrum method based on J Earthq Eng 17:859–878
inelastic demand spectra. Earthq Eng Struct Dyn Jalayer F, Cornell CA (2009) Alternative non-linear
28:979–993 demand estimation methods for probability-based
Fajfar P (2002) Structural analysis in earthquake seismic assessments. Earthq Eng Struct Dyn
engineering – a breakthrough of simplified non-linear 38:951–972
methods. In: Proceedings of the 12th European confer- Jennings PC, Husid R (1968) Collapse of yielding struc-
ence on earthquake engineering, London, 9-13 Sep- tures during earthquakes. J Eng Mech Div Proc ASCE
tember 2002, CD-ROM paper, Paper ref. 843. 94:1045–1065
Elsevier, p 20 Krawinkler H, Zareian F, Lignos DG, Ibarra LF
Fardis MN, Biskinis DE (2003) Deformation capacity of (2009) Prediction of collapse of structures under earth-
RC members, as controlled by flexure or shear. In: quake excitations. In: Papadrakakis M, Lagaros ND,
Toshimi K, Hitoshi S (eds) Proceedings of the interna- Fragiadakis M (eds) Proceedings of the 2nd interna-
tional symposium on performance-based engineering tional conference on computational methods in struc-
for earthquake resistant reinforced structures, a vol- tural dynamics and earthquake engineering
ume honoring Prof. Shunsuke Otani, Tokyo, 8-9 Sep- (COMPDYN 2009), Rhodes, 22–24 June 2009,
tember 2003, pp 511–530 CD-ROM paper, paper no. CD449
FEMA P-695 (2009) Quantification of building seismic Kunnath SK, Reinhorn AM, Park YJ (1990) Analytical
performance factors. Federal Emergency Management modeling of inelastic response of R/C structures.
Agency, Washington, DC J Struct Eng ASCE 116:996–1027
Gatto KS, Uang CM (2002) Effects of loading protocol Liel AB, Deierlein GG (2013) Cost-benefit evaluation of
and rate of loading on woodframe shearwall response. seismic mitigation alternatives for older reinforced con-
In: Seventh U.S. national conference on earthquake crete frame buildings. Earthq Spectra 29:1392–1411
engineering (7 NCEE), Boston, MA, 21-25 July 2002 Liel AB, Haselton CB, Deierlein GG, Baker JW
Gupta A, Krawinkler H (2000) Dynamic P-delta effects (2009) Incorporating modeling uncertainties in the
for flexible inelastic steel structures. J Struct Eng assessment of seismic collapse risk of buildings. Struct
126:145–154 Saf 31:197–211 S
Haselton CB, Liel AB, Deierlein GG (2009) Simulating Lignos DG, Krawinkler H (2012) Sidesway collapse of
structural collapse due to earthquakes: model idealiza- deteriorating structural systems under seismic excita-
tion, model calibration, and numerical simulation tions. Report no. TR 177, The John A. Blume Earth-
algorithms. In: Papadrakakis M, Lagaros ND, quake Engineering Research Center, Department of
Fragiadakis M (eds) Proceedings of the 2nd interna- Civil and Environmental Engineering, Stanford Uni-
tional conference on computational methods in struc- versity, Stanford
tural dynamics and earthquake engineering Lignos DG, Krawinkler H (2013) Development and utili-
(COMPDYN 2009), Rhodes, 22–24 June 2009, zation of structural component databases for
CD-ROM paper, paper no. CD497 performance-based earthquake engineering. J Struct
Haselton CB, Liel AB, Deierlein GG, Dean BS, Chou JH Eng ASCE 139:1382–1394
(2011) Seismic collapse safety of reinforced concrete MacRae GA (1994) P-D effects on single-degree-of-
buildings. I: assessment of ductile moment frames. freedom structures in earthquakes. Earthq Spectra
J Struct Eng 137:481–491 10:539–568
2752 Seismic Design of Dams
Medina RA, Krawinkler H (2003) Seismic demands for

nondeteriorating frame structures and their depen- Seismic Design of Dams
dence on ground motions. Report no. 144. The John
A. Blume Earthquake Engineering Research Center,
Department of Civil and Environmental Engineering, Guoxi Wu
Stanford University, Stanford Wutec Geotechnical International, New
Melchers RE (1999) Structural reliability analysis and Westminster, Metro Vancouver, BC, Canada
prediction, 2nd edn. Wiley, Chichester
Moehle J, Deierlein GG (2004) A framework methodol-
ogy for performance-based earthquake engineering.
In: Proceedings of the 13th world conference on earth-
quake engineering (13 WCEE), Vancouver, 1–6 Aug
Synonyms
2004, digital paper
NIST GCR 10-917-8 (2010) (ATC-76-1). Evaluation CSR; Design seismic parameters; Earthfill dams;
of the FEMA P-695 methodology for quantification Factor of safety; Granular and plastic soils; Limit
of building seismic performance factors.
National Institute of Standards and Technology,
equilibrium; Newmark sliding block; Nonlinear
Gaithersburg finite element analysis; Performance-based
Panagiotakos TB, Fardis MN (2001) Deformations of design; PWP model; Risk analysis; Soil liquefac-
reinforced concrete members at yielding and ultimate. tion; SPT; Time history
ACI Struct J 98:135–148
Park Y, Ang A (1985) Mechanistic seismic damage model
for reinforced concrete. J Struct Eng ASCE
111:722–739 Introduction
Rosenblueth E (1981) Point estimates for probability.
Appl Math Model 5:329–335
Shafei B, Zareian F, Lignos DG (2011) A simplified
Design of new dams or safety evaluation of
method for collapse capacity assessment of moment- existing dams for seismic loads is standard prac-
resisting frame and shear wall structural systems. Eng tice and routinely required. In a broad term, dams
Struct 33:1107–1116 can be classified into three types – concrete dam,
Shome N, Cornell CA (1999) Probabilistic seismic
demand analysis of nonlinear structures, reliability of
rock-fill dam, and earthfill dam – and seismic
marine structures program report no. RMS-35, Depart- design can be at a dam safety level and/or at a
ment of Civil and Environmental Engineering, serviceability level.
Stanford University This entry provides contents and discusses
Sivaselvan MV, Reinhorn AM (2000) Hysteretic models
for deteriorating inelastic structures. J Eng Mech
methods for design and analysis of earthfill
126:633–640 dams at a dam safety level where the ultimate
Song J, Pincheira J (2000) Spectral displacement demands limit state is applied for the highest level of
of stiffness and strength degrading systems. Earthq design earthquake loads. For dam safety, dam
Spectra 16:817–851
Ugurhan B, Baker JW, Deierlein GG (2014) Uncertainty
failure is the primary concern; for serviceability
estimation in seismic collapse assessment of modern consideration, the dam should remain functional
reinforced concrete moment frame buildings. In: Pro- and any damages should be easily repairable
ceedings of the tenth U.S. national conference on under this level of earthquake loads.
earthquake engineering (10 NCEE), Anchorage
Vamvatsikos D, Cornell CA (2002) Incremental dynamic
analysis. Earthq Eng Struct Dyn 31:491–514
Vamvatsikos D, Fragiadakis M (2010) Incremental Dam Performance in Past Earthquakes
dynamic analysis for estimating seismic performance
sensitivity and uncertainty. Earthq Eng Struct Dyn
39:141–163
Damages to dams under earthquakes can result
Villaverde R (2007) Methods to assess the seismic col- from ground shaking, soil liquefaction, ground
lapse capacity of building structures: state of the art. cracking, ground displacements (lateral spread-
J Struct Eng 133:57–66 ing and settlement), and in extreme cases surface
Zareian F, Krawinkler H, Ibarra L, Lignos D (2010) Basic
concepts and performance measures in prediction of
rupture along an earthquake-active fault. Exam-
collapse of buildings under earthquake ground ples of recent big and devastating earthquakes
motions. Struct Des Tall Spec Build 19:167–181 include the 2008 Sichuan earthquake (a crustal
Seismic Design of Dams 2753
earthquake with a magnitude of M8.0 occurred occurring. This is mainly because the societal risk
in China) and the 2010 Chile earthquake tolerance on a relative scale increases in order
(a subduction earthquake with M8.8 occurred from nuclear radiation leak to flood from a dam
off the coast of central Chile). breach, to collapse of a residential building, and to
Historical data show that dams around the loss of a temporary structure.
world have performed well and satisfactory and The first factor contributing to seismic risk,
that the probability of a dam failure under strong such as from an earthquake event with a PGA of
ground motion shaking is low. Nearly all well- 0.7 g, is the occurrence rate of such earthquake
built and well-compacted embankment dams can which is often measured by annual exceedance
withstand moderate earthquake shaking with probability (AEP). The other factor is the conse-
peak ground accelerations (PGA) greater than quence as a result of an earthquake, which is
0.2 g. Dams constructed of clay soils on clay, or ultimately measured in terms of loss of life and
on rock, or on overburden foundations resistant to sometimes economic loss. At a conceptual level,
liquefaction have withstood (with no apparent seismic risk can be expressed as the product of
damages) extremely strong shaking with PGAs seismic hazard probability and consequence, and
from 0.35 to 0.8 g. it represents the probabilistic expectation of the
Soil liquefaction, either in the dam fills or in the consequence.
foundation, is the most damaging factor that affects Ideally, seismic design of a dam should be
the performance of dams under earthquakes. Dams based on a risk analysis, including calculation of
built of sandy soils, especially hydraulic or semi- the actual probability of a dam failure and its
hydraulic fills, or built on foundations of loose (low consequence. The adequacy of the dam would
density) sandy soils are highly susceptible to earth- then be judged from the seismic risk (such as
quake damage or vulnerable to failures due to the annual probability of single or multiple fatalities)
potential for soil liquefaction. A famous case his- that is acceptable to the society for the loss of life
tory of such is the near failure of the Lower San involved in a dam failure.
Fernando Dam in California, USA, in a 1971 earth- The risk-based approach to dam safety evalu-
quake (Seed 1979). ation should balance the public risk and the lim-
ited societal resources available to manage the
particular risk. As shown in Fig. 1, it is consid-
ered generally acceptable that the maximum level
Design Seismic Loads Based on a Risk of societal risk to fatality is less than 103 per
Analysis annum for loss of one life and the risk is less than
105 per annum when more than 100 lives would
Public tolerance to seismic risk for the conse- be lost in the event of a dam failure. The principle
quence of dam failure ultimately determines the that the risks should be as low as reasonably S
adequacy of an existing dam or the criteria for the practicable (ALARP) is generally followed in
design of a new dam. For instance, at a specific practice, and it is thus reasonable to use an annual
location with the same tectonic setting or geolog- probability of 105 and 106 for 100 and more
ical condition and thus the same earthquake hazard fatalities.
condition, a nuclear reactor facility may be In mathematical terms, seismic risk to loss of
designed using a PGA of 1.0 g for a very low life (fatality) is calculated using the following
probability earthquake event, a hydroelectric dam two equations:
using a PGA of 0.7 g for a low probability earth-
quake event, a residential building using a PGA of
0.5 g for a median probability of happening, and a PFatality ¼ PFailure PFatality=Failure
temporary bridge for construction traffic using a and
PGA of 0.3 g for a relatively high probability of PFailure ¼ PEarthquake PFailure=Earthquake
Seismic Design of Dams,

Fig. 1 An example
relationship to manage
societal risk for dam safety
where: liquefaction capacity (Level 4), dam crest settle-

ment magnitude (Level 5), reservoir water level
PFatality = unconditional probability of fatality (Level 6), and dam damage level (Level 7). For
from earthquake, i.e., seismic risk. illustration purpose, at Level 1 for earthquake
PEarthquake = unconditional probability for an acceleration, it may have four scenarios with a
earthquake to occur, such as the one with a PGA of 1.0, 0.7, 0.5, and 0.3 g, respectively, and
PGA of 0.7 g. each PGA scenario has its probability of occur-
PFailure/Earthquake = conditional probability of a rence (AEP). At Level 3 assuming two models for
dam failure in event of the earthquake. liquefaction analysis, the weighting factors (e.g.,
PFatality/Failure = conditional probability of loss of 0.55 and 0.45) would be assigned to each model
life in the event of a dam failure. to make a total weighting of 1.
In many cases where a dam consequence class
The probabilities caused by all dam failure is available after completion of a life safety
initiating events (failure modes) need to be aggre- model and analysis, it can be conservatively
gated or compounded in order to obtain the total assumed that the conditional probability
probability to a dam failure. Failure modes (such (PFatality/Failure) equals to 1.
as soil liquefaction) for earthquake loading should Although in recent years the risk-based seis-
be identified in advance in order to perform a risk mic dam safety evaluation is increasingly used in
calculation. The risk calculation for a particular dam safety management, it is not widely used in
failure mode (PFailure) is carried out using an event engineering design or seismic safety evaluation
tree approach where significant events are due to the limited ability to perform such a com-
sequenced in levels. Construction of an event plex risk-based analysis. As described below, the
tree for a dam requires special knowledge in geo- standard-based or traditional approach is more
technical earthquake engineering and a compre- commonly used in seismic design.
hensive understanding of the dam.
Using “soil liquefaction” as an example fail- Standard-Based Method and Design
ure mode, an event tree for dam failure could Earthquake AEP
consist of earthquake acceleration (Level 1), con-
tributing earthquake magnitude (Level 2), lique- The standard-based method is a semi-
faction analysis method (Level 3), soil probabilistic method that defines the seismic
hazard using the probabilistic approach, but does Once a seismic hazard level is selected, con-
not explicitly calculate the probability of a dam ventional analyses such as limit equilibrium or
failure and its consequence. Without any quanti- finite element analyses are conducted using seis-
fiable risk calculation, this is an empirical or mic loads corresponding to the selected AEP. The
experience-based approach to risk evaluation results (such as stresses, ground displacements,
and management. The consequence, i.e., condi- or stability in a dam or its foundation) and con-
tional fatalities in the event of a dam failure, is sequences are evaluated deterministically using
indirectly evaluated by using a dam consequence standards, specifications, and design codes. The
class scheme (FEMA 1998; CDA 2007). In some potential for a dam failure is evaluated by com-
countries the highest consequence class is paring deterministically the resulting stresses and
“Extreme” when potential (expected) fatalities displacements with ultimate stability and
are 100 and more. established failure criteria. The evaluations of
With this approach, a seismic hazard level results are primarily done by empirical evidence,
(AEP) is selected according to the consequence past experience, and engineering judgment.
class of a dam so that the selected AEP conforms
to the societal acceptable risk level. For a dam
under an “Extreme” class, it would be reasonable Seismic Hazard Evaluations
to target a dam failure probability of 105 to 106
per annum, and an earthquake with an AEP of A deterministic seismic hazard analysis (DSHA)
105 or higher could be adequate, taking into is normally conducted for each known earth-
account the satisfactory post-earthquake perfor- quake source to determine site ground motion
mance of dams around the world, i.e., the low parameters such as PGA, response spectra at
conditional probability of a dam failure under 50th and 84th percentile, magnitude of earth-
strong shaking. For background information, the quake, and duration of strong shaking. This
nuclear industry uses the 84th percentile spectra would include assessment of potential seismic
at AEP of 104 or approximately median-mean hazards from earthquake activities along local
spectra at a probability of 105 per annum. With or regional known faults for the life of a dam.
less stringent safety criteria than the nuclear A probabilistic seismic hazard analysis
industry, dam safety design can adopt a design (PSHA) is usually conducted to determine design
earthquake with an AEP of 1/10,000 (i.e., 104) seismic parameters at various AEP levels, in
based on the mean spectra. The mean is the additional to DSHA where known active faults
expected value given the epistemic uncertainties, or subduction zones are identified. A PSHA nor-
and the mean hazard value typically such as in mally consists of identification of earthquake
Canada varies from 65th to 75th percentiles in the sources by developing a source model and the
hazard distribution. earthquake occurrence rate for each source zone S
The design earthquake AEP for an “Extreme” with the understanding of local geology and
class dam can vary from one region to another or regional past earthquake history, application of
from one country to another depending on the risk ground motion prediction equations (Earthquake
tolerance ability, and it is normally jointly Spectra 2008) that are appropriate to the region
selected by the dam owner and the government (e.g., hardness and shear wave velocity of the
regulatory agency. In some developing countries bedrock) and the types of seismic sources, and
with lower societal reliability than the developed determination of earthquake response spectra at
countries, an earthquake AEP of 1/1,000 is used various levels of probability by the integration of
for dam safety design. On the other hand, even in hazard contributions over all earthquake magni-
developed countries such as Canada, an earth- tudes and distances for all seismic source zones.
quake AEP of 1/2,475 is used for building safety The response spectra, either from DSHA or
design as a result to balance public risk and from PSHA, are normally defined for a series of
economic cost. discrete natural periods or frequencies (such as

Fig. 2 Example results of
seismic hazard analyses (a)
probabilistic method and
(b) deterministic approach
0.1, 0.2, 0.5, 1, 2, 5, 10, and 33 Hz; the frequency typical response spectra from DSHA and PSHA at
for PGAs can range from 33 to 100 Hz) that are the most commonly used damping ratio of 5 %.
used in the hazard calculation, and they are
always associated with an uncertainty level
(e.g., median, mean, or 84th percentile). Design Seismic Parameters
A seismic hazard analysis determines the poten-
tial intensity of seismic loads that could hit a Seismic parameters for design of dams consist
damsite; and it always proceeds to, but does not mainly of peak ground accelerations (PGAs),
always relate to, the design of a new dam or perfor- site-specific response spectra for horizontal and
mance evaluation of an old dam for earthquake vertical accelerations, magnitude and site-source
loading. However, seismic design of dams uses distance for the design earthquake, duration of
results of a seismic hazard analysis. Figure 2 shows strong shaking, ground motion time histories as
required for dynamic analyses, and fault dis- Input Ground Motion Time Histories
placements in rare situations when a dam is The current trend in seismic design of dams
(or to be) on an active fault. is to conduct linear or nonlinear time-history
analysis to obtain dynamic response of dam
Design Response Spectra and PGAs to earthquake loads. Time-history analysis
Response spectra and PGAs from DSHA can be of dam requires input ground motion time
used directly in seismic design, and they are histories (acceleration, velocity, and
conventionally computed at an 84th percentile displacement).
value in the hazard distribution, i.e., one standard For dynamic analysis of earthfill dams, multi-
deviation (using e = 1 in ground motion predic- ple ground motion records (five to eight com-
tion equations) above the predicted median value. monly adopted) from past earthquake are
In some cases for seismic faults with low rate of selected, and each record is then linearly scaled
activities, the 50th percentile values (the median) to fit the design response spectra (such as a
are used in design. UHRS) over the range of natural periods that
When seismic hazard is evaluated using a are appropriate for the dam. Using this method,
probabilistic approach, response spectra and an individual ground motion record is multiplied
PGAs from PSHA are used for seismic safety by a single scale factor to increase or decrease the
design of dams, using the results for the selected magnitude of the motion and its spectrum, with-
AEP (e.g., mean spectra with AEP of 1/10,000). out modifying the shape of its spectrum or its
In current practice, the vertical ground frequency contents.
motions do not seem to have much effect on the The selected ground motion records should be
performance of earthfill dams and thus are not consistent with the seismic parameters and rep-
normally included in design analysis. resentative to the design earthquake in terms of
magnitude, site-source distance, duration of
Design Earthquake Magnitude and Site- strong shaking, tectonic setting and source mech-
Source Distance anism, and consistency of site conditions
The PSHA results represent at a specific AEP between the recording station and the dam.
level a composite of hazard contributions from In some cases when uniform scaling of
earthquakes of all magnitudes and distances. The recorded ground motions is unable to meet the
response spectra from a PSHA are also called the requirements, the design earthquake ground
uniform (or equal) hazard response spectra motions would be obtained by modifying a
(UHRS). recorded ground motion in the time or frequency
Deaggregation of the composite seismic haz- domain.
ard is performed to identify relative contributions
of individual earthquakes or scenario earthquakes S
with various magnitudes and distances. For seis- General Approach to Seismic Design and
mic design of a dam, representative earthquake Analysis
magnitudes (M) and site-source distances (D) are
obtained by deaggregation of the uniform hazard For dams under static conditions, limit equilib-
at natural periods that are significant and critical rium analyses for slope stability are normally
to the dam seismic response. conducted for long-term operating reservoir con-
The representative M and D for the design ditions (steady-state seepage) and for short-term
earthquake are then used in seismic analysis reservoir drawdown conditions. In general, for
where magnitude and duration of the earthquake satisfactory stability factor of safety (FS) would
are needed such as in the soil liquefaction analy- be minimum 1.5 for long-term conditions
sis, or for selection of ground motion records (FS > 1.5) and 1.3 for short-term conditions
needed in a time-history analysis. (FS > 1.3).
Seismic Design of Dams, Fig. 3 Types of seismic analyses based on liquefaction susceptibility (PWP = excess pore
water pressures from earthquake loads)
For dams under seismic loading, two primary analysis, a post-earthquake static stability analy-
modes of dam failure must be addressed, the sis, and a finite element dynamic analysis for
overtopping failure caused by excessive settle- computing permanent ground deformations.
ment of dam crest and the internal erosion and
piping failure caused by cracks in dams, damage
of filter layer upstream, or drains downstream of Evaluation of Soil Liquefaction
the core zones.
Some of the common design measures to mit- Liquefaction of Sands Under Cyclic Loads
igate these failures include the following: remove Sand liquefaction is a fluidization process of sat-
or improve problematic foundation soils by urated sand mass subject to cycles of shear stress.
ground treatment and adequately compact the Under shaking, it can be easily observed that
dam fills, use wide core zones of plastic soils loose sands in a dry container will experience
that are resistant to erosion, use well-graded volume contraction and settle to a more compact
wide filter zones upstream of the core, construct state. In a saturated condition, immediate volume
chimney drains downstream of the core to lessen change of the sand would not occur because water
soil saturation and reduce downstream seepage, in the pore does not drain quickly enough under
and use dam crest and downstream slope details the rapid earthquake loading. Instead, the poten-
to provide protection of the dam in the event of an tial for volume change translates into a quick
overtopping. increase in excess pore water pressure (PWP) in
Seismic analyses appropriate to a dam and its the sand mass. Liquefaction occurs when the
foundation are conducted using design seismic PWP exceeds a threshold value that the pore
parameters to provide adequate information on water effectively suspends sand particles. Sand
expected stresses and ground displacements for boiling to ground surface is a surficial expression
evaluation of expected performance of the dam. of liquefaction of sands in the ground.
Seismic analyses (see Fig. 3) can include a soil Liquefaction strength, or capacity, or resis-
liquefaction evaluation, a pseudo-static stability tance (more commonly used) of sands can be
analysis, a Newmark sliding block deformation measured by laboratory testing using

Fig. 4 Liquefaction
resistance trend curves for
clean sands (<5 % fines)
reconstituted samples prepared to a target relative sampling. A key result of SPT is SPT N value
density. It is found (see Fig. 4) that liquefaction that is the number of blow counts for a penetra-
resistance generally increases with relative den- tion depth of 305 mm (1 ft.), and (N1)60 is defined
sity as expected, and under the same relative as the N value corrected for the soil under an in
density, it decreases with more cycles of shear situ effective vertical normal stress of about
stress used in the testing. 100 kPa (1 atmospheric pressure) and to the stan-
Because liquefaction is triggered by multiple dard SPT hammer energy level (i.e., at 60 % of
cycles of shear stresses, liquefaction resistance of the theoretic maximum hammer energy). The
soils is represented using both the cyclic shear general understanding is that the measured SPT
stress level (tcyc) and the number of cycles for N value is corrected downward (i.e., reduced)
such uniform (constant amplitude) stresses to when the test is performed for soil at a depth
trigger a liquefaction failure. Cyclic stress ratio with a confining stress higher than 100 kPa
(CSR) refers to the ratio of the cyclic shear stress (e.g., at 300 kPa).
to the initial vertical normal stress (s0 v0) at which Current technology or methodology for
the sample is under cyclic shearing, i.e., CSR = evaluating liquefaction resistance of sandy
tcyc/s0 v0. For earthquake loading, the number of soils was developed empirically by indirect stud-
cycles is related to earthquake magnitude and ies of performance case histories of dams in his-
duration. In engineering analysis, nonuniform torical earthquakes, where in some cases soil
cycles of shear stress from a M7.5 earthquake liquefaction was assessed to have occurred and
are artificially set equal to, in terms of net effect in other cases liquefaction was believed not S
on soil, a shear stress level with 15 uniform triggered.
cycles. Thus, liquefaction resistance of soil is Field soil liquefaction resistance of sands
typically characterized by CSR at 15 cycles or (particle size from 0.075 to 4.75 mm) has been
CSR15 as shown in Fig. 5. derived from field performance of soils under
various levels of historical earthquakes, and
Liquefaction Resistance of Sandy Soils CSR15 from case history studies is plotted with
Except with ground-frozen sampling technique in situ measured (N1)60 (Seed and Harder 1990)
that is practically not applicable, liquefaction as an equivalence to the relative density of sand.
resistance of in situ sands is considered not mea- Similar to the curve shown in Fig. 5, sands with
surable from laboratory testing. (N1)60 > 30 are considered to be dense and gen-
The most widely adopted test for field evalu- erally not liquefiable; sands with (N1)60 < 10 are
ation of relative density for sands is the standard generally loose and highly susceptible to
penetration test (SPT) with split-barrel soil liquefaction.

Fig. 5 An example
liquefaction resistance
curve for clean sands
Unless completely confined with no drainage 1980s and 1990s of the last century, soil index
path, gravels (particle size from 4.75 to 75 mm) parameters (water content, liquid limit, PI, and
are not so much vulnerable to soil liquefaction fines content) were used as basis for liquefaction
due to its high permeability that allows fast dis- assessment, but the current trend method is to use
sipation of excess pore water pressures. How- results of laboratory cyclic tests (Finn and Wu
ever, sandy soils containing some amount of 2013), especially for evaluation of more critical
gravels are liquefiable, and their liquefaction structures such as for dam safety.
resistance can be tested in the field using a large Liquefaction resistance of in situ plastic soils
penetration hammer such as the Becker Hammer can be measured directly by laboratory testing on
Penetration Test (Youd et al. 2001). in situ Shelby samples. Soil samples in a cyclic
At a given relative density or (N1)60, sandy direct simple shear test are normally first consol-
soils containing significant amount of fines idated to its pre-consolidation pressure (the
(particle size <0.075 mm) are known to have highest historic pressure experienced by the soil
higher liquefaction resistance than clean sands or s0 p) and then sheared by cycles of tcyc under
with no fines (Youd et al. 2001). Nonplastic s0 v0 and initial shear stress (tst) similar to the in
silts, containing 100 % fines but with very low situ stresses of the sample.
plasticity (such as plasticity index PI < 5 %), are As compared to sands where the relative den-
considered to behave under cyclic loading in a sity or (N1)60 is the single parameter, there are
similar manner as sands. The liquefaction resis- more parameters and factors affecting and con-
tance of nonplastic silts is evaluated as sands with tributing to the liquefaction resistance for plastic
upward correction on fines content. soils (see Fig. 6). The curves for plastic soils are
generally flatter than curves for sands; that is, the
Liquefaction Resistance of Plastic Soils resistance decreases less with increasing number
A more descriptive terminology for liquefaction of cycles. The two key factors are the over-
failure of plastic soils (silts or clayey silts), consolidation ratio (OCR = s0 v0/s0 p) and
containing 100 % fines and with relatively high PI. Liquefaction resistance of plastic soils gener-
PI (such as PI > 10 %), is “cyclic strain soften- ally increases with increasing OCR and PI but
ing” which emphasizes more on the structural decreases with increasing tst.
breakdown of the material under cyclic loading As a general observation, plastic soils having
than on the buildup of excess pore water pressure. CSR15 > 0.4 (for 15 cycles) are considered to be
Evaluation of liquefaction resistance for plastic hard, insensitive to earthquake loading, and
soils is an area that requires more research. In the unlikely to experience much strength loss. Plastic

Fig. 6 Liquefaction
resistance trend curves for
plastic soils, generally
flatter than sands
soils having CSR15 near or less than 0.15 may be analysis for a low dam or using a 2D finite ele-
sensitive to shaking and can experience signifi- ment dynamic analysis to model the geometric
cant loss of strengths in a strong earthquake. effect for a high dam. A site response analysis,
either 1D or 2D model, can be conducted using
Earthquake-Induced CSR the equivalent linear method (Idriss et al. 1974) or
For liquefaction analysis, seismic load in a soil a true nonlinear method (Finn et al. 1986) for the
element is represented by time history of shear simulation of shear modulus degradation
stresses (tdyn) imposed by earthquake shaking, (decrease) and damping ratio increase with the
and they can be calculated from a site response increase of shear strain. A true nonlinear
analysis using the input ground motions at the approach is considered more appropriate when
damsite and dynamic properties of dam fills and the ground shaking level is high, such as with a
foundation soils. PGA of 0.4 g or higher.
Soil dynamic properties for each of the soil
zones in the dam and its foundation primarily Factor of Safety Against Liquefaction
consist of low-strain shear modulus (Gmax) and Upon completion of a site response analysis,
the damping characteristics (Seed et al. 1986). cyclic stress ratio (CSR = tdyn/s0 v0) in each soil
The low-strain shear modulus (Gmax) is com- zone is calculated by converting the nonuniform
monly computed from shear wave velocity (Vs) cycles of shear stresses from the earthquake to
that can be measured by seismic downhole or equivalent cycles of uniform shear stresses
crosshole surveys. Shear modulus would reduce (Wu 2001) and then normalized to the in situ S
from Gmax as shear strain increases; reduction of vertical normal stress (s0 v0).
soil shear modulus (stiffness) with increasing The liquefaction resistance (CSR15) of soils is
strains is more significant for gravels than for determined, for sands using (N1)60 from SPT and
sands and more for sands than for clays. Soil for plastic soils using liquefaction resistance
damping ratio would be less than 5 % at low curves from laboratory cyclic shear tests.
strain, but it increases with increasing strain. A factor of safety against soil liquefaction for
The maximum damping ratio is about 25 % for each soil zone is calculated to be the ratio of
gravels and sands and in the order of 20 % for liquefaction resistance of soil and seismic shear
clays and silts. stress from earthquake, i.e., FSLIQUEFACTION =
A seismic site response analysis (▶ Site CSR15/CSR. A computed FSLIQUEFACTION near
Response: 1-D Time Domain Analyses) is com- or less than 1.0 indicates triggering of soil lique-
monly performed in a total stress analysis, with- faction, and FSLIQUEFACTION > 1.5 means not
out including PWP effect, using a 1D soil column liquefiable.
Pseudo-Static Limit Equilibrium safety, should be considered to be tolerable or

Stability Analysis not for dam performance assessment. The
Newmark deformation method assumes that
A pseudo-static analysis is a limit equilibrium deformation of a dam is modeled by a rigid
method which includes additional seismic inertia block sliding on an assumed failure surface
forces in a conventional static slope stability under the design ground motions at the damsite.
analysis. Seismic coefficients (kh for horizontal Sliding deformation is assumed to occur
and kv for vertical) are often used in such analy- whenever the acceleration of a sliding block
sis, and they are normally taken as the peak exceeds the yield acceleration, which is the hor-
ground accelerations (PGAs) as a fraction of the izontal acceleration that results in a factor of
gravity acceleration. The seismic inertia forces safety of 1 in a pseudo-static slope stability anal-
are represented by khW and kvW in horizontal ysis, and the sliding stops when the acceleration
and vertical direction, respectively, where W is falls below the yield acceleration. Mathemati-
the weight of a sliding block. Due to the nature of cally, the Newmark sliding block displacements
seismic shaking, shear strengths of soils for rapid are computed by double integration of accelera-
loading conditions are conventionally used for tion time history on the sliding block for the
characterization of soil resistance to earthquake portion of accelerations that exceeds the yield
loadings. acceleration (Fig. 7).
In a scenario that soils in a dam and its foun- Although it has been widely used at a time
dation would not develop significant PWP from when there is lack of direct methods for comput-
shaking, a pseudo-static factor of safety greater ing seismic ground deformations, the Newmark
than 1 (FS-pseudo >1) is a very strong indication sliding block model is generally recognized to be
that there would be little or no damage to the dam not a very good representation of how embank-
from an earthquake. ment dams deform under earthquake, especially
However, a pseudo-static factor of safety less for dams with low factors of safety. Due to its
than 1 (FS-pseudo <1) does not necessarily rep- simplicity, the method may provide a range of
resent dam instability or unsatisfactory perfor- expected ground deformations, but it would nei-
mance; instead due to the transient nature of ther give distribution of ground displacements in
earthquake motions, the dam would undergo the dam nor provide direct calculation of dam
some deformations for the short time interval crest settlement.
when the ground acceleration exceeds the yield
acceleration. In such case seismic ground defor-
mations are computed and used for performance Post-Earthquake Static Stability
assessment. Analysis
It is not possible to predict failure of a dam or
estimate seismic ground deformations by a A direct consequence of soil liquefaction is a
pseudo-static analysis, and therefore other types great reduction of shear strength, in particular
of more comprehensive analysis are required to for loose and contractive sands (Fig. 8), for
provide a more reliable basis for seismic design which the shear strength may become a small
of dams. fraction of its static and drained value due to the
buildup of excess pore water pressures and fluid-
ization. In design practice, shear strength of liq-
Newmark Sliding Block Seismic uefied soil is also described as residual strength
Deformation Analysis and sometime as post-liquefaction strength.
As a frictional material and under static load-
Newmark (1965) pointed out that for seismic ing, loose sand can have a friction angle possibly
loading permanent ground deformations of a ranging from 30 to 35 ; however, the post-
dam, in addition to the pseudo-static factor of liquefaction strength of the same loose sand

Fig. 7 Newmark sliding
block deformation
calculation diagram
shaking disturbance, it can eventually decrease to

the remolded strength value. Depending on sen-
sitivity, remolded strength of silts and clayey silts
may range from about 25 % of its peak static
strength for low-sensitivity soils to only 10 %
for high-sensitivity soils.
A static limit equilibrium analysis, without
seismic loads but using the post-earthquake soil
strength, is conducted in order to assess the post-
earthquake static stability of the dam. The analy- S
sis would use residual strength for liquefied soils
and reduced strength in soils that do not liquefy
but with the buildup of excess pore water pres-
Seismic Design of Dams, Fig. 8 Shear stress – strain
response indicating great strength loss after liquefaction sures from the design earthquake shaking.
and some strength loss with cyclic mobility A factor of safety less than 1 from a post-
earthquake static stability analysis (FS-post <1)
may decrease to an equivalent friction angle of 5 is a strong indication of dam instability and a
or perhaps less (Seed and Harder 1990), indicat- potential failure during or immediately after the
ing the degree of shear strength reduction due to design earthquake. As such, the design for a new
liquefaction. dam would be improved and ground strengthen-
Residual strength of plastic soils after initial ing for an existing dam would be required, in
strain softening could decrease to 50–80 % of its order to increase the post-earthquake dam stabil-
pre-earthquake static value, and after sufficient ity and meet the dam safety requirements.
If FS-post >1 is calculated, ground deforma- loads. Excess pore water pressures (PWP) caused
tions from the design earthquake are often com- by earthquake loads can also be computed using
puted for dam performance evaluation, effective stress models with calculation of factor
preferably using a nonlinear finite element of safety against liquefaction for each soil ele-
dynamic analysis. The Newmark sliding block ment. In addition, the procedure provides two
deformation analysis should be avoided for com- options for applying input ground motions:
puting deformations of dams involving soil (1) ground accelerations applied at the model’s
liquefaction. rigid base and (2) outcropping ground velocities
applied at the model’s viscous and elastic base.
The elastic base option is often used when the
Seismic Deformations by Finite Element model base stiffness is not rigid relatively to the
Dynamic Analysis model body that prevents incident seismic waves
from effectively reflecting back to the
A more rigorous and direct method for computing model body.
earthquake-induced ground deformations, with As an example of Finn-Wu finite element pro-
and without soil liquefaction, is a nonlinear finite cedure, factors of safety against liquefaction
element dynamic analysis (Finn et al. 1994) that computed from a dynamic analysis using a sub-
overcomes the uncertainties associated with the duction earthquake input motion are shown in
approximation embedded in the pseudo-static Fig. 9 (Finn and Wu 2013). The results showed
stability calculation and in the Newmark that almost the entire saturated zone (below the
deformation analysis. With PWP calculations water table shown in blue) under the downstream
built into effective stress soil models, the slope of the dam would undergo large cyclic
finite element method has the capability to simu- strains with FSLIQUEFACTION < 1. A computed
late the coupled effects from dynamic site deformed mesh of the dam, with soil material
response, generation of excess pore water pres- zones shown in colors, immediately after the
sures, soil liquefaction, and post-liquefaction earthquake using the same subduction ground
behavior, and this is a trend method for quantita- motion is shown in Fig. 10, indicating a deep-
tive evaluation of earthquake-induced ground seated ground deformation pattern and significant
deformations. lateral movement and settlement of the down-
A well-known analysis procedure for the finite stream dam crest. The original dam surface is
element approach is the one developed by Finn outlined in blue in the figure.
et al. (1986) using soil data from centrifuge tests
and then verified using soil data and field mea-
sured deformations of the Upper San Fernando Seismic Performance Evaluation
Dam in California, USA, in a 1971 earthquake
(Wu 2001), and recently extended to include a In some cases, expected performance of dam can
dilative silt model (Finn and Wu 2013). This be assessed to either clearly safe or clearly
analysis procedure is highly regarded by practic- unsafe, based on results of simple limit equilib-
ing engineers and referenced for seismic design rium stability analyses. In majority cases, how-
of dams by the US Federal Emergency Manage- ever, dam performance is assessed from
ment Agency (FEMA 2005). estimated ground deformations and strains. Ulti-
The Finn-Wu finite element procedure is sim- mately, the expected ground deformations and
ple and well suitable for practical engineering the associated ground strains, either from a
analysis. This analysis method adopts Mohr- Newmark analysis or a nonlinear finite element
Coulomb criteria for modeling shear strength of analysis, should not exceed what the dam can
soils and uses a nonlinear hyperbolic model to safely withstand without catastrophic release of
simulate hysteresis response of soils under cyclic the reservoir water due to overtopping or internal
> 1.5 1.0 to 1.1

150.0
1.1 to 1.5 not used
140.0 0.8 to 1.0
130.0 0.5 to 0.8

< 0.5
120.0
110.0
100.0
90.0
–40.0 –20.0 0.0 20.0 40.0 60.0 80.0 100.0 120.0 140.0 160.0 180.0
Seismic Design of Dams, Fig. 9 Computed FSLIQUEFACTION shown in colors for an example finite element model of
an earthfill dam (undeformed, partial model, dimension in m)
150.0
140.0
130.0
120.0
110.0
100.0
90.0
–80.0 –60.0 –40.0 –20.0 0.0 20.0 40.0 60.0 80.0 100.0 120.0 140.0 160.0 180.0
Horizontal Distance (m)
Seismic Design of Dams, Fig. 10 Deformed dam cross section after a subduction earthquake (displacement scale
factor 1.0, elevation for vertical axis in m)
erosion. In the process, uncertainties associated Overtopping of a dam in earthquake can be

with the calculated factor of safety or with the caused either by slope instability (failure) or by
computed ground deformation need to be under- excess ground deformations (settlement) at dam
stood to achieve the confidence on the expected crest. If post-earthquake stability analyses indi-
performance. In some cases, the consequences of cate a factor of safety well above 1 such as
misjudging uncertainty levels would also be FS-post > 1.3, historical dam performance expe- S
taken into considerations. rience in earthquakes would indicate that the dam
Two key factors contributing to uncertainties will have limited or small deformations and will
in predicting the expected performance of dam perform satisfactorily. When liquefaction is not
under earthquake are related to soil liquefaction. involved, factor of safety greater than 1 with seis-
The first is “Will liquefaction be triggered?”, and mic force in a pseudo-static analysis
the second is “What is the residual strength after (FS-pseudo > 1) would also indicate that the
soil liquefaction?” Ideally, a risk analysis would dam will perform well in a design earthquake.
be beneficial to quantify the uncertainties. How- On the other hand, confidence in dam safety
ever, state of practice is mostly based on a deter- decreases and probability of a dam failure due to
ministic approach for dam performance overtopping or internal erosion increases when a
assessment, such as using analysis steps outlined post-earthquake factor of safety (FS-post) near or
in Fig. 11. less than 1.0 is calculated using residual strengths
Seismic Design of Dams, Fig. 11 Evaluation of dam performance for critical failure modes under earthquake
for liquefied soils. A FS-post less than 1.0 would Summary

indicate slope instability under post-earthquake
static conditions, and a dam slope failure can This entry describes basic principles and meth-
occur immediately after the earthquake. odology used in seismic analysis and design for
In general, when a wedge or circular sliding earthfill dams.
surface has a low post-earthquake factor of The risk-based approach for dam safety eval-
safety, the deformations along the slip surface uation is introduced to indicate that seismic
would be large or sometime excessive. In these design criteria are governed by societal tolerance
cases, ground deformations calculated preferably of seismic risk which consists of both seismic
from a finite element dynamic analysis are used hazard and its consequence. In engineering prac-
to assess the potential of a dam failure by tice, a full risk analysis is not commonly
overtopping, or internal erosion, or loss of soils performed due to its complexity; instead, conse-
in piping. quence of a dam failure in an earthquake is often
Engineering judgment is carefully applied in represented using a dam consequence class
assessing the level of uncertainties in design scheme. A design level of seismic hazards is
parameters and analytical methodology and thus then selected from the consequence class for a
the confidence level in the use of calculated dam. Seismic hazard of a damsite can be evalu-
deformations. When soil liquefaction is not an ated on a probabilistic approach. Once design
issue, deformations estimated from a Newmark seismic parameters are determined using the
sliding block analysis would prove to be adequate selected seismic hazard level, seismic analysis
for many cases. For dams with liquefaction and design are conducted using a deterministic
issues, seismic design or evaluation would use approach where seismic demands are compared
deformations computed from a finite element with the ultimate capacity of soils.
dynamic analysis, which has adequate soil State-of-practice methodology for evaluation
models for simulation of soil liquefaction and of soil liquefaction is described, which include
post-liquefaction behavior and also can compute measurement of liquefaction resistance of soils
large-strain ground deformations. Even using a by laboratory tests, assessment of liquefaction
finite element approach, computed deformations resistance from in situ tests in the field, soil
for dams without involving soil liquefaction parameters that impact the liquefaction resis-
would be more reliable (with less uncertainties) tance, and typical values of liquefaction resis-
than those involving liquefaction. tance. Liquefaction potential of a soil element is
Seismic Design of Earth-Retaining Structures 2767
evaluated by comparing the seismic shear stress Finn WDL, Ledbetter RH, Marcuson WF (1994a) Seismic
from earthquake with the liquefaction resistance deformations in embankments and slopes. In:
Balasubramaniam AS et al (eds) Proceedings sympo-
of the soil. sium on developments in geotechnical engineering
Seismic analyses discussed in this entry from Harvard to New Delhi 1936–1994, Bangkok.
include pseudo-static stability analysis, Balkema Publication. ISBN 90 5410
Newmark sliding block deformation analysis, 522 4, pp 233–264
Finn WDL, Ledbetter RH, Wu G (1994b) Liquefaction in
post-earthquake static stability analysis, and silty soils: design and analysis. Ground failures under
finite element dynamic analysis for computing seismic conditions. ASCE Geotech Spec Publ
permanent ground deformations. 44:51–76
It is pointed out that soil liquefaction is the key Idriss IM, Seed HB, Serff N (1974) Seismic response by
variable damping finite elements. J Geotech Eng
factor contributing to uncertainties in predicting ASCE 100(1):1–13
the expected performance of dam under earth- Newmark NM (1965) Effects of earthquakes on dams and
quake, which are based on results of limit equi- embankments. Geotechnique 15(2):139–160
librium stability analyses and in majority cases Seed HB (1979) 19th Rankine lecture: considerations in
the earthquake resistant design of earth and rockfill
from estimated ground deformations. Engineer- dams. Geotechnique 29(3):215–263
ing judgment should be applied in assessing the Seed RB, Harder LF (1990) SPT-based analysis of cyclic
level of uncertainties in design parameters and pore pressure generation and undrained residual
analytical methodology and thus the confidence strength. In: Proceedings of H. Bolton seed memorial
symposium, vol 2. BiTech Publishers, Vancouver,
level in the use of calculated factors of safety or pp 351–376
ground deformations. Seed HB, Wong RT, Idriss IM, Tokimatsu K (1986)
Moduli and damping factors for dynamic analyses of
cohesionless soils. J Geotech Eng ASCE
112(11):1016–1032
Cross-References Wu G (2001) Earthquake induced deformation
analyses of the Upper San Fernando dam under the
▶ Probabilistic Seismic Hazard Models 1971 San Fernando earthquake. Can Geotech
J 38:1–15
▶ Site Response: 1-D Time Domain Analyses
Youd TL et al (2001) Liquefaction resistance of soils:
summary report from the 1996 NCEER and 1998
NCEER/NSF workshops on evaluation of liquefaction
References resistance of soils. J Geotech Geo-Environ Eng ASCE
127(10):817–833 (21 authors)
Canadian Dam Association (CDA) (2007) Dam safety
guidelines. CDA
Earthquake Spectra (2008) Special issue on the next gen-
eration attenuation project. Earthq Spectra Spec Issue
24(1):639–644
Federal Emergency Management Agency (FEMA)
Seismic Design of Earth-Retaining
(1998) Federal guidelines for dam safety: hazard Structures S
potential classification systems for dams, FEMA
333, Oct 1998. Reprinted April 2004 Luigi Callisto
Federal Emergency Management Agency (FEMA)
Department of Structural and Geotechnical
(2005) Federal guidelines for dam safety: earthquake
analyses and design of dams, FEMA 65, May 2005 Engineering, Sapienza Università di Roma,
Finn WDL, Wu G (2013) Dynamic analyses of an earthfill Rome, Italy
dam on over-consolidated silt with cyclic strain soft-
ening, Keynote lecture. In: 7th international confer-
ence on case histories in geotechnical engineering,
Chicago, Apr 2013 Introduction
Finn WDL, Yogendrakumar M, Yoshida N, Yoshida
H (1986) TARA-3: a computer program to compute Earth-retaining structures provide support to
the response of 2-D embankments and soil structure
excavations or earth fills: the earth pressure is
interaction systems to seismic loadings, Department of
Civil Engineering, University of British Columbia, either transmitted to the soil located below the
Vancouver foundations of the structure, as in gravity
2768 Seismic Design of Earth-Retaining Structures
retaining walls, or balanced by a combination of Conceptual Model for the Seismic

horizontal passive forces in the ground and the Behavior of Retaining Structures
reaction forces of additional structural elements,
as in the case of embedded retaining structures. During an earthquake, the structure and the
Seismic forces have a detrimental effect on these retained soil are accelerated and are therefore
structures, producing an increase in the earth subjected to the corresponding inertial forces
pressure and a decrease in the available soil resis- that, combining with the preexisting gravitational
tance. Commonly, the seismic design of an earth- forces, produce additional loading on the
retaining structure is carried out using a pseudo- retaining structure. It is important to note that
static approach, in which the seismic forces are the seismic inertial forces are different from the
derived from a uniform acceleration field and are gravitational forces, as they vary cyclically in
applied statically to the soil-structure system. amplitude and direction, and have a transient
This approach has the advantage of being rela- nature. This has important consequences on the
tively simple to implement, but the rationale for seismic design of a retaining structure.
the choice of the constant acceleration values
needs careful consideration. Therefore, the fol- Displacing Retaining Structures
lowing section provides a simple conceptual An effective illustration of the seismic behavior
model explaining how the pseudo-static acceler- of a retaining structure is provided by the simple
ation values can be derived from the estimated scheme of a rigid block with a mass m resting on a
maximum acceleration and from the acceptable horizontal base (Fig. 1a). The base is subjected to
seismic performance of the structure. The a simple dynamic excitation, consisting of a hor-
remaining sections provide the essential formulas izontal acceleration a0 that is kept constant over a
for the evaluation of the seismic earth pressure time interval t0, and is then removed (Newmark
and deal with specific issues relative to the seis- 1965). If ab is the block’s acceleration, then the
mic design of gravity and embedded retaining corresponding inertial force is mab that must be
structures. balanced by the resultant force T applied by the
Seismic Design of Earth- a c a

Retaining Structures,
ab a0 base
Fig. 1 Schematic
representation of a rigid
block subjected to an a0 ac block
acceleration pulse at ab
the base b
mac t
d v t0 tm
Tlim v0
a0
f
ab ab ur
ac
a0 t
e ur
g
mac Sa(ac)
Tlim t
base to the bottom surface of the block (Fig. 1b). from the dissipation of energy taking place during
If the capacity at the block-base contact is Tlim, the activation of the plastic mechanism.
then the acceleration of the block cannot be larger In addition to the inertial forces related to its
that its critical value ac ¼ T lim =m: own mass, a retaining structure is loaded by the
Now suppose that a0 is larger than ac (Fig. 1c): earth thrust that in turn increases because the
since the acceleration of the block cannot go masses of the retained soil are accelerated. There-
beyond ac, a relative acceleration occurs between fore, a scheme more appropriate to a retaining
base and block and the corresponding velocity structure is that of Fig. 1f, where the block is
plots tend to diverge, as shown in Fig. 1d. The loaded additionally by Sa(ab), which is the resul-
shaded area between the two velocity plots rep- tant force of the total earth pressures
resents the relative displacement ur between base corresponding to the acceleration ab.
and block; if the horizontal acceleration at the (Expressions for Sa(ab) are given in the next sec-
base of the block were kept constant at a0 for a tion.) Hence in this case a given acceleration
long period of time, the system would be loaded produces larger inertial forces. The critical accel-
permanently by the inertial forces, and the rela- eration ac is derived from the equilibrium equa-
tive displacement would increase rapidly with an tion (Fig. 1g):
upward concavity. This would be by definition a
collapse mechanism: the system capacity is m ac þ Sa ðac Þ ¼ T lim (1)
attained under constant forces, and the displace-
ments accelerate. and is smaller than that of the previous scheme of
In the present case, since a0 is maintained only Fig. 1a: the conceptual model is essentially the
for the time interval t0, it can be seen that as t gets same, but from Fig. 1d and e, it is evident that a
larger than t0, the sign of the relative acceleration smaller critical acceleration corresponds to larger
changes, and the relative velocity decreases. relative velocities and hence to larger
Therefore, the relative displacements increase displacements.
with a downward concavity and eventually stop The schematic model of Fig. 1 evidences an
when the relative velocity equals zero (Fig. 1e): at important principle: since the block’s accelera-
the end of this specific event, since the system’s tion cannot exceed ac, the internal forces acting
capacity has been attained only transiently, the within the block cannot go beyond those
system has not collapsed, but rather suffered a evaluated for a = ac. Since ac is related to the
damage in the form of a permanent relative dis- seismic resistance of the system, it follows that
placement. Whether this damage is acceptable or stronger systems are called to resist larger inter-
not depends on the specific function of the nal forces.
system. With reference to the seismic design of a
This schematic illustration shows that because retaining structure, one can depict two different S
of the transient nature of the seismic forces, the cases:
attainment of the full strength of the system dur-
ing an earthquake does not imply a collapse, but 1. The critical acceleration of the system ac is
rather a damage deriving from the temporary larger than the maximum acceleration at the
activation of a plastic mechanism. In principle, base a0; in this case, the maximum accelera-
it would be possible to design the structure in tion of the block is a0, there are no relative
such a way that its resistance be never attained displacements, and the maximum internal
during the seismic event (ac = Tlim/m > a0), but forces are evaluated with a = a0.
this approach to the design is not particularly 2. The critical acceleration of the system ac is
desirable, especially for severe, low probability smaller than the maximum acceleration at the
events, because it may be uneconomical and also base a0; in this second case, the maximum
because the seismic behavior would not benefit acceleration of the block is ac; the
displacements increase as ac decreases; the

a
internal forces are evaluated with a = ac and ac
increase with ac.
t
This conceptual model can be used to illustrate
a performance-based approach to the design of
earth-retaining structures. A given limit state cor-
vr
responds to a certain design seismic action that
must be found accounting for the local site
response. The required seismic performance of
a retaining structure can be expressed by its
permanent displacement and can be chosen by t
the designer for compatibility with the limit ur
state under consideration. The design consists
on one hand in endowing the system with a
critical acceleration sufficiently large to produce
displacements that do not exceed the specified t
limit and on the other hand in protecting the
structural elements from a premature yielding Seismic Design of Earth-Retaining Structures,
Fig. 2 Illustration of the Newmark (1965) integration
by making sure that their structural capacity is scheme
sufficient to carry the maximum internal forces
occurring during the development of the
displacements. characterization of the soil deposits, starting from
the bedrock.
Evaluation of Seismic Displacements For simple earth-retaining structures, the out-
The conceptual model of Fig. 1 can be used crop seismic action is more commonly expressed
directly to evaluate, for a given acceleration as an elastic response spectrum, and the local site
time history at the base, the seismic displace- response is considered through simple amplifica-
ments of a retaining structure. This is accom- tion coefficients, usually based on the classifica-
plished as shown in Fig. 2: whenever the base tion of the subsoil according to the shear wave
acceleration a(t) exceeds ac, the time history of velocity of the topmost, say, 30 m. In this case,
the relative acceleration a(t) – ac is integrated to acceleration time histories are not available: a
provide the relative velocity vr. The integration is direct calculation of the displacements is not pos-
halted when vr equals zero and is restarted when sible, and it is necessary to use a simplified
a(t) becomes again larger than ac. It is recognized approach.
that a retaining wall can move only downhill; The final displacement induced by the accel-
therefore, the positive and the negative portions eration time history of Fig. 1 can be evaluated as
of the acceleration time histories are considered

one at a time. The displacements are obtained v20 ac
ur ¼ 1 (2)
from the integration of the time history of the 2ac a0
relative velocity.
The direct evaluation of the displacements where v0 = a0 t0 is the maximum acceleration at
depicted in Fig. 2 is very simple, but can be the base. Newmark (1965) argued that a given
performed only if the seismic action is described acceleration time history can be approximated as
as a set of accelerograms. This implies the eval- amax/ac cycles equivalent to the simple impulse of
uation of the outcrop seismic motion that must be Fig. 1, where amax is the maximum recorded
inserted into a detailed site response calculation, acceleration. This assumption results in the fol-
which in turn requires an accurate geotechnical lowing expression for the final displacement:
Franklin & Chang (1977) previous relationships or of the diagram of

amax = 0.5 g
Fig. 3 is quite immediate; the critical acceleration
vmax = 0.76 m/s
expresses the resistance of the retaining structure
10
to the seismic action and can be evaluated using
the earth pressure formulas given in the follow-
ing; amax is the maximum expected acceleration
for the limit state under consideration; the maxi-
1 mum velocity vmax can be estimated, for instance,
ur (m)
from the constant-velocity portion of the elastic

spectra provided by the construction codes (e.g.,
EN 1998-5 2003).
However, Fig. 3 shows that for a given ratio
0.1 Newmark (1965)
ac/amax, the available relationships result in a
Richards & Elms (1979)
very large uncertainty for the displacement. The
Wong (1982) - average
reason for this substantial scatter is that the max-
Whitman (1990) - 95%
confidence
imum recorded acceleration alone is not suffi-
0.01 cient to represent the characteristics of an
0.01 0.1 1 accelerogram. While in principle it is possible to
ac/amax
produce more accurate expressions that relate the
Seismic Design of Earth-Retaining Structures,
displacements to additional properties of the seis-
Fig. 3 Seismic displacements plotted as a function of mic motion (frequency content, Arias intensity,
the critical to maximum acceleration ratio duration, etc.), often the designer that wishes to
adopt a simplified procedure is provided only

with an estimate of amax. An alternative way to
v2max ac amax obtain a more accurate prediction of the seismic
ur ¼ 1 (3)
2ac amax ac displacement is to use the structure of the above
predictive equations for the development of local
where vmax is the maximum recorded relationships based on regional seismic records.
acceleration.
Franklin and Chang (1977) selected a number Equivalent Seismic Action on Retaining
of accelerograms scaled to amax = 0.5 g and Structures
vmax = 0.76 m/s; they produced a parametric Consider a relationship between the displacement
integration of these accelerograms for different ur and the ratio ac/amax that can be deemed appro-
values of the ratio ac/amax. Richards and Elms priate to the local seismicity (Fig. 4). If u0 is the
(1979), based on the same acceleration time his- tolerable displacement for the limit state under S
tories, produced the following relationship for the consideration and b is the corresponding value of
prediction of the seismic displacements: the ratio ac/amax, then the requisite that u ur

implies that the critical acceleration ac must be
v2max amax 4 larger than b amax (with b < 1). This means that a
ur ¼ 0:087 (4)
amax ac check that the desired performance is met can
also be done by controlling that under the con-
The displacements predicted by the above rela- ventional seismic forces associated to the accel-
tionships are plotted in Fig. 3 as a function of the eration bamax, the resistance of the system is not
ratio ac/amax, together with the results by Franklin exceeded.
and Chang (1977) and with the curves proposed In other words, the performance of the
by Wong (1982) for the average values and by retaining structure that in principle should be
Whitman (1990) for the values associated to a evaluated in terms of displacements can also be
95 % confidence. In principle, the use of the checked in terms of equivalent forces. This is
In the absence of displacements, the coefficient

b of Eq. 5 is equal to 1, and therefore the seismic
0.1 earth pressure can be evaluated using the formulas
given in the following section, using a horizontal
seismic coefficient corresponding to the maximum
acceleration. However, there are cases in which a
ur (m)
different approach is necessary, and the soil should

0.01 be regarded as an elastic material. These are dealt
with at the end of the next section.
Evaluation of Earth Pressure

0.001
0 0.2 0.4 0.6 0.8 Earth Pressure in Active and Passive Limit
β = ac/amax
States
When a plastic mechanism is mobilized, forces
Fig. 4 Relationship between seismic displacements and and/or stresses transmitted by the soil in contact
the ratio ac/amax developed by Rampello et al. (2010) for with the retaining structure may be evaluated by
stiff soils based on the Italian seismicity assuming that the soil is in limit conditions, that
is, its shear strength is completely mobilized. If
the retaining structure displaces away from the
commonly expressed in terms of seismic coeffi- soil, the soil is in an active limit condition; con-
cients: seismic forces deriving from a horizontal versely, if the soil is contrasting the displacement
seismic coefficient, of the retaining structure, it is in a passive limit
condition.
kh ¼ bamax =g; (5) For a purely frictional strength criterion, the
resultant force applied by the soil to the retaining
generally used in conjunction with a vertical structure may be found using the Mononobe-
seismic coefficient expressed as a fraction of kh, Okabe approach (Okabe 1924; Mononobe and
must be such that the system capacity is not Matsuo 1929), which is an extension to seismic
exceeded. conditions of the Coulomb’s limit equilibrium
Note that no global or partial safety coefficient method. In active limit conditions, the total
should be used in this check, as this is not an earth thrust SaE is provided by a volume of soil
assessment of the safety with respect to a collapse that slides downward along a planar surface with
mechanism, but rather an indirect evaluation of an inclination a to the horizontal; this mechanism
the seismic performance of the system. is shown in Fig. 5, where j is the angle of shear-
ing resistance of the soil and d is the angle of
Non-displacing Retaining Structures friction at the soil-structure contact. The earth
In some circumstances, a retaining structure can- thrust is found by maximizing with respect to a
not undergo permanent deformation resulting the force that ensures the translational equilib-
from the activation of a plastic mechanism. This rium for the soil wedge of Fig. 5, subjected to
may be due either to the design requirements for the bulk force WE and to the resultant force
the specific limit state implying a negligible dam- R applied by the in situ soil onto the sliding
age to the structure or to the retaining structure plane. Since the shear strength of the soil is
being significantly constrained by additional fully mobilized along the sliding surface, R is
structural elements, like props, rakers, etc. inclined by j with respect to the normal to the
Seismic Design of Earth-

ε
Fig. 5 Illustration of the
Mononobe-Okabe khW SaE
approach for the evaluation
of the active seismic force θ WE
WE W We
on a retaining structure SaE θ
ρ R
δ kvW
ϕ –kvW
SaE
R khg
α
kvg
sliding plane and SaE has an inclination d with 1

SaE ¼ g ð1 kv Þ H 2 K aE (6)
respect to the normal to the retaining structure. 2
The seismic forces are included in the analysis by
considering the horizontal and vertical seismic where g is the soil unit weight, H is the height of
coefficients kh and kv in the evaluation of WE. the retaining structure, and KaE is the coefficient
The value of SaE, maximized with respect to a, of active thrust, evaluated with the following
is given by the expression equation:
cos2 ðj b yÞ
K aE ¼ " # (7)
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2
sin ðj þ dÞ sin ðj e yÞ
cos y cos b cos ðd þ b þ yÞ 1 þ cos ðd þ b þ yÞ cos ðe bÞ
2
in which the angles b and e have the meaning The inclination acr of the sliding surface that
depicted in Fig. 5 while the angle y defines the maximizes SaE can be evaluated with the follow-
direction of WE: ing expression:
kh
y ¼ arctan (8)
1 kv
" 1
# S
ftan aðtan a þ cot bÞ ½1 þ tan ðd þ b þ yÞ cot bg2 tan a
acr ¼ ðj yÞ þ arctan (9)
1 þ tan ðd þ b þ yÞ ð tan a þ cot bÞ
with the dependency of KaE on kh (Eqs. 7 and 8);

therefore, it can be regarded as the function
a¼jey Sa(ab) introduced in the previous section and
(10)
b¼jby used accordingly.
Equations 6, 7, and 8 imply quite obviously
The active pressure SaE given by Eq. 6 is a that SaE increases with the horizontal seismic
function of the horizontal acceleration, through coefficient kh. Conversely, it can be seen from
Eqs. 9 and 10 that the angle acr decreases with kh, the assumption of a planar sliding surface is not
implying that the dimensions of the sliding wedge conservative, especially when the soil-wall fric-
increase with the earthquake intensity. In other tional coefficient d approaches j.
words, for larger accelerations, the volume of soil A more general approach to evaluate the seis-
that interacts with the retaining structure is larger, mic pressure in active and passive limit condi-
and this finding has several implications, for tions is to use the lower-bound theorem of limit
instance, in the selection of the relevant strength analysis. A convenient expression for vertical
parameters or for the design of a drainage system walls was derived by Lancellotta (2007) for pas-
that needs to be effective in controlling the pore sive limit conditions and extended by Rampello
pressure within the entire volume of soil et al. (2011) to active limit conditions. The effec-
interacting with a retaining wall. tive normal stress acting on the retaining struc-
Since the Mononobe-Okabe method considers ture in active or passive limit conditions s0 a,p is
only translational equilibrium, it provides no direct expressed as the product of a suitable thrust coef-
information on the point of application of SaE. ficient Ka,p and the notional vertical effective
For passive limit conditions, the Mononobe- stress s0 v; the thrust coefficient has the
Okabe approach is not recommended, because expression
2 3

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
6 cos d 7 cos e
K a, p ¼ 4 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi cos d sin2 j sin2 d 5 ð1 kv Þ e2c tan j
cos ðe yÞ sin j sin ðe yÞ
2 2 cos y
(11)
with is better evaluated using the above lower-bound

solution. The total contact stress sa,p is equal to

sin d sin ðe yÞ the sum of the effective contact stress s0 a,p and
c ¼ 0:5 arcsin arcsin
sin j sin j the pore pressure u:

d þ ðe yÞ þ 2y
sa,p ¼ K a,p s0v þ u ¼ K a,p ðsv uÞ þ u (13)
(12)
The change in total contact stresses sa,p is not
In the above expressions, the upper and lower caused only by the introduction of u in the above
operators apply to active and passive conditions, expression but also by the coefficients of earth
respectively. Equations 11 and 12 imply that Ka pressure in seismic condition depending on the
increases and Kp decreases with y, that is, with pore water pressure.
the seismic acceleration. For most practical purposes, it can be assumed
Providing normal stresses that are propor- that during the seismic motion, the pore water
tional to the vertical effective stresses, this vibrates in phase with the soil skeleton
lower-bound solution yields the entire distribu- (Matsuzawa et al. 1984 suggest that this is the
tion of soil-wall contact stresses that can be used case for permeability coefficients smaller than
to study the equilibrium of the retaining structure 105 m/s). The evaluation of the seismic pressure
for the selected limit state. should consider that the inertial forces in the soil
are proportional to the total mass of the soil
Effect of Pore Water Pressure skeleton and the pore water while the shear
The presence of pore water pressures produces a strength is proportional to the effective stresses.
change in the seismic contact stresses. This effect Callisto and Aversa (2008) showed that this
effect can be taken into account by modifying the If a wall retains an excavation in a fine-grained
expression of y in Eq. 8, which becomes soil, in the long-term, the removal of soil pro-
duces a significant over-consolidation and the

0 sv kh volumetric response of the soil becomes dilatant.
y ¼ arctan 0 (14)
sv 1 kv In this case, the seismic event is likely to produce
a decrease of the pore water pressures (Du < 0),
where sv is the total vertical stress. This expres- and this beneficial effect may be neglected. How-
sion leads to larger values of Ka and to smaller ever, for relatively loose coarse-grained soils, the
values of Kp, showing that pore water pressures earthquake may produce a significant increase Du
have a detrimental effect on the coefficients of of the pore water pressure. The evaluation of Du
earth pressure. Expression (14) implies that Ka for coarse-grained material is beyond the scope of
and Kp may vary with depth even in a homoge- this entry; the reader may refer to the sections
neous soil. For the particular, and quite uncom- devoted to liquefaction and to Seed and
mon, case of a hydrostatic pore water pressure Booker (1977).
distribution with the total hydraulic head at the An excavation in fine-grained material can be
ground surface, expression (14) becomes inde- taken to occur in approximately undrained con-
pendent of depth: ditions, generating negative excess pore water
pressures. Traditionally, the static design of an

excavation in these conditions is carried out
0 g kh
y ¼ arctan 0 (15) expressing the shear strength of the soil in terms
g 1 kv
of total stresses, using the Tresca criterion and a
limiting shear stress equal to the undrained shear
where g is the unit weight and g0 is the submerged strength Su. Yet for the analysis of the seismic
weight of the soil. Use of expression (15) is conditions, a total stress analysis is seldom appro-
explicitly suggested by Eurocode 8 part priate, because the time interval corresponding to
5 (EN 1998-5 2003). a significant dissipation of the excess pore water
pressure may be considered to be much smaller
Effect of Cohesion that the return period of any severe seismic event.
The relationships listed above were developed for Therefore, the seismic design is controlled by the
a purely frictional material. For some of these seismic events that, having a large return period
solutions, the possibility exists to account rigor- (e.g., 475 years for a 10 % probability over a life
ously for the effect of cohesion. However, small span of 50 years), may be deemed to produce
values of the cohesion c0 can be incorporated in effects on a retaining structure that is already in
the analysis with sufficient accuracy using the drained conditions.
Rankine expression: S
Evaluation of Earth Pressure for
pffiffiffiffiffiffiffiffi
s0a,p ¼ 2c0 K a,p þ K a,p s0v (16) Non-displacing Retaining Structures
Although the case of zero displacement can be
addressed simply by inserting b = 1 into Eq. 5
Drainage Conditions and therefore evaluating the earth pressure using
For a saturated soil, it can be assumed that the the maximum predicted acceleration, there are
seismic event occurs in undrained conditions, cases in which active limit conditions may not
even for coarse-grained materials. The excess develop at the rear of a retaining structure,
pore water pressures Du generated during the because of the very small deformation allowed
earthquake can be easily incorporated into during the excavation or because the wall has
Eq. 13, to provide (for a purely friction material) been backfilled after its construction. In these
situations, if the deformation of the retaining
sa,p ¼ K a,p ½sv ðu þ DuÞ þ ðu þ DuÞ (17) structure is impeded during the seismic event, a
different approach is in order to evaluate the earth together with Eqs. 18 and 19, may be used to
pressure, based on the hypothesis that the soil evaluate the design earth pressures.
does not attain its strength and can be modeled
as a linearly elastic material. The increase in the
horizontal stress produced by the earthquake can Gravity Retaining Walls
then be expressed as
In static conditions, gravity retaining walls trans-
amax fer the sub-horizontal earth thrust S to the foun-
DshE ¼gH f (18)
g dation soil, thanks to their considerable self-
weight. Figure 7 shows that the weight of the
where H is the height of the retaining structure. wall W combines with S to yield a resultant
For an infinitely rigid wall, the coefficient f may force R that (i) goes through the base of the wall
be taken equal to 1 (Wood 1973). For a flexible and (ii) has a small inclination to the vertical. As
wall, the coefficient f is a function of the soil-wall shown in Fig. 7, different types of gravity
relative stiffness dp, expressed as (Younan and retaining walls have a similar global behavior,
Veletsos 2000) the only difference being the relative amount of
soil contributing to the total weight of the
Gs H 3 structure.
dp ¼ (19)
ðEI Þw Since these structures are endowed with a
significant mass, seismic accelerations produce
In the above relationship, Gs is the shear modulus large inertial forces that must be taken into
of the backfill, and the quantity (EI)w is the bend- account, considering both the horizontal and ver-
ing stiffness of the retaining structure (per out-of- tical acceleration components.
plane length). Figure 6 shows the coefficient Construction codes provide the values of the
f plotted as a function of the relative stiffness dp seismic coefficient that should be used for the
as obtained by Younan and Veletsos (2000) for design, essentially in the form of the reduction
two different restraining conditions. This graph, coefficient b of Eq. 5, depending on the allowable
displacement at the end of the earthquake. Using
the appropriate seismic coefficient, the design of
1 a gravity retaining wall is carried out studying
global and local mechanisms. The global mecha-
nisms are essentially the same for the different
0.8 types of retaining structures depicted in Fig. 7:
(a) sliding along the base, (b) bearing capacity of
the foundation, (c) overturning, and (d) overall
0.6
slip-circle mechanism (Fig. 8). Local mecha-
nisms are relative to the attainment of the struc-
f
0.4 tural strength or to the mobilization of the

reinforcement strength in a reinforced earth
structure.
0.2
The details of each safety check are given by
the construction codes and therefore are not dealt
0 with here. However, a general appreciation of the
0 10 20 30 40 50 seismic design of a retaining wall is provided by
dp
the example of Fig. 9, relative to a cantilever
retaining wall with a height H = 3.5 m subjected
Fig. 6 Coefficient f of Eq. 18 plotted as a function of to a seismic event characterized by a maximum
the relative stiffness dp acceleration amax = 0.25 g. Restricting the
S S S
W W W
R R R
Seismic Design of Earth-Retaining Structures, Fig. 7 Transfer of earth pressure to the foundation soils for gravity
retaining walls
a b
c
d
Seismic Design of Earth-Retaining Structures, Fig. 8 Global limit states for a gravity retaining wall
S
attention to the sliding mechanism only, Fig. 9 with a large base to have a particularly good
shows the values of the critical acceleration ac seismic performance, it must be endowed with a
plotted as a function of the width B of the wall correspondingly large structural strength to resist
base and of the corresponding displacements the internal forces associated with its large criti-
evaluated using the relationship of Fig. 4. It is cal acceleration. Following this line of thought, it
evident that wider walls have a larger critical is evident that a wall with a critical acceleration
acceleration and therefore undergo smaller dis- equal to amax will suffer negligible displacement
placements. The figure also shows the maximum and will be subjected to internal forces that can-
bending moment M(ac) in the wall stem, evalu- not be larger than those evaluated with an accel-
ated when the wall is critically accelerated. It is eration equal to amax.
evident that since the bending moments increase It should be noted that any “safe” provision,
with ac, they must show an increasing trend such as the use of strength or resistance factors, or
with B. Hence, if a retaining wall is designed an underestimation of the soil strength, leads to
a b 160
ac = amax
140
M (amax)
0.2
120
ac (g), ur (m)
M (kNm/m)
γ = 20 kN/m 3
c' = 0,ϕ' = 32° 100

H = 3.5 m
amax = 0.25 g 0.1 80
ac
u 60
M (ac)
c' = 0,ϕ' = 24° 0.6 B
0 40
B 1.5 2.0 2.5
B (m)
Seismic Design of Earth-Retaining Structures, Fig. 9 (a) Example of a retaining wall; (b) critical acceleration,
displacement, and maximum bending moment plotted as a function of base width
Seismic Design of Earth- a b

Fig. 10 Reference
schemes of a cantilevered
(a) and a singly propped (b) H
embedded retaining wall, H
showing the direction of the
inertial forces activating the kc γ/g L
plastic mechanisms L kc γ/g
d
d
design acceleration smaller than the critical one. possesses a large mass and the corresponding
Figure 9 shows clearly that a larger critical accel- inertial forces are small; the effect of vertical
eration implies on one hand smaller displace- acceleration can usually be neglected.
ments and on the other hand larger internal On the other hand, cantilevered or singly
forces. For the structural design of the wall, it is propped walls (Fig. 10) bear some similarities
then recommendable to evaluate always the inter- with gravity retaining walls, because if the
nal forces corresponding to the actual critical strength of the soil interacting with the wall is
acceleration of the wall. attained during the seismic event, a plastic mech-
anism may develop, and the earth pressure rela-
tionships may be used in a limit equilibrium
Embedded Retaining Walls computation to evaluate the critical acceleration
that activates the mechanisms. Figure 11 shows
The seismic behavior of embedded retaining an example in which a cantilevered retaining wall
walls is different from that of gravity retaining is subjected to a maximum acceleration
walls, in that the earth thrust is resisted by the amax = 0.5 g. For this wall, the critical accelera-
passive strength of the soil in front of the wall and tion can be obtained studying the rotational
by additional constraints (anchors, props); there- mechanism of Fig. 10 using, for instance, the
fore, it is not necessary that the structure Lancellotta (2007) closed form expressions for
a b
0.5 ac
u (Rampello et al. 2010)
0.4 u (Richards & Elms 1979)
M (ac) 300
ac (g), u (m)
H=4m
M (kNm/m)
γ = 20 kN/m3 0.3
ϕ' = 35° 200
0.2
δ = 20°
L
0.1
100
d
0
3.0 4.0 5.0
d (m)
Seismic Design of Earth-Retaining Structures, Fig. 11 (a) Example of an embedded retaining wall; (b) critical
acceleration, displacement, and maximum bending moment plotted as a function of embedded length
the active and passive pressures. The results are standard method for the static design of multi-
very similar to those developed for the example constrained retaining wall (Fig. 12a). While there
retaining wall of Fig. 9, showing that increasing is not a general consensus on the use of this
the embedded length of the wall the critical accel- method in seismic conditions, a simple strategy
eration increases, and therefore the permanent for taking into account the inertial force consists
displacement decreases but the internal forces in of the following steps:
the wall become larger. For these wall types, it is
recommended that the internal forces be evalu- (i) Carry out a static analysis to model the
ated considering the critical value of the acceler- construction sequence (Fig. 12b).
ation, that is, considering the activation of the (ii) Evaluate the additional forces due to the
plastic mechanisms depicted in Fig. 10. earthquake and the reduction of the coeffi-
When the structure has several additional con- cient of passive resistance, as a function of
straints, the attainment of the soil strength is not the maximum acceleration estimated for
sufficient for the activation of a plastic mecha- the site.
nism: if the design requires that the capacity of (iii) Apply these forces as external loads on the
the wall itself and of the constraints cannot be subgrade reaction model, and reduce the
reached, then these should be classified as a coefficient of passive resistance (Fig. 12c).
non-displacing wall and the earth pressure should (iv) Run the program for equilibrium and com- S
be evaluated accordingly, as explained in a pre- patibility (Fig. 12d).
ceding section. In this perspective, no consider-
ation of plastic mechanisms is needed, and the
objective of the design is essentially the evalua- Summary
tion of the maximum internal forces in the struc-
tural members that should derive necessarily The current trend in the design of earthquake-
from a study of the soil-structure interaction. resistant structures is based on the evaluation of
However, the analysis of the soil-structure the seismic performance of the system. For a
interaction under seismic conditions is not retaining structure, the seismic performance is
straightforward and can be carried out at different expressed by its cumulative displacement at the
levels of complexity. A very basic representation end of the earthquake. Although it would be
of the soil-structure interaction is represented by desirable to evaluate directly the performance of
the subgrade reaction methods that constitute a the structure, it has been shown that a design
a b c d
subgrade find contact stresses find new contact stresses

reduce Kp
reaction model resulting from excavation and internal forces
add seimic forces
Seismic Design of Earth-Retaining Structures, Fig. 12 Use of a subgrade reaction model for the seismic analysis of
a multi-propped retaining wall
requirement limiting the seismic displacements Matsuzawa H, Ishibashi I, Kawamura M (1984) Dynamic
can be transformed into an equivalent require- soil and water pressures of submerged soils. J Geotech
Eng 111(10):1161–1176
ment on the seismic action that needs to be con- Mononobe N, Matsuo H (1929) On the determination of
sidered in a conventional design process, which is earth pressure during earthquakes. In: Proceedings of the
commonly based on forces. Specifically, the con- world engineering congress, Tokio, vol 9, pp 177–185
ventional seismic coefficient used in a pseudo- Newmark NM (1965) Effects of earthquakes on dams and
embankments. Fifth Rankine lecture. Géotechnique
static calculation can be made to decrease as the 15(2):139–193
allowable displacement decreases. Specific cases Okabe S (1924) General theory on earth pressure and
of non-displacing walls have been evidenced that seismic stability of retaining wall and dam. J Jpn Civ
may call for a different approach, in which the Eng Soc 10(5):1277–1323
Rampello S, Callisto L, Fargnoli P (2010) Evaluation of
internal forces are evaluated assuming that slope performance under earthquake loading condi-
the soil does not mobilize its strength during the tions. Ital Geotech J 44(4):29–41
earthquake. Rampello S, Callisto L, Masini L (2011) Spinta delle terre
sulle strutture di sostegno. Atti delle XXIII Conferenze
Geotecniche di Torino
Richards R, Elms DG (1979) Seismic behavior of gravity
retaining walls. J Geotech Eng Div ASCE
References 105(GT4):449–464
Seed HB, Booker JR (1977) Stabilization of potentially
Callisto L, Aversa S (2008) Dimensionamento di opere di liquefiable sand deposits using gravel drains.
sostegno soggette ad azioni sismiche. In: Opere J Geotech Eng 103(7):757–768
geotecniche in condizioni sismiche, MIR 2008. Pàtron, Whitman RV (1990) Seismic design and behavior of grav-
Bologna, pp 273–308 ity retaining walls. In: Proceedings of the design and
EN 1998-5 (2003) Eurocode 8: design of structures for performance of earth retaining structures, Ithaca,
earthquake resistance – part 5: foundations, retaining ASCE GSP 25, pp 817–842
structures and geotechnical aspects. CEN European Wong CP (1982) Seismic analysis and improved seismic
Committee for Standardization, Brussels design procedure for gravity retaining walls. MSc the-
Franklin AG, Chang PK (1977) Earthquake resistance of sis, Department of Civil Engineering, M.I.T., Cam-
earth and rockfill dams. Report 5, Permanent displace- bridge, MA
ment of earth embankments by Newmark sliding block Wood JH (1973) Earthquake-induced soil pressures on
analysis. MP S-71-17, Soils and Pavements Labora- structures. PhD thesis, California Institute of Technol-
tory, US Army Waterways Experiment Station, ogy, Pasadena
Vicksburg Younan AH, Veletsos AS (2000) Dynamic response of
Lancellotta R (2007) Lower-bound approach for seismic flexible retaining walls. Earthq Eng Struct Dyn
passive earth resistance. Géotechnique 57(3):319–321 29:1815–1844
Seismic Design of Pipelines 2781
pipelines which hindered firefighting efforts.

Seismic Design of Pipelines Several years later, in the 1971 San Fernando
earthquake, about 80 % of the reported destruc-
Hany El Naggar tion were in welded steel buried pipelines. Fol-
Department of Civil and Resource Engineering, lowing the 1989 Loma Prieta earthquake, the East
Dalhousie University, Halifax, NS, Canada Bay Municipal Utilities District identified over
120 water pipeline breaks, and the San Jose
Water Company reported another 155 pipe
Synonyms breaks. However, most of the serious damage
reported by Pacific Gas and Electric Company
Earthquakes; Lifelines; Liquefied soil; PGD; (PG&E) occurred to natural gas mains and ser-
Pipeline breakings; Seismic-induced settlement vice lines. During the Northridge earthquake
(1994) in California, some natural gas pipelines
were severely damaged and the leak of contain-
Introduction ment fluid caused a large explosion in the
Northridge town resulted in several deaths (Lau
Pipelines are a cost-effective means for the trans- et al. 1995; O’Rourke and Palmer 1996). In the
portation of water supply and sewers, in addition 1995 Hyogo-Ken Nanbu earthquake (in Japan),
to commercial fluids such as oil and gas. The the natural gas leakage from buried pipelines
designation of pipeline systems as “lifelines” resulted in numerous number of fires (531 cases
indicates that they should be designed to function reported) which started primarily due to gas
and operate at all times for public safety and well- release and electricity sparks affecting areas of
being and also for economic reasons. Engineers over one square kilometer totally burnt
must consider the different types of loads that are (Scawthorn and Yanev 1995). More recently,
going to be imposed on the pipeline, the environ- the 1999 Chi-Chi earthquake (in Taiwan) also
ment that the pipeline will travel in, and the type caused severe damage to natural gas distribution
of material that the pipeline is going to convey. systems. More than 100,000 customers were
Consequently, pipelines must be designed for affected after the earthquake, and the estimated
different loads such as stresses due to pressure economic loss of five major natural gas compa-
generated by the flow (internal pressure), external nies was approximately US$ 25 million (Chen
pressure generated by the weight of earth and by et al. 2000). In the 1999 Izmit earthquake
live loads for buried pipelines, or external pres- (in Turkey), Tupras refinery suffered serious fire
sure generated by fluid if the pipe is submerged damage as the pipeline conveying water from
underwater. Also, seismic loads (earthquake neighbor lake was damaged, so the refinery was
loads) in medium and high seismicity zones are dependent on internal water reservoirs which S
detrimental and should be considered. In general, were insufficient. In 2001 two earthquakes
pipeline design includes several general steps: occurred (1 month apart) in San Salvador; several
(1) load determination, (2) critical performance pipelines in rural areas experienced severe
evaluation (determining the critical stress and/or destruction caused by huge landslides. Till now,
deformation), (3) comparison of the critical per- in several very recent earthquakes occurred in
formance with the limiting criteria established by Chile (2010), New Zealand (Darfield, 2010),
adopted codes and standards, and finally (4) selec- Japan (Tohoku, 2011), and Italy (earthquake of
tion of the pipe. Emilia, 2012), serious damages to pipelines have
Over the past century, several catastrophic been witnessed.
earthquakes caused severe damage to buried Accordingly, pipelines should be designed to
pipelines. In the 1906 San Francisco earthquake, function and operate during and following design
one of the main reasons that caused the extensive earthquakes for life safety and economic reasons.
damage was due to the failure of several water As mentioned above several earthquakes in the
2782 Seismic Design of Pipelines
Seismic Design of Pipelines, Fig. 1 Main types of faults
last few decades resulted in too many pipeline presents methods to calculate the amount of
breakings, and that extensive pipe breakage has PGD as well as the extent of the PGD zone.
the potential to lead to great economic harm to Equations to quantify the amount of PGD are
our urban communities. provided. Also, useful observations and notes to
The main focus of this chapter will be centered determine the extent of the PGD zone are
on the seismic analysis and design of pipelines. presented and discussed.
Fault
Permanent Ground Deformation and Stresses in the earth’s crust push the two sides of
Seismic Wave Propagation Hazards the fault. Eventually enough stress builds up and
the rocks slip suddenly releasing energy in waves
In seismic events, buried pipelines can be dam- that travel through the rock to cause earthquake.
aged mainly by either the permanent ground Accordingly, earthquakes occur on faults. A fault
deformations (PGD) or by the transient seismic is a thin zone of crushed rock separating blocks of
wave propagation. PGD movements include the earth’s crust. When an earthquake occurs on
faulting, landslides, lateral spreading due to liq- one of these faults, the rock on one side of the
uefaction, and seismic settlement. Even though fault slips with respect to the other. The fault
PGD risks are usually restricted to small regions surface can be vertical, horizontal, or oblique to
within the pipeline, the chances of them causing the surface of the earth. If the earthquake magni-
severe damage is substantial since they impose tude is large enough, the offset along the fault
large deformations. On the other hand, the seis- will propagate all the way to the earth’s surface
mic wave propagation risks typically affect the causing surface rupture (fault offset). Figure 1
whole pipeline, but with lower damage rates below shows the main types of faults. In normal
(as the total deformations are general less and and reverse faults, the major ground displace-
not permanent). For example, only 5 % of the ment is vertical with a minor horizontal displace-
affected area experienced lateral spreading dur- ment. These ground displacements pose axial
ing the 1906 San Francisco earthquake; approxi- tension/compression and bending stresses in the
mately 52 % of all pipeline breaks occurred pipeline depending on the direction of move-
within one city block of the lateral spreading ment. On the other hand, in strike slip fault the
(O’Rourke et al. 1985). main deformation (the offset) occurs in the hori-
zontal plane, which poses axial tension/compres-
Permanent Ground Deformation (PGD) sion and bending stresses in the pipeline
This section describes in details the four different depending on the intersection angle of the pipe-
forms of permanent ground deformation and line and the fault.
The following empirical equations by Wells ac ¼ g ðFOS 1Þ sin a (4)

and Coppersmith (1994) are the most recognized
equations to estimate the average fault displace- where g is the acceleration due to gravity, FOS
ment relative to the size of the considered is the factor of safety, and a is the angle of the
earthquake: slope.
Then the displacement of the block can be
For the normal fault ! log df ¼ 4:45 þ 0:63 M calculated by double integrating the ground
acceleration. Jibson and Keefer (1993) proposed
(1)
the following equation to estimate the Newmark
displacement, ds, in centimeters as
For the reverse fault ! log df ¼ 0:74 þ 0:08 M
(2) Log dS ¼ 1:460 log I a 6:642 ac þ 1:546 (5)
For the strike slip fault ! log df ¼ 6:32 þ 0:90 M where ac is the critical acceleration in gs and Ιa is
(3) the Arias Intensity in m/s.
Arias Intensity can be calculated using the
where df is the average fault displacement in following relationship developed by Wilson and
meters and M is the moment magnitude of the Keefer (1983):
earthquake.
Log I a ¼ M 2 log R 4:1 (6)
Landslides
where M is the earthquake magnitude and R is
Seismically induced landslide involves a wide
source distance in kilometers.
range of downslope mass ground movements,
which can occur in offshore, coastal, and onshore
Lateral Spreading Due to Liquefaction
environments.
Seismic shaking may trigger the liquefaction of
saturated loose cohesionless soils. The liquefac-
Offshore Landslides In offshore landslides, the
tion process increases the pore water pressure in
response of the pipeline is mainly governed by
the ground to a level where the effective stress
the orientation of the pipeline to the direction of
approaches zero at which point the soil loses
ground movement. The imposed displacements
entirely its shear strength, which in turn results
for undersea slides are so large that pipeline
in the lateral movement of the soil (lateral spread-
response is likely controlled by the maximum
ing). In past major earthquakes, large number of
force available at the soil-pipe interface.
pipelines suffered massive damage caused by
liquefaction-induced ground failures (Zhang
Onshore Landslides There are several types of S
et al. 2004).
onshore landslides based on soil movements,
Youd et al. (2002) proposed two empirical
geometry of the slide, and the types of material
equations to approximately estimate the PGD
involved. The main types are rock falls, rock
due to liquefaction:
topples, slides, and lateral spreads. Rock fall
For lateral spreads down gentle ground
and rock topple can cause direct damage to
slopes (GS),
above-ground pipelines by the impact of falling
rock fragments. In an earth slide the earth moves
relatively as a block; they typically develop along LogdL ¼ 16:213 þ 1:532 M 1:406 log R
natural slopes and embankments. 0:012 R þ 0:338 log S þ 0:54log T 15
Based on Newmark’s Block model for land-
þ 3:413 logð100 F15 Þ
slides (Newmark 1965), the critical acceleration
at which the slide will be triggered ac can then be 0:795logðD5015 0:1Þ
determined from (7)
Seismic Design of B B
Pipelines, Fig. 2 Sketch
showing the ground slope
and the free face lateral A
A Y = 100 A/B
spreads S = 100 A/B Slip surface
Ground slope Free face
For lateral spreads at a free face (FF), For leveled sites
Log dL ¼ 16:710 þ 1:532 M 1:406 log R amax

dGS ¼ 0:30 H 1 þ2 (11)
N
0:012 R þ 0:592 log Y þ 0:54logT 15
þ 3:413 logð100 F15 Þ where dGS is the liquefaction-induced settlement
0:795logðD5015 0:1Þ in centimeters, H1 is the thickness of saturated
(8) cohesionless soil layer in meters, H2 is the height
of embankment in meters, N is the SPT N-value in
where dL is the PGD in meters, M is the earthquake the cohesionless layer, and amax is the ground
magnitude, S is the ground slope %, g is the free acceleration in gals.
face ratio (in % see Fig. 2), T15 is the thickness in
meters of the saturated cohesionless soil layer with Seismic Wave Propagation
a corrected standard penetration test (SPT value of For the seismic analysis and design of buried pipe-
less than 15), F15 is the percentage of average fines lines, the effect of the seismic wave propagation on
contents in T15 (in %), D5015 is the mean grain size the pipeline is usually characterized by the induced
in mm in T15, and R* is an adjusted distance ground strain and curvature. Newmark (1967)
parameter in kilometers given by developed a straightforward method to estimate
the ground strain. The general form of a traveling
R ¼ R þ 10 eð0:89M5:64Þ (9) wave in Newmark’s method is given by
where R is the horizontal distance in kilometers t x

from the site of interest to the nearest bound of the U¼f þ (12)
T l
seismic energy source (do not use less than 0.5 km).
where U is the function of the separation distance
Seismic Settlement between the two points, x, and the speed of the
Seismic-induced settlement may be caused by seismic wave; T is the period of the repeating
densification of cohesionless soils, consolidation motion; and l is the wavelength.
of cohesive soils, or consolidation of liquefied For particle motion parallel to the direction of
soil. In this section we will only discuss propagation (R-waves), eg, the ground strain along
liquefaction-induced ground settlement as it can the direction of propagation can be calculated as
cause larger settlement and hence higher poten-
tial for damage to buried pipelines. V max
eg ¼ (13)
Takada and Tanabe (1988) proposed the two CR
following empirical equations to calculate
liquefaction-induced settlement at embankments where Vmax is the maximum horizontal ground
and leveled sites: velocity in the direction of wave propagation and
For embankments CR is the propagation velocity of the R-wave.
For particle motion perpendicular to the direc-
amax tion of propagation (S-waves), Cg, the curvature
dGS ¼ 0:11 H 1 H2 þ 20 (10)
N can be calculated as
Seismic Design of Lc
Pipelines,
Fig. 3 Kennedy
et al. (1977) model for one
side of the fault Tu d f sin b/2
Pu
Rc
Amax small compared to the axial strains induced in

Cg ¼ (14)
CS 2 the ground by the traveling waves.
where Amax is the maximum ground acceleration

perpendicular to the direction of wave propaga- Pipelines Response to Faulting
tion and Cs is the propagation velocity of the
S-wave. This section presents the response of continuous
Wave propagation with particle motion paral- pipelines subject to fault offsets. PGD due to
lel to the pipeline (R-wave) would induce large faulting can be resolved into two components:
axial strain in the pipeline. On the other hand, longitudinal PGD (parallel to the pipeline) and
S-waves propagating parallel to the pipeline with transverse PGD (perpendicular to the pipeline)
particle motion perpendicular to the pipeline axis. In the case of “normal fault” type (see
direction would induce only bending strains. Fig. 1), the pipeline will be subjected to bending
For S-waves traveling in a horizontal plane axial tensile force, caused by the transverse and
and at an angle with respect to the pipeline, the longitudinal components, respectively. In this
S-wave particle motion (perpendicular to its case, tensile rupture would be the most probable
direction of propagation) would have one com- failure mechanism. In the “reverse fault”-type
ponent parallel to the pipeline inducing axial case, the pipeline will be subjected to bending
strain, eg, and another component perpendicular axial compressive force, caused again by the
to the pipeline inducing bending strain, eb, which transverse and longitudinal components, respec-
can be calculated as tively. In this case, buckling would be the most
likely failure mechanism. In the last case of
V max “strike slip fault,” the pipeline can be subjected
eg ¼ (15) to either tension or compression depending on the
2 Cs S
intersection angle between the pipeline and the
fault and pipe and the relative movement at the
and
pD V max fault.
eb ¼ (16) Kennedy et al. (1977) proposed a simplified
l Cs
method to analyze the tensile and bending behav-
where eb is the upper bound bending strain ior of pipelines due to fault movements. Figure 3
(maximum), Vmax is the peak ground velocity, shows the Kennedy et al. model for one side of
Cs is the apparent propagation velocity of the the fault.
S-wave, l is the wavelength, and D is the pipe According to Kennedy et al. (1977), the total
diameter. strain in the pipe (bending + tensile) is given by
The bending strains induced in a pipeline due
to traveling waves (S-waves, L-waves, or the DL D
e ¼ ea þ eb ¼ þ (17)
vertical component of R-waves) are generally L 2Rc
where DL is the total elongation of the pipeline, horizontal bearing capacity factors for sand and
L is the total length of the pipeline, D is the pipe clay after Hansen (1961).
diameter, and Rc is the radius of curvature of the The total elongation of the pipeline, DL, can be
curved portion. Rc can be estimated by estimated using the following equation:
spDt 2
Rc ¼ (18) df sin b
Pu DL ¼ df cos b þ (21)
3Lc
where s is the axial stress at fault crossing and Pu
is the peak lateral pipeline-soil interaction force where df is the average fault displacement, b is
per unit length which can be calculated using the the fault angle, and Lc is the horizontal projection
1984 ASCE guideline using the following rela- length of the laterally deformed pipeline (see
tions for sand and clay, respectively: Fig. 3).
For sand Lc can be approximately calculated using the
following simplified equation:
Pu ¼ g H N qh D (19)
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Lc ¼ Rc df sin b (22)
For clay
Pu ¼ cu N ch D (20) Pipelines Response to Longitudinal PGD
where g is the unit weight of the soil, H is the This section presents the response of continuous
embedment depth of the pipeline, and Nqh and pipelines subject to longitudinal permanent
Nch are the horizontal bearing capacity factors for ground deformations, PGD, where the soil move-
sand and clay, respectively. Figure 4 presents ment is in the same direction as the pipeline.
Seismic Design of Nqh Nch

Pipelines,
2 10 50 2 10
Fig. 4 Horizontal bearing
0
capacity factors for sand
and clay (Reproduced after
2
Hansen 1961)
4
Depth to Diameter Ratio H/D
8
j = 20° j = 0°
10
25°
12
30°
14
35°
16
40°
18
45°
20
Sand Clay

O’Rourke et al. (1995) proposed an inelastic b p x2 2 n bp x r
pipeline model to analyze the response of pipe- dð x Þ ¼ 1þ
E 2 þ r 1 þ r sy
lines to longitudinal PGD. Figure 5 shows the
(23)
considered model. To develop this model
O’Rourke et al. utilized an idealized block
where n and r are Ramberg-Osgood parameters
pattern in which a mass of soil having length
L moves down a slight incline. The soil displace- (given in Table 1 below), E is the modulus of
ment on either side of the PGD zone is zero, elasticity of steel, sg is the effective yield stress,
and bp is the pipe burial parameter, defined as the
while the soil displacement within the zone is a
constant value d. friction force per unit length tu divided by the
In O’Rourke et al. model, a block of soil pipe cross-sectional area A.
The pipe burial parameter bp can be obtained
between Points B and D moves to the right
pulling the pipe laterally with it with the soil from the following:
forces acting on the pipeline within the PGD For sand
zone to the right. On the other hand, the soil
between Points A and B and the soil between tan ’ g H
bp ¼ (24)
Points D and E resist the pipeline movement t
and these soil restraint forces are directed to the
left. The combined pipeline-soil interaction For clay
forces result in a region of pipe axial tension
between points A and B and a region of axial a cu
bp ¼ (25)
compression between points D and E. The condi- t
tions outlined in Fig. 5 correspond to the case
where the PGD, d, is comparatively large and where ’ is the angle of shear resistance, g is the
the length of the PGD zone, L, is comparatively effective unit weight of the soil, H is the embed-
short. In that case, the maximum pipe displacement depth of the pipeline, a is the adhesion
ment is less than the ground displacement and the factor for clay, cu is the undrained cohesion of
pipe strain is controlled by L. the clay (see Fig. 7 below), and t is the pipe wall
Figure 6 presents the other possibility of the thickness.
O’Rourke et al. model where the length of the
PGD zone is relatively large while the amount of Wrinkling of the Pipe Wall in Compression
PGD is comparatively short. Also in this case, Substituting a critical local buckling strain into
there is still axial pipe tension between points Eq. 22, one can obtain the critical length of PGD
A and B and axial compression between points zone Lcr. This can then be used to calculate the
D and E; however, the zone is long enough that critical ground movement dcr from Eq. 23. The S
the pipe displacement matches that of the ground critical strain in compression may be taken as
between Points C and D where the axial force and 0.175 t/R.
strain in the pipe are zero.
As it can be seen from Figs. 5 and 6, the axial
force in the pipeline in the segment AB is linearly Pipeline Response to Transverse PGD
proportional to the distance from Point
A. Accordingly, the pipelines strain and displace- This section presents the response of continuous
ment can be evaluated using the following pipelines subject to transverse permanent ground
Ramberg-Osgood model relations: deformations, PGD, where the soil movement is
perpendicular to the pipeline. When subjected to

transverse PGD, the pipeline will stretch and
bp x n bp x r bend as it attempts to conform to the transverse
eðxÞ ¼ 1þ (22)
E 1 þ r sy ground movement profile. In this case, the failure
Seismic Design of Pipelines, Fig. 5 O’Rourke et al. model to analyze the response of pipelines to longitudinal PGD
mechanism of the pipeline will be governed by In general, the response to transverse PGD is a
the relative magnitude of the axial tensile strain function of the magnitude of PGD, the width of
and the bending strain. If the tensile strain is the PGD zone, and the pattern of ground defor-
relatively small, the pipe wall may buckle in mation. Two types of transverse ground deforma-
compression due to excessive bending. Instead, tion patterns are discussed here: the spatially
if tensile strain is relatively large, the pipe may distributed transverse PGD pattern and the abrupt
rupture in tension due to the combined effects of transverse PGD pattern. Figure 8 shows sketch of
the tensile and bending stresses. the considered patterns.
Seismic Design of Pipelines, Fig. 6 O’Rourke et al. model to analyze the response of pipelines to longitudinal PGD
(PGD zone is relatively large, while the amount of PGD is comparatively short)
Spatially Distributed Transverse PGD pipeline is relatively flexible and its lateral dis-
O’Rourke (1989) proposed a simple model to placement is assumed to closely conform to the
analyze the response of pipelines to spatially soil outline. Accordingly, the pipeline strain is
distributed transverse PGD. In this model expected to be mainly due to the ground curva-
O’Rourke considered two types of response ture (i.e., displacement controlled). On the other
(wide and narrow width PGD zones) as shown hand, for the narrow width PGD case, the pipeline
in Fig. 9 below. In the wide width PGD zone, the is relatively stiff and the pipeline lateral
displacement is significantly smaller than that of – For the wide width PGD zone
the soil. Hence, the pipeline strain is anticipated
to be due to loading at the pipeline-soil interface p 2
d 2
(i.e., load controlled). ea ¼ (28)
2 W
The maximum bending strain, eb, in the pipe-
line is given by the following: – For the narrow width PGD zone
– For the wide width PGD zone The axial tension in this case is small and
neglected.
p2 dD where d is the magnitude of the PGD, D is the
eb ¼ (26)
W2 pipe diameter, W is the length of the PGD zone,
Pu is the maximum lateral force per unit length at
– For the narrow width PGD zone the pipeline-soil interface, E is the elastic modu-
lus of the pipeline material, and t is the pipe
pu W 2 wall’s thickness.
eb ¼ (27)
3pEtW 2
Abrupt Transverse PGD
And the average axial tensile strain, ea, in the Parker et al. (2008) proposed a simple model to
pipeline is estimated by the following: analyze the response of pipelines to transverse
PGD. Figure 10 presents the geometric and
Seismic Design of Pipelines, Table 1 Yield stress and force details of the model. In this model the
Ramberg-Osgood parameters for mild steel and X-grade
width of the abrupt transverse PGD is 2W1.
steel
Within this width the pipeline is subject to a
Grade X- X- X- X- X-
lateral force per unit length Pu1. This lateral
B 42 52 60 70 80
Yield 241 290 359 414 483 552
load is resisted by soil resistance forces Pu2 over
stress a distance W2 on each side of the abrupt trans-
(MPa) verse PGD zone (see Fig. 10). Therefore, from
n 10 15 9 10 5.5 16 horizontal equilibrium in the direction of pipe-
r 100 32 10 12 16.6 16 line, we get
Seismic Design of
Pipelines,
Fig. 7 Adhesion factors
(Reproduced after
Honegger and Nyman
2004)
Pu1 W 1 ¼ Pu2 W 2 (29) (Point B to Point F) can be estimated using the

following equation:
The tensile force in the pipeline is assumed to be a
constant value To within the PGD zone. Beyond 1 2
DL ¼ P W 3 ð1 þ Pu1 =Pu2 Þ =T o 2 (30)
the margins the pipeline axial tension decreases 3 u1 1
linearly at Points C and E to zero at Points A and
G (see Fig. 10). Thus, the axial pipe strain is calculated as
According to Parker et al. (2008), the total
elongation due to the pipeline deformation DL
ea ¼ (31)
L
and the bending strain in the pipeline can be

evaluated as
D Pu1
eb ¼ (32)
2 To
Pipelines in Liquefied Soil
It is common practice that pipelines are buried

at depths of 2 m or less from the ground
surface. Thus, the top of the susceptible
liquefiable soil layer is commonly located
below the bottom of the pipeline. However,
in some cases when the pipeline is buried at a
river bed in saturated sand, for example, the
Seismic Design of Pipelines, Fig. 8 Sketch of the con- soil surrounding the pipeline may liquefy dur-
sidered patterns ing strong seismic shaking event. In this case,
Seismic Design of
Pipelines,
Fig. 9 O’Rourke (1989)
model to analyze the
response of pipelines to
spatially distributed S
transverse PGD
Seismic Design of Pipelines, Fig. 10 Geometric and force details of the Parker et al. (2008) model
the pipeline will probably deform laterally p D2

following the flow of the liquefied soil down- Puplift ¼ ðgsoil gcontents Þ
4
ward a mild slope or move upward due to
p D t gpipe (33)
buoyancy, particularly when something
restrains the pipeline at one point or a com-
where D is the pipe diameter, gsoil is the unit
pressive load acts on the pipeline.
weight of the liquefied soil, gcontents is the unit
weight of the contents inside the pipe (water, oil,
Horizontal Movement
gas, etc.), gpipe is the unit weight of the pipe
When a pipeline is surrounded by liquefied soil,
material, and t is the pipe wall’s thickness.
the pipeline may move laterally due to the flow of
The maximum uplift displacement and/or the
liquefied soil downslope. The response of a bur-
spacing for pipeline restraints is given by (see
ied pipe surrounded by liquefied soil subject to
Fig. 11 below)
spatially distributed transverse PGD can be ana-
lyzed using the O’Rourke (1989) method 16Puplift W s 4
presented earlier. A dmax 3 þ 16 I dmax (34)
E p5
Vertical Movement and the maximum strain in the pipeline is then
When a pipeline is surrounded by liquefied soil, given by
the pipeline may uplift due to the buoyancy and
moves upward. Hou et al. (1990) proposed ana- p2 dmax D p2 dmax 2
emax ¼ þ (35)
lytical method to analyze the response of pipe- Ws2 4 Ws2
lines subjected to vertical movements. According
to Hou et al. (1990), the uplifting force per unit where A is the cross-section area, I is the moment
length, Puplift, acting on the pipeline within the of inertia, Ws is the spacing of the restraints, and
liquefied zone is given by E is the elastic modulus of the pipe’s material.
Seismic Design of Pipelines, Fig. 11 Profile of pipeline crossing liquefied zone
Seismic Design of
Pipelines,
Fig. 12 External forces on
the buried pipeline by the
adjacent soil during seismic
shaking
Seismic Design Guidelines and Pipelines in Competent Non-liquefied Soil

Pipeline-Soil Interaction The 1984 ASCE Guidelines suggest for the pur-
pose of analysis idealized elastoplastic models
The pipeline-soil interaction effect exerts exter- for the force-deformation relationship at the
nal forces on the buried pipeline by the adjacent pipeline-soil interface (see Fig. 13 below).
soil when seismic motions are provided. The type
of soil around the pipeline plays a significant role Longitudinal Movement
in its seismic behavior. In general, the soil dis- In this case the relative movement is parallel to S
placement will produce friction like forces at the the pipeline which results in axial forces at the
pipeline-soil interface (see Fig. 12 below). The pipeline-soil interface. The1984 ASCE Guide-
overall seismic performance of buried pipeline is lines provide relations for both cohesionless and
strongly related to the force-deformation rela- cohesive soils. For cohesionless materials, the
tionship at the pipeline-soil interface (the p-y longitudinal resistance is due to the friction in
curves). For cohesionless soils, the probability the longitudinal direction at the pipeline-soil
of liquefaction becomes higher in loose mate- interface. The normal pressure which leads to
rials. In cohesive soils, softer soils will undergo the axial friction is the overburden and the lateral
greater differential settlement due to consolida- soil pressures. In the 1984 ASCE Guidelines, the
tion and higher amplification effects and accord- normal pressure is taken as the average of the
ingly greater interaction forces. vertical and at rest lateral soil pressures acting
Seismic Design of Pipelines, Fig. 13 Idealized elastoplastic models for the force-deformation relationship at the
pipeline-soil interface (Reproduced after ASCE 1984)
on the pipeline. (The 1984 ASCE Guidelines Pu ¼ g H D N qh (38)

assumes Xu ffi 2:5 5:0 mm).

and the maximum elastic relative displacement in
1 þ ko horizontal transverse direction is
tu ¼ p D g H tan k’ (36)
2 8
< ð0:07 0:10ÞðH þ D=2Þ for loose sand
>
For cohesive materials, the longitudinal resis- yu ¼ ð0:03 0:05ÞðH þ D=2Þ for medium sand
tance is proportional to the adhesion at the >
:
ð0:02 0:03ÞðH þ D=2Þ for dense sand
pipeline-soil interface:
(39)
tu ¼ p D a cu (37) For cohesive materials, the maximum soil resis-

tance in horizontal transverse direction may be
where D is the pipe diameter, g is the effective evaluated using the following equation:
unit weight of the soil, H is the depth of the
pipeline, ’ is the angle of shear resistance, ko is Pu ¼ cu N ch D (40)
the coefficient of lateral soil pressure at rest, k is a
friction factor, a is the adhesion factor (given in The maximum elastic relative displacement in
Fig. 7), and cu is the undrained shear strength of horizontal transverse direction is
the soil.
yu ¼ ð0:03 0:05ÞðH þ D=2Þ (41)
Transverse-Horizontal Movement
In this case the relative movement is perpendic- where g is the unit weight of the soil, H is the
ular to the pipeline which results in transverse- embedment depth of the pipeline, and Nqh and
horizontal forces at the pipeline-soil interface. Nch are the horizontal bearing capacity factors for
The 1984 ASCE Guidelines provide relations sand and clay, respectively (Fig. 4).
for both cohesionless and cohesive soils. For
cohesionless materials, the maximum soil resis- Pipelines in Liquefied Soil
tance in horizontal transverse direction may be The response of continuous pipelines buried in
calculated using the following equation: liquefied soil layer is very sensitive to the
stiffness of the soil (Suzuki et al. 1988; Miyajima Cross-References

and Kitaura 1989). Based on several experimen-
tal results, it is recommended that the stiffness ▶ Earthquake Magnitude Estimation
of liquefied soil ranges from 1/100 to 3/100 ▶ Earthquake Mechanisms and Tectonics
of that for non-liquefied soil (Yoshida and ▶ Earthquake Response Spectra and Design
Uematsu 1978; Matsumoto et al. 1987; Yasuda Spectra
et al. 1987; Tanabe 1988). Accordingly, ▶ Liquefaction: Performance of Building
the reduced stiffness will be used at the Foundation Systems
pipeline-soil interface. Analysis is then ▶ Seismic Vulnerability Assessment: Lifelines
performed using the same procedures as in the ▶ Soil-Structure Interaction
competent non-liquefied soil case (on the
conservative side).
References
American Society of Civil Engineers (ASCE)

Summary (1984) Guidelines for the seismic design of oil and
gas pipeline systems. Committee on Gas and Liquid
Fuel Lifelines, ASCE
This chapter focused on the seismic analysis and
Chen WW, Shih BJ, Wu CW, Chen YC (2000) Natural
design of pipelines. In seismic events, buried gas pipeline system damages in the Ci Ci earthquake
pipelines can be damaged mainly by either the (The City of Nantou). In: Proceedings of the 6th inter-
permanent ground deformations (PGD) or by the national conference on seismic zonation, Palm
Springs, California.
transient seismic wave propagation. Hence, dif-
Hansen JB (1961) The ultimate resistance of rigid piles
ferent analysis methods for pipelines subjected to against transversal forces. Bulletin 12, Danish Geo-
permanent ground deformation (PGD) or tran- technical Institute, Copenhagen
sient seismic wave propagation hazards were Honegger, D.G. and Nyman, D. (2004). PRCI guidelines
for the seismic design and assessment of natural gas
presented and discussed in details. In addition,
and liquid hydrocarbon pipelines. Pipeline Research
the response of continuous pipelines subject to Council International, Published by Technical Tool-
fault offsets was discussed. Also, several boxes, Houston, Texas
methods were introduced to predict the response Hou Z, Cai J, Liu X (1990) Response calculation of oil
pipeline subjected to permanent ground movement
of pipelines to either longitudinal or
induced by soil liquefaction. In: Proceedings of the
transverse PGD. China-Japan symposium on lifeline earthquake engi-
Earthquake shaking may trigger the lique- neering, Beijing, pp 107–114
faction of saturated loose cohesionless soils. Jibson RW, Keefer DK (1993) Analysis of the
seismic origin of landslides: examples from the
The liquefaction process increases the pore
new Madrid seismic zone. Geol Soc Am Bull
water pressure in the ground to a level 105:521–536 S
where the effective stress approaches zero at Kennedy RP, Chow AW, Williamson RA (1977) Fault
which point the soil loses entirely its shear movement effects on buried oil pipeline. J Transp
Eng Div ASCE 103(TE5):617–633
strength. A separate section in this chapter
Lau DL, Tang A, Pierre J-R (1995) Performance of life-
presented and discussed methods of analyzing lines during the 1994 Northridge earthquake. Can J Civ
and designing pipelines buried in liquefiable Eng 22:438–451
soils. Matsumoto H, Sasaki Y, Kondo M (1987) Coefficient of
subgrade reaction on pile in liquefied ground. In: Pro-
The overall seismic performance of buried
ceedings of the second national conference on soil
pipelines is strongly related to the pipeline-soil mechanics and foundation engineering, Tokyo,
interaction. Consequently, this chapter dedicated Japan, pp 827–828 (in Japanese)
a section to provide guidelines for the seismic Miyajima M, Kitaura M (1989) Effects of liquefaction-
induced ground movement on pipeline. In: Proceed-
design of pipelines considering the pipeline-soil
ings of the Second U.S.-Japan workshop on liquefac-
interaction effects as it plays a significant role in tion, large ground deformation and their effects on
its seismic behavior. lifelines, Buffalo, Technical report NCEER-89-0032,
2796 Seismic Design of Tunnels
Multidisciplinary Center for Earthquake Engineering earthquake engineering symposium, Tokyo, Japan,
Research, Buffalo, pp 386–400 pp 657–663 (in Japanese)
Newmark NM (1965) Effects of earthquakes on dams and Youd TL, Hansen C, Bartlett S (2002) Revised multilinear
embankments. Geotechnique 15(2):139–160 regression equations for prediction of lateral spread
Newmark NM (1967) Problems in wave propagation in displacement. J Geotech Geoenviron Eng ASCE
soil and rocks. In: Proceedings of the international 128(12):1007–1017
symposium on wave propagation and dynamic prop- Zhang G, Robertson PK, Brachman RWI (2004)
erties of earth materials, University of New Mexico Estimating liquefaction-induced lateral displacements
Press, pp 7–26 using the standard penetration test or cone
O’Rourke MJ (1989) Approximate analysis procedures penetration test. J Geotech Geoenviron Eng ASCE
for permanent ground deformation effects on buried 130:861–871
pipelines. In: Proceedings of the second U.S.-Japan
workshop on liquefaction, large ground deformation
and their effects on lifelines, Buffalo. Technical report
NCEER-89-0032, Multidisciplinary Center for Earth-
quake Engineering Research, Buffalo, pp 336–347
O’Rourke TD, Palmer MC (1996) Earthquake perfor-
Seismic Design of Tunnels
mance of gas transmission pipelines. Earthq Spectra
12(3):493–527 Youssef M. A. Hashash1 and
O’Rourke TD, Grigoriu MD, Khater MM (1985) Seismic Maria Ines Romero-Arduz2
response of buried pipes. Pressure vessel and piping 1
technology – a decade of progress, ASME, pp 281–323
O’Rourke MJ, Liu XJ, Flores-Berrones R (1995) Steel Engineering, 2230c Newmark Civil Engineering
pipe wrinkling due to longitudinal permanent ground Laboratory, University of Illinois at
deformation. J Transp Eng 121(5):443–451 Urbana-Champaign, Urbana, IL, USA
Parker E, Moore R, Evans T, Usher N (2008) Evaluation 2
of landslide impact on deepwater submarine pipelines.
OTC 19459, Offshore Technology Conference, Engineering, University of Illinois at
Houston Urbana-Champaign, Urbana, IL, USA
Scawthorn C, Yanev PI (1995) Preliminary report 17 Jan-
uary 1995, Hyogoken Nambu, Japanese earthquake.
Eng Struct 17(3):14657
Suzuki H (1988) Damage to buried pipes caused by large Synonyms
ground displacement. In: Proceedings of the first
Japan-U.S. workshop on liquefaction, large ground Cut-and-cover structures; Earthquakes; Free-
deformation and their effects on lifeline facilities,
field deformations; Immersed-tube tunnels;
Tokyo, pp 127–132
Takada S, Tanabe K (1988) Estimation of earthquake Nonlinear response; Racking; Soil-structure
induced settlements for lifeline engineering. In: Pro- interaction; Tunnels
ceedings of the ninth world conference earthquake
engineering, vol VII, pp 109–114
Tanabe K (1988) Fundamental study on seismic assess-
ment and design of buried pipelines subjected to Introduction
ground failure during earthquake. Doctoral disserta-
tion, Kobe University (in Japanese) The complete enclosure in soil or rock makes
Wells DL, Coppersmith KJ (1994) New empirical rela-
their seismic behavior different than that of
tionships among magnitude, rupture length, rupture
width, rupture area, and surface displacement. Bull aboveground structures or superstructures.
Seismol Soc Am 84(4):974–1002 Underground structure seismic response is
Wilson RC, Keefer DK (1983) Dynamic analysis of a constrained by seismic response of the surround-
slope failure from the 6 August 1979 Coyote Lake,
ing soil and cannot experience free vibrations as
California earthquake. Bull Seismol Soc Am
73(3):863–877 is the case for aboveground structures. This sec-
Yasuda S, Saito K, Suzuki N (1987) Soil spring constant tion focuses on the seismic analysis and design of
on pipe in liquefied ground. Proceedings of the nine- large linear underground structures commonly
teenth JSCE conference on earthquake engineering,
used for metro structures, highway tunnels, and
Tokyo, Japan, pp 189–192 (in Japanese)
Yoshida T, Uematsu M (1978) Dynamic behavior of a pile large water/sewage transportation ducts in urban
in liquefaction sand. Proceedings of the fifth Japan areas and can be grouped into three broad
Seismic Design of Tunnels 2797
Seismic Design of Tunnels, Fig. 1 Typical cross sections of underground structures
categories: (1) bored or mined tunnels, (2) cut- in recent large magnitude earthquakes (e.g.,
and-cover tunnels, and (3) immersed tunnels Tohoku, Japan, 2011; Maule, Chile, 2010),
(Fig. 1). underground structures suffer appreciably less
The section starts with a selected review of damage than surface structures. However, dam-
performance of underground structures during age or failure of a limited section of an under-
seismic events. This is followed by presentation ground structure can be disruptive to post-
of a performance-based framework for design earthquake recovery or operation in densely pop-
and analysis of underground structures consider- ulated urban areas as such damage can interrupt
ing both permanent and transient deformations. the function of an entire system whether it is part
A number of seismic design issues are then of a mass transit or vehicular transportation net-
discussed including vertical ground shaking and work or large water or sewage transportation
response, interaction of temporary and perma- tunnels. S
nent structures, impact of superstructure and Damage is related to a number of parameters
adjacent structures, transitions and tunnel joints, including ground motion intensity, ground con-
seismic retrofit of existing facilities, design conditions, and structural support system. Shallow
siderations for structural support members, and tunnels tend to be more vulnerable to earthquake
precast tunnel lining. shaking than deep tunnels, and those constructed
in soil can undergo more deformation than those
constructed in competent rock. Circular bored
Performance of Underground Facilities tunnels are less susceptible to earthquake damage
During Seismic Events than cut-and-cover tunnels. Shaking damage can
be reduced by stabilizing the ground around the
Based on several studies that documented earth- tunnel and by improving the contact between the
quake damage to underground facilities in the lining and the surrounding soil using grouting.
past and the behavior of underground facilities Stiffening the lining without stabilizing the
Seismic Design of
Tunnels, Fig. 2 Street
view of Dakai subway
station collapse (Iida
et al. 1996)
surrounding poor ground may only result in Underground Structures in Japan

excessive seismic forces in the lining. Damage The Dakai subway collapse (Fig. 2) during the
at tunnel portals may be caused by slope instabil- 1995 Hyogoken-Nambu earthquake in Kobe,
ity (Hashash et al. 2001). Japan, was the first collapse of an urban under-
ground structure due to earthquakes shaking. The
Underground Structures in the United States collapse experienced by the center concrete col-
The Bay Area Rapid Transit system (BART) in umns (Fig. 3) was due to lack of shear reinforce-
San Francisco, California, sustained the 1989 ment, leading to a collapse of the ceiling slab and
Loma Prieta earthquake without damage and settlement of the soil cover. The 1962 station
was operational after the earthquake. It consists design did not include specific seismic provi-
of underground stations and tunnels embedded in sions. However, in the 2011 Tohoku earthquake,
soft bay mud deposits connected to Oakland via the underground subways in Sendai experienced
the transbay-immersed-tube tunnel. It was one of strong shaking with no reports of damage. Only a
the first underground facilities designed with water distribution plant in Kashima City and a
seismic considerations: special seismic joints wastewater treatment plant in Itako City were
were designed to accommodate differential observed to have been damaged by liquefaction.
movements. Limited displacements were mea- The damages include uplift of buried tanks, off-
sured at that joint. sets in underground tunnels, damage to support
During the same earthquake, the Alameda utilities, and damage to major trunk lines on and
Tubes, a pair of immersed-tube tunnels that con- off the site (Ashford et al. 2011).
nect Alameda Island to Oakland in California,
experienced structural cracking on the ventilation Underground Structures in Taiwan
buildings and limited water leakage due to lique- The 1999 Chi-Chi earthquake did not cause vis-
faction of loose deposits above the tubes. ible signs of damage in several highway tunnels
The 1994 Northridge earthquake caused no located in central Taiwan. The main damage
damage to concrete lining of bored tunnels of occurred at tunnel portals because of slope insta-
the metro system in Los Angeles, California bility. No damage was reported in the Taipei
(Hashash et al. 2001). subway, located 100 km from the ruptured fault
gravels, the Santiago Metro tunnels and under-

ground stations, and highway tunnels on the
southbound of Route 5 appeared to be
undamaged by the 2010 Maule earthquake in
Chile, as shown in Fig. 6. In the northbound of
Route 5, “La Calavera” tunnel near Calera has a
rock block dislodged, but the tunnel was old and
had problems before the earthquake (Elnashai
et al. 2010).
In summary, well-engineered underground
structures performed well even under strong
shaking in recent earthquakes in different parts
of the world. However, underground structures
are vulnerable to permanent ground displace-
ments such as liquefaction, slope stability, and
fault displacement. There is also vulnerability to
transient ground motions when insufficient struc-
tural detailing is provided, the underground struc-
ture is constructed in loose ground, masonry
lining is used, or near field effects are present.
Performance-Based Seismic Evaluation

Seismic Design of Tunnels, Fig. 3 Dakai subway sta-
Framework
tion collapse (Iida et al. 1996)
Underground structures under earthquake effects
zone (Fig. 4). However, some tunnels in moun- can undergo permanent deformations and/or tran-
tainous areas were severely damage due to slope sient deformations. Factors influencing these
failure (Fig. 5). effects include shape, dimensions, and depth of
the structure, properties of the surrounding soil or
Underground Structures in Turkey rock, properties of the structure, and severity of
The August 17, 1999, Kocaeli earthquake had ground shaking (Hashash et al. 2001). Table 1
minimal impact on the Bolu twin tunnels, a 1.5- summarizes a performance-based framework for
billion-dollar project under construction at that seismic design and analysis of underground
time. It had an excavated section of 15 m tall by structures. The framework consists of three S
16 m wide and crossed several minor faults par- main steps: definition of seismic environment,
allel to the North Anatolian Fault. After the earth- evaluation of ground response to shaking, and
quake, continuous monitoring showed no assessment of structure response due to seismic
movement due to the earthquake. The November shaking.
12, 1999, earthquake caused collapse of both
tunnels 300 m from its eastern portal, in a clay Step 1: Definition of Seismic Environment
gauge material in the unfinished section (Hashash Seismic analysis of underground structures starts
et al. 2001). with site-specific definition of its seismic envi-
ronment. A detailed field and laboratory investi-
Underground Structures in Chile gation program is necessary; the field
Cut-and-cover highway box structures with three investigation program should include definition
lanes of traffic in one direction and approxi- of the site stratigraphy and direct measurements
mately 1 km long constructed in relatively stiff of shear wave velocity profiles and cone
Seismic Design of
Tunnels, Fig. 4 Slope
failure at tunnel portal,
Chi-Chi earthquake, central
Taiwan (Hashash
et al. 2001)
penetration resistance of soft soils as well as based on analysis findings to answer the question
assessment of potential geo-hazards including of what is feasible and at what cost as illustrated in
slope instability, fault displacement, and lateral Fig. 7.
spreading. Appropriate static and cyclic labora- Often a two-level criterion is adopted: operat-
tory tests for major soil units are also required. ing design earthquake (ODE) and maximum
Site-specific probabilistic and/or deterministic design earthquake (MDE). Those are defined
seismic hazard analyses, as well as hazard ana- using response spectra developed in the seismic
lyses using conditional (mean) spectra, are hazard analysis. A suite of three component
needed to define seismic hazard levels for perma- motions is needed for each of the design earth-
nent condition of the structure (operational and quake levels for site response analysis and soil-
maximum levels). Increasingly seismic hazard, structure interaction modeling. It is preferred to
using a shorter return period, is being considered use recorded motions instead of synthetic
for temporary conditions during construction. motions to spectrally match the target spectra.
Seismic performance criteria selection is a cru- Ground motion spatial incoherence must be
cial aspect in the design of an underground struc- taken into account for long structures including
ture. Performance objectives include explicit (1) wave passage, (2) extended source effects,
target performance of the structure and system (3) ray-path effects, and (4) local site effects.
performance. Performance objectives are not One-dimensional equivalent linear and
purely technical requirements and should include nonlinear site response analyses are conducted
owner and user requirements, policy consider- to assess how the ground motion is affected by
ations, and life-cycle costs. It is an iterative process the soil column. One-dimensional site response
Seismic Design of Tunnels, Fig. 5 Chi-Shue tunnel before and after Chi-Chi earthquake (Wang et al. 2001)
Seismic Design of Tunnels, Fig. 6 Left: Highway box structures in Santiago, Right: Highway tunnel in Route 5 South
Seismic Design of Tunnels, Table 1 Performance-based framework for design and analysis of underground
structures
Step 2: Evaluation of Ground Response to Step 3: Assessment of structure behavior due to

Step 1: Definition of Seismic Environment
Shaking seismic shaking
Permanent deformations Seismic Design Loading Criteria

• Liquefaction • Loading Criteria for MDE
Field and laboratory investigations
• Slope Instability • Loading Criteria for ODE
• Fault Displacement
Seismic hazard analysis Transient deformations Underground Structure Response to Ground

• Longitudinal Extension/Compression Deformations
• Deterministic Seismic Hazard Analysis (DSHA)
• Probabilistic Seismic Hazard Analysis (PSHA) • Longitudinal Bending • Free-Field Deformation Approach
• Racking/Ovaling • Soil-structure Interaction Approach
Seismic performance criteria

• Maximum Design Earthquake (MDE)
Additional Seismic Design Issues
• Operating Design Earthquake (ODE)
Seismic input motions

• Acceleration, Velocity, Displacement Amplitudes
• Target Response Spectra, Motion Time History
• Spatial Incoherence of Ground Motion
Site response analysis
analyses are used in the analysis of underground Permanent deformations or ground failure
structures to: includes liquefaction, slope instability, and fault
displacement. Liquefaction, prevalent in loose
(a) Obtain free-field racking deformations sand and fill deposits, can result in generation of
(differential sideways movements between sand boils, loss of shear strength, lateral spread-
the top and bottom elevations of rectangular ing, and slope failure. Tunnels in liquefiable
structures) along the underground structure deposits can experience increased lateral pres-
height which can be used in pseudo-static sures, loss of lateral passive resistance, flotation
soil-structure interaction. or sinking, lateral displacements if lateral spread-
(b) Obtain input motions for dynamic soil ing happens, permanent settlement, and compres-
for dynamic soil-structure interaction sion/tension failure after soil consolidation.
analysis. A landslide intercepting a tunnel can result in
(c) Obtain strain-compatible soil properties for concentrated shearing displacements and col-
use in pseudo-static and dynamic soil- lapse of a cross section. The potential for these
structure interaction. failures is greatest when a pre-existing landslide
(d) Assess potential liquefaction and ground intersects the tunnel, in shallower parts of tunnel,
failure. and at tunnel portals. An underground structure
may have to pass across an active fault zone; in
Step 2: Evaluation of Ground Response to these situations the tunnel must tolerate the
Shaking expected displacements. The design for perma-
Evaluation of ground response to shaking can be nent deformations is discussed in section “Design
divided into permanent deformations or ground for Permanent Deformations.”
failure and transient deformations or ground Transient deformations can be quite complex
shaking. due to interaction of seismic waves with surficial
OWNER
Select Preliminary
DESIGNER Performance Objectives
PEER REVIEWERS
BUILING OFFICIAL
Develop Preliminary Design
Assess
Performance Peer
Capability Reviewers
Revise Design
and/or
Performance Objective
BUILDING OFFICIAL
&
PEER REVIEWERS
No Does Yes
Performance DONE
Meet Objectives?
Seismic Design of Tunnels, Fig. 7 Performance-based design procedure after Hamburger and Hooper (2011)
deposits. Underground structures undergo three interaction of temporary and permanent struc-
primary modes of deformation during seismic tures; permanent changes in state of stress of
shaking: compression-extension, longitudinal soil; impact of superstructure and adjacent struc-
bending, and ovaling/racking. The design for tures; tunnel joints such as portals, stations, and
transient deformations is discussed in section tunnel segment; seismic retrofit of existing facil-
“Performance Evaluation Under Transient ities; design considerations for structural support
Ground Deformations.” members; precast tunnel lining; and seismic S
design of buried reservoirs.
Step 3: Assessment of Structure Behavior Due
to Seismic Shaking
The evaluation of structure behavior will be pri- Design for Permanent Deformations
marily a deformation controlled soil-structure
interaction problem. Pseudo-static and dynamic Designing underground structures for permanent
soil-structure interaction approaches have been deformations may not be viable, but ground sta-
used in the evaluation of structure response and bilization techniques can help prevent large
are discussion in section “Performance Evalua- deformations. Some solutions include ground
tion Under Transient Ground Deformations.” improvement, drainage, soil reinforcement,
Section “Additional Seismic Performance grouting, earth retaining systems, or even remov-
Issues” discusses additional seismic design issues ing problematic soils or relocating the tunnel
such as vertical ground shaking and response; alignment.
Underground Structures Crossing Active (b) If large displacements are concentrated in a

Faults narrow zone, retrofit will consist of enlarging
In the Century Area Tunneling Safety and Fault the tunnel section across and beyond the dis-
Investigations TAP report for the Los Angeles placement zone. The length over which the
Metro (Cording et al. 2011), some recommenda- enlargement is made is a function of fault
tions are provided for consideration when an displacement and permissible curvature of
underground tunnel segment crosses an active the road or track; the longer the enlarged
fault: tunnel, the smaller the post-earthquake cur-
vature (Power et al. 1996). This solution has
– The segments must be designed to accommo- been implemented in the San Francisco
date fault displacement without collapse and BART system and Los Angeles Metro rapid
with the capability of being repaired. transit tunnel system. Concrete-encased steel
– The alignments should be selected so that ribs provide sufficient ductility to accommo-
the tunnel crosses at a relatively sharp angle date distortions with little strength degrada-
to the fault zone to minimize the length tion. Under axial displacements, even though
of tunnel that must accommodate fault compression is more damaging to the tunnel
displacements. lining than extension, both will result in
– Methods employed to allow fault displace- unacceptable water inflow. A solution for
ment on a tunnel lining include excavating to water tightness is flexible couplings (Wang
a larger section at the crossing to facilitate 1993), used for the Southwest Ocean Outfall
realignment, providing the tunnel with a in San Francisco. Cellular concrete may also
strong but flexible lining like ductile steel seg- be used within the enlarged tunnel, because it
ments or articulated joints, and placing crush- has a low yield strength that can minimize the
able backpacking material around the loads on the tunnel liner while also providing
structural lining. adequate resistance for normal soil pressures
– As there is no precedent of placing metro and other seismic loads.
underground stations on active faults, design-
ing for it will be extremely difficult and cost- Estimating fault displacement is a key issue to
prohibitive. If possible, it is advisable to avoid design tunnels crossing active faults. One option
this scenario. to estimate fault displacement is using empirical
relationships that express expected displace-
The design approach for tunnels crossing ments in terms of some source parameter. Deter-
active faults will change depending on the dis- ministic and probabilistic fault displacement
placement magnitude and the tunnel width over hazard analyses can be used to assess fault dis-
which the displacement is distributed: placement hazard where a displacement attenua-
tion function is used in a probabilistic seismic
(a) If fault displacements are small and/or dis- hazard analysis (Coppersmith and Youngs 2000;
tributed over a relatively wide zone, provid- Youngs et al. 2003).
ing articulation of the tunnel lining through
ductile joints is a possible solution. The Flotation in Liquefiable Deposits
closer the joint spacing, the better the tunnel Liquefaction evaluation is discussed elsewhere.
performance will be; this is more viable in If liquefaction is limited to soil layers above the
soft soils where displacements can be effec- underground structure, then it is unlikely to influ-
tively redistributed over the tunnel lining. ence the racking deformations of the structure.
The tunnel can then deform in an S-shape However, if the structure is partially or entirely
through the fault zone without rupture. It is embedded in liquefiable soil, additional evalua-
always necessary to keep the tunnel water- tions are required. Underground structures may
tight when using joints. experience flotation in liquefiable deposits.
Seismic Design of Tunnels, Fig. 8 Isolation principle, use of cutoff walls to prevent tunnel uplift (Schmidt and
Hashash 1999)
As shown in Fig. 8a, when the tunnel experiences Performance Evaluation Under
uplift due to flotation, the liquefied soil moves Transient Ground Deformations
underneath the displaced tunnel and lifts it further
up (Schmidt and Hashash 1999). Uplift can be If permanent deformations are not expected, then
prevented through isolation using cutoff walls, the underground structure must be designed for
such as sheet pile walls; stone columns transient ground deformations. In this case, the
(Fig. 8b); or jet grout columns (Fig. 8c). Sheet underground structure response is controlled by
piles with drainage capability can also reduce the ground deformation and the peak ground
excess pore water pressure. The rise in excess velocity.
pore water pressure is prevented at the bottom The focus of underground structure seismic
of the tunnel and in the soil underneath with these design is on free-field deformations of the ground S
barrier walls. With longer barrier walls and a and their interaction with the structure, since the
wider structure, uplift is more difficult. After the inertia of the surrounding soil is large relative to
liquefaction potential is mitigated, flexible joints the inertia of the structure. Figure 9 shows response
can be used to allow for differential displace- of underground structures to seismic motions:
ments at tunnel connection joints. axial compression and extension, longitudinal
bending, and ovaling/racking. Axial deformations
Slope Instability and Lateral Spreading are due to seismic waves producing motions par-
Stabilizing the soil or changing the alignment is allel to the tunnel axis, bending is due to seismic
often the most practical way to mitigate slope waves producing particle motion perpendicular to
instability and lateral spreading. It is challenging the longitudinal axis, and ovaling/racking is due to
to design an underground structure to resist or take shear waves propagating normally to the tunnel
these movements unless the hazard is localized axis. Design considerations for axial and bending
and the movement is small (Power et al. 1996). deformations are generally in the direction along
a b c
d e f
Seismic Design of Tunnels, Fig. 9 Deformation modes of tunnels due to seismic waves
the tunnel axis and in the transverse direction for dynamic soil-structure interaction analyses
ovaling/racking. within a reasonable amount of time. These types
There are four main approaches to tackle the of analyses allow problems with complicated
design for transient deformations: free-field tunnel geometry and ground conditions to be
deformation methods, pseudo-static soil- solved efficiently. However, the selection of
structure interaction analyses, dynamic soil- parameters for a complex problem requires
structure interaction finite-element analyses, and expertise; therefore, it is important to always
dynamic earth pressure methods. The merits and verify the computer model solution with simpler
drawbacks of these methods are summarized in pseudo-static or closed-form solutions.
Table 2. The presence of a rectangular frame structure
Free-field deformation methods assume that in the ground will induce dynamic earth pressures
the underground structure deformations are iden- acting upon the structure. Complex shear and
tical to those of the surrounding ground. They do normal stress distributions along the exterior sur-
not take into account soil-underground structure faces of the structures are expected, but quantify-
interaction and are most appropriate when the ing those distributions require rigorous dynamic
structure (racking) stiffness is equivalent to that soil-structure interaction, since they heavily
of the surrounding ground. depend on how the interface is modeled.
Pseudo-static soil-structure interaction models In the past, the Mononobe-Okabe method was
account for the kinematic interaction between the used to calculate the seismic-induced dynamic
soil and the underground structure neglecting earth pressures on underground structures. The
inertial interaction. They are often used for prac- method assumes the earthquake load is caused by
tical design purposes when the structure is not too inertial forces of the surrounding soil and calcu-
complex (NCHRP 611, Anderson et al. (2008)). lates the load using soil properties and a deter-
Nowadays, the ease of access to high perfor- mined seismic coefficient. This method is not
mance computers makes it possible to perform applicable in the case of underground structures,
Seismic Design of Tunnels, Table 2 Comparison of seismic design approaches

Approaches Advantages Disadvantages Applicability
Free-field 1. Comparatively easy to formulate, 1. Nonconservative for tunnel For tunnel
deformation many 1D wave propagation structure more flexible than structures with
methods programs available ground equal stiffness to
2. Conservative for tunnel ground
structure stiffer than ground
3. Overly conservative for
tunnel structures significantly
stiffer than ground
4. Less precision with highly
variable ground conditions
Pseudo-static 1. Good approximation of soil- 1. Ignores inertial effects Most conditions
soil-structure structure interaction 2. Less precision with highly except for variable
interaction 2. Comparatively easy to formulate variable ground soil profile, shallow
methods 3. Reasonable accuracy in 3. Shear displacement not structures
determining structure response transmitted uniformly to
4. Computationally efficient shallow box structures
5. Sensitivity analysis can be easily
performed
Dynamic soil- 1. Best representation of soil- 1. Computationally demanding All conditions
structure structure system 2. Uncertainty of design seismic
interaction finite- 2. Best accuracy in determining input parameters may be several
element analysis structure response times the uncertainty of the
3. Capable of solving problems with analysis
complicated tunnel geometry and
ground conditions (significant
variations in soil stiffness)
Dynamic earth 1. Serve as additional safety 1. Lack of rigorous theoretical None
pressure methods measures against seismic loading basis
2. Resulting in excessive
deformations for tunnels with
significant burial
3. Use limited to certain types of
ground properties
since they will move with the ground and will not estimate of the underground structure deforma-
form an active wedge. tion. Imposing the free-field deformations
When designing underground structures for directly on the underground structure can under-
transient deformations, sufficient ductility is estimate or overestimate the structure S
needed to absorb imposed deformations without deformations.
losing the capacity to carry static loads. Care
should be exercised in not increasing the Closed-Form Elastic Solutions
stiffness of the structure as this tends to attract Initial estimates of strains and deformations in a
additional loads thus increasing the demand on tunnel can be developed using simplified closed-
the structure. form solutions. There are many assumptions
within the formulation of these methods:
Free-Field Deformation Approach
Free-field deformations are the deformations (a) The seismic wave field is considered as a
caused by seismic waves on a given soil profile plane of wave with the same amplitude at
in the absence of structures or excavations. The all locations along the tunnel, differing only
interaction between the soil and the underground in their arrival time. Wave scattering and 3D
structure is neglected, but provides a first-order wave propagation are neglected. The results
Seismic Design of
Tunnels, Fig. 10 Seismic
waves causing longitudinal
axial and bending strains
(Power et al. 1996)
of these analyses should be interpreted with caused predominately by vertically propagating

care (Power et al. 1996). shear waves (Wang 1993). Ground shear distor-
(b) Harmonic wave propagating at a given angle tions can be defined assuming a non-perforated
of incidence in a homogeneous, isotropic, ground or a perforated ground. As shown in
elastic medium. The critical incidence angle Fig. 11, both cases ignore the tunnel lining (soil-
resulting in the maximum strain is typically structure interaction), where the maximum dia-
used (Newmark 1967). However, the strain metric strain is in terms of the maximum free-
order of magnitude estimated by this method field shear strain (gmax) and the Poisson ratio (vm).
is useful as initial design tool and design The first can be used to approximate the behavior
verification method. of a tunnel lining whose stiffness is equal to the
(c) St John and Zahrah (1987) developed free- medium it replaces. The second can be used to
field solutions for axial and curvature strains approximate the behavior of a tunnel lining
due to compression, shear, and Rayleigh whose stiffness can be neglected in comparison
waves. Figure 10 shows the seismic waves with the stiffness of the medium.
causing the strains. Treating the tunnel as an A rectangular box structure will undergo
elastic beam allows the calculation of com- transverse racking deformations (Fig. 12) when
bined axial and curvature deformations. subjected to earthquake shear distortions.
Racking deformations are defined as the differ-
The strain bending component is relatively ential sideway movements between the top and
small compared to axial strains, but if the tunnel bottom elevations of rectangular structures. If an
radius increases, the curvature contribution initial calculation is needed, it can be calculated
increases. Tunnel cracks may open and then based on St John and Zahrah (1987) equations.
close in the lining due to the cyclic nature of the
axial strains. As long as the cracks are small, are Numerical Analysis
uniformly distributed, and do not affect the per- Many computer programs are available to esti-
formance of the tunnel, even unreinforced con- mate free-field shear distortions: SHAKE
crete linings are considered adequate. It is (Schnabel et al. 1972), FLUSH (Lysmer
important to emphasize that the p- and s-wave et al. 1975), D-MOD (Matasovic 1993), and
velocities used are those of the deep rock. The DEEPSOIL (Hashash et al. 2011; Hashash and
range for s-wave is between 2 and 4 km/s and Park 2001), among others. One-dimensional site
p-wave between 4 and 8 km/s (Power et al. 1996). response analyses can be used to characterize the
change in the propagating ground motions on
Ovaling and Racking Deformation variable soil profiles, but these analyses only
Ovaling deformations, developed by waves act- take into consideration vertically propagating
ing perpendicular to the circular tunnel lining, are shear waves. However, these are the waves that
Seismic Design of
g max
Tunnels, Fig. 11 Free- Δd = ±
field shear distortions of d 2
perforated and
non-perforated ground Δd
2
Δd = ± 2g
max(1–um)
d
Δd
2
Seismic Design of Horizontal Shear Deformation, Δ (ft) S

Tunnels, Fig. 12 Typical 0.1 0.2 0.3 0.4 Ground Surface
free-field racking
deformation imposed on a ΔB ΔA
rectangular frame (Wang
1993)
Depth Below Surface (ft)
50 Δdiff Δdiff
A
100
Δdiff = ΔA – ΔB
B
150
Soil Deformation Racking Deformation
Profile of a Box Structure
carry most of the seismic energy. The analyses Transverse: Ovaling Deformations of Circular
can be performed in equivalent linear frequency Tunnels
domain or nonlinear time domain. The resulting Peck et al. (1972) proposed closed-form solutions
free-field shear distortion can be expressed in the in terms of thrust, bending moments, and dis-
form of shear strain or shear deformation profile placement under external loading. The lining
with depth. response was a function of structure compress-
ibility and flexibility ratios, in situ overburden
Applicability of Free-Field Deformation Approach pressure, and at-rest earth coefficient. To adapt
The free-field deformation approach is a simple to seismic loading, the free-field shear stress
and effective design tool when earthquake- replaces the in situ overburden pressure and
induced ground motions are small. However, in earth coefficient. The stiffness of the tunnel rela-
structures located within soft soil profiles, the tive to the ground is measured by the compress-
method gives overly conservative designs, ibility (C) and flexibility (F) ratios. Those are the
because free-field ground distortions in these extensional stiffness and flexural stiffness of the
soils are large. It also neglects the difference in medium relative to the lining.
stiffness between the lining and the surrounding Under this framework, Wang (1993)
soil. The presence of an underground structure presented solutions for the diametric strain, the
modifies the free-field deformations; methods to maximum thrust, and the bending moment under
model this interaction will be described in the full-slip conditions, meaning normal force but no
following sections. tangential shear force are present between the
lining and the medium. For most cases the inter-
Pseudo-static Soil-Structure Interaction face condition is between full slip and no slip.
In pseudo-static soil-structure interaction Slip interface can only happen in tunnels in soft
analyses, the soil and structure inertia due to soils or cases of severe seismic loading and full-
seismic shaking is neglected. The problem is slip assumption may lead to underestimation of
simplified to that of a structure in a soil medium the maximum thrust. As shown in Wang (1993)
subjected to simple shear on horizontal and and NCHRP 611 (Anderson et al. 2008), for
vertical planes. ground Poisson’s ratio less than 0.5, thrusts
The beam-on-elastic foundation approach is decrease with decreasing compressibility ratio,
used to model soil-structure interaction effects. but for Poisson’s ratio of 0.5, the thrust response
Both the lining and the medium are assumed to be is independent of compressibility. The normal-
linear elastic. Wang (1993) presents a summary ized lining distortion can be a plotted as function
of closed-form elastic solutions for axial force of flexibility ratio, as shown in Fig. 13. When
and moment developed for circular tunnels due F < 1.0, the lining is considered stiffer than the
to seismic waves propagating along and perpen- ground and deforms less than the ground. When
dicular to the tunnel axis. Adding stiffness and F > 1.0, the lining is expected to deform more
strength to the structure may attract more forces, than the free field with an upper limit equal to the
so a better solution would be to add ductility. perforated ground case as described in Table 3.
These solutions are dependent on the estimates Penzien (2000) provides an analytical procedure
of appropriate spring coefficients compatible to evaluate racking deformations of rectangular
with anticipated displacements and wave lengths. and circular tunnels. His solutions for ovaling
They are often limited to idealized seismic wave deformations in terms of thrust and moment are
forms. very close to those of Wang (1993) for full-slip
Most pseudo-static SSI analyses focus on the condition. However, the value of thrust for
interaction of vertically propagating shear waves no-slip condition is much smaller in Penzien
with the transverse section of a tunnel. These (2000) than in Wang (1993), differing in one
analysis approaches are described next. order of magnitude.
Seismic Design of Tunnels, Fig. 13 Normalized lining deflection vs. flexibility ratio, full-slip interface, circular
lining (Wang 1993)
Seismic Design of Tunnels, Table 3 Explanation of The results from the numerical analyses agree
flexibility ratio with Wang (1993) solutions, highlighting the
Flexibility limitations of the other analytical solution.
ratio F Meaning Sedarat et al. (2009) show that interface condition
F!0 The structure is rigid, so it will not between the tunnel lining and the surrounding
rack regardless of the distortion of the
ground soil has an important impact on the computed
F<1 The structure is stiff relative to the thrust in the lining but limited impact on com-
medium and will therefore deform puted lining deformation.
less. Usually soft soil, and racking
deformations are large Transverse: Racking Deformations of Rectangular
F=1 The structure and medium have equal
Tunnels
stiffness, so the structure will undergo
approximately free-field distortions Box-shaped cut-and-cover tunnels, common for S
F>1 The structure racking stiffness is transportation tunnels, have seismic characteris-
smaller than that of the soil. Usually tics different from circular tunnels because the
stiff soil, and racking deformations walls and slabs of the box-shaped tunnels
are small
are stiffer. They are also often placed at
F!1 The structure has no stiffness, so it
will undergo deformations identical to shallower depths compared to circular tunnels.
the perforated ground Therefore, it is important to carefully consider
the soil-structure interaction due to increased
stiffness and the increased seismic ground
Hashash et al. (2005) compared the two ana- deformations at shallow depths (Hashash et al.
lytical solutions to finite-element method numer- 2001). Numerical analyses are often employed
ical analyses to validate which of the solutions to compute the response of the tunnel struc-
provide the correct solution to this problem. ture to deformation of the surrounding soil.
Seismic Design of Tunnels, Fig. 14 Racking ratio, free-field and structure racking (Hashash et al. 2010)
Wang (1993) and Anderson et al. (2008) employed performing a simple frame analysis; even for
such techniques to develop relationships between simple one-barrel frames, no computer analysis
racking ratio and flexibility (Fig. 14). is needed.
Huo et al. (2006) present an analytical solution The racking ratio (R) is the ratio structure
for deep rectangular structures with a far-field racking distortion to the free-field ground
shear stress. Complex variable theory and con- distortion:
formal mapping were used to develop the solu-
tion of structures in homogeneous, isotropic, Dstructure
R¼
elastic medium. Dfree-field
The relative stiffness between soil and structure,
structure geometry, input earthquake motions, and NCHRP 611 (Anderson et al. 2008) developed
tunnel embedment depth are factors that contribute a relationship between the racking ratio (R) of
to the soil-structure interaction effect. The most rectangular conduits and the flexibility ratio (F),
important of those is the flexibility ratio (F) that whereby
refers to the shear stiffness of the soil relative to the
structure that replaces it (Wang 1993): 2F
R¼
1þF
Gm B
F¼ Hashash et al. (2010) describes the procedure
PH
to be used for performing 2D pseudo-static
where Gmis the shear modulus of soil or rock racking analysis of rectangular tunnels (Fig. 15).
medium, B is the width of the structure, P is the In this type of analysis, the soil is assumed to be
force required to cause a unit racking deflection massless and the section deforms in pure shear. In
of a rectangular frame structure, and H is the a first step, Fig. 15a, the free-field racking defor-
height of the tunnel as illustrated in Fig. 14. For mation time history is computed from 1D site
a rectangular frame with an arbitrary configura- response analyses using hazard-compatible
tion, the flexibility ratio can be determined by ground motion time histories.
Seismic Design of Tunnels, Fig. 15 2D pseudo-static numerical analysis (Hashash et al. 2010)
To perform the 2D numerical analysis, the 1–3 m above and below the structure should be
second step (Fig. 15b) is to define the elastic included. Shear modulus values can be selected
properties of a uniform soil medium as the aver- using the strain-compatible shear wave velocities
age strain-compatible elastic properties of the from site response analysis in the selected layers,
selected soil layers (Anderson et al. 2008). from which the average shear modulus over the
As recommended in Hashash et al. (2010), layers selected layers can be calculated. In this step, the
structural member properties are needed: E (stiff- assemblage of distinct blocks (rigid or deform-
ness), I (moment of inertia), and A (cross- able) with a prescribed constitutive relationship.
sectional area). One of the advantages of these dynamic soil-
Finally, in a 2D numerical analysis, the lateral structure interaction analyses is that the soil
displacement (dim) time histories obtained from layers are modeled to reflect the idealized site
1D site response analysis are applied at the left, stratigraphy; each soil layer can be either
right and top boundaries of the model to impose modeled as a linear elastic material with strain-
the free-field racking calculated in the first step on compatible shear modulus and damping values or
the model, as shown in Fig. 15c. With the numer- characterized via soil constitutive models that
ical analysis, the soil medium will transmit shear- represent soil nonlinearity and hysteretic
ing deformations to the box structure and the box response at small strains. However, the use of
racking deformation time history can be obtained. the nonlinear constitutive models requires careful
This is used to obtain the racking ratio (R). selection of input parameters and thus more
Two-dimensional pseudo-static numerical advanced testing to define those input parame-
analyses can be a very useful tool, but they have ters. The nonlinear behavior and the frequency
some limitations. The ground surface shear dis- content of the free-field environment contribute
placements for shallow box structures cannot be to the structural racking behavior.
transmitted uniformly. The model can be artifi- Hashash et al. (2010) provides a simplified 2D
cially extended to address this problem. Racking dynamic soil-structure interaction procedure that
deformations are assumed to vary uniformly over makes computational effort manageable for
the height of the box structure. The response of design purposes of transverse response of rectan-
individual layers is not represented. This gular tunnels. The first step is to perform a 1D site
becomes a problem when layers with very differ- response analysis to obtain the acceleration and
ent stiffness are part of the soil profile. displacement time history throughout the soil
profile and then obtain the strain-compatible
shear wave velocities and damping ratios for the
Dynamic Soil-Structure Interaction
2D model layers.
The complex soil-structure interaction of under-
The numerical analysis involves applying the
ground structures during seismic loading can be
displacement time history at the model base,
simulated using numerical analysis tools which
then propagating the ground motion through
include lumped mass/stiffness methods and
the soil to simulate the soil-box interaction.
finite-element/difference methods.
The displacement time histories must be
Lumped mass/stiffness methods are useful to
obtained at four monitored points, as
analyze the 3D behavior of a tunnel lining in a
shown in Fig. 16, to calculate the relative box

simplified manner. Many parameters for the
displacement Dbox ¼ max abs dh, C dh, D
springs that represent the structure stiffness and
and the free-field relative displacement
the soil stiffness must be defined to have a real-
Df f ¼ max abs dh, A dh, B to then calculate
istic model.
In finite-element/difference models, the tunnel the racking ratio R ¼ Dbox
Df f . The complete soil
structure is discretized and the soil surrounding profile does not need to be included; a limited
the tunnel can be either discretized or represented thickness of soil that captures the characteristics
by springs. 2D and 3D models can also be used to of wave propagation is sufficient. In the dynamic
analyze the inelastic sections of the tunnel cross soil-structure analysis, the soil profile
section. Discrete element models are useful when represented in the analysis reflects the actual
weak planes in the soil profile exist, since in this soil profile, as there is no need to use an average
method the soil/rock mass is represented by an soil layer needed in pseudo-static analyses.
Seismic Design of Tunnels, Fig. 16 2D dynamic soil-structure interaction (Hashash et al. 2010)
Comparison of Pseudo-static and Dynamic performed to verify and supplement the results of
Racking Soil-Structure Interaction Analyses pseudo-static soil-structure interaction.
Hashash et al. (2010) performed a series of
pseudo-static and dynamic soil-structure interac-
tion analyses of single- and double-box structures Performance Evaluation for Immersed-
in stiff and soft soil profiles using equivalent Tube Tunnels
linear and nonlinear site response analysis with
14 ground motion time histories. The results are Pseudo-static longitudinal 3D models can be used
shown in Fig. 17. The results of the study found to analyze axial and bending deformations in
that for F < 1 the dynamic analysis and pseudo- immersed-tube tunnels. In a lumped mass analy-
static analysis appear to be quite similar and sis approach, the tunnel lining is divided into
slightly above NCHRP 611 (Anderson individual segments with different masses and
et al. 2008). For 4 < F < 9, the dynamic results stiffnesses. The masses are then connected by
show more scatter and higher racking ratios com- springs that represent the axial, shear, and bend-
pared to pseudo-static results, both above ing stiffness of the tunnel as shown in Fig. 18
NCHRP611 (Anderson et al. 2008). Dynamic (Hashash et al. 1998). Free-field displacement S
analyses for 10 < F < 13 show slightly lower time histories that consider effects of wave pas-
racking ratios than pseudo-static analysis, both sage/phase shift and incoherence are calculated at
below NCHRP611 (Anderson et al. 2008). selected locations along the tunnel’s length. The
Some of the main conclusions from Hashash computed free-field displacement time histories
et al. (2010) are: (a) numerical approaches pro- are then applied at the ends of springs to represent
vide results and trends that are consistent with soil-tunnel interaction in a quasi-static analysis.
results obtained from simplified closed-form If a dynamic analysis is needed, appropriate
solutions, (b) there is a need to account for vari- damping factors need to be incorporated into the
ability in the input ground motions and site structure as well as springs to represent the soil.
response analysis methods as they affect the flex- Recent work from Anastasopoulos
ibility ratio (F), and (c) dynamic analyses must be et al. (2007) focuses on nonlinear response to
Seismic Design of Tunnels, Fig. 17 Results from comparison of 2D pseudo-static and dynamic numerical analyses
(Hashash et al. 2010)
Seismic Design of Alameda Vertical

Tunnels, Fig. 18 3D Portal
Transverse Oakland
model for global response
Portal
of immersed-tube tunnel Soil Spring
Vertical Tube
Longitudinal
Transverse Displacement Time
Tube Section History Input
strong seismic shaking of deep immersed tunnels gaskets and relatively small segment lengths.
(
70 m). The free-field acceleration time histories However, it is important to note that this study
are computed at the base of the tunnel through 1D did not examine the heterogeneous nature of the
wave propagation analysis using equivalent linear soil conditions and ground motion incoherency
and nonlinear analyses. The computed free-field and did not investigate the time-dependent stress
acceleration time histories are then imposed on the relaxation on the rubber gasket or the effect of
supports of the tunnel in the form of excitation. tectonic displacements from fault rupturing.
The tunnel is modeled as a multi-segment beam
connected to the ground through calibrated
springs, dashpots, and sliders. Wave passage Additional Seismic Performance Issues
effects are taken into account using Eurocode
8 (EC8 2002); however, the geometric incoher- A number of additional items must be considered
ence was not considered because it did not make a when evaluating the seismic performance of box
difference when added to wave passage. The soil structures. The detailed treatment of all these
is assumed to be uniform along the tunnel. items is beyond the scope of this document, but
A finite-element analysis is used to perform a a brief overview of some of them is presented.
nonlinear dynamic transient analysis of the tun-
nel. Tunnel segments are modeled using beam Response to Vertical Ground Shaking
elements that take into account shear rigidity. Significant vertical ground motions often due to
The joints are modeled with nonlinear near-fault effects can impose vertical loads on
hyperelastic elements. The bored tunnels at the the roof of box structures. Figure 19 shows a
end of both segments are incorporated in the schematic of the vertical loading on relatively
analysis as beams on viscoelastic foundation. shallow box structure (Hashash et al. 2010). S
Influence of segment length and joint properties Two types of analyses can be performed to assess
was then investigated parametrically. the vertical loads. The vertical acceleration near
The results from Anastasopoulos et al. (2007) the ground surface can be estimated as part of the
show that seismic response of immersed tunnel seismic hazard analysis and then be used to com-
correlates better with PGV than PGA consistent puter pseudo-static inertial load of the soil on the
with prior studies. There are some key conclu- tunnel roof. Alternatively, 2D dynamic soil-
sions applicable to immersed tunneling projects. structure interaction analysis representing both
First, a properly designed immersed tunnel can the underground structure and the soil can be
resist near-fault soil-amplified excitation with a performed to compute the vertical inertial load-
PGA as large as 0.6 g and PGV as large as 80 m/s ing on the roof. There remain significant uncer-
and containing long period pulses. Also, the net tainties in selecting appropriate soil properties
tension and excessive compression between seg- for propagating vertical ground motion in such
ments can be avoided by suitable design of joint a model.
Seismic Design of
Tunnels, Fig. 19 Vertical
ground shaking effects
Interaction of Temporary and Permanent reduces the racking ratio of the tunnel structure,
Structures and hence neglecting the presence of the tempo-
Braced excavations are used when space is rary walls is acceptable. An increase between
needed to construct a shallow underground struc- 15 % and 20 % in the racking ratios is estimated
ture. Temporary excavations in highly seismic when the temporary shoring wall is modeled
urban areas are also being seismically designed above and below the permanent structure. The
though for a lower level of seismic shaking than a shoring walls act as extended wings, transferring
permanent structure. Dynamic soil-structure soil loads to the structure from above and below
interaction analyses are preferable to analyze the structure and therefore creating more racking
racking in temporary structures, because it allows deformations (Hashash et al. 2010). Therefore,
the dynamic load increments and levels of defor- neglecting the contribution of the shoring walls
mation on the temporary system to be estimated in this case will lead to underestimating of the
without significant simplifications. racking deformation and is thus not advisable.
It is customary to neglect the contribution of
the temporary shoring wall on the permanent Impact of Superstructure and Adjacent
tunnel box. This is based on the assumption that Structures
the presence of the temporary walls will enhance Shallow underground structures for public trans-
the performance of the system and neglecting it is portation are a key component of sustainable
prudent. The effect of the presence of shoring cities. In dense urban environments, underground
walls on the seismic response of permanent box structures are built in close proximity to high-rise
structures was studied in Hashash et al. (2010). building foundations. Tall buildings have the
For many projects only a few feet of the top of potential to change the ground motions in the
shoring is system is cut off after completion of the foundation soil and therefore transmit significant
box construction while the rest of the wall forces and base shear to adjacent underground
remains in place. The study analyzed a single structures. It is therefore important to evaluate
box surrounded by soft clay using both 2D the impact of transmitted forces from the super-
pseudo-static and dynamic methodologies, as structure to the soil to the underground structure
discussed earlier. Three cases were considered under earthquake loading.
in Fig. 20a. If aboveground structures are built over
Figure 20b shows that modeling the shoring existing underground structures, the interaction
wall over the height of the structure only slightly between the structures must be evaluated.
Seismic Design of Tunnels, Fig. 20 Effect of shoring walls on permanent structures (Hashash et al. 2010)
The global system performance can be evaluated junctions, tunnels crossing distinct geologic
by means of a numerical analysis of seismic soil- media, and local restraints of any type (Hashash
box-structure interaction. et al. 2001).
A related item is that the placement of wide The most common solution to these interface S
box structures in deep excavations in soft soils problems is the use of flexible joints. The first
may lead to significant changes in the soil prop- step in the design process is to use closed-form
erties beneath the structure. The influence of solutions or numerical analyses to determine the
those changes on the long-term dynamic response required allowable rotation and differential
of the structure should be taken into account movements in the longitudinal, transverse direc-
(Hashash et al. 2010). tions. The joints also must be designed to support
static and dynamic earth and water loads before
Tunnel Joints: Portals, Stations, and Tunnel and during the earthquake while remaining
Segments watertight. If a continuous design is
Stiffness differences in the tunnel structure may implemented, very large forces and moments
generate differential movements and stress con- are likely to be generated (Hashash et al. 2001).
centration. Some examples include tunnel- Tunnel portals and vent structures have part of
building or tunnel-station connections, tunnel their structure above ground; therefore, the
Z
Z
Ring “a” Key segment Key segment
X Y
X Y left rings
Ring “b”
Key segment Key segment
right rings
Ring “b” Ring “a”
Seismic Design of Tunnels, Fig. 21 FEM lining models for trapezoidal and rectangular geometric configurations
boundary conditions (Kramer et al. 2007)
Soil
Gap
Concrete Liner
X Y
Concrete Tunnel Linear
Soil Block
80-ft by 80-ft by 20-ft
Boundary Conditions at Sides and Top:
Controlled Horizontal Movement
Controlled Vertical Movement
Boundary Condition at the Base:

Fixed at All Directions
Seismic Design of Tunnels, Fig. 22 Details of mesh development and boundary conditions (Kramer et al. 2007)
seismic design should account for inertial effects connections must therefore be designed to
or potential pounding. It is always preferable that accommodate anticipated deformations. The
those structures are isolated from the tunnel joint behavior can be kept elastic or allow inelas-
through flexible joints (Hashash et al. 2001). ticity if proper detailing is performed. Numerical
For soil-rock transitions, it is recommended analyses are often used to evaluate the seismic
that the tunnel not be cast directly against the performance of segmental lining. Kramer
rock. Instead, there should be some over- et al. (2007) performed 3D finite-element ana-
excavation filled with soil or aggregate. If that is lyses, Fig. 21, to compute radial and circumfer-
not possible in bored tunnels, then a flexible ential joint behavior during seismic ovaling and
lining can be installed (Hashash et al. 2001). wave propagation for the Silicon Valley Rapid
Transit (SVRT) Project in San Jose, California.
Precast Tunnel Lining These analyses incorporate inelastic constitutive
When a tunneling machine is used to excavate a soil behavior, cracked concrete properties, and
tunnel, the lining usually is erected in segments. no-tension, frictional segment joint surfaces
Single pass precast concrete segmental lining (Fig. 22). Kramer et al. (2007) provide guidelines
systems are often employed. The segment joint for precast lining design in high seismicity zones.
Seismic Retrofit of Existing Facilities seismic evaluation is different from aboveground

When considering seismic retrofit of an existing structures.
tunnel structure, the retrofit strategy depends on Seismic evaluation of underground structures
the structure damage mode. If there is a gross includes evaluation for permanent and transient
stability problem, the seismic retrofit strategies ground deformations. Pseudo-static and dynamic
involve strengthening the structure itself or the analysis approaches can be used to estimate the
surrounding geologic materials. deformation of underground structures due to
One concern for a circular tunnel is the contact transient ground deformations and are presented.
quality between the liner and the surrounding Additional seismic design issues were
geologic media. Strengthening this interface discussed, including vertical ground shaking and
includes replacing the lining, increasing the lin- response, interaction of temporary and perma-
ing thickness by adding reinforced concrete, or nent structures, impact of superstructure and
adding reinforcing bars or internal steel liner. adjacent structures, tunnel joints, and seismic
Increasing lining thickness is not always a good retrofit of existing facilities.
solution, as it tends to attract more load. Methods
that increase ductility as well as strength are more
effective (Power et al. 1996). Adding circumfer- Cross-References
ential joints along the tunnel axis can also reduce
the stresses and strains induced by longitudinal ▶ Conditional Spectra
propagating waves. However, the value of adding ▶ Dynamic Soil Properties: In Situ
joints must be weighed against the expected per- Characterization Using Penetration Tests
formance of the liner without joints. It is impor- ▶ Liquefaction: Countermeasures to Mitigate
tant to verify in the retrofit design that there will Risk
be no additional water leakage and that the joints ▶ Nonlinear Dynamic Seismic Analysis
will not become weak spots (Power et al. 1996). ▶ Nonlinear Finite Element Analysis
In cut-and-cover structures, some seismic ret- ▶ Nonlinear Seismic Ground Response Analysis
rofit strategies include increasing ductility of of Local Site Effects with Three-dimensional
reinforced concrete lining, adding confinement High-fidelity Model
at existing columns, and adding steel plate jackets ▶ Performance-Based Design Procedure for
at joints. Structures with Magneto-Rheological
Dampers
▶ Probabilistic Seismic Hazard Models
Summary ▶ Seismic Actions Due to Near-Fault Ground
Motion
The performance of the underground facilities ▶ Site Response for Seismic Hazard Assessment S
during recent seismic events showed that the ▶ Site Response: 1-D Time Domain Analyses
underground structures have suffered appreciably ▶ Site Response: Comparison Between Theory
less damage than surface structures. However, and Observation
the failure of even one of these underground ▶ Soil-Structure Interaction
structures can be detrimental to the proper post-
earthquake operation of a tunnel network in urban
areas. Thus, evaluation of underground structures References
to seismic shaking is necessary.
A performance-based approach is recommended Anastasopoulos I, Gerolymos N, Drosos V, Kourkoulis R,
for seismic evaluation of underground structures. Georgarakos T, Gazetas G (2007) Nonlinear
response of deep immersed tunnel to strong seismic
Underground structures seismic response is con- shaking. J Geotech Geoenviron Eng 133(9):
trolled by the deformation of the soil or rock 1067–1090. doi:10.1061/(ASCE)1090-0241(2007)
medium in which they are embedded and their 133:9(1067)
Anderson DG, Martin GR, Lam I, Wang JN (2008) Seis- to far-field shear stresses. Tunn Undergr Space
mic analysis and design of retaining walls, buried Technol 21(6):613–625
structures, slopes and embankments, vol NCHRP Iida H, Hiroto T, Yoshida N, Iwafuji M (1996) Damage to
611: National Cooperative Highway Research Pro- the Daikai subway station. Soils and Foundations
gram, Transportation Research Board 283–300. Special issue on geotechnical aspects of the
Ashford SA, Boulanger RW, Donahue JL, Stewart JP January 17, 1995 Hyogoken-Nambu Earthquake,
(2011) Geotechnical quick report on the Kanto Plain Tokyo, Japan
Region during the March 11, 2011, Off Pacific Coast Kramer GJE, Sedarat H, Kozak A, Liu A, Chai J (2007)
of Tohoku Earthquake, Japan GEER Association Seismic response of precast tunnel linings. Paper
report no. GEER-025a: Geotechnical Extreme Events presented at the proceedings of the 2007 rapid excava-
Reconnaissance (GEER) tion and tunneling conference, Toronto. Publisher and
Coppersmith KJ, Youngs RR (2000) Data needs for prob- copyright holder: Society for Mining, Metallurgy &
abilistic fault displacement hazard analysis. In: Exploration (SME). www.smenet.org
Cello G, Tondi E (eds) The resolution of geological Lysmer J, Udaka T, Tsai CF, Seed HB (1975) FLUSH: a
analysis and models for earthquake. Pergamon Press, computer program for approximate 3-D analysis of
Oxford soil-structure interaction problems. Earthquake Engi-
Cording EJ, Martin G, Parker H (2011) Century city area neering Research Center, Berkeley
tunneling safety and fault investigations: tunnel advi- Matasovic N (1993) Seismic response of composite
sory panel. Los Angeles County Metropolitan Trans- horizontally-layered soil deposits. PhD thesis, Univer-
portation Authority, Los Angeles sity of California, Los Angeles. Retrieved from
EC8 (2002) Eurocode 8: design of structures for earth- http://wwwlib.umi.com/pqdd2/search/do?query=au%
quake resistance. European Committee for Standardi- 28Matasovic%2C%20Neven%2C%29%20and%20da
zation (CEN), Brussels %281993%29
Elnashai AS, Gencturk B, Kwon O-S, Al-Qadi IL, Newmark NM (1967) Problems in wave propagation in
Hashash Y, Roesler JR, et al. (2010) The Maule soil and rock. Paper presented at the international sym-
(Chile) earthquake of February 27, 2010 consequence posium on wave propagation and dynamic properties
assessment and case studies: Mid-America Earthquake of earth materials, Albuquerque, NM
Center Peck RB, Hendron AJ, Mohraz B (1972) State of the
Hamburger RO, Hooper JD (2011) Performance-based art in soft ground tunneling. Paper presented at
seismic design. Mod Steel Constr 4:36–39 the rapid excavation and tunneling conference,
Hashash YMA, Park D (2001) Non-linear New York
one-dimensional seismic ground motion propagation Penzien J (2000) Seismically induced racking of tunnel
in the Mississippi embayment. Eng Geol linings. Earthq Eng Struct Dyn 29(5):683–691.
62(1–3):185–206 doi:10.1002/(sici)1096-9845(200005)29:5<683::aid-
Hashash YMA, Tseng WS, Krimotat A (1998) Seismic eqe932>3.0.co;2-1
soil-structure interaction analysis for immersed tube Power MS, Rosidi D, Kaneshiro J (1996) Volume III
tunnels retrofit. Paper presented at the proceedings of Strawman: screening, evaluation, and retrofit design
the 1998 conference on geotechnical earthquake engi- of tunnels. National Center for Earthquake Engineer-
neering and soil dynamics III. Part 2 (of 2), Seattle, 3–6 ing Research, Buffalo
Aug 1998 Schmidt B, Hashash YMA (1999) Preventing tunnel flo-
Hashash YMA, Hook JJ, Schmidt B, Yao JI-C (2001) Seis- tation due to liquefaction. Paper presented at the pro-
mic design and analysis of underground structures. ceedings of the second international conference on
Tunn Undergr Space Technol 16:247–293 earthquake geotechnical engineering, Lisbon
Hashash YMA, Park D, Yao JIC (2005) Ovaling deforma- Schnabel PB, Lysmer JL, Seed HB (1972) SHAKE: a
tions of circular tunnels under seismic loading, an computer program for earthquake response analysis
update on seismic design and analysis of underground of horizontally layered sites. Earthquake Engineering
structures. Tunn Undergr Space Technol Research Center, Berkeley
20(5):435–441. doi:10.1016/j.tust.2005.02.004 Sedarat H, Kozak A, Hashash YMA, Shamsabadi A,
Hashash YMA, Karina K, Koutsoftas D, O’Riordan Krimotat A (2009) Contact interface in seismic analy-
N (2010) Seismic design considerations for under- sis of circular tunnels. Tunn Undergr Space Technol
ground box structures. Paper presented at the 2010 24:482–490
earth retention conference – earth retention conference St John CM, Zahrah TF (1987) Aseismic design of under-
3, Bellevue, 1–4 Aug 2010 ground structures. Tunn Undergr Space Technol
Hashash YMA, Groholski D, Musgrove M, Park D, 2(2):165–197
Phillips C, Tsai C-C (2011) DEEPSOIL V5.0, manual Wang JN (1993) Seismic design of tunnels: a state-of-the-
and tutorial. Board of Trustees of University of Illinois art approach, Monograph 7. Parsons Brinckerhoff
at Urbana-Champaign, Urbana Quade & Douglas, New York
Huo H, Bobet A, Fernández G, Ramı́rez J (2006) Analyt- Wang WL, Wang TT, Su JJ, Lin CH, Seng CR, Huang TH
ical solution for deep rectangular structures subjected (2001) Assessment of damage in mountain tunnels due
Seismic Design of Waste Containment Systems 2823
to the Taiwan Chi-Chi earthquake. Tunn Undergr of liquids and gas, as drainage layers for landfill
Space Technol 16(3):133–150 leachate and surface water infiltration, and as
Youngs RR, Arabasz WJ, Anderson RE, Ramelli AR, Ake
JP, Slemmons DB, et al. (2003) A methodology for protection for other components of the waste
probabilistic fault displacement hazard analysis containment system. Geosynthetic waste contain-
(PFDHA). Earthq Spectra 19(1):191–219. ment system elements, including geomembrane
doi:10.1193/1.1542891 and geosynthetic clay liner (GCL) barrier layers,
geonet drainage layers, and geotextile filter, sep-
aration, and cushion (geomembrane protection)
layers, are typically only found in containment
Seismic Design of Waste systems constructed within the last 25 years,
Containment Systems appearing with increasing frequency over that
time span.
Edward Kavazanjian Jr. Figure 1 illustrates the components of munic-
School of Sustainable Engineering and the Built ipal solid waste (MSW) landfill waste contain-
Environment, Arizona State University, ment systems. MSW landfills are by far the most
Tempe, AZ, USA common type of engineered waste containment
system, followed (with respect to frequency of
occurrence) by cover systems (caps) for
Synonyms uncontrolled dumps and hazardous waste sites
and then by hazardous waste landfills. Caps and
Equivalent linear analysis; Geosynthetic; Land- containment systems for hazardous waste land-
fill; Performance-based design; Site response; fills employ similar elements to MSW landfills.
Solid waste; Waste containment systems The performance of ancillary facilities at a waste
disposal site subject to seismic loading, including
leachate and gas treatment facilities, surface
Introduction water control systems, access roadways, and
landfill monitoring systems, is also an important
Maintaining the integrity of waste containment consideration but will not be addressed herein.
systems subject to seismic loading is not only Following a brief review of the performance
important with respect to protecting human of landfills in earthquakes, this entry summarizes
health and the environment from wastes and the state of knowledge on waste mass seismic
waste by-products, it is also important with response and the properties of the waste and
respect to the availability of facilities essential geosynthetic materials relevant to seismic analy-
for post-earthquake recovery, i.e., of landfills for sis and design. Current approaches to the seismic
proper disposal of the large volumes of design of landfills are then discussed, culminat- S
earthquake-generated waste. Important considering in a discussion of the value of performance-
ations in the seismic design of waste containment based seismic design for these important
systems include the response of the waste mass facilities.
itself to seismic excitation, the stability of the
waste mass (including its foundation), and
the integrity of the engineered components of Seismic Performance of Waste
the waste containment system. Engineered com- Containment Systems
ponents of waste containment systems include
base and side-slope liner (barrier) systems, leach- In general, the seismic performance of waste
ate and gas collection and removal systems, and containment systems subject to strong ground
final cover systems. Liner and cover systems motion from earthquakes has been acceptable,
include both natural soil and geosynthetic ele- i.e., has not resulted in a harmful discharge of
ments that serve as barriers to advective transport waste or waste by-products to the environment.
2824 Seismic Design of Waste Containment Systems
VERTICAL GAS EXTRACTION WELL

VEGETATED SOIL COVER
VEGETATED SOIL COVER CLAY BARRIER LAYER
DRAINAGE LAYER GAS COLLECTION/
GEOMEMBRANE FOUNDATION LAYER
GEOSYNTHETIC CLAY LINER
FOUNDATION LAYER
HORIZONTAL GAS EXTRACTION WELL SUBSURFACE BARRIER
GROUNDWATER (LEACHATE)
EXTRACTION WELL
WASTE
WASTE
LEACHATE REMOVAL SIDE SLOPE RISER

LEACHATE COLLECTION LAYER
GEOMEMBRANE
COMPACTED CLAY
LEACHATE SUMP
VADOSE ZONE MONITORING PROBE
GROUNDWATER MONITORING WELL
Seismic Design of Waste Containment Systems, Fig. 1 Containment system components for modern (left) and
older (right) landfills (Kavazanjian et al. 1998)
In the epicentral region of the 1989 M 6.9 Loma OII Landfill nor any of the landfills subject to
Prieta earthquake, several MSW landfills with strong ground shaking in the Loma Prieta earth-
2H:1V slopes rising up to 60 m above grade quake employed geosynthetic containment sys-
withstood earthquake motions with peak ground tem elements (Kavazanjian et al. 2013b;
accelerations estimated to be on the order of 0.6 g Matasovic et al. 1998), Tensile rupture of a
with only minor cracking in the soil cover, e.g., geomembrane barrier layer was observed at two
observed cracks on the order of a 50 mm or less separate locations at the Chiquita Canyon Land-
(Buranek and Prasad 1991; Johnson et al. 1991; fill in the Northridge earthquake, one location at
Orr and Finch 1990). However, there were obser- the crest of a lined side slope and a second loca-
vations of more significant cracking in the soil tion near the crest of a lined side slope (EMCON
cover on the very steep north slope of the Oper- 1994). Figure 2a shows the cover soil cracking
ating Industries, Inc. (OII) Landfill (average observed at the OII Landfill, and Fig. 2b shows
inclination, 1.5H:1V; maximum inclination, one of the geomembrane tears observed at the
1.3H:1V) in the 1987 M 6.1 Whittier Narrows Chiquita Canyon Landfill following the
earthquake and the 1994 M 6.7 Northridge earth- Northridge earthquake.
quake when they were subject to peak ground Geosynthetic waste containment system ele-
accelerations estimated to be less than half the ments are of particular concern with respect to
0.6 g level experienced by landfills in the Loma seismic loading because they create the potential
Prieta event (Kavazanjian et al. 2013b; for planes of weakness along which slippage and
Matasovic et al. 1998). Furthermore, neither the instability can occur (due to a relatively low
Seismic Design of Waste Containment Systems, benches of the OII Landfill (Courtesy or Raymond
Fig. 2 Damage to landfills observed following the B. Seed); (b) tear in the side-slope liner at the Chiquita
Northridge earthquake: (a) cracking in soil cover on the Canyon Landfill (Courtesy of Robert M. Koerner)
interface or in-plane shear resistance) and Waste Mass Seismic Response

because of their potential for tensile rupture
which could result in breaching of barrier, filter, Overview: The seismic response of the waste
and protective layers and disruption of drainage mass is an important consideration in waste con-
layers. Stability failures of landfills have tainment system seismic design as it controls the
occurred under static loading due to the low loading on the containment system elements as
interface shear strength of a geosynthetic waste well as the global stability of the waste mass.
containment system element (Koerner and Soong Because there is often a significant impedance
2000; Mitchell et al. 1990; Seed et al. 1990) and contrast between the waste mass and the under-
the tensile rupture of the side-slope lying foundation material (soil or rock), landfill S
geomembranes at the Chiquita Canyon Landfill seismic response is susceptible to amplification
were cited previously. Furthermore, there are no of free-field motions. Amplification of earth-
case histories of cover systems with quake ground motions is reported to have
geomembrane elements subject to free-field occurred at the one landfill at which strong
ground motions in excess of 0.2 g, and cover ground motions have been recorded to date, OII
systems are of particular concern due to the Landfill (Kavazanjian et al. 2013b), and Bray and
potential for amplification of seismic motions Rathje (1998) have conducted analyses
by the waste mass. There is also a potential for suggesting that the amplification potential of
hidden damage to containment system elements municipal solid waste landfills is similar to the
beneath the surface of the landfill, where damage amplification reported by Harder (1991) between
cannot be directly observed or readily detected by the base and crest of earth dams in the transverse
the monitoring systems employed in current direction. Figure 3 compares the amplification of
practice. peak ground acceleration from the base to the
Seismic Design of Waste

Containment Systems,
Fig. 3 Amplification of
peak ground acceleration at
the OII Landfill in the
Northridge earthquake
(Kavazanjian et al. 2013b)
crest of the OII Landfill (as determined by back Most seismic response analyses in landfill
analysis) to the amplification of the transverse practice are conducted using the equivalent linear
peak ground acceleration from the base to the method. Waste mass properties of importance in
crest of earth dams as reported by Harder (1991). an equivalent linear seismic response analysis
Waste Properties: Seismic response of a include the waste total unit weight (or mass den-
waste mass is controlled by waste mass proper- sity), the small-strain shear modulus, and equiv-
ties, the waste thickness, and the impedance con- alent linear shear modulus reduction and
trast between the base of the landfill and the waste damping curves. Poisson’s ratio may also be of
mass (which depends upon the properties of the interest if two-dimensional equivalent linear
waste mass and the underlying foundation mate- response analyses are to be conducted. Small
rial) as well as the characteristics of the free-field strain shear modulus is generally established
ground motion. For landfills with geosynthetic based upon shear wave velocity and unit weight
elements in the liner and cover system, slip or mass density. Waste mass shear strength and
along the interface of a geosynthetic element the interface shear strength of geosynthetic ele-
may also affect the seismic response. However, ments (or in the case of a geosynthetic clay liner,
slip at geosynthetic interfaces is usually ignored the in-plane shear strength) are essential proper-
in practice when evaluating landfill seismic ties in seismic analysis of waste containment
response. Analyses in which the potential for systems for waste mass stability and cover system
slip at the interface is ignored are termed assessments. The properties of natural soil mate-
“decoupled” analyses as landfill seismic response rials used in landfill construction are also impor-
is decoupled from the geosynthetic interface tant in a landfill seismic stability assessment but
behavior and potential displacement at the are not discussed herein.
interface. A decoupled analysis is generally Figure 4, from Zekkos et al. (2006), provides a
considered conservative with respect to family of typical unit weight profiles for MSW
predicting landfill seismic performance, though landfills that depend upon the amount of cover
there are relatively rare situations where this has soil and compaction effort provided by the land-
been shown not to be the case (Bray and Rathje fill operator during waste placement. Due to dif-
1998). ficulties in measuring the waste unit weight in the
Seismic Design of Waste

Fig. 4 Typical MSW total
unit weight vs. depth
profiles (Zekkos et al. 2006)
field, Fig. 4 or some other relationship for typical three sources: (1) from back analysis of strong
unit values is usually used in landfill seismic ground motion records from the OII Landfill
analysis and design rather than direct measure- alone (e.g., Idriss et al. 1995), in some cases
ment, even for existing facilities. supplemented with cyclic laboratory testing of
Shear wave velocity and the small-strain mod- waste (Matasovic and Kavazanjian 1998);
ulus of MSW have been measured in the field and (2) from large-diameter cyclic triaxial testing by
in the laboratory. Zekkos et al. (2008) and Yuan Zekkos et al. (2008) on reconstituted waste spec-
et al. (2011) report shear wave velocity and imens from the Tri Cities Landfill for three dif-
small-strain modulus values measured in labora- ferent composition ratios; and (3) from large-
tory testing of MSW reconstituted at three scale laboratory simple shear testing by Yuan
different ratios of refuse (MSW particles et al. (2011) on reconstituted waste specimens
>20 mm, i.e., retained on a 20-mm sieve) to from the Tri Cities Landfill for the same three
soil-sized (MSW particles <20 mm, i.e., passing composition ratios employed by Zekkos et al.
a 20-mm sieve) material (note that these investi- (2008). Figure 6 presents the equivalent linear
gators used the same MSW and same composi- shear modulus reduction (Fig. 6a) and damping
tion ratios but different laboratory devices). (Fig. 6b) curves from Zekkos et al. (2008) from S
However, unlike unit weight, waste mass shear large-scale triaxial testing, termed the UCB data,
wave velocity is readily measured in the field, and from Yuan et al. (2011) from large-scale
e.g., using noninvasive surface wave techniques. simple shear testing, termed the ASU data.
Field data is more reliable than laboratory testing These two data sets are relatively consistent
on reconstituted specimens, and case histories of with each other and with the OII field data and
site-specific shear wave velocity measurements represent the best available information on these
for landfills are being reported in the literature property relationship.
with increasing frequency. Figure 5 provides a Poisson’s ratio values for MSW have been
summary of shear wave velocity profiles for derived from field measurements of shear and
MSW landfills worldwide as compiled by compressional wave velocity at the OII Landfill
Ramaiah (2013). (Matasovic and Kavazanjian 1998) and from lab-
Equivalent linear shear modulus reduction and oratory measurements of axial and radial strain
damping curves for MSW are available from (Zekkos 2005). Poisson’s ratio can also be
Vs (m/s)
0 50 100 150 200 250 300 350 400
0
5
10
15
Depth (m)
20
25
30
35
40
45
50
Legend Landfill Name Location Method Reference
Puente Hills landfill California, USA Down-hole Earth Technology 1988
Puente Hills landfill California, USA Cross-hole Earth Technology 1988
Richmond landfill California, USA Down-hole Sharma et al. 1990
North west Regional Arizona, USA Down-hole Houston et al. 1995
Landfill Facility
Live Oka landfill Georgia, USA SASW Haker et al. 1997
Sanifil landfill Georgia, USA SASW Haker et al. 1997
Bolton Raod landfill Atlanta, USA SASW Haker et al. 1997
Olympic View Sanitary Washington, USA SASW Matasovic & Kavazanjian 2006
Landfill (OVSL)-Old
OVSL-West Washington, USA SASW Matasovic & Kavazanjian 2006
OVSL -South Washington, USA SASW Matasovic & Kavazanjian 2006
Villalba landfill Madrid, Spain SASW Cuellar et al. 1998
Valdemingomez landfill Madrid, Spain SASW Pereira et al. 2002
Lasdehasas landfill Madrid, Spain SASW Mateos 2006
Maldegem landfill Ghent, Belgium SASW Van Impe & Bouzza 1996
Site 1 (6-8 years old) Melbourne, Australia CSSASW Bouzza & Kavazanjian 2000
Site 2 (7-8 years old) Melbourne, Australia CSSASW Bouzza & Kavazanjian 2000
Spadina landfill Saskatoon, Canada MASW Fleming et al. 2011
Gazhipur landfill Delhi, India SASW Ramaiah et al. 2013
Okhla landfill Delhi, India SASW Ramaiah et al. 2013
Recommended Vs profile in absence of site specific data Kavazanjian et al. 1995
Range of Vs for Southern California Landfills Kavazanjian et al. 1996

Zekkos 2005
Vs range for Tri-Cities, Altamont and Redwood landfills
Vs range for Michigan landfills (13 profiles from 4 landfill sites) Sahadewa et al. 2011
Seismic Design of Waste Containment Systems, Fig. 5 Shear wave velocity at MSW landfills worldwide
(Ramaiah 2013)
Seismic Design of Waste a 1.2

Fig. 6 Equivalent linear
Normalized shear modulus reduction, G/Gmax

property relationships for
waste specimens 1.0
reconstituted at different
composition ratios (Yuan
et al. 2011): (a) shear 0.8
modulus reduction; (b)
damping
0.6
ASU 100%<20mm
0.4
ASU 65%<20mm
ASU 35%<20mm
Zekkos et al 100&<20 mm
0.2 Zekkos et al 62-76%<20mm
Zekkos et al 8-25%<20mm
0.0
0.0001 0.001 0.01 0.1 1 10
Cyclic shear strain (%)
b 30
ASU 100%<20mm
ASU 65%<20mm
25 ASU 35%<20mm
UCB 100%<20mm
UCB 62-76%<20mm
Material damping (%)
20 UCB 8-25%<20mm
15
10
0
0.0001 0.001 0.01 0.1 1 10 S
Cyclic shear strain (%)
inferred from the field measurements of the lat- analyses of landfill failures under static loading
eral earth pressure at rest in MSW reported by to develop shear strength envelopes. Kavazanjian
Dixon et al. (1999). The data on Poisson’s ratio et al. (1996) employed back analysis of both
from these studies is scattered, both within and landfill failures and steep stable landfill slopes
among the studies, but it appears that values on to develop a bilinear strength envelope in which
the order of 0.2–0.4 are appropriate for MSW was assigned a cohesion of 25 kPa at low
unsaturated MSW. normal stresses and a friction angle of 33 at
Due to difficulties associated with conven- higher confining pressures. Based upon back ana-
tional laboratory and field testing, studies of lyses of failed landfill slopes, Eid et al. (2000)
MSW shear strength have employed back proposed that typical MSW shear strength could
be characterized by a linear failure envelope with Furthermore, hazardous waste disposed of in

a friction angle of 35 and a cohesion intercept landfills is often a hazardous material (e.g., pes-
and that the lower- and upper-bound shear ticide residue, organic solvents) or mixed waste
strength should be characterized using zero and (e.g., potentially radioactive construction and
50-kPa cohesion, respectively. Based upon labo- demolition debris) in a soil matrix. Hence, it is
ratory triaxial and direct testing on relatively common to assume that the properties of the
large-sized specimens (e.g., 150 mm-diameter waste mass at a hazardous waste site are the
triaxial specimens), Bray et al. (2009) suggested same as the properties of the backfill soil (or of
that MSW shear strength was dependent on the the native soil at a hazardous material spill site)
normal stress and characterized by a cohesion of (Kavazanjian and Matasovic 2001).
15 kPa and a friction angle that was equal to 36 Geosynthetic Properties: The interface shear
at a normal stress less than or equal to one atmo- strength of geosynthetic materials and the
sphere and that decreased by 5 for every log in-plane internal strength of GCLs can vary
cycle of normal stress. Both Bray et al. (2009) widely, depending upon the type of geosynthetic
and Kavazanjian et al. (2013b) have noted the material, the nature of the material in contact with
anisotropic structure and shear strength of the geosynthetic material, and for GCLs the inter-
MSW, with the strengths reported above applica- nal reinforcement and moisture content
ble to the horizontal plane, the weakest plane in a (Zornberg et al. 2005; McCartney et al. 2009).
waste mass due to the preferred alignment of the Furthermore, many geosynthetic materials have
long axes of waste constituents in this direction, significantly different peak and large-
and with substantially greater shear strength on displacement shear strengths.
non-horizontal planes that cut across the long Hydrated unreinforced or poorly reinforced
axes of the constituents. Based upon analysis of GCLs can have extremely low in-plane shear
landfill performance in the Northridge earth- strengths, in some cases as low as the hydrated
quake, Augello et al. (1995) suggested that the strength of their sodium bentonite core (e.g., a
shear strength of MSW subject to seismic loading friction angle, f, equal to 4 at high normal
was greater than the static strength. Bray stresses). However, such materials are rarely
et al. (2009) suggest that the dynamic shear used in landfill construction. Hydrated reinforced
strength of MSW is at least 20 % greater than GCLs can have a substantial peak internal shear
the static shear strength. However, in most seis- resistance, sometimes in excess of f = 30 , and
mic evaluations, the static shear strength has been still maintain a relatively high strength when
found to be sufficient to demonstrate landfill seis- subject to large deformation (i.e., f > 15 ),
mic stability. depending on the quality of the reinforcement.
The data on material properties for waste Smooth geomembranes in contact with cohe-
materials other than typical MSW is rather sive soils (e.g., compacted clay) can have very
sparse. Hazardous waste in modern landfills is low shear strength (f = on the order of 10 ), but
generally stabilized and containerized prior to textured geomembranes can provide substan-
placement in a landfill. The containers are then tially higher interface strengths, even when sub-
stacked and backfilled with soil. In older hazard- ject to a large displacement (f sometimes on the
ous waste landfills, the waste may have been order of 16–20 for the large displacements).
containerized, typically in partially filled steel Granular materials in contact with smooth
drums, without stabilization. Cone penetrometer geomembranes may have an interface strength
testing at a 20–30-year-old hazardous waste land- exceeding 70 % of the shear strength of the gran-
fill where the waste was placed in partially filled ular soil, while the shear strength of a granular
steel drums without stabilization showed no evi- material in contact with a textured geomembrane
dence that the drums were still intact or of voids may approach the shear strength of the soil. Non-
(as would be expected at the top of a partially woven geotextile placed against a smooth
filled drum) (Matasovic et al. 2005). geomembrane will have a relatively low shear
strength but can have a peak interface strength of seismic response analysis is a one-dimensional
exceeding that of most natural soils if placed equivalent linear analysis of a representative col-
against a textured geomembrane. Woven umn or columns for the waste mass. For important
geotextile placed against a smooth geomembrane projects, nonlinear one-dimensional analysis and
will typically have a higher interface strength two-dimensional equivalent linear analyses are
than a nonwoven geotextile against a smooth sometimes performed. Nonlinear two-dimensional
geomembrane. Woven geotextiles are not typi- analyses are primarily used for research purposes.
cally placed against textured geomembranes. The integrity assessment is usually based upon the
Geonets placed upon a geomembrane (textured calculated seismic response, permanent seismic
or smooth) may have a low interface strength displacement calculated in a Newmark-type seis-
(f = 10 ) but can be heat bonded to a nonwoven mic displacement analysis, and empirical rules on
geotextile and then placed against the allowable displacement for the various elements of
geomembrane if a higher interface strength is the containment system. The Newmark displace-
desired. ment analysis is typically conducted using shear
Minimum required interface strengths are typ- stress time histories from the seismic response anal-
ically established by design analyses and then ysis and a yield acceleration calculated using con-
specified in design documents. Laboratory testing ventional limit equilibrium analysis. The standard
is often conducted during design to demonstrate of practice is to use the large-displacement static
that the required strengths are achievable with shear strength of any interface engaged in the limit
available materials. The geosynthetic materials equilibrium analysis. If more than one interface is
used in construction are then generally subject aligned with a slip surface engaged in the limit
to conformance testing prior to the start of con- equilibrium analysis, the large-displacement static
struction and to quality assurance testing during strength of the interface with the lowest static peak
construction to demonstrate that the specific strength is employed.
materials used in construction can provide the Due to the interrelationship between the yield
specified interface shear strength. acceleration, shear stress time history, and calcu-
Other geosynthetic properties of importance lated seismic displacement, multiple potential
to seismic design include the tensile strength sliding mechanisms must be checked in a landfill
and the strain at yield and at failure of the seismic stability assessment. Potential sliding
geosynthetic elements of the containment sys- mechanisms include veneer failures of the landfill
tem. For advanced analyses, the shear and axial cover along interfaces with containment system
stiffness of geosynthetic material elements may elements (both geosynthetic and natural soil) and
be required. between the cover soil and the waste, circular
surfaces through the waste mass and/or the foun-
dation, and sliding block surfaces along planes of S
Design Analyses weakness in waste, including along interfaces
with geosynthetic and natural soil containment
Decoupled Analysis: Most seismic analyses of system elements and other potential planes of
landfill systems conducted in practice today employ weakness. Figure 7 provides an example of a
a decoupled approach in which the response of the landfill seismic stability analysis in which multi-
landfill to the seismic loading is considered sepa- ple potential sliding surfaces were evaluated
rately from integrity assessments for the waste mass (Kavazanjian et al. 2013b). In this case, the hor-
and containment system components. The seismic izontal planes of weakness were continuous
response analysis is conducted assuming that the layers of daily and interim covers hypothesized
containment system components and waste mass to be present within the waste mass. The calcu-
maintain their integrity, that there is no relative lated seismic displacement on each of these fail-
displacement at geosynthetic interfaces, and that ure surfaces is considered independently in the
no instabilities develop. The most common type integrity assessment.
Seismic Design of Waste Containment Systems, represent the peak average acceleration for each potential
Fig. 7 Potential sliding mechanisms assumed in seismic sliding mechanism (Kavazanjian et al. 2013b)
stability assessment for the OII Landfill: g-values
Due to the conservatism associated with the damage can be readily detected in a post-
decoupled approach and the use of large- earthquake inspection and readily repaired.
displacement static shear strength, the calculated It should be noted that these general guidelines
permanent seismic displacement from this type of are based primarily upon back analyses of the
analysis is considered merely an index of seismic performance of slopes, embankments, and land-
performance and not the expected permanent fills in earthquakes using conventional decoupled
seismic displacement along the hypothesized analysis. If a more sophisticated seismic analysis
sliding surface. A calculated seismic displace- is conducted, e.g., a two-dimensional nonlinear
ment of 150 mm or less is typically considered coupled analysis of seismic response and dis-
indicative of minimal permanent seismic displacement, these guidelines may not be appropri-
placement in the design earthquake, and calcu- ate, particularly with respect to damage to the
lated displacements of up to 1 m are considered geosynthetic elements of the containment
indicative of limited displacement. Calculated system.
displacements of greater than 1 m are considered Coupled Analysis: Coupled analysis of land-
indicative of the potential for large, possibly fill seismic response and waste containment sys-
uncontrolled displacements in the design tem integrity is rare and complex due to
earthquake. difficulties in modeling slip at interfaces and
Table 1 summarizes typical values of accept- geosynthetic element stress–strain behavior.
able calculated seismic displacement for Bray and Rathje (1998) developed a
non-geosynthetic components of landfill contain- one-dimensional model for landfill response that
ment systems from a decoupled analysis, along included slip at the interface. Kavazanjian
with the anticipated duration for the interim and et al. (2012) describe a fully coupled
final repairs of these components. For the two-dimensional model for performance-based
geosynthetic elements of waste containment sys- seismic design of waste containment systems
tems, 150 mm of calculated displacement is gen- that accounts for both slip at an interface and
erally considered to be indicative of no damage, geosynthetic element stress–strain behavior. The
and up to 3 m of displacement may be considered interface model allows for a post-peak decrease
acceptable for geosynthetic elements in which in the interface shear strength. Geosynthetic
Seismic Design of Waste Containment Systems, Table 1 MSW landfill seismic design criteria and performance
standards for decoupled analysis (Kavazanjian et al 1998)
Repair to
Design criteria and Interim remediation to restore pre-earthquake
Cover system component performance standard compliance condition
Final cover
Soil monocover on side 150 mm of soil 3 months to strip vegetation and 12 months to
slopes deformation; partial regrade and recompact areas of restore vegetation
failure contained on site cracking
Landfill gas control
Collection wells Up to 25 % of wellheads 1 month to route headers around 12 months to repair/
broken broken wellheads replace broken
wellheads
Headers Up to 25 % of header pipes 1 month to bypass broken header 3 months to repair/
cracked or broken pipes replace broken
headers
Vacuum pumps Power loss; no structural None required 1 month to restore
damage off-site power
Leachate transmission Acceptable breakage of 1 month to bypass broken pipes 3 months to repair
pipes pipes with double broken pipes
containment
Surface water management
Conveyance systems Cracking and up to 2 months to completely restore 9 months to replace/
(bench channels, down 300 mm of displacement surface pathways rebuild surface
drains, culverts) pathways
Sedimentation basin Minor cracking of 2 weeks to 1 month to patch the 9 months to rebuild
concrete cracks the basin
(if needed)
Access roads 300 mm displacement 2 months to patch the cracks 12 months for full
(cracking) repair
elements of the waste containment system are tensile strength. However, a back analysis of the
modeled as structural beam elements with zero geomembrane tears observed at two separate
moment of inertia (to allow for buckling) and locations at the Chiquita Canyon Landfill (cited
with a parabolic stress–strain model, allowing previously) has suggested that damage can occur
for explicit calculation of forces and strains in to geosynthetic waste containment system ele-
the geosynthetic elements of the containment ments when the calculated Newmark displace-
system. Parametric studies using this model sug- ment is less than 150 mm and when there are S
gest that in some cases the geosynthetic elements irregularities in the containment system, e.g.,
of a waste containment system may not sustain due to strain concentrations around patches in
significant damage even though a decoupled the geomembrane at locations where coupons
Newmark analysis results in a calculated dis- were removed for destructive testing as part of
placement of greater than 150 mm. In particular, the construction quality assurance (CQA) pro-
when the interface shear strength on the top of a gram (Kavazanjian et al. 2013).
geosynthetic element is less than the interface Performance-Based Design: Figure 8 illus-
shear strength on the bottom of the geosynthetic trates the strain in the geomembrane for a landfill
element and there are no irregularities side-slope liner as predicted by the performance-
(penetrations, seams) in the element, the interface based seismic design model described by
may be able to sustain large sliding displace- Kavazanjian et al. (2012). While this model has
ments without incurring damage to the yet to be fully validated, back analysis of the
geosynthetic element, i.e., without exceeding its performance of the Chiquita Canyon Landfill in
Seismic Design of Waste Tensile Force

Fig. 8 Strain induced in a Geomembrane
side-slope liner system 3.09kN/m
geomembrane by seismic
loading (Kavazanjian 1.8% Compressive Strain
et al. 2012)
Tensile Strain
Compressive Stress
Compressive Stress
the Northridge earthquake (Kavazanjian et al. response of the waste mass and methods for eval-
2013a) indicates that strains due to the anchoring uating the integrity of containment system com-
of the geomembrane at the crest of the slope and ponents subject to seismic loading. Observations
strain concentration factors due to seams in the of landfill performance in earthquakes indicate
geomembrane (Giroud 2005) at locations where that the waste mass may amplify free-field
coupons were removed for CQA testing must be ground motions and that containment system ele-
considered in order to predict the tears observed ments, including geosynthetic liner system com-
in the liner following the earthquake. These ponents and soil cover, are susceptible to
recent findings suggest that, at a minimum, new earthquake-induced damage. Waste mass seismic
guidelines need to be developed for acceptable response depends to a large extent upon the waste
locations for removal of coupons for CQA testing mass material properties, including the small-
and for minimizing other irregularities in the strain stiffness, strain-softening behavior, and
geosynthetic elements of the containment sys- internal damping in the waste. For landfills with
tem. However, ultimately performance-based geosynthetic liner and cover systems, the inter-
design using fully coupled models that consider face and in-plane shear strength of the
interface and geosynthetic material behavior to geosynthetic components is also an important
explicitly predict strain and forces in consideration with respect to seismic response.
geosynthetic elements of modern waste contain- Most containment system performance analyses
ment systems is likely to become the standard of employed in engineering practice are decoupled
practice, at least for critical facilities in areas of analyses in which waste mass seismic response is
high seismicity. calculated using the equivalent linear method and
the performance of liner system components is
assessed based upon the calculated seismic dis-
Summary placement from a Newmark-type analysis. The
seismic performance of containment system
Seismic design of waste containment systems is components is then assessed based upon empiri-
an important consideration with respect to both cal rules for allowable calculated displacement.
maintaining the protection afforded by the con- While more sophisticated fully coupled
tainment system for human health and the envi- performance-based analyses in which the forces,
ronment and providing safe and secure capacity strains, and displacements induced by the earth-
for waste disposal during earthquake response quake in containment system components are
and recovery operations. Important engineering explicitly calculated are possible, such analyses
considerations with respect to the seismic design have yet to make their way into engineering
of waste containment systems include the seismic practice.
Cross-References Idriss IM, Fiegel G, Hudson MB, Mundy PK, Herzig

R (1995) Seismic response of the operating industries
landfill. Earthquake design and performance of solid
▶ Dynamic Soil Properties: In Situ waste landfills. Geotechnical special publication
Characterization Using Penetration Tests no. vol 54. ASCE, Reston, VA, pp 83–118
▶ Geotechnical Earthquake Engineering: Johnson ME, Lundy J, Lew M, Ray ME (1991) Investiga-
Damage Mechanism Observed tion of sanitary landfill slope performance during
strong ground motion from the Loma Prieta Earth-
▶ Nonlinear Dynamic Seismic Analysis quake of October 17, 1989. In: Proceedings of the
▶ Performance-Based Design Procedure for second international conference on recent advances
Structures with Magneto-Rheological in geotechnical earthquake engineering and soil
Dampers dynamics. University of Missouri, Rolla,
pp 1701–1706
▶ Seismic Design of Dams Kavazanjian E Jr, Matasovic N (2001) Seismic design of
▶ Site Response: 1-D Time Domain Analyses mixed and hazardous waste landfills. In: Proceedings
of the fourth international conference on recent
advances in geotechnical earthquake engineering and
soil dynamics. University of Missouri-Rolla, Rolla
References (on CD ROM)
Kavazanjian E Jr, Matasovic N, Stokoe KH II, Bray JD
Augello AJ, Matasovic N, Bray JD, Kavazanjian E Jr, (1996) In-Situ shear wave velocity of solid waste from
Seed RB (1995) Evaluation of solid waste landfill surface wave measurements. In: Proceedings of the
performance during the Northridge earthquake. In: 2nd international congress environmental geotechnics,
Earthquake design and performance of solid waste Osaka, Balkema, vol 1, pp 97–104
landfills. Geotechnical special publication no. 54. Kavazanjian E Jr, Matasovic N, Caldwell J (1998) Dam-
ASCE, Reston, VA, pp 17–50 age criteria for solid waste landfills. In: Proceedings of
Bray JD, Rathje EM (1998) Earthquake-induced displace- the sixth US national conference on earthquake engi-
ments of solid-waste landfills. J Geotech Geoenviron neering, EERI (on CD ROM)
Eng ASCE 124(3):242–253 Kavazanjian E Jr, Arab MG, Matasovic N (2012) Perfor-
Bray JD, Zekkos D, Kavazanjian E Jr, Athanasopoulos mance based design for seismic design of
GA, Reimer M (2009) Shear strength of municipal geosynthetics-lined waste containment systems. In:
solid waste. J Geotech Geoenviron Eng ASCE 135 Proceedings of the second international conference
(6):709–722 on performance-based earthquake geotechnical engi-
Buranek D, Prasad S (1991) Sanitary landfill performance neering, State of the Art – State of the Practice
during the Loma Prieta earthquake. In: Proceedings of Volume, Keynote Lecture No. 19, University of
the second international conference on recent Catania, p 23 (on CD ROM)
advances in geotechnical earthquake engineering and Kavazanjian E Jr, Arab M, Matasovic N (2013a)
soil dynamics. University of Missouri, Rolla, Performance of Two Geosynthetics-Lined Landfills
pp 1655–1660 in the Northridge Earthquake. In: Proceedings of
Dixon N, Jones DRV, WhittleRW (1999) Mechanical the seventh international conference on case
properties of household waste: in situ assessment histories in geotechnical engineering, Missouri
using pressuremeters. In: Proceedings of the Sar- University of Science and Technology, p 14
dinia’99: seventh international waste management (on CD ROM)
and landfill symposium, vol III, pp 453–460 Kavazanjian E Jr, Matasovic N, Bachus RC (2013b) S
Eid HT, Stark TD, Douglas WD, Sherry PE (2000) Munic- The 11th Peck lecture: pre-design geotechnical
ipal solid waste slope failure. 1: waste and foundation investigation for the OII superfund site landfill.
properties. J Geotech Geoenviron Eng, ASCE 126 J Geotech Geoenviron Eng. ASCE 139(10):
(5):397–407 1839–1863
EMCON (1994) Northridge earthquake seismic evalua- Koerner RL, Soong T-Y (2000) Stability assessment of ten
tion, Chiquita Canyon landfill. Technical report pre- large landfill failures. In: Advances in transportation
pared for Laidlaw Environmental Systems by EMCON and geoenvironmental systems using geosynthetics.
Associates, San Jose Geotechnical special publication no. 103. ASCE,
Giroud J P (2005) Quantification of geosynthetic behav- Reston, VA, pp 1–38
ior. Geosynt Int 12(1):2–27, Special Issue on the Matasovic N, Kavazanjian E Jr (1998) Cyclic characteri-
Giroud Lectures zation of oII landfill solid waste. J Geotech Geoenviron
Harder LS Jr (1991) Performance of earth dams during the Eng ASCE 124(3):197–210
Loma Prieta earthquake. In: Proceedings of the second Matasovic N, Kavazanjian E Jr, Anderson RL (1998)
international conference on recent advances in geo- Performance of solid waste landfills in earthquakes.
technical earthquake engineering and soil dynamics. Earthquake Spectra EERI 14(2):319–334
University of Missouri, Rolla, pp 1673–1691
2836 Seismic Event Detection
Matasovic M, Kavazanjian E Jr, De A, Dunn RJ (2005) - Introduction

CPT-based seismic stability assessment of a hazardous
waste site. Int J Soil Dynam Earth Eng 26(2):201–208
McCartney JS, Zornberg JG, Swan RH (2009) Analysis of Seismic stations record signals continuously;
a large database of GCL-Geomembrane interface however, the main interest for the user is the
shear strength results. J Geotech Geoenviron Eng recordings of seismic events. The events will
ASCE 134(2):209–223 normally be contained in a very small fraction
Mitchell JK, Seed RB, Seed HB (1990) Kettleman hills
waste landfill slope failure. I: liner‐system properties. of the actually recorded signal (Fig. 1).
J Geotech Eng ASCE 116(4):647–668 As it is seen in this 24 h seismogram, there
Orr WR, Finch MO (1990) Solid waste landfill perfor- are only a few sections with real seismic events
mance during Loma Prieta Earthquake. In: and a lot of other wiggles which is just noise.
Geotechnics of waste fills: theory and practice,
vol 1070, Special technical publication. ASTM Inter- The more or less continuous noise seen
national, Conshohocken, pp 22–30 throughout the seismogram is the earth’s back-
Ramaiah BJ (2013) Static and seismic stability assessment ground noise (see below and Figs. 2 and 4).
of municipal solid waste landfills. Ph.D dissertation, In the old days when all the signals were
Department of Civil Engineering, Indian Institute of
Technology Delhi, Hauzkhas recorded on paper (on seismograms), the events
Seed RB, Mitchell JK, Seed HB (1990) Kettleman hills were detected easily by eye, particularly for
waste landfill slope failure. II: stability analyses. small local events with high-frequency content.
J Geotech Eng 116(4):669–690 This method is sometimes also used with
Yuan P, Kavazanjian E Jr, Chen W, Seo B (2011) Com-
positional effects on the dynamic properties of munic- today’s digital recordings which can be
ipal solid waste. Waste Manag 31:2380–2390 displayed on screen or printed on paper. But
Zekkos Zeccos D P (2005) Evaluation of static and the majority of event detection is done auto-
dynamic properties of municipal solid-waste. PhD the- matically, either in real time to limit the
sis, Department of Civil and Environmental Engineer-
ing, University of California, Berkeley amount of data recorded or off-line when
Zekkos D, Bray JD, Kavazanjian E Jr, Matasovic N, large data sets have been collected, e.g., from
Rathje E, Reimer M, Stokoe KH III (2006) Unit weight field experiments. Over the years a few simple
of municipal solid waste. J Geotech Geoenviron Eng and very many sophisticated algorithms have
ASCE 132(10):1250–1261
Zekkos D, Bray JD, Riemer MF (2008) Shear modulus and been developed to automatically detect a “real”
material damping of municipal solid waste based on signal in the presence of noise. Few, if any,
large-scale triaxial testing. Can Geotech J 45:45–58 have been developed that can beat the human
Zornberg JG, McCartney JS, Swan RH Jr (2005) Analysis eye, which still is used to evaluate the effi-
of a large database of GCL internal shear strength
results. J Geotech Geoenviron Eng ASCE 131(3):1–14 ciency of the detector. The use of a trigger or
a detector can have two purposes:
1. Determine the onset of real seismic events in

Seismic Event Detection a large data set. These onset times will be used
for further processing.
Jens Havskov1 and Gerardo Alguacil2 2. Extract out the waveform time segments
1
Department of Earth Science, University of of a larger data set which contains seis-
Bergen, Bergen, Norway mic events, either in a field recorder to
2
Instituto Andaluz de Geofı́sica, University of limit the amount of data recorded or to
Granada, Granada, Spain extract out just the useful events for
processing.
Synonyms In this entry, the most common event detec-

tion methods will be described. An overview of
Earthquakes; Seismic events; Seismic signals; detection methods is found in Sharma
Seismogram; Volcanic events et al. (2010).
Seismic Event Detection 2837
OKLAHOMA GEOLOGICAL SURVEY OBSERVATORY, LEONARD, OK

DEC 19 (353), 2001 Low Frequency [0.0143-0.057 Hertz] Z [vertical] earth velocity
00
03
06
UTC/GMT HOUR
09
12
15
18
21
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58
MINUTES AFTER EACH UTC/GMT HOUR JL
Seismic Event Detection, Fig. 1 A seismogram for the distance from the station to the event in degrees, h is
a 24 h period. Detected events are indicated with the the depth of the earthquake, and M is the magnitude
name of the region where the earthquake occurred, D is (Figure from www.okgeosurvey1.gov/seis/20011219.gif)
Examples of Seismic Signals
Detecting real seismic events in a continuous data

stream can be quite a challenge, particularly if
there is nonstationary noise like noise from cars
passing by or wind gusts and in some cases it is
virtually impossible to distinguish between
man-made noise and real seismic events. S
In addition, there is an ever present so-called
microseismic background noise (see Figs. 4 and
1). This noise is caused by standing waves in the
ocean and wave action near the coasts. It is pre-
sent all over the world, strong near the coast, and
weaker far from the coast. It has a maximum
amplitude at a period of about 4–6 s. Figure 2
shows a typical curve with the noise as a function
or period.
Seismic Event Detection, Fig. 2 Typical amplitude
of the microseismic background noise as a function of
Seismic events can have quite a range of
period(s) (The data for the plot are taken from Brune and appearances and frequency content depending
Oliver 1959) on the type of seismic sensor used and the
Seismic Event Detection,

Fig. 3 Seismogram of
a local event (M = 1.5) at
5 stations. Sensors are short
period and the traces are not
filtered. Each trace is auto
scaled. The first arrival
phase P and second arrival
phase S are indicated. The
epicentral distances
(distance from the
earthquake to the station
measured along the
surface) are given above the
traces to the right (The
figure is from Havskov and
Ottemöller 2010)
Seismic Event Detection, Fig. 4 A magnitude 1.5 event recorded on a broadband station. The top trace is the original
trace and the bottom trace has been filtered from 5 to 10 Hz
distance to the event and its magnitude. Essen- microseismic noise, is the harmonic signal dom-
tially there are three types of seismic sensors: inating the record. It has a period of 4.5 s. A small
short-period measuring ground velocity high-frequency event will “drown” in this noise
(frequencies above 1 Hz), broadband measuring and can practically not be seen on the original
ground velocity (frequencies typically from 0.01 record. By filtering the signal from 5 to 10 Hz, the
to 50 Hz), and accelerometers measuring ground local event is clearly seen. This illustrates that
acceleration from 0 to 100 Hz. Some examples event triggering often will have to work on fil-
will be given of seismic events. tered traces. So how did humans detect this kind
Local small earthquake recorded with short- of event before digital recording? Simply by
period sensors (Fig. 3). The signals are high recording the signals in two different frequency
frequency, and P- and S-phases are clearly seen bands, short period above 1 Hz and long period
although the P-phase is getting close to the noise below 0.1 Hz, and thereby filtering out the noise
level for station ASK. around the microseismic peak (Fig. 2).
Small event recorded on a broadband station Distant event recorded on a broadband station
(Fig. 4). The high background noise level, the (Fig. 5). This magnitude 6.2 event is recorded at

Fig. 5 Recording of an
earthquake at a large
distance (4,200 km). The
insert above left shows
a zoom of the noise before
the P
a distance 4,200 km. It is seen that a large distant

event has much lower frequencies than small
local events, partially due the source nature of
earthquakes and partly due to filtering out the
high frequencies along the path.
The unfiltered trace shows, before the P, the
microseismic noise. For smaller events the
P might drown in the noise. The microseismic
noise is filtered away in the middle trace and the
phases appear very clear. The bottom trace is
filtered so the signal looks like it would have
been recorded on a short-period seismograph. In
this case a detector would not be able to find the
true extent of the seismic event since the long-
period S- and surface waves are not seen.
Volcanic events (Fig. 6). Volcanoes present
a special problem since the signals from the
events often are very emergent. Figure 6 shows
the most common types. S
Seismic Event Detection, Fig. 6 Major types of seis-
The tectonic event is like any other earthquake mic signals from volcanoes (From http://vulcan.wr.usgs.
and is easy to detect. The shallow events are very gov/Imgs/Gif/Monitoring/Seismic/quakes.gif)
harmonic and monochromatic and of sufficient
amplitude to be easy to detect. The surface event Trigger Methods
can be very difficult to detect automatically since
it is very emergent and might be mistaken for A trigger or detector is an algorithm that checks
noise. The harmonic tremor is a continuous series the signal for variations that could indicate an
of harmonic events and cannot be detected. event. There are several trigger algorithms avail-
Many events close in time: Figure 7 shows able, some very sophisticated using neural net-
swarm activity of small earthquakes in a 24 h works (e.g., Romeo et al. 1995) and pattern
period. The human eye can easily detect the indi- recognition (e.g., Köhler et al. 2010). In the
vidual events while an automatic trigger would hands of an expert, they can significantly improve
have problems with separating the events. the event detections/false trigger ratio,
2840
IG.AEP2..HHZ
IG.AEP2..HHZ
00:30
01:30
02:30
03:30
04:30
05:30
06:30
07:30
08:30
09:30
10:30
11:30
12:30
13:30
14:30
15:30
16:30
17:30
18:30
19:30
20:30
21:30
22:30
23:30
0 2m 4m 6m 8m 10m 12m 14m 16m 18m 20m 22m 24m 26m 28m
Seismic Event Detection, Fig. 7 Swarms of earthquakes. A 24 h record on 5 February 2013 of a temporary broadband station during the seismic swarm of Torreperogil
(Southern Spain)
Seismic Event Detection

Fig. 8 An example of level
trigger. The plot shows
a horizontal component of
the accelerogram of a small
local earthquake. The
trigger level is marked with
two horizontal lines. In this
case the de-trigger level is
the same as the trigger
level. Usually the triggers
of the three components are
combined, and an event is
declared if at least two or
three channels are triggered
within a given time window
(e.g., 1 s)
particularly for a given type of seismic events. the level trigger will also trigger on a single high
However, the sophisticated adjustments of oper- value (a spike), which obviously is not a seismic
ational parameters to the actual signal and seis- signal.
mic noise conditions, at each seismic site, which This algorithm is often used in strong-motion
these triggers require, have often been proven seismic instruments (recording large ground
unwieldy and subject to error. Hence, for practi- accelerations), where high sensitivity is not an
cal purposes, only two types of triggers are issue and where consequently man-made and
widely used, the level trigger and the STA/LTA natural seismic noise is not critical. The level
trigger (see below), and only those will be trigger is now largely replaced by the STA/LTA
described here. trigger (see next section), since computer power
The level trigger: This trigger is also called the no longer is an issue. However, many instruments
amplitude threshold trigger and simply searches still have the level option.
for any amplitude exceeding a preset threshold. The short-term average – long-term average
The start of the event is declared when this trigger (STA/LTA) is the most frequently used
threshold is reached and the end of the event is trigger algorithm. A single channel of seismic S
declared when the level reaches the de-trigger signal is typically processed as follows: The sig-
level, usually smaller than the trigger level. nal is band-pass filtered, and the absolute average
When this type of trigger is installed in STA (short-term average) over the STA time win-
a recorder, that only shall record the real events, dow is determined. Typically, the STA time win-
the recording window will be extended with a dow is 0.5 s for local earthquakes. A short STA
pre-event time before the trigger and a post- time makes the trigger more sensitive to short-
event time after the trigger (Fig. 8). This is to term variations in the signal, while a longer STA
ensure that the whole event is recorded. is better at averaging out short-term fluctuations
The level trigger has some obvious problems. and it can be compared to a low-pass filter. If
If DC (a constant voltage) is present in the signal there are spikes in the signal, a long STA must
and not removed before triggering, it can prevent be used in order to average out the interference.
triggering or keep the instrument in continuous This will of course reduce the sensitivity to short-
trigger mode. If no averaging or filtering is done, lasting signals, but if a longer-lasting signal is the
objective, STA can be as long as the duration of filtering, which is desirable also for other reasons
the main P-wave train. The most common mis- (see below). The STA/LTA trigger algorithm is
take in setting the STA is to use a too short value. well suited to cope with fluctuations of natural
The same filtered signal is also used to calcu- seismic noise, which are slow in nature. It is less
late the LTA (long-term average) over the LTA effective in situations where man-made or natural
time window, which is typically 50–500 s. Thus, seismic noise of a bursting or spiky nature (e.g.,
LTA will give the long-term background signal wind gusts) is present. At sites with high, irregu-
level, while the STA will respond to short-term lar seismic noise, the STA/LTA trigger usually
signal variations. The ratio between STA and LTA does not function well, meaning there are too
is constantly monitored, and once it exceeds many false triggers.
a given threshold, the trigger level, the start of A typical implementation will now be
an event is declared for that trace. The LTA described. The trigger must be running continu-
should be short enough to adapt to slow changes ously in the computer, so the STA and LTA values
of the background noise and long enough to avoid are calculated as running averages:
reducing the sensitivity to triggering on
low-amplitude emergent signals. This can be jxi j STAi1
STAi ¼ STAi1 þ
a problem with very emergent signal where the NSTA
STA might increase fast enough to prevent trig-
gering (see the surface event in Fig. 6). Since the jxi j LTAi1
LTAi ¼ LTAi1 þ (1)
natural background noise usually changes very NLTA
slowly, it is generally better to use a too long
than a too short LTA time. If LTA is not STAi
Ri ¼
frozen during triggering (see next section), the LTAi
LTA time must be at least as long as the event
duration to prevent premature de-triggering. where xi is the signal (filtered or unfiltered); STA,
Using a too short LTA time combined with the short-term average; LTA, the long-term aver-
a high trigger ratio might even prevent triggering age; R, the STA-LTA ratio; and NSTA and NLTA,
completely since the LTA is adjusted upward the number of points in the STA and LTA win-
so fast that STA/LTA never exceeds the trigger dows, respectively. In this case, x is the unfiltered
ratio. signal. This works if the signal has no DC com-
Once the event starts, the LTA is usually fro- ponent, but if x has a large DC component, this
zen, so that the reference level is not affected by will be transferred to the STA and LTA and they
the event signal. The end of the event is declared might be so large that R never reaches the trigger
when the STA/LTA ratio reaches the de-trigger ratio.
level. Trigger levels and de-trigger levels of 4.0 There are two common variations of the trig-
and 2.0, respectively, are typical values. In order ger algorithm Eq. 1. One is to use the squared xi
to get the complete event, the recording will start instead of the absolute value. Using the squared
pre-event time before the trigger time. Likewise, xi, makes the trigger more sensitive to changes.
to avoid the truncation of the signal, recording The other is the Allen trigger (Allen 1978), in
continues some time after the de-triggering, the which the absolute value of xi is replaced by the
post-event time. The maximum time of recording characteristic function ci. In a simplified form, ci
can usually be limited by a maximum recording is calculated as
time setting, and it is also sometimes possible to
discard triggers lasting less than a certain time, ci ¼ x2i þ k ðxi xi1 Þ2 (2)
the minimum trigger time.
STA and LTA must be calculated on signals where k is a constant. The second term is propor-
without a DC component in order to reflect the tional to the squared first time derivative of the
real signal. The DC is normally removed by signal. The characteristic function is sensitive to

Fig. 9 Below:
A spectrogram of the signal
plotted above, with three
local earthquakes closely
spaced in time. It is
apparent that the amplitude
of high frequencies decays
faster than lower
frequencies after the first
arrival of each event. The
colors indicate spectral
amplitudes
both amplitude variations and frequency and its onset time. So the human analyst would
changes. have no problem detecting the events in Fig. 7.
All the above described trigger methods have Figure 10 shows an example of how the algo-
in common that they are not-looking-forward rithm STA/LTA works for the standard trigger
algorithms, i.e., the present value of the trigger using the absolute value.
parameters depends only on present and past The original signal is a small high-frequency
values of the signal. This makes them easy to earthquake with a medium-level microseismic
implement with simple recursive, real-time background noise, which is reflected in the slow
operations. variations of the STA and more smoothed in the
If the seismic events are very close in time LTA. At about time 12 s, the ratio gets above the
(Fig. 7), it is obvious that the STA/LTA would threshold of 3 (1. trigger). It immediately gets
produce many failures. If one event is not finished below again, but since the de-trigger ratio is
before the next event starts, the first event would 2, the trigger remains activated. In the above
simply be prolonged so correct start times of example, the LTA is frozen once an event has
events would only be obtained when a new trig- been declared, which can be seen from the flat
ger occurs after the previous has finished. A more LTA level. A few seconds after the de-triggering,
sophisticated trigger might also monitor the fre- the ratio again rises above the trigger level and
quency content of the signal. Since the end of the remains in trigger state for about 1 s (2. trigger). S
event usually has a lower-frequency content than This signal has a good signal-to-noise ratio, so
the initial P-wave, this could be used to declare why has the trigger not been activated much
the start of a new event before the previous event longer? The first reason is simply that the DC
is finished. This can be illustrated with has not been removed before calculating STA
a spectrogram which is a time-frequency repre- and LTA. Since STA and LTA before triggering
sentation of the spectral content of a signal. are about 220, the event will de-trigger when
Figure 9 shows an example with three events STA is 440 and the maximum ratio is 7 (Fig. 10
occurring within 70 s. and Table 1). The DC of this signal is only 200 as
A human analyst works in a quite different compared to the maximum value of 3,400; yet if
way: He detects an event looking at the entire it is not removed, it will have a large influence on
record and evaluates the past, present, and subse- the trigger performance.
quent values of the trace (or something like the When the DC is removed, the figure
envelope of it) to decide the presence of an event corresponding to Fig. 10 will look almost the

Fig. 10 The seismic
trigger. Top trace is the
original signal; the second
trace, the STA; the third
trace, the LTA; and the last
trace, the STA/LTA ratio.
The STA window is 0.4 s
and the LTA window is 50 s.
The trigger ratio is 3 and the
de-trigger ratio 2. All
amplitudes are counts (the
raw number generated by
the recorder) (Figure from
Havskov and Alguacil
(2010))
Seismic Event Detection, Table 1 Effect of DC and This is particularly important with broadband
filter on STA, LTA, and STA/LTA. The units of the seismometers, where small earthquake signals
numbers are in counts
often are buried in dominant 0.2–0.3 Hz seismic
DC not DC Filtered noise and a trigger without a filter simply would
removed removed 5–10 Hz
not work except for the largest events (Fig. 4).
STA and LTA 220 40 3
before trigger Some recorders allow several trigger parameter
Maximum STA 1,500 1,450 600 sets to be used simultaneously. This is needed if,
Maximum 7 36 200 e.g., a broadband station has to trigger on micro-
STA/LTA earthquakes, teleseismic P-waves, and surface
Duration of 15 50 105 waves which each requires separate filters, STA,
trigger (s) and LTA.
For more detail of how to set and use
STA/LTA triggers, see Trnkoczy, information
same, but since both noise, STA and LTA, are sheet 8.1 in New Manual of Seismological Obser-
smaller due to the removal of the DC, STA/LTA vatory Practice (NMSOP) (Bormann 2012).
will be much higher for the earthquake signal,
and the event will remain in trigger state for 50 s Improved Triggering Using Several Stations,
instead of 15 s (Table 1). The performance can be the Network Trigger
further improved by filtering. The signal has Whatever trigger method is used with single
a typical microseismic background noise channels or single three-component stations,
superimposed, and this gives the main contribu- there is always likely to be quite a lot of false
tion to the STA and LTA before triggering. By triggers, all depending on the local conditions and
using a band-pass filter from 5 to 10 Hz, the how sensitive the trigger has been set up. When
performance is further improved (Table 1) and several stations within a single network are
the trigger lasts for 105 s. processed together, the number of false triggers
So, the conclusion is that every trigger must can be significantly reduced setting up
have an adjustable band-pass filter in front of the a requirement of concurrent triggering. This
trigger algorithm, which has both the function of means that for a network trigger to be declared,
removing the DC and making the trigger algo- there must be a minimum number of triggers
rithm sensitive to the frequency band of interest. within a given time window, the array

Fig. 11 A volcanic event
recorded on an array of
vertical sensors. The
bottom trace (black) is the
sum of the individual
channels (color), once they
have been time aligned
using the P-wave relative
propagation times. The
improvement on the signal-
to-noise ratio is apparent
propagation window. The array propagation win- will be enhanced and the supposedly incoherent
dow is calculated as the time the waves take to noise will cancel out. Running a seismic trigger
propagate across the network. If, e.g., the net- algorithm on the summed signal will therefore be
work is 100 km in diameter and assuming the more sensitive and more reliable than on any of
case that triggering takes place on the S-waves the single stations. In a way this is similar to the
only and the event is outside the network, then the network trigger where the same event must be
array propagation window should be 100/Vs present in all channels. An example is shown in
where Vs is the S-wave velocity. If some stations Fig. 11.
trigger on P and others on S, the array propaga-
tion window should be even larger by the S-P Triggering Using Three Channels
time so in general the array propagation time is Most seismic stations today have three channel
set to a value larger than 100/Vs. In case there is sensors oriented vertically, north–south and
swarm activity (Fig. 7), a network trigger might east–west. As a first small improvement in the
not help a lot since the events come so close in triggering, it can be required that all three chan-
time that it would be hard to associate events, that nels trigger on the seismic event. Since external S
is, if the events are correctly triggered in the first disturbances affect all three channels, this
place. requirement might only improve the detection
marginally. With three channels, it is possible to
Triggering Using a Seismic Array make a polarization analysis or in other words
A seismic array consists of multiple sensors in determine the particle motion of the ground (see
a small enough area that signals are correlated, Fig. 12).
and the seismic array is therefore also sometimes For simplicity, the polarization is only shown
called a seismic antenna. in the north-vertical plane. It is seen that the
The signal-to-noise ratio of a seismic signal is P-wave is not polarized in the noise and highly
improved by summing the signals of each station polarized in the P-wave signal. By continuously
in the array using a delay corresponding to ray calculating the degree of polarization, it is possi-
travel time across the array so the signals get ble to distinguish a P-wave from background
lined up. In this way the coherent seismic signal noise which would have no polarization.
This property has been used to improve the detec-

tion of P- and S-waves (Roberts et al. 1989; With-
ers et al. 1998).
Triggering Using a Known Waveform

Many seismic events have very similar wave-
forms if they originate from the same or nearby
sites. This could also be the case for volcanic
events even though they might not be from
exactly the same area. A “known” waveform
shape can then be used to detect unknown arrivals
by cross correlating the known wavelet with the
signal of unknown events. An example of similar
events is seen in Fig. 13. Despite the large dis-
tances, the signal shapes are nearly identical for
the two events.
The only trigger parameter would then be
the minimum correlation needed to declare an
event. Using this technique, it is often possible
Seismic Event Detection, Fig. 12 Top: the three com- to detect events which otherwise could not be
ponents of ground velocity for a local earthquake. Bottom:
detected. When using a seismic array, this
the particle motion in the north-vertical plane at two different
times as indicated on the figure. To the left is the analysis of method is particularly powerful (Gibbons and
the noise and to the right the analysis of the P-wave signal Ringdal 2006).
Seismic Event Detection, Fig. 13 Left: the first couple The top trace event has magnitude 1.8 and the bottom trace
of seconds of the P-waves for the two events occurring in event magnitude 1.6 (Figure from Havskov and
the same area as recorded with three different stations. Ottemöller (2010))
Seismic Fragility Analysis 2847
Summary Romeo G, Mele F, Morelli A (1995) Neural networks and

discrimination of seismic signals. Comput Geosci
21:279–288
Today, huge quantities of continuous seismic data Sharma BK, Kumar A, Murthy VM (2010) Evaluation of
are recorded, and automatic detection of the seis- seismic event detection algorithms. J Geol Soc India
mic events is indispensable. Over the years very 75:533–538
many algorithms have been developed to detect Withers M, Aster R, Young C, Beiriger J, Harris M,
Moore S, Trujillo J (1998) A comparison of select
“real” seismic events; however the overall most trigger algorithms for automated global seismic
used and reliable detectors are based on the stan- phase and event detection. Bull Seismol Soc Am
dard STA/LTA method. Combined with a network 88:95–106
trigger, this becomes a simple and robust event
detector, in particular considering that the number
of stations is continuously increasing so more use
of the network trigger can be done, making the
requirement for the single-channel detector less Seismic Fragility Analysis
stringent.
Murat Altug Erberik
Civil Engineering Department, Middle East
Cross-References Technical University, Ankara, Turkey
▶ Passive Seismometers
▶ Principles of Broadband Seismometry Synonyms
▶ Seismometer Arrays
▶ Seismic Noise Damage probability matrix; Fragility curve; Fra-
gility surface; Limit state; Numerical simulation;
Seismic capacity; Seismic demand; Uncertainty
References
Allen RV (1978) Automatic earthquake recognition and Introduction

timing from single traces. Bull Seismol Soc Am
68:1521–1532
Bormann, P. (ed) (2012) New Manual of Seismological
Seismic fragility can be defined as the proneness
Observatory Practice (NMSOP-2), IASPEI, GFZ Ger- of a structural component or a system to fail to
man Research Centre for Geosciences, Potsdam; perform satisfactorily under a predefined limit
nmsop.gfz-potsdam.de state when subjected to an extensive range of
Brune JN, Oliver J (1959) The seismic noise at the earth’s
surface. Bull Seismol Soc Am 49:349–353
seismic action. In accordance with the above
Gibbons SJ, Ringdal F (2006) The detection of low mag- definition, seismic fragility analysis can be S
nitude seismic events using array-based waveform regarded as a probabilistic measure for seismic
correlation. Geophys J Int 165:149–66 performance assessment of structural compo-
Havskov J, Alguacil G (2010) Instrumentation in earth-
quake seismology, 3rd edn. Springer, Dordrecht,
nents or systems. There are two different end
358 pp products of seismic fragility analysis: damage
Havskov J, Ottemöller L (2010) Routine data probability matrix and fragility curve.
processing in earthquake seismology. Springer, Damage probability matrix (DPM) is a table
Dordrecht, 347 pp
Köhler A, Ohrnberger M, Scherbaum F (2010)
that provides discrete values of damage state
Unsupervised pattern recognition in continuous seis- probabilities for specified levels of ground
mic wavefield records using self-organizing maps. motion intensities. Each column of DPM stands
Geophys J Int 182:1619–1630 for a constant level of ground motion intensity
Roberts RG, Christofferson A, Cassedy F (1989) Real
time event detection, phase identification and source
whereas each row of DPM denotes the probabil-
location using single station 3 component seismic data ity of being in a predefined damage state. Hence,
and a small PC. Geophys J 97:471–480 any element of DPM represents the probability of
2848 Seismic Fragility Analysis
experiencing a certain damage state for a specific ground motion intensity parameters can be used
level of ground motion intensity. in the generation of fragility curves. As seen from
Fragility curves are continuous functions that the figure, four limit states (LSs) have been
represent the probability of exceeding predefined assigned, which in turn means that there exist
limit (or performance) states for specific levels of five damage states (DSs). Attainment of limit
ground motion intensity. Fragility curves can be states is an important phase of fragility curve
developed for structural and nonstructural com- generation and is discussed in following sections.
ponents, or they can be generated for structural The columns of DPM are obtained by intersecting
systems and assemblies. In the field of earthquake the fragility curve set with vertical lines (dashed
engineering, the most common application is to lines in Fig. 1) at specific intensity levels and
generate fragility curves for building structures. calculating the damage state probabilities,
Depending on the characteristics of the building which are corresponding portions between any
stock in the region of interest, the fragility curves two limit states in these vertical alignments.
can be derived for many different structural types
(i.e., steel, reinforced concrete, or masonry build-
ings; low-, mid-, or high-rise construction; frame, Different Approaches for Seismic
wall, or mixed structures; etc.). Other than build- Fragility
ing structures, fragility curves have also been
generated for infrastructures (like bridges, There are four main resources to generate the
dams), lifeline systems (like transportation, seismic fragility of a structural system or compo-
power, water supply networks), unrestrained nent: expert opinion, field data, experimental
equipment, etc. data, and analytical simulations.
Fragility curves can be used for different pur- One of these main data sources for the deriva-
poses. The main purpose is to determine the seis- tion of fragility curves is based on the opinion of
mic performance of new or existing structural the experts from the field of earthquake engineer-
systems. In the former case, the fragility informa- ing that are invited to make a prediction about the
tion is obtained for optimal design and in the probable damage state of a structure or a class of
latter case for condition assessment. Fragility structures under different levels of seismic
curves can also be employed in order to assess action. The fragility curves generated by this
the efficiency of different intervention techniques judgment-based data are called as “judgmental
on existing structural systems. When the curves fragility curves.” Until recently, expert opinion
are derived to represent a certain class of struc- was the main source to generate fragility infor-
tural systems, they can also be used for regional mation especially for a wide range of structures
seismic damage and loss estimation studies. Such and performance limits in the United States by
studies are very popular nowadays since they can the introduction of the ATC-13 document
be employed for pre-earthquake mitigation (Applied Technology Council 1985) and
efforts and postearthquake decision-making HAZUS earthquake loss estimation methodology
processes. (National Institute of Building Sciences 1999).
There is a close relationship between fragility However, the main issue that impairs the use of
curves and DPMs so that the information expert opinion is the subjectivity and bias of the
obtained from fragility curves can be converted collected data since it depends on the unique
to construct DPMs or vice versa. Figure 1 dem- experience of the experts involved and it is very
onstrates how to generate DPM from a given set difficult to quantify this subjectivity or eliminate
of fragility curves. The vertical axis, simply the bias.
named as “probability,” actually stands for the Fragility curves can also be derived by using
probability of exceeding a limit state. The hori- the structural damage data obtained from field
zontal axis represents the complete range of observations after earthquakes. The fragility
ground motion intensity levels (ILs). Different curves obtained by this approach are called as
Seismic Fragility Analysis, Fig. 1 Conversion from a set of fragility curves to DPM (Prepared by using the data from
Erberik and Elnashai 2004)
“empirical fragility curves.” Although it is diffi- issue is the nonuniformity of field data in terms
cult to construct empirical fragility curves from of ground motion intensities and damage state
field observations, the obtained fragility curves definitions since it is being collected from many
are invaluable in the sense that they reflect the different resources. In this case, it becomes cru-
exact behavior of a structure or a class of struc- cial to convert all field data to a standard format S
tures. The difficulty comes from the fact that for in order to use in the generation of empirical
empirical fragility curve generation, an extensive fragility curves.
field data from many earthquakes with different Experimental data can be employed in the
magnitudes should be gathered. For instance, generation of fragility curves by using a similar
Rossetto and Elnashai (2003) gathered a huge approach as in the case of empirical fragility
database that is composed of 340,000 buildings curve generation based on field data. However,
inspected after 19 different earthquakes and loca- in order to obtain sufficient data for fragility
tions in order to construct the empirical fragility curve generation, it may be required to conduct
curves for European-type reinforced concrete a large number of laboratory tests with different
frame buildings. Furthermore, it is not an easy levels of loading intensities and material proper-
task to estimate the spatial distribution of earth- ties. Hence, this way of generating fragility
quake intensity at different locations where the curves is generally an expensive solution and
buildings under consideration reside. Another therefore not popular when compared to the
generation of fragility curves based on analytical beyond the elastic range of response, like
simulations. unreinforced masonry structures. The best
The most commonly used data source for method is always nonlinear dynamic approach,
quantification of seismic fragility comes from but in practice, this may not be the case due to
the analytical simulations since most of the issues computational difficulties as discussed above.
encountered in the case of judgmental, empirical,
and experimental-based approaches are handled
by the use of numerical data obtained through Components of Seismic Fragility
analytical modeling. It is possible to generate Analysis
response statistics for any type of structure or
class of structures subjected to a wide range of Basically, seismic fragility analysis is the com-
ground motion intensities. However, in this case, parison of seismic capacity and demand and to
there are two main issues: computational effort estimate whether the seismic capacity is
and limitations in analytical modeling due to exceeded for a well-defined performance level
idealization of actual structures. Although the when the structural system is subjected to speci-
computing technology is very powerful nowa- fied levels of ground motion intensity. Due to the
days, generation of analytical fragility curves probabilistic nature of seismic fragility analysis,
can take considerable time and effort depending both seismic capacity and demand are defined by
on the level of detail in analytical modeling. In probability functions in terms of certain random
some cases, it can even become unfeasible such variables to quantify all the uncertainties
as when it is required to run a dynamic analysis involved in the process.
thousands of times on a detailed finite element In the field of earthquake engineering, there is
model of a special structure. Hence, this means no consensus on the best method for seismic
that it is inevitable to introduce some simplifica- fragility analysis. However, the general frame-
tions and idealizations to the analytical model. work of seismic fragility analysis can be more
But then, the issues arising from limitations of or less presented as shown in Fig. 2. Each element
analytical modeling begin to play an important of the flowchart defines a different component of
role in seismic fragility analysis. This may cause seismic fragility analysis procedure. The ele-
the deviation of the response of the analytical ments enclosed by boxes with dashed lines can
model from the actual behavior of the structure be regarded as a single component in the analysis.
under concern. As a result, analytical approach to Each component of seismic fragility analysis is
quantify seismic fragility always possesses a explained in detail in the following paragraphs.
trade-off between accuracy and computational The first component is the development of the
effort. In this regard, there are different alterna- analytical model and generation of structural
tives to be used as the analysis method for seismic simulations to account for structural variability.
fragility analysis. These can be listed as linear At this stage, the major challenge is the selection
static, linear dynamic, nonlinear static, and of the analytical model. Depending on the level of
nonlinear dynamic methods from the simplest to idealization and simplification, a wide range of
the most complex approach. Among these, linear models from equivalent single-degree-of-
static method is not generally used in seismic freedom (SDOF) systems to detailed finite ele-
fragility analysis due to the dynamic nature of ment (FE) meshes can be employed in order to
the actual response. However, nonlinear static carry out the structural simulations (Fig. 3). Since
analysis is an alternative when it is difficult to a huge number of analyses are required to con-
carry out series of dynamic analyses to generate struct the response statistics and the fragility
fragility functions. Linear dynamic analysis can curves, the use of a simple model (like a SDOF
sometimes be used, especially when the struc- model) becomes very advantageous in terms of
tural system under concern exhibits brittle behav- computational effort. However, it should be
ior under seismic action and does not go far noted that such a simple model can only realize
Seismic Fragility Analysis, Fig. 2 General framework of seismic fragility analysis
Seismic Fragility Analysis, Fig. 3 Different levels of idealization for analytical modeling of structural systems from
the simplest to the most sophisticated
the global behavior of the structural system to if they are dynamic in nature) on this finite ele-
some extent and it is blind to local structural ment mesh within a reasonable period of time.
characteristics or construction details. Hence, Briefly, the problem of selecting the appropriate
such models are generally used to generate the modeling strategy is multidimensional, and one S
fragility curves of a class of structures (for has to make the decision by considering all the
instance, mid-rise reinforced concrete frame pros and cons.
structures) in order to predict regional damage After the selection of the analytical model, the
or loss. On the other hand, it is not possible to next step is to generate the structural simulations.
use simplified models for special structures, in Due to the probabilistic nature of seismic fragility
which the structural and construction details are analysis, some of the major structural parameters
important and should be reflected in the seismic within the analytical model are considered as
fragility analysis. A good example for this case is random variables with appropriate probability
historical masonry structures, which are unique density functions assigned to them. Normal or
and possess their own structural characteristics. lognormal distributions are commonly used for
In such cases, finite element mesh modeling convenience. These can be mechanical properties
seems to be the best solution, but then the prob- like stiffness or strength to account for the mate-
lem is to carry out numerous analyses (especially rial variability or geometric properties like
length, height, or cross-sectional dimensions. In earthquake ground motions. Descriptive parame-

order to generate the population of analytical ters like Modified Mercalli Intensity (MMI) had
simulations, a sampling method is required. The been used in the past, but they are no more pop-
most popular method is Monte Carlo sampling ular due to the fact that they are judgmental and
(Metropolis and Ulam 1949). It is a robust and a subjective measures. In addition, they do not
straightforward method to generate structural correlate well with damage. The most commonly
simulations, but the disadvantage is that it used group of parameters are single-valued
requires a very large sample size, which makes (or peak-valued) parameters like peak ground
the method unfeasible to use in the case of acceleration (PGA) and peak ground velocity
detailed and complex analytical models. This (PGV). These parameters can be easily obtained
shortcoming opens a door to constrained sam- from the ground motion time history. However,
pling methods to reduce the sample size, like single-valued parameters are insensitive to fre-
the Latin hypercube model (McKay et al. 1979). quency content and duration of the ground
Since dynamic analysis is an indispensable motion record, which is extremely important for
tool for seismic fragility analysis, characteriza- obtaining the seismic performance of a structural
tion of ground motion intensity becomes an system in some cases. To overcome the short-
important component to generate fragility infor- comings of single-valued parameters, spectral
mation. To conduct dynamic analyses, there are parameters like spectral acceleration (Sa) and
two options: to employ (1) actual or (2) synthetic spectral displacement (Sd) can also be employed
ground motion records. If it is possible to find a since these are enhanced parameters which are
sufficiently large number of records from differ- functions of both the ground motion characteris-
ent earthquakes in the region of interest, then the tics and SDOF system properties. Although rare,
first option seems to be the best solution to reflect there are also cases where involved parameters
the effect of regional seismicity and local geo- like effective peak acceleration, Housner’s spec-
logic characteristics. Ground motion records are tral intensity, and input energy have been used as
generally selected to cover the whole range of the intensity parameter in seismic fragility
intensity levels. However, this may induce the analysis.
need to scale ground motion records to fill the After generating the structural simulations and
gaps of seismic intensity levels, especially for selecting the ground motion records, the next step
high intensities. Scaling ground motion records is to carry out a large number of analyses in order
is a different research topic in earthquake engi- to build up the response statistics. These are
neering, and there are many different methods scattered plots in which the vertical axis stands
that one can employ. The second option, using for a response parameter like displacement, drift,
synthetic ground motion records, becomes cru- or force, whereas the horizontal axis represents
cial if there exists an insufficient number of actual the ground motion intensity parameter (see
ground motion records in the study region. Gen- Fig. 4a). All the uncertainty and the variability
erating synthetic ground motion records is arising from the ground motion characterization
another huge area of research in earthquake engi- and the analytical simulation is reflected in this
neering, and there are many different approaches plot. Although scattered, the general trend in this
in the literature. However, it should be noted that plot is that as the ground motion intensity
in order to generate reliable synthetic ground increases, seismic demand characterized by the
motion records in a region, geologic and seismo- response parameter also increases.
logical parameters should be well studied and The response statistics can be employed to
known beforehand. generate fragility curves if some limit states are
Another challenge in seismic fragility analysis defined as a function of the response parameter
is the selection of the ground motion intensity under consideration. This is another important
parameter since it is difficult to determine a stage in seismic fragility analysis because the
unique parameter that fully characterizes definition of a limit state directly affects the
Seismic Fragility Analysis, Fig. 4 Schematic represen- specific ground motion intensity, (c) discrete fragility
tation of the fragility curve generation procedure information for a certain limit state, (d) continuous fragil-
(Adopted from Ugurhan et al. 2011) (a) response ity function
statistics, (b) probability of exceeding a limit state at a
corresponding fragility curve. There are many FEMA-273 document (Federal Emergency Man-
different ways to attain the limit states. Limit agement Agency 1997), three limit states were
state values can be directly obtained from the defined as “Immediate Occupancy,” “Life
previous studies in the literature if the structural Safety,” and “Collapse Prevention” in terms of
type under consideration had been studied before. interstory drift. The limit states are generally
However, in most of the cases, the local and considered as deterministic parameters. How- S
special characteristics of the structural system ever, there are cases in literature in which limit
requires the attainment of new limit states. states are also considered in a probabilistic man-
In this case, the best method is to rely on field ner to account for the variability in seismic
observations or experimental findings if any. capacity.
In the absence of such data, analytical approaches Each vertical line of scattered data in response
can be used to determine the limit states. For statistics stands for a specific ground motion
instance, nonlinear static analysis, in particular intensity level with its own statistical distribu-
pushover method, is a good candidate for this tion. If the limit states are also added to the
task. The instantaneous capacity of the structural same plot, then it becomes possible to determine
system and the progression of damage can be the probabilities of exceeding the limit states at
monitored in this type of analysis. specific ground motion intensity levels. This is
Generally, two or more limit states are defined illustrated in Fig. 4b, in which LSi represents the
in seismic fragility analysis. For instance, in ith limit state and GMIj represents the jth intensity
level. Hence, the shaded area in the given distri- increased until collapse at each run by using
bution denotes the probability of exceeding the ith scaling methods (Vamvatsikos and Cornell
limit state at the jth ground motion intensity level, 2002). Hence, it becomes possible to monitor
i.e., P[D LSi|GMIj]. The value obtained repre- the complete system response under seismic
sents one of the data points in Fig. 4c. If this action from linear elastic behavior to collapse
process is repeated for different intensity levels by using IDA approach, which makes it a good
for the whole range, the discrete fragility infor- candidate for seismic fragility analysis.
mation shown in Fig. 4c can be obtained for a In literature, there are two different nonlinear
certain limit state. Then, repeating the process for static procedures that have replaced dynamic
different limit states reveals a set of fragility analyses in the generation of fragility curves,
information. The last step is to convert this dis- named as capacity spectrum method (CSM) and
crete information into a continuous function by displacement coefficient method (DCM). The
curve fitting. The continuous fragility functions CSM was first developed by Freeman et al.
are generally represented by cumulative lognor- (1975) and then became more publicized thanks
mal distribution (Fig. 4d), although other forms to the introduced in the ATC-40 document
of equations (like exponential function) have also (Applied Technology Council 1996). It has been
been used in the literature. The shape of very popular since then in performance based
the fragility curve is dependent on the overall earthquake engineering. The CSM method is
uncertainties involved in the generation process. based on the idea to compare a nonlinear capacity
Less variability means a steeper fragility curve. curve (obtained from a pushover analysis) in
In the limit, the fragility curve with no uncer- acceleration-displacement response spectrum
tainty (i.e., deterministic case) turns out to be a (ADRS) format with an elastic demand spectrum
step function. that is reduced to account for equivalent
damping. The intersection point is obtained by
trial and error and it is called as “performance
Additional Information on Seismic point”. The second method, DCM, was first intro-
Fragility Analysis duced by the FEMA-273 document (Federal
Emergency Management Agency 1997). Just
The topic of seismic fragility analysis has been a like the CSM, it is based on constructing the
very popular and active research area in the field pushover curve of the structural system under
of earthquake engineering for the last two concern. Then, this information is converted
decades. Hence, there are many additional issues into equivalent SDOF response. The final step is
to be discussed related to this topic. Some of these to modify the linear elastic response of this equiv-
are mentioned in the following subsections. alent SDOF system by the help of some empirical
coefficients that account for MDOF response,
Different Analysis Approaches in the inelastic behavior, degradation, and P-D effect.
Literature The final product is an estimate of the maximum
In the previous section, a fragility curve genera- displacement of the structural system, called as
tion procedure that makes use of time history “the target displacement.” By using both
analyses was introduced. However, it should be methods, it is much easier to generate fragility
stated that this is not the only alternative to gen- curves since it is only a matter of repeating the
erate fragility curves. For instance, instead of pushover analysis rather than carrying out com-
using time history analysis, incremental dynamic plicated nonlinear time history analyses.
analysis (IDA) can also be used for the generation There is also a different class of analysis
of fragility curves. In this approach, the structural methods, called as response surface methods,
system under consideration is subjected to a which are employed to reduce the large number
series of nonlinear time history analyses, in of computations in seismic fragility analysis. In
which the intensity of the ground motion is these methods, a response function is generated
in terms of the major random variables (ground Combined Fragility Curves

motion intensity parameter, geometric or Fragility curves for individual structures can be
mechanical properties) to predict the seismic combined together to yield the fragility informa-
response of structural systems without the need tion of a class of structures with similar proper-
to carry out complex dynamic analyses. The other ties. This is especially required for regional
advantage of these methods comes from the fact damage and loss estimation, to obtain the fragil-
that both seismic demand and capacity can be ities of different construction types such as
directly incorporated with their inherent joint reinforced concrete frame structures,
probability distributions in the response surface, unreinforced masonry structures, etc. In this
which can only be achieved by making qualita- case, the combined fragility curve can be
tive physical assumptions in the other analysis obtained by using (Shinozuka et al. 2000)
approaches.
XM
FC ð X Þ ¼ i¼1
Pi Fi ð X Þ (1)
Quantification of Uncertainty in Seismic
Fragility Analysis
where
Seismic fragility analysis requires the treatment
of uncertainties involved in the process since both
Ni
seismic demand and capacity possess different Pi ¼ X M (2)
sources of uncertainty. The uncertainties may i¼1
Ni
either arise from inherent randomness in nature
(aleatoric uncertainty), which cannot be reduced, In these equations, FC stands for the combined
or from lack of knowledge or data (epistemic fragility, whereas Fi represents the fragility curve
uncertainty). From another point of view, the of the ith individual structure and X denotes any
uncertainties in seismic fragility analysis may ground motion intensity parameter used in the
belong to ground motion characteristics and the generation of the fragility curve. Parameter Ni is
demand on the structural system or the structural the number of structures that the fragility infor-
capacity under seismic action, performance level mation Fi belongs to, and the summation in the
identification, and the limitations in analytical denominator of Eq. 2 represents the total number
modeling. All sources of uncertainty should be of structures in that specific class for which the
quantified and reduced (if possible) in the gener- information will be combined.
ation of fragility curves.
Most of the researchers have stated that the Fragility Surfaces
uncertainties arising from ground motion charac- In some cases, it becomes difficult to obtain the
teristics outweigh the uncertainties due to struc- seismic fragility of a structural system by using a
tural system characteristics and analytical single ground motion intensity parameter since S
modeling. But, of course, this may not be always there is no such unique parameter that is perfectly
the case, depending on the specific properties of correlated with seismic damage as mentioned
seismic action and the structural system under above. Then, it is possible to use two different
concern. ground motion intensity parameters together in
The reliability of the generated fragility curves the development of fragility information, which
can be tested by plotting the confidence intervals is called as fragility surface rather than fragility
for selected levels of confidence with upper and curve. In this case, there are two horizontal axes
lower bounds. If the uncertainties arising from that define the ground motion intensity and one
the components of seismic fragility analysis are vertical axis that defines the probability of
significant, the confidence intervals will eventu- exceeding a certain limit state. This means the
ally be large, indicating a poor estimate of per- response statistics obtained can be plotted as a
formance that is represented by the fragility surface, not a curve. The dual ground motion
information. intensity parameters for fragility surfaces can be
2856 Seismic Fragility of Aging Offshore Platforms
selected as magnitude and distance, PGA and Applied Technology Council (1996) Report ATC-40:
PGV, spectral acceleration or displacement at seismic evaluation and retrofit of concrete buildings.
Applied Technology Council, Redwood City
two different vibration periods of the structural Erberik MA, Elnashai A (2004) Fragility analysis of flat-
system (for instance, the first and the second slab structures. Eng Struct 26:937–948
vibration mode), etc. Federal Emergency Management Agency (1997) NEHRP
guidelines for the seismic rehabilitation of buildings.
FEMA Publication 273. Federal Emergency Manage-
ment Agency, Washington, DC
Summary Freeman SA, Nicoletti JP, Tyrell JV (1975) Evaluations of
existing buildings for seismic risk – a case study of
Seismic fragility analysis can be regarded as a Puget Sound Naval Shipyard, Bremerton, Washington.
In: Proceedings of the US National conference on
probabilistic measure for seismic performance earthquake engineering, Berkeley, pp 113–122
assessment of structural components or systems. McKay MD, Conover WJ, Beckman RJ
This analysis concept is very popular since it can (1979) A comparison of three methods for selecting
be used in different research areas of earthquake values of input variables in the analysis of output from
a computer code. Technometrics 221:239–245
engineering. Due to its probabilistic nature, it Metropolis N, Ulam S (1949) The Monte Carlo method.
requires the use of random variables to charac- J Am Stat Assoc 44(247):335–341
terize both seismic demand and capacity. In addi- National Institute of Building Sciences (1999) HAZUS
tion to this, there exist challenges such as the technical manual. Prepared for Federal Emergency
Management Agency, Washington, DC
idealization of structure through an analytical Rossetto T, Elnashai A (2003) Derivation of vulnerability
model, selection of the ground motion intensity functions for European-type RC structures based on
parameters and the analysis approach, attainment observational data. Eng Struct 25:1241–1263
of the limit states, and quantification of the Shinozuka M, Feng MQ, Lee J, Naganuma T (2000) Sta-
tistical analysis of fragility curves. J Eng Mech
involved uncertainties. That is why the 126(12):1224–1231
researchers in this field have used many different Ugurhan B, Askan A, Erberik MA (2011) A methodology
approaches and methods to obtain their own seis- for seismic loss estimation in urban regions based on
mic fragility information. But the main goal is ground-motion simulations. Bull Seismol Soc Am
101(2):710–725
always to predict the performance of structural Vamvatsikos D, Cornell CA (2002) Incremental dynamic
systems with a quantified level of confidence analysis. Earthq Eng Struct D 31(3):491–514
under different levels of seismic action.
Cross-References Seismic Fragility of Aging Offshore

Platforms
▶ Analytic Fragility and Limit States [P(EDP|
IM)]: Nonlinear Dynamic Procedures Viswanath Kammula1 and Srinivas Sriramula2
1
▶ Analytic Fragility and Limit States [P(EDP| Structural Engineer-in-Training, Soscia
IM)]: Nonlinear Static Procedures Engineering Limited, Toronto, Canada
2
▶ Empirical Fragility Lloyd’s Register Foundation (LRF) Centre for
▶ Estimation of Potential Seismic Damage in Safety and Reliability Engineering, School of
Urban Areas Engineering, University of Aberdeen, Fraser
▶ Seismic Loss Assessment Noble Building, Aberdeen, UK
References Synonyms
Applied Technology Council (1985) Report ATC-13:
earthquake damage evaluation data for California. Aging infrastructure; Fragility analysis; Offshore
Applied Technology Council, Palo Alto structures; Probabilistic assessment; Seismic risk
Seismic Fragility of Aging Offshore Platforms 2857
Introduction However, as the resources in shallow water

have started to extinguish, the exploration activ-
The importance of infrastructure to both the fab- ities have moved to deeper waters. Nonetheless,
ric of society and its economy is becoming there are many fixed steel jacket platforms that
increasingly more apparent. The robustness of are still in service.
infrastructure systems can be judged by their When the FSJ platforms were originally
capacity to accommodate change over time. designed, the design life was considered as
However, the rapid rate at which our infrastruc- 25 years. However, due to the continued avail-
ture is aging requires new solutions for providing ability of hydrocarbons at some locations, the
a resilient infrastructure that can last for future operators are looking at various options to extend
generations. the design life. For example, in the United King-
Earthquakes are paramount among the natural dom Continental Shelf (UKCS) sector of the
hazards impacting the infrastructure. The past sev- North Sea, out of the 288 offshore installations,
eral decades have witnessed a series of costly and about 50 % of the fixed platforms have exceeded
damaging earthquakes. The events would have their design life (Stacey et al. 2008). When it
even more devastating effects if a major earth- comes to extending the life of these structures,
quake were to hit an old and unprepared infrastruc- there are various key limit states one needs to
ture. The challenge that engineers today face is not consider, depending on the geographical loca-
only on how to construct new infrastructure with tion. For example, the effects of wave fatigue on
proper resistance to earthquakes but rather to offshore structures are a key consideration in one
update older structures that do not have the proper zone, whereas the structural response under
defense should a natural disaster occur. Failing to earthquakes is a serious limit state at a different
modernize older infrastructure could result in very location.
serious consequences. Considering the complexity and popularity of
Offshore platforms are one of the key infra- these structures, this entry looks at ways to char-
structure systems housing the facilities for explo- acterize the probabilistic response of FSJ plat-
ration and production of oil and gas. In addition to forms under earthquake loads. The seismic
acting as accommodation for offshore workers, vulnerability curve or seismic fragility curve is
these platforms contain a range of equipment one of the most important elements in the assess-
from power generation to oil processing. ment of seismic damage and the evaluation of the
Assessing the reliability of these structural sys- performance of different structural systems. The
tems is highly complicated due to the complexity fragility curves are used to represent the proba-
of interaction between various elements. It is bilities of structural damage due to earthquakes
possible to classify the offshore platforms further as a function of ground motion indices such as
as floating and fixed offshore platforms (Sadeghi peak ground acceleration (PGA), spectral accel- S
2007), with usage depending on the depth of eration (Sa), and spectral displacement (Sd). The
water at the installation location. These structures characterization of earthquake excitation and the
are popular in the waters of California, the Gulf of identification of different damage levels have an
Mexico, Nigeria, the North Sea, and the important role in developing the fragility curves
Persian Gulf. and understanding the seismic performance of
This entry will focus on a very specific and these structures.
popular type of offshore platform, the fixed steel This entry provides details of the development
jacket platform (FSJ). According to Ferreira of fragility curve for a typical FSJ offshore plat-
(2003), there are around 7,500 FSJ platforms form using the SAC-FEMA technique. The fol-
distributed around the world. These platforms lowing sections provide mathematical details of
have been popular because of their relatively the SAC-FEMA technique, an overview on the
simple design and the availability of oil at shal- reference structure and establishment of dynamic
low waters in the initial days of exploration. characteristics. Then, the ground motions used in
fragility analysis and the analysis results are method, effective fragility analysis method, and
discussed. The entry concludes with suggestions methods based on response surface (Pinto
to formulate a better inspection and maintenance et al. 2004). The SAC-FEMA method, which
framework based on the summary of findings. provides the basis for the FEMA-350 (2000) for
the seismic design and assessment of steel
moment-resisting frames, is the most practical
Seismic Fragility Analysis of Structures approach among these options. The SAC-FEMA
method is efficient when the response is predom-
Theory of Fragility Analysis inantly dominated by the first mode and when
failure is influenced by a single scalar variable,
Introduction for instance, the interstory drift. However, when
Seismic damage and loss estimations are essen- failure involves several random variables, then
tial for disaster planning and formulating one would need to use methods based on effective
damage-reduction and insurance policies. fragility analysis techniques or response surfaces.
Regional loss estimation can be mathematically
described through Eq. 1, as shown below (Kwon Mathematical Details of SAC-FEMA Technique
2007): The SAC-FEMA method (Pinto et al. 2004) is
one of the popular methods to perform vulnera-
XXX bility analysis. In the SAC-FEMA approach, the
P½Loss ¼ P½Loss jD ¼ d:P½D
seismic hazard P½IM ¼ s is defined in terms of
s LS d (1)
spectral acceleration ordinates, calculated at the
¼ djLS:P½LSjIM ¼ s:P½IM ¼ s
fundamental period of the structure. In this
method the failure occurs when the maximum
where P[▪] = probability of events in the demand over the duration of the seismic excita-
brackets. The term Loss refers to direct or indirect tion exceeds the corresponding capacity. The
losses from a seismic event; IM is an intensity seismic hazard is combined with the drift demand
measure of a seismic hazard such as spectral to define drift hazard as follows:
acceleration or peak ground acceleration (PGA);
ð
and s is a realization of the intensity measure. P
H D ðd Þ ¼ P½D d jSa ¼ x j dHðxÞ j (2)
[LS|IM = s] is a conditional probability of
attaining structural limit state (LS), and P[D =
d|LS] is a conditional probability of attaining where |dH(x)| means the absolute value of the
damage (minor, moderate, or major), which is a derivative of the site’s spectral acceleration haz-
qualitative measure. The loss of a region is deter- ard curve times dx. The seismic demand is related
mined based on the damage state. Among these to the hazard as follows:
elements, the term, P[LS|IM = s], refers to strong,
motion-shaking severity influencing the proba- ^ ¼ aSa b
D (3)
bility of reaching or exceeding a specified perfor-
mance limit state. This term is referred as where D ^ is the median value of demand D. The
vulnerability or fragility. The strong, motion- constants, a and b, are determined from a regres-
shaking severity may be expressed as an inten- sion analysis of structural demands calculated
sity, peak ground parameters (PGA, PGV, or from the dynamic analyses. The term bD, stan-
PGD), or spectral ordinates (Sa, Sv, or Sd) dard deviation of the natural logarithm of demand
corresponding to a fundamental structural period. D, is calculated about the median value of
demand D. Ideally, at each spectral acceleration
Analysis Techniques value, there would be multiple demand values, as
Among others, some practical approaches to per- two ground motions with same spectral accelera-
form fragility analysis are the SAC/FEMA tion could result in different drifts. Hence, the
!
median value over each set of demands is ln d=C^
corresponding to a spectral ordinate value that P½ C d ¼ F (8)
bC
needs to be calculated. Also every set is associ-
ated with a unique dispersion. To account for the
dispersion, a random variable ϵ, with unit median Substituting Eqs. 6 and 8 into 7, the probability of
and dispersion equal to bD, is introduced. Equa- annual failure is given as
tion 4 then updates to:
1 k 2
^
PPL ¼ H sa exp
C
b2
þ bC
2
(9)
D ¼ aSa b ϵ (4) 2 b2 DjSa
However, for simplicity, only one demand is used Previous studies have shown that PPL is dictated
for a given spectral acceleration, and it is by the hazard and not by the uncertainties in the
assumed that this value would be the median demand and capacity. The above procedure, sum-
demand value, and the dispersion is assumed to marized in Eq. 9, fully couples hazard, demand,
be the same across all the spectral acceleration and capacity. However, as highlighted in Kwon
values. The demand D is assumed to be (2007), the derivation of vulnerability curves is
log-normally distributed about the median, with more advantageous than calculating fully
standard deviation of the natural logarithm equal coupled risks.
to bD. For a given intensity level, the probability
of reaching a structural demand, d, can be calcu- Application to a Typical Fixed Steel Jacket
lated using cumulative distribution function of Platform
log-normal distribution, given as:
Analytical Modeling
! Fragility analysis was performed on a typical FSJ
ln d=axb
P½d Dj Sa ¼ x ¼ 1 F (5) offshore platform that was designed according to
bDjSa the API guidelines and presented by Nordal
et al. (1987). It is assumed that all the lateral
Equations 4 and 5 result in a closed form solution forces are transferred to the foundation mainly
of drift hazard as shown below: by the diagonal braces. Furthermore, at each
level, the horizontal X-braces transfer the forces
2
d 1k 2 among the four vertical frames. In this study, the
H D ðdÞ ¼ P½D d ¼ H Sa exp b performance of the platform is estimated through
2 b2 DjSa
fragility analysis when subjected to seismic
(6)
forces. A 2-D model of the X-braced vertical
frame on grid line A is developed in ETABS S
The drift hazard should be combined with
(2008); the section sizes are available in Nordal
the drift capacity, C, to evaluate the annual prob-
et al. (1987). A screenshot of this analytical
ability of the performance level not being
model is shown in Fig. 1.
met, PPL.
The numerical model of the structure is devel-
ð oped based on the following assumptions:
PPL ¼ P½C d jdH D ðdÞj (7)
• All the connections are assumed to be pinned
connections.
The drift capacity C is assumed to follow • All the members are modeled as truss
log-normal distribution with a median value of elements.
C^ and dispersion bC. Therefore, the probability of • Rigid diaphragm condition is assumed only
the capacity, C, being lower than the demand, d, for the level 4, where the mass is defined.
is expressed as • All supports are assigned pin conditions.
Seismic Fragility of
Aging Offshore
Platforms,
Fig. 1 Analytical model in
ETABS (a) 3D view (b)
elevation
• The primary lateral force-resisting members Evaluation of Dynamic Properties

are the braces. Using the above defined mass and section sizes, a
• Nonlinearity effects are not modeled. modal analysis is performed to obtain the funda-
mental period. First two mode shapes of this
Based on the work of Stear and Bea (1997), structure along with their corresponding periods
the total deck load being supported is taken as are shown in Fig. 2. The fundamental period was
51 MN, which would be shared by the four also determined using static pushover analysis to
frames in the y-direction. Hence, a mass of verify the period obtained using the modal anal-
12.75 MN (12,750 KN) is assigned to the top ysis. The fundamental period obtained using both
node of the frame. In addition, as discussed the analyses are in good agreement.
in the modeling assumptions, rigid diaphragm
condition (all the nodes in a given floor Selection of Ground Motions
are constrained together) is assigned for the top For demonstrative purposes, ground motions are
level. selected from the Pacific Earthquake Engineering
acceleration using the following equation

(Chopra 2010):

2
2p
A ¼ o2n D ¼ D (10)
Tn
where on is the natural frequency of the structure,

D is the peak deformation of the structure due to
earthquake, Tn is the natural period of the struc-
ture, and A is the pseudo-spectral acceleration
corresponding to Tn. The peak displacement of
the top node, obtained above, is divided with total
story height to obtain the peak drifts. The graph of
peak drift versus spectral acceleration is shown in
Fig. 4 for the ground motions listed above.
In general, the pseudo-spectral acceleration is
very low because of the higher fundamental
period. When compared to buildings, the percent-
Seismic Fragility of Aging Offshore Platforms,
Fig. 2 Mode shapes of the frame (a) Mode 1 (T =
age increment in the mass must be higher than the
3.71 s) (b) Mode 2 (T = 0.123 s) percentage increment in stiffness for the offshore
structures. Having obtained the seismic hazard
expressed in spectral acceleration and the
Center (PEER) New Generation Attenuation demand expressed in interstory drift, a reliability
(NGA) database. However, it is recommended analysis can now be performed. The limit states
to select the ground motions based on the site are discussed in the following section.
location, if available. The database is composed
of 3,551 ground motion recordings that represent Formulation of Limit States
over 160 seismic events (including aftershock The performance limit states are of prime impor-
events) ranging in magnitude from M4.2 to tance in the fragility analysis and need to be
M7.9 (Chiou et al. 2008). A total of 30 far-field carefully determined. The three performance
ground motions are selected from the PEER NGA levels that were used for the offshore steel-braced
database. The PGA range for the final selected frame are given in Table 2. However, more inves-
suite of ground motions, listed in Table 1, is 0.2 g tigation is required to ascertain the applicability
to 1.8 g with an average of 0.5 g. The ground of these limit states to offshore structures. The S
motion suite includes earthquakes that occurred limits corresponding to each level are also given
from 1971 to 1999. The spectral acceleration at in Table 2.
the fundamental period of the structure, 3.71 s
corresponding to 5 % damping, is obtained by Fragility Analysis Results
constructing the elastic response spectra of each Using the outlined procedure in section “Mathe-
ground motion (Seismosoft 2011). For instance, matical Details of SAC-FEMA Technique,” the
for the first ground motion in Table 1, Loma fragility curves are developed for the three limit
Prieta earthquake (A01090), the pseudo- states discussed in the previous section and are
acceleration elastic spectra is shown in Fig. 3. shown in Fig. 5. It can be inferred that the prob-
ability of reaching the higher limit state, charac-
Calculation of Story Drift terized with drift 2 %, is almost 100 % for a
For a linear elastic system, the peak story drift seismic hazard as low as 1.2 g, which signifies
can be obtained from the elastic pseudo-spectral the non-robust seismic design of the structure.
Seismic Fragility of Aging Offshore Platforms, Table 1 Considered ground motions (PEER: NGA Database 2008)
No Name PGA, g Earthquake name Year Station name
1 A01090 0.294 Loma Prieta 1989 Foster City – APEEL 1
2 ABRD130 0.313 Whittier Narrows-01 1987 Brea Dam (Downstream)
3 AING000 0.299 Whittier Narrows-01 1987 Inglewood – Union Oil
4 AND250 0.244 Loma Prieta 1989 Anderson Dam (Downstream)
5 ATAR090 0.644 Whittier Narrows-01 1987 Tarzana – Cedar Hill
6 CH10270 0.269 Loma Prieta 1989 Oakland – Outer Harbor Wharf
7 CHY014W 0.229 Chi-Chi, Taiwan 1999 CHY014
8 CHY080N 0.218 Chi-Chi, Taiwan-03 1999 CHY080
9 G03090 0.367 Loma Prieta 1989 Gilroy Array #3
10 Go2000 0.363 Loma Prieta 1989 Gilroy Array #2
11 Go2090 0.321 Loma Prieta 1989 Gilroy Array #2
12 H-Z14090 0.274 Coalinga-01 1983 Parkfield – Fault Zone 14
13 HDA165 0.269 Loma Prieta 1989 Hollister Diff. Array
14 HDA255 0.279 Loma Prieta 1989 Hollister Diff. Array
15 HEC090 0.337 Hector Mine 1999 Hector
16 LAC180 0.316 Northridge-01 1994 LA – City Terrace
17 CHY028-E 0.653 Chi-Chi, Taiwan 1999 CHY028
18 CHY028-N 0.795 Chi-Chi, Taiwan 1999 CHY028
19 H-PVY045 0.592 Coalinga-01 1983 Pleasant Valley P.P. – yard
20 PET090 0.662 Cape Mendocino 1992 Petrolia
21 TAR360 0.990 Northridge-01 1994 Tarzana – Cedar Hill A
22 TCU074-E 0.597 Chi-Chi, Taiwan 1999 TCU074
23 ORR090 0.568 Northridge-01 1994 Castaic – Old Ridge Route
24 ORR291 0.268 San Fernando 1971 Castaic – Old Ridge Route
25 ORR360 0.514 Northridge-01 1994 Castaic – Old Ridge Route
26 PEL090 0.231 San Fernando 1971 LA – Hollywood Stor FF
27 PEL180 0.174 San Fernando 1971 LA – Hollywood Stor FF
28 PEL360 0.358 Northridge-01 1994 LA – Hollywood Stor FF
29 TCU071-N 0.380 Chi-Chi, Taiwan-03 1999 TCU071
30 TAR090 1.779 Northridge-01 1994 Tarzana – Cedar Hill A
Seismic Fragility of 1.4

Aging Offshore
Spectral Acceleration (g)
Platforms, 1.2
Fig. 3 Pseudo-
acceleration elastic 1
response spectra (z = 5 %) A01090
0.8
0.6
0.4
0.2
0
0 1 2 3 4
Tn (sec)
Seismic Fragility of 0.90

Aging Offshore ζ = 5%
0.80
Platforms, Fig. 4 Peak
drift versus spectral 0.70
Peak Drift (%)

acceleration 0.60
0.50
0.40
0.30
0.20
0.10
0.00
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
Sa (g)
Seismic Fragility of Aging Offshore Platforms, Table 2 Structural performance levels and damage, expressed in
terms of drift (Source: ASCE 2007)
Structural performance levels
Elements Type Collapse prevention Life safety Immediate occupancy
Braced steel Primary Extensive yielding and Many braces yield or buckle but Minor yielding or
frames buckling of braces do not totally fail buckling of braces
Many braces and their Many connections may fail
connections may fail
Secondary Same as primary Same as primary Same as primary
Drift 2 % transient 1.5 % transient 0.5 % transient
or permanent 0.5 % permanent negligible permanent
The slope of the fragility curve decreases with force demand. Though the fragility curves are
the limit state which implies that relatively derived where the demand is expressed in terms
higher hazard increment is necessary to reach of interstory drift, similar curves can be derived
the probability of failure as the limit state is when the demand is expressed in terms of accel-
increased. eration, residual drift, etc.
The procedure described herein to perform
fragility analysis is simple and practical. Hence,
Summary when assessing the performance of similar FSJ
platforms, practicing engineers can implement it S
In this entry, the seismic performance of a typical easily. The above results may also be used to
fixed steel jacket platform is evaluated through prioritize the retrofitting strategy of aging plat-
fragility analysis. The analysis has shown that the forms. Fragility curves provide a quick, yet, suf-
considered offshore platform cannot withstand ficient insight in to the performance robustness of
even the seismic hazard corresponding to the the structures and help to determine the members
collapse prevention performance level. This that need the utmost attention. This is also ideal
undermines the robustness of such structures for planning the inspection and maintenance
which are aging, for the ever-increasing seismic strategies for these structures. It is to be noted
hazard. Even though the results were shown on a that other significant limit states also influence
simple elastic system, the authors believe that the such decisions for FSJ platforms; the presented
implications could be more serious when an approach is recommended for characterizing the
inelastic system was used, as the drift demand seismic effects along with appropriate consider-
would be much higher albeit with a reduction in ation of other limit states.
Seismic Fragility of 1
Aging Offshore
Platforms, 0.9
Fig. 5 Seismic fragility LS = 0.5
0.8 LS = 1.5
Probability of Exceedance
curves for interstory drift
0.7 LS = 2.0
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Spectral Acceleration (g)
Acknowledgments Sriramula’s work within the Lloyd’s Ferreira DD (2003) Fiscal treatment decommissioning
Register Foundation Centre for Safety and Reliability and bonds: anticipating impacts of financial
Engineering at the University of Aberdeen is supported assurance requirements for offshore decommissioning,
by Lloyd’s Register Foundation, a UK-registered charity a decision model for the oil industry. Doctoral
and sole shareholder of Lloyd’s Register Group Ltd. dissertation, State University Campinas, Brazil,
pp 164–168
Kwon O (2007) Probabilistic seismic assessment of struc-
Cross-References ture, foundation, and soil interacting systems. PhD
dissertation, University of Illinois at Urbana-
Champaign
▶ Analytic Fragility and Limit States [P(EDP| Pacific Earthquake Engineering Research Center (PEER):
IM)]: Nonlinear Static Procedures NGA Database (2005) The Regents of the University
▶ Modal Analysis of California (Regents). http://peer.berkeley.edu/nga/.
Accessed 10 Nov 2013
▶ Performance-Based Design Procedure for Nordal H, Cornell CA, Karamchandani A (1987) A struc-
Structures with Magneto-Rheological Dampers tural system reliability care study of an eight-leg steel
▶ Seismic Collapse Assessment jacket offshore production platform. In: Proceedings
▶ Seismic Fragility Analysis of The Marine Structural Reliability Symposium,
SNAME, Arlington, VA, October 1987.
▶ Seismic Risk Assessment, Cascading Effects Pinto PE, Giannini R, Franchin P (2004) Seismic reliabil-
ity analysis of structures. IUSS Press, Pavia
Sadeghi K (2007) An overview of design, analysis, con-
References struction and installation of offshore petroleum plat-
forms suitable for cyprus oil/gas fields. GAU J Soc
ASCE (2007) Seismic rehabilitation of existing buildings. Appl Sci 2(4):1–16
ASCE/SEI Standard 41-06, Reston SeismoSignal “Version 4.3.0” (Software) (2011)
Chiou B, Darragh R, Gregor N, Silva W (2008) NGA Seismosoft, Pavia. www.seismosoft.com
project strong-motion database. Earthq Spectra Stacey A, Birkinshaw, M, Sharp JV (2008) Life
24(1):23–44 extension issues for ageing offshore installations. In:
Chopra AK (2010) Dynamics of structures–theory and American Society of Mechanical Engineers, proceedings
applications to earthquake engineering. Prentice-hall, of the international conference on offshore mechanics
NewJersey 07458, U.S.A. and arctic engineering, vol 5, pp 199–216
ETABS “Version 9.5.0” (Software) (2008) Computers Stear JD, Bea RG (1997) Earthquake analysis of offshore
and Structures, Berkeley. www.csiberkeley.com platform, screening methodologies. Technical Report
FEMA 350 (2000) Recommended seismic design criteria to Joint Industry Project Sponsors, Project Phase III,
for new steel moment-frame buildings. NEHRP Department of Civil Engineering, University of
Program California at Berkeley
Seismic Instrument Response, Correction for 2865
amplitude displacement X(o), the output ampli-

Seismic Instrument Response, tude Y(o) can be calculated as
Correction for
Y ðoÞ ¼ XðoÞ AðoÞ (1)
Jens Havskov1 and Gerardo Alguacil2
1
Department of Earth Science, University of where A(o) is the displacement amplitude
Bergen, Bergen, Norway response, then the displacement, X(o), of the
2
Instituto Andaluz de Geofı́sica, University of ground can simply be calculated as
Granada, Granada, Spain
XðoÞ ¼ AðoÞ=Y ðoÞ (2)
Synonyms
A(o) has traditionally been called the magni-
Accelerometer; Amplitude response; Digitizer; fication, since for an analog recorder A(o) gives
Fourier transform; Phase response; Poles and how many times the signal is magnified at differ-
zeros; Response function; Seismometer ent frequencies.
In this document, a description of the response
Recording seismic signals, whether in digital or function for different kinds of sensors will be
analog form, gives a number in a computer or an described, as well as methods used for making
amplitude on paper (Fig. 1). For velocity sensors a correction for the instrument response to arrive
this number is proportional to ground velocity at the true ground motion, whether in time
above a certain frequency, and for accelerometers domain or frequency domain.
the number is proportional to ground
acceleration.
What is really needed is the true motion of the The Elements of a Seismic Station
ground in displacement, velocity, or acceleration.
So considering Fig. 1, the signal to the right is A seismic station consists of a sensor connected
recorded, and the signal to the left is what is to a recorder. The sensor has an output in volts
needed. They are obviously different. Seismolo- proportional to the ground velocity (velocity
gists most often use displacement while engineers sensor) or to the ground acceleration
prefer acceleration. In most cases the correction is (an accelerometer), and this output is generally
not as easy as just multiplying with a constant to only constant in a specific frequency range. The
get the ground motion for two reasons: proportionality constant is called the generator
constant G and has units of V/(m s1) and V/g
– The seismic sensor is not recording the desired (g is the gravity acceleration) for the two types of S
ground motion magnitude: displacement, sensors, respectively. The sensor output signal
velocity, or acceleration and a transformation enters the digital recorder which will have some
must be done. filter built in at high frequency to avoid aliasing.
– The response of the seismic sensor to the The digitizer will give out a number (called
ground motion, as well as the response of the counts) proportional to the input voltage, and
recorder, is in general frequency dependent, so this defines the digitizer sensitivity in terms of
a frequency-dependent correction must be counts/V.
made (Fig. 1).
It is thus essential to know how the seismic Frequency Response of a Linear System
sensor and recorder modify the input signal, the
ground motion, to produce the recorded output A seismic sensor and associated electronics in the
signal. If, e.g., for given harmonic ground seismic recorder is assumed to behave as a linear
2866 Seismic Instrument Response, Correction for
Input max 0.02 mm displacement Output max 1 250 000 counts
In Out
Acclerograph
1 sec
Seismic Instrument Response, Correction for, accelerometer and becomes a new signal with
Fig. 1 The input is the ground moving with a maximum a maximum number (count) of 1,250,000. Note also that
displacement 0.02 mm. This “signal” goes through an the frequency content is different for input and output
system. The linearity means that there is a linear

relationship between input signal and output sig-
nal. If the input signal is x(t) and the output y(t),
then multiplying x(t) with a constant will result in
an output signal multiplied with the same con-
stant. For example, if the ground velocity is dou-
bled, then the output from the seismometer is also
doubled. If two signals of different frequency and
Seismic Instrument Response, Correction for,
amplitude are input, then also two signals with
Fig. 2 RC low-pass filter. R is the resistor and C the
the same frequencies (with different amplitude capacitor. Input is x(t) and output y(t)
and phase) are output. The frequency-dependent
relation between the output and the input is called
the frequency response function. For a more com- output signal amplitude Y(o) can be written as
plete description, see Scherbaum (2007). (e.g., Havskov and Alguacil 2010)
A very simple linear system is the RC filter
1 1
consisting of a resistor R and a capacitor C. Y ðoÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi XðoÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi2ffi XðoÞ
An RC low-pass filter is seen in Fig. 2. 1 þ o2 R2 C2 1 þ o2 =o0
This circuit lets low frequencies pass without (4)
attenuation, while high frequencies are attenu-
ated due to the frequency-dependent impedance where o0 = 1/RC. o0 is also called the corner or
of the capacitor. For a sine wave signal of fre- offiffi=
cutoff frequency of the filter, and for p ffi o0, the
quency f, the capacitor impedance is the relation amplitude has been reduced to 1= 2 ¼ 0:707
between the voltage amplitude at the capacitor (see Fig. 4). From Eq. 4 the amplitude frequency
terminals and the amplitude of the current response function of the filter A(o) (since only
flowing through the capacitor. amplitudes are considered) is
The impedance Zc of a capacitor seen by the
input for a sine wave signal of frequency f is 1 Y ðo Þ
AðoÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi2ffi ¼ (5)
1 þ o =o0 XðoÞ
2
1 1
Zc ¼ ¼ (3)
2pfC oC If A(o) is completely known, then the amplitude
of the harmonic input signal X(o) can be calcu-
where C is the capacitance (F) and o = 2pf. The lated from the measured signal as
voltage over the capacitor is delayed by one
fourth cycle (90 ) with respect to the current XðoÞ ¼ Y ðoÞ = AðoÞ (6)
through it. Considering that the RC filter is a
frequency-dependent voltage divider, using The amplitude response can be measured very
a monochromatic signal of angular frequency o simply as shown in Fig. 3. By varying the fre-
and amplitude X(o), x(t) = X(o)cos(ot), the quency, both input and output amplitudes can be
Seismic Instrument Response, Correction for, goes directly to channel 1 (Ch1, top trace) on the oscillo-
Fig. 3 Measuring the amplitude and phase response func- scope and to channel 2 (Ch2, bottom trace) through the
tion of an RC filter. The signal from a signal generator filter so both input and output is measured
measured at different frequencies to produce the

amplitude response function.
From Fig. 3, it is seen that the output signal not
only has been changed in amplitude but also has
been delayed a little relative to the input signal; in
other words, there has been a phase shift; see
definition below.
In this example, the phase shift is negative (see
definition in Eq. 7). The complete frequency
response of the filter therefore consists of both
the amplitude response function and the phase
response function F(o); see Fig. 4. Considering
a general input harmonic waveform x(o,t) =
X(o) cos(ot) at frequency o, the output can be
written as
yðo, tÞ ¼ XðoÞ AðoÞ cos ðot þ FðoÞÞ Seismic Instrument Response, Correction for,
Fig. 4 Amplitude and phase response of an RC filter
(7) with a corner frequency of 1 Hz. The phase response
(or phase shift) is given in degrees
The phase shift is here defined as a quantity being
added to the phase as seen above. Thus compar-
ing Fig. 4 and Eq. 7, it is seen that the phase shift
is negative. This is the most common way of eiot ¼ cos ðotÞ þ i sin ðotÞ (8)
S
defining the phase shift, but the opposite sign is
sometimes seen, and it may then be called phase Equation 7 can now be written as
delay or phase lag. So it is very important to know
which definition has been used. yðo, tÞ ¼ XðoÞAðoÞeiðotþFðoÞÞ
Using Eq. 7 to correct the amplitude and ¼ XðoÞAðoÞeiot eiFðoÞ (9)
phase of a single monochromatic signal in time
domain is easy, but to make the correction for y(o,t) is now a complex number of which the real
a whole signal with different frequencies is not part is the actual output. This can be further
simple. simplified considering that any complex number
This can be considerably simplified if the Z can be written as
complex representation of harmonic signals is
used. Instead of writing cos(ot), the real part of pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
the exponential function can be used: Z ¼ a þ ib ¼ a2 þ b2 eiF ¼ jZ jeiF (10)
where F = tan 1(b/a) which also follows from Velocity Sensor

Eq. 8. The complex frequency response T(o) can The complex response function for a velocity
then be defined as sensor for displacement, velocity, and accelera-
tion is the following (Havskov and Alguacil
T ðoÞ ¼ AðoÞeiFðoÞ ¼ jT ðoÞjeiFðoÞ (11) 2010).
and Eq. 9 can be written as Displacement
Z ðo Þ io3 G
yðo, tÞ ¼ XðoÞT ðoÞeiðotÞ (12) T d ðoÞ ¼ G¼ 2 (13)
U ðoÞ o0 o2 þ i2oo0 h
Now only one complex function, T(o), that

includes the phase shift can therefore completely
describe the instrument frequency response. So Velocity
far only monochromatic signals have been used
and the corrections can easily be made with both o2 G
T v ðoÞ ¼ (14)
the real and complex representations Eqs. 7 and o20 o2 þ i2oo0 h
12, so it might be hard to see why the complex
representation of the response function is needed. Acceleration
Later it is going to be shown how an observed ioG
seismogram is corrected, not only for one ampli- T a ðo Þ ¼ (15)
o20 o2 þ i2oo0 h
tude at a time, but actually for the whole signal
(containing a range of frequencies), which is pos-
where G is the generator constant, o0 is the nat-
sible with digital data. Obviously this cannot be
ural frequency of the sensor (the frequency at
done with one amplitude at a time, but will have
which it will oscillate if no damping), and h is
to be done with the amplitude and frequency
the damping coefficient (how the motion of the
content of the whole signal, in other words, mak-
mass is damped; see Fig. 5). It is seen that the
ing spectral analysis. It will hopefully then be
only difference between the respective three
clear why a complex representation is needed.
types of response curves is the factor io.
For the velocity response, the amplitude and
phase response are (Fig. 5)
Frequency Response of Seismic Sensors
Go2
There are two important types of seismic sensors: Av ðoÞ ¼ jT v ðoÞj ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2 ffi
o0 o2 þ 4h2 o2 o20
1. Velocity sensor (seismometer). This sensor (16)
has an output proportional to velocity in
a limited frequency band; see Fig. 5. This is
both the case for the traditional passive seis- ImðT v ðoÞÞ 2hoo0
Fv ðoÞ ¼ a tan ¼ a tan
mometers like geophones (for frequencies ReðT v ðoÞÞ o20 o2
down to about 1 Hz) and modern broadband (17)
seismometers (down to frequencies of
0.003–0.03 Hz). It is seen that as the damping decreases, the
2. Accelerometer. This sensor usually has an sensor response to velocity becomes constant
output proportional to ground acceleration for frequencies above the natural frequency. It is
from 0 Hz (DC) to some cutoff frequency also seen that if there is little damping, the sensor
(see Fig. 6). In some cases there will also be will have a high gain at the natural frequency, so
a lower cutoff, e.g., at 0.1 Hz. damping is usually set to 0.707.
Seismic Instrument 101

Response, Correction
for, Fig. 5 The amplitude
and phase response
functions for a velocity
seismometer with a natural 100
frequency of 1 Hz. Curves
AMPLITUDE
for various levels of
damping h are shown. It is
assumed that the generator 10–1
constant is 1
h = 0.1
h = 0.3
10–2
h = 0.7
h=1
h=2
h=4
10–3
0.1 1 10 100
FREQUENCY (Hz)
180
h = 0.1
160 h = 0.3
h = 0.7
140 h=1
h=2
120 h=4
PHASE degrees
100
80
60
40
20
0
0.1 1 10 100
FREQUENCY Hz
Accelerometer this frequency (usually above 50 Hz). The instru-

The acceleration response curve for an FBA ment correction is then very easy since it is
accelerometer is (Havskov and Alguacil 2010) assumed that the feedback natural frequency of
the sensor is very high compared to the frequen-
G=o2f cies of interest so the response for acceleration is
T a ðoÞ ¼ (18) simply approximated by multiplying with the
o2f o2 þ i2oof h
generator constant
where of is the effective natural (angular) fre- T a ðo Þ ¼ G (19)

quency with feedback. The amplitude of this
response has a flat value of G from DC up to The velocity and displacement response is then:
Velocity Complete Frequency Response

T v ðoÞ ¼ G io (20)
For passive sensors, the above equations repre-
sent exactly the seismometer response, and for
Displacement many broadband sensors, this is also the case up
to a particular frequency. A recording of
T d ðoÞ ¼ G o2 (21)
a seismic signal also involves the seismic
recorder which also has a frequency response.
So for engineering purposes, where acceleration Modern recorders have little amplitude change
is needed, it is very easy to correct for the accel- or phase shift in the main frequency band used,
erometer response since it is just a question of so the frequency dependence of the recorder is
multiplying with a constant, while seismologist often ignored. Some recorders might modify the
wanting displacement must make the transforma- signal enough for a correction to be made.
tion (corresponding to integrating the signal In order to completely describe the response for
twice). active sensors and possibly also include the
In practice, many accelerograms, response of the recorder, a more general repre-
especially – but not only – those recorded with sentation of the response function must be used.
earlier accelerographs, are affected by offset or It turns out that T(o) for all systems made from
baseline shifts that may influence the estimation discrete mechanical or electrical components
of characteristic parameters of strong ground (masses, springs, coils, capacitors, resistors,
motion (Boore 2001). The source of this long- semiconductors, etc.) can be represented exactly
period noise may be true ground motions (e.g., Scherbaum 2007) by rational functions
(rotation, tilt, local deformation) or instrumental of io:
(mechanical nonlinearity, electronic effects) and
is not yet fully understood (Akkar and Boore ðio z1 Þðio z2 Þðio z3 Þ:::::
T ðoÞ ¼ c
2009). Prior to any integration to compute the ðio p1 Þðio p2 Þðio p3 Þ:::::
ground velocity or displacement, the (22)
accelerogram should be corrected from these
effects. where c is the combined normalization constant
In many practical cases, this correction may for nominator and denominator polynomials,
consist simply in applying a high-pass filter z are the zeros (or roots) of the nominator poly-
with a suitable cutoff frequency, as it will be nomial, while the zeros of the denominator poly-
shown below. In some cases, a particular nomial (poles) are p. Using Eq. 22 to represent
processing scheme adapted to an individual T(o) is the so-called poles and zeros representa-
record is required. In fact, most accelerogram tion, which has become the most standard way.
databases (e.g., Internet Site for European For example, for the seismometer response
Strong-motion Data, www.isesd.hi.is) include function for velocity, Tv(o) can be written as
a “processed” version of each accelerogram,
which usually consists of baseline correction, ðio 0Þðio 0Þ
band-pass filtering, and conversion to ground T v ðo Þ ¼ (23)
ðio p1 Þðio p2 Þ
motion units.
It is interesting to note that whether where
a velocity sensor or an accelerometer is used,
it is possible to calculate displacement, veloc- pffiffiffiffiffiffiffiffiffiffiffiffiffi
ity, and acceleration from both types of sen- p1 ¼ o0 h þ h2 1 (24)
sors; however, as is shown later, in practice,
pffiffiffiffiffiffiffiffiffiffiffiffiffi
this will be frequency limited due to noise in p2 ¼ o0 h h2 1
the system.
So in addition to the poles p1 and p2, the seis- just a question of adding to the equation in the
mometer response function has a double zero at z pole and zero representation.
= 0 and the normalization constant is 1. Note In addition to the frequency-dependent ele-
that since h usually is smaller than 1, the poles are ments, a complete seismograph or accelerograph
usually complex. Complex poles always appear also has the frequency-independent elements, the
as conjugate pairs. For the displacement generator constant, and the digitizer gain which
response, the equation is has to be multiplied with the normalization
constant.
ðio 0Þðio 0Þðio 0Þ
T d ðo Þ ¼ (25) Example
ðio p1 Þðio p2Þ
A seismometer has a generator constant G =
and there is thus one more zero. For the 300 V/ms1 and the digitizer has a sensitivity
standard accelerometer, the displacement of 100,000 counts/V. The velocity sensitivity
response is simply (io)2 which corresponds to of the seismograph in the passband (flat for
two zeros. velocity) is then 100,000 counts/V 300 V/
ms1 = 3 107 counts/ms1. This number
For active sensors, there might be more poles must then be multiplied with the normaliza-
and zeros representing the filtering inherent in tion constant for the rest of the system which is
active sensors. the sensor and possibly some filters. For dis-
The frequency dependence of a recorder can placement, the amplitude displacement
equally be represented by poles and zeros, so it is increases with frequency (Fig. 6), but that is
DISPLACEMENT VELOCITY ACCELERATION

104 102
SEISMOMETER
104
2
V/(m/s2)
V/(m/s)
0 –1
V/m
102 100 1
102 3
2
100 100 10–2
1 10 100 1 10 100 1 10 100
FREQUENCY Hz
104 102
ACCELEROMETER
105
S
–1
V/(m/s2)
V/(m/s)
0
V/m
2 102 100
103 1
–2
101 100 10–2

1 10 100 1 10 100 1 10 100
FREQUENCY Hz
Seismic Instrument Response, Correction for, slope for each segment is indicated. Note how one curve
Fig. 6 Amplitude frequency response of a 1 Hz velocity translates into another by just adding or subtracting one
sensor (top) and an accelerometer (bottom) with a natural unit of the slope corresponding to a multiplication or
frequency of 100 Hz. From left to right, the figures show division by frequency. The response curves assume
the sensor response for ground displacement, velocity, and a generator constant G = 100 V/(m/s) for the seismometer
acceleration, respectively. The axes are logarithmic and and G = 1 V/(m/s2) for the accelerometer. Please note that
the horizontal axes show frequency in Hz. The asymptotic the vertical scales are different
Seismic Instrument Response, Correction for, is recorded on a broadband station and the amplitude is
Fig. 7 Top: Seismogram of a local earthquake; the posi- therefore proportional to ground velocity. Bottom: The
tion of the maximum amplitude is indicated. The fre- amplitude displacement response function. The gain
quency of the maximum amplitude is 1.4 Hz. The signal corresponding to the frequency of 1.4 Hz is indicated
taken care of with the io factor so the normal- shift was not considered since only one partic-
ization constant still has to be multiplied with ular amplitude was used.
the same number. For example, at 10 Hz, the In the more general case, it is desirable to
gain for displacement will be 2p 10s1 3 make instrument correction for the complete seis-
107 counts/ms1 = 1.88 109 counts/m. mogram. The seismogram seen in Fig. 7 is pro-
portional to ground velocity in its passband, and
a general procedure is therefore needed to calcu-
How to Correct for the Response late the corresponding displacement seismogram.
Obviously it is not possible to deal with one
In the simplest case, there is only a need to correct frequency at a time and the amplitude and fre-
for a single amplitude reading. In Fig. 7, a signal quency content of the whole signal must be used.
from a local earthquake (M = 5.2) recorded on In other words, a spectral analysis must be used.
a broadband station is seen. Figure 7 shows the The complex frequency response function
corresponding amplitude response curve for T(o) is now assumed known and the complex
displacement. Notice that around 8 Hz, there spectra of the input and output signals x(t) and
is a filter cutting off the high frequencies. This y(t) can be defined as X(o) and Y(o), respec-
filter is part of the sensor electronics and will be tively, using the definition for the Fourier spec-
represented by poles and zeros in the response trum (e.g., Oppenheim and Schafer 1989).
function. In the seismogram the maximum Knowing the complete complex output spectrum,
amplitude is 1,105,000 counts at a frequency the complete complex input spectrum can be
of 1.4 Hz. From the response curve the gain of obtained:
the system is 8 counts/nm or 8 109 counts/m.
So the maximum ground displacement ampli- XðoÞ ¼ Y ðoÞ=T ðoÞ (26)
tude is 1,205,000/8 = 138,000 nm. In practical
seismology, nm is mostly used for ground If T(o) is a seismic instrument response, the
motion since it is a more convenient unit than instrument-corrected ground motion spectrum is
m. In this example, there was no need for now obtained, and since X(o) is complex, this
a complex response function and the phase also includes the correction for phase.
The separate amplitude and phase spectra of the is used for both forward and inverse, the normal-
seismic signal can then be obtained as ization constants will cancel out. In practice,
since the signal consists of discrete numbers, the
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
spectrum also is done with discrete numbers; see,
Asignal ðoÞ ¼ ReðXðoÞÞ2 þ ImðxðoÞÞ2 (27)
e.g., Oppenheim and Schafer (1989).

1 ImðXðoÞÞ
Fsignal ðoÞ ¼ tan (28)
ReðXðoÞÞ Response Correction in Practice
Thus in practice, the ground displacement spec- In theory it is possible to recover the ground
trum would be calculated by taking the complex motion (displacement, velocity, or acceleration)
Fourier spectrum of the output signal and dividing at any frequency knowing the instrument response.
it with the complex displacement response func- In practice, one has to be careful to only do this in
tion and finally taking the absolute part of the the frequency band where the instrument records
complex ground displacement spectrum. This is real ground motion and not just electronic noise,
basically what is done to calculate source spectra since the instrument correction then becomes
for earthquakes (Havskov and Ottemöller 2010). unstable and the output has nothing to do with
All of this could of course easily have been the real seismic signal. Figure 8 shows an example.
done with the corresponding noncomplex equa- The figure shows the influence of filtering,
tions since, so far, the phase information has not when estimating the ground displacement signal.
been used and usually phase spectra for earth- In the frequency band 1–10 Hz, the signal looks
quake signals are not used. However, the next very much like the original signal although a bit
step is to obtain the true ground displacement more low frequency, since it is converted to dis-
time domain signal instead of the frequency placement and can nearly be considered an inte-
domain signal. With the knowledge about Fourier gration of the original signal. In the 0.1–10 Hz
transforms, this is now easy, since, when all the range, the earthquake signal almost disappears in
frequency domain coefficients or the spectrum the microseismic background noise. Why is this
has been calculated, the corresponding time signal considered seismic noise and not instru-
domain signal can be generated by the inverse mental generated noise? First, the earthquake
Fourier transform. Since the corrected signal con- signal has about the same amplitude as above;
sists of a sum of cosine signals with different second, it “looks” like seismic background noise;
phases, each of them delayed differently due to and third, the amplitude at 1,290 nm is at a period
the instrument response, the shape of the output of 5 s (peak amplitude of microseismic noise)
signal will depend on the phase correction. So which looks reasonable compared to worldwide
now life is easy, since the corrected complex observations. Note that this is how the earthquake S
displacement spectrum is already corrected for signal would have looked being recorded on
phase and the ground motion can be obtained as a broadband sensor, hardly noticeable. The last
trace shows the calculation of the displacement
ð
1
without filtering so the lowest frequency used is
1
xðtÞ ¼ Y ðoÞ=T ðoÞ eiot do (29) 1/T, where T is the length of the window, here
2p
1 80 s (only 65 s shown) so f = 0.0125 Hz. The
amplitude is now more than 50,000 nm and the
While it is possible to use only half of the positive signal looks “funny.” The large amplitude obvi-
frequencies for making the amplitude spectrum, ously cannot be right since the microseismic
both positive and negative frequencies must be noise has its largest amplitude around 3–8 s and
used for the inverse transformation. Normaliza- it was 1,290 nm. So this is a clear case of trying to
tion constants of the Fourier transform can be make a displacement signal at frequencies lower
defined in different ways, but if the same routine than where seismic signals exist in the data.
P S 7300 counts
Original
260 nm
Displacement 1–10 Hz
Displacement 0.1–10 Hz
1290 nm
50200 nm
Displacement, no filter
40 50 60 10 20 30
Seconds
Seismic Instrument Response, Correction for, filters. The third trace from the top shows harmonic
Fig. 8 Instrument corrections in different filter bands. waves at a frequency of about 0.2–0.3 Hz which is typical
The top trace is the original recording of a small earth- microseismic background noise originating in the ocean.
quake with a 1 Hz seismometer. The three bottom traces The amplitudes to the right are maximum amplitudes
have been converted to displacement using different (Figure from Havskov and Alguacil (2010))
Original max=11300 counts
S
A
max=420 counts
Acceleration max=690 000 nm/(s*s)

S
A
max=700 000 nm/(s*s)
Displacement max=330 nm
S
A
max=340 nm
1.0 second
Seismic Instrument Response, Correction for, the accelerometer signal but of lower dominant frequen-
Fig. 9 Acceleration and displacement. The seismogram cies, and the amplitudes are different. The middle traces
in the figure is the first few seconds of a P-wave. On the show the two signals converted to accelerations and the
site there is also an accelerometer installed (A) next to the bottom traces converted to displacement (frequency band
seismometer (S). The top traces show the original records 1–20 Hz) (Figure from Havskov and Alguacil (2010))
in counts. The signal from the seismometer is similar to
Seismic Loss Assessment 2875
The ratio of the displacement gain for a 1 Hz motion, whether in displacement, velocity, or
seismometer at 1 Hz and 0.0125 Hz is acceleration
1/0.01253 = 5 105. In other words, if the gain at – How the noise influences the correction mak-
1 Hz is 1.0, the signal must be multiplied by 1.0 to ing it necessary to limit the correction to
get the displacement, while at 0.0125 Hz, it must a particular frequency range
be multiplied by 5 105. So any tiny amount of
instrumental noise present at low frequencies will
Cross-References
blow up in the instrument correction. In the above
example, it seems that the displacement signal can
▶ Passive Seismometers
be recovered down to 0.1 Hz with a 1.0 Hz sensor.
▶ Principles of Broadband Seismometry
Engineers often think accelerometers are the
▶ Recording Seismic Signals
only instrument to use and seismologist normally
▶ Seismic Accelerometers
will only use the weak motion velocity sensor
▶ Sensors, Calibration of
(seismometer). However, knowing the response
curve and the limitation in the data, it does not
really matter which instrument is used; see Fig. 9.
References
In this example, a seismic signal has been
recorded on a 1 Hz seismometer and an acceler- Akkar S, Boore D (2009) On baseline corrections and
ometer at the same site, and using the respective uncertainty in response spectra for baseline variations
response function, both signals have been commonly encountered in digital accelerograph
converted to displacement and acceleration. The records. Bull Seism Soc Am 99:1671–1690.
doi:10.1785/0120080206
only limitation is that a seismometer may get Boore D (2001) Effect of baseline corrections on displace-
saturated by a strong motion and an accelerome- ments and response spectra for several recordings of
ter will not be sensitive enough for recording the 1999 Chi-Chi, Taiwan, Earthquake. Bull Seism
weak motions (e.g., from distant earthquakes). Soc Am 91:1199–1211
The corrected signals are now very similar and quake seismology. Springer, Dordrecht, 358 pp
of the same amplitude. This example clearly Havskov J, Otemöller L (2010) Routine data processing in
demonstrates that, with modern instruments and earthquake seismology. With sample data, exercises
processing techniques, both accelerometers and and software. Springer, Dordrecht, 347 pp
Oppenheim AV, Schafer RW (1989) Discrete-time
seismometers can be used to get the same result. signal processing. Prentice-Hall, Englewood Cliffs,
870 p
Scherbaum F (2007) Of poles and zeros, fundamentals
Summary of digital seismology, revised 2nd edn. Springer,
Dordrecht, 271 pp
Recorded seismic signals represent a number pro- S

portional with the ground motion. The most com-
mon seismic stations have a velocity sensor with
an output proportional to ground velocity in some Seismic Loss Assessment
frequency band and/or an acceleration sensor
with output proportional with ground accelera- Nilesh Shome
tion in combination with a digital recorder. All Model Development, Risk Management
these elements will have some frequency depen- Solutions, Newark, CA, USA
dence and it has been shown:
– How to calculate the system frequency Synonyms

response
– How to correct for both the sensor and the Damage; Insurance; Loss; Nonstructural compo-
recorder in order to obtain the true ground nent; Risk; Structural component; Vulnerability
2876 Seismic Loss Assessment
Introduction This provides monetary losses to buildings in

terms of the average annual loss (AAL) and
It is observed that earthquakes, particularly those losses at various return periods.
of large magnitudes, can cause significant dam- In 1985, Applied Technology Council (ATC)
age to structures over a large region surrounding first published a comprehensive report (ATC-13
the rupture zone of earthquakes. The top three 1985) to estimate the earthquake loss of existing
costliest earthquakes in the world during buildings by developing a suite of VFs based on
1980–2012 are the Mw 9.0 Tohoku-Oki earth- expert opinions. ATC considered this approach
quake (2011), Mw 6.8 Kobe earthquake (1995), since very limited earthquake damage or loss data
and Mw 6.7 Northridge earthquake (1994) caus- were available at that time. The report developed
ing overall economic losses when occurred about earthquake loss estimates as a function of Modi-
$210B, $100B, and $44B respectively (III 2013). fied Mercalli Intensity (MMI) for different facil-
The structures those are damaged due to earth- ity classes (e.g., low-rise wood frame). Recently,
quakes are not only buildings, bridges, etc., but Global Earthquake Model (GEM) project has
also utilities like water, gas and sewer lines, and also followed a similar approach (Jaiswal
lifelines like train tracks and roads. The earth- et al. 2013) for developing VFs.
quakes also cause landslides due to movement of Professor Karl Steinbrugge based on engineer-
grounds or liquefaction due to loss of strength of ing investigations, data collection, and field stud-
the soil underneath the structures. In addition, ies of earthquake-damaged buildings developed
earthquakes can cause fire and tsunami leading the first, well-accepted VFs providing an estimate
to significant damage to structures. All the histor- of the functional relationships between building
ical losses illustrate that the potential for earth- damage and earthquake intensities (Steinbrugge
quake loss can be significant in a specific region 1982). These functions provided a firsthand
or to portfolio of buildings for insurance compa- insight to the damageability of buildings to the
nies. Hence it is important for the federal and insurance companies in the 1990s. Since engi-
state governments or for the insurance companies neers were interested primarily in the life safety
to estimate accurately the seismic losses to the of buildings till the end of the twentieth century,
built environment or to the insured properties to there was limited interest among the engineers to
manage earthquake risks. compile the damage data of all the slightly and
The performance of buildings nowadays is moderately damaged buildings for assessing per-
expressed in terms of monetary loss due to dam- formance of buildings in earthquakes. Northridge
age, downtime, and number of injuries including earthquake (1994) was the first earthquake that
fatalities due to earthquakes. See, for example, provided a large number of insurance claims data,
performance-based earthquake engineering which are used for performance assessment of
(PBEE) framework developed by Pacific Earth- single-family low-rise wood-frame buildings in
quake Engineering Research (PEER) center high seismic zones of California, for example,
(Cornell and Krawinkler 2000) or that developed Los Angeles and San Francisco (Wesson
by ATC-58 (2011). The loss assessment of build- et al. 2004).
ings is often carried out using a vulnerability In 1999, the Federal Emergency Management
function (also known as damage function), Agency (FEMA) first published a comprehensive
which defines the distribution of the ratio of the methodology to calculate analytically earthquake
building loss to the building value as a function of damage of model buildings for a given peak
ground-motion (GM) intensities. In this chapter, response to estimate potential losses from earth-
different approaches of development of building quakes (HAZUS 2003). Later on, the assembly-
vulnerability functions (VF) will be discussed. based vulnerability (ABV) was developed by
This will be followed by calculations of the risk Porter et al. (2001), and this was adopted by the
of building loss due to earthquakes by integrating Consortium of Universities for Research in Earth-
the seismic hazard results with the building VFs. quake Engineering (CUREE) project (2001)
for developing vulnerability functions for wood- uncertainty can be divided into two different cat-
frame buildings. This is a significantly improved egories: (1) aleatory and (2) epistemic (Benjamin
analytical approach over the HAZUS. The and Cornell 1970). Aleatory uncertainty is the
CUREE approach, however, did not consider inherent variability in the physical system; it is
more realistic representations of structures (e.g., stochastic and cannot be reduced by improving
2D/3D nonlinear FEM model as has been consid- the current approach of loss estimation. Episte-
ered lately in the latest PEER or ATC projects), mic uncertainty, on the other hand, is associated
has limited set of fragility functions compared to with lack of knowledge; it can be subjective and
those considered lately in the ATC-58 project is reducible with additional information. There
(2011), and also did not consider the correlation are significant epistemic uncertainties associated
of response or damage between different compo- with the loss estimates due to uncertainty in the
nents of buildings for accurate estimation of vari- estimation of different parameters, and this
ability of total loss of buildings. should be considered in the decision-making pro-
Recently, researchers are focused on develop- cess. FEMA-355F (2000) and ATC-58 (2011)
ing analytical VFs based on detailed nonlinear have recommended typical uncertainties of a
analysis of buildings. For example, PEER large number of parameters based on extensive
(2011) considered 3D models of 40+ story tall research, and Jayaram et al. (2012) have devel-
buildings (Shome et al. 2014; Ramirez and oped a comprehensive approach to estimate epi-
Miranda 2009; Shome and Bazzurro 2009), and stemic uncertainty of VFs based on the
many others have considered 2D plane frame uncertainty of those parameters.
model for loss estimation of buildings. In addi-
tion, ATC-58 (2011) has developed a number of
approaches – simple to very detailed and Methodology
complex – for loss assessment of buildings.
Majority of the researchers has adopted a story- The seismic loss estimation of buildings is depen-
based loss assessment procedure to take into dent on the lateral load resistance characteristics
account the variation in loss over the height, of buildings. Since building codes are being
which is particularly important for mid- to high- improved over the years, load resistance is depen-
rise buildings. All these researchers have consid- dent on the age of buildings. In addition, the
ered a large number of ground-motion records for resistance also depends on: (1) construction
nonlinear response analysis of buildings, and a class (e.g., steel moment-resisting frame),
few of those made appropriate correction to the (2) occupancy class (e.g., residential), and
response results because of possible introduction (3) region (e.g., high seismic regions in Califor-
of bias due to scaling of those records for nia). Although lateral resistance ideally should
nonlinear analysis (Haselton 2009). It is observed not depend on the occupancy class, it is generally S
that standard approach of scaling does not take observed that commercial buildings are generally
into account dependence of building response on well designed and constructed than residential
the spectral shape and this introduces bias in the buildings leading to better lateral resistance
response results. In this context, the framework capacity of those buildings. The loss estimation
developed by Jayaram et al. (2012) will be of portfolio of buildings uses this type of generic
discussed in detail. This approach considered information (e.g., HAZUS 2003) by grouping the
story-based loss assessment, took into account buildings with similar load resistance capacities.
the bias in the response results due to spectral The loss estimation of individual buildings (e.g.,
shape, and considered correlation of response, ATC-58 2011), on the other hand, uses the
damage, and repair cost of different components detailed building information to estimate accu-
of buildings at different stories. rately the lateral load resistance capacity of build-
It is well known that there is significant uncer- ings for calculating seismic losses. The seismic
tainty in the estimation of loss results and this loss estimation approaches that will be discussed
Seismic Loss Assessment, Fig. 1 Illustrative five-parameter Beta distribution
in this chapter can be applied easily for loss model, and Beta distribution is widely used for
estimation of both portfolios and individual building loss calculations following ATC-13 rec-
buildings. The VFs are developed based on ommendation (1985). The parameters of the dis-
mainly three approaches: (1) empirical approach tribution can be estimated from the mean and
which is generally expert-opinion-based devel- variance of the sample losses (Benjamin and Cor-
opment, (2) observation-based approach which nell 1970). The insurance claims data, however,
is based on building loss data, and (3) analytical show that the Beta distribution combined with
approach. All these approaches as well as the Dirac delta functions at 0 % loss and 100 % loss
advantages and disadvantages will be to model the probabilities of zero (F0) and com-
discussed here. plete (F1) loss fits the observed loss distribution
Earthquakes not only cause damage to struc- data better. In addition, insurance companies gen-
tural and nonstructural components of buildings erally pay to completely replace a property (i.e.,
but also damage building contents, such as furni- 100 % loss) beyond a particular threshold (a) of
tures, and cause closure of buildings. The closure loss (e.g., 70 % loss). Hence, more accurate rep-
incurs losses due to the expenses to the occupants resentation of distribution requires three addi-
for temporary rents or hotel stay (called additional tional parameters, which are F0, F1, and a, in
living expenses, ALE) or loss of income to busi- addition to the conventional mean and variance.
nesses (called business interruption, BI, loss). The The 5-parameter distribution for losses is shown
losses to contents and due to closure of buildings in Fig. 1 for illustration. Note that this type of
will not be discussed in this chapter. This chapter distribution may be more appropriate to model
will focus only on the loss due to property damage the insurance losses. The methodology for esti-
of buildings from shaking of earthquakes. mating the five parameters of the distribution
from loss data will be discussed in detail later on.
Loss Distribution
Since there is significant uncertainty (aleatory Seismic Risk
and epistemic) in estimating building losses at a Seismic risk assessment of buildings requires
given intensity of GM, VFs provide the loss dis- estimating the probability of losses for all the
tribution parameters as a function of intensities. possible future earthquakes over a specified
It is generally assumed that the loss distribution period of time or estimating losses for a scenario
can be represented by parametric distribution earthquake. The scenario earthquake can be a
historical earthquake, or maximum credible followed by Applied Technology Council (ATC)

earthquake (MCE), or standard design basis in the ATC-13 (1985) and (2) Cook’s approach
earthquake (DBE), or more frequent service followed recently in PAGER and GEM project
level earthquake (SLE). The standard risk man- (Jaiswal et al. 2013). Although the information
agement of buildings entails estimation of aver- generated from these approaches may not be very
age annual loss (AAL) for all the possible reliable, this type of approach will continue to be
earthquakes to determine the insurance premium used in other projects in the future because of
of buildings. This is the amount of modeled pre- ease in developing VFs when information or
mium that an insurer needs to collect in order to resource is limited.
cover the average loss over time. The other risk
quantity that is commonly employed is estima- ATC-13 (1985) Approach
tion of probability of exceeding certain loss for One of the first systematic attempts of developing
all the possible earthquakes. When this is esti- building vulnerability was carried out by ATC-13
mated for different amount of losses, the result is for assessing seismic risk in the state of California
known in the insurance industry as the exceed- for the Seismic Safety Commission (SSC). The
ance probability or EP curve. The scenario loss project was funded by the Federal Emergency
for a given earthquake, on the other hand, is Management Agency (FEMA). Since very lim-
useful for deterministic seismic risk management ited earthquake loss data was available at that
by estimating the Probable Maximum Loss time and there was no acceptable analytical
(PML). PML is the largest possible loss which approach available for estimating building loss
may occur, in regard to a particular risk, given the calculations, ATC derived VFs based on experi-
worst combination of circumstances. Tradition- ence and judgment of the experts – structural
ally, PML is the expected shake damage loss engineers, builders, and the like – by estimating
given the maximum size earthquake. There is, the expected percentage of damage of typical
however, considerable uncertainty over the buildings for a specific construction type when
severity of loss that might arise under a broad these buildings are subjected to a Modified
variety of events. So an improved PML is a con- Mercalli Intensity (MMI). Based on personal
servative deterministic upper bound, which is knowledge and experience, the experts responded
typically 90 % upper confidence bound on the to a formal questionnaire with their best estimates
portfolio loss from a risk-based MCE, for exam- of percent physical damage or damage ratio.
ple, 2,475-year return period event on a fault. All In order to develop the VFs, ATC established
these risk calculations require distribution of an advisory project engineering panel (PEP) pro-
building loss as a function of intensity of GM viding necessary feedback in order to develop
and this is obtained by developing the VFs of consensus of damage or loss estimates of differ-
buildings. The following section will discuss dif- ent facility classes. This was modeled after the S
ferent approaches of development of VFs. Delphi method for expert-opinion solicitations,
which was originally developed for the Air
Force by RAND Corporation in the 1950s. The
Empirical Approach: Expert-Opinion- tasks in Delphi method consist of formulating
Based Vulnerability Functions questionnaires, getting answers to those from
experts, iterating the answers one or more times,
There is significant advantage in developing and aggregating the responses by following a
expert-opinion-based VFs. This approach can be suitable statistical approach. The questionnaire
followed to develop VFs relatively quickly and asked each expert to provide low, best, and high
easily without requiring any loss data or develop- estimates of the damage ratio for different con-
ing any analytical approach. Researchers over the struction classes or engineering facilities at dif-
years have followed primarily two approaches for ferent MMI. In addition, the experts were also
developing VFs: (1) Delphi method which was asked to provide their level of experience with the
Seismic Loss
Assessment,
Fig. 2 Vulnerability
functions of single-family
wood-frame buildings in
California. (a) Mean
damage ratio (MDR).
(b) Coefficient of variation
(CoV)
construction class and the degree of uncertainty Catastrophe modelers and risk engineers adopted
in their estimates. Subsequently, two other sets of the ATC-13 damage curves until the 1994
questionnaires were developed in order to reach a Northridge earthquake. It will be shown later on
consensus on the estimation of different damage that there is significant difference in the results
estimates. The questionnaire results were based on ATC-13 and that observed in the
processed to develop damage probability matri- Northridge earthquake. In addition, the ATC-13
ces (DPM) by fitting Beta distribution to those approach is subjective. The damage functions
results. In DPM, the damage to a construction from this approach are based exclusively on the
class for a given intensity is described by a series opinion of experts. For this reason, it is difficult to
of damage states (DS) and the probability that the calibrate or modify those in order to incorporate
DS will occur at that intensity. The VF of single- new data or new design (e.g., adjustment for new
family wood-frame buildings in California fol- seismic code or other seismic regions). These
lowing this approach is shown in Fig. 2 for functions can also be biased as observed in the
illustration. Northridge earthquake. Hence, in the recent
Because it was the first systematic attempt to years, researchers have focused on developing
develop VFs, ATC-13 quickly became the stan- VFs either based on observed building-level
dard for assessing earthquake risk of buildings. damage data (Wesson et al. 2004) or from
analytical approaches. This approach combines and so this type of data is not available in suffi-
the experimental damage data and the observed cient quantity for developing the VFs. On the
damage data of different components of buildings other hand, insurance claims data (e.g., those
(Jayaram et al. 2012). from Northridge earthquake) can provide a large
quantity of damage data, which can be used for
GEM/PAGER Approach estimating VFs. But these data typically contain
Recently, US Geological Survey (USGS) has limited information about the buildings (typically
adopted Cook’s model for developing VFs for provides construction class – primarily materials
the PAGER system and for the GEM project of construction, e.g., steel, etc., height, and age).
(Jaiswal et al. 2013). This approach involves The earliest attempt in developing VFs fol-
identifying experts, constructing seed and target lowing this approach was by Steinbrugge
questions, eliciting expert judgments, and com- (1982). The VFs were developed based on lim-
bining multiple judgments. In this approach, ited damage data for a handful of earthquakes.
unlike the Delphi approach (adopted by Since insurance industry was not active in
ATC-13), the judgment of scoring is calibrated collecting damage data before the 1990s and
objectively and classical statistical test is engineers were focused on studying a few impor-
employed for hypothesis testing of the scores. tant damaged buildings following earthquakes,
This involves assigning weight to an expert Steinbrugge applied his experience and judgment
based on two separate scores: (1) calibration in order to develop the functions based on limited
score which measures the accuracy of the scores damage data that he collected over the years from
by the experts based on background reference small number of earthquakes. The developed VF
density functions of the scores and (2) informa- for single-family wood-frame buildings is shown
tion score which represents the degree with in Fig. 2. The results show that the vulnerability
which the distribution of the expert is concen- of residential wood-frame buildings is signifi-
trated with respect to some background distribu- cantly higher than that predicted by ATC-13 at
tion. The product of these two scores is the lower intensities. This function was used exten-
overall score of each of the experts. Since the sively by the insurance companies in the
weights can be recomputed objectively with 1980s–1990s as a guideline for managing earth-
new observation or addition of new experts, the quake risks. Later on, this function will be com-
weights can be recomputed in the future for pared with that based on extensive damage data
updating the VFs. from Northridge earthquake.
In this section, Northridge insurance claims
data, which lacks detailed information on the
Observation-Based Approach: lateral load resistance characteristics of build-
Vulnerability Functions from Building ings, will be used for developing the VF. The S
Damage Data other type of data, which would be a preferred
one, is the damage data that provides the detailed
The most direct approach of developing the VFs lateral load resistance characteristics of build-
would be based on building damage data ings. An example on this would be the data of
observed in historical earthquakes. This approach fractured connections of 167 welded steel
requires damage data of large number of build- moment-frame (SMF) buildings collected in the
ings for different construction classes, height, SAC joint venture project. This data was col-
age, and other characteristics, which are impor- lected in a very detailed and systematic way
tant in defining lateral load resistance of build- following Northridge earthquake, and those data
ings. If the damage data is based on detailed were used for developing empirical VFs for
inspection of buildings following earthquakes, building loss calculations in FEMA-351 Rapid
VFs can be developed accurately. But this Method (Bonowitz and Maison 2003). The pro-
requires significant amount of time and resources ject inspected about 18,000 connections and
found that 13 % of those had some kind of dam- the structure, year of construction of the build-
age. Although this approach provided very accu- ings, and the construction class. The scatter plot
rate information about the performance of welded of insurance loss data for single-family wood-
connections of SMF buildings, seismic loss esti- frame dwellings (SFD) constructed after 1975 is
mation requires repair cost of those connections shown in Fig. 3a, and the same for steel moment-
as well as other structural and nonstructural com- frame (SMF) buildings based on detailed inspec-
ponents of those buildings. Note that the value of tion of beam-column connections is shown in
the nonstructural components can be more than Fig. 3b for visual comparison of the trend and
70 % of total building value and so the repair cost uncertainty in these two data sets. The figure
of these components is significant, particularly at illustrates that the uncertainty conditioned on
low intensities. Since damage of nonstructural earthquake intensity (here PGA for SFD and
components were not collected, the empirical Sa(T = 2 s) for SMF) is very high. The vertical
VF developed in SAC project is primarily for stripe of data in the Fig. 3a represents the varia-
the structural components, and the VF will pro- tion of loss ratio in a specific zip code since
vide a low estimate of damage at low intensities. Northridge data is available only at the zip code
In addition, the uncertainty in the damage or loss level. Note that although the insurance claims
of buildings at a given intensity cannot be esti- data provides losses for content and additional
mated from this limited data, and this information living expenses (ALE), those are not considered
is essential for seismic risk calculations. in this section. The building loss data here
Since insurance claims data lacks detailed includes damage to the structural as well as
building information, the VFs developed from nonstructural components of buildings.
this data are for a broad category of buildings, Since the Northridge losses are available by
like low-rise wood-frame buildings constructed zip code, Wesson et al. (2004) estimated loss
in the era of moderately developed building code distribution by zip code by fitting a 2-parameter
(e.g., 1950–1975 in high seismic regions in Gamma distribution to the claims data. Since
California). This would lead to higher uncertainty insurance data does not capture the losses below
in the estimation of losses relative to the losses deductible, the maximum likelihood approach
for a specific building. The other issue with the was followed for estimating the parameters with
insurance claims data is that the data captures the constraint that the fraction of losses below
losses only after applying the deductibles deductible must be equal to the observed fraction
(known as gross loss). Hence the building loss is of no loss. The parameters of the distribution for
partially known from insurance claims data each of those zip codes are then fitted indepen-
(formally known as “censored” data), and this dently against the intensity measure to get the
missing information is critical for accurate esti- distribution of loss as a function of intensity mea-
mation of building vulnerabilities at low sure. The VFs based on this approach are shown
intensities. in Fig. 2. It is observed that there is significant
difference in the results between the vulnerability
Development of Vulnerability Functions functions based on Northridge insurance claims
Based on Claims Data data and that estimated earlier in ATC-13 follow-
In this section, insurance claims data will be used ing expert opinion or by Steinbrugge based on
for developing VFs, which estimates the proba- limited damage data.
bility distribution of building loss as a function of A more direct approach for the estimation of
ground-motion intensity. This approach is illus- the distribution parameter would be to estimate
trated here by using the Northridge claims data, those directly from the data by grouping those as
which is even today the best source of a large a function of intensity. In the earlier approach,
number of high-quality insurance claims data for since each of the parameters of the distribution
earthquakes. The data were available only by zip was fitted independently, the correlation between
codes and provided the claim amount, value of the parameters was not preserved. In addition, the
Seismic Loss
Assessment,
Fig. 3 Distribution of
damage ratio (ratio of
claims amount or repair
cost and building
replacement cost)
distribution as a function
earthquake intensity as
observed in Northridge
earthquake. (a) Post-75
single-family wood-frame
buildings. (b) Steel-frame
buildings inspected in SAC
project
regression analysis gave equal weight to all the the method of moments is not suitable for this
parameters for each of the zip codes irrespective type of data. Hence maximum likelihood
of the number of data points in the zip codes. approach will be followed to estimate the param-
Since Beta distribution is commonly employed eters (see Shome et al. 2012 for details).
for describing the earthquake loss distribution In general, the building VFs based on
and a five-parameter Beta distribution is deemed observed damage/loss data from one region can- S
to be the most suitable for modeling seismic loss not be directly extrapolated to other regions since
distribution, the fitting of five-parameter distribu- the building design and construction practice
tion model to the loss data will be discussed here. varies among different regions. Since the fre-
The five parameters of the distribution are: mean, quency of large-magnitude earthquakes is small
variance, probability of no loss (F0), probability and the occurrence of those events near large
of complete loss (F1), and the threshold (a) population center is even rarer, it is not possible
beyond which losses are considered to be 100 % to get a large number of damage data for different
by insurance companies. The parameters of the construction classes in order to develop VFs.
probability distribution can be estimated by using Also this approach needs high-quality data, i.e.,
two methodologies, namely, the method of building-level loss results like those obtained in
moments and the maximum likelihood approach. 1994 Northridge earthquake. The number of data
Since the insurance loss data is available after should be high for estimating VFs with high
applying the insurance limits and deductibles, confidence.
Maximum Likelihood Estimation (MLE) of below deductible is not available and is

Parameters reported as zero. If F0 is simply estimated as
The objective here is to estimate the parameters the fraction of data points with zero ground-up
of the five-parameter Beta distribution for the loss, F0 will be severely overestimated when
actual loss paid by the building owners for the deductibles are large particularly at low
repairing buildings (i.e., distribution for intensities. Note that the data is right censored
ground-up (GU) loss). The GU loss is computed as well because losses beyond the insurance
by adding the gross loss, the deductible, and the policy limits are truncated to the limit. This is,
losses beyond the limits if applied. But this however, much rarer since it only occurs at
cannot be done for the cases when the gross very large damage ratios induced by very
loss equals zero. This is a typical case of left- high-intensity GM.
censored data set where the ground-up losses The density function for damage ratio f(Di) at
are accurate above deductible, but the data a given intensity equals the following:
8
< F0 for Di ¼ 0
f ðD i Þ ¼ F0 þ ð1 F0 F1 ÞBPDF ðDi , a, bÞ for 0 < Di < a (1)
:
½F1 þ ð1 F0 F1 Þð1 BCDF ða, a, bÞÞ for Di a
where: where the parameters, F0 and F1, correspond to a

range of GM intensities for which loss data is
a = The threshold beyond which losses are accumulated for fitting and these parameters are
treated as 100 %. Note that this is governed estimated by maximizing L. The first term in the
by the underwriting policy of an insurance likelihood estimate corresponds to data points
company. where the observed damage ratio equals zero.
BCDF(Di,a,b) = Beta cumulative distribution This includes likelihood of damage ratio less
function (CDF) corresponding to sample dam- than deductible dedi, but reported as zero in the
age ratio Di (=the ratio of the loss to the claims dataset, and the true zero damage. The
building’s value). second term corresponds to data points where
BPDF() = Beta probability density function (PDF). the observed damage ratio equals 1. This corre-
a and b = Parameters for the Beta distribution of sponds to the likelihood of observing complete
the sample damage ratios for a GM loss and the losses above the threshold, a. The
intensity bin. third term corresponds to the rest of the data
points where the damage ratio is more than
Hence the likelihood of observing damage deductible, but less than a. The parameters of
ratios Dis equals: the distribution F0, F1, a, and b are obtained by
maximizing the likelihood of observing all the
" # damages, Di, which is shown in Eq. 2 as the
product of likelihood of the damages.
L ¼ ∏ F0 þ ð1 F0 F1 ÞBCDF ðded i , a, bÞ
Di ¼0 The results of fit of the five-parameter distri-
" # bution to the Northridge loss data at high-
∏ F1 þ ð1 F0 F1 Þð1 BCDF ða, a, bÞÞ intensity GM are shown in Fig. 4. The claims
Di a
" # data used for fitting the distribution is from a
number of insurance companies and the informa-
∏ F0 þ ð1 F0 F1 ÞBPDF ðDi , a, bÞ tion of the threshold (a) is not available from
Di ϵ ð0, aÞ
those companies. Hence for simplicity, it is
(2) assumed that a equals to 100 % damage.
Seismic Loss
Assessment,
distribution of loss at high-
intensity ground motion
between the data and
MLE fit
The results show that the fitting captures well the approach, the damage is calculated for the indi-
objective of probability mass at zero and 100 % vidual building assemblies, e.g., partition wall,
damage. Note that it would be difficult to estimate based on the response results from nonlinear
reliably all the parameters of the distribution time-history analysis of the model buildings to
when data is limited. In those cases, it is advisable estimate mean and variance of damage of the
to bring in other sources of data to define the loss model buildings as a function of GM intensity.
distribution. For example, one can use building The individual buildings in this approach were
collapse results – analytical and/or represented by deterministic simplified structural
experimental – to define F1 of a specific building models, where building components were ideal-
or a class of buildings. ized as springs and masses. Recently, Jayaram
et al. (2012) have developed a systematic simu-
lation approach based on the PEER-PBEE loss
Analytical Approach of Development of assessment framework to develop building-
Vulnerability Functions specific VFs from component-level damage at
each story of buildings. In this approach, the
There is limited building damage data from earth- total building loss is estimated by using the
quakes and that too from small number of con- response from detailed nonlinear time-history
struction classes. In addition, building damage analysis of buildings. The approach took into
characteristics based on the damage data from account the dependence of response on Sa at S
one region may not be extrapolated to other multiple periods, the correlation of losses
regions. Hence, VFs for majority of construction between different components to improve the
classes in most of the regions in the world can be estimation of uncertainty of losses, and on cap-
developed only from an analytical approach. tured the effects of epistemic and aleatory uncer-
Hence HAZUS (2003) first developed a compre- tainties of different parameters considered in loss
hensive analytical methodology to estimate the calculations, such as GMs, structural response
building seismic losses. HAZUS adopted capac- parameters, loss costs, etc. All these steps help
ity spectrum method to estimate response of to get the unbiased loss due to earthquakes and
generic buildings and estimated only mean loss quantify accurately the uncertainty in the loss
based on damage of different building compo- estimates.
nents. Later on, CUREE-Caltech (2001) developed The analytical loss calculation in general con-
the assembly-based vulnerability (ABV) method- sists of (1) repair cost of different building com-
ology for developing building-specific VFs. In this ponents associated with a damage state or cost
due to demolition, (2) damage state given a for the type and the severity associated with that
demand parameter (EDP), and (3) EDP given an state. This is predicted based on the fragility
intensity of GM. A brief description of these functions of the subsystem or component. The
parameters is given below and the details can be damage states are defined in HAZUS (2003) and
found in HAZUS (2003). these are none, minor (DS1), moderate (DS2),
extensive (DS3), and complete damage (DS4).
Building Components For example, moderate structural damage of con-
In order to carry out the seismic loss calcula- crete shear-wall buildings is defined by the diag-
tions, building components are divided broadly onal cracks on most of the shear-wall surfaces
into two categories: structural and and large diagonal cracks with spalling of con-
nonstructural. The loss estimation approach crete at the wall ends of some shear walls. The
either estimates damage of the individual com- moderate damage of the NSD components or
ponents (e.g., CUREE-Caltech 2001) or dam- subsystems represents large and extensive cracks
age of a group of components or subsystems of the partition walls requiring repair and
(e.g., HAZUS 2003). The structural component repainting as well as replacement of a few
(SD) resists gravity, earthquake, wind, and walls. The moderate damage state of the NSA
other types of loads. This includes components subsystem or component represents extensive
like columns, beams, load-bearing walls, falling of the suspended-ceiling tiles with discon-
etc. For buildings planned by design profes- nected and/or buckled ceiling support framing
sionals, the structural components are typically (T-bars) at a few locations.
designed and tailored to building-specific con-
figurations and loading conditions. The Repair Costs
nonstructural components include a large vari- The loss from earthquakes is associated with
ety of architectural, mechanical, and electrical the repair due to damage of different compo-
components, which can be further divided into nents or subsystems as well as building replace-
as either “drift-sensitive” (NSD) (e.g., ment cost associated with demolition and
non-load-bearing partition walls, exterior cur- collapse. The uncertainty in repair cost arises
tain walls, etc.) or “acceleration-sensitive” primarily due to lack of information about the
(NSA) components (e.g., suspended ceilings, component. The cost of repair works is gener-
HVAC systems, etc.). These components are ally significantly higher than the cost of new
usually neither designed by the professionals construction, and this is more significant for the
nor tailored to specific buildings even when structural components than the nonstructural
designed. These are tested under some generic components. For example, the repair of beam-
load conditions to satisfy the acceptability column connections of the structural compo-
criteria of the building codes under some ide- nent requires removing the suspended ceilings,
alized conditions. Hence the uncertainty in the inspection of the joints which may require non-
performance of structural components is rela- destructive testing, closure of part of the floor
tively low when designed and constructed by of the building, etc. All these expenses have to
professionals compared to those of the be incurred on top of the repair cost of the
nonstructural components. Drift is used to pre- connections. Insurance claims data also suggest
dict the damage states of the SD and the NSD relatively higher uncertainty in repair cost at
subsystems or components. Acceleration, on low intensities. In order to estimate the repair
the other hand, is used to predict the damage cost, it is generally assumed that the insurer
states of the NSA components. will pay the cost of repair or replacement of
the damaged buildings with the materials of
Damage States like kind and quality without any deduction
The damage state of each subsystem or compo- for depreciation, which is called replacement
nent represents a consequence in terms of repair cost method for valuation.
HAZUS-Based Vulnerability
The HAZUS methodology is developed to esti-
mate losses for different federal, state, and local
agencies for planning earthquake risk mitigation,
emergency preparedness, and many other uses.
Although the methodology can be used to esti-
mate losses for individual buildings, it is devel-
oped to estimate losses of portfolio of buildings
distributed over a large region. The methodology
provides an estimate of the probability of mean
loss for none, slight, moderate, extensive, and
complete damage of building components at a
given level of ground shaking. The building dam-
Seismic Loss Assessment, Fig. 5 Mean fragility func-
age characteristic is defined by the following tions and its uncertainty for extensive damage of
curves: (1) fragility curves that describe the prob- nonstructural drift-sensitive subsystem (NSD)
ability of reaching or exceeding different states of
damage given the peak-building response and
(2) building capacity (pushover) curves that are The component fragility functions are esti-
used to determine the peak-building response for mated from the results of laboratory experiments
the demand spectrum. or from observed damage data. Since there is
limited observed damage data or experimental
Fragility Functions data, past studies recommended fragility func-
The fragility curve is defined by a median tions primarily based on engineering judgment
value of the engineering demand parameter (e.g., HAZUS). Recently, ATC-58 (2011) has
(EDP), which is, for example, spectral dis- made significant effort to develop a suite of fra-
placement in HAZUS and maximum interstory gility functions so that loss calculations can be
drift ratio (IDR) in Jayaram et al. 2012, and by carried out in day-to-day building design work.
the variability associated with the estimation of Also Network for Earthquake Engineering Sim-
that damage state for that EDP. The median ulation (NEES) Program is actively pursuing
EDP corresponds to the threshold of the dam- experimental research (http://www.nees-
age state. The curve is assumed to follow the nonstructural.org/) to develop fragility functions
cumulative lognormal distribution function as for different nonstructural components of build-
shown below: ings. All these research provide a range of fragil-
ity functions giving an estimate of the mean and
the standard deviation of the fragility functions. S
lnðEDP=mÞ
PðDS dsi jEDPÞ ¼ F (3) The mean fragility function for extensive damage
b
of the generic NSD subsystem, which is an
assembly of a number of components such as
where m denotes the median value of EDP at partitions, exterior glazing systems, etc., is
which building component or subsystem reaches shown in Fig. 5. The epistemic uncertainty in
the threshold of the damage state dsi, b denotes the estimation of the fragility function for that
the corresponding dispersion, and F() is the damage state is shown by the 90 % confidence
cumulative density function for standard normal band.
distribution. The dispersion considers (1) uncer-
tainty in the damage state threshold, (2) variabil- Building Response Calculations
ity in the capacity (response) of buildings, and The building responses in HAZUS are estimated
(3) uncertainty in building response or EDP due based on the capacity spectrum method (CSM).
to the spatial variability of ground motion. The building capacity curve (also known as a
Seismic Loss
Assessment,
Fig. 6 Example building
capacity curve, demand
spectrum, and building
response for wood-frame
buildings in California for
input spectrum from M7
earthquake on Hayward
fault at a site close to the
fault
pushover curve) describes the lateral load resis- response particularly at higher intensities.
tance of a building as a function of its character- Hence this approach may not be suitable for
istic lateral displacement. The curve is derived mid-rise and high-rise buildings and for types of
from the results of static-equivalent base shear buildings at high intensities.
and roof displacement and is defined by three
control points: (1) design capacity, (2) yield Building Loss Calculations
capacity, and (3) ultimate capacity as shown in The fragility curves define all the four damage
Fig. 6. It is assumed that the building capacity states for each of the three building subsystems
will have a range of possible properties and those based on the EDPs estimated from the intersec-
are lognormally distributed as a function of the tion of the capacity curve and the seismic demand
ultimate capacity. The peak-building response is curve. It is assumed that the nonstructural dam-
the intersection of the building capacity curve age states are independent of structural damage
and the response spectrum, which defines the states. The repair cost is estimated based on the
seismic demand. Since buildings go beyond its damage of different subsystems of buildings.
elastic limit, the demand spectrum defined by the Note that although the results from HAZUS loss
elastic damping is reduced due to the higher estimation methodology are intended to assess
effective damping. The effective damping the socioeconomic impact of earthquakes in a
includes both the elastic damping and the hyster- region, the result has a number of limitations.
etic damping associated with the post-yield Earthquake impact on a society is a very complex
cyclic response of buildings. Figure 6 illustrates process and the economic consequences cannot
the capacity curve, demand spectrum curve, and be considered in a simplistic approach. The dollar
the peak-building response at the point of inter- losses are calculated for all the three coverages:
section of those curves. The main drawback of (1) building repair and replacement costs, (2) con-
CSM is that the MDOF structures are idealized by tents losses, and (3) additional living expenses
an equivalent SDOF system using an increased (ALE) for residential occupancies or business
period and damping in order to estimate the seis- interruption (BI) losses for commercial or indus-
mic demand of the MDOF structures by the trial occupancies. In this chapter, only building
capacity spectrum approach. In addition, this repair loss calculations are discussed for brevity.
method is a nonlinear static procedure, whereas This loss considers repair and replacement due to
the nonlinear time-history analysis generally pro- both structural and nonstructural damages, and
vides a more accurate estimation of building this is calculated as the product of the probability
of a component in a given damage state and the function to estimate the probability of that assem-
mean repair costs for that damage state, summed bly in different damage states as a function of
over all the building subsystems and the damage structural response. Test data from the CUREE
states as given below: Project, experimental data from other projects,
and observation of damage from different earth-
quakes are used to develop the fragility functions.
3 X
X 5
Li ¼ RCi Pijk DRijk (4) The losses were estimated for repair of damage of
k¼1 j¼2 different assemblies following standard construc-
tion cost-estimation principles. This approach
where RCi is the replacement cost of an entire estimates mean and uncertainties (which is also
model building i (e.g., wood frame), Pijk is the improvement over HAZUS) in loss estimate by
probability of model building i and subsystem capturing the uncertainties in: ground motion,
k (e.g., NSD) being in damage state j (e.g., mod- structural characteristics, component damageabil-
erate), and DRijk is the mean damage ratio for ity, and repair costs. This approach followed a
building i and subsystem k in damage state j. simulation approach in order to quantify the
The ratio of loss of building Li and the replace- (aleatory) uncertainty in the loss results. The
ment cost of RCi is the mean damage ratio (MDR) losses are calculated for individual components
of the building i for the intensity of GM at which of the buildings instead of a group of components
the losses are calculated. This calculation is like in HAZUS to get the building loss.
repeated at different intensities to develop the The structural response which is calculated
VF(s) for seismic loss calculations of an individ- from the building simulation results is used as an
ual building or a portfolio of buildings. Note that input to the fragility function to simulate the dam-
although HAZUS considers only mean repair age state of the component assemblies. The com-
cost, simulation-based approach, which will be ponents are idealized as an assembly of springs
discussed later on, considers the uncertainty in and masses to calculate the response. The masses
repair cost as well to calculate the uncertainty in represent the building components that have sig-
the VFs. Figure 2 shows the VF following nificant weight, such as the walls, ceiling, floor,
HAZUS approach of residential light wood- etc., and the springs represent building elements,
frame buildings constructed during the period of such as the walls, beams, etc., that resist deforma-
1940–1975 following a modern building code tion. The idealization took into account for
(e.g., Uniform Building Code) in the high seismic strength and stiffness degradation of the compo-
zone of California for a suite of response spec- nents, along with many other complex nonlinear-
trum for different magnitude of earthquakes at ities, to represent accurately the spring element.
different distances based on the disaggregation The response parameters used in this approach are
of USGS hazard results in Los Angeles. maximum interstory drift ratio (IDR), peak floor S
acceleration (PFA), peak transient horizontal shear
Assembly-Based Vulnerability (ABV)/CUREE- strain, peak member forces, and residual drift ratio.
Caltech Approach The total repair cost is calculated by summing
In this approach (CUREE 2001), the nonlinear the cost of repairing the damaged assemblies. The
building response was calculated based on cost of repairing a damaged assembly includes
nonlinear time-history analysis of model buildings labor costs, material requirements, debris
representing the best estimate of mass, stiffness removal, and equipment rental. The cost calcula-
(load-deformation characteristics),damping, and tion includes the cost of repair of the undamaged
building properties required for the analysis. The assemblies (or line of sight) due to damage of an
buildings are modeled as a collection of assem- assembly. The cost also considers the case of
blies of components, such as a gypsum wallboard collapse of the building. The cost is simulated
partition, a floor diaphragm, beams, etc., and each from the distribution of the cost for repair of
of the assembly is represented by a fragility each damaged assembly. The simulation
provides total repair or replacement cost at a earthquake engineering (PBEE) framework. The
given level of seismic shaking intensity. The pro- steps involved in the procedure are quantifying
cess is repeated many times at a given intensity to ground-motion hazard using a vector of spectral
estimate the probability distribution of cost for accelerations; predicting building response
calculating the building-specific VF. The plot of parameters such as story drifts, floor accelerations,
MDR and coefficient of variation (CoV) of typi- and residual drifts under the quantified hazard;
cal wood-frame buildings of average size as a estimating structural collapse and demolition;
function of PGA based on CUREE-Caltech pro- and calculating story-wise losses based on the
ject is shown in Fig. 2. The plot shows that the responses. The total building loss is calculated
prediction of MDR and uncertainty are signifi- by summing the story losses. The procedure cap-
cantly less than those observed in Northridge tures the effects of epistemic and aleatory uncer-
earthquake. tainties in different parameters considered in loss
calculations, such as ground motions, structural
PEER-PBEE-Based Approach response parameters, loss costs, etc., in order to
Recently, Jayaram et al. (2012) have developed a quantify the uncertainty in the final loss estimate.
generic procedure for developing the building The mathematical framework of the develop-
VFs by following a systematic simulation ment of the mean vulnerability function based on
approach based on the PEER performance-based this approach is given below:
ðððð

EðLjSa ðT 0 Þ ¼ L:dGðLjDMÞjjdGðDMjEDPÞjjdGðEDPjSa ðTÞjjdGðSa ðTÞÞjSa ðT 0 Þj (5)
where L is the building loss; Sa(T0) is the spectral model of a structure. The fundamental period is a
acceleration at a reference period; E(L|Sa(T0)) is good choice when the structural performance is
the mean loss at Sa(T0); dG(L|DM) denotes the reasonably elastic and dominated by the first
derivative of the probability of exceedance of the mode, but it is not a good choice for tall buildings
building loss given a damage measure (DM); in high seismic regions. The response of these
dG(DM|EDP) is the derivative of the probability buildings is nonlinear when damaged during
of exceedance of the DM given an EDP (e.g., earthquakes, and the response is determined by
story drift ratio); dG(EDP|Sa(T)) is the derivative multiple frequencies. The equation shows that the
of the probability of exceedance of the EDP given structural response is predicted based on a vector
a vector of spectral acceleration, Sa(T); and of spectral accelerations, but the vulnerability is
dG(Sa(T)|Sa(T0)) is the derivative of the probabil- represented as a function of scalar intensity mea-
ity of exceedance of spectral accelerations at sure for simplicity of use, but without sacrificing
multiple periods, Sa(T), given Sa(T0). Note that any accuracy.
the reference period, denoted here T0, is not nec- The variance of the vulnerability function can
essarily the fundamental period of the analytical be estimated as follows:
ðððð
Var ðLjSa ðT 0 ÞÞ ¼ ðL EðLjSa ðT 0 ÞÞ2 jdGðLjDMÞjjdGðDMjEDPÞjjdGðEDPjSa ðTÞÞjjdGðSa ðTÞÞjSa ðT 0 Þj
(6)
The above integrals Eqs. 5 and 6 can be evaluated (MCS). When the number of variables is large,
conveniently by using Monte Carlo Simulation particularly when the intensity measure is of
Seismic Loss
Assessment,
Fig. 7 Response spectrum
conditioned on
Sa(T0) = 0.22 g at
T0 = 2.5 s. (a) Exponential
of the conditional mean of
lnSa(T). (b) Conditional
standard deviation of
lnSa(T)
large dimension (i.e., vector of Sa), it is efficient is important to select the records in such a way
and convenient to use MCS. The MCS approach that the distribution of Sa(T) at all the periods
involves simulating all the random variables in conditioned on Sa(T0) of the scaled records is
Eqs. 5 and 6 (which are Sa(T), EDP, DM, and L) appropriate. The mean and variance of a typical
and then computing the mean and the variance of conditional spectrum in Los Angeles are shown
L for a wide range of Sa(T0) values. The MCS in Fig. 7. It is observed that variance is zero at the
approach provides flexibility and allows consid- conditioning period T0, but the variance increases
eration of accurate but complex models for dif- rapidly as the period of Sa moves further from T0. S
ferent parameters as well as correlation between In general, the risk of exceeding Sa(T0) at a site is
these parameters. It also allows for the rigorous caused by earthquakes of several possible com-
consideration of collapse and demolition. binations of M-R. Since the shape of response
spectrum depends on the M-R of earthquakes,
Conditional Spectral Accelerations (Sa(T)|Sa(T0)) there are different possible shapes of response
Although it is quite common to scale a suite of spectrum conditioned on Sa(T0) due to the differ-
ground-motion records to Sa(T0) for nonlinear ent combinations of M-R of earthquakes that can
analysis of buildings for estimating EDPs, studies strike a site. The distribution of M-R conditioned
have shown that this introduces bias in the on Sa(T0) at a building site can be obtained by
results – particularly for mid-rise to high-rise deaggregation of hazard for the site, and this
buildings (e.g., Haselton 2009). This is due to different shaped response spectra should be con-
the fact that the response of these buildings sidered in the simulation of building response to
depends on Sa at periods other than T0. Hence it improve accuracy. In the simulation of loss
Seismic Loss
Assessment,
Fig. 8 Response spectra
for an event (MW = 7 and
R = 10 km) scaled to the
median Sa(T0) = 0.12 g of a
set of 98 records and the
median spectrum for the
set. The median is
compared with the average
of the four NGA
attenuation models
(Bozorgnia et al. 2014).
The median spectra are
closely matched by
carefully selecting the set of
records
results, first the M and R pairs are sampled based Hence the purpose of computing these EDPs by
on the distribution from the deaggregation nonlinear time-history analysis for different
of hazard results, and subsequently, spectral recorded ground motions is to develop regression
acceleration at multiple periods, Sa(T), condi- models to predict EDPs as functions of Sa(T)
tioned on Sa(T0), M, and R, is simulated. while carrying out simulations.
An example of simulated response spectrum During earthquakes, there is a possibility of
is shown in Fig. 8. collapse of structures. Two different types of
structural collapses can be observed during the
Calculation of EDP analysis of buildings. The first type of collapse,
The EDPs required in this approach for loss cal- simulated collapse, is the collapses observed in
culations are interstory drift ratio (SDR), peak the analyses. This occurs due to excessive lateral
floor accelerations (PFA), and residual drift drifts and deterioration that lead to loss of the
ratio (ResDR). The SDR values are used to pre- gravity load resistance of structures. This is
dict the damage of SD and NSD components at modeled using a collapse probability curve,
each story level, the PFA values are used to which provides the probability of collapse of
predict those of the NSA components, and buildings as a function of Sa(T0). The other type
ResDR is used to determine if a building would is referred to as virtual collapse, which captures
have to be demolished after an earthquake. The those cases when a structure in nonlinear analyses
EDPs are first calculated based on nonlinear time- undergoes a very large drift without collapse due
history analysis of buildings for a suite of ground- to limitations in numerical modeling. Both these
motion records. But it is nearly impossible to collapses are considered here for developing vul-
select records that satisfy conditional distribution nerability functions. The collapse probability
of Sa at multiple periods (which is can be more curve is developed using logistic regression,
than ten for tall buildings) (see, e.g., Jayaram which is commonly used for regressing binary
et al. 2012). Hence it is efficient and also accurate data (Shome and Cornell 2000), and can be
to fit regression models to predict independently expressed as follows:
the different EDPs at different stories and use
these fitted models subsequently to estimate the
p
EDPs for the simulated response spectra, which ln ¼ b0 þ b1 Sa ðT 0 Þ þ e (7)
have the right conditional distribution of Sa(T). 1p
Seismic Loss
Assessment,
Fig. 9 Collapse fragility
curve for a modern 20-story
steel moment-frame
building in Los Angeles
where p = probability of collapse, b0 and distribution (the multivariate nature is assumed).

b1 = regression coefficients, and e = regression The distribution is defined by the means and
error. The collapse fragility curve for a 20-story dispersions of the capacity based on the fragility
modern steel-frame building is shown in Fig. 9. functions and the correlation between those
In addition to fitting the EDPs at each story capacities.
for estimating the mean and standard deviation The repair costs conditioned on DM are also
of the loss results conditioned on Sa(T), the modeled as a multivariate lognormal random vari-
correlation between different EDPs at the same able with the median and dispersion and correla-
story and between EDPs at different stories tions of repair costs between different damage
needs to be estimated to jointly simulate the states over the height of buildings. The losses of
EDPs. The extent of these correlations is com- different components across different damage
puted using the regression residuals. It is assumed states are expected to be correlated because of
that the EDPs follow multivariate lognormal common factors involved in repairs, such as the
distribution. contractor hired for the repairs and the material
costs. The parameters for the distribution of repair
Damage Measure and Repair Cost cost of buildings are based on the study by Ramirez
In this approach, building component losses are and Miranda (2009) on the cost of contractors for a
simulated based on the repair costs associated large number of building projects. Shome
with different damage states, which are obtained et al. (2014) have shown how the information of S
from the component fragility functions. Since the repair cost of the contractors can be simulated to
damage states are defined to be discrete, the sim- estimate the lognormal parameters of repair cost
ulation of correlated discrete damage states is a for a specific building or a group of buildings.
difficult task. It is shown by Shome et al. (2014) The total loss is computed by summing up the
that the correlation between discrete damage sample components losses at each story. This is
states can be easily established by representing normalized by the building value to get the dam-
the fragility functions using a “damage-capacity” age ratio for the building. The scatter plot of
formulation. In this approach, instead of defining simulation of a 20-story modern steel building
the correlation between the discrete damage in Los Angeles, California, is shown in Fig. 10
states, the correlation between the damage capac- as a function of Sa(T0). The mean and the CoV of
ity of different components is defined. The dam- the damage ratios are then computed by binning
age capacities of all the components can be the data points into Sa(T0) bins to obtain the
jointly simulated from a multivariate lognormal building VF.
Seismic Loss
Assessment,
Fig. 10 Damage ratios of a
modern 20-story SMF in
Los Angeles for all the
simulated Sa(T) vectors
corresponding to a wide
range of Sa(T0) values
Seismic Loss
Assessment,
Fig. 11 Impact of
different sources of
epistemic uncertainties in
the loss results of a modern
20-story steel moment-
frame building in Los
Angeles. (a) Average
annual loss (AAL). (b)
500-year loss
Seismic Loss Results and Uncertainty in developed by Jayaram et al. (2012) can be used
Those Results to account for epistemic uncertainties in the vul-
nerability functions by developing a suite of func-
The VFs developed in the previous sections are tions representing the uncertainty. The losses are
used to carry out seismic risk assessments of then sampled from these functions to compute the
buildings. The VFs are integrated with the hazard epistemic uncertainty of different loss matrices
curves to compute the AAL and the return period used for building risk assessment like AAL, EP
or EP losses. As discussed, the methodology loss, etc. Figure 11 illustrates the normalized
64 % confidence interval (CI) of the AAL and the results cannot be extrapolated from one region to
500-year return period loss of a modern 20-story the another because of regional building design
steel moment-frame building in Los Angeles. and construction practices, which are generally
The relative extents of contributions of the unique. Hence this chapter described develop-
epistemic uncertainties in vulnerability and ment of building vulnerability functions
hazard to the total epistemic uncertainty based on expert opinion. It is found that this
would, however, depend on whether the loss approach can provide significantly biased results
assessment is carried out for a single building or and so this is used rarely nowadays. This
a portfolio of buildings. For a single building, as approach, however, would be used for assessing
seen in the figure, the VF uncertainty has a large seismic risk in the future when time and resources
contribution to the total uncertainty. For a port- are limited.
folio of buildings, however, the uncertainty in Over the years, different analytical approaches
losses across different locations will be mostly have been developed and only a few of those have
independent and will average out. The hazard been discussed in this chapter due to limitations
uncertainty, however, is highly correlated across in space. This chapter discussed in detail devel-
multiple locations and will not average out. opment of vulnerability functions following
Therefore, for a portfolio, hazard uncertainty capacity spectrum method by HAZUS,
will have higher contribution to the total assembly-based vulnerability, and a comprehen-
uncertainty. sive approach following Jayaram et al. (2012)
Analytical approaches have de facto become stan-
dard in the recent years for seismic risk calcula-
Summary tions. Although these approaches require a large
number of parameters, researchers are actively
Seismic loss assessment of buildings plays an estimating those based on laboratory experiment
important role in managing earthquake risk for and observed damage data. It is well known that
public agencies, insurance companies, as well as there is significant uncertainty in the estimation of
individual stakeholders. The loss assessment pro- those parameters vis-à-vis VFs based on those
cedure has evolved over the years – from expert parameters, but the uncertainty in VF can be easily
opinion to very rigorous analytical approach. estimated based on the analytical approaches. The
Since the analytical procedure is very complex importance of consideration of epistemic and ale-
requiring accurate information of a large number atory uncertainty and the methodology in order to
of parameters, there is significant uncertainty in estimate those in building vulnerability has also
accurate estimation of building losses. Hence loss been discussed here.
estimation based on observed damage character-
istics of buildings in historical earthquakes only S
can provide accurate loss results. This chapter has Cross-References
discussed in detail the development of building
vulnerability functions (VF) based on observed ▶ Analytic Fragility and Limit States [P(EDP|
loss data. The data, particularly those that are IM)]: Nonlinear Dynamic Procedures
from insurance companies, can be censored ▶ Empirical Fragility
because of insurance deductibles. This is tradi- ▶ Insurance and Reinsurance Models for
tionally dealt with maximum likelihood approach Earthquake
and this approach has been described here. But ▶ Seismic Risk Assessment, Cascading Effects
this approach requires a large number of very ▶ Seismic Vulnerability Assessment: Masonry
high-quality building-specific loss data, which is Structures
generally not available. The other disadvantage ▶ Seismic Vulnerability Assessment: Reinforced
of data-based vulnerability functions is that the Concrete Structures
2896 Seismic Monitoring of Volcanoes
References performance-based earthquake engineering, report no

171. The John A. Blume Earthquake Engineering Cen-
ATC-13 (1985) Earthquake damage evaluation data for ter, Stanford University, Stanford
California. Applied Technology Council, Redwood Shome N, Bazzurro P (2009) Comparison of vulnerability
City of a new high-rise concrete moment frame structure
ATC-58 (2011) Guidelines for seismic performance using HAZUS and nonlinear dynamic analysis. In:
assessment of buildings (75 % draft). Applied Tech- 10th international conference on structural safety and
nology Council, Redwood City reliability (ICOSSAR), Osaka
Benjamin JR, Cornell CA (1970) Probability, statistics Shome N, Cornell CA (2000) Structural seismic demand
and decisions for civil engineers. McGraw-Hill, analysis: consideration of collapses. In: 8th ASCE
New York specialty conference on probabilistic mechanics and
Bonowitz D, Maison BF (2003) Northridge welded steel structural reliability, St Luis
moment-frame damage data and its use for rapid loss Shome N, Jayaram N, Rahnama M (2012) Uncertainty and
estimation. Earthquake Spectra 19(2):335–364 spatial correlation models for earthquake losses. In:
Bozorgnia Y, Abrahamson NA, Atik LA et al (2014) 15th world conference on earthquake engineering
NGA-West2 research project. Earthquake Spectra. (WCEE), Lisbon
doi:10.1193/072113EQS209M Shome N, Jayaram N, Krawinkler H et al (2014) Loss
Cornell CA, Krawinkler H (2000) Progress and challenges estimation of tall buildings designed for the PEER
in seismic performance assessment. PEER Center tall building initiative project. Earthquake Spectra
News 3(2), http://peer.berkeley.edu/news/2000spring/ (accepted for publication)
CUREE (2001) Improving loss estimation for woodframe Steinbrugge KV (1982) Earthquakes, volcanoes and
buildings: CUREE-Caltech woodframe project. In: tsunamis – an anatomy of hazards. Skandia America
Porter KA, Beck JL, Seligson HA, et al (eds) Element Group, New York
4, economic aspects. Consortium of Universities for Wesson RL, Perkins DM, Leyendecker EV et al (2004)
Research in Earthquake Engineering, Richmond Losses to single-family housing from ground motions
FEMA 355F (2000) State of the art report on performance in the 1994 Northridge, California, Earthquake.
prediction and evaluation of steel moment-frame Earthquake Spectra 20:1021–1045
buildings. Federal Emergency Management Agency,
Washington, DC
Haselton CB (ed) (2009) Evaluation of ground motion
selection and modification methods: predicting
median interstory drift response of buildings, PEER
report 2009/01. PEER GMSM, Pacific Engineering Seismic Monitoring of Volcanoes
Research Center, University of California, Berkeley
HAZUS-MH MR3 (2003) Multi-hazard loss estimation Glenn Thompson
methodology: earthquake model. Department of
Homeland Security, Washington, DC
School of Geosciences, University of South
III (2013) Earthquakes: risk and insurance issues. Insurance Florida, Tampa, FL, USA
Information Institute. http://www.iii.org/issue_updates/
eathquakes-risk-and-insuance-issues.html. Accessed
8 Nov 2013
Jaiswal KS, Wald DJ, Perkins D et al (2013) Estimating
Synonyms
structural collapse fragility of generic building typol-
ogies using expert judgment. In: 11th international Volcano-seismic monitoring
conference on structural safety and reliability
(ICOSSAR), New York
Jayaram N, Shome N, Rahnama M (2012) Development of
earthquake damage functions for tall buildings. Earth- Introduction
quake Eng Struct Dyn 41(11):1495–1514
PEER-2011/05 (2011) Case studies of the seismic perfor- Seismicity beneath a volcano usually increases
mance of tall buildings designed by alternative means.
In: Moehle J, Bozorgnia Y, Jayaram N et al (eds) Task
before an eruption because magma and volcanic
12 report for the tall buildings initiative. Pacific Earth- gas must first force their way up through fractures
quake Engineering Research Center Report, Berkeley and passageways. When magma and volcanic
Porter KA, Kiremidjian AS, LeGrue JS (2001) Assembly‐ gases or fluids move, they will either cause
based vulnerability of buildings and its use in perfor-
mance evaluation. Earthquake Spectra 17(2):291–312
rocks to break or cracks to vibrate. When rocks
Ramirez CM, Miranda E (2009) Building-specific loss break, high-frequency earthquakes are triggered.
estimation methods & tools for simplified When cracks vibrate, either low-frequency
Seismic Monitoring of Volcanoes 2897
earthquakes or a continuous shaking called vol- understanding of the volcano or lead to improved
canic tremor, which can last from minutes to monitoring tools.
days, occurs. Much of the equipment and software used for
Volcanic earthquakes often occur in swarms, volcano-seismic monitoring were originally
which are clusters in time and space of similar developed for regional earthquake monitoring.
earthquakes without an obvious mainshock. Vol- The main difference is that the software needs
cano seismologists look for changes in the rate, to be augmented because of the diverse range of
size, and location of earthquakes and for the seismic signals recorded at volcanoes, many of
occurrence of swarms and tremor to forecast which elude standard earthquake detection and
eruptions and to evaluate whether a volcanic location techniques. These signals, and the
eruption is intensifying or ending. Volcanic haz- framework in which they are commonly
ards including explosive eruptions, rockfall, interpreted, are discussed in the next section.
pyroclastic flows, and lahars also cause The sections that follow describe volcano obser-
ground vibrations and can be identified by their vatories including the operations room which is
seismic signatures. This makes it possible, in the command center during a crisis, the design of
principle, to detect them as they happen and volcano-seismic networks, the real-time seismic
issue warnings. monitoring and data analysis tools found in
This entry discusses volcano-seismic monitor- most observatories, and the challenges of
ing from the viewpoint of a scientist leading a maintaining this infrastructure and managing
seismic monitoring program at a volcano obser- data effectively. In the final section, other moni-
vatory. Seismic monitoring records continuous, toring techniques employed at observatories are
high-sample-rate data on the internal state of the briefly discussed.
volcano. Other monitoring techniques require
manual labor to collect or process data, rely on
daylight and good weather conditions, only
detect volcanic activity once it has reached the Volcano-Seismic Signals
surface, or have a low sample rate. For these
reasons, seismic monitoring is the backbone of Volcano seismologists are most interested in sig-
most volcano observatories. A volcano-seismic nals which are anomalous, because they have
monitoring program comprises of a seismic net- high amplitudes, characteristic waveforms, or
work, a telemetry system, data acquisition, and unusual frequency behavior. Given the variety
alarm, analysis, and archival systems. At large of volcanoes around the world, with staggeringly
observatories, personnel may include electronics different eruptive styles, magma compositions,
engineers, volcano seismologists, seismic ana- and viscosities, it is perhaps remarkable that
lysts, software developers, and network adminis- some common types of seismic signals are S
trators. At small observatories, one person may observed. These signals can be broken down
cover most or all of these roles. into three broad categories: volcanic earthquakes
A volcano observatory has to be able to pro- (“events”), continuous signals, and surface sig-
cess large quantities of streaming data, detect nals. Each of these is discussed below. Any vol-
changes in the volcanic system immediately, cano may exhibit some or all of these at some
and respond without having to do a lot of manual time and perhaps may also exhibit exotic signals
analysis. This requires high levels of automation unique to that volcano. Classification is important
and systems engineering. While top priority is because each type of signal may represent a dif-
interpreting real-time data using established ferent physical mechanism. However, classifica-
knowledge from the scientific field of volcano tion is problematic.
seismology, other important roles are trouble- By analyzing the spatial and temporal patterns
shooting data problems and conducting between different signal types and volcanic activ-
(or promoting) research that may lead to a better ity, a greater understanding of a particular
volcano may be revealed. There are common earthquake events with a particularly monotonic
patterns of behavior, gleaned from analyzing the appearance and an exponentially decaying tail.
seismic data from many volcanic eruptions. LP earthquakes lack discernable P or S phases;
These are encapsulated in the Generic Volcanic consequently, most are not located. Evidence
Earthquake Swarm Model (McNutt 1996), a from Soufrière Hills Volcano (Montserrat) that
sequence observed at many volcanoes, the main they are often associated with venting from the
features of which are volcano-tectonic earth- surface of the dome and often trigger rockfall
quake swarms, followed by low-frequency earth- suggests they may originate at depths of less
quake swarms, tremor, and eruption. than 1 km. LP earthquake focal mechanisms
For a more detailed discussion of volcano- reveal a volumetric component which is evidence
seismic signals and many other topics mentioned of a fluid phase. LP earthquake swarms in volca-
in this chapter, Wasserman (2012) is highly nic systems are often associated with eruptions or
recommended. intrusions and are believed to be due to processes
such as pressure-induced oscillations of fluid-
Volcanic Earthquakes filled cracks in magmatic and hydrothermal
systems.
Volcano-Tectonic Earthquakes
Volcano-tectonic (VT) earthquakes are tectonic Hybrid Earthquakes
earthquakes that occur near active volcanoes. The Hybrid earthquakes have a high-frequency
physical mechanism is shear failure and slip on a P-wave onset, typical of a VT earthquake,
fault plane triggered by magma ascent or the followed by a long-period tail. They may repre-
relaxation that occurs after magma is erupted. sent triggering of an LP earthquake by a VT
VT earthquakes have clear P- and S-waves earthquake. They typically occur in swarms and
(if recorded with a good signal-to-noise ratio) may be indicative of magma intrusion or
and high-frequency content (>5 Hz). VT earth- extrusion.
quakes can be located in the same way as other
tectonic earthquakes, i.e., by using the differen- Low-Frequency Earthquakes
tial travel times between the P and S phases Collectively, long-period and hybrid earthquakes
across a seismic network. However, most VT are referred to as low-frequency earthquakes.
earthquakes are small (ML = 0.5–1.5), and There may be a continuum between long-period
poor signal-to-noise ratio often prevents identifi- and hybrid earthquakes.
cation of phases, so in practice many (perhaps
most) VT earthquakes cannot be located. Veloc- Deep Long-Period Earthquakes
ity models for volcanoes are often poorly deter- Some volcanoes produce low-frequency earth-
mined, making absolute depths unreliable. quakes (e.g., 20–40 km depth), particularly in
Trends in relative depths are a useful diagnostic the early stages of unrest. These have been called
tool, however, and may indicate the rise of deep long-period (DLP) earthquakes. These have
magma toward the surface. VT earthquakes fre- emergent P and S phases, are rich in frequencies
quently occur in swarms that consist of many below 5 Hz, and are inferred to represent move-
similar-sized events and do not occur in ment of deep-seated magma and associated fluids
mainshock-aftershock sequences where one sin- in the mid-to-lower crust. They look like VT
gle event dominates. earthquakes with the high frequencies filtered
out, perhaps because they occur in a highly atten-
Long-Period Earthquakes uating region.
Long-period (LP) earthquakes are unique to vol-
canic regions. They have emergent onsets and a Explosion Quakes and Very-Long-Period Signals
narrow frequency range with a peak frequency Explosion quakes are signals that accompany
typically from 1 to 4 Hz. Tornillos are LP Strombolian or other (larger) explosive eruptions.
These signals are identified by the occurrence of Benoit and McNutt (1996) examined the
an airwave which is caused by expanding gas reports of over 600 swarms to compile the Global
accelerated at the vent exit. This wave mainly Volcanic Earthquake Swarm Database. They
travels through the air with the typical speed of identify three main types of volcanic earthquake
sound (330 m/s at 20 C). swarm. Type 1 swarms begin before an eruption
With the advent of broadband volcanic seis- and have a mean (and mode) duration of 8 days.
mology, very-long-period (VLP) signals have Type 2 swarms begin coincident with, or during,
been observed in seismograms from some volca- an eruption. Type 3 swarms (39 % of records)
noes. Many of these are broadband versions of were not associated with eruptions and have a
explosion quakes. VLP signals may be produced mean duration of 3.5 days and a mode of
by the rapid expansion of a large gas volume at 1.5 days. The most common depth for swarms is
shallow depth within the conduit. The gas expan- 2–3 km. Their study does not distinguish between
sion might result from shallow gas coalescence different types of volcanic earthquakes. They
and expansion or from expansion of a gas slug suggest it may also be biased by the
formed at greater depth. underreporting of swarms not associated with
eruptions.
Continuous Signals
Tremor
Swarms Volcanic tremor is a narrowband (usually
Earthquake swarms are sequences of earth- 1–4 Hz), continuous vibration thought to be due
quakes closely clustered in space and time with- to sustained subsurface movement of magma or
out a single mainshock. Volcanic earthquakes volatiles and is often observed before explosive
often occur in swarms, whereas nonvolcanic eruptions. It may last from a few minutes to
earthquakes usually follow a mainshock- months in duration. Tremor has similar spectral
aftershock sequence. The Gutenberg-Richter characteristics as low-frequency earthquakes.
law describes the relative frequency of occur- Harmonic tremor shows one or many regularly
rence of earthquakes of different magnitudes, spaced overtones in addition to a fundamental
and this is encapsulated in a parameter called frequency. Sometimes spectral peaks in har-
the b-value (see Sanchez et al., “▶ Frequency- monic tremor glide upward (or downward) in
Magnitude Distribution of Seismicity in Volca- frequency over as little as a few minutes. Erup-
nic Regions,” this volume). Volcanic earthquake tion tremor, a continuous vibration coincident
swarms typically have b-values higher than one with explosive eruptions, has a wider frequency
(most of the energy is in small earthquakes), range (0.5–10 Hz). Dome collapses, lahars,
whereas mainshock-aftershock sequences typi- weather storms, and telemetry problems can all
cally have b 1 (most of the energy is in a produce signals that could be confused with S
single mainshock). Volcanic earthquakes occur tremor.
in hot, highly heterogeneous material containing Tremor may be the result of continuous exci-
many small faults. Nonvolcanic earthquakes tation of the source that produces low-frequency
tend to occur in more homogeneous material earthquakes. The resonance of a fluid-filled con-
with failure on a single larger fault, which then duit is one model for the origin of tremor, as it is
loads adjacent faults, causing many of them to for low-frequency earthquakes. Interface waves
fail producing aftershocks. Low-frequency traveling along the crack or conduit wall can
earthquake swarms often contain one or produce overtones, like an organ pipe. Gliding
more families of repeating earthquakes. Each lines could be caused by a change in the sound
family is identified by a unique waveform and speed of the fluid (e.g., due to a change in bubble
is the result of the same nondestructive source density) or a change in a length of the section of
process being activated repeatedly in the same conduit that is resonating (e.g., a change in nucle-
location. ation depth).
Tremor may also be a superposition of observed simultaneously with a small LP signal.

low-frequency earthquakes. There are many Second, free-falling blocks impact the slope
observations of low-frequency earthquake below (and fragment) simultaneous with a
swarms merging into volcanic tremor. Modeling 0.5 Hz signal. Third, the fragmented material
has shown that harmonic tremor could appear generates a rockfall signal.
when overlapping earthquake signals occur at Dome collapse signals are the superposition of
regular time intervals that differ by less than many rockfall events, which may be occurring
2 % (Powell and Neuberg 2003). As this regular simultaneously on different flanks of the
rate of earthquakes gradually increases volcano. Dome collapses may last minutes to
(or decreases), gliding spectral lines are pro- hours. From studying dome collapse signals, it
duced. In this model, tremor is a low-frequency is possible to tell how long dome collapses lasted
earthquake swarm. and which phases of the collapse were most
energetic. Several major dome collapses at
Surface Signals Soufrière Hills Volcano (Montserrat) and Merapi
Surface signals are those generated by hazardous Volcano (Indonesia) have been triggered by
surface processes such as rockfall, pyroclastic intense rainfall.
flows, and lahars. Since these processes occur at
the surface, they mainly generate surface waves. Lahar Signals
The composition of lahars varies from muddy
Rockfall and Pyroclastic Flow Signals water to highly erosive, dense mixtures of wet
Actively growing lava domes are highly unstable, ash, rocks, and boulders that set like concrete.
and blocks can be observed falling almost con- Lahars usually occur during or immediately
tinuously, disintegrating into smaller blocks and after heavy rainfall. Barclay et al. (2006) found
plumes of hot ash. Most of these rockfalls are that about 2 cm per hour of rain falling on uncon-
small, but many generate detectable seismic sig- solidated materials was sufficient to trigger lahars
nals. If larger parts of the dome collapse, or if on Soufrière Hills Volcano (Montserrat). How-
there is an explosive component destabilizing the ever, hot volcanic material can generate lahars by
blocks, a pyroclastic flow may be generated. mixing with crater lakes or with snow and ice on
Pyroclastic flows are more energetic, produce glaciated volcanoes. Volcanic activity may also
vigorously convecting ash clouds, behave more melt the base of a glacier and cause an outbreak
like a fluid (less friction), and are therefore able to flood that briefly rivals the force of a major river.
travel farther and at higher speeds. There is a These are particularly common in Iceland and are
continuum from the smallest rockfalls to the larg- known in Icelandic as jökulhlaups.
est pyroclastic flows, and they mostly are Lahar signals are tremor-like signals that can
spawned from the part of the dome that is actively be distinguished from pyroclastic flow signals by
growing. Pyroclastic flows can also be generated their duration (tens of minutes to hours), higher-
by the collapse of an eruption column. frequency content (6–10 Hz, up to 100 Hz in
Rockfall signals are emergent, contain a wide some cases), and slower speeds; they show ampli-
range of frequencies (1–10 Hz, peaking 3–4 Hz), tude peaks on different stations perhaps minutes
and are dominated by surface waves. They have apart as they travel down valley. To improve the
been located by exploiting the amplitude distri- detection of lahar signals, seismic stations, video
bution of rockfalls signals across the seismic net- cameras, and tripwires can be added at various
work. On Soufrière Hills Volcano (Montserrat), positions adjacent to (or within) the valley.
some rockfall and many pyroclastic flows have a
long-period precursor, and it is possible that this Difficulties of Event Classification
is an explosive degassing signal. At Unzen Vol- Classification of volcano-seismic signals is diffi-
cano (Japan), pyroclastic flow signals are found cult. The signal recorded is a convolution of the
to comprise three parts. First, a dome collapse is source signature, propagation and site effects,
and the instrument response, superimposed on record. The first sign of unrest is often an increase
noise which varies spatially and temporally. The in the rate of VT earthquakes, indicating rock
source signature may be the result of complex fracture due to changes in stresses caused by
interactions between a multiphase fluid and an rising magma. In some cases it may be the occur-
unknown geometry of dikes and conduits. Further rence of DLP earthquakes.
from the source, the signal-to-noise ratio When magma reaches shallower depths, vola-
decreases, and increased scattering and separa- tiles exsolve, and LP earthquakes are recorded.
tion between P, S, and surface waves lengthen the Volatiles cause a reduction in the acoustic veloc-
signal, making it less clear. Inelastic attenuation ity and an increase in the impedance contrast of
may be greatest at shallow depths where poorly magma cavities, trapping energy and leading to
consolidated, highly fractured material results in longer, lower frequency signals. As the volatile
higher frequencies being filtered out. Attenuation content increases, more energy is trapped, and
may also be significant at depths of 20 km or this may lead to LP earthquake events called
more, because of high heat flow and partial tornillos. Narrowband tremor may be generated
melt. Similar-looking signals at two different by continuous vesiculation as magma rises fur-
volcanoes, or even at the same volcano, might ther or by the boiling of groundwater.
be caused by different physical mechanisms. If groundwater is heated rapidly, a phreatic
Classification is also subjective. Different explosion may result.
seismic analysts may disagree on the classifica- At any time in the sequence, magma rise may
tion of a particular event. Discrepancies in clas- stall. This may be the end of the unrest, or it may
sification also appear in published literature: an recommence as more magma is injected at depth
LP in Lahr et al. (1994) looks similar to a hybrid adding more heat and volatiles to the system.
in Luckett et al. (2007). Classification might arti- Tremor due to groundwater boiling may subside
ficially separate signals that lie along a contin- as the system dries out, leading to a period of
uum. The classification scheme may also evolve relative quiescence. For basaltic systems, an
during volcanic unrest as a wider variety of sig- eruption may then occur without further warning.
nals are recorded. For these reasons it is helpful to For andesitic to rhyolitic systems, hybrid
reassess the event classes used before interpreting earthquake swarms often occur due to repeated
trends in their rates of occurrence or locations. shear failure of viscous magma as it gets close to
Terminology also differs. Some of the terms the surface. Hybrid earthquake swarms are fre-
describe the frequency content of the signal and quently associated with growth of a lava dome.
others imply a physical mechanism. VT earth- If the events merge into a continuous tremor, har-
quakes are also called high-frequency events monic tremor and gliding spectral lines might be
and A-type events. LP earthquakes are also called observed and are the strongest signs that an explo-
B-type events. Collectively, long-period and sive eruption may be about to occur. During an S
hybrid events are called low-frequency explosive eruption, violent ground shaking typi-
(LF) events, because they are both dominated cally manifests as a broadband tremor signal.
by low-frequency coda. Low-frequency seismic- Explosion earthquakes may be recorded, identifi-
ity includes low-frequency events and volcanic able by the shockwave that travels through the air
tremor. and is coupled back into the ground.
Explosive eruptions are short-lived, and hope-
Interpreting Volcanic Seismicity fully before they occur, any vulnerable
A common pattern of volcanic seismicity is populations will have been evacuated. But extru-
described by the Generic Volcanic Earthquake sive eruptions may last for years or decades, with
Swarm Model (McNutt 1996). Background seis- varying levels of activity, spawning rockfall,
micity varies from volcano to volcano and can pyroclastic flows, and lahars. Monitoring these
only be established for a particular volcano long-lived eruptions presents challenges of its
through a long-established seismic monitoring own, since communities are often in close
proximity. Escalations in activity are often pre- volcanoes in parallel, how to interpret the data
ceded by short sequences of VT earthquakes, without being able to see the volcanoes, how to
low-frequency earthquakes, and tremor, perhaps maintain so many networks, and how to manage
suggesting a batch of new magma rising toward all the data.
the surface. However, volcano-seismic data have Participation in a volcanic crisis, is an oppor-
also been used to track the rate, energy, and tunity to help society and also witness some of
location of debris flows and estimate extrusion nature’s most spectacular phenomena. Having
rates. the right team is crucial. The leader of a volcano
An important question to answer once an erup- observatory must be experienced in volcano
tion is underway is when will it stop. The first monitoring and be committed to the job. The
sign may be the cessation of low-frequency seis- work can be stressful, the hours long, and hard
micity, indicating that volatiles are no longer decisions have to be made in the heat of the
present in the system, which may be confirmed moment, e.g., regarding evacuations. Weekly
with gas flux measurements. More significant meetings play a vital role at many observatories
may be the occurrence of VT earthquakes at by pulling the team together, integrating data
depth with fault-plane solutions consistent from many monitoring techniques, identifying
with magma withdrawal (Roman et al. 2006). technical problems, and prioritizing work. All
Geodetic data may also indicate deflation of the observatory staff must conduct themselves pro-
volcanic edifice. fessionally and have the authority to perform the
A more detailed discussion of the interpreta- roles assigned to them. Fostering good commu-
tion of volcano-seismic data is beyond the scope nications within the observatory and with the
of this entry. There are several excellent summa- authorities and the public builds trust and
ries of our evolving understanding of volcano- enhances public safety.
seismology including Chouet and Matoza (2013), The backbone of a volcano observatory is the
Chouet (2003), and McNutt (2000, 2002, 2005). seismic monitoring program, and the heart of the
observatory is the operations room. It is from
there that seismologists track live data streams,
Observatories coordinate field teams and warn them of hazard-
ous activity, and alert the authorities (and the
Volcano observatories vary greatly in their level public) to escalations in activity. When volcanoes
of sophistication. The simplest observatory may are at background level, the operations room may
be a hut on the flank of a volcano, with a single be unoccupied. Periodic data checks, perhaps
seismograph recording on paper on a revolving coupled with automatic alarm systems, may be
drum. A modern observatory, however, will have enough to keep seismologists abreast of signifi-
a network of seismic stations with data cant changes in seismicity. When unrest begins,
telemetered to the observatory using a variety of periodic checks may become more frequent. At
communication systems such as FM radio, satel- some point the operations room is activated and
lite, and cellular networks. The observatory may manned 24 h a day.
be on the flanks of the volcano, within a few miles These different stages of volcanic activity,
of the active vent. Or it may be tens or hundreds from background to unrest to impending erup-
of miles away, colocated with a university or a tion, greatly impact the level of seismic monitor-
government agency. Many observatories monitor ing that can be done. At background level,
a single volcano, but some monitor several. The seismologists may be able to conduct research.
Alaska Volcano Observatory has operated seis- When a crisis is underway, it will be difficult to
mic networks on as many as 32 volcanoes. The do anything more than interpret available real-
Japan Meterological Agency monitors 47 volca- time data using operational tools that are in place.
noes with real-time seismic data. This poses Part of the job of an observatory seismologist is to
unique challenges: how to monitor so many engineer the infrastructure of the seismic
monitoring program and operations room to be for a GPS clock and can time stamp data on site
able to respond effectively during a crisis. This (which is more accurate). On-site data recording
infrastructure includes seismic networks and enables data to be retrieved or retransmitted later
software systems used to monitor volcanic seis- if there is a communication outage. Two-way
micity. These are the subjects of the next two telemetry allows data packets to be resent auto-
sections. matically if they are not received intact and
allows troubleshooting and reconfiguring of sta-
tions from the observatory without the expense
Volcano-Seismic Networks and delay of a site visit.
While a digital broadband network offers
Around 550 volcanoes have erupted in historical many advantages over an analog short-period
times, and about 200 of these are seismically network, the equipment is more expensive and
monitored. The best networks tend to be around power requirements are 2–3 times higher, mean-
volcanoes that pose a particularly high threat to ing that additional solar panels and batteries are
large population centers, e.g., Etna and Vesuvius needed. Broadband sensors also require precise
(Italy), Rainier and St. Helens (continental USA), leveling. Many analog networks have been
and Kīlauea (Hawaii). Many volcanoes are mon- upgraded to digital telemetry to improve dynamic
itored only by a single station, often as part of a range, and most new networks use digital telem-
regional network. Nevertheless, by tracking the etry, but it is still common to find analog telem-
number and cumulative energy of different types etry at volcano observatories.
of seismic signals recorded each day on a single Designing a volcano-seismic monitoring net-
station, a volcanic eruption can be anticipated. work is complicated and involves multiple trade-
It may then be possible to rapidly deploy sufficient offs. For the same budget a volcano observatory
additional stations to locate earthquakes and fore- may be able to install a few digitally telemetred
cast where and when an eruption may take place. broadband stations or many more analog-
The most commonly used seismometers for telemetered short-period stations. The best
volcano monitoring are short-period sensors. choice depends on the goal.
These have a corner frequency of 0.5–2 Hz and For a volcano that is far removed from popu-
come in single (vertical) and three component lation centers, the goal might be to detect erup-
varieties. Short-period seismometers are usually tions so that aviation authorities can be warned of
deployed with analog telemetry. Both have a hazardous ash clouds, requiring a minimal level
dynamic range (the ratio between the largest of monitoring. For a densely populated region
and smallest amplitudes that can be represented) around a frequently active volcano, or one capa-
of only a few thousand. This results in signal ble of devastating eruptions, higher quality mon-
amplitudes often being “clipped,” which greatly itoring is needed to provide as much lead time as S
diminishes the monitoring and research value of possible. The magnitude detection threshold may
the data. Three-component portable broadband vary from 2 for a regional network to below 0 for
seismometers, available since the late 1980s, a dense volcano-seismic network. Moran
use a force-feedback circuit and typically have a et al. (2008) identified four levels of volcano-
corner frequency of 30–120 s and a dynamic seismic monitoring and the number of short-
range of several million, allowing them to record period, broadband, strong motion, and infrasound
signals on-scale even right up on the volcano sensors needed within different radii to make
summit. They have to be coupled with 24-bit those levels of monitoring achievable. Their find-
digital telemetry systems, which have a similar ings are summarized in Table 1.
dynamic range. A key consideration for locating earthquakes
Analog-telemetered data are time stamped at accurately is to minimize the azimuthal gap.
the observatory by a time signal from a GPS A ring of stations, which encloses most of the
clock, but modern field digitizers have an input volcanic-earthquake epicenters, is ideal with a
Seismic Monitoring of Volcanoes, Table 1 Recommended instrumentation to achieve different levels of volcano-
seismic monitoring from Moran et al. (2008)
Level Goal Recommendation
1 Minimal monitoring/eruption detection Site a total of five seismic stations within 200 km,
Detect M > 1.5 earthquakes; crudely locate M > including two within 50 km of the volcanic center
3 earthquakes
2 Limited monitoring/unrest detection Site a total of five seismic stations within 50 km,
Detect M > 1 earthquakes; crudely locate M > including two within 10 km of the volcanic center
2 earthquakes; determine event type; detect
energetic seismic tremor
3 Basic real-time monitoring Site six to eight seismic stations within 20 km of the
Detect M > 0.5 earthquakes; accurately locate M > volcanic center, including two or three stations with
1 earthquakes; determine event type; detect seismic at least one three-component sensor and at least one
tremor; on-scale recording of energetic seismicity on broadband station, within 5 km
at least one station; detect very-long-period events
Detect changes in travel time; detect broad-scale
changes in seismic velocity
Use fault-plane solutions and b-values to determine
generalized stress fields near the volcanic center
4 Advanced real-time monitoring Site 12 to 20 seismic stations within 20 km of the
Detect and accurately locate M > 0 earthquakes; volcanic center, including at least six broadband
determine event type; detect and crudely locate stations, as many as possible within 5 km; at least
seismic tremor; on-scale recordings of energetic one strong-motion station; and at least two
seismicity on multiple stations; detect and crudely infrasonic stations (with at least two infrasonic
locate very-long-period events and other very-low- sensors per station) at erupting, restless, and
frequency seismicity (or) frequently active remote volcanic centers
Determine detailed source properties of tornillos;
construct 3-D velocity models (provided local
seismicity is sufficient)
Detect explosions and possible infrasonic precursors
to explosions at restless and (or) frequently active
volcanoes
Detect detailed stress-field changes by calculating
well-constrained fault-plane solutions and
(or) moment tensors, mapping b-values at high
spatial resolution, and detecting changes in S-wave-
splitting directions over time
few stations closer to the volcanic center where they match the true locations of the earthquake
they can help detect smaller seismic signals and will depend on the velocity model and to a lesser
be used to locate summit events with greater degree, on the algorithm used to locate the events.
precision and improve depth resolution. For stra- The sophistication of the velocity model may
tovolcanoes, all the earthquakes may be within vary from a simple one-dimensional regional
1–2 km of the volcanic center, whereas for cal- model with constant velocity layers to 3-D
deras the seismicity might be diffused over an models determined from seismic tomography.
area 15–20 km in radius. It is useful also to have Site selection depends on the geology, topog-
real-time data from at least one more distant raphy, accessibility, and noise. It is preferable to
station (which may be part of a regional network) install seismometers in solid bedrock, but volca-
to help constrain the depths of deeper events, and noes are typically comprised of layers of ash,
to discriminate more effectively between flow deposits, and boulders. Repeaters may be
regional and local volcanic earthquakes. These required on ridges to rebroadcast signals or
steps make it possible to locate earthquake rea- boost them if transmitting over tens of kilome-
sonably well relative to each other, but how well ters, but represent additional expense and
Seismic Monitoring of Volcanoes, Table 2 Some of the best volcano-seismic monitoring networks. The numbers in
parentheses indicate number of borehole instruments. All of these networks surpass level 4 as defined in Table 1 of
Moran et al. (2008)
Stations within 5 km Stations within 20 km
Volcano Broadband Short period Total Broadband Short period Total
St. Helens (USA) 2 7 9 2 16 (4) 18 (4)
Soufrière Hills (Montserrat) 7 (1) 0 7 (1) 9 (1) 6 (4) 15 (5)
Piton de la Fournaise (Reunion) 14 6 20 16 17 33
Halemaumau, (Hawaii) 12 5 17 19 14 33
Stromboli (Italy) 18 0 18 18 0 18
Etna (Italy) ? ? 13 ? ? 34
Vesuvius (Italy) ? ? 17 ? ? 21
Erebus (Antartica) 6 6 12 6 6 12
potential points of failure. Sites that require heli- Real-Time Monitoring

copter access will be expensive to maintain, so
often it is better to find sites that can be easily Volcanic activity may escalate suddenly, and the
reached with four-wheel-drive vehicles or on ability to rapidly identify anomalous volcano
foot. Sites should be far from traffic and seismicity is critical. Real-time data visualization
other human noise and also away from any tall systems allow scientists to rapidly assess multiple
obstacles that may be vibrated by the wind, parameters derived from seismic data such as
such as trees, cliffs, and radio towers. Burying hypocenters, magnitudes, event rates, spectral
sensors a few feet below the surface helps variations in tremor, etc. Alarm systems alert
suppress high-frequency wind noise and scientists to large events and the occurrence of
low-frequency noise due to temperature and pres- high amplitude tremor and volcanic earthquake
sure variations. swarms. Data-rich websites and remote desktop
Among the best volcano-seismic networks are connections allow scientists to respond rapidly to
those used to monitor Piton de la Fournaise alarm: This is convenient and enhances safety.
(Reunion), Halemaumau (Hawaii), St. Helens The US Geological Survey, through its Earth-
(USA), and Soufrière Hills (Montserrat) volca- quake and Volcano Hazards Programs, has
noes (Table 2). According to the criteria in played a pivotal role in disseminating free,
Table 1, these are all level four networks. open-source software to aid in volcano-seismic
A map of the digital seismic network that has monitoring. This section describes common tech-
been operational in Montserrat since 2006 is niques and software used to examine continuous
shown in Fig. 1. This utilizes a mixture of Guralp seismic data and events in more detail. S
CMG-40T broadband and Mark L4-C short-
period seismometers, coupled with Guralp Continuous Data
DM24 digitizers and FreeWave Spread Spectrum
serial port radios. One of the broadband Digital Helicorder Plots
stations has a CMG-3 T seismometer in a 30-m The most basic form of real-time data display is
borehole and less than 4-km from the dome. Four the helical drum recorder (often called
more sites feature a dilatometer and 2-Hz “helicorders” or “drums”). A pen etches a trace
seismometer in 200-m boreholes, coupled with on a smoked sheet (or draws an ink trace on a
surface GPS. The Piton de la Fournaise network blank sheet) of paper wrapped around a cylindri-
also uses CMG-40T and L4-C seismometers, cal drum. Helical drum recorders have played a
coupled with Kinemetrics Q330 digitizers and a vital role in volcano-seismic monitoring,
combination of wireless internet and FM radio allowing rapid visualization of seismic ampli-
telemetry. tudes and event identification of signal types.
Seismic Monitoring of
Volcanoes, Fig. 1 The
Montserrat seismic
network from April 2006
(Modified after Luckett
et al. (2007)). Telemetry
paths are shown as lines.
The observatory is marked
with MVO. Repeater sites
are marked with a gray
circle. Short-period stations
are marked with a triangle.
Broadband stations are
marked with a star, except
for the borehole broadband,
MBLY, marked with a
square. Not shown are the
four 200-m borehole short-
period stations, operated by
the CALIPSO consortium
since 2002
Helicorders were ubiquitous until the 1990s usgs.gov/software/swarm/index.php) is an excel-

and are still used at many observatories today, lent solution that allows data to be plotted dynam-
but they have numerous drawbacks. They require ically: The user can select the time range and the
considerable maintenance and provide limited scale; the data can be filtered, and short segments
dynamic range. Adjacent traces often overlap, of data can be highlighted and replotted as spectra
making them hard to read. The data are not ame- or spectrograms. Many observatories have now
nable to other forms of analysis, and the sheets replaced large collections of helicorders with
require storage space. So there has long been a multiple screens showing SWARM displays
desire to replace helicorders with a software from different seismic channels. Figure 2 shows
equivalent. Seismic Waveform Analysis and 48 h of seismicity at the Soufrière Hills Volcano
Real-time Monitor (SWARM) (http://volcanoes. in June 1997.
Seismic Monitoring of Volcanoes, Fig. 2 Screenshot In June 25, a moderate dome collapse sent pyroclastic
of the software Swarm, showing a digital helicorder plot flows the northern flanks of the Soufrière Hills Volcano,
from station MBWH channel SHZ for 48 h from June claiming 19 lives
23, 1997. Remarkable cyclic hybrid swarms are visible.
RSAM Plots weather, and instrumental difficulties are recog-

Continuous data are often downsampled to nized as distinct patterns in RSAM datasets. S
one sample per minute, which makes it easier RSAM data for dome-building episodes gradu-
to identify long-term trends in continuous ally develop into exponential increases that ter-
seismic amplitude. The Real-time Seismic minate just before the time of magma extrusion.
Amplitude Measurement system (Endo and Volcanic earthquakes and rockfall show up as
Murray 1991) took the average amplitude of isolated spikes on RSAM plots for seismic sta-
the seismic signal in each 1-min time window, tions close to the edifice, but seldom for more
and recorded these data into a file. RSAM data distant stations. Weather-related noise shows up
do not discriminate between different types of as low-level, long-term disturbances on all seis-
seismicity. mic stations, regardless of distance from the vol-
RSAM data show cyclic seismicity clearly cano. The RSAM system proved valuable in
(Fig. 3a). Changes in earthquake activity associ- providing up-to-date information on seismic
ated with dome-building episodes (Fig. 3b), activity for three Mount St. Helens eruptive
Seismic Monitoring of a x 10
4
Volcanoes, Fig. 3 (a) 2.5
Plot of RSAM data with
one sample per minute for 2.25
station MBWH channel
SHZ from June 22 to 2
29, 1997. The cyclic hybrid
swarms mentioned in Fig. 2 1.75
RSAM on MBWH.SHZ
are shown clearly. The
spikes represent discrete 1.5
events, e.g., pyroclastic
flows. (b) RSAM plot 1.25
showing the major dome
collapse of the Soufrière 1
Hills Volcano (Montserrat)
which occurred on July 0.75
29, 2001
0.5
0.25
06/23 06/24 06/25 06/26 06/27 06/28 06/29

Date in 1997
b x 104
6
5.5
4.5
4
RSAM MBWH.SHZ
3.5
2.5
1.5
0.5
15:00 18:00 21:00 00:00 03:00 06:00

Date (UTC) 29-30 July 2001
episodes from 1985 to 1986 and in numerous Computing the median (rather than mean) ampli-
eruptions at other volcanoes since. Exponential tude in each time window provides a measure-
increases in RSAM data commonly ment of tremor amplitude less biased by events or
precede explosive eruptions. Inspired by spikes in the data - but is no longer strictly
RSAM, many other parameters have been ‘RSAM’ data. Corrected for the instrument
computed on 1-min (or 10-min) timescale. response and geometrical spreading, and then
For example, RSEM provides a relative measure- integrating the data, produces reduced displace-
ment of energy on a 1-min (or 10-min) timescale. ment. Since 1996 the Alaska Volcano
Observatory has recorded an instrument 1991), and this is now incorporated in Earthworm
corrected seismic amplitude and the peak fre- (discussed below). The seismologist can choose
quency for each vertical-component seismic which stations to monitor and define an amplitude
channel, on a 10-min timescale. threshold and a duration for each station. Both
must be exceeded to trigger a station. The seis-
Spectrograms mologist also must define the number of stations
Changes in tremor spectra may indicate different that must trigger simultaneously to declare an
flow regimes or changes in source parameters alarm. There is also a mechanism which reduces
such as geometry, sound speed, or ascent rate the false alarms due to regional or teleseismic
(Thompson et al. 2002). Harmonic tremor and earthquakes: A far away station can be chosen
gliding spectral lines are frequently followed by which prevents an alarm being declared when it
eruptions. Spectral monitoring can help differen- exceeds an amplitude threshold. The tremor
tiate between VT and low-frequency earth- alarm is typically configured for a moderate
quakes, or tremor; wind noise; and electronic amplitude and a duration of several minutes (the
noise. RSAM system also included an event alarm,
The Alaska Volcano Observatory (AVO) which would typically be configured for high
began experimenting with high-resolution spec- amplitude and duration of a few to tens of
trograms in 1996. For up to eight stations, a seconds).
spectrogram is plotted underneath a normalized The RSAM tremor alarm system was used to
seismic trace. The time resolution is about 10 s monitor the Soufrière Hills Volcano (Montserrat)
and the frequency resolution 0.1 Hz. Each image from 1999 to 2003. Figure 6 shows the number of
file (Fig. 4) displays a 10-min seismic trace and alarms issued over this period.
spectrogram for up to eight seismic-data chan-
nels. These are ordered in increasing distance Events
from the volcano. The data are instrument
corrected and color coded, so that ground Real-Time Seismic Event Catalogs
motions at one station can be easily compared During a volcanic crisis, it is vital to have near-
with those at another station, even at a different real-time information about the rates, sizes, and
volcano. locations of seismic events, and for this a real-
These web-based spectrograms have been a time event catalog is needed. Providing there are
core AVO monitoring tool since 1998, and a at least four stations, event detection software
seismologist reviews them twice a day. A conve- automates the process of capturing anomalous
nient web interface (http://www.aeic.alaska.edu/ signals in large volumes of continuous data.
spectrograms/mosaicMaker.php) shows mosaics Each channel of seismic data is detrended, fil-
of 10-min spectrogram image files (Fig. 5) and tered, and rectified, and a running short-term S
enables 12 h of data from more than 20 average (typically 1 to a few seconds) and long-
volcanoes to be reviewed in just 30–45 min. Sim- term average (typically a few tens of seconds) are
ilar web-based spectrograms are used for computed. Where the STA (short-term average)
monitoring volcanic seismicity in the Cascades to LTA (long-term average) ratio exceeds a
and Hawaii. threshold, a candidate P or S arrival is declared.
Association is the process whereby candidate
Tremor Alarms arrivals’ “picks” are grouped together based on
Tremor is common precursor to eruptions, and similar arrival times and geographic location to
strong tremor frequently accompanies eruptions. declare an event. Location techniques typically
Observatory seismologists therefore need to be use a 3-D grid search to minimize the difference
aware when tremor is being recorded. The earli- between measured and theoretical travel times,
est widely used volcano-seismic alarm system given a particular velocity model. For volcanic
was part of the RSAM system (Endo and Murray earthquakes, it is most common to compute a
Seismic Monitoring of Volcanoes, Fig. 4 Seismic UTC. The hottest colors shown (pink) suggest an S-wave
traces and corresponding spectrograms for the vertical ground velocity of 0.02 mm/s. Near-real-time spectro-
components of seven seismic stations near Akutan Vol- gram plots like this have been linked to the AVO internal
cano, corresponding to 10 min of data on April 22, 2014. website since 1996. The menu at the top of the screen
A regional earthquake is visible clearly around 20:13 allows easy navigation to other volcanoes or time periods
duration magnitude due to the prevalence of helpful to view event rate and energy rate plots
clipped signals from short-period analog teleme- (Fig. 7) and hypocenter plots (Fig. 8).
try. For on-scale recordings, local magnitude is There are some caveats with event catalogs.
often also computed. To visualize catalogs, it is Discrete events are only one aspect of volcanic
Seismic Monitoring of Volcanoes, Fig. 5 A spectro- visible at 20:13 (Fig. 4) and 22:15. A calibration signal
gram mosaic, part of the same application shown in Fig. 4. occurs on the top station (AKV) at 22:03. The apparent
Three hours of data are shown. Regional earthquakes are data dropouts are due to data latency
seismicity. Signals such as precursory earthquake magnitude at which earthquakes become detect-
swarms and tremor episodes, or those generated able. Times when an event catalog suggests there
by explosive eruptions, dome collapses, pyro- is little seismicity may actually be times of high
clastic density currents, and lahars, are of the seismicity. Real-time catalogs also suffer from
greatest interest from a hazard perspective. Such poorly resolved locations and magnitudes – also S
signals generally evade standard earthquake- a review by an analyst later corrects for these.
detection schemes based on comparing the
short-term and long-term signal averages and Swarm Alarms
are not systematically cataloged. The vast major- Earthquake swarms, like tremor, are a common
ity of volcano-seismic signals lack identifiable precursor to volcanic eruptions. A real-time event
phases and so cannot be located with differential catalog can be used as the basis for detecting
travel-time techniques. Earthquake catalogs earthquake swarms.
therefore present only a narrow view of volcanic Okmok Volcano (Alaska) erupted ash to a
seismicity. Many volcano-seismic signals are height of 15 km on July 12, 2008, after less than
emergent or long in duration, so STA/LTA detec- 5 hours of precursory seismicity which included
tors break down. Events may be masked by high an earthquake swarm. Concerned by this, the
background signals (e.g., tremor, wind, or elec- Alaska Volcano Observatory developed a
tronic noise). Station outages also increase the swarm alarm system. The system identified the
RSAM event alarms

4000
cumulative # of alarms
3000
2000
1000
0
2000 2001 2002 2003
RSAM tremor alarms

1000
cumulative # of alarms
800
600
400
200
0
2000 2001 2002 2003
Date
Seismic Monitoring of Volcanoes, Fig. 6 RSAM The alarm thresholds configured for each station were
event alarms (top panel) and tremor alarms (bottom raised throughout 2000 and 2001 as activity increased.
panel) issued at Montserrat Volcano Observatory between Event alarms typically corresponded to pyroclastic flows
February 1999 and May 2003. In total there were about and regional earthquakes, and tremor alarms to dome
4,000 alarms, an average of almost three alarms per day. collapses (i.e., a series of pyroclastic flows)
start and end of five swarms that occurred today, in conjunction with other tools that
between February and April 2009, as well as expand its real-time monitoring capabilities.
significant escalations in the rate of earthquakes Two reasons for this are that all US volcano
and energy release during those swarms observatories use them, and the USGS Volcano
(Thompson and West 2010). Disaster Assistance Program deployed them in
many countries at the request of foreign
Software governments.
Earthworm (http://folkworm.ceri.memphis.edu/ Earthworm is not the only widely used seismic
ew-doc/), is the most widely used system for monitoring systems: Alternatives are SeisComP3
earthquake and volcano-seismic monitoring (https://www.seiscomp3.org/wiki/doc) and Ante-
today. It is an open-source, data acquisition, and lope (http://www.brtt.com/software.html). Ante-
earthquake detection framework developed by lope is an excellent framework for developing
the US Geological Survey. It comprises of mod- new monitoring and research tools, but requires
ules that communicate via messages on shared a commercial license except for US higher edu-
memory rings. It is used in the National Earth- cation institutions. SeisComP3 provides a rich set
quake Information Center, tsunami warning cen- of graphical user interfaces for global and
ters, and regional seismic networks. Earthworm regional seismic monitoring and is particularly
is in widespread use at volcano observatories suited for tsunami early warning. However,
# Events per hour

1500
100
Cumulative
1000
50
500
0 0
03/21 03/22
x 107 Energy per hour x 108

5
Cumulative
0 0
03/21 03/22
Seismic Monitoring of Volcanoes, Fig. 7 Two com- energy release rate plots (computed from magnitude data).
mon ways of examining trends in detected events are The data shown are from Redoubt volcano, Alaska, from
(upper panel) event rate (also called ‘counts’- the number 2009/03/20 to 2009/03/23 and are produced using the
of detected events per unit time) plots, and (lower panel) GISMO toolbox. Energy units are arbitrary
neither currently has adaptations for volcano- used to regenerate plots on a private web site used
seismic monitoring. by observatory scientists.
Earthworm, SeisComP3, and Antelope are Earthworm has the ability to import data from
capable of generating real-time earthquake cata- a wide variety of field instrumentation and data
logs within a few minutes of event occurrence, servers and includes a variety of earthquake
and they work well for regional earthquakes and detection, location, and magnitude algorithms.
volcano-tectonic earthquakes. However, other There are modules which compute RSAM data
volcano-seismic signals lack clear P- and and create daily helicorder and spectrogram
S-waves required to compute even an approxi- plots. Tremor and swarm alarm systems are now S
mate location. Fortunately, Earthworm includes included in Earthworm. Other modules can har-
another mode of event detection called “subnet vest messages about detected events, arrivals,
triggering,” which determines only an approxi- locations, and magnitudes to create a real-time
mate event time, and does not attempt to locate an event catalog, serve up plots of event rates and
event. This approach is employed by many vol- Google Maps of epicenters, and serve data in
cano observatories. Custom code can be added to QuakeML format.
monitor these events and trigger other calcula- SWARM can either draw data from a
tions. For example, at the Montserrat Volcano SeedLink server, an Earthworm wave server, or
Observatory from 2000 to 2003, real-time mea- an FDSN (International Federation of Digital
surements of the amplitude, energy, and fre- Seismograph Networks) web service. Another
quency content of new events were made, and option is a Winston wave server, which emulates
Redoubt Volcano Seismicity
60°N 39⬘
60.6
60°N 33⬘ L
a
t
i 60.5
t
u
d
60°N 27⬘ e
60.4
60°N 21⬘
60.3
152°W 54⬘ 152°W 42⬘ 152°W 30⬘ 152°W 18⬘ 0 10 20
Depth (km)
Mag. Symb. Size
−5
0
0 1
Depth (km)
5 2
3
10
4
15 5
20
−153.0 −152.8 −152.6 −152.4
Longitude
−5
0
Depth (km)
5
10
15
20
Jan 2009 Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Date
Seismic Monitoring of Volcanoes, Fig. 8 A common in map view, and adjacent panels show depth versus
way of presenting 3-D hypocenter data is as a set of 2-D latitude and depth versus longitude. These help scientists
slices through the data. In this plot, generated by VolPlot, identify spatial relationships in hypocenter data. At the
a program used for many years at the Alaska Volcano bottom, depth versus time is shown, which helps scientists
Observatory, earthquake hypocenters from Redoubt Vol- recognize if earthquakes are getting closer to the surface.
cano in 2009 are shown. The main panel shows epicenters The size of the symbols indicates magnitude
an Earthworm wave server, but stores data in a different event classes. For example, at the Mont-
MySQL database, providing rapid access and a serrat Volcano Observatory from 2000 to 2003,
deep data archive. Winston also provides a web rockfall signals were directed into an automated
interface that allows plotting of helicorder rockfall location system.
images and RSAM data (RSAM data are Automated techniques for classifying
precomputed on import to Winston). events have been tried at some observatories,
but remain rare. Many researchers have had
success with frequency-based analysis, artifi-
Data Analysis cial neural networks, and hidden Markov
chains, but typically for only certain types of
Some analyses are not performed in real time, events recorded at a particular volcano on a
because it cannot be fully automated or just particular station. Neither Earthworm, Ante-
because the tools to do it well have not yet been lope, or SeisComP3 currently include an
developed; but they nevertheless provide timely auto-classification module.
information that can affect the way scientists
interpret activity. Swarm Analysis
While event detection software captures many
Analyst-Reviewed Event Catalogs (perhaps most) volcanic earthquake signals, addi-
A real-time event catalog is valuable, but it can- tional events can be found by using a match filter.
not approach the quality of an analyst-reviewed This technique is particularly useful for tracking
catalog. A seismic analyst will periodically repeating, low-frequency earthquakes, since the
review the detected events, classify them, and emergent onsets of this type of event make them
delete false events. They might also pick P and notoriously difficult to trigger using an STA/LTA
S phases, run software to locate events, and com- detector. A known event waveform (or part of it)
pute magnitudes. Such a catalog is essential for is cross-correlated against continuous data to find
research, but can be difficult to produce in a additional events in the same “family.” Each
timely manner as volcanic unrest intensifies and family represents a source which is repeatedly
decisions need to be made on timescales of activated in a volume that is small compared to
minutes and hours, rather than days. During the wavelength of the correlated signal.
earthquake swarms (and major aftershock A number of volcanic processes are thought to
sequences), human analysts may be be able to generate repeating earthquakes, includ-
overwhelmed, increasing latency further just ing repeated resonances of fluid-filled cracks, a
when it matters most. propagating crack tip that is driven by intruding
While Antelope and SeisComP3 include an magma, or even repeated shear failure within a
analyst review capability, Earthworm does not: body of magma. Additional analyses of the earth- S
So it is commonly paired with Seisan (http://seis. quake locations and the waveform frequency
geus.net/software/seisan/seisan.pdf), which is an spectra are needed in order to distinguish between
excellent free software package for processing these various models. Double-difference reloca-
and editing event catalogs. Seisan can estimate tion can be used to image the spatial evolution of
fault-plane solutions, moment tensors, and the swarm more clearly.
b-values too. Seisan includes some adaptations A related technique is to cross-correlate wave-
for volcano-seismic analysis: It supports event forms from all events in a catalog against each
classes such as VT, LP, hybrid, and rockfall and other, to discover all event families. The
can generate event-rate plots. Event waveforms MATLAB toolbox “GISMO” includes tools to
are typically examined as spectra as well as time automate this process. During unrest at Redoubt
series. However, automated event classification Volcano in 2009, this was used in near real time
is highly desirable because different real-time to monitor the evolution of swarms (Buurman
processing schemes could then be applied to et al. 2012) (Fig. 9).
Feb 26. 08:00 Mar 21 09:00

Time
a b
Mar 22 00:00
Feb 26. 12:00
Mar 22 12:00
Feb 27. 00:00 C

ro
ss
co
rre 1
la
tio
n
*1 Mar 23 00:00
0.9
Feb 27. 11:00 897 events Mar 23 05:00 2000 events
0.8
Mar 27 01:00 c Mar 29 07:50
d
Time
0.7
Correlation coefficient
0.6
0.5
Mar 27 06:00
0.4
0.3
Mar 27 07:30 438 events 37 events

Mar 29 08:45
0.2
Apr 02 21:45 May 03 00:00

Apr 03 00:00 e f 0.1
May 03 12:00
Time
May 04 00:00
May 04 12:00 0
May 05 00:00
Apr 03 12:00
May 05 12:00
May 06 00:00
Apr 04 00:00
May 06 12:00
May 07 00:00
Apr 04 12:00 1949 events May 07 12:00 7470 events
Time Time
Seismic Monitoring of Volcanoes, Fig. 9 Cross- top to bottom and left to right in each panel. The top-left to
correlation plots for six swarms during the crisis at lower-right diagonal represents the autocorrelation of
Redoubt Volcano in 2009 (reproduced from Buurman each waveform. The swarms began on: (a) February
et al. (2012)). Each pixel represents of pair of waveforms 26, (b) March 21, (c) March 27, (d) March 29, (e) April
that have been cross-correlated. Color represents the max- 2, and (f) May 2
imum cross-correlation coefficient. Time progresses from
Locating Phaseless Seismic Signals known precisely when the wire is cut. Disad-
The tremor alarm systems mentioned previously vantages are that trip wires only work once
are very simple and they lack the ability to locate and a dilute flow may leave the wire intact.
tremor. Some observatories have exploited seis- Sophisticated lahar-monitoring systems have
mic data to track the location of tremor and map been installed at many volcanoes including
debris-flow trajectories, which could help miti- Rainier, Merapi, and Nevado del Ruiz.
gate hazards.
Phaseless signals, once identified (via an Imaging Spatial Changes in Volcanic
STA/LTA algorithm or visual inspection), can Seismicity
be approximately located by exploiting the man- There are many techniques for imaging magma
ner in which seismic wave amplitudes decay with storage regions. A volume devoid of volcano-
increasing distance from the source, due to geo- tectonic earthquakes may indicate a large
metrical spreading and attenuation (Jolly magma storage body, incapable of shear failure.
et al. 2002). Alternatively, they can be located Seismic tomography can reveal volumes with
by cross-correlating data to determine travel-time low P- or S-wave propagation speeds. B-value
differences and then using traditional differential mapping (see Sanchez et al., “▶ Frequency-
travel-time techniques to locate the source Magnitude Distribution of Seismicity in Volcanic
(assuming the wave type is known). The Wave- Regions,” this volume) can reveal volumes
form Envelope Clustering and Correlation incapable of sustaining large failures, which are
(WECC) system (Wech and Creager 2008) often inferred as being associated with magma. In
reverses this approach. It continuously locates the region around Uturuncu Volcano (Bolivia),
envelopes of continuous waveform data, and if the presence of a sill at around 15–20-km depth
the location errors are small, it declares an event. has been inferred by an S-wave shadow zone.
Designed to locate the episodic tremor associated Roman et al. (2006) found that changes in the
with the slow slip, it is now being applied to orientation of fault-plane solutions could be used
volcanoes. (retrospectively) to differentiate between
volcano-tectonic earthquakes related to the injec-
Detecting Lahars tion of new magma and posteruptive relaxation.
Lahar signals may appear as high-frequency Ambient noise tomography is another tech-
tremor signals on a volcano-seismic network, nique, which can be used to detect velocity
but may be difficult to detect and locate without changes. In active source seismology, an impul-
designing a network for lahar monitoring. High- sive source is generated at one location and
frequency surface waves attenuate rapidly with recorded at another. The resulting seismogram
increasing distance, so the amplitude of high- is the Green’s function of the path between the
frequency signals can be indicative of how close sites. If this were done repeatedly, seismic veloc- S
a flow is to a seismometer. More dilute flows are ity changes along the travel path could be
also richer in higher frequencies. An existing detected. At a volcano, a drop in velocity could
volcano-seismic network can be adapted for be indicative of increased heat flow, faulting,
lahar monitoring by adding stations close to fluid injection, or expansion of the volcanic edi-
the flow channel at various points, enabling fice. MSNoise (http://www.msnoise.org/) is a
lahar signals to be more easily resolved and package which has been developed to automate
crudely located using the times of the peak the computation of Green’s functions between all
amplitudes on each station, and also the station pairs on a daily basis, allowing velocity
amplitude distribution across the seismic net- changes as small as 0.1 % to be identified.
work. Trip wires can be positioned across the At Piton de la Fournaise Volcano, the locations
valley at different points and at different of dike injections and eruptive fissures have been
heights, and the arrival time and location are inferred.
Detecting Explosions Problems will usually be apparent with tools

Cleveland Volcano (Alaska) has been erupting volcano seismologists use every day. RSAM
frequently in recent years, generating ash clouds plots may reveal diurnal cycles in the data
that threaten aviation travelling between North which may indicate undercharging of batteries
America and Asia, and, without a functioning or anthropogenic noise. Spectrograms and
seismic station within a few tens of kilometers, seismograms may show a flat signal (power
many of these eruptions were previously missed loss), a one-sided signal (stuck seismometer),
or only recognized later in satellite images. De large offset (seismometer not level), white noise
Angelis et al. (2012) describe a system which (loss of seismic signal), or spikes and dropouts
exploits data from a seismic network on Okmok (interference). These can cause havoc with auto-
Volcano, operated by the Alaska Volcano Obser- mated detection and alarm systems, leading to
vatory, to detect explosions at Cleveland Volcano corrupted event catalogs and false alarms, so
(about 120 km away). An STA/LTA detector one strategy is to identify and eliminate affected
creates candidate arrivals. For those that fall stations from automated processing schemes.
within a 2-min sliding window, the differential SeisNetWatch (http://www.isti.com/products/
travel times are computed by cross-correlating seisnetwatch/) is used by many seismic network
envelopes of the seismograms and then operators to monitor all ip-addressible nodes in a
inverted for apparent slowness. If this is consis- network, including field digitizers, to monitor
tent with a sound speed of 340 +/30 m/s in the data latency and other state-of-health parameters.
direction of Cleveland, an alarm is sent by email. The volume of data flowing into an observa-
An alternative to the Cleveland explosion alarm tory can be overwhelming. A single channel of
mentioned above looks for a coherent signal 24-bit, three-component data recorded at 100 Hz
across an infrasound array instead. The advan- requires 620 MB of storage per day
tage is that infrasound sensors record explosion (uncompressed). Twenty channels of data, a typ-
signals less ambiguously, and the technique ical amount for a small volcano observatory, are
could be used to detect explosions in other about 1.5 TB per year. Although this would fit on
regions where installation of a local seismic net- a single hard drive today, it is about 100 times the
work is unrealistic. capacity of a hard drive available 20 years ago
(the storage capacity of hard drives increases by a
factor of about 10 every decade). Continuous
Technical Challenges digital volcano-seismic data before the mid
1990s are rare but, remarkably, the Pacific North-
One of the major challenges for observatories and west Seismic Network recently recovered contin-
seismic networks is monitoring the state of health uous data saved to tape during the 1980 eruption
of a seismic network. There is only one chance to of St. Helens Volcano.
gather data, and in a crisis, data not collected in To improve the reliability of data acquisition
real-time have little value. Seismic and repeater and processing systems, several measures can be
stations need to be engineered to withstand taken. To protect against power failures, all
weather variations which may include heavy mission-critical computers can be connected to
rain, snow and ice accumulation, strong winds, uninterruptible power supplies and configured to
and months of little solar energy. Spare automatically reboot and restart critical processes
equipment needs to be available so that stations once power is restored. No one wants to arrive at
can be fixed quickly when a seismic station com- an observatory on a Monday morning and finds
ponent breaks. Vandalization by humans, dam- that the alarm computer suffered a power outage
age by animals, and theft are other on Friday night and has not been running since,
considerations. During unrest, ash build-up on especially if a significant event occurred. Another
solar panels can quickly cause power loss at a measure is to run all mission-critical computers
seismic station. in parallel, providing automatic failover or at
least have pre-configured spare computers ready an expensive field visit or lengthy data outage.
to plug in whenever a primary computer failed. Free, community supported software is available
All potential points of failure can be monitored for data acquisition, event detection and location
by a diagnostic alarm system. Finally, a daily (e.g. Earthworm), real-time data visualization
checklist helps catch any other problems. All of (e.g. SWARM), event processing and catalog
these measures were taken at the Montserrat Vol- production (e.g. Seisan), instead of each observa-
cano Observatory in 2000, leading to an improve- tory reinventing the wheel. Archiving data -
ment in the data capture rate and public safety. including continuous waveform data, derived
Once captured, data need to be managed effec- data such as RSAM, event catalogs and station
tively to preserve them for further analysis and metadata – is a crucial task that has become much
research. Given the capacity of modern hard easier thanks to the expanding capacity of hard
drives, it is now possible to keep many years of drives, faster internet, and data management cen-
data “online.” Copying data to an off-site location ters such as the IRIS DMC.
is also becoming more common. Many observa- Real-time monitoring may rely on an opera-
tories now transmit real-time data to the IRIS tions room which is manned 24 h a day. Or it may
Data Management Center, which not only serves rely on automated alarm systems that alert obser-
as a backup, but outsources the resources needed vatory staff to escalations in seismicity (and data
to disseminate the data more easily to the outages) outside of normal office hours. Research
research community. and development is often driven by the need for
better real-time monitoring tools – automated
classification of volcano-seismic signals, for
Summary example. Ambient noise tomography and detec-
tion of multiplets are examples of techniques that
Volcano-seismologists forecast volcanic activity might become routine at volcano observatories as
by analyzing the rates, energy release and spatial computational power continues to increase. As
distribution of different types of characteristic automated monitoring becomes increasingly
seismic signals recorded in the vicinity of volca- sophisticated, observatory seismologists will
noes. The most commonly identified signals are need to spend less time on troubleshooting and
VT, LP or hybrid earthquakes. Explosive and development and more time analyzing and
effusive eruptions can cause rockfalls and pyro- interpreting volcanic-seismicity, and collaborat-
clastic flows, which also generate characteristic ing on research.
seismic signals. Volcanic earthquakes frequently
occur in swarms. Swarms and tremor are both
common precursors to escalations in volcanic
activity. While there is still much that is not Cross-References S
understood – the origin of LP earthquakes for
example - no other technique provides such ▶ Earthquake Location
detailed information about the internal state of a ▶ Earthquake Swarms
volcano, or about debris flows as they occur, or ▶ Frequency-Magnitude Distribution of Seismic-
provides such a detailed chronology of an ity in Volcanic regions
eruption. ▶ Principles of Broadband Seismometry
Volcano-seismic monitoring provides high- ▶ Rockfall Seismicity Accompanying Dome-
sample-rate data, 24 hours a day. Short-period Building Eruptions
seismometers and analog telemetry are gradually ▶ Seismic Anisotropy in Volcanic Regions
being phased out and replaced with modern ▶ Seismic Event Detection
broadband seismometers and field digitizers. ▶ Seismic Tomography of Volcanoes
Two-way telemetry allows observatory staff to ▶ Very-Long-Period Seismicity at Active Volca-
diagnose and fix problems sometimes without noes: Source Mechanisms
2920 Seismic Network and Data Quality
▶ Volcanic Eruptions, Real-Time Forecasting of international handbook of earthquake and engineering

▶ Volcano-Tectonic Seismicity of Soufriere seismology, IASPEI, Palo Alto, CA, 81A: 383–406
McNutt SR (2005) Volcanic seismology. Annu Rev Earth
Hills Volcano, Montserrat Planet Sci 33(1):461–491. doi:10.1146/annurev.
earth.33.092203.122459
Moran SC, Freymueller JT, LaHusen RG, McGee KA,
References Poland MP, Power JA, White RA (2008) Instrumenta-
tion recommendations for volcano monitoring at
Barclay J, Johnstone J, Matthews A (2006) Meteorological U.S. Volcanoes under the national volcano early warn-
monitoring of an active volcano: implications for ing system: scientific investigations report
eruption prediction. J Volcanol Geotherm Res 2008 – 5114, p 47. Retrieved from http://pubs.usgs.
150(4):339–358. doi:10.1016/j.jvolgeores.2005.07.020 gov/sir/2008/5114/
Benoit JP, McNutt SR (1996) Global volcanic earthquake Powell TW, Neuberg JW (2003) Time dependent features
swarm database and prelimary analysis of volcanic in tremor spectra. J Volcanol Geotherm Res 128(1–3):
earthquake swarm duration. Annali Di Geofisica 177–185. doi:10.1016/S0377-0273(03)00253-1
39(2):221–229 Roman DC, Neuberg J, Luckett RR (2006) Assessing the
Buurman H, West ME, Thompson G (2012) The seismic- likelihood of volcanic eruption through analysis of
ity of the 2009 redoubt eruption. J Volcanol Geotherm volcanotectonic earthquake fault–plane solutions.
Res. doi:10.1016/j.jvolgeores.2012.04.024 Earth Planet Sci Lett 248(1–2):244–252.
Chouet B (2003) Volcano seismology. Pure Appl Geophys doi:10.1016/j.epsl.2006.05.029
160(3):739–788. doi:10.1007/PL00012556 Thompson G, McNutt SR, Tytgat G (2002) Three distinct
Chouet BA, Matoza RS (2013) A multi-decadal view of regimes of volcanic tremor associated with the erup-
seismic methods for detecting precursors of magma tion of Shishaldin Volcano, Alaska 1999. Bull
movement and eruption. J Volcanol Geotherm Res Volcanol 64(8):535–547. doi:10.1007/s00445-002-
252:108–175. doi:10.1016/j.jvolgeores.2012.11.013 0228-z
De Angelis S, Fee D, Haney MM, Schneider DJ Thompson G, West ME (2010) Real-time detection of
(2012) Detecting hidden volcanic explosions from earthquake swarms at redoubt Volcano, 2009.
Mt. Cleveland Volcano, Alaska with infrasound and Seismol Res Lett 81(3):505–513. doi:10.1785/
ground-coupled airwaves. Geophys Res Lett 39:1–6. gssrl.81.3.505
doi:10.1029/2012GL053635 Wassermann J (2012) Volcano seismology. In: Peter B
Endo ET, Murray TL (1991) Volcanology a volcano mon- (ed) IASPEI new manual of seismological observatory
itoring and prediction tool. Bull Volcanol 53:533–545 practice 2 (NMSOP-2), second. Potsdam : Deutsches
Jolly AD, Thompson G, Norton G (2002) Locating pyro- GeoForschungsZentrum GFZ, Potsdam, pp 1–77.
clastic flows on Soufriere Hills Volcano, Montserrat, doi:10.2312/GFZ.NMSOP-2_ch13
West Indies, using amplitude signals from high Wech AG, Creager KC (2008) Automated detection and
dynamic range instruments. J Volcanol Geotherm location of Cascadia tremor. Geophys Res Lett 35(20),
Res 118(3–4):299–317. doi:10.1016/S0377-0273(02) L20302. doi:10.1029/2008GL035458
00299-8
Lahr JC, Chouet BA, Stephens CD, Power JA, Page RA
(1994) Earthquake classification, location, and error
analysis in a volcanic environment: implications for
the magmatic system of the 1989–1990 eruptions at
redoubt volcano, Alaska. J Volcanol Geotherm Res Seismic Network and Data Quality
62(1–4):137–151. doi:10.1016/0377-0273(94)90031-0
Luckett R, Baptie B, Ottemoller L, Thompson G (2007) Peter H. Voss1 and Lars Ottemöller2
Seismic monitoring of the Soufriere Hills Volcano, 1
Montserrat. Seismol Res Lett 78(2):192–200.
Geological Survey of Denmark and
doi:10.1785/gssrl.78.2.192 Greenland – GEUS, Copenhagen K, Denmark
2
McNutt SR (1996) Seismic monitoring of volcanoes: A Department of Earth Science, University of
review of the state-of-the-art and recent trends. In: Bergen, Bergen, Norway
Scarpa R, Tilling R (eds) Monitoring and mitigation
of volcano hazards, Chapter 3, Springer-Verlag,
Berlin, pp 99–146
McNutt SR (2000) Volcanic seismicity. In: Houghton HB, Synonyms
McNutt SR, Rymer H, Stix J (eds) Chapter 63 of
encyclopedia of volcanoes. Academic Press, San
Diego CA, 1015-1033
Earthquake monitoring; Seismometer sensitivity;
McNutt SR (2002) Volcano seismology. In: Lee WHK, Instrument response; Ambient noise; Data
Kanamori H, Jennings PC (eds), Chapter 25 of latency
Seismic Network and Data Quality 2921
Introduction earthquake monitoring services to research of the

Earth interior. Examples of seismic networks are
Measuring of ground motion provides the most global seismograph networks (Fig. 1) that are
essential observations made from earthquakes operated to primarily detect large earthquakes
and the basis for research within seismology and and other events globally, regional or national
earthquake engineering. The measurements are seismic networks to detect events within a coun-
done by seismic stations that are equipped with try, microseismic arrays to detect induced earth-
sophisticated instruments and are placed at spe- quakes at a water reservoir or hydrocarbon field,
cific selected locations. The good quality of these and dense monitoring at volcanoes to measure the
data is obviously important but can only be ongoing activity. Individual seismic stations are
achieved if many different factors are considered. commonly part of several networks as sharing of
Often, the approach is adjusted to the scientific or data is common practice. Seismic networks can
monitoring needs and the available funding. This be either permanent or temporary (e.g., http://
section deals with systems of seismic stations, www.usarray.org/), i.e., deployed for a shorter
so-called seismic networks and data quality- limited time period often linked to very specific
related issues. research projects. Normally, less effort is put into
the installation of temporary networks.
Seismic Networks
Seismic Stations
Seismic networks are systems of individual seis-
mic stations with the purpose to monitor ground The main elements of a seismic station are the
motion. The size and configurations of seismic vault and housing of the equipment, the power
networks depend on the focus area and type of supply to the equipment, the sensor, the digitizer
signals that are to be recorded. The data provided with a timing device and recorder, and commu-
by seismic networks (seismograms; see Fig. 3) nication. Figure 2 shows how these units can be
can be used for many different applications from connected at a seismic station, and Fig. 5 shows
Seismic Network and Data Quality, Fig. 1 Map of the Global Seismograph Network (GSN)
GPS
Router
Internet
Datalogger
Sensor
Digitizer
Seismic Network and Data Quality, Fig. 2 Equipment data. A router and Internet connection are standard equip-
used at a typical seismic station. The analog signal from ment for the data transmission. Some units will have
the sensor is converted to a computer-readable format in integrated datalogger and digitizer or integrated digitizer
the digitizer that also time stamp the data with a GPS; a and sensor. The units will naturally also need power
datalogger is storing and handling the transmission the supply
an example of a seismic station. All these ele- sensor. Accelerometers remain on-scale even
ments will have an effect on the data quality and for the largest earthquakes and are the most com-
have to be considered carefully. When building a mon instrument for earthquake engineering
seismic station, there are normally financial con- applications.
straints, and one has to find the optimum solution The main considerations when choosing a dig-
also considering the main monitoring purpose. itizer are the dynamic range and the timing accu-
The main choice for the sensor is between seis- racy. An important setting on the digitizer is the
mometer, also referred to as weak motion sensor, sampling rate, which according to the Nyquist
and accelerometer, also called strong motion sen- theorem allows to resolve frequencies less than
sor. The difference is that seismometers are more half the sampling frequency. A normal sampling
sensitive, particularly at lower frequencies, and rate at a local seismic network is 100 Hz. Infor-
the accelerometer will have a higher clip level mation of earthquake signals above 50 Hz is
than the seismometer. At some seismic stations, therefore not recorded, due to the Nyquist theo-
both a broadband seismometer and an accelerom- rem. The sampling rate is thus set according to
eter are installed to improve the dynamic range of the expected signal frequencies but also restricted
the seismic station. by limitations in data storage and communica-
The main parameters describing the seismom- tion. Communication to a seismic station can be
eter are the sensitivity, the natural frequency, and achieved in many different ways (e.g., cell phone,
the damping. The response of the seismometer is satellite, the Internet, digital radios) and is an
given by its transfer function (impulse response) important element as this allows data to be avail-
that has an amplitude and phase-delay part. Seis- able close to real time. Most seismic stations are
mometers are available with different natural able to send data in this way to one or several
periods. Traditionally, long-period and short- central recording systems, and this is what today
period sensors were used, where the maximum defines a seismic network.
gain would have been at 20 s and 1 s, respec- Before one selects instrumentation for a
tively. Today, broadband sensors are the most seismic station, it should be considered if the
common and typically have the high sensitivity instrument will be able to properly record
between 120 s and 50 Hz. Seismometers are able the earthquakes it is meant for. A way to
to record the lowest possible expected ground verify these requirements is to display the
noise levels within a given frequency band instrument sensitivity and clip levels in
(limited by the instrument self-noise), and this is frequency–amplitude diagrams, with the
the most important technical specification of the expected response of likely earthquakes. Figure 3
100
10
Octave wide bandpassed acceleration [ms−2] 0.1
0.01
0.001
M 7.5
0.0001
1e−05
M 5.5
1e−06 M 3.5
1e−07 M 1.5
1e−08
1e−09
1e−10
1e−11
0.001 0.01 0.1 1 10 100
Frequency [Hz]
Seismic Network and Data Quality, and the upper acceleration amplitude limit in the band
Fig. 3 Frequency–amplitude plot for octave-wide band 0.001–50 Hz is the clip level for the sensor. Sensor limits
passes of ground-motion acceleration. The blue polygon are scaled down to account for the band passing of the
represents the sensitivity limits of an STS-2 broadband data. The red lines show the possible acceleration levels
seismometer, where the frequency limit at 50 Hz is given from different magnitude regional earthquakes that can be
by the Nyquist frequency for a sample rate of 100 Hz. The recorded at a distance of approximately 100 km. The two
lower acceleration amplitude limit in the band black lines show the USGS high- and low-noise models
0.001–50 Hz is the minimum sensitivity of the sensor, (Peterson 1993) (See also Clinton and Heaton 2002)
shows such a diagram for an STS-2 seismometer to the exchange of digital data through the Inter-
with regional earthquakes of magnitudes 1.5, 3.5, net, seismologists would mail the seismograms S
5.5, and 7.5. Note that a magnitude 7.5 is above on paper or microfilm to each other. Today, it is
the clip level in this case, so another type of still common practice (with some exceptions) to
sensor should be considered for earthquakes of exchange and share data between the now enor-
this strength. Figure 4 shows an example of a mous number of seismic stations. This sharing of
clipped seismic signal from a regional data is essential for research projects as well as
earthquake. for seismic monitoring services as earthquakes
often occur in between different networks and
only a combination of the networks allows good
Data Exchange determination of even the basic earthquake
parameters. Data (including waveform data and
The exchange of seismic data has been practiced derived data) can be exchanged in many different
since the installation of the very first stations as ways from email to data downloads from a server
little could be done based on single stations. Prior and receiving data in near real time. There are a
Plot start time: 2008-12-16

05:20:11.6
30 s
4.5 s
Seismic Network and Data Quality, Fig. 4 Seismogram the Z, N, and E, respectively. The signal is clipped since
showing a magnitude 4.2 earthquake recorded at a dis- the signal exceeded the digitizer clip level. The instrument
tance of 87 km on the GID station in Denmark. The is the short-period 4.5 Hz GeoSIG GBV-315 (Data are
channels represent individual sensors oriented in the ver- from the Geological Survey of Denmark and Greenland)
tical, north–south, and east–west directions, given by
number of different public domain proprietary on earthquake (e.g., phase travel times and ampli-
protocols for the near real-time exchange of tudes), is often carried out between neighboring
data between the stations and data centers and networks. On a global scale, the main center for
also for the exchange between data centers. the parametric data is the International Seismo-
Most of these allow for data exchange in near logical Center (ISC) (see http://www.isc.ac.uk).
real time and the recovery of missing data that
may occur due to transmission errors. The public
domain SeedLink protocol is one of the most used Data Processing
for real-time data transmission, especially in
between data centers (Hanka et al. 2000). The The evaluation or processing of the data recorded
SeedLink protocol transmits data in 512 byte by the seismic network normally takes place at
packages, where station, channel, location, or the central recording site. Depending on the pur-
network codes can be requested. The SeedLink pose of a seismic network, different automatic
protocol ensures that data are not lost in case of processing routines are applied to the data when
disconnected transmissions for shorter periods. it comes in. For most networks, the main focus is
This is also common in most commercial soft- on detecting seismic events. This is done by
ware solutions for data transmission in seismic checking the seismic signal and detecting
networks. The largest data center receiving and changes from the background noise in the signal
providing seismic waveform data globally in a at individual stations. The information from a
number of ways is the Incorporated Research number of stations is combined to detect, locate,
Institutions for Seismology (IRIS) Data Manage- and estimate magnitude of the seismic events
ment System (DMS) (see http://www.iris.edu/hq/ using the entire network. This is done for net-
programs/dms). It is also becoming possible to works with different dimensions and monitoring
operate distributed data centers as, for example, purpose requiring different approaches. The out-
done in Europe through the ORFEUS European come from the automatic processing is normally
Integrated Data Archive (EIDA) initiative (http:// inspected before dissemination to the public in
www.orfeus-eu.org/eida/eida.html). Exchange of so-called earthquake early warning systems. This
parametric data, the data that contain information type of warning or information dissemination is
Seismic Network and Data Quality, Fig. 5 Example of Communication and power are available from a nearby
a seismic station installed in southern Norway. The vault hut (right). (Photo taken by Ole Meyer, University of
housing (with the seismometer and digitizer) is Bergen)
constructed from prefabricated concrete rings. (left).
an important responsibility of seismic monitoring (SEED). One part of SEED refers to the wave-
systems. At a later stage, the data are further form data, known as miniSEED; the other part of
processed through manual interactive data SEED, known as the dataless SEED, contains the
processing. For both the automatic and manual metadata. The network codes for SEED are orga-
data processing, there are a number of public nized by the International Federation of Digital
domains as well as commercial software packages. Seismograph Networks (FDSN, http://www.fdsn.
Figure 6 shows phase readings done with the org/). In addition to each component of the seis-
SEISAN software package (http://seisan.info/). mic data, the complete instrument description
makes the data complete. This information
makes it possible to convert the output recorded
Metadata at the seismic station into true ground motion.
Incorrect metadata is often the source to failed
The meta- or descriptive data are an essential part analysis of earthquake records.
of the seismic data as without these the data is not
complete. Data are identified by a so-called
SCNL code (giving stations, component, net- Virtual Seismic Networks
work, and location) and the timing of the data. S
The station code represents the locality of the Virtual seismic networks or virtual nets are the
seismic station, the component code reflects the type of network where data from seismic stations
instrumentation, the seismic network code iden- in different seismic networks are collected and
tifies the operator of the station, and the location processed as a new “virtual” seismic network.
code allows distinguishing between two instru- The virtual seismic networks are often seen in
ments of the same type at one station. The meta- regions in between two or more seismic networks
data associated to the data streams will include where a joint processing of data from the region
information such as the location of the station will improve the detection capabilities. An exam-
(latitude, longitude, elevation), the orientation ple of such virtual seismic network is the Green-
of the sensors (azimuth and dip), the sample land Ice Sheet Monitoring Network (http://www.
rate, and the instrument response. glisn.info/), where data from nine different seis-
The main standard for seismic data is the mic networks are collected and processed jointly
Standard for the Exchange of Earthquake Data to improve the detection capability of glacial
Seismic Network and

Data Quality, Plot start time: 2012-03-14
Fig. 6 Seismogram P 19:22:35.410
displaying an earthquake
recorded at a distance of Z
about 50 km. The 9.5 s of
data shows the three
channels of the seismic
station BLS5 in Norway.
The channels represent
individual sensors oriented
in the vertical, north–south, S
and east–west directions,
given by the Z, N, and E, N
respectively. The P and the
S phase are seen at
19:22:36.5 and 19:22:42.5,
respectively (The data is
from the National
Norwegian Seismic
Network. Data are
unfiltered)
22m36 37 38 39 40 41 42 43 44 SEC
earthquakes in Greenland. Other virtual seismic When talking about “data quality,” the question
networks collect data after quality or instrument often is whether it is sufficient or insufficient. This
type; see, e.g., the _REALTIME and _STS-1 depends on the application of the data. For exam-
virtual networks operated by IRIS (http://www. ple, a seismic station may perform well for local
iris.edu/vnets). A precondition for such earthquake recording, but not provide sufficient
networks to operate is the use of common stan- data for global earthquake monitoring, if the used
dards for real-time data transmission as described instruments are designed only to measure seismic
above. signals above 5 Hz or if the ambient noise level is
very high at the lower frequencies. If data quality is
sufficient, it means that the collected data provides
Seismic Noise and Data Quality the information needed to answer the questions
asked prior to the collection of the data. On the
With seismic waveform data, a number of param- other hand, if data quality is insufficient, it means
eters can be used to quantify data quality. The that the collected data cannot answer the questions
main parameters are ambient noise levels, timing at hand. Data quality is therefore a term used to
accuracy, and data completeness. The noise levels describe if data is adequate for its intended use.
at a seismic station depend on a number of factors The border between sufficient and insufficient data
including the true ambient ground noise, the vault quality is not sharp since data are often collected to
construction, and the instrumentation. Increased answer multiple questions or since the data quality
noise levels at a seismic station result in a lower might change over time during the collection of
signal to noise ratio, which means that a seismic the data.
station is worse in detecting and observing seismic Improving data quality is first of all achieved
events. by reducing the noise in the data, both for seismic
Seismic Network and Data Quality, Fig. 7 Ambient respectively. Body wave and surface wave signals show
noise probability density functions (PDF) of data from the total power of all the earthquake records in the period
the HLID station show how cars, calibrations, and other of the PDF. LNM and HNM are the low- and high-noise
sources are clearly distinguished from the natural ambient models, respectively. Note the peak in the noise at about
noise (McNamara and Boaz 2005). Auto mass re-center 5 s period; this is due to the ocean-generated noise (See
and calibration signals are due to the station operator also http://geohazards.cr.usgs.gov/staffweb/mcnamara/
re-leveling the sensor and checking the amplitudes, PDFweb/Noise_PDFs.html)
stations designed to record local or distant earth- installation process. The instruments should
quakes. The ambient or microseismic noise has always be well grounded to avoid static electric-
both natural and man-made sources. Typical ity. Also ambient variations in the magnetic field
man-made sources are roads, machinery, airports, should be avoided since it can limit the quality
railways, etc., and within some constraints it is (Forbriger et al. 2010). S
often possible to stay away from these. The main The most common way to present the micro-
natural source of seismic noise is ocean waves, seismic noise at a station is to compute accelera-
and while stations can be deployed at some dis- tion power density spectra and to display the
tance from the coastline, it will not be possible to probability of noise levels over a longer time
get away from this noise completely. Wind and period on probability density plots (McNamara
atmospheric changes also result in increased and Buland 2004; McNamara and Boaz
noise, and this can mostly be avoided by building 2005). An example is shown in Fig. 7. This type
better seismic vaults, by putting sensors in of plot allows the identification of the different
pressure-tight enclosures, and also by increasing noise sources that act at different frequencies.
the depth of installation. The seismic instrumen- However, one needs to be aware that the
tation, both seismometer and digitizer, also has “real” ground noise is only resolved in a
self-noise, and this can be minimized when specific frequency range depending on the
choosing the equipment but also during the equipment.
The noise at frequencies greater than 2 Hz is data completeness, gaps (lack of data in a time
related to cultural sources. The local maxima at interval), noise levels, and data latency (data
around 4–8 s and 10–16 s are related to standing arriving late). Most seismic networks record data
waves and waves breaking near the shore, respec- continuously, and measuring data completeness is
tively; see Fig. 7. Even more long-period noise is an efficient tool to evaluate the overall network
related to atmospheric changes. There can be performance. Checking for gaps is typically done
strong differences at individual sites between to identify issues with the communication or
day and night related to the changes in cultural equipment.
activity but also between different seasons. For One parameter that can be checked in near real
example, in northern Europe the noise caused by time is the data latency. An increase in latency will
the ocean is much higher during the winter com- indicate unstable communication. If the latency
pared to the summer. increases continuously, it will indicate that the
Borehole sensors are often used to improve connection to the station is lost or worse the
data quality, since borehole sensors are not as recorder, digitizer, or other units at the station are
affected by noise generated at the surface, as broken. Figure 8 shows an example on latency
sensors placed at or near the surface. The effect where the connection to the seismic station was
is documented by Carter et al. (1991) stating that down for a few days. Monitoring of latency will
945 m below the surface, noise levels between also allow evaluating the quality of communication
15 and 40 Hz are reduced by 10 dB. For periods of during regular operation.
between 30 s and 100 s, Hutt et al. (2002) report It is important that seismometers are installed
that the noise reduction is most prominent at the with correct horizontal orientation so that the
first few tens of meters. The disadvantage of north and east components are oriented within
borehole installations with respect to surface the true geographic system. It is also possible to
installations is that the instruments required are have different orientation, but this needs to be
more expensive to buy and install and servicing recorded with the station’s metadata. The orien-
the sensor is more difficult. tation of sensors can be computed retrospectively
from the recordings of seismic events. For exam-
ple, the “LDEO Seismology Project: Waveform
Network Maintenance and Quality Quality Center” measures misorientation by
Control comparing computed to observed polarization of
surface waves (http://www.ldeo.columbia.edu/%
Seismic networks require a significant amount of 7Eekstrom/Projects/WQC.html). They also
maintenance work and attention to detail. Moni- attempt to measure changes in gain of seismom-
toring of the data quality provides an essential eters over time from recorded data. To record and
tool in the identification of problems with seismic possibly avoid changes in gain, regular calibra-
stations. Being able to identify problems and tion of the seismic equipment should be part of
respond quickly is key to successful operation the network routine operation.
of seismic networks. Recording of changes to When plotting the noise spectra color scaled
instrumentation is also important. over time, one obtains spectrograms that are use-
Data quality can be looked at while the data is ful in evaluating the changes of noise. Failure of
coming in, later to evaluate the performance of a instrumentation will be seen easily as it will look
network or at the end of a temporary deployment. like an apparent sudden change in noise levels. At
Seismic networks make use of tools to monitor the ORFEUS Data Center, PSD plots are gener-
parameters for quality control. One such system is ated in yearly intervals showing frequency bands
QUACK (Quality Analysis Control Kit) available around 0.01, 0.05, 0.5, and 2 Hz (http://www.
at IRIS (http://www.iris.washington.edu/servlet/ orfeus-eu.org/data/data_quality.html). An exam-
quackquery/welcome.do). QUACK provides infor- ple showing a 1-year PSD from the broadband
mation on several parameters such as availability, station SFJD in Greenland is seen in Fig. 9.
Seismic Network and Data Quality, Fig. 8 Latency of record and GMT. The connection to the station was lost on
data streams from the DBG station in Greenland. The April 18th and regained on the 22nd (This figure is gen-
latency is the time between the last sample in the data erated at http://www.iris.edu/servlet/quackquery/)
Seismic Network and IU.SFJD.00.BHZ PSD at selected frequencies in 2011

Data Quality, −80
Fig. 9 Power spectral
0.01 Hz 0.05 Hz 0.05 Hz 2.0 Hz
density (PSD) plot of data
from the SFJD station in −100
PSD (10° log (m2/s−1/Hz))
Greenland, covering the

year 2011. The black, red, −120
green, and blue lines show
the frequencies 0.01 Hz,
−140
0.05 Hz, 0.5 Hz, and 2.0 Hz, S
respectively ( With
permission from Dr. −160
Sleeman, ORFEUS. This
figure is obtained from
−180
http://www.orfeus-eu.org/)
−200
0 50 100 150 200 250 300 350
Julian day
Site Selection and Vault Construction quality. One can install instruments of the highest
quality at sites where a high ambient noise level
The choice of site, selected to record seismic will make the data useless for the purpose they
data, is essential to obtain data with a high were intended for. Before deciding on a specific
Seismic Network and

Data Quality,
Fig. 10 Example of BB
vaults from the GEOFON
network. (a) Underground
bunker vault for remote
recording, (b) wide and
shallow borehole type, and
(c and d) simple bunker
construction. Note that b–d
allow onsite recording
since there is a separate
recording room (See also
Fig. 7.55 by W. Hanka in
NMSOP, Vol.1, Bormann
(Ed.), 2002)
site, it is common practice to carry out a site relatively deep substantial vaults to rather shal-
survey where all relevant criteria are considered low and simple installations (Fig. 10). The cou-
and noise measurements are done for a few days pling of the sensor to bedrock is very important in
to identify potential noise sources. If the source of all types of installation. If the budget is not lim-
the seismic signal that one wishes to record is ited, one would attempt to build a pier placed on
known, for example, a small area that is known to bedrock several meters below the surface.
be seismically active, one should aim to get as Around the pier, a room would be constructed
close to the source as possible. One should also that would be isolated from the pier. It is essential
aim to have stations installed with the highest to achieve rather stable temperatures and pressure
azimuthal coverage possible. It is of course around the sensor. Another room may be placed
important to stay away from the man-made on top to house the remaining equipment. In a
sources of noise mentioned above. Site selection simpler setup, one can use a concrete structure
is often dependent on other factors than the ambi- with a smaller diameter that has the sensor at the
ent noise, such as access to power and communi- bottom, a few meters below the surface. Care has
cation as well as the physical access to the site. to be taken to either drain water away or avoid
One might therefore not always be able to use the water ingress completely. Many such designs
site with the lowest noise level, since it does not have been made taking into account local condi-
meet these requirements. tions and available materials. The type of vault
Constructing good vaults is also very impor- chosen should also be based on the specifications
tant, and possibilities range from boreholes to of the equipment that will be used.
Seismic Network and

Data Quality,
Fig. 11 Detection
capability of the UK
seismograph network. Unit
is magnitude on the Richter
scale. Green triangles are
broadband stations, red
triangles are short-period
stations, and blue triangles
are broadband stations
operated by other agencies
(http://www.earthquakes.
bgs.ac.uk/)
Detection Capability References
If detection of earthquakes is the main purpose of Carter JA, Barstow N, Pomeroy PW, Chael EP, Leahy PJ
(1991) High-frequency seismic noise as a function of
a seismic network, knowledge on the detection depth. Bull Seismol Soc Am 81:1101–1114
capability of the network is essential when the Clinton J, Heaton T (2002) Potential advantages of a
network is designed and evaluated. The detection strong-motion velocity meter as opposed to a strong S
capability depends on the level of the background motion accelerometer. Seismol Res Lett 73:332–342
Forbriger T, Widmer-Schnidrig R, Wielandt E, Mark
noise, the type of instrumentation, and the con-
Hayman M, Ackerley N (2010) Magnetic field back-
figuration of the network. The detection capabil- ground variations can limit the resolution of seismic
ity is mostly improved by lower background broad-band sensors. Geophys J Int 183(1):303–312.
noise in the frequency range of interest, the dis- doi:10.1111/j.1365-246X.2010.04719.x
Hanka W, Heinloo A, Jaeckel KH (2000) Networked seis-
tribution of stations, and their distance to the area
mographs: GEOFON real-time data distribution.
of interest. An example of detection capability is ORFEUS Newsl 2(3):1–24
shown in Fig. 11. Here it is assumed that the Havskov J, Alguacil G (2010) Instrumentation in earth-
amplitude threshold for the detections of an quake seismology. Springer, Dordrecht, p 349
Havskov J, Ottemöller L (2010) Routine data processing
earthquake must be twice the background noise in earthquake seismology with sample data, exercises
level and that the earthquake is recorded on at and software. Springer, New York, p 380
least four stations (http://www.earthquakes.bgs. Hutt CR, Bolton HF, Holcomb LG (2002) Seismograph
ac.uk/monitoring/detection_capability.html). networks. In: Lee WHK, Jennings P, Kisslinger C,
2932 Seismic Noise
Kanamori H (eds) International handbook of earth- seismic source. They can be used to locate the
quake & engineering seismology, part 1. Academic source, to analyze source processes, and to
Press, p 1200
McNamara DE, Boaz RI (2005) Seismic noise analysis study the structure of the medium of wave
system, power spectral density probability density func- propagation. In contrast, the term “seismic
tion: stand-alone software package, United States geo- noise” designates undesired components of
logical survey open file report, No. 2005–1438, p 30 ground motion that do not fit in our conceptual
McNamara DE, Buland RP (2004) Ambient noise levels in
the continental United States. Bull Seismol Soc Am model of the signal under investigation. What
94(4):1517–1527 we identify and treat as noise in seismic studies
Peterson J (1993) Observations and modeling of back- depends on the available data, on the aim of
ground seismic noise. Open file report 92–302, our study, and on the method of analysis.
U.S. Geological Survey, Albuquerque
Scherbaum F (2001) Of poles and zeros. Fundamentals of Accordingly, data treated as noise in one con-
digital seismology, vol 15, 2nd edn, Modern text may be considered as useful signals in
approaches in geophysics. Kluwer, Boston, p 269 other applications. For example, short-period
noise (frequency >0.2 Hz) can be used for
Further Reading investigating the potential amplification of
The most complete coverage of these topics is probably earthquake-generated ground motions due to
given in the New Manual of Seismological Observa-
tory Practice (http://http/nmsop.gfz-potsdam.de), giv- local geological site conditions, also termed
ing an overview, the theoretical background, as well as microzonation (e.g., Parolai 2012), and gener-
examples and information sheets. Textbooks on seis- ally, both short- and long-period noise
mic instrumentation are Havskov and Alguacil (frequency < 0.1 Hz) can also be used for
(2010) and Scherbaum (2001). A textbook on data
processing for earthquake seismology is Havskov and tomographic studies of inhomogeneities in the
Ottemöller (2010). Stein and Wysession (2003) and Earth’s crust (e.g., ▶ Noise-Based Seismic
Lay and Wallace (1995) also have relevant chapters. Imaging and Monitoring of Volcanoes).
According to Bormann and Wielandt (2013),
disturbing noise in seismic records in a wider
sense may comprise:
Seismic Noise
• Ambient vibrations due to natural sources
Peter Bormann (like ocean microseisms, wind, etc.)
Formerly GFZ German Research Center for • Man-made vibrations (from industry, traffic,
Geosciences, Potsdam, Germany etc.)
• Secondary signals resulting from Seismic
Wave Propagation in inhomogeneous medium
Synonyms (due to wave scattering and therefore also
termed signal-generated noise)
Ambient noise; Cultural noise; Digitization • Effects of gravity (like Newtonian attraction
noise; Instrumental noise; Man-made noise; of atmosphere, horizontal accelerations due to
Microseismic noise; Microseisms; Self-noise; surface tilt)
Signal-generated noise • Signals resulting from the sensitivity of
▶ Broadband Seismometers to ambient condi-
tions (like temperature, air pressure, magnetic
Introduction field, etc.)
• Signals due to technical imperfections or dete-
Seismic signals are usually transient waveforms rioration of the seismic ▶ Passive Seismome-
radiated from a localized natural or man-made ters (corrosion, leakage currents, defective
semiconductors, etc.)
• Intrinsic ▶ Seismometer Self-Noise and
Peter Bormann: deceased. Measuring Methods of the Seismic
Seismic Noise 2933
Instrumentation (like Brownian noise, elec- interest in ▶ Ocean Bottom Seismometer (OSB)
tronic and quantization noise) installations in order to improve the global ray
• Artifacts from data processing path coverage for tomographic studies of the
Earth and for regional studies of seismicity and
However, only the first four items are micro- Earth structure in the ocean environment. But
seismic noise in a proper sense and elaborated in OBS deployment, operation, and maintenance
the following whereas the other ones relate to are rather expensive and logistically demanding,
seismometry and data processing and are referred and the noise conditions on the ocean floor are
to only en passant. usually worse than on land (Shinohara et al. 2011;
The study of earthquakes or the imaging of Webb 2002). Therefore, this essay will deal, also
the Earth with seismic wave arrivals requires because of strict space limitations, only with
▶ Seismic Event Detection above background microseismic noise conditions on land.
noise. Levels of natural ambient noise may More details about such noise in both environ-
vary by 60 dB (i.e., a factor of 1,000 in ampli- ments as well as on the specifics of seismic sig-
tude) depending on location, season, time of nals in different bandwidth ranges and related
day, and weather conditions. This corresponds noise-dependent local and global detection
to differences in detection thresholds for seis- thresholds can be found in Bormann and
mic arrivals by about three magnitude units. Wielandt (2013).
Therefore, prior investigation of noise levels With respect to historical aspects of the
at potential recording sites is of utmost impor- recording and treatment of microseismic noise,
tance, as is environmental shielding and proper it has to be considered that most of the early
installation of seismic Sensors so as to reduce twentieth-century seismographs by Wiechert,
the influence of various sources of natural and Mainka, Galizyn, Bosch-Omori, Milne-Shaw,
man-made (cultural) noise (see sections and others were mechanical medium-period
“Causes and Basic Characteristics of Ambient broadband sensors of low magnification of the
Noise Observed on Land” to “Improving the ground motion. Only the more sensitive ones
Signal-to-Noise Ratio SNR by Data with 100–500 times magnification were already
Processing” below; Trnkoczy et al. 2011; able to record the ever-present ground motions
Forbriger 2012). with peak amplitudes around 6 s which constitute
The essay will brief on: the dominating background noise for any seismic
measurement with maximum amplitudes in the m
• Different mathematical representations of micrometer range (Fig. 1). They were termed
noise and signals microseisms. Such recordings were first reported
• Origins of natural and man-made ambient by Algue in 1900. Wiechert (1904) wrote that
noise these microseisms are caused by ocean waves S
• Basic features of noise in the wide frequency on coasts. Later it was found that it is necessary
range of interest for the recording of local to discriminate between: (a) the smaller primary
micro-earthquakes to normal modes of the ocean microseisms with periods between 10 and
Earth (0.2 mHz < f < 100 Hz) 20 s, typically around 14 2 s and (b) the sec-
• Influence of filter procedures and the sensor ondary or double frequency microseism which is
environment on the appearance of signals and related to the main noise peak around 6 2 s
noise in seismic records (Figs. 1 and 2).
• Task-dependent procedures for signal-to- Other types of seismic as well as cultural and
noise ratio (SNR) improvement of earthquake instrumental noise became more or less
and explosion records disturbing only when the sensitivity of electro-
magnetic seismographs, and later of electroni-
More than 98 % of all seismological record- cally amplified seismographs, was raised so
ings are made on land, despite the growing much that they were able to resolve even ground
2934 Seismic Noise
Seismic Noise, Fig. 1 STS-2 velocity broadband record frequency waves of a local earthquake of magnitude
of the 6 s secondary ocean microseisms at the Black Forest ML = 2.2 that occurred 32 km away (Figure by courtesy
Observatory (BFO), Germany, superposed by high- of Thomas Forbriger for NMSOP Chapter 4 # IASPEI)
motion amplitudes in the size range of molecules 30–80 dB, but they are much larger in velocity
and atoms (109 to 1010 m). power (>130 dB) and displacement power
Noise conditions at seismic recording sites (>250 dB). The latter is much more than the
vary globally in the wide range of periods of dynamic range of common modern digital seis-
seismological interest between some upper and mographs with 24 bit analog-to-digital con-
lower level, termed the New High Noise Model verter (ADC) and a dynamic range of
(NHNM) and the New Low Noise Model 144 dB. In order to resolve, at good sites, at
(NLNM) according to Peterson (1993). These least the ambient noise with a few bits, even at
noise models are expressed in dB of power spec- the NLNM level, and still record without clip-
tral density (see section “Nature and Theoretical ping also seismic signals from events in a rea-
Representation of Signals and Noise”) of accel- sonably large range of magnitudes and
eration related to 1 (m/s2)2/Hz (Fig. 2). However, epicentral distance, current broadband seismo-
the actual noise at any recording site corresponds graphs used are velocity sensors such as the
nowhere in the whole range of periods to either STS-1 and STS-2 ▶ Recording Seismic Signals.
the NHNM or the NLNM. Some sites and seis- Their dynamic range is given in Fig. 3 with
mographs are more affected by long-period tilt- respect to the NLNM when measured in a con-
noise at a period above 30 s, and this differently stant relative frequency bandwidth of 1/3
in each record component, while others are octave and represented by “average peak”
mostly disturbed by high-frequency (f > 1Hz) amplitudes (see section “Approximate Conver-
man-made noise. sion of Power Densities into Recording Ampli-
According to Fig. 2, variations in noise accel- tudes”). In contrast, classical analog records
eration power are typically within about on paper or film had only a dynamic range of
Seismic Noise 2935
Seismic Noise,
Fig. 2 Envelope curves of
acceleration noise power
spectral density Pa (in units
of dB relative to 1 (m/s2)2/
Hz) as a function of noise
period (according to
Peterson 1993). They
define the new global high
(NHNM) and low noise
models (NLNM) which are
currently the accepted
standard curves for
generally expected limits of
microseismic noise
depending on period. No
single station site, however,
meets in the whole period
range the NLNM or
NHNM. However, some
exceptional seasonal,
diurnal, or local site noise
conditions may exceed
these limits. Also shown are
for the NLNM the related
curves calculated for the
displacement and velocity
power spectral density Pd
and Pv in units of dB
relative to 1 (m/s)2/Hz and
1 m2/Hz (Modified version
of Fig. 2 from P. Bormann
1998 # Springer; with kind
permission from Springer
Science + Business Media)
some 40 dB. They could cover only a very Nature and Theoretical representation
limited range of noise and record signal ampli- of Signals and Noise
tudes between about 1 mm and 100 mm. Cov-
ering a range of 144 dB instead would Fourier Transformation of Continuous and S
correspond to covering an amplitude range Sampled Transient Signals
between 1 mm and 16 km! However, what the Seismograms contain noise and the desired sig-
actual usable dynamic range of a seismograph nals. Sometimes the signal is completely masked
means has been discussed in Sect. 4.2.2 of by the noise. One of the main tasks in applied
Bormann and Wielandt (2013). seismology is to ensure a good signal-to-noise
The reasons for the main types of natural and ratio (SNR) both by selecting recording sites
cultural microseismic noise observed on land and with low ambient noise and by data processing.
strategies for their avoidance or reduction in seis- The success of the latter largely depends on our
mic records are discussed in the following sec- understanding of the ways in which seismic sig-
tions after a brief discussion on the principal nals and noise differ. The basic mathematical tool
difference between seismic signals and noise, for this purpose is the harmonic or Fourier anal-
their mathematical representation, and the correct ysis, that is, the decomposition of a signal into
way of comparison. sine waves. Depending on the type of signal
2936 Seismic Noise
Seismic Noise, Fig. 3 A representation of the bandwidth (also Sect. 4.3.2 in Bormann and Wielandt 2013). The
and dynamic range of conventional analog World Wide scale is in decibels (dB) relative to (1 m/s2)2. Noise is
Standard Seismograph Network (WWSSN) short- and measured in a constant relative bandwidth of one third
long-period and digital broadband seismographs (STS1 octave and represented by “average peak” amplitudes
and STS2 with good shielding - see Wielandt 2011 and equal to 1.253 times the root-mean-square (RMS) ampli-
Forbriger 2012) with respect to the equivalent Earth peak tude. NLNM relates to the global New Low Noise Model
acceleration generated by solid Earth tides due to the according to Peterson (1993; see Fig. 2) The requirement
attraction forces exerted by the Sun and the Moon, by a to be fulfilled for presenting seismic signal, noise, and clip
magnitude 8 earthquake at about 3,300 km distance or by a levels in one diagram has been outlined in Sect. 4.3.3 of
local earthquake of magnitude 4 at about 10 km distance Bormann and Wielandt (2013). Copy of Fig. 4.7 in
(see ▶ Earthquake Magnitude Estimation). The depicted Bormann and Wielandt (2013), compiled by using data
lower bound is determined by the Instrumental Self-noise of W. Hanka and J. M. Steim; # IASPEI
(transient or stationary, continuous or sampled) ð

1
1
different mathematical formulations must be f ðtÞ ¼ ð2pÞ FðoÞexpðiotÞdo (1)
used which have been outlined in detail in Sect. 1
4.2 of Bormann and Wielandt (2013). The
essence is that according to the Fourier theorem, ð
1
any arbitrary transient function f(t) in the FðoÞ ¼ jFðoÞjexpðifðoÞÞ ¼ f ðtÞexpðiotÞdt

time domain can be represented by an equivalent 1
function F(o) in the frequency domain. The (2)
mathematical relationship between the two
domains is defined by the Fourier integral where o (in units radian/s) = 2pn is the circular
transformation: frequency with n = common frequency
Seismic Noise 2937
(in units s1), |F(o)| = amplitude spectrum, and The concept of frequency filtering (e.g., high-,
f(o) = phase spectrum. However, the Fourier low-, or band-pass filtration) has its roots in ana-
integral transformation can only be applied to log signal processing and is therefore intimately
transient signals (i.e., signals that disappear or connected to the Fourier integral transformation.
decay after some time so that they have a finite Practical computer procedures of the discrete
energy). For other signals, the integral may Fourier transformation, however, do not use
diverge. Moreover, the signals should have been integrals and are often only approximations to
radiated by localized finite sources thus being the corresponding analog procedures. Supple-
coherent signals with a defined phase relation- mentary information on different aspects of the
ship. In the case of digital data, the discrete trans- Fourier transformation can also be found in Sect.
formation has to be applied. It also requires a 5.2.3 of Wielandt (2011).
sampled signal of finite duration. In contrast, A general problem is that frequency filtering
signals which do not vary much within the inter- reduces the amplitudes of harmonic components
val of time over which earthquake signals are of a signal in part of the frequency spectrum,
typically analyzed are mathematically treated as usually in order to suppress noise. Examples are
stationary signals, that is, signals whose statisti- given in section “Frequency Filtering.” But filter-
cal properties do not change with time. Such ing may also change the waveform of the desired
signals require a different mathematical signal in the time domain, the more so the
treatment. narrower bandwidth. Systematic differences
Normally, the detailed waveform of station- between various magnitude scales as well as
ary signals is of no interest but rather magnitude saturation effects are the consequence
subjected to efforts to remove the disturbing of filtering the input signals with different
signal from the record. The exception is the responses in different and often rather limited
above-mentioned application in microzonation frequency ranges (see ▶ Earthquake Magnitude
studies (or noise tomography). However, it is Estimation).
of general interest to know how strong such
stationary signals are in different parts of the Spectral Analysis of Stationary Signals (Noise)
frequency spectrum. This information is The analysis of stationary noise usually aims at
contained in a quantity named power spectral assessing “how strong” the noise is or whether
density (PSD). it will prevent the detection of a specific earth-
Transient signals such as earthquake quake signal. Frequency filters may help to
seismograms may have a complicated waveform suppress noise in one part of the spectrum but
which is determined by the earthquake source, still preserve signals in another part. Therefore
the structure of the Earth, and the properties it is necessary to consider the spectral distribu-
of the seismograph. The amplitudes and phases tion rather than the total strength of both types S
of the harmonic components of the signal are not of signals. However, the formal measures of
random numbers but follow certain mathematical “strength” for transient and stationary signals
relationships. Such waveforms or spectra are are incompatible and cannot easily be com-
sometimes called deterministic, in contrast to pared. Mathematically speaking, transient sig-
the “stochastic” or “random” waveform of nals have a finite energy and zero power (in the
noise. However, in a strict sense, even microseis- average over all times) while stationary signals
mic noise is deterministic as well. But the differ- have an infinite energy but a finite power. The
ence is that its sources and propagation paths are same holds for the spectral densities. For a
not known well enough to predict or interpret the quantification of stationary noise, the concept
waveform, and therefore it is assumed that the of power spectral density (PSD) is therefore
spectral amplitudes and phases are random num- appropriate while transient signals are more
bers. Again, this is a mathematical simplification adequately described by their energy spectral
that may or may not be appropriate. density (ESD).
2938 Seismic Noise
The Fourier integral transformation as The time derivative f˙(t) of a signal f(t) has the
formulated in Eqs. 1 and 2 has the mathematical Fourier transform joF(o). Thus, if the signal was
property (known as Rayleigh’s or Parseval’s originally measured as a displacement and the
theorem) Fourier transform of the velocity is of interest
then it is obtained by multiplying the original
ð
1 ð
1 ð
1 Fourier transform with jo ¼ j 2pn and the Fou-
2 jFðoÞj2
f ðtÞ dt ¼ do ¼ jFð2pvÞj2 dv rier transform of the acceleration is then
2p
1 1 1 o2 FðoÞ . Going back from acceleration or
(3) velocity to displacement is achieved by dividing
with the appropriate factor. The ESD and PSD are
The left side represents, in data processing, positive quadratic functions of the Fourier ampli-
the total energy of a signal f(t) which is not the tudes. Therefore, differentiating the signal brings
physical energy but proportional to it. The a factor o2, integration a factor o2, etc. A PSD
right side represents the energy as an integral of of velocity can thus easily be converted into that
E = |F(2pn)|2 over all frequencies n. Thus, of displacement or acceleration.
according to the above definition, the integrand
is the energy spectral density E of the signal.
Note that negative frequencies must be Representing PSDs in Decibels
included in the integration. For real signals, the As shown in Fig. 2, it has become a standard to
ESD is a symmetric function, E(2pn) = E(2pn). measure microseismic noise as a PSD of acceler-
Therefore, by multiplying by 2, the integration ation, that is, in (m/s2)2 per Hz or m2s3. But
can be carried out over positive angular frequen- “power” in this context is not the physical power
cies only. This convention is normally used in in watts but simply the mean square amplitude of
engineering and other practical applications; the ground acceleration. As in acoustics, the power
ESD is then given as twice the “mathematical” ratio r = (a2/a1)2 between two signals of ampli-
value. The same applies to the PSD. It is impor- tude a1 and a2 is often expressed in decibels (dB).
tant to clarify which convention is being used. This is a logarithmic measure that expresses a ratio
Power is energy per time, usually expressed as as a difference of logarithms. The difference in dB
the average power over at least one cycle of an is 10 log10[(a2/a1)2] = 20 log10[(a2/a1)]. When
approximately periodic signal. The concept of expressing the power spectral density Pa of accel-
power is mainly applied to stationary signals eration relative to the metric unit 1 (m/s2)2/Hz, it
such as sine waves or seismic noise whose aver- holds that
age power is constant or varies slowly. A Fourier
transform in a strict sense is not defined for h 2 i
such signals, but the signal over any finite time Pa ½dB ¼ 10 log10 Pa =1 m=s2 =Hz (4)
interval of length T can be Fourier analyzed.
When both sides of Eq. 3 are divided by T, the
left side represents the average power, and In Fig. 2, however, a comparison has been made
P = E/T = |F(2pn)|2/T = |F(o)|2/T on the right of the respective numerical values for physically
side is then the power spectral density P(o) for incommensurable quantities such as acceleration
the interval T. But the PSD is not equivalent to a power density with velocity and displacement
Fourier transform! The signal cannot be power density. Therefore, Eqs. 2 and 3 and the
reconstructed from it with an inverse transforma- following two text lines do not imply that the
tion because the phase information is lost. The decibel units on both sides of the equal sign
smoothed and decimated PSD is however nor- mean the same. They only look the same because
mally more useful because it contains all essen- the incommensurable reference levels were omit-
tial information in a concentrated form and ted. For this reason, three decibel scales for accel-
eliminates arbitrary normalization factors. eration power, velocity power, and displacement
Seismic Noise 2939
power have been given in Fig. 2. This clarifica- therefore meaningful to ask whether the signal
tion notwithstanding, it holds that numerically recorded at one location is coherent with the same
signal recorded at another location. This aspect of

Pv ½dB ¼ Pa ½dB þ 20 log o1 (5) coherence is usually expressed as a “coherence
length” over which the recorded signals have a
and specified coherence. The coherence length is only
a fraction of a wavelength for omnidirectional

Pd ½dB ¼ Pa ½dB þ 40 log10 o1 noise (noise propagating in all directions) but
can be much longer when noise originates in a
¼ Pv ½dB þ 20 log o1 (6) limited source area and propagates essentially in
one direction towards the station (Fig. 5). In a
Consequently, for o = 1 (period T = 2p/o plane wave, the coherence length is theoretically
= 6.28 s) it holds that Pa = Pv = Pd (see infinite.
Fig. 2) although only the numbers are equal!
Also, (Pd Pa) = 2 (Pv Pa) = constant
for any given period, negative for T < 2p s, and Comparing Spectra and Amplitudes of
positive for T > 2p s. Transient and Stationary Signals
Cross-correlation and Coherence of Seismic Bandwidth and Amplitude
Signals and Noise Transient signals may be impulsive and oscilla-
The cross-correlation of two continuous station- tory, and stationary signals either random signals
ary signals f(t) and g(t) at the time lag t is defined (noise) or monochromatic sine waves. A typical
as the expectation value (practically, the mean seismogram belongs to the “oscillatory” cate-
value) of the sliding dot product f(t) g(t + t): gory. The seismic signal near the focus of an
ð1 earthquake should be impulsive. Ambient noise
ðf ⋆gÞ ðtÞ ¼
def
f
ðtÞ gðt þ tÞdt; (7) is often a superposition of all types, the “random”
1 components being due to a large number of inde-
pendent sources (traffic, wind, waves in the
where f * denotes the complex conjugate of f. ocean), “transient” components being caused by
Properly normalized, the cross-correlation is a identifiable sources such as passing cars, and
measure of the similarity between the two signals nearly monochromatic components originating
and is called their coherence. Like the power spec- from machinery such as sawmills and pumps.
tral density, it can be calculated via the Fourier When the PSD of a stationary signal is con-
transformation and specified as a function of fre- stant over the bandwidth of a seismograph or a
quency, in which case the name coherence spec- filter then, when reducing the filter bandwidth, S
trum is more adequate. Each signal is fully the signal power is proportional to the bandwidth
coherent with itself (auto-correlation with the because it is the integral of a constant power
coherence equaling 1). Independent random sig- density over the bandwidth. The rms amplitude
nals are incoherent (their coherence equals 0) of the signal, which is the square root of the
although strictly this is only true in the limit of an power, must therefore be proportional to the
infinite duration and after removing the mean. square root of the bandwidth. This explains why
More interesting is the case that signals are par- a “spectral amplitude density” cannot be gener-
tially coherent. Then, with some (frequency- ally defined. Such a definition would imply that
dependent) fraction of the power, a common signal the amplitude is generally proportional to the
may be hidden in them, for example, a seismic bandwidth, which it is not (see Fig. 4.6 in
signal masked by instrumental noise of the sensors. Bormann and Wielandt 2013). It may neverthe-
Seismic signals depend not only on time but less be so in special cases. Consequences of band-
also on the location where they are recorded. It is width differences of seismographs or filter
2940 Seismic Noise
procedures applied for the evaluation of Stationary signals must be characterized by their
seismograms have been discussed in detail in PSD, and specifying noise by its RMS amplitudes
Sect. 4.5 of Bormann and Wielandt (2013). is meaningless without definition of the band-
width. The values given by the NLNM and
Approximate Conversion of Power Densities NHNM in Fig. 2 follow the engineering power
into Recording Amplitudes (Pe) convention, and the following related formu-
According to Aki and Richards (1980), the max- las as well.
imum amplitude of a wavelet f(t) near t = 0 can Often it is necessary to represent the amplitude
be roughly approximated by the product of of noise (whether RMS or average-peak) in a
what they call the “amplitude spectral density” constant relative bandwidth RBW over the whole
F(o)| (see Eq. 2) and the bandwidth of the frequency range. This means that the bandwidth B
wavelet, i.e., around each frequency n0 is a fixed fraction or
multiple of n0. The lower and upper band limits,
f ðtÞt¼0 ¼ jFðoÞj2ðvu v1 Þ (8) fl and fu, are chosen so that f0 is the geometric mean
of the two (i.e., n0 appears in the middle between nl
with nu and nl being the upper and lower corner and nu on a logarithmic frequency scale). When
frequencies of the band-passed signal. Such an the relative bandwidth is n octaves or m decades
approximation is only possible for signals whose (n and m normally being fractions, not integers),
energy is maximally concentrated in time. Like- then the following relationships hold:
wise, if the power spectral density P of noise is
defined as v1 ¼ 2n=2 v0 , vu ¼ 2n=2 v0 , RBW
ð
1 ð
1
2 ð
1 ¼ 2n=2 2n=2 and B
jFðoÞj
f ðtÞ2 dt ¼ do ¼ jFð2pvÞj2 dv ¼ 2n=2 2n=2 v0 (13)
2p
1 1 1
(9)
and (12) can then be written as
then the mean square amplitude of noise in the h i1=2
time domain is aRMS ¼¼ Pe v0 2n=2 2n=2
< f 2 ðtÞ >¼ 2Pðvu v1 Þ (10) ¼ ðPe v0 RBWÞ1=2 (14)
This is of course a simplified version of the gen- Octaves can be converted easily into decades and
eral rule that the power spectral density (PSD) vice versa by using the relation
must be integrated over the passband of a filter to
obtain the power (or mean square amplitude) at m ¼ n log10 ð2Þ ¼ n 0:3010 (15)
the output of the filter. The square root of this
power is then the root mean square (RMS) or Typical response curves of short-period narrow-
effective amplitude band analog seismographs for recording transient
teleseismic body waves (i.e., waves that have trav-
aRMS ¼ ½2P ðvu v1 Þ1=2 (11) elled more than 2000 km either as longitudinal
waves, oscillating in the propagation direction, or
However, if the power spectral density is defined transverse waves, oscillating perpendicular to the
according to the engineering approach only by its direction of propagation) typically have band-
positive frequencies as Pe = 2P, then widths between 1 to 2 octaves. Different authors
have used an integration bandwidth of 1/2 to 1/3
octave (a standard bandwidth in acoustics) for
aRMS ¼ ½Pe ðvu v1 Þ1=2 (12)
computing RMS amplitudes from PSD. According
Seismic Noise 2941
Seismic Noise, Fig. 4 Schemes for the generation of (a) distance between the coast and X, p water pressure, Dp
primary and (b) secondary microseisms (for explanations water pressure fluctuation proportional to wave amplitude
see text). L cyclone low-pressure area, X area of interfer- (Copy of Fig. 12 from Friedrich et al. 1998 # Springer;
ence where standing waves with half the period of ocean with kind permission from Springer Science + Business
waves develop, r distance between L and the coast, Dr Media)
to the above relationships, 1/3 octave RMS ampli- constructively, are termed coherent. This is usu-
tudes will be only about 82 %, 70 %, and 50 % of ally the case for seismic signals generated and
the respective RMS amplitudes calculated for 1/2, radiated by a common source process. The degree
2/3, or 4/3 octave bandwidth, respectively. of coherence is defined by the ratio between the
There is a 95 % probability that the instanta- cross-correlation and the autocorrelation of the
neous amplitude of random noise with a Gaussian time series. The cross-correlation is squared and
amplitude distribution will lie within a range of divided by both autocorrelations to render the
2aRMS. Peterson (1993) showed that both result independent of the signal amplitudes.
broadband and long-period noise amplitudes fol- It may vary between 0 and 1. For noise, it shows
low closely a Gaussian probability distribution. distinct frequency dependence. Spatial coherence
This is true for largely natural ambient noise, may be rather high for long-period ocean micro-
often not, however, for short-period and broad- seisms. Accordingly, the correlation radius, i.e.,
band noise in urban and industrialized areas. the longest distance between two seismographs
There the noise is often dominated by transient for which the noise recorded in certain spectral
or periodic signals (Groos and Ritter 2009 and ranges is still correlated, increases with the noise
section “Microseismic Noise in Urban Environ- period. It may be several kilometers to tens of
ments”). Yet, in case of a Gaussian probability kilometers for f < 1 Hz but drops to just a few
distribution, the absolute peak amplitudes of the tens of meters or even less for f > 50 Hz.
narrowband filtered signal envelopes should fol- For random noise, it is usually not larger than a S
low a Rayleigh distribution. In the case of an few wavelengths.
ideal Rayleigh distribution, the theoretical aver-
age peak amplitudes (APA) are 1.253 aRMS. Primary and Secondary Microseisms
Therefore, RMS amplitudes in 1/6 decade band- Primary ocean microseisms are generated only in
width correspond approximately to average peak shallow waters in coastal regions. Here the wave
amplitudes in 1/3 octave bandwidth. energy can be converted directly into seismic
energy either through vertical pressure variations
(Fig. 4a) or by the smashing surf on the shores.
Causes and Basic Characteristics of Therefore, primary ocean microseisms have the
Ambient Noise Observed on Land same period as the dominating water waves
(T 10 to 16 s). Haubrich et al. (1963) com-
Signals which have a similar waveform and pared the spectra of microseisms and of swell at
polarization, so that they can interfere the beaches and could demonstrate a close
2942 Seismic Noise
relationship between the two data sets. The rela- This allows to locate the source areas and track
tive stability in amplitude for continental sites their movement by means of seismic arrays (e.g.,
around the globe suggests persistent, stable Cessaro, 1994; Friedrich et al. 1998, Fig. 5). This
sources. possibility has already been used decades ago by
Contrary to the primary ocean microseisms, some countries, e.g., in India, for tracking
the secondary ocean microseisms could be approaching monsoons with seismic networks
explained by Longuet-Higgins (1950) and under the auspices of the Indian Meteorological
Hasselmann (1963) as being generated by the Survey. While near-shore areas may be the
superposition of ocean waves of equal period source of both primary and secondary micro-
traveling in opposite directions, thus generating seisms, the pelagic sources of secondary micro-
standing gravity waves of half the period. These seisms meander within the synoptic region of
standing waves cause nonlinear pressure pertur- peak storm wave activity.
bations that propagate without attenuation to the Note that the noise peak of secondary micro-
ocean bottom and there couple into an elastic seisms has a shorter period when generated in
wave of much larger wavelength and higher shallower inland seas or lakes (T 2–4 s)
phase velocity if the propagation directions are instead of in deep oceans because of generally
nearly opposite. The area of interference shorter wavelengths and periods of the water
X (Fig. 4b) may be off-shore where the forward waves in shallower and smaller water bodies.
propagating waves, generated by a low-pressure Also, off-shore interference patterns largely
area L, superpose with the waves traveling back depend on coastal geometries and the latter may
after being reflected from the coast. But it may allow the development of internal resonance phe-
also be in the deep ocean when the waves, excited nomena in bays, fjords, or channels (Fig. 6),
earlier on the front side of the low-pressure zone, which affect the detailed spectrum of micro-
interfere later with the waves generated on the seisms. In fact, certain coastlines may be distin-
back-side of the propagating cyclone. Thus, sec- guished by unique “spectral fingerprints” of
ondary ocean microseisms are mostly recorded microseisms.
with periods between 5 and 8 s and are the result
of energetic ocean waves between the 10 and 16 s Long-Period Noise
period. Horizontal and vertical noise amplitudes Ground noise on land observed at frequencies
of marine microseisms are similar. The particle between 0.2 and 50 mHz, i.e., lower than the
motion is dominatingly of Rayleigh-wave type, microseism peak, is usually associated with
i.e., elliptical polarization of the particle motion atmospheric pressure fluctuations. At these fre-
in the vertical propagation plane. But some quencies, vertical component seismometers react
energy is present also as body waves, and hetero- to changes in gravity as the mass of the atmo-
geneous Earth structure may even couple some sphere above a site changes with atmospheric
energy into Love waves, i.e., surface waves that pressure. If seismometers are not fully sealed,
oscillate parallel to the Earth surface and perpen- they may also experience apparent accelerations
dicular to the direction of propagation with due to buoyancy effects on the seismometer
amplitudes that decay exponentially with depth mass. Pressure fluctuations may also cause strong
below the surface. longperiod horizontal component noise by
Medium-period ocean/sea microseisms expe- distorting the case of the seismometer and/or of
rience low attenuation. They may therefore prop- the walls of the vault.
agate hundreds of kilometers inland. Since they Long-period seismometers are also sensitive
are generated in relatively localized source areas, to temperature changes. Therefore, careful instal-
they have, when looked at from afar, despite the lations aim at sealing the sensor from both pres-
inherent randomness of the source process, a sure and temperature fluctuations (Forbriger
rather well-developed coherent part, at least in 2012). The vertical component noise levels
the most energetic and prominent component. between 0.2 and 1.7 mHz can be reduced by
Seismic Noise 2943
Seismic Noise, Fig. 5 An example of good spatial coher- backazimuth determination changed from one day to the
ence of medium-period secondary ocean microseisms at next, while (b) shows the location of the two storm areas
larger distance from the source area which, in this case, and of the GRF seismic array. Observations by at least two
allows rather reliable determination of the direction from arrays permit localization and tracking of the noise-
the station towards the source (backazimut angle = angle generating low-pressure areas (Copy of Fig. 7 from Frie-
measured clockwise from the north) by means of array drich et al. 1998 # Springer; with kind permission from
analysis (see ▶ Seismometer Arrays). (a) Shows how the Springer Science + Business Media)
more than 10 dB by subtracting the scaled, locally Bormann et al. 2012 and Fig. 4.26 in Bormann
recorded pressure signal from the vertical accel- and Wielandt 2013).
eration record, thereby improving the SNR for Turbulence in the atmospheric boundary layer
normal mode observations (e.g., Fig. 2.23 in produces pressure fluctuations at the Earth’s
2944 Seismic Noise
Seismic Noise, Fig. 6 Examples for coastline geome- from Friedrich et al. 1998 # Springer; with kind permis-
tries that provide suitable interference conditions for the sion from Springer Science+Business Media)
generation of secondary microseisms (Copy of Fig. 13
surface that cause significant deformation to Short-Period Noise (0.5–50 Hz)

depths of a few tens of meters below the surface. Ambient noise at frequencies higher than those
This noise source is a primary reason why broad- associated with the microseisms peak may have
band seismometers at permanent stations are natural causes such as wind (wind friction over
often installed beneath the Earth’s surface in rough terrain, trees, and other vegetation or built-
shallow boreholes (see Sect. 7.4.5 and Fig. 7.59 up objects swinging or vibrating in the wind),
in Trnkoczy et al. (2011)). For instruments on the rushing waters (waterfalls or rapids in rivers and
surface, short-wavelength atmospheric pressure creeks), etc. Wind-generated noise is broadband,
fluctuations produce a large tilt signal that ranging from about 0.5 Hz up to about 15–60 Hz.
makes the horizontal components particularly But the dominant sources of high-frequency
noisy. Tilt couples gravity into the horizontal noise are man-made (power plants, factories,
components but much less into the vertical com- rotating or hammering machinery, road and rail
ponent (Chap. 5, Sect. 5.3.3 and Figs. 5.8 and 5.9 traffic, etc.; see Sects. 7.1 and 7.2 in Trnkoczy
in Wielandt 2011). The ratio increases with the et al. 2011.). Wind-generated noise couples
period and may reach a factor of up to 300 (about mostly into surface-wave modes; cultural noise,
50 dB). Horizontal component local tilt noise however, couples at least partly into body waves
may also be caused by traffic and wind pressure. that can propagate also to greater depth. Cultural
At very long periods, deformation of the noise in records is in principle avoidable,
ground, of the seismometer vault, or of the build- although it is often impractical to place seismic
ing due to insolation gradients (uneven heating by station sites at sufficient distances away from
sunshine) may become noticeable as well. Gen- cities or highways. Some rules of thumb suggest
erally, recording sites poorly coupled to the that it may be necessary to site short-period
ground or on poorly consolidated ground will recording stations of high magnification as far
tend to be noisier, both at long and short periods. as 25 km from power plants or rock-crushing
Other reasons for increased long-period noise machinery, 15 km from railways, 6 km from
may be air circulation in the seismometer vault highways, and 1 km or more from smaller
itself or underneath the sensor cover. Special care roads. And with respect to moving waters, it is
in seismometer installation and shielding is there- recommended to place stations away from mov-
fore required in order to reduce drifts and long- ing water between some 60 km for very large
period environmental noise (Wielandt 2011; waterfalls and dams to 15 km for smaller rapids
Trnkoczy et al. 2011; Forbriger 2012). (Table 2.1 in Willmore 1979, http://www.seismo.
Seismic Noise 2945
Seismic Noise, Fig. 7 Comparison of the minimum and Seismic Network (GRSN) (Color version of Fig. 9 from
maximum levels of short-period and long-period ambient Bormann et al. 1997 # Springer; with kind permission
noise power observed at stations of the German Regional from Springer Science + Business Media) S
com/msop/msop79/sta/tab_2.1.gif or Table 7.5 in The particle motion of short-period noise is there-

Trnkoczy et al. 2011). However, noise travels fore more erratic than for long-period ocean noise
further across competent rock than through allu- and may vary more (by about 30 dB) than long-
vial filled valleys. Therefore, “safe” distances period noise (by about 10 dB) from station to
may be from 1/2 to 2/3 of the values above, station within a seismic network. Very pro-
depending on propagation path. nounced is the seasonal variability of the second-
Most of the short-period natural and ary ocean microseisms by about 30 dB (Fig. 7).
man-made sources are distributed, stationary, or Nevertheless, polarization analysis (the inves-
moving. Their contributions, coming from vari- tigation of the different oscillation directions of
ous directions, superpose to a rather complex, the various types of seismic waves in space),
more or less stationary random noise field. averaged over moving time-windows, sometimes
2946 Seismic Noise
reveals preferred azimuths of the main axis of

horizontal particle motion, which hints at local-
ized noise sources. Moreover, when the vertical
component particle motion in three-component
records is clearly developed and averaged, then
this indicates the presence of fundamental
Rayleigh-wave type polarization. A rather popu-
lar and cost-effective microzonation method is
based on this assumption. It derives information
about the fundamental resonant frequency of the
soft-soil cover and estimates local site amplifica-
tion of ground motion from the peak in the hori-
zontal to vertical component spectral noise ratio
(H/V method, e.g., Parolai 2012).
According to the above, every recording site
has different noise characteristics depending on
its distance from the ocean, industrial and settle-
ment areas, major infrastructure facilities, the
wind climate of the site, and the depth of burial
of the sensor. At some locations, a seasonal cycle Seismic Noise, Fig. 8 Displacement power spectral den-
sity calculated from 6 moving windows, 50% overlapping,
is evident due to variations in wind or water flow of short-period noise records, 4,096 samples long each,
in nearby rivers. For example, according to Fyen i.e., from a total record length of about 80 s at a rather quiet
(1990), spring runoff raises noise levels between site in NW Iran (Copy of Fig. 4.6 in Bormann (2002)
0.5 and 15 Hz at the NORESS array by up to # IASPEI)
15 dB. In settlement and industrial areas, a diur-
nal cycle may dominate due to variable human pressure fluctuations associated with wind can
activity during the day (Fig. 12). directly drive motions of the ground.
High-frequency noise spectra on land during Generally, noise levels increase with higher
quiet intervals resemble each other at quieter wind speeds. However, there is apparently no
sites. The rather featureless displacement power linear relationship between wind speed and the
spectrum follows a power-law dependence in amplitude of wind-generated noise. Wind noise
frequency, proportional to f2 (Fig. 8). This increases dramatically at wind speeds greater
spectral appearance changes drastically due to than 3–4 m/s and may reach down to several
the strong impact of cultural noise, especially in hundred meters depth below the surface at wind
the frequency band between 1 and 10 Hz (Figs. 13 speeds > 8 m/s (Young et al. 1996). But gener-
and 14). Wind noise depends on the strength of ally, the level and variability of wind noise is
the wind and the character of the site. Recorded much higher at or near the surface and is reduced
noise power near 1 Hz in a surface vault may significantly with depth (Fig. 9). Moreover, at
increase by 10–20 dB, as compared to records wind speeds below 3–4 m/s, the background
on calm days, when wind speed reaches about noise may be omnidirectional with relatively
5 m/s. Short-period wind noise becomes detect- large coherence length at frequencies below
able in records at surface sites far from cultural 15 Hz, but the coherence length is strongly
sources at wind velocities above about 3 m/s and reduced at higher wind speeds with increased air
at 4 m/s for subsurface sensors as well (Fig. 9). turbulence and local wind pressure fluctuations
The primary mechanism by which wind couples (Withers et al. 1996).
into highfrequency noise is probably by the direct Comparisons of noise spectra from surface
action of the wind on trees, bushes, and other vaults and subsurface installations in mines
structures, although at lower frequency the (Fig. 9) or boreholes (Fig. 10) have repeatedly
Seismic Noise 2947
Seismic Noise,
Fig. 9 Displacement
power noise spectra
measured at the surface
(upper curves) and at 420 m
below the surface in a
disbanded salt mine at
Morsleben, Germany
(lower curves), on a very
quiet day (hatched lines)
and on a day with light
wind on the surface (wind
speed about 4 m/s; full
lines) (Copy of Fig. 4.24 in
Bormann (2002) #
IASPEI)
Seismic Noise,
Fig. 10 Velocity power
density spectra as obtained
for noise records at the
surface (top) and at 300 m
depth in a borehole (below)
near Gorleben, Germany
(Figure by courtesy of
M. Henger; copy of
Fig. 4.23 in Bormann
(2002) # IASPEI)
shown significant improvements in SNR for sensor deployments at several hundred meters
high-frequency phases recorded at depths, some- depth.
times even as shallow as a few meters. Large In case of Fig. 10, the noise power at 300 m
reductions in noise level may occur already in depth was reduced, as compared to the surface, by
the first 5 m between the weathered rock of the about 10 dB, at f = 0.5 Hz, 20 dB at 1 Hz and
surface and the more competent rock in the bore- 35 dB at 10 Hz. The surface-wave nature of
hole. Note, however, that for periods longer than ambient noise (including ocean noise) is the rea-
2 s the noise reduction is only marginal, even for son for the exponential and frequency-dependent
2948 Seismic Noise
Seismic Noise, Fig. 11 Records of short-period noise due to the surface amplification effect which may reach a
(left) and signals (right) at the surface and at different factor of two for vertical incident waves but is less for
depth levels of a downhole seismometer array. Note the incidence angles >> 0 (Fig. copied from Bormann,
significantly larger Pn amplitude closer to the surface. It is 1966; modified from Broding et al. 1964)
decay of noise amplitudes with depth. This is not Cultural noise may vary strongly, up to about
the case for body waves (Fig. 11). Since the 30 or even 50 dB at some higher frequencies
penetration depth of surface waves increases between sites in different environments (Fig. 14)
with wavelength, high-frequency noise attenu- but also from day to day or with the time of the day
ates more rapidly with depth. But both noise at a given site in a busy urban environment (e.g.,
reduction and signal behavior with depth depend Fig. 12). Industrial and traffic noise levels are
also on local geological conditions (e.g., section usually lowest on weekends but increase during
“Installations in Subsurface Mines, Tunnels, and holidays because of higher traffic. Sources such as
Boreholes”). Because of the surface-wave char- power plants, transformer stations, etc. generate
acter of short- and medium-period noise, the hor- narrowband noise, producing energetic spectral
izontal propagation velocity of ambient lines at 50 or 60 Hz as well as related harmonics
microseismic noise depends on frequency. It is and subharmonics (e.g., at 30 Hz, 25 Hz, 12.5 Hz)
close to the shear-wave velocity in the uppermost depending on location (Fig. 13). The NORESS
crustal layers, which is about 2.5–3.5 km/s for array data included lines at 6, 12, and 17 Hz due
outcropping hard rock and about 300–650 m/s for to reciprocating saws in a nearby sawmill, and
unconsolidated sedimentary cover. This is rather records at Moxa (MOX) station in Germany have
different from the apparent horizontal propaga- been spoiled over decades by a pronounced 2 Hz
tion velocity of P waves and all other steeply peak due to a heavy steam engine 16 km away
emerging teleseismic body-wave onsets. driving a steel-press for rails. It may be possible to
Broadband spectral levels fall also with remove by frequency filtering such narrowband
increasing borehole depth, but narrow spectral noise at frequencies that differ from the dominat-
lines due to cultural sources and or waveguide ing signal frequency (see section “Frequency Fil-
trapping may show less depth dependence. This tering”). Other types of machinery, however, may
may result in very different noise decay and also produce narrowband peaks in the noise spec-
signal-to-noise ratio behavior with depth (see tra, but with time-variable frequencies as motor
section “Installations in Subsurface Mines, Tun- speed may vary. These noise sources are more
nels, and Boreholes”). difficult to remove.
Seismic Noise 2949
Seismic Noise, Fig. 12 Comparison of relatively quiet Iran. The measurements were made at different times of
sections of vertical component noise records (left, without the day (Copy of Fig. 7.15 from Bormann in Trnkoczy
strong transients) and related power spectra (right) at a et al. (2002) # IASPEI)
reference site in the district capital town of Miane in NW
Microseismic Noise in Urban Environments industrialized city such as Hamburg, the micro-
In urban, densely populated, and industrialized seismic noise level may be 20–40 dB higher for
environment both natural and cultural noise com- frequencies of about 0.5–30 Hz and may be
ponents may play a major role in a wide range of reduced during night time by less than 10 dB
frequencies. Measuring and analyzing urban seis- only between 1 Hz and 10 Hz. In the metropolitan
mic noise (USN) requires broadband recordings city of Berlin, the “cultural noise” patterns differ
and data analysis. Figure 14 compares the noise from that of HAM (see Fig. 7.31 in Trnkoczy S
power density spectrum measured at the former et al. 2011) and may again be different in towns
seismological station in the city of Hamburg, with more pronounced diurnal variability of
Germany, with the spectrum measured on a hard human and industrial activities than in mega-
rock site in a quiet Spa village some 70 km away, cities, as in Fig. 12.
which is now the alternative station site in the Groos and Ritter (2009) stated that the gener-
GRSN. ally high variability of USN does not allow to
From Fig. 14, the following conclusions can characterize a sample time series of USN com-
be drawn: While in a remote quiet urban setting prehensively by single measures such as the
the microseismic noise spectrum may decay at standard deviation of a noise series or the PSD
night almost monotonously towards higher fre- at a given frequency. For the metropolitan
quencies, the cultural noise activity may increase city of Bucharest, Romania, they calculated
the daytime PSD for frequencies above 1 Hz by therefore long-term spectrograms of up to
some 10 dB. In contrast, in a very busy 28 days duration from broadband seismic
2950 Seismic Noise
Seismic Noise, Fig. 13 Noise records and related power and its lower harmonics and/or caused by the vibration of
spectra near to a transformer house and power line. Note the transformer (Copy of Fig. 7.19 from Bormann in
the monochromatic spectral lines around 13, 30, and Trnkoczy et al. (2002) # IASPEI)
50–60 Hz, either induced by the AC current frequency
recordings. Spectrograms, or sonograms, are information about ambient noise for selecting suit-
visual 2D representations of the spectrum of fre- able time windows for H/V noise studies (Parolai
quencies contained in a seismic, sound, or other 2012) or ambient noise tomography.
signal as they vary with time or with some other
variable. The aim was to identify the frequency-
dependent behavior of the time-variable Signal and SNR Variations due to Local
processes that contribute to USN. Based on the Site Conditions
spectral analysis of the data in eight frequency
ranges between 8 mHz and 45 Hz, they proposed Compared to hard rock sites, both noise and sig-
a time-domain classification which allows to iden- nals may be amplified when recorded on soft soil
tify both Gaussian distributed noise time series as cover. This signal amplification may partly or
well as time series that are dominated by transient even fully outweigh the higher noise observed
or periodic signals due to traffic, rotating machin- on such sites. Signal strength observed for a
ery, etc. In the case of Bucharest, only some 40 % given event may vary strongly (up to a factor of
of the analyzed time series are characterized by about 10–30) within a given array or station net-
Gaussian distributed random noise. The most work, even if its aperture is much smaller than the
common deviations are due to large-amplitude epicentral distance to the event (<10–20%), so
transient signals. Moreover, significant variations that differences in backazimuth and amplitude-
of the statistical noise properties during daytime distance relationship are negligible (Figure 15
were found in the wide frequency range between and Fig. 4.36 in Bormann and Wielandt 2013).
0.04 and 45 Hz, pointing to a broadband human Also, while one station of a network may record
influence on USN. The authors recommend the events rather weakly from a certain source area,
use of such automatically derived broadband the same station may do as well as other stations
Seismic Noise 2951
Seismic Noise, Fig. 14 Noise power spectra at HAM salt-anhydrite diapir with very high contrast of its acoustic
(upper two curves, red) and BSEG (lower two curves, impedance (= product of wave velocity in the medium
blue) determined from vertical-component records at day- times density of the medium) against that of the surround-
time and nighttime. Note that the noise reduction is almost ing alluvial softrock layers. This prevents much of the
negligible for periods longer than 2 s. The drastic reduc- short-wavelength noise energy to penetrate into this
tion at higher frequencies has been achieved both by the local hard rock anomaly (Color version of Fig. 18 from
reduced cultural noise sources at Bad Segeberg, as com- Bormann et al. 1997 # Springer; with kind permission
pared to the big city of Hamburg, and by installation of the from Springer Science + Business Media)
broadband seismograph in an outcropping local hardrock
(or even better) for events from another region, tilt noise then have the highest priority (see
azimuth, or distance (e.g., station GWS in Fig. 16 Forbriger 2012).
left and right, respectively).
Figure 16 compares, for regional and Installations in Subsurface Mines, Tunnels,
teleseismic events, the short-period P-wave and Boreholes
amplitude ratio (left) and SNR (right) of two As shown in Figs 13–15, short-period microseis-
stations of the German Regional Seismic Net- mic noise is strongly reduced with the depth of
work (GRSN). In the same azimuth range, but at sensor installations in mines and boreholes.
different epicentral distances, station BRG may Withers et al. (1996) found that for rather high S
record both > 3 times larger as well as > 3 times frequencies monitoring, as in very local weak
smaller amplitudes than station MOX. This cor- motion and industrial seismology measurements
responds to differences in station magnitude up to between 10 and 20 Hz, the SNR could be
one unit! The same applies to the differences in improved between 10 and 20 dB and for frequen-
SNR. Therefore, optimal site selection cannot be cies between 23 and 55 Hz as much as 20–40 dB
made only on the basis of noise measurements. by deploying a short-period sensor at only 43 m
Also, the signal conditions at possible alternative below the surface. However, depending on geo-
sites should be compared. logical conditions and related peculiarities of the
However, differences in local SNR conditions local noise field as well as the decreasing free-
may become negligible in long-period recordings surface amplification effect of incident body
(Fig. 4.39 in Bormann and Wielandt 2013) and waves, the SNR does not necessarily increase
thus play a lesser role in site selection for broad- steadily with depth. A striking example have
band records. Thermal shielding and reduction of been short-period signal and noise measurements
2952 Seismic Noise
Seismic Noise, Fig. 15 Short-period records of under- the differences in signal amplitudes both amongst the sta-
ground nuclear explosions at the test sites of Semipalatinsk tions for a given event and for the same station, when
(left, D 41 1 ) and Nevada (right, D 81 1 ) at comparing events in different backazimuth and distance
stations of the former East German seismic network. Note (Copy of Fig. 4.35 in Bormann (2002) # IASPEI)
Seismic Noise, Fig. 16 Pattern of the relative short- (170 km apart) in a distance-azimuth polar diagram
period P-wave amplitudes (left) and of the related SNR (Modified version of Fig. 7 from Bormann et al. 1992
(right) at station BRG normalized to those of station MOX # Elsevier; with permission from Elsevier)
Seismic Noise 2953
in the early 1960s with 1 Hz borehole seismome- period noise with periods between 30 s and
ters in two abounded deep oil drills in the USA 1,000 s (see Fig. 7.59 in Sect. 7.4.5 of Trnkoczy
(Douze, 1964). In the Texas hole, the average ratio et al. 2011). Since the cost of drilling and instal-
of seismic signal in the borehole to that measured lation increases greatly with depth, no permanent
at the surface decreased to 1/10th at about 1,500 m seismic borehole installations deeper than
depth and then increased again to ½ of the surface 100–150 m have yet been made. In any event,
value at 3,000 m depth. In contrast, in an Okla- the borehole should be drilled through the soil or
homa borehole, it dropped to about 1/3 at about cover of weathered rock and penetrate well into
1,000 m depth and then remained roughly constant the compacted underlying rock formations. For
thereafter. Accordingly, despite the noise reduc- more information about tunnel and long-period
tion with depth, there was no SNR improvement borehole installations, see Sects. 7.4.3 and 7.3.5
(on average) in the Texas borehole down to about in Trnkoczy et al. 2011.
1,000 m depth, but then the SNR increased to a Thus, differences in the frequency spectrum,
factor of about 15 at 3,000 m depth. Contrary to horizontal wave-propagation velocity, degree of
this, the SNR increased by a factor of 3 in the coherence, and depth dependence between
Oklahoma borehole within the first 800 m but (short- and medium-period) microseismic noise
then remained roughly constant (ranging between and seismic waves allows to improve the signal-
1 and 5) up to 3,000 m depth (see Fig. 4.40 in to-noise-ratio (SNR) by installing seismic sen-
Bormann and Wielandt 2013). sors either at reasonable depth below the surface
Nevertheless, generally a significant SNR or by way of data processing.
improvement is to be expected within the first
few hundred meters depth. This applies particu-
larly to borehole installations of long-period and Improving the Signal-to-Noise Ratio by
broadband sensors which benefit already at much Data Processing
shallower depth greatly from the very stable tem-
perature conditions and strongly reduced atmo- Frequency Filtering
spheric pressure fluctuations and related tilt If the dominating frequency content of the seis-
noise, also in hard rock tunnels. In mines or mic signal differs from that of the disturbing
boreholes, a depth of 100 m below surface is noise, then band-pass, high-pass, or low-pass fil-
generally sufficient to achieve most of the prac- tering may significantly improve the SNR
ticable reduction by 20 to 30 dB of long- (Figs. 17 and 18).
Seismic Noise, Fig. 17 Top: Velocity broadband record noise ratio is much improved and several later phases are
in the period range between 0.5 and 360 s of a teleseismic clearly recognizable (Figure by courtesy of J. Havskov for
earthquake. Only the P-phase might be identified. Bottom: NMSOP editions # IASPEI)
On the long-period band-pass filtered trace the signal-to-
2954 Seismic Noise
ML 2.2 earthquake at 32km distance

STS–2 BFO HHZ; raw data
0
ground velocity / μm s−1
−2
−4
16h20m 22m 24m 26m 28m 30m 32m
time after 19.01.2007 00:00:00 UT
ML 2.2 earthquake at 32km distance
STS–2 BFO HHZ; 4–pole Butterworth high–pass at 2. Hz
2
ground velocity / μm s−1
−1
16h26m10s 20s 30s 40s 50s 27m0

time after 19.01.2007 00:00:00 UT
Seismic Noise, Fig. 18 Isolation of a local ML = 2.2 STS-2 record (top trace), thus eliminating the dominating
earthquake (bottom trace) that occurred 32 km away from 6 s ocean microseisms (Figure by courtesy of Thomas
the Black Forest Observatory (BFO), Germany, by 2-Hz Forbriger for NMSOP Chapter 4 # IASPEI)
high-pass filtering of the raw data of a velocity broadband
Velocity Filtering and Beamforming velocity of noise, being dominatingly surface

Often the dominant signal frequencies may coin- waves of Rayleigh-wave type, is much lower
cide with that of strong noise. Then frequency than that of P waves and also lower than that
filtering does not help to improve the SNR. On of teleseismic S waves with a steep angle of
the other hand, the horizontal propagation incidence. Then frequency-wavenumber (f-k)
Seismic Noise 2955
trace. A weak underground nuclear explosion at

a distance of 143.6 , not recognizable in any of
the single traces, is very evident in the beam
trace.
Noise Prediction Error, Polarization, and

Other Multi-parameter Methods of Filtering
In near real time, it is possible to use a moving
time-window to determine the characteristics of a
given noise field by means of cross- and autocor-
relation of array sensor outputs. This allows the
prediction of the expected random noise in a
subsequent time interval. Subtracting the
predicted noise time series from the actual record
results in a much reduced noise level. Weak seis-
mic signals, originally buried in the noise but not
predicted by the noise “forecast” of the
prediction-error filter (NPEF), may then stand
out clearly. NPEFs have several advantages as
compared to frequency filtering:
• No assumptions on the frequency spectrum of

noise are required since actual noise properties
are determined by the correlation of array
Seismic Noise, Fig. 19 Detection of a weak under-
ground nuclear explosion in the 10 kt range at the Mururoa
sensor outputs.
Atoll test site (D = 145 ) by beam forming (top trace). No • While frequency differences between signal
signal is recognizable in any of the 13 individual record and noise are lost in narrowband filtering,
traces from stations of the Gräfenberg array, Germany they are largely preserved in the case of the
(below) (From Buttkus 1986 # Schweizerbart http://
www.schweizerbart.de/)
NPEF. This may aid signal identification and
onset-time picking.
• Signal first-motion polarity is preserved in the
filtering is suitable to improve the SNR in the NPEF whereas it is no longer certain after
beam trace of seismic array records (Schweitzer narrow-band or zero-phase band-pass
et al. 2011). Assuming that the noise within the filtering.
array is random while the signal is coherent, even S
a simple direct summation of the n sensor outputs Three-component recordings allow to recon-
would already produce some modest SNR struct the ground particle motion and to deter-
improvement. When the direction and velocity mine its polarization which may differ between
of travel of a signal through an array is known, different types of seismic waves and noise. One
it is possible to compensate for the differences in great advantage of polarization filtering is that it
arrival time at the individual sensors and then is independent of differences in the frequency and
sum up all the n record traces (beam forming). velocity spectrum of signal and noise and thus
This increases the signal amplitude by a factor n can be applied in concert with other procedures
while the random noise amplitudes increase in for SNR improvement.
the beam trace only by √n, thus improving the Early examples of NPEF and polarization fil-
SNR by √n. Figure 19 compares the (normalized) tering have been given in Bormann and Wielandt
individual records of 13 stations of the (2013). For a briefing on modern three-
Gräfenberg array, Germany, with the beam component broadband procedures of SNR
2956 Seismic Noise
improvement, signal detection, phase identifica- ▶ Seismometer Arrays

tion, and event location that have been ▶ Sensors, Calibration of
implemented in procedures for fully automatic ▶ Site Response: 1-D Time Domain Analyses
near real-time multi-parameter signal analysis, ▶ Site Response: Comparison Between Theory
see Sect. 11.9 in Bormann et al. (2014). These and Observation
procedures apply cross-correlation methods, use ▶ Stochastic Ground Motion Simulation
differences in the complexity and polarization
properties between signal and noise as well as a
compressed signal representation by a sequence
References
of n-dimensional feature vectors which increase
the robustness of the procedures. Some of these Aki K, Richards PG (1980) Quantitative
versions also adapt the signal detector to tempo- seismology – theory and methods, vol II. Freeman,
ral/seasonal variations in the nature and power of San Francisco
Bormann P (1998) Conversion and comparability of data
ambient noise.
presentations on seismic background noise. J Seismol
2(1):37–45
Bormann P (2002) Chapter 4: Seismic signals and noise.
Summary In: Bormann P (ed) IASPEI new manual of seismolog-
ical observatory practice (NMSOP),
vol 1. GeoForschungsZentrum, Potsdam, 34 pp
The essay defines the term of “seismic noise” and Bormann P, Wielandt E (2013) Chapter 4: Seismic signals
explains why coherent seismic signals and inco- and noise. In: Bormann P (ed) New manual of seismo-
herent random noise are not commensurate, thus logical observatory practice (NMSOP-2) (2012).
IASPEI, GFZ German Research Centre for
requiring different mathematical treatment and
Geosciences, Potsdam, 63 pp. doi:10.2312/GFZ.
analysis. Also described are the different causes NMSOP-2_CH4. http://nmsop.gfz-potsdam.de
and ways of analysis of ambient microseismic Bormann P, Wylegalla K, Strauch W, Baumbach
noise as recorded at stations on land in the very M (1992) Potsdam seismological station network:
processing facilities, noise conditions, detection
wide frequency range between some mHz and
threshold and localization accuracy. Phys Earth Planet
100 Hz. The article sketches some essential pro- Int 69:311–321
cedures of noise reduction by way of proper sen- Bormann P, Wylegalla K, Klinge K (1997) Analysis of
sor shielding and installation, also at depth, as broadband seismic noise at the German Regional Seis-
mic Network and search for improved alternative sta-
well as feature and problem adapted filtering
tion sites. J Seismol 1:357–381
aimed at improved signal-to-noise ratio and thus Bormann P, Engdahl ER, Kind R (2012) Chapter 2: Seis-
signal detection and identification. Many refer- mic waves and Earth models. In: Bormann P (ed) New
ences for further readings are given, amongst manual of seismological observatory practice
(NMSOP-2) (2012). IASPEI, GFZ German Research
them to relevant IASPEI Manual articles acces-
Centre for Geosciences, Potsdam, 105 pp.
sible via http://nmsop.gfz-potsdam.de. doi:10.2312/GFZ.NMSOP-2_CH2. http://nmsop.gfz-
potsdam.de
Bormann P, Klinge K, Wendt S (2014) Chapter 11: Data
analysis and seismogram interpretation. In: Bormann
Cross-References P (ed) New manual of seismological observatory prac-
tice (NMSOP-2) (2012). IASPEI, GFZ German
▶ Downhole Seismometers Research Centre for Geosciences, Potsdam,
▶ Noise-Based Seismic Imaging and Monitoring doi:10.2312/GFZ.NMSOP-2_CH11. http://nmsop.
gfz-potsdam.de
of Volcanoes
Broding RA, Bentley-Llewellyn NJ, Hearn DP (1964)
▶ Ocean-Bottom Seismometer A study of three dimensional seismic detection system.
▶ Seismic Event Detection Geophysics 29(2):221–249
▶ Seismic Instrument Response, Correction for Buttkus B (ed) (1986) Ten years of the Gräfenberg array:
defining the frontiers of broadband seismology.
▶ Seismic Network and Data Quality
Geologisches Jahrbuch E-35, Hannover, 135 pp
▶ Seismometer Self-Noise and Measuring Cessaro RK (1994) Sources of primary and secondary
Methods microseisms. Bull Seismol Soc Am 84(1):142–148
Seismic Reliability Assessment, Alternative Methods for 2957
Douze EJ (1964) Signal and noise in deep wells. Geophys- Webb SC (2002) Seismic noise on land and on the sea
ics 29(5):721–732 floor. In: Lee WHK, Kanamori H, Jennings PC,
Forbriger Th (2012) IS 5.4: Recommendations for seis- Kisslinger C (eds) International handbook of earth-
mometer deployment and shielding. In: Bormann quake and engineering seismology, Part A. Academic,
P (ed) New manual of seismological observatory prac- Amsterdam, pp 305–318
tice (NMSOP-2) (2012). IASPEI, GFZ German Wiechert E (1904) Verhandlungen der zweiten
Research Centre for Geosciences, Potsdam, 10 pp. internationalen Seismologischen Konferenz. Beitr
doi:10.2312/GFZ.NMSOP-2_IS_5.4. http://nmsop. Geophys, Ergänzungsband II, pp 41–43
gfz-potsdam.de Wielandt E (2011) Chapter 5: Seismic sensors and their
Friedrich A, Klinge K, Kr€ uger F (1998) Ocean-generated calibration. In: Bormann P (ed) New manual of seis-
microseismic noise located with the Gräfenberg array. mological observatory practice (NMSOP-2) (2012).
J Seismol 2:47–64 IASPEI, GFZ German Research Centre for
Fyen J (1990) Diurnal and seasonal variations in the Geosciences, Potsdam, 51 pp. doi:10.2312/GFZ.
microseismic noise observed at the NORESS array. NMSOP-2_CH5. http://nmsop.gfz-potsdam.de
Phys Earth Planet In 63:252–268 Withers MM, Aster RC, Young CJ, Chael EP
Groos JC, Ritter JRR (2009) Time domain classification (1996) High-frequency analysis of seismic back-
and quantification of seismic noise in an urban envi- ground noise as a function of wind speed and shallow
ronment. Geophys J Int 179:1213–1231 depth. Bull Seismol Soc Am 86(5):1507–1515
Hasselmann KA (1963) A statistical analysis of the gen- Young CJ, Chael EP, Withers MW, Aster RC (1996)
eration of microseisms. Rev Geophys 1:177–209 A comparison of the high- frequency (>1Hz) surface
Haubrich RA, Munk WH, Snodgrass FE (1963) Compara- and subsurface noise environment at three sites in the
tive spectra of microseisms and swell. Bull Seismol United States. Bull Seismol Soc Am 86(5):1516–1528
Soc Am 53:27–37
Longuet-Higgins MS (1950) A theory for the generation
of microseisms. Phil Trans R Soc Lond A 243:1–35
Parolai St (2012) Chapter 14: Investigation of site
response in urban areas by using earthquake data and
seismic noise. In: Bormann P (ed) New manual of Seismic Reliability Assessment,
seismological observatory practice (NMSOP-2) Alternative Methods for
(2012). IASPEI, GFZ German Research Centre for
Geosciences, Potsdam, 38 pp. doi:10.2312/GFZ.
Hossein Ebrahimian and Raffaele De Risi
NMSOP-2_CH14. http://nmsop.gfz-potsdam.de
Peterson J (1993) Observations and modeling of seismic Department of Structures for Engineering and
background noise. U.S. Geological Survey Open-File Architecture, University of Naples Federico II,
Report, 93–322, 95 pp Naples, Italy
Schweitzer J, Fyen J, Mykkeltveit S, Gibbons SJ, Pirli M,
K€uhn D, Kværna T (2011) Chapter 9: Seismic arrays.
In: Bormann P (ed) (2012) New manual of seismolog-
ical observatory practice (NMSOP-2) (2012). IASPEI, Synonyms
GFZ German Research Centre for Geosciences, Pots-
dam, 80 pp. http://nmsop.gfz-potsdam.de
Bayesian method; Fragility curve; Limit state;
Shinohara M, Suyehiro K, Shiobara H (2011) Chapter 7,
section 7.5: Marine seismic observation. In: Mean annual frequency of exceedance; Monte
Bormann P (ed) (2012) pp 112–135. http://nmsop. Carlo simulation; PEER performance-based S
gfz-potsdam.de approach; Reliability assessment; Second-
Trnkoczy A, Bormann P, Hanka W, Holcomb LG, Nigbor
moment reliability methods; Simulation-based
RL (2002) Chapter 7: Site selection, preparation and
installation of seismic stations. In: Bormann P reliability; System reliability
(ed) IASPEI New Manual of Seismological Observa-
tory Practice (NMSOP), vol 1. GeoForschungsZentrum
Potsdam, Potsdam, 108 pp
Trnkoczy A, Bormann P, Hanka W, Holcomb LG, Nigbor
A Perspective into the Seismic
RL, Shinohara M, Suyehiro K, Shiobara H (2011) Reliability Assessment
Chapter 7: Site selection, preparation and installation
of seismic stations. In: Bormann P (ed) New manual of The treatment of structural behavior under seis-
seismological observatory practice (NMSOP-2)
mic actions can be considered as one of the most
(2012). IASPEI, GFZ German Research Centre for
Geosciences, Potsdam. doi:10.2312/GFZ.NMSOP- challenging aspects of structural performance
2_CH7. http://nmsop.gfz-potsdam.de assessment and design. It encompasses numerous
2958 Seismic Reliability Assessment, Alternative Methods for
sources of uncertainty associated with the seismic structures under seismic actions. This issue is
action, structural response, and structural capac- addressed by means of defining a prescribed set
ities. The various parameters involved in the seis- of limit states, which describe specific stages of
mic performance assessment problem can be structural behavior and its consequences (e.g.,
generically designated in terms of seismic states associated with consequences in terms of
demand (D) and capacity (C) of the structure. costs, loss of lives, impact on the environment, or
The seismic demand and capacity can be charac- serviceability; see Bozorgnia and Bertero 2004).
terized as a functional (i.e., function of function) Reaching a limit state denotes that an unaccept-
of a set of uncertain parameters (also denoted as able behavior for the structure occurs. This con-
“random variables”). Hence, probabilistic dition is generically referred to as failure. The
methods are necessary in order to take into limit state exceedance is verified by comparing
account the uncertainties in demand and capacity a measurable quantity, associated with the
within the performance assessment or design. structural response, denoted as the demand D
Traditionally, seismic-design codes have (also referred to as the engineering demand
addressed the uncertainties by allowing some parameter (EDP), e.g., maximum inter-story
degree of conservatism in evaluating demand drift ratio, maximum chord rotation, maximum
and capacity at the level of structural compo- component shear, etc.), with the corresponding
nents; nevertheless, the link to the overall perfor- limit value, denoted as the capacity C (e.g.,
mance of the structure remains unclear. Strictly yielding rotation, ultimate rotation, shear
speaking, the reliability of the structure to with- strength, etc.). As a result, the inequality D
stand future events remains more-or-less C, also called safety margin formulation,
unknown to the designer while employing vari- defines acceptable performance respect to
ous established code-based approaches. There- the current limit state in mathematical terms
fore, in the past decades, several research efforts (Pinto et al. 2007). This formulation is gener-
have been carried out and substantial progress has ally expressed as limit state function
been made towards the inclusion of various (or performance function) G = C-D. This type
sources of uncertainty into structural perfor- of formulation is also employed in the load and
mance assessment and design frameworks. resistance factor design (LRFD) format
This entry introduces the main sources of (Galambos et al. 1982), which is familiar to
uncertainty in demand and capacity, as well as engineers as the safety checking equation.
their treatments through alternative reliability As an alternative to the safety margin formula-
methods. In this regard, three general families tion, the safety factor formulation is introduced
of methods are briefly summarized in this entry: as G = ln(C/D) (DNV 1992). For both the safety
the second-moment reliability methods, the margin formulation and the safety factor formu-
simulation-based reliability approaches, and lation, failure is defined as G 0.
finally the probabilistic performance-based meth- As mentioned before, the evaluation of the
odology developed within the joint efforts of structural performance in terms of demand and
Pacific Earthquake Engineering Research Center capacity may encompass several sources of
(PEER). The closing section provides a brief uncertainty. In such context, the performance
discussion about future challenges in seismic reli- with respect to a given limit state is evaluated
ability assessment. in terms of the probability that the limit state
function will be less than zero (probability of
failure PF) in a specified reference time
An Introduction to Seismic Reliability period. Identifying X = [X1,X2,. . .,Xn] to be
Assessment the group of n uncertain parameters involved
in the reliability problem, the probability of
Seismic structural reliability is concerned with failure may be determined by the following
the probabilistic performance evaluation of integral:
ð
methods provide the following classification of
PF ¼ P½GðXÞ 0 ¼ f X ðXÞdx (1)
GðXÞ0 uncertainties in seismic structural engineering:
(1) inherent randomness due to the intrinsic var-
where fX is the joint probability distribution for iability in material mechanical properties as well
uncertain parameters X. This means that one can as environmental actions (i.e., loads or seismic
evaluate the probability of failure by integrating excitation); (2) statistical uncertainties due to the
the joint density function over the failure domain lack of knowledge associated with the estimation
defined as G(X) 0. of the parameters of probability distributions
Conventionally, the generalized reliability from observational samples of limited size; and
(or safety) index b associated with a given limit finally (3) model uncertainties due to the imper-
state function is defined by one-to-one mapping fections of the mathematical models used to
relationship with failure probability as follow: describe complex physical phenomena (i.e.,
model describing load or element capacity) (see
also Der Kiureghian 1996; Ditlevsen and Madsen
b ¼ F1 ðPF Þ (2) 1996). Some other reliability methods (mostly
simulation based) classify the sources of uncer-
where F(∙) is the standard normal Cumulative tainty into two general categories: those related to
Distribution Function (CDF). Equation 2, how- the representation of the seismic action and those
ever, does not mean to imply that uncertain related to the structural modeling (see, e.g.,
parameters are jointly normal. The inverse of Papadimitriou et al. 2001). Another useful clas-
the standard normal CDF simply provides a con- sification is based on the definition of the aleatory
venient one-to-one mapping between the com- and epistemic uncertainties (Cornell et al. 2002;
puted probability of failure and a reliability Der Kiureghian and Ditlevsen 2009). These two
index. categories are also known as inherent random-
The evaluation of the integral in Eq. 1 can be ness (i.e., uncertainty impossible to mitigate) and
computationally difficult; some examples are as statistical uncertainties (i.e., uncertainties that
follows: fX is often not well-defined because of can be mitigated with reasonable effort), respec-
the incompleteness of the statistical information tively (McGuire 2004). On the other hand, in the
available; G(X) may have a nonlinear form; the Bayesian probabilistic framework, the various
computation of the multifold integral can be very sources of uncertainty are treated in a unified
difficult if the number of uncertain parameters is manner. In subsequent sections, the uncertainties
high. Various methods have been proposed for in the seismic reliability problems are divided
solving the integral form in Eq. 1. These (if required) into those related to the representa-
approaches range from the classical moment tion of the seismic action, also referred to as the
methods for structural reliability (e.g., first- record-to-record (RTR) variability, and other S
order second-moment reliability method) to the sources of uncertainty. As mentioned above, the
simulation-based approaches (i.e., Monte Carlo set of uncertain parameters involved in the seis-
family of methods), and also the PEER approach, mic performance assessment or design problem
which is quite different compared to the other two are denoted hereafter by the vector of
techniques. In this entry, alternative methods for parameters X.
estimating the probability of failure are
described. Structural Component and Structural
System Reliability
Types of Uncertainty A structural component is a basic structural ele-
There are alternative ways for classifying the ment whose performance is defined by a single
sources of uncertainty. Arguably, such classifica- limit state function G(X) = 0 (Fig. 1a). In the
tion depends to some extent on the reliability hyperspace of the designated uncertain parame-
method adopted. Some classical reliability ters X, the limit state function represents the
a b c
x2 x2 x2
_0
G2(X)< _0
G2(X)<
_0
G1(X)< _0
_0
G1(X)< G1(X)<
1 1 2 1
2
x1 x1 x1
Seismic Reliability Assessment, Alternative Methods (b) series system, (c) parallel system. The gray shaded
for, Fig. 1 Reliability problems for the special case of area represents the side of the failure domain
only two uncertain parameters x1 and x2: (a) Component,
boundary (i.e., limit state surface) between safe x2 _0

G3(X)<
and failure zones (the latter is also called the
failure domain), as shown in Fig. 1a for the spe- _0
G1(X)<
cial case of only two uncertain variables
(Ditlevsen and Madsen 1996).
A structural system is an assemblage of com-
ponents, whose performance is described by indi-
vidual component limit state function. Therefore, 1 3
in a structural system reliability problem, the
_0
G2(X)<
failure domain may be described as the union 2
and/or intersection of several limit state surfaces.
In particular, a series system (also known as the x1
weakest-link) reliability problem is defined as the
union of the failure domains (Fig. 1b), and a Seismic Reliability Assessment, Alternative Methods
parallel system reliability problem is defined as for, Fig. 2 Reliability problems for the special case of
only two uncertain parameters x1 and x2: cut-set system.
the intersection of the failure domains (Fig. 1c). The gray area represents the side of the failure domain
In analytic terms, the probability of failure for
series and parallel systems with n components is
N nj
defined, respectively, as PF, cutset ¼ P [ \ ðGi ðXÞ 0Þ (4)
j¼1 i¼1

n
PF, series ¼ P [ ðGi ðXÞ 0Þ PF, parallel where N defines the associated number of the
i¼1

subsystem that is connected in series to the
n other subsystems and nj is the number of compo-
¼ P \ ðGi ðXÞ 0Þ (3)
i¼1 nents connected in parallel belonging to the jth
subsystem.
A cut-set is made as a combination of series and
parallel systems generating a generalized system Seismic Reliability Assessment, a Time
reliability problem. A cut-set (Fig. 2) is any set of Variant Nonlinear Problem
components whose joint failure represents a fail- If the limit state function reveals time-dependent
ure of the system (Ditlevsen and Madsen 1996; properties, then the reliability problem can be
Au and Beck 2003b; Jalayer et al. 2007a). Ana- referred to as time variant. Hence, given the sto-
lytically, the probability of failure for a cut-set chastic nature of the seismic excitation, the seis-
system is evaluated as mic reliability assessment can be classified as a
time-variant problem. Therefore, the failure moments of G(X) in terms of the first two
event constitutes the outcrossing of the time- (central) moments of the vector of uncertain
variant limit state surface. This represents the parameters X:
so-called first-passage (or first-excursion) prob-
lem (e.g., Au and Beck 2001b). Moreover, in mG ¼ ½GðXÞ ¼ a0 þ aT ½X ¼ a0 þ aT mX
many practical cases, the limit state function is h i
nonlinear. Therefore, the seismic reliability prob- s2G ¼  ðGðXÞ mG Þ2
h i
lems can be classified, in general, as time-variant ¼ aT  ðX mX ÞðX mX ÞT a 5 aT SX a
nonlinear problems.
(6)
The Second-Moment Reliability One of the simplest forms of reliability measure

Methods was proposed by Cornell (1969), in which the
second-moment reliability index is defined as
In general, the nth moment of a mono-variate the ratio between the mean value and the standard
distribution (i.e., a distribution with a single sca- deviation of the performance function G (i.e., the
lar random variable) is defined as the expected so-called “Cornell index”):
value of the variable raised to the power of n.
mG
Using moments of finite order for describing the b¼ (7)
sG
uncertainty in given parameter is a well-known
problem in statistics referred to as the moment
Assuming that the unacceptable performance is
methods. A comprehensive review of the moment
reached when G 0, it is possible to establish a
methods in the structural reliability analysis is
one-to-one mapping between the Cornell index
presented in Zhao and Ono (2001). The second-
and the probability of failure, as already shown in
moment reliability methods can be dated back to
Eq. 2:
the 1960s (some historical notes can be found in
Ditlevsen and Madsen 1996). Since then, this
family of methods has been refined and extended PF ¼ P½GðXÞ 0
significantly. At present, they represent one of the
GðXÞ mG 0 mG
most important and widespread approaches for ¼P
sG sG
seismic reliability evaluation of structures. As
m
reflected by their title, the second-moment reli- ¼ F G ¼ FðbÞ (8)
ability methods employ the first two moments of sG
the vector of uncertain parameters X (i.e.,
expected value [X] and covariance SX). This mapping is exact if the variables are normal. S
Otherwise, Eq. 8 returns an approximate value of
Linear (First-Order) Limit State Function and the failure probability.
Its First Second Moments
The simplest mathematical form for a perfor- First-Order Second-Moment (FOSM) Method
mance function is the linear (first-order) in Component Reliability
polynomial: In the case of a nonlinear limit state surface, the
reliability index fails to be constant under differ-
ent but mechanically equivalent formulations of
GðXÞ ¼ a0 þ a1 X1 þ . . . þ an Xn ¼ a0 þ aT X
the performance function. The issue of a lack of
(5) invariance was first recognized by Ditlevsen
(1973) and later resolved by Hasofer and Lind
where the coefficients a0 and a = [a1,. . .,an] are (1974). In order to overcome this issue, they
constants. It is possible to write the two central proposed the transformation of the uncertain
Seismic Reliability a b
Assessment, Alternative FORM
x2 _0
G(X)< u2 approximation
Methods for, Fig. 3 The α
limit state function and the _0
g(U)<
joint-normal PDF of the G(X)>0 U0
uncertain parameters in (a) g(U)>0
nonstandard space of β
correlated uncertain u1
parameters and (b) standard
space of the uncorrelated
uncertain parameters contours of fX (x) contours of φU (u)
x1
parameters X into a set of uncorrelated standard The reliability index in Eq. 11 can be interpreted
normal deviates (variables) U (i.e., mU = 0, as the distance from the origin to the hyperplane
SU = I). To this aim, the basic random variables, (in standard normal space) defined by b0 +
X, can be transformed into a set of uncorrelated bTU = 0. In other words, the reliability index
standard variates U through the linear for a linear limit state function is numerically
transformation: equivalent to the distance from the origin to the
limit state hyperplane in the space of the standard
U ¼ L1 1
x D x ðX m x Þ or X 5 DX LX U þ mX normal variables. Using this geometric interpre-
(9) tation, Hasofer and Lind (1974) defined the reli-
ability index as the minimum distance between
where DX is diagonal matrix of the standard devi- the origin of standard normal space and the func-
ations of the basic random variables and LX is the tion g(U) (Fig. 4a).
lower triangular matrix obtained as the Cholesky The point U0 belonging to the limit state sur-
decomposition of the correlation coefficient face that has the minimum distance from the origin
matrix RX (i.e., RX = LXLXT). It is then possible of standard normal space is called design point
to define the joint-normal PDF fU(u) and (or performance point). The latter is the most
g(U) = G[X(U)] as the joint-normal distribution likely failure point. Strictly speaking, U0 is the
of the uncertain parameters and the limit state vector of uncertain parameters that has the maxi-
function, respectively, transformed in the space mum contribution towards the failure probability.
of the standard uncorrelated variables (Fig. 3). By As also mentioned before, the limit state func-
considering a linear performance surface (see tion is generally nonlinear. Therefore, the func-
Eq. 5), g(U) can be expressed as tion g(U) usually cannot be characterized by a
first-order polynomial. An approximate way to
gðUÞ ¼ G½XðUÞ ¼ a0 þ aT ðDX LX U þ mX Þ solve this problem is to replace the nonlinear
function with a first-order Taylor series expan-
¼ a 0 þ a T m X þ a T D X L X U 5 b 0 þ bT U
sion. In other words, the performance surface, in
(10) the neighborhood of the design point U0, is
approximated by the tangent hyperplane at U0
It can be easily demonstrated that the first and (Fig. 4b). In analytical terms, the approximate
second moments of this performance function are limit state function gF(U) becomes
mg = b0 and s2g = bTb. Substituting these two
terms in Eq. 7, the reliability index in the standard
space can be obtained as follows: gF ðUÞ ¼ gðU0 Þ þ ∇gT ðU0 Þ ðU U0 Þ (12)
mg b0
b¼ ¼ pffiffiffiffiffiffiffiffi (11) where ∇g(U0) is the gradient of g evaluated at the
sg bT b performance point and the subscript F represents
a u3 b u3 g(U)=0 c u3
g(U)=0
g(U)=0
U0 u2 U0 u2 U0 gS(U)=0 u2
u1 u1 gF(U)=0 u1
Seismic Reliability Assessment, Alternative Methods approximation, gF(U) = 0 at performance point, (c)
for, Fig. 4 Different types of limit state surface and their nonlinear surface approximated with a paraboloid,
approximations: (a) linear, (b) nonlinear with linear gS(U) = 0
the first-order approximation to g. In particular, point U0 (Fig. 3b). The components of a are
the first two moments of g(U) can be calculated as called the sensitivity factors of the reliability
follows (by using the relationships in Eq. 6): index b with respect to uncertain parameters U.
The smaller is ai, the smaller is the contribution of
mg ¼ gðU0 Þ þ ∇gT ðU0 ÞðmU U0 Þ the related uncertain parameter Ui to the failure
probability assessment.
¼ ∇gT ðU0 ÞðmU U0 Þs2g
Clearly, the determination of b requires the
¼ ∇gT ðU0 ÞSU ∇gðU0 Þ finding of the design point U0. Therefore, one
has to solve the following constrained optimiza-
¼ ∇gT ðU0 Þ∇gðU0 Þ (13)
tion problem:
where g(U0) is equal to zero since U0 is on the

limit state function, and the mean is equal to zero b ¼ minjUj
(15)
gð U Þ ¼ 0
(mU = 0) and the covariance is equal to one
(SU = I) by definition. For a nonlinear perfor-
mance function, the first-order reliability index There are many algorithms that are able to
can be defined as equal to the distance of the solve this kind of problems (see, e.g., Liu and
linearized failure surface (at the performance Der Kiureghian 1991). The most common
point U0) to the origin of the standardized algorithm used in structural reliability is the one
uncorrelated variables (Fig. 3b). With reference developed by Rackwitz and Flessler (1978).
to Eq. 12, it is possible to calculate the first A comprehensive description of the algorithm
two moments of the linearized limit state and relevant examples can be found in Pinto S
function (according to Eq. 13) and obtain the et al. (2007), while a first review of the efficiency
following reliability index (in literature also of the available algorithms was presented in Liu
referred to as first-order second-moment reliabil- and Der Kiureghian (1991). Once b is deter-
ity index bFOSM or Hasofer-Lind reliability mined, the probability of failure can be approxi-
index bHL): mated by Eq. 8.
mg ∇gT ðU0 Þ U0 Extension to System Reliability

b¼ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ aT U0 (14) It is noteworthy that in order to compute a system
sg
∇gðU0 ÞT ∇gðU0 Þ reliability problem, one must be able to compute
(or approximate) the probability of a union of
where a is the unit normal (directed towards N events (series system) or the probability of an
the failure region) on the hyperplane that is tan- intersection of N events (parallel systems).
gent to the limit state surface at the performance The probability of failure for these two cases
can be approximated as (Hohenbichler and therefore often called advanced or extended

Rackwitz 1983) FOSM, since the limit state function is approxi-
mated with a first-order approximation, while the
probability distributions are no longer approxi-
PF, series 1 FN ðb, RÞ PF, parallel FN ðb, RÞ
mated only with the first two central moments. In
(16) particular, if the limit state function is approxi-
mated with its first-order Taylor expansion, these
where FN is the multi-normal CDF, b is the methods are known as FORM (First-Order Reli-
N-vector of the reliability indexes associated ability Methods), and the analytical relations to
with the structural components, and R is the calculate the reliability index are similar to those
N∙N correlation matrix, in which the correlation reported from Eqs. 12 to 15. If the limit state
coefficients, rij = aiTaj, quantify the correlation function is approximated with its second-order
between failure modes i and j. In general, the Taylor expansion, these methods are known as
evaluation of FN(b, R) and FN(b, R) in SORM (Second-Order Reliability Methods).
Eq. 16 is not trivial; however, an efficient In both cases, the Taylor expansion is done at
Monte Carlo simulation algorithm has been the design point U0.
developed for this purpose (Ambartzumian A SORM analysis reduces the error associated
et al. 1998). The previous results can be com- with the non-flatness of the limit state function at
bined to estimate the probability of failure of a the design point in a FORM approximation.
general system (Ditlevsen and Madsen 1996). In particular, this approach fits a second-order
surface (usually a paraboloid) at the performance
point as shown in Fig. 4c. A parabolic approxi-
FORM and SORM Method in Component mation to the limit state surface at the design
Reliability point is
For the reliability index presented so far, only the gS ðUÞ ¼ gðU0 Þ þ ∇gT ðU0 Þ ðU U0 Þ
two moments of each uncertain parameter have
1
been considered. However, in many applications, þ ðU U0 ÞT HðU0 ÞðU U0 Þ (17)
information beyond the second moments is avail- 2
able. The methods that make use of that addi-
tional information are called full distribution where H(U0) is the Hessian matrix of g evaluated
reliability methods. If the probability distribution at U0 and the subscript S denotes the approxima-
is available for some or all the basic variables, it tion to the second order of g. In this case, it is
would be sensible to incorporate it in the reliabil- more difficult to express the first two moments of
ity analysis. If the joint probability distribution the limit state function (for more details, see Pinto
for the vector of uncertain parameters in the orig- et al. 2007); instead, the probability of failure is
inal space is not normal and the variables are not approximated in terms of b and the n1 principal
independent, the multifold integral presented in curvatures ki of the paraboloid in the standard-
Eq. 1 becomes very hard to solve. In fact, several ized uncorrelated n-space defined by vector U.
efficient methods for approximating the integral The exact expression of this probability is given
take advantage of the special properties of the in terms of single-fold integral defined by Tvedt
standard normal distribution of uncorrelated vari- (1990) which can be solved numerically in lack
ables. Therefore, the transformation of indepen- of closed-form solution. A simpler approxima-
dent/dependent non-normal uncertain parameters tion was derived on the basis of asymptotic anal-
to independent joint-normal uncertain parameters ysis by Breitung (1984):
is quite important.
n1 1
Historically, the transformation approach was PF FðbÞ ∏ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (18)
developed as an extension of the FOSM and is i¼1 1 þ bki
where b is calculated as the minimum distance where mR and mS denote the mean resistance and
between the origin of the standard normal space load; a0 R and a0 S are the sensitivity coefficients
and the paraboloid. (function of the sensitivity factors presented in
Eq. 14) for the resistance and load corresponding
to R and S, respectively; b is the FOSM reliability
Reliability Methods in Design Codes: index associated to the target reliability (obtained
The Safety Factors in LRFD from the inverse of the standard normal cumula-
tive distribution); and finally, dR and dS are the
Until the 1960s, most design codes did not provide coefficient of variation (i.e., the standard devia-
an explicit implementation of reliability concepts. tion divided by the mean) of the resistance and
The recommended methods mainly involved load. For example, assuming that the nominal
stress checks in critical sections, which aimed to values are equal to the mean values, it can be
verify that the unit stresses under the action of observed that the safety factor is only a function
working, or service, loads do not exceed of a, b, and d.
predesignated allowable values (i.e., values pre- Clearly, the safety factors depend on the cho-
scribed by the building code to provide a factor of sen limit states. For the ultimate limit state, for
safety against attainment of some limiting stress, instance, most of design codes provide safety
such as the minimum specified yield stress or the factors that are calibrated with respect to nominal
stress at which the buckling occurs); this design values. These nominal values are chosen as the
philosophy is also known as the working stress value with a predefined probability of being
design. The issue of uncertainty in the various exceeded; typically a lower (usually 5 %) percen-
parameters was implicitly addressed by employing tile is selected for the resistance and an upper
safety factors much larger than unity on the mate- (usually 95 %) percentile is selected for the load
rial strength. During the 1960s and 1970s, reliabil- effect. The European standard (CEN 2002) and
ity analyses were addressed more explicitly by the ASCE standard (ASCE 2010) both define the
codes, giving rise to reliability-based design. acceptable values of the reliability index. In par-
However, because professional design engineers ticular, the European standard defines minimum
found the application of the reliability-based values that are established as a function of the
methods somewhat difficult to access, a semi- consequences related to the reaching of a given
probabilistic code-based approach known as the limit state (with b varying from 3.3 to 4.3 with
load and resistance factor design (LRFD) format respect to 50 years of service period), whereas the
was proposed (Galambos et al. 1982): ASCE standard defines the acceptable reliability
index (which is related to a specified failure prob-
R=gR gS S or Rd Sd (19) ability) as a function of the structural damage and
occupancy category of the construction (with b S
where Rd and Sd are the design resistance and varying from 2.5 to 4.5 for 50 years of service
load, respectively; R and S are the nominal resis- period). This approach, however, allows for reli-
tance and nominal action (due to loads); and gR ability checking at the member level rather than
and gS are the associated resistance and load at the system level. As mentioned before,
safety factors. The codes aim to link these coef- establishing the link between component reliabil-
ficients with the target (admissible) reliability ity and system reliability is not a trivial problem.
and with the amount of uncertainty in each uncer-
tain parameter. The two safety factors in Eq. 19
have the general form Alternative Methods Based on Time-
Invariant Reliability Methods
mR m
gR ¼ 1 þ a0R bdR or gS ¼ S 1 þ a0S bdS
R S When an equivalent linear static procedure or a
(20) response spectrum analysis is performed, the
reliability analysis can be carried out in a straight- or has several design points that make significant
forward manner using the time-invariant reliabil- contributions to the failure probability (Faber
ity methods. However, in the case of dynamic 2012; Pinto et al. 2007). Although a large variety
nonlinear analyses, the problem becomes much of simulation techniques may be found in the
more complicated. In fact, in the latter case, it is literature, they generally have their origin in the
generally difficult to express the limit state of so-called Monte Carlo (MC) method (Rubinstein
interest directly in terms of time-invariant basic 1981; Fishman 1996). The general principles of
load and resistance uncertain parameters. There- standard MC simulation technique are outlined in
fore, the combination of the different source of the following section. To have a further insight
uncertainties becomes an intricate problem to into the basis of simulation techniques, the prob-
tackle. Nowadays, it is possible to combine the ability integral in Eq. 1 is rewritten as follows:
reliability methods and the finite element analysis
ð
to overcome this problem (Der Kiureghian 1996).
PF ¼ IF ðxÞf X ðxÞdx ¼ ½IF ðXÞ (21)
Currently, useful specialized software is avail-
OX
able to address this issue (Der Kiureghian
et al. 2006). Alternatively, a few hybrid
approaches have been recently proposed in where Ox denotes the entire sample space of
order to address the time variance in reliability X and IF(x) is an indicator function that equals
problem. one when G(x) 0 and zero otherwise. The
These methods combine the RTR variability integral in Eq. 21 is the expected value of the
(taken into account by alternative methods like indicator function denoted as [IF(x)]. In other
the PEER approach, which is discussed subse- words, PF is the expected value of IF(x) based on
quently in this entry) with other source of uncer- the joint PDF fX(x). It is noteworthy that through-
tainties (treated by current FORM/SORM out section “Simulation-Based Reliability
reliability methods). Typical hybrid approaches Methods,” p(x) might be used in place of fX(x)
consist of the confidence interval approach interchangeably. This is done in order to homog-
(Cornell et al. 2002; Ellingwood and Kinali enize and to simplify the probability notations
2009) and the mean estimate approach (Cornell within different equations. Hence, the term p(x)
et al. 2002). Other relevant examples can be is used in cases where the conditioning on a
found in Ibarra and Krawinkler (2005), Haselton preposition is defined, the use of fX might create
and Deierlein (2007), Dolšek (2009), Liel further confusions, or in the robust reliability
et al. (2009), Vamvatsikos and Fragiadakis expressions defined subsequently in section
(2010), and Celarec and Dolšek (2013). In these “Robust Reliability Assessment Using
studies, many approximations are presented; for Simulation-Based Approaches.” However, the
instance, if the limit state function does not have a marginal PDFs are expressed generally by fX.
defined functional form, the required gradients of By employing the MC method, various realiza-
the linearized limit state function can be obtained tions of the uncertain parameters x (i.e., xk, k = 1:
through perturbation of individual uncertain NS) can be independently sampled so that the
parameters in a series of sensitivity analyses. multidimensional integral in Eq. 21 is estimated
as follows:
Simulation-Based Reliability Methods 1 X NS

NF
PF ¼ ½IF ðXÞ ffi IF ðxk Þ ¼ (22)
N S k¼1 NS
The general integral form in Eq. 1 can also be
estimated by employing various simulation tech- where NS is the total number of simulations
niques. These techniques are especially efficient and NF is the number of realizations for which
when the limit state function is non-differentiable IF(xk) = 1 (i.e., the realizations leading to failure).
It can be shown that Eq. 22 is an unbiased esti- In this procedure, the analytic integration
mator for PF since is replaced with summation over a suitably large
set of samples. Generally, the approximation can
1 X NS
N S PF be made as accurate as needed by increasing NS.
ð½IF ðXÞÞ ¼ ½IF ðxk Þ ¼ ¼ PF
N S k¼1 NS
Fundamentals of MC Method
(23) The MC family of methods employs various sam-
pling techniques in order to generate realizations
The variance of the estimator can be calculated
of the prescribed uncertain parameters. The stan-
by considering that IF(xk)’s are independently
dard MC simulation uses the random sampling
distributed Bernoulli variables with variance
approach in order to generate a large amount of
equal to PF(1–PF):
realizations of uncertain parameters. The proba-
! bility of failure can then be estimated by directly
1 X NS
implementing Eq. 22 regardless of the complex-
ℝð½IF ðXÞÞ ¼ ℝ I F ð xk Þ
N S k¼1 ity of the problem. Checking whether the struc-
ture has failed for each sample usually requires
N S PF ð 1 PF Þ
¼ a structural analysis. To simulate k = 1:NS
NS2
outcomes of the joint density function
PF ð1 PF Þ (corresponding to Eqs. 21–22 or Eqs. 25–26) in
¼ (24)
NS standard MC simulation, a random number,
denoted as zj, k, between 0 and 1 is generated for
This equation reveals that the uncertainty in the each of the components in xk (i.e., xj, k, j = 1:n).
estimator decreases with increasing NS. There- Assuming that Xj’s are independent, the random
fore, in order to accurately quantify the likelihood numbers zj, k are mapped to xj, k by
of rare events (with small PF), which is a charac-
teristic feature in seismic assessment, NS should
be increased. Besides the integral form in Eq. 21 xj, k ¼ F1
Xj zj, k (27)
for the structural reliability formulation, many
problems in probabilistic inference require the
where FXj is the marginal Cumulative Distribu-
calculation of multidimensional integrals or sum-
tion Function (CDF) associated with the uncer-
mations over very large outcome spaces. In the
tain parameter Xj. Generally, the standard MC
general form, they can be expressed as
simulation is not justifiable computationally for
ð adequate estimation of very small failure proba-
R¼ hðxÞf X ðxÞdx ¼ ½hðXÞ (25) bilities as a very large number of simulations are
OX required. S
The Latin hypercube sampling technique
where R is a performance measure of interest and (LHS; Helton and Davis 2003) can be employed
h(x) is a response quantity of interest. The general in order to reduce the number of simulations, NS,
idea of MC integration is used to approximate the in addition to achieving an acceptable level of
expectation. Hence, a set of NS samples can be accuracy for the statistical characteristics of
drawn independently from distribution p(x), and response. The LHS is a special type of MC sim-
the expectations can be approximated by a finite ulation that uses the stratification of the theoreti-
sum: cal CDFs of uncertain parameters. Stratification
divides the CDF curve into NS equal intervals on
1 X NS the probability scale (i.e., 0.0 to 1.0). A sample is
R ¼ ½hðXÞ ffi hð xk Þ (26) then randomly drawn from each interval or “strat-
N S k¼1
ification” of the input CDFs based on the
technique of “sampling without replacement.” statistical correlation (Vořechovský and Novák

According to this technique, once a sample is 2009): (1) diminishing undesired and spurious
taken from a designated stratification, this strati- correlation between uncertain parameters gener-
fication is not sampled again. ated during sampling procedure, particularly in
As a result, the samples based on LHS reflect the case of a very small number of simulations,
more accurately the distribution of values in the and (2) introducing the prescribed statistical cor-
input probability distribution. LHS offers great relations between pairs of uncertain parameters
benefits in terms of increased sampling efficiency defined by the target correlation. Hence, in order
and faster runtimes (due to fewer NS that is to impose a prescribed correlation into the sam-
required). Its roles in different aspects of reliabil- pling scheme, an optimization problem for mini-
ity engineering have been described by, e.g., mizing the difference between the target
Novák et al. (1998) and Olsson et al. (2003). correlation and the actual correlation (estimated
Stein (1987) has shown that the LHS reduces from samples) should be solved. For this purpose,
the variance of the response function compared a stochastic optimization approach called simu-
to the crude Monte Carlo method. lated annealing (SA) has been recently proposed
Various methodologies were proposed for (Vořechovský and Novák 2009; Golafshani
sampling based on LHS method; however, the et al. 2011) for application of LHS together with
most efficient strategy to perform sample selec- the SA methodology in structural reliability
tion, which deals also with samples of the tails of assessment).
the PDF, is the sampling of interval mean values The main drawback of the standard MC sim-
denoted as LHS mean (Huntington and Lyrintzis ulation methods is that it is not computationally
1998). In this context, consider that the range of efficient for problems dealing with small proba-
probability distribution FXj between 0.0 and 1.0 is bilities of failure (i.e., PF 103). The main
divided into NS equidistant intervals each with reason is due to the fact that the number of sam-
equal probability of 1/NS. The expression xj, k is ples and hence the number of structural analyses,
the value of the uncertain parameter associated which are required to achieve a given accuracy, is
with the kth stratification with the probability inversely proportional to PF. It requires informa-
k/NS where k = 1:NS: tion from rare samples that lead to failure, and on
average, many samples are required before one
xj, k ¼ F1
Xj ðk=N S Þ (28) such failure sample occurs. For instance, in case
of achieving the PF in order of 104 (which is
The sampled data xj, k within the interval [xj, k–1, typical in seismic reliability assessment) with a
xj, k] can be derived as coefficient of variation around 10 %, nearly 106
simulations are required. Essentially, for rare
ð xj, k events, the chance of generating samples from
x f Xj ðxÞdx fX(x) for which IF(x) equals one is extremely
xj, k1
xj, k ¼ ð xj, k small (i.e., success rate is slow); therefore, many
f Xj ðxÞdx samples are needed to fulfill this requirement.
xj, k1 To overcome this obstacle, different tech-
ð xj, k niques have been proposed in order to reduce
¼ NS x f Xj ðxÞdx, j ¼ 1, . . . , n , the number of simulations. One of the most com-
xj, k1
monly applied techniques is namely the impor-
k ¼ 1, . . . , N S tance sampling method, which is described in the
(29) next section.
where fXj is the PDF of the uncertain parameter Importance Sampling Simulation Technique
Xj. In spite of high efficiency of LHS technique, The underlying idea of the importance sampling
there are generally two issues concerning simulation (Rubinstein 1981; Shinozuka 1983;
Schuëller and Stix 1987; Au and Beck 1999) is to considerable attention (see Au and Beck (1999)
carry out Monte Carlo simulation with samples for a complete discussion and suggestion of an
having a higher rate of falling in the failure region efficient kernel method).
since these samples contribute only to the evalu- Au and Beck (2001b) have also developed a
ation of PF. In this class of methods, the integral very efficient importance sampling method for
is written in the following form: the first-excursion problem for linear dynamical
systems under Gaussian stochastic excitation.
ð
pðxÞ pðXÞ However, for time-dependent problems, which

PF ¼ IF ðxÞ pIS ðxÞdx ¼  IF ðXÞ
pIS ðxÞ pIS ðXÞ are often characterized by a large number of
OX
uncertain parameters with complexity arising
(30) from their dynamic nature, the application of
importance sampling is much more difficult
where PIS(x) is denoted the importance sampling (Au and Beck 2003a).
density function to increase the probability of Hence, Au and Beck (2001a) proposed a sub-
sampling from the failure region. Applying set simulation technique that is capable of effi-
Monte Carlo simulation to the integral, PF is ciently computing the small failure probabilities
then estimated by encountered in engineering reliability analysis of
general dynamical systems.
1 X NS
pð xk Þ
PF ffi IF ðxk Þ (31) Calculation of Failure Probability Using the
N S k¼1 pIS ðxk Þ
Subset Simulation
Subset simulation is an efficient and adaptive
The efficiency of this method relies on a proper stochastic simulation algorithm for estimating
choice of the importance sampling distribution, small failure probabilities in high dimensions
which inevitably requires some knowledge about (Au and Beck 2001a, 2003b; Zuev et al. 2012).
failure zone. Many past studies successfully The underlying idea is to express the small failure
applied importance sampling to time-invariant probability as a product of larger probabilities
reliability problems where the number of uncer- conditional on some intermediate events. This
tain parameters is not too large (see, e.g., allows converting the simulation of a rare event
Schuëller and Stix 1987; Papadimitriou into a sequence of simulations of more frequent
et al. 1997; Der Kiureghian and Dakessian events (see also Jalayer and Beck 2008; Jalayer
1998; Au and Beck 1999). et al. 2010).
For instance, Schuëller and Stix (1987) pro- In this methodology, the failure region is
posed the selection of an importance sampling modeled as the last element in a sequence
density function as an N-dimensional joint- of embedded failure regions (i.e., decreasing S
normal PDF with uncorrelated components, nested sequence of failure regions)
where the mean value equals the design point as F ¼ Fm . . . F2 F1 . F1 is the first element
obtained from FORM analysis and standard devi- in the failure sequence (i.e., largest failure
ation values corresponding to the standard devi- region), and F = Fm is the target failure region
ations of the components of X. Although design and the last element in the failure sequence.
points are a good notion to characterize the Therefore, by considering the product rule in
important region in X and hence importance sam- probability, the failure probability can be derived
pling densities centered on them are often a good as follows:
choice, they are not a necessary ingredient for
forming a good importance sampling density. PF ¼ PðFm Þ ¼ PðFm jFm1 ÞPðFm1 Þ
Hence, adaptive or iterative schemes and in par- m1
ticular the kernel density estimators, which do not ¼ PðF1 Þ ∏ PðFiþ1 jFi Þ (32)
i¼1
involve the notion of design points, have attracted
According to Eq. 32, the rough idea of subset of the sequence of embedded failure regions,
simulation is to estimate the failure probability Fi, is a key issue to be addressed. It can signifi-
as a product of a sequence of conditional proba- cantly affect the efficiency of the subset
bilities denoted as {P(Fi+1|Fi):i = 1:m–1} and simulation procedure. If F is introduced with
P(F1). Hence, although PF is small, the condi- a single parameter, the sequence of
tional probabilities can be made sufficiently large intermediate failure regions Fi can be generated
by appropriately adopting m. by adaptively varying that parameter. The
Various simulation techniques can be used (scalar) demand-to-capacity ratio, which is suit-
for sampling from PDFs conditional on a failure able within the structural reliability assessment,
region Fi. Subset simulation proceeds in is defined as (Au and Beck 2003b; Jalayer
advance by simulating NS samples x(0) from et al. 2007a):
p(x) by MC simulation (note that superscript
“0” denotes that the samples correspond to the
N mech nj Di, j ðxÞ
first level): Y ðxÞ ¼ max min (35)
j¼1 i¼1 Ci, j ðxÞ
ð
1 X NS
ð0Þ
Pð F1 Þ ¼ IF1 ðxÞpðxÞdx ffi I F 1 xk where Nmech is the number of potential failure
N S k¼1
Ox mechanisms and nj is the number of components
(33) in the jth mechanism (cf. Eq. 4). This generalized
performance variable can be interpreted as the
From these samples, one can readily obtain new component demand-to-capacity ratio that brings
samples that are distributed as p(x|F1) (i.e., these the system closer to the desired limit state. It is
samples lie in F1). These samples can be used to interesting to note that the performance
estimate P(F2|F1) by sample averaging similar to variable in Eq. 35 takes into account a range of
what has already done in Eq. 33. These samples potential failure mechanisms (cut-sets) as noted
provide seeds for simulating more samples in section “Structural Component and Structural
according to p(x|F2) and hence for estimating System Reliability.” As a result, the target
P(F3|F2). Repeating this process, the conditional failure region can be rewritten as F = {x:
probabilities of the higher-conditional levels can Y(x) > 1}, and the sequence of embedded
be computed until the failure region of interest Fm intermediate failure regions can be generated as
is attained. In the ith conditional level, where 1 Fi = {x:Y(x) > yi} where 0 < y1 < ∙∙∙ < ym = 1 are
i m1, let x(i) be the samples with distribution the intermediate thresholds. Au and Beck
p(x|Fi); the conditional probability can be (2003b) proposed an adaptive methodology for
expressed as choosing yi values so that the estimated condi-
tional probabilities are equal to a fixed value.
ð
1 X NS
ði Þ
They found that a fixed value equal to 0.1 yields
PðFiþ1 jFi Þ IFiþ1 ðxÞpðxjFi Þdx ffi IFiþ1 xk good efficiency.
N S k¼1
Ox This section attempts to briefly introduce the
(34) main concepts of the subset simulation as an
advanced stochastic simulation method for esti-
The subset simulation is shown to be especially mation of small probabilities corresponding to
efficient for modeling rare failure events (i.e., rare failure events. However, a detailed introduc-
when the probability of failure is very small). tory description of methodology can be found
The statistical properties of each conditional fail- in the chapter “▶ Subset Simulation Method
ure probability as well as the PF are derived in Au for Rare Event Estimation: An Introduction”
and Beck (2001a, 2003b). of the general section “Reliability and Robust-
Considering that the target failure region ness” of the Encyclopedia of Earthquake
can be stated as F = {x: D(x) > C(x)}, the choice Engineering.
ð
Robust Reliability Assessment Using
PF ¼ pðFjD, MÞ ¼ pðFjx, D, MÞpðxjD, MÞdx
Simulation-Based Approaches
The characterization of uncertainties can be OX
ð
treated in two levels: (1) prior probability distri- ¼ Ι F ðxÞpðxjD, MÞdx
butions for the uncertain parameters based on
OX
available information and/or (qualitative) profes-
sional judgment and (2) the results of in situ tests, (36)
various types of inspections, pseudo-dynamic
where D denotes some test data and M is the set of
health-monitoring tests, and the consideration of
possible structural models used to specify (both
the corresponding measurement errors that can be
the structural and the probabilistic) modeling
used to update the prior distributions of uncertain
assumptions in the analysis; p(x|D, M) is the
parameters. The treatment of the latter issue is
posterior PDF of the model parameters, x, based
tackled by the Bayesian updating procedure (e.g.,
on the observed data and assumed structural
Box and Tiao 1973; Jaynes 2003).
model; and P(F|x, D, M) is the failure probability
In the Bayesian approach, the probability is
given the model parameters defined by x, which
always conditional on the amount of information
can be reduced to a deterministic index
available. In this probabilistic framework, prob-
function, IF. This index is a function of the
ability represents the degree of belief in a certain
model parameters which equals one if failure
outcome based on the amount of information.
occurs and zero otherwise. The conditioning on
The Bayesian approach was not assessed objec-
M is included here to stress the fact that all prob-
tively by many statisticians until late last century
abilities involved in model updating are always
because of the absence of a strong rationale
conditional on the choice of the modeling
behind the theory (see Jaynes 2003; Zuev
assumptions.
et al. 2012 for a brief summary of development
The Bayesian framework is used herein to
corresponding to the Bayesian approach). How-
provide a rigorous method for updating the plau-
ever, the seminal work of the physicist Cox
sibility of each of the models in representing the
(1961) expounded by the physicist Jaynes
structure that is quantified by the probability dis-
(2003) has significantly enhanced the Bayesian
tribution over the vector of model parameters or
probability theory as a convenient mathematical
the posterior PDF, p(x|D, M):
language for inference and uncertainty quantifi-
cation. The Bayesian approach usually leads to
high-dimensional integrals that often can be eval- pðxjD, MÞ ¼ pD ðxÞ ¼ c1 pðDjx, MÞpðxjMÞ
uated neither analytically nor numerically by (37)
straightforward quadrature. Nevertheless, the
development of new algorithms as well as where p(x|M) is the prior probability distribution S
increasing computing power has led to an explo- for x specified by M, which reflects the
sive growth of using Bayesian approach in differ- relative plausibility of each model before utiliz-
ent branches of science. ing the data D, p(D|x, M) is the (updated)
In the robust (updated) reliability approach, probability distribution (also known as the
the plausibility of all the possible structural likelihood function) for observed data D based
models conditional on the amount of information on a model specified by the model parameters x,
available is taken into account in a Bayesian and c1 is a normalizing constant. Thus, the
framework (for more details, see Jaynes 2003; robust failure probability in Eq. 36 can be
Papadimitriou et al. 2001; Beck and Au 2002; shown as
Jalayer et al. 2010; Zuev et al. 2012). The robust
ð
failure probability then can be calculated 1
PF ¼ c Ι F ðx, MÞpðDjx, MÞpðxjMÞdx (38)
(in comparison to Eq. 21) by the following
integral: OX
Similarly, the general performance measure of be generated efficiently by sampling from the
Eq. 25 can also be rewritten based on the robust prior PDF using direct MC method. Of course,
reliability concept as follows: if the posterior PDF pD is evaluated first from
Eq. 37, then there would be no problem in using
ð ð
the standard MC from Eq. 36. However, in this
RD ¼ hðxÞpðxjD, MÞdx ¼ hðxÞpD ðxÞdx
case, the factor c1 needs to be calculated.
OX OX According to the drawbacks of the current MC
(39) methods as well as the importance sampling tech-
niques in solving robust reliability problems, Mar-
Substituting Eq. 37 into 39, RD can also be kov chain Monte Carlo (MCMC) simulation
expressed as techniques, in particular, the Metropolis-Hastings
(MH) algorithm (Metropolis et al. 1953; Hastings
ð
1 1970), are discussed briefly in the next section.
RD ¼ c hðxÞpðDjx, MÞpðxjMÞdx (40)
It offers a feasible and powerful way for simulat-
OX ing samples according to an arbitrary distribution
(when the target PDF is known only up to a scaling
Equations 36 and 38 (similarly, Eqs. 39 and 40) constant), at the expense of introducing depen-
suggest two ways of evaluating the robust reli- dence among the samples.
ability by simulation. Equations 36 and 39 sug-
gest estimating the PF or RD as the average of IF Generating Samples in Robust Reliability
or h over samples simulated from pD, while Based on Adaptive MCMC Simulation
Eqs. 38 and 40 indicate that the integral and the Technique
normalizing constant should be estimated, indi- The goal of MCMC is to design a Markov chain
vidually and then combined. such that the stationary distribution of the chain is
Particular difficulties are encountered in the exactly the distribution that we are interested
evaluation of PF or RD based on the expressions in sampling from (i.e., the target distribution).
in Eq. 36 or 39, respectively. The updated prob- The idea is to use specially designed methods
ability density function pD is known only up to a for setting up the transition function, such that
multiplicative constant (see Eq. 37), since the no matter how each chain is initialized, it will
normalizing constant c1 is usually given by a converge to the target distribution. The MH algo-
high-dimensional integral that is also difficult to rithm is normally used to generate samples
evaluate (Beck and Au 2002; Jalayer et al. 2010). according to an unscaled PDF when the target
Hence, the application of MC simulation or PDF is known only up to a scaling constant, i.e.,
importance sampling is not generally feasible in there is no need to know the normalizing constant
this case, because these methods cannot simulate (see Eq. 33) in advance. The fact that this proce-
independent samples from pD. dure allows us to sample from un-normalized
Additionally, when using importance sam- distributions is one of its major attractions espe-
pling (see Eq. 30), it is necessary to choose a cially in case of Bayesian model updating. The
sampling density that is concentrated in the afore- MH algorithm was originally developed by
mentioned small zone; otherwise similar prob- Metropolis and his coworkers (Metropolis
lems, as in the MC simulation, will arise. et al. 1953) in statistical physics and later gener-
However, this task is not trivial to accomplish alized by Hastings (1970) in Bayesian statistics
since information about the zone, where p(D|x, (see Fishman 1996 for comprehensive discus-
M) is concentrated, is not directly available (see sions on the MCMC methods). However, its
Beck and Au 2002 for more details). Strictly potential use for solving reliability problems in
speaking, the posterior PDF pD occupies a much structural engineering has been only recently
smaller volume than that of the prior PDF p(x|M), demonstrated (see Au and Beck 1999, 2001a,
so samples in its high probability region cannot 2003b; Beck and Au 2002).
The MH algorithm can be used to generate and assessment of building structures has been
samples according to the target updated PDF developed by the Pacific Earthquake Engineering
pD. Using the Markov chain samples generated Research Center (PEER; Cornell and Krawinkler
from the MH procedure, PF or RD is estimated as 2000; Moehle and Deierlein 2004). There are
the average over the samples, which is the same several stages to this process: (1) calculation of
approach as the usual MC method, except that the ground motion hazard by quantifying uncertainty
samples are simulated from a Markov chain in ground motion with a probabilistic model for a
instead of being independent and identically dis- parameter related to ground motion and known as
tributed. In order to reduce the initial effect of the the intensity measure (IM); (2) estimation of the
choice of the transient probability distributions uncertainty in structural response D (e.g., force
(e.g., non-normalized target PDF) on the esti- and deformation engineering parameters), for
mate, the first few samples are often not included each IM level; (3) estimation of the uncertainty
in the sample averaging. in damage measure DM (i.e., physical states of
In the MH procedure, a proposal PDF is cho- damage) given response level D; and (4) estima-
sen to generate a candidate point that is condition of the uncertainty in the decision variable DV
tional on the previous state of the sampler. The as the resulting consequences (e.g., financial
next step is to either accept the proposal or reject losses, fatalities, and business interruption)
it. However, as noted previously, the updated given damage measure.
posterior PDF pD is concentrated in a small Each stage of the process is performed and
zone, and direct adaptation using a proposal executed (more or less) independently and then
PDF, which varies with a vastly different length linked back together, as expressed in the follow-
scale from that of the target PDF, will not be ing integral form:
effective (see Beck and Au 2002 for more
ððð
details). Therefore, a sequence of intermediate dG½DMjD
lðDV Þ ¼ G½DV jDM
proposal PDFs that vary gradually between the dDM
prior PDF and the target PDF pD are introduced
dG½DjIM dlðIMÞ
(Au and Beck 1999; Beck and Au 2002).
dIM
dD dIM
According to the proposed methodology, how-
ever, the proposal PDF is chosen as the kernel dD dDM (41)
sampling density constructed using the Markov
chain samples from the previous simulation level where G[Y|X] denotes generically the conditional
(for the first simulation level, the prior PDF is Complementary Cumulative Distribution Func-
used as the proposal PDF). Thus, the adaptation is tion (CCDF) of Y given a certain value of X and
done from one simulation level to the next. This l(Y) denotes the mean annual exceedance rate
methodology is especially useful in the subset (mean annual frequency) of Y. As it can be S
simulation described previously in section “Cal- observed from Eq. 41, the PEER framework
culation of Failure Probability Using the Subset enjoys a modular structure and benefits from the
Simulation.” Accordingly, a modified MH algo- (hypothetical) conditional independence between
rithm was developed by Au and Beck (2003b) the main parameters (i.e., the conditional inde-
that can obtain samples from a posterior PDF pendence of DV|DM from D and IM and other
conditioned on a failure region Fi. ground motion parameters, DM|D from IM and
other ground motion parameters, and D|IM from
other ground motion parameters, such as, but not
PEER Performance-Based Approach limited to, magnitude and distance).
This (hypothetic) Markovian independence
In recent years, an effective foundation for the between various intermediate parameters is one
development of probabilistic Performance-Based of the main factors that distinguish the PEER
Earthquake Engineering (PBEE) for the design approach from the two other approaches
discussed in this entry, namely, the classic and The main advantage of this methodology is
the simulation-based reliability methods. In these that it evaluates the limit state frequency, lLS,
approaches, the probability of failure (which can by decoupling the ground motion hazard and
generically represent a decision variable DV nonlinear dynamic analyses via a site- and
exceeding a certain threshold value) is calculated structure-specific intermediate variable known
directly and without considering intermediate as the ground motion intensity measure, IM (see
parameters such DM, D, and IM. Another impor- also Shome et al. 1998; Shome and Cornell 1999;
tant factor that distinguishes the PEER frame- Cornell and Krawinkler 2000; Luco 2002; Jalayer
work equation (Eq. 41) from the other two 2003; Jalayer and Cornell 2009). The ground
approaches, discussed herein, is that Eq. 41 yields motion IM serves as link between seismic hazard
a rate of exceedance and not a probability. This is analysis, typically provided by seismologists, and
while the classic and the simulation-based reli- structural analysis conducted by engineers.
ability methods lead to a probability of exceed- This IM approach is appealing because it
ance. In other words, the rate of exceedance allows the first two analysis stages in the PEER
reported in Eq. 41 needs to be translated into PBEE methodology to be performed (almost)
probability by assuming an underlying probabil- independently. The benefit of this approach is
ity model (see Der Kiureghian 2005). that the number of analyses needed can be sub-
In this section, the second stage of the PEER stantially reduced because most of the uncer-
integral expression dedicated to estimating the tainties are concentrated in ground motion
conditional probability of exceeding the engineer- hazard in terms of the MAF of exceeding a cer-
ing demand parameter D given IM, denoted as G tain level of IM, lIM, which can directly be esti-
[D|IM], is discussed in detail. The estimation of the mated from probabilistic seismic hazard analysis
mean annual frequency of exceeding a specific IM (PSHA; Cornell 1968; McGuire 2004) for the
values, denoted as l(IM), is the main focus of the area where the structure is located.
chapter “▶ Probabilistic Seismic Hazard Models.” The conditional probability of D given IM can
Finally, an outlook into the whole framework is be predicted through alternative nonlinear
provided in the chapter entitled “▶ Performance- dynamic analysis procedures. Herein, three dis-
Based Design Procedure for Structures with tinct nonlinear dynamic procedures are
Magneto-Rheological Dampers”. described: (1) the so-called cloud analysis
Taking into account other sources of uncer- method, in which the structure is subjected to a
tainty imposes an overall increase in the disper- set of (as-recorded) ground motions (Cornell
sion for the failure probability (Jalayer 2003; et al. 2002; Jalayer and Cornell 2003; Elefante
Jalayer and Cornell 2003), while an eventual et al. 2010), and subsequently, the distribution of
shift in the median failure probability is not D given IM (D and IM are denoted as cloud data)
envisioned in this method (see the “hybrid” is directly obtained by performing a logarithmic
methods discussed in section “Alternative linear regression on the cloud data; (2) the incre-
Methods Based on Time-Invariant Reliability mental dynamic analysis (Vamvatsikos and Cor-
Methods”). nell 2004) where a selected suite of ground
In the PEER PBEE methodology, the evalua- motions are incrementally scaled to different
tion of structural performance is defined by prob- levels of the IM; and (3) multiple-stripe analysis
abilistic performance objectives, which can be (MSA; Jalayer 2003; Jalayer and Cornell 2009),
expressed in terms of the mean annual frequency where analysis is performed at a specified set of
(MAF) of exceeding a specified limit state (also IM levels known as stripes, each of which may be
referred to as the limit state frequency) denoted as performed for a different suite of ground motions
lLS. Therefore, it is worth noting that the usual (i.e., structural dynamic analyses for multiple
outcome of this framework is an estimate of the stripes of IM through selected set of records for
rate of exceeding a designated limit state thresh- each IM level). It is noteworthy that although
old rather than a probability. IDA uses the same set of ground motions for all
IM levels, the MSA approach allows the 1

re-selection of ground motions at each IM level
P[IMcap<IM/IM=im]
P[D>CLS|IM=im]
to be consistent with the seismicity of the desig-
nated site (based on the disaggregation of seismic
hazard; see, e.g., Baker and Cornell 2005). Thus, 50th percentile
or
the distribution of D given IM is estimated for
multiple IM levels.
The structural fragility curve for a range of IM
values is a useful probabilistic structural response η: median IM
0
quantity that can directly be estimated from the IM
aforementioned nonlinear dynamic analysis pro-
cedures. The structural fragility for a given limit Seismic Reliability Assessment, Alternative Methods
for, Fig. 5 Typical fragility curve
state, LS, can be defined as the conditional prob-
ability of exceeding the limit state capacity for a
given level of IM (Ellingwood 2001; Jalayer
2003; Baker and Cornell 2005; Jalayer and Cor-
representing the capacity of structure in terms of
nell 2009; Baker 2014). If the considered limit
IM values (distribution of IM values) that result in
state corresponds with the collapse condition, the
a D level equal to CLS, and the probability terms
associated fragility curve is known as the collapse
denote the structural fragilities for the desired
capacity. Subsequently, this result can be com-
limit state. The typical shape of the fragility
bined with a ground motion hazard, i.e., lIM, to
curve is presented in Fig. 5.
compute the MAF of exceeding any given limit
The fragility curve can be estimated by cloud
state (Shome and Cornell 1999; Jalayer 2003;
analysis, MSA, or IDA.
Jalayer and Cornell 2003; Ibarra and Krawinkler
2005; Haselton and Deierlein 2007).
Cloud Analysis Approach
Mean Annual Frequency of Exceeding a The probability term presented in Eq. 42 can be
Limit State derived by assuming D to be a lognormal vari-
The MAF of exceeding a specified limit state, able. This is a common assumption that has been
denoted as lLS, can be obtained from two confirmed as reasonable in many past studies
approaches: (a) the IM-based approach (also (e.g., Porter et al. 2007; Jalayer and Cornell
known as the fragility/hazard format) and 2009):
(b) the D-based (i.e., demand-based) approach,
which in the literature is also called the engineer-
!
ing demand parameter (EDP)-based approach lnCLS lnDjIM ðimÞ S
(Cornell and Krawinkler 2000). The two P½D > CLS jIM im ¼ 1 F
bDjIM ðimÞ
approaches are expressed by the two following
(44)
equations, respectively:
ð
l LS ¼ P ½D > CLS j IM ¼ im jdl IM ðimÞj (42) where F is the standardized Gaussian CDF and
and b are conditional median and standard devi-
im
ð ation (dispersion) of the natural logarithm of

l LS ¼ P IMcap < IMj IM ¼ im jdl IM ðimÞj D given IM. In order to estimate the statistical
properties of the cloud data, i.e., conditional
im
median and dispersion, conventional linear
(43)
regression (using least squares method) is used
where CLS is the capacity associated with the in the natural logarithmic scale, as shown in
given LS, IMcap is an uncertain parameter Fig. 6. Thus,
Seismic Reliability
102
Assessment, Alternative
Methods for,
Fig. 6 Typical logarithmic
1
linear regression on 100
b
ln (IM)
cloud data β
10−2
10−4 −2
10 10−1 100 101
ln (D)
lnDjIM ðimÞ ¼ ln a þ b lnðimÞ, analytical pdf

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
!ffi
uNcloud
uX C
u ln
LS (45)
u a imbj
t j¼1 Single IDA curve
bDjM ¼
N cloud 2
where a and b are the coefficients of the logarith-
IM
mic linear regression and Ncloud is the number of

the cloud data.
Rare events may exert a very large demand
D (i.e., even cause instability in the nonlinear
dynamic analysis) that can affect the general
trend in the linear regression of the cloud data.
In such cases, a logistic regression (Agresti 2002)
can generally be used. In this methodology, D LS threshold
binary values of 0 (for non-collapse data) and
Seismic Reliability Assessment, Alternative Methods
1 (for collapse cases) are assigned for each IM,
for, Fig. 7 Typical IDA curves and fitting of the empir-
and the logistic regression allows the construc- ical distribution of the IM related to the overcoming of
tion of the associated fragility curve based on limit state
these binary data (Baker and Cornell 2005;
Elefante et al. 2010).
associated with collapse. By repeating this pro-
IDA Approach cess for a set of ground motions, one can obtain a
In a great number of research works (see, e.g., set of IM values corresponding to the onset of
Jalayer 2003; Vamvatsikos and Cornell 2004; collapse. Subsequently, the probability of col-
Baker and Cornell 2005; Ibarra and Krawinkler lapse for a specific IM equal to im can be esti-
2005; Zareian and Krawinkler 2007; Haselton mated as the fraction of records for which
and Deierlein 2007; Jalayer and Cornell 2009; collapse occurs at a level lower than the
Baker 2014), the collapse capacity is estimated predefined value, im. A lognormal distribution
by repeatedly scaling a ground motion using the is often fitted to the empirical distribution of
IDA procedure until the ground motion causes IM levels that cause the structure to collapse.
collapse of the structure; hence, each ground In Fig. 7, the black dashed line represents the
motion can be associated with a single IM value LS threshold, and the red stars illustrate the
IMcap values that cause the onset of the limit state. fragility function are shown schematically in
On the same diagram, the histogram of the IMcap Fig. 8 (Ebrahimian 2012).
data and the related empirical distribution are
shown as well. Demand and Capacity Factor Design (DCFD)
Format (IM-Based Version)
Multiple-Stripe Analysis Approach The IM-based closed-form solution for the
In cases where different records are used at each annual frequency of exceeding a desired limit
IM level, the IDA method (section “IDA state, lLS, has been derived by Jalayer (2003)
Approach”) cannot be used to estimate the prob- (see also Jalayer and Cornell 2003) under a set
ability of collapse. A substitute for the conven- of simplifying assumptions. It is used as a
tional IDA approach should be utilized in this performance-oriented design procedure proposed
particular case. Therefore, instead of an IM in the SAC/FEMA Steel Project (Cornell
value associated with the onset of collapse for et al. 2002) as
each record (i.e., the IDA approach), the fraction
of records that cause collapse at each IM level can 1
lLS ¼ lIM IMcap exp k2 b2IMcap (46)
be taken into account. The latter can be obtained 2
from individual stripes by calculating the number
of accelerograms for which the limit state has where IMcap and bIMcap are the parameters of the
been exceeded with respect to the total number estimated lognormal distribution assuming that
of records (Baker and Cornell 2005; Ebrahimian the IMcap capacity is a lognormal variable and
2012; Baker 2014). k reflects the steepness of the hazard curve,
As a result, the empirical distribution of the lIM, in the vicinity of the point IMcap (the hazard
collapse capacity for a set of IM values can be curve is approximated in the region of interest
obtained. Once the ratio of the number of collapse by the power-law relationship as lIM(IMcap) =
cases to the total number of ground motions is ko(IMcap)k). The first term in Eq. 46 shows
calculated for each IM level, a fragility curve can the first-order approximation to the limit state
be generated through the logistic regression. The probability, and the exponential expression
observed fraction of collapse and the estimated is a magnifying factor that accounts for the
a b
1 1
Logistic regression
9 failure on
10 accelerograms S
P[D>CLS|IM=im]
LS threshold
D
0 0
IM IM
Seismic Reliability Assessment, Alternative Methods for, Fig. 8 (a) Typical MSA output and (b) fragility curve
obtained through the logistic regression applied on the MSA results
sensitivity of the limit state probability to the uncertainty, applying any of the aforementioned
randomness in the IMcap. methods is not a trivial task.
The closed-from expression for the limit state Arguably, the first-order and second-order
frequency can represent an alternative interpreta- reliability methods cannot be adopted in a
tion for the seismic design or assessment consid- straightforward manner when solving time-
ering an IM-based design criterion. A certain dependent nonlinear reliability problems. On the
design criterion is to check whether the MAF of other hand, the PEER approach works best in
exceeding a certain limit state (limit state fre- cases where the primary source of uncertainty is
quency) is less than or equal to an allowable reflected in the RTR variability (Der Kiureghian
annual frequency, lo. Thus, 2005). Therefore, in the recent years, various
hybrid methods (section “Alternative Methods
1
Based on Time-Invariant Reliability Methods”)
lLS ¼ lIM IMcap exp k bIMcap lo (47)
2 2
2 have been proposed (see, e.g., Liel et al. 2009).
Other approaches include those developed by
After some simple rearrangements with the Vamvatsikos and Fragiadakis (2010) and Celarec
objective of allocating the parameters and Dolšek (2013) where the PEER approach is
corresponding to the capacity to one side, mixed with other reliability methods (FOSM and
Eq. 47 takes the following form (Jalayer 2003; Latin hypercube sampling). As a result, the treat-
Jalayer and Cornell 2003): ment of various sources of uncertainty in both
aforementioned reliability methods deserves fur-
1 ther research and investigations.
imlo IMcap exp k2 b2IMcap (48)
2 It is noteworthy that both the modern con-
struction codes and the professional community
where imlo is the indicator IM for a hazard level of engineers around the world are becoming more
equal to the allowable annual frequency, lo. The inclined towards the application of nonlinear
right-hand side of this expression represents the dynamic analyses. This raises the professional
factored capacity (expressed in terms of the interests to the PEER approach, since it provides
adopted IM). Therefore, this expression illus- a more accessible way of taking into account and
trates a design criterion based on the fragility/ propagating the RTR variability in design and
hazard format in terms of the IM-based factored assessment problems. Moreover, this approach
capacity being greater than or equal to the IM for leads directly to the estimation of the economic
a given allowable annual rate. losses, which are viewed as essential risk metrics
in decision making.
In order to take into account various sources of
Summary and Future Challenges uncertainty (e.g., modeling, RTR, etc.) in the
PEER approach while maintaining a low number
A summary of general methodologies used in the of structural analyses, Jalayer et al. (2013) pro-
seismic reliability assessment has been posed a method inspired from the concept of
presented. In particular, the historical evolution robust reliability discussed in section “Calcula-
of each method that allows for accounting uncer- tion of Failure Probability Using the Subset Sim-
tainties in demand and capacity is briefly illus- ulation.” Conditioned on a prescribed analytical
trated. It is noted that considerable progress has fragility model, they suggest using the standard
been made in the development of these method- MC simulation (with few samples) in order to
ologies, and great efforts have been performed to create various structural model realizations
implement them with the aim of loss reduction based on the modeling uncertainties present in
and risk mitigation. Considering that the seismic the problem. A suite of recorded ground motions
reliability of structures is a specific time- employed to represent the RTR variability is then
dependent problem with many sources of applied to each of the above-mentioned structural
models in order to calculate a desired structural The Bayesian network methodologies have
response. In other words, the number of structural been developed during the past 25 years mostly
response realizations is equal to the number of in the field of artificial intelligence (Russell and
ground motion records present in the suite of Norvig 2003). This method has recently been
records. These structural response realizations applied in order to solve high-dimensional struc-
are then used as data in order to both update the tural reliability problems such as infrastructure
joint probability distribution for the parameters of seismic risk assessment for spatially distributed
the adopted analytic fragility model and also to systems (see, e.g., Grêt-Regamey and Straub
directly obtain the robust fragility. 2006; Straub and Der Kiureghian 2010). The
In the simulation-based approach, consider- Bayesian networks rely on graphical visualiza-
ation of RTR variability requires the adoption of tion of the uncertain parameters and their corre-
a suitable stochastic ground motion model (see lation structure. They are arguably going to
the chapter “▶ Stochastic Ground Motion Simu- become efficient means of resolving systemic
lation”) conditional on seismic source parameters reliability problems due to their propensity for
(see Au and Beck 2003b; Jalayer and Beck 2008). being automatized and also due to the fact that
Accordingly, the consideration of RTR variabil- they lend themselves quite well to the implemen-
ity together with other sources of uncertainties tation of advanced simulation-based methods.
will provide a complete probabilistic treatment of
the structural response (see, e.g., Jalayer
et al. 2007b where RTR variability and modeling
uncertainty are considered together in the seismic
References
reliability analysis of RC frames based on the
Agresti A (2002) Categorical data analysis. Wiley, New
subset simulation). One of the main challenges York, 710 pp
in applying this approach is to ensure that the Ambartzumian R, Der Kiureghian A, Ohanian V,
stochastic ground motion model provides a real- Sukiasian H (1998) Multinormal probability by
sequential conditioned importance sampling: theory
istic description of the characteristics of the
and application. Probab Eng Mech 13(4):299–308
ground motions expected to happen at the build- ASCE American Society of Civil Engineers
ing site (see, e.g., Atkinson and Silva 2000; (2010) ASCE/SEI 7–10, Minimum design loads for
Rezaeian and Der Kiureghian 2010). Moreover, buildings and other structures, New-York
Atkinson G, Silva W (2000) Stochastic modelling of
there are difficulties in the application of
California ground motions. Bull Seismic Soc Am
advanced simulation-based algorithms (e.g., sub- 90(2):255–274
set simulation) so that these techniques cannot be Au SK, Beck JL (1999) A new adaptive importance sam-
performed in a straightforward and practical pling scheme for reliability calculations. Struct Saf
21(2):135–158
manner. However, the fast and incredible
Au SK, Beck JL (2001a) Estimation of small failure prob-
increase in computational power is going to ren- abilities in high dimensions by subset simulation. S
der the simulation-based methods ever more Probab Eng Mech 16(4):263–277
accessible for resolving high-dimensional reli- Au SK, Beck JL (2001b) First excursion probabilities for
linear systems by very efficient importance sampling.
ability problems.
Probab Eng Mech 16(3):193–207
In comparison to the PEER approach, the Au SK, Beck JL (2003a) Important sampling in high
advanced simulation-based methods have the dimensions. Struct Saf 25(2):139–163
advantage of being able to directly estimate Au SK, Beck JL (2003b) Subset simulation and its appli-
cation to seismic risk based on dynamic analysis. J Eng
losses without passing through multiple stages Mech 129(8):901–917
(e.g., hazard, demand, damage, losses). Undoubt- Baker JW (2014) Efficient analytical fragility function
edly, the application of advanced simulation- fitting using dynamic structural analysis. Earthq Spec-
based techniques is another major concern in tra doi: http://dx.doi.org/10.1193/021113EQS025M
Baker JW, Cornell CA (2005) Vector-valued ground
the future of reliability methods due to the usually motion intensity measures for probabilistic seismic
high dimension of the vector of uncertain demand analysis. Report No. 150, John A. Blume
parameters. Earthquake Engineering Center, Stanford, 321 pp
Beck JL, Au SK (2002) Bayesian updating of structural Elefante L, Jalayer F, Iervolino I, Manfredi G (2010)
models and reliability using Markov chain Monte Disaggregation-based response weighting scheme for
Carlo simulation. J Eng Mech 128(4):380–391 seismic risk assessment of structures. Soil Dyn Earthq
Box GEP, Tiao GC (1973) Bayesian inference in statisti- Eng 30(12):1513–1527
cal analysis. Wiley, New York Ellingwood BR (2001) Earthquake risk assessment of
Bozorgnia Y, Bertero VV (2004) Earthquake engineering: building structures. Reliab Eng Syst Saf 74(3):251–262
from engineering seismology to performance-based Ellingwood BR, Kinali K (2009) Quantifying and com-
engineering. CRC Press, ISBN 0-8493-1349-9 municating uncertainty in seismic risk assessment.
Breitung K (1984) Asymptotic approximations for Struct Saf 31(2):179–187
multinormal integrals. J Eng Mech 110(3):357–366 Faber MH (2012) Statistics and probability theory: in
Celarec D, Dolšek M (2013) The impact of modelling pursuit of engineering decision support, vol 18.
uncertainties on the seismic performance assessment Springer, New York
of reinforced concrete frame buildings. Eng Struct Fishman GS (1996) Monte Carlo: concepts, algorithms,
52:340–354 and applications. Springer, New York
CEN (2002) European Committee for Standardisation, Galambos TV, Ellingwood B, MacGregor JG, Cornell CA
Eurocode-Basis of Structural Design, EN 1990:2002 (1982) Probability based load criteria: assessment of
Cornell CA (1968) Engineering seismic risk analysis. Bull current design practice. J Struct Div 108(5):959–977
Seismol Soc Am 58(5):1583–1606 Golafshani AA, Ebrahimian H, Bagheri V, Holmas
Cornell CA (1969) A Probability-based structural code. T (2011) Assessment of offshore platforms under
ACI J 66(12):974–985 extreme waves by probabilistic incremental wave
Cornell CA, Krawinkler H (2000) Progress and challenges analysis. J Constr Steel Res 67(5):759–769
in seismic performance assessment. PEER Cent News Grêt-Regamey A, Straub D (2006) Spatially explicit ava-
3(2):1–3 lanche risk assessment linking Bayesian networks to a
Cornell CA, Jalayer F, Hamburger RO, Foutch DA GIS. Nat Hazards Earth Syst Sci 6(6):911–926
(2002) Probabilistic basis for 2000 SAC Federal Emer- Haselton CB, Deierlein GG (2007) Assessing Seismic
gency Management Agency steel moment frame Collapse Safety of Modern Reinforced Concrete
guidelines. J Struct Eng 128(4):526–533 Moment-Frame Buildings, PEER Report 2007/08,
Cox RT (1961) The algebra of probable inference. Johns Pacific Engineering Research Center. University of
Hopkins Press, Baltimore California, Berkeley
Der Kiureghian A (1996) Structural reliability methods Hasofer AM, Lind NC (1974) Exact and invariant second-
for seismic safety assessment: a review. Eng Struct moment code format. J Eng Mech Div 100(1):111–121
18(6):412–424 Hastings WK (1970) Monte Carlo sampling methods
Der Kiureghian A (2005) Non‐ergodicity and PEER’s using Markov chains and their applications.
framework formula. Earthq Eng Struct Dyn Biometrika 57(1):97–109
34(13):1643–1652 Helton JC, Davis FJ (2003) Latin hypercube sampling and
Der Kiureghian A, Dakessian T (1998) Multiple design the propagation of uncertainty in analyses of complex
points in first and second-order reliability. Struct Saf systems. Reliab Eng Syst Saf 81(1):23–69
20(1):37–49 Hohenbichler M, Rackwitz R (1983) First-order concepts
Der Kiureghian A, Ditlevsen O (2009) Aleatory or episte- in system reliability. Struct Saf 1(3):177–188
mic? Does it matter? Struct Saf 31(2):105–112 Huntington DE, Lyrintzis CS (1998) Improvements to and
Der Kiureghian A, Haukaas T, Fujimura K (2006) Struc- limitations of Latin hypercube sampling. Probab Eng
tural reliability software at the University of Califor- Mech 13(4):245–253
nia, Berkeley. Struct Saf 28(1):44–67 Ibarra LF, Krawinkler H (2005) Global collapse of frame
Det Norske Veritas (1992) Structural reliability structures under seismic excitations. John A. Blume
analysis of marine structures, classification notes Earthquake Engineering Center, Report no
no 30.6 152, Department of Civil and Environmental Engi-
Ditlevsen O (1973) Structural reliability and the invari- neering, Stanford University, Stanford, 324 pp
ance problem. Research report no 22, Solid Mechanics Jalayer F (2003) Direct probabilistic seismic analysis:
Division, University of Waterloo, Waterloo implementing non-linear dynamic assessments. Ph.-
Ditlevsen O, Madsen HO (1996) Structural reliability D. dissertation, Department of Civil and Environmen-
methods. Wiley, Chichester tal Engineering, Stanford University, Stanford, 244 pp
Dolšek M (2009) Incremental dynamic analysis with con- Jalayer F, Beck JL (2008) Effects of two alternative rep-
sideration of modeling uncertainties. Earthq Eng resentations of ground‐motion uncertainty on probabi-
Struct Dyn 38(6):805–825 listic seismic demand assessment of structures. Earthq
Ebrahimian H (2012) Assessment of existing offshore Eng Struct Dyn 37(1):61–79
platforms in Persian Gulf region for life extension or Jalayer F, Cornell CA (2003) A technical framework for
operative conditions. Ph.D. dissertation, Department probability-based demand and capacity factor design
of Civil and Environmental Engineering, Sharif Uni- (DCFD) seismic formats. Pacific Earthquake Engi-
versity of Technology, Tehran, 363 pp neering Research Center (PEER) 2003/08
Jalayer F, Cornell CA (2009) Alternative non-linear Papadimitriou C, Beck JL, Katafygiotis LS

demand estimation methods for probability-based seis- (2001) Updating robust reliability using structural
mic assessments. Earthq Eng Struct Dyn 38(8):951–972 test data. Probab Eng Mech 16(2):103–113
Jalayer F, Franchin P, Pinto PE (2007a) A scalar damage Pinto PE, Giannini R, Franchin P (2007) Seismic reliabil-
measure for seismic reliability analysis of RC frames. ity of structures. IUSS Press, Pavia
Earthq Eng Struct Dyn 36(13):2059–2079 Porter K, Kennedy R, Bachman R (2007) Creating fragil-
Jalayer F, Franchin P, Pinto PE (2007b) Considering ity functions for performance-based earthquake engi-
structural modelling uncertainty in seismic reliability neering. Earthq Spectra 23(2):471–489
analysis of RC frames: use of advanced simulation Rackwitz R, Flessler B (1978) Structural reliability under
methods. In: Proceedings of the conference: combined random load sequences. Comput Struct
COMPDYN 2007, paper ID: 1218, Crete, Greece, 9(5):489–494
13–16 June 2007 Rezaeian S, Der Kiureghian A (2010) Simulation of synthetic
Jalayer F, Iervolino I, Manfredi G (2010) Structural ground motions for specified earthquake and site charac-
modeling uncertainties and their influence on seismic teristics. Earthq Eng Struct Dyn 39(10):1155–1180
assessment of existing RC structures. Struct Saf Rubinstein RY (1981) Simulation and the Monte Carlo
32(3):220–228 method. Wiley, New York
Jalayer F, De Risi R, Elefante L, Manfredi G (2013) Russell SJ, Norvig P (2003) Artificial intelligence: a modern
Robust fragility assessment using Bayesian parameter approach, 2nd edn. Prentice-Hall, Upper Saddle River
estimation. Paper presented at the Vienna Congress on Schuëller GI, Stix R (1987) A critical appraisal of methods
Recent Advances in Earthquake Engineering and to determine failure probabilities. Struct Saf
Structural Dynamics 2013 (VEESD 2013), Vienna, 4(4):293–309
28–30 Aug Shinozuka M (1983) Basic analysis of structural safety.
Jaynes ET (2003) Probability theory: the logic of science. J Struct Eng 109(3):721–740
Cambridge University Press, New York Shome N, Cornell CA (1999) Probabilistic seismic
Liel AB, Haselton CB, Deierlein GG, Baker JW demand analysis of nonlinear structures. Report no
(2009) Incorporating modeling uncertainties in the RMS-35, Department of Civil and Environmental
assessment of seismic collapse risk of buildings. Struct Engineering, Stanford University, Stanford, 320 pp
Saf 31(2):197–211 Shome N, Cornell CA, Bazzurro P, Carballo JE
Liu P-L, Der Kiureghian A (1991) Optimization algo- (1998) Earthquakes, records, and nonlinear responses.
rithms for structural reliability. Struct Saf Earthq Spectra 14(3):469–500
9(3):161–177 Stein M (1987) Large sample properties of simulations
Luco N (2002) Probabilistic seismic demand analysis, using Latin hypercube sampling. Technometrics
SMRF connection fractures, and near-source effects. 29(2):143–151
Probabilistic seismic demand analysis, SMRF connec- Straub D, Der Kiureghian A (2010) Bayesian network
tion fractures, and near-source effects. Ph.D. disserta- enhanced with structural reliability methods: method-
tion, Department of Civil and Environmental ology. J Eng Mech 136(10):1248–1258
Engineering, Stanford University, Stanford, 260 pp Tvedt L (1990) Distribution of quadratic forms in normal
McGuire RK (2004) Seismic hazard and risk analysis. space-application to structural reliability. J Eng Mech
Earthquake Engineering Research Institute, Berkeley, 116(6):1183–1197
240 pp Vamvatsikos D, Cornell CA (2004) Applied incremental
Metropolis N, Rosenbluth AW, Rosenbluth MN, Teller dynamic analysis. Earthq Spectra 20(2):523–553
AH, Teller E (1953) Equation of state calculations by Vamvatsikos D, Fragiadakis M (2010) Incremental
fast computing machines. J Chem Phys 21:1087–1092 dynamic analysis for estimating seismic performance
Moehle J, Deierlein GG (2004) A framework methodol- sensitivity and uncertainty. Earthq Eng Struct Dyn S
ogy for performance-based earthquake engineering. 39(2):141–163
In: Proceedings of the 13th world conference on earth- Vořechovský M, Novák D (2009) Correlation control in
quake engineering, Vancouver, 1–6 Aug 2004, 13 pp small-sample Monte Carlo type simulations I: a simu-
Novák D, Teplý B, Keršner Z (1998) The role of Latin lated annealing approach. Probab Eng Mech
Hypercube Sampling method in reliability engineer- 24(3):452–462
ing. In: Shiraishi N, Shinozuka M, Wen YK (eds) The Zareian F, Krawinkler H (2007) Assessment of probability
7th international conference on structural safety and of collapse and design for collapse safety. Earthq Eng
reliability, ICOSSAR 97, vol 2. Balkema, Rotterdam Struct Dyn 36(13):1901–1914
(Kyoto, Japan) Zhao Y-G, Ono T (2001) Moment methods for structural
Olsson A, Sandberg G, Dahlblom O (2003) On Latin reliability. Struct Saf 23(1):47–75
hypercube sampling for structural reliability analysis. Zuev KM, Beck JL, Au S-K, Katafygiotis LS
Struct Saf 25(1):47–68 (2012) Bayesian post-processor and other enhance-
Papadimitriou C, Beck JL, Katafygiotis LS (1997) Asymp- ments of Subset Simulation for estimating failure
totic expansions for reliability and moments of uncer- probabilities in high dimensions. Comput Struct
tain systems. J Eng Mech 123(12):1219–1229 92:283–296
2982 Seismic Resilience
“▶ Earthquakes and Their Socio-economic

Seismic Resilience Consequences”).
Contemporary cities can be interpreted as
Domenico Asprone and Gaetano Manfredi complex systems, composed of dynamic relation-
Department of Structures for Engineering and ships between physical environment, i.e., infra-
Architecture, University of Naples “Federico II”, structural systems (e.g., utility and transportation
Naples, Italy networks) and more in general all lifelines, natu-
ral environment and social environment,
consisting of communities and their internal rela-
Synonyms tionships. Hence, according to a general defini-
tion, cities can be considered resilient if able to
City resilience; Community resilience; Disaster cope with extreme events without suffering dev-
resilience; Ecosystem resilience; Engineering astating losses and damages to their physical
resilience; Urban resilience systems or reduced quality of life for the inhabi-
tants (Godschalk 2003). However, a comprehen-
sive definition is still not available, given the
Introduction complexity in defining the properties of urban
systems and the response of cities to extreme
Resilience has several meanings in academic dis- events.
course. It is derived from the Latin term resilire, What are the real operations taking place in
which means “to bounce back.” It is used in urban systems? What about the dynamic equilib-
multiple scientific contexts to identify the capa- rium at the basis of the urban system operations?
bility to recover, absorb shocks, and restore equi- What is meant by limited damages and preserva-
librium after a perturbation. First, the concept of tion of functionality for urban systems after
resilience was introduced in the nineteenth cen- extreme events? Does the optimal response of
tury in physics to indicate the ability of materials urban systems to extreme events, i.e., the “resil-
to withstand impulsive loads without suffering ient” response, depend on the type of extreme
damages. Then, resilience was also used in med- event? These are just some of the questions that
icine (Pfeiffer 1929) and psychology (Werner make the concept of resilience exploding with
1971; Garmezy 1973). different and multidisciplinary meanings.
Recently, resilience is triggering increasing
interest in other scientific contexts, referred to
communities, urban systems, and built environ- Defining Resilience
ment, as the capability to recover from natural
and human-induced disasters. The advent of the The extreme events that urban systems can be
concept of resilience in this context is the result of subjected into can be divided into four categories
an increasing need for a response to new and (O’Brien et al. 2006), and each of them may need
intense threats to modern societies. Increasing different approaches to define a resilient
interdependence and complexity of contempo- response:
rary cities along with more severe events induced
by climate change is making modern societies – Natural events, such as earthquakes, tsunamis,
ask for prevention, preparation, reduced impacts floods, etc.
and damages, and rapid recovery; that is resil- – Technological events, or human-induced
ience. This urgent need is pushing scientific events, such as accidents on transport net-
community to discuss about the best approach to works, industrial accidents, terrorist attacks,
resilience against disasters and, first of all, to etc. In developed countries, these events are
define disaster resilience (see “▶ Earthquake extensively considered in risk management
Disaster Recovery: Leadership and Governance”; strategies, aimed at public awareness of risks,
Seismic Resilience 2983
risk mitigation, and building capacity to with- of its component populations, or even the ability to
stand and recover from extreme events. maintain a steady ecological state
– Humanitarian emergencies, i.e., man-made or
• Wildavsky (1991)
natural events, hitting very vulnerable and
weak populations, such as droughts, famines, Resilience is the capacity to cope with unantici-
pated dangers after they have become manifest,
epidemics, wars, etc. learning to bounce back
– Events induced by climate changes, i.e., events
that may still be counted as natural events • Dovers and Handmer (1992)
(floods, landslides, etc.) or as humanitarian emer- Re-active and pro-active resilience of society
gencies (drought, heat waves, etc.) but that are can be distinguished, based on the major
induced by recent climate changes and can affect difference between ecosystems (that react to dis-
unprepared populations with unexpected inten- turbances) and societies (that can plan in advance,
due to human capacity for anticipation and
sity. In addition, for the climate-induced events, learning)
the most effective mitigation techniques are indi-
rect, undertaken on a global scale, through the • Adger (1997, 2000)
reduction of greenhouse gases.
Social resilience could be measured through prox-
ies of institutional change and economic structure,
Hence, resilience of urban systems against property rights, access to resources, and demo-
different events varies, and different graphic change
strategies can be implemented to build resilient • Horne and Orr (1998)
cities, in terms of risk mitigation actions, shock
Resilience is a fundamental quality of individuals,
preparedness, and recovery capability from
groups and organizations, and systems as a whole
damages. to respond productively to significant change that
Given the multiplicity of points of view to disrupts the expected pattern of events without
disaster resilience, numerous definitions of resil- engaging in an extended period of regressive
behavior
ience are available in literature; an excellent
review of these definitions is presented by Zhou • Mallak (1998)
and coworkers (Zhou et al. 2010), who elaborated
Resilience is the ability of an individual or
the most of the following list: organization to expeditiously design and
implement positive adaptive behaviors matched to
• Holling (1973, 1986), Holling et al. (1995) the immediate situation, while enduring minimal
stress
Resilience is defined as the amount of disturbance
that can be sustained by a system before a change in
system control or structure occurs. It could be mea- • Miletti (1999)
sured by the magnitude of disturbance the system can Local resiliency with regard to disasters means that
tolerate, still persisting in its pre-disturbance state a locale is able to withstand an extreme natural
S
event without suffering devastating losses, dam-
• Timmerman (1981) age, diminished productivity or quality of life and
Resilience is the ability of human communities to without a large amount of assistance from outside
withstand external shocks or perturbations to their the community
infrastructure and to recover from such perturbations
• Comfort (1999)
• Pimm (1984)
The capacity to adapt existing resources and skills
Resilience is the speed with which a system returns to new systems and operating conditions
to its original state following a perturbation
• Miletti (1999), Geis (2000), Chen et al. (2008)
• Pimm (1984), Holling et al. (1995),
Gunderson et al. (1997) In the context of disaster management, resilience is
used to describe the ability to resist or adapt to
Resilience of an ecological system relates to the stress from hazards, and the ability to recover
functioning of the system, rather than the stability quickly
• Adger (2000), Kimhi and Shamai (2004) services; and (4) economic, the capacity to reduce
both direct and indirect economic losses. Resil-
Social resilience is understood as having three ience has four dimensions: (1) robustness, strength
properties: resistance, recovery and creativity, in to withstand a given level of stress without loss
which (1) resistance relates to a social entity’s of function; (2) redundancy, the extent to which
efforts to withstand a disturbance and its conse- elements and systems are substitutable;
quences, and can be understood in terms of the (3) resourcefulness, the capacity to identify prob-
degree of disruption that can be accommodated lems, establish priorities and mobilize resources;
without social entity undergoing long-term change; and (4) rapidity, the capacity to meet priorities and
(2) Recovery relates to an entity’s ability to pull achieve goals in a timely manner. A resilient sys-
through the disturbance, and can be understood in tem has: (1) reduced probability of failures;
terms of the time taken for an entity to recover from (2) reduced consequences from failures; and
a disruption. (3) Creativity is represented by a gain (3) reduced time to recovery
in resilience achieved as part of the recovery pro-
cess, and it can be attained by adapting to new • Kendra and Wachtendorf (2003)
circumstances and learning from the disturbance
experience The ability to respond to singular or unique events
• Cardona (2003)
• Carpenter et al. (2001)
The capacity of the damaged ecosystem or com-
The Resilience Alliance consistently refers to
munity to absorb negative impacts and recover
social-ecological systems (SES) and defines their
from these
resilience by considering three distinct dimensions:
(1) the amount of disturbance a system can absorb
and still remain within the same state or domain of • Pelling (2003)
attraction; (2) the degree to which the system is The ability of an actor to cope with or adapt to
capable of self-organization; (3) the degree to hazard stress
which the system can build and increase the capac-
ity for learning and adaptation
• Rockstrom (2003)
• Paton et al. (2000) Strategies of social resilience building include
manageable strategies, such as institutional devel-
Resilience describes an active process of self- opment, land reform, land tenure, diversification,
righting, learned resourcefulness and growth—the marketing, human capacity building, and
ability to function psychologically at a level far unmanageable ones, such as relief food, cereal
greater than expected given the individual’s capa- banks, social networks, virtual water imports
bilities and previous experiences
• Rose (2004, 2007)
• UN/ISDR (2009)
Resilience includes inherent resilience (ability
The ability of a system, community or society under normal circumstances) and adaptive resil-
exposed to hazards to resist, absorb, accommodate ience (ability in crisis situations due to ingenuity
to and recover from the effects of a hazard in a or extra effort)
timely and efficient manner, including through the
preservation and restoration of its essential basic
• Aguirre (2006)
structures and functions. Resilience means the abil-
ity to “resile from” or “spring back from” a shock. A resilient social entity absorbs, responds and
The resilience of a community in respect to poten- recovers from the shock; and improvises and inno-
tial hazard events is determined by the degree to vates in response to disturbances
which the community has the necessary resources
and is capable of organizing itself both prior to and • Maguire and Hagan (2007)
during times of need.
In broad terms, social resilience is the capacity of a
• Bruneau et al. (2003) social entity (e.g., a group or community) to
bounce back or respond positively to adversity
An analysis of seismic resilience at four levels:
(1) technical, how physical systems perform when • Kang et al. (2007)
subjected to earthquake forces; (2) organizational,
the ability to respond to emergencies and carry out Resilience is the ability of the system to recover
critical functions; (3) social, the capacity to reduce once hazard has occurred and can be measured by
the negative social consequences of loss of critical the duration of an unsatisfactory condition.
• Asprone and Manfredi (2014) Furthermore, the social value of the towers,
An extreme event and the resulting changes, mov- representing a crucial symbol for the collective
ing urban systems to new dynamic equilibrium, identity of the city of New York, has been pre-
represent a phase in urban life cycle; resilience served by reconfiguring the city in a different
represents the sustainability of this phase, from dynamic equilibrium. The towers’ values still
the economic, social and environmental point of
view, for all the present and future actors, directly exist and their physical absence was recovered
and indirectly involved in the recovery process. from the social and cultural point of view. Nev-
ertheless, engineering resilience is also extremely
meaningful. In fact, one could argue that a com-
plex and dynamic system, as the city, is always
Resilience of Ecosystems and able to reach a state of equilibrium after a shock,
Engineering Resilience because the ability of cities to adapt to changes is
extremely high. But the new post-event dynamic
To apply the concept of resilience to complex equilibrium could be “worse” than the previous
systems, such as cities, two approaches can be one. In this case only with an engineering resil-
followed: the resilience of ecosystems (a) and the ience approach a “negative” response can be
engineering resilience (b). In the first, proposed appreciated; for example, quality and perfor-
and developed by Holling (1973, 1986, 2001), mance indicators of the urban system can be
resilience can be defined as the ability of a system used for this scope. Hence, aiming at synthesizing
in dynamic equilibrium, subject to external the different approaches, it can be concluded that
shocks, to move back to a dynamic equilibrium the urban system is resilient if, after a shock, it
state. On the contrary, engineering resilience, can reach a dynamic equilibrium state, even if
developed by Pimm and other authors (Pimm different from the previous one, but the urban
1984; Bruneau et al. 2003) can be defined as the system is really resilient if, at the same time,
ability of a system to absorb an external shock certain indicators of quality and performance of
and quickly return to the initial state. the system return to pre-shock values. This con-
Apparently the first definition may be more cept was also introduced by Dalziell and
complete and suitable for urban systems; in fact, McManus (2004), which affirmed the need to
moving from the fact that a complex system in introduce metrics of resilience. Hence, the ques-
dynamic equilibrium (as the urban system, which tions the scientific community is currently deal-
consists of physical and social subsystems linked ing with refer to the indicator system that should
by a dynamic network of relationships) can pre- be used. Are the indicators currently in use to
sent different equilibrium states (i.e., can “work”) assess city sustainability suitable for the scope?
in various configurations, it can be concluded that Which quality and performance indices can
a positive response to a negative external shock describe the “effectiveness” of the response to S
can be also represented by a new equilibrium external shocks?
state, different from the previous one. For exam- A further approach to the concept of resilience
ple, looking at the terrorist attack on the World of cities leads to the definition of the social resil-
Trade Center in New York on 11 September ience. It moves from the centrality of communi-
2001, it can be said that the city of New York ties in urban systems, i.e., the predominance of
had a “resilient” response. New York quickly the social over the physical system; according to
recovered from the social and economic effects this approach, social resilience is defined as the
induced by the event, even if a new equilibrium ability of communities to deal with external
was reached in a different configuration of shocks, managing the changes induced on infra-
the physical system, i.e., without rebuilding the structures, on the external environment, and on
World Trade Center towers and relocating the the economic and social systems (Adger 1997,
activities that took place there, that is, 2000). Social resilience, according to Adger, can
re-thinking that space (i.e., “ground zero” site). be measured by three characteristics: resistance
to external shocks (a), the ability to recover from The most important methodologies available
external shocks (b), and creativity (c), that is, the in literature can be divided into two categories:
ability to adapt to new circumstances (Adger (a) the physical resilience approach and (b) the
2000; Kimhi and Shamai 2004). With a similar social-economic resilience approach. In the for-
approach, Lorenz (2010) affirms that social resil- mer, attention is focused on performances of
ience consists of the adaptive capacity (a), that is, physical systems, e.g., single structures, urban
the ability to change in order to withstand the lifelines, and transportation systems. In this
external shocks; the coping capacity (b), that is, case, resilience is measured as the capability of
the ability to preserve and give continuity to the the physical components and systems to recover
system of relationships, given the external their functionality. Mainly, these methods are
shocks; and the participative capacity (c), that developed and proposed within engineering com-
is, the self-organization ability, aimed at coping munity. In the latter, attention is focused on social
with external shocks. Hence, the approach to systems, and resilience is measured as the capa-
social resilience affirms the centrality of commu- bility of communities to recover a good life qual-
nities, able to manage the other physical elements ity level. Mainly, these methods are proposed in
and determine resilient urban systems. social sciences community.
In all the approaches so far analyzed, however,
resilience is perceived as the ability of the city to
have a “positive” response, when exposed to an Physical Resilience
external shock, as an extreme event. The main
issue is the need to define a “positive” response: Bruneau et al. (2003) define resilience as charac-
the ability to return to the previous equilibrium terized by four main properties: robustness, rapid-
configuration or even a different reconfigured ity, redundancy, and resourcefulness (4 R’s), to be
equilibrium state. And, in a complex and managed and computed as proxies of resilience.
dynamic system, what is an equilibrium state? Robustness is related to the “strength, or the ability
Furthermore, is resilience to be reached sepa- of elements, systems, and other units of analysis to
rately both in the social and the physical systems, withstand a given level of stress or demand with-
or does social system resilience entail physical out suffering degradation or loss of function.”
system resilience? Thus, is the physical system Rapidity is “the capacity to meet priorities and
resilience condensed in community resilience? achieve goals in a timely manner in order to con-
Hence, is community, representing the only deci- tain losses and avoid future disruption.” Redun-
sion maker for urban management, the only mas- dancy refers to the availability of substitutable
ter of a city’s destiny, the key to a resilient city? elements or systems in the aftermath of a disrup-
tion, and resourcefulness is the capacity to mobi-
lize material and human resources. Within this
Quantifying Resilience approach, different methods have been proposed,
whose final scope is to compute resilience as the
The increasing interest in resilience requires ability to cope with degradation in system perfor-
methodological frameworks to measure and mance Q(t), over time. Numerically, resilience
assess it. Measuring disaster resilience would R is often computed as the area underneath func-
help understand and improve resilience of urban tion Q(t), divided by the time to restore the
systems against risks and implement the most pre-event performance (Fig. 1):
effective strategies to “bounce back” from disas-
ters. Aimed at this goal, different studies have ðt1
been developed, proposing measurement frame- QðtÞdt
works of disaster resilience and other properties R¼
t0
(1)
related to resilience. t1 t0
distribution of income among people are funda-

mental aspects of resilience (Adger 2000) and are
often used to compute resilience. Attitude to
System Performance (Q)
mobility and migration or amount of young peo-

ple is also related to resilience (Ruitenbeek 1996;
Adger 2000). Social memory of past changes and
r0
impacts (Olick and Robbins 1998) also relates to
the capacity of communities to adapt and cope
with disasters, that is, resilience. Hence, different
authors refer to this kind of variables to estimate
Time
community resilience, in terms of preparedness
t0 t1
and copying capacity to disasters. Specific indi-
Seismic Resilience, Fig. 1 Physical resilience cators have been also developed, moving from
social-economic variable. This is the case of the
Disaster Deficit Index (DDI), proposed by
Cardona et al. (2008), measuring country resil-
Being t0 the time of the event and t1 the time of the ience against disasters from a macroeconomic
total recovery of the pre-event performance. perspective:
This approach has been applied to buildings
(Bruneau and Reinhorn 2007), bridges (Decò LR
DDI ¼ (2)
et al. 2013), road networks (Arcidiacono RE
et al. 2012), and urban infrastructure systems
(Ouyang and Dueñas-Osorio 2012; Franchin and being LR the maximum expected direct economic
Cavalieri 2013), using different performance func- impact of possible disasters and RE the available
tions Q(t). internal and external resources that can be made
Further works are increasingly conducted on available to face disasters.
this topic and made available in literature, focus- A recent attempt to integrate physical and
ing on different urban systems and different per- social-economic perspectives of resilience has
formance functions Q(t); however, the most of been done with the PEOPLES Resilience
the works in recent literature share the Framework (Renschler et al. 2010), linking dif-
theoretical scheme in Eq. 1 to compute resilience. ferent resilience dimensions (technical, organiza-
Recently, Cavallaro et al. (2014) and Franchin tional, societal, and economic) and resilience
and Cavalieri (2013) applied this approach to properties (robustness, redundancy, resourceful-
social-physical graphs (merging physical net- ness, and rapidity) as proposed by Bruneau
works and social components) using as perfor- et al. (2003). S
mance Q(t) the efficiency of the network in the
“social” nodes, aiming at measuring the capabil-
ity of the physical systems to serve their end Summary
users.
An increasing research interest in scientific com-
munity is focusing on disaster resilience of urban
Social-Economic Resilience environment, from different scientific perspec-
tives, including those from risk engineering and
Studies aimed at computing resilience from a social sciences communities. A comprehensive
social perspective focus on economic, demo- approach to resilience is still not available, but
graphic, and institutional variables, in time and the need for a general methodology to manage
space. In example, economic growth and the and measure urban resilience is urged due to the
increasing interdependence and complexity of quantitatively assess and enhance seismic resilience
contemporary cities along with more severe of communities. Earthq Spectra 19:733–752
Cardona OD (2003) The notions of disaster risk: concep-
events induced by climate change. tual framework for integrated management. Informa-
This entry deals with the most recent tion and indicators program for disaster risk
approaches to urban disaster resilience. It moves management. Inter-American Development Bank,
from the description of a large variety of defini- Manizales
Cardona OD, Ordaz MG, Marulanda MC, Barbat AH
tions of disaster resilience, as proposed by (2008) Estimation of probabilistic seismic losses and
researchers of different disciplines. The differ- the public economic resilience – an approach for a
ences between engineering resilience and resil- macroeconomic impact evaluation. J Earthq Eng
ience of ecosystems are discussed. Thus, 12(S2):60–70
Carpenter S, Walker B, Anderies JM, Abel N (2001) From
different approaches to compute and measure metaphor to measurement: resilience of what to what?
resilience are described, focusing on physical Ecosystems (N Y, Print) 4(8):765–781
resilience, i.e., resilience of infrastructures and Cavallaro M, Asprone D, Latora V, Manfredi G, Nicosia
lifelines, and social-economic resilience, i.e., V (2014) City ecosystem resilience analysis in case of
disasters. Comput-Aided Civ Infrastruct Eng
resilience of social systems and communities. 29(8):608–625
Chen SC, Ferng JW, Wang YT, Wu TY, Wang JJ
(2008) Assessment of disaster resilience capacity of
Cross-References hillslope communities with high risk for geological
hazards. Eng Geol 98(3–4):86–101
Comfort L (1999) Shared risk: complex systems in seis-
▶ Earthquake Disaster Recovery: Leadership and mic response. Pergamon, New York
Governance Dalziell EP, McManus ST (2004) Resilience, vulnerabil-
▶ Earthquakes and Their Socio-economic ity, and adaptive capacity: implications for system
performance. Stoos, Switzerland: 1st International
Consequences Forum for Engineering Decision Making (IFED),
17 pp
Decò A, Bocchini P, Frangopol DM (2013) A probabilistic
References approach for the prediction of seismic resilience of
bridges. Earthq Eng Struct Dyn 42(10):1469–1487
Adger WN (1997) Sustainability and social resilience in Dovers SR, Handmer JW (1992) Uncertainty,
coastal resource use. CSERGE working paper series, sustainability and change. Glob Environ Chang
Centre for Social and Economic Research on the 2(4):262–276
Global Environment, University of East Anglia, Franchin P, Cavalieri F (2013) A framework for physical
Norwich and University College London simulation of critical infrastructures, accounting for
Adger WN (2000) Social and ecological resilience: are interdependencies and uncertainty. In: 11th interna-
they related? Prog Hum Geogr 24(3):347–364 tional conference on structural safety & reliability
Aguirre B (2006) On the concept of resilience. ICOSSAR, Taylor & Francis, New York
Disaster Research Center, University of Delaware, Garmezy N (1973) Competence and adaptation in adult
Delaware schizophrenic patients and children at risk. In: Dean
Arcidiacono V, Cimellaro GP, Infuso A, Reinhorn AM, SR (ed) Schizophrenia: the first ten Dean award lec-
Road network’s disaster resilience assessment meth- tures. MSS In- formation Corporation, New York
odology (2012) Bridge maintenance, safety, manage- Geis DE (2000) By design: the disaster resistant and
ment, resilience and sustainability. In: Proceedings of quality-of-life community. Nat Hazard Rev
the sixth international conference on bridge mainte- 1(3):106–120
nance, safety and management, Como (Italy), Godschalk D (2003) Urban hazard mitigation: creating
pp 1912–1919 resilient cities. Nat Hazard Rev 4:136–143
Asprone D, Manfredi G (2014) Linking disaster resilience Gunderson LH, Holling CS, Pritchard L, Peterson GD
and urban sustainability: a global approach for future (1997) Resilience in ecosystems, institutions and soci-
cities. Disasters J 39(1):s96–s111 eties. Discussion paper 95. Beijer International Insti-
Bruneau M, Reinhorn AM (2007) Exploring the concept tute of Ecological Economics, Stockholm
of seismic resilience for acute care facilities. Earthq Holling CS (1973) Resilience and stability of ecological
Spectra 23(1):41–62 systems. Annu Rev Ecol Syst 4:1–23
Bruneau M, Chang S, Eguchi R, Lee G, O’Rourke T, Holling CS (1986) The resilience of terrestrial ecosys-
Reinhorn A, Shinozuka M, Tierney K, Wallace W, tems: local surprise and global change. In: Clark WC,
von Winterfeldt D (2003) A framework to Munn RE (eds) Sustainable development of the
Seismic Response Prediction of Degrading Structures 2989
biosphere. Cambridge University Press, Cambridge, scale: the PEOPLES resilience framework. MCEER
pp 292–317 technical report –MCEER-10-006, pp 91, University
Holling CS (2001) Understanding the complexity of eco- at Buffalo (SUNY), The State University of New York,
nomic, ecological, and social system. Ecosystems Buffalo
(N Y, Print) 4:390–405 Rockstrom J (2003) Resilience building and water
Holling CS, Schindler DW, Walker BW, Roughgarden demand management for drought mitigation. Phys
J (1995) Biodiversity in the functioning of ecosystems: Chem Earth 28:869–877
an ecological synthesis. In: Perrings C, Ma¨ler KG, Rose A (2004) Defining and measuring economic resil-
Folke C, Holling CS, Jansson BO (eds) Biodiversity ience to disasters. Disaster Prev Manag 13:307–314
loss: economic and ecological issues. Cambridge Uni- Rose A (2007) Economic resilience to natural and
versity Press, Cambridge, pp 44–83 man-made disasters: multidisciplinary origins and
Horne JF, Orr JE (1998) Assessing behaviors that create contextual dimensions. Environ Hazards 7:383–398
resilient organizations. Employ Relat Today Ruitenbeek HJ (1996) Distribution of ecological entitle-
24(4):29–39 ments: implications for economic security and popu-
Kang B, Lee SJ, Kang DH, Kim YO (2007) A flood risk lation movement. Ecol Econ 17:49–64
projection for Yongdam dam against future climate Timmerman P (1981) Vulnerability, resilience and the
change. J Hydro Environ Res 1(2):118–125 collapse of society: a review of models and possible
Kendra MJ, Wachtendorf T (2003) Elements of resilience climatic applications. Institute for Environmental
after the world trade center disaster: reconstructing Studies, University of Toronto, Toronto
New York city’s emergency operation center. Disas- United Nations International Strategy for Disaster Reduc-
ters 27(1):37–53 tion (UNISDR) (2009) Global assessment report on
Kimhi S, Shamai M (2004) Community resilience and the disaster risk reduction: risk and poverty in a changing
impact of stress: adult response to Israel’s withdrawal climate. UNISDR, Geneva
from Lebanon. J Community Psychol 32(4):439–451 Werner E (1971) The children of Kauai. A longitudinal
Lorenz DF (2010) The diversity of resilience: contribu- study from the prenatal period to age ten. University of
tions from a social science perspective. Nat Hazards Hawaii Press, Honolulu
2010:1–18 Wildavsky A (1991) Searching for safety. Transaction,
Maguire B, Hagan P (2007) Disasters and communities: New Brunswick
understanding social resilience. Aust J Emerg Manag Zhou HJ, Wang JA, Wan JH et al (2010) Resilience to
22(2):16–20 natural hazards: a geographic perspective. Nat Hazards
Mallak L (1998) Resilience in the healthcare industry. In: 53(1):21–41
Paper presented at the seventh annual engineering
research conference, Banff, 9–10 May
Miletti D (1999) Disasters by design: a reassessment of
natural hazards in the United States. Joseph Henry
Press, Washington, DC
O’Brien G, O’Keefe P, Rose J, Wisner B (2006) Climate
Seismic Response Prediction of
change and disaster management. Disasters Degrading Structures
30(1):64–80
Olick JK, Robbins J (1998) Social memory studies: from Ching Hang Ng1, Nopdanai Ajavakom2 and
“collective memory” to historical sociology of mne-
monic practices. Annu Rev Sociol 24:105–140
Fai Ma3
1
Ouyang M, Dueñas-Osorio L (2012) Time-dependent US Nuclear Regulatory Commission,
resilience assessment and improvement of urban infra- Office of Nuclear Reactor Regulation, S
structure systems. Chaos 22(3):art no 033122 Washington, DC, USA
Paton D, Smith L, Violanti J (2000) Disasters response: 2
risk, vulnerabilities and resilience. Disaster Prev
Department of Mechanical Engineering,
Manag 9(3):173–179 Chulalongkorn University, Bangkok, Thailand
3
Pelling M (2003) The vulnerability of cities: natural disas- Department of Mechanical Engineering,
ters and social resilience. Earthscan, London University of California, Berkeley, Berkeley,
Pfeiffer K (1929) Untersuchungen € uber die Resilienz der
durch die Prothesen beanspruchten Gewebe und ihre
CA, USA
Bedeutung f€ur die Okklusion der Prothesen.
Berichthaus, Z€urich
Pimm SL (1984) The complexity and stability of ecosys- Synonyms
tems. Nature 307:321–326
Renschler C, Frazier A, Arendt L, Cimellaro GP,
Reinhorn AM, Bruneau M (2010) Framework for Degrading structures; Hysteresis; Nonlinear
defining and measuring resilience at the community response; System identification
2990 Seismic Response Prediction of Degrading Structures
Introduction hysteretic models of a deteriorating structure from

its experimental load–displacement traces. Sec-
All structures degrade when acted upon by cyclic ond, it will be shown that a hysteretic model
forces such as those associated with earthquakes, obtained by system identification can be used to
high winds, and sea waves. Development of a predict the future performance of the same deteri-
practical model of degrading structures that orating structure. The organization of this entry is
would match experimental observations is an as follows. In section “Differential Model of
important task. Furthermore, identification and Hysteresis,” the smoothly varying Bouc–Wen
prediction of deterioration is a problem of con- differential model of hysteresis in both its classical
siderable practical significance. Under cyclic (non-degrading) and generalized forms is
excitation, degradation manifests itself in described. A robust identification algorithm is
the evolution of the associated hysteresis constructed in section “System Identification” to
loops. Theoretical research in internal friction in generate hysteretic models of a deteriorating struc-
the last few decades has noticeably increased the ture from its experimental load–displacement
conceptual understanding of hysteresis (Kojic traces. This algorithm is based upon the general-
and Bathe 2005; Krasnoselskii and Pokrovskii ized Bouc–Wen model and differential evolution,
1989). Practical issues related to internal friction, streamlined through global sensitivity analysis.
however, have not been adequately addressed. The model thus generated can account for degra-
The lack of a practical theory of hysteretic evo- dation and pinching effects, which are prominent
lution is at present a major barrier to successful features of real-life structural deformation. To
design of structures against performance obtain experimental data for model validation,
deterioration. cyclic performance tests of simple joints and
In the past 30 years, cyclic performance test- subassemblies are reported in section “Prediction
ing of structural joints and subassemblies around of Performance.” Using an experimental
the world has generated a substantial amount of load–displacement trace, a working hysteretic
experimental data on load–displacement traces. model is identified. It will be shown that the
In the same period, the generalization of the hysteretic model obtained by identification may
Bouc–Wen differential model of hysteresis per- be used for predicting the nonlinear response of
mits curve fitting of practically any hysteretic the same structure when driven by other cyclic
trace with a suitable choice of its 13 control loads. Finally, the requirements for accurate pre-
parameters (Bouc 1967; Wen 1976). Using sys- diction of system response will be discussed. The
tem identification techniques, it appears highly terms “Bouc–Wen model” and “differential
feasible to utilize the generalized differential model” will be used interchangeably throughout
model of hysteresis and the extensive database the entry.
of experimental hysteretic traces to deduce a
working model for degrading structures.
A fundamental objective of this research project Differential Model of Hysteresis
is to do just that.
Many questions naturally spring to mind. What When a structure is subjected to severe cyclic
are the advantages of the differential model over loading, the hysteresis loops associated with the
other empirical models of hysteresis? How can an structural response are memory dependent. That
empirical model of hysteresis be used for predic- means the evolution of hysteresis loops depends
tion of nonlinear system response? What are the not only on the instantaneous deformation but
factors that would affect the precision of predic- also on the history of deformation. This memory
tion? These questions will be answered along the nature of degradation makes modeling and anal-
way. Two principal tasks in connection with hys- ysis challenging. The generalized Bouc–Wen
teretic evolution will be addressed. First, a robust model is one of the widely used empirical models
identification algorithm will be used to generate capable of modeling the memory effects.
Suppose the equation of motion of a multi- f(t)

degree-of-freedom system can be decoupled
and, along the direction of the generalized coor- x
dinate x, the system is governed by
m
x þ cx_ þ r ðx, zÞ ¼ f ðtÞ
m€ (1)
where m and c are, respectively, the mass and z

damping coefficients, z is an imaginary hysteretic
displacement, and r(x, z) is the total restoring c k
force. It is assumed that the excitation f(t) is
cyclic. In the development of differential model,
the restoring force r(x, z) is separated into an
elastic (linear) component and a hysteretic
(nonlinear) component by
r ðx, zÞ ¼ akx þ ð1 aÞkz (2) Seismic Response Prediction of Degrading Struc-

tures, Fig. 1 Schematic diagram of a hysteretic
where k is the stiffness coefficient and 0 a 1 degrading system
is a weighting parameter. Obviously, the restor-
ing force is purely hysteretic if a ¼ 0; it is purely is a pinching shape function. In general, degrada-
elastic if a ¼ 1. A diagrammatic representation of tion depends on the response duration and sever-
the system is shown in Fig. 1. ity. A convenient measure of such combined
Hysteresis loops may be generated if the hys- effect is the energy
teretic displacement z and the total displacement
x are connected by the nonlinear differential ðt
equation (Bouc 1967; Wen 1976) _
EðtÞ ¼ ð1 aÞkzxdt (5)
0
z_ ¼ Ax_ bjx_jjzjn1 z gx_jzjn (3)
dissipated through hysteresis from initial time
There are five unspecified loop parameters A, a, t ¼ 0 to present time t. Since
b, g, and n in Eqs. 2 and 3, which together repre-
sent the classical Bouc–Wen model. The param- ðt
eters A, b, and g are basic hysteresis shape _
eðtÞ ¼ z xdt (6)
parameters. The parameter n controls the sharp- 0 S
ness of yield. Over the years, the original
Bouc–Wen model has been extended, and new is proportional to E(t), it may also be used as a
parameters have been added to fit hysteretic measure of response duration and severity. Both
shapes arising from deteriorating systems. The degradation shape functions n and are assumed
result is a contemporary model with thirteen con- to depend linearly on e as the system evolves:
trol parameters given by
8 9 n ð e Þ ¼ 1 þ dn e (7)
<Ax_ n bjx_jjzjn1 z þ gx_jzjn =
z_ ¼ hðzÞ (4) ð e Þ ¼ 1 þ d e (8)
: ;
Two unspecified degradation parameters dn and
In the above expression, n and are degradation d are thus introduced. Under cyclic excitation,
shape functions (Baber and Wen 1981), and h(z) the pinching of hysteresis loops is often observed.
For example, pinching may be associated with Nonetheless, it should be noted that the differen-
slippage of longitudinal reinforcement in tial model is one-dimensional and cannot account
reinforced concrete or with X-braced steel frames for interactions between loadings in different
driven by high-shear loads. Baber and Noori directions. There are generalizations in which
(1986) introduced a slip-lock element behaving the differential model is augmented to a multi-
quite similarly to a hardening nonlinear spring, degree linear system (Roberts and Spanos 1990).
with the special characteristics that the “slip” In such generalizations, however, it is also
zone stiffness is nearly zero, while the “lock” assumed that the system could be decoupled.
zone stiffness is infinite. The “slip” zone stiffness Since no new physics is obtained, these general-
is initially zero and increases continuously as the izations would not be pursued herein. Finally, it
system degrades with time. Thus, the pinching must be emphasized that the extended Bouc–Wen
shape function h(z) takes the form (Foliente model of hysteresis is an empirical model. As
1995) such, it is not derivable from the fundamental
postulates of mechanics and the exact physical
sgnðx_Þq zu 2 =z22
meanings of its thirteen parameters are not fully
hðzÞ ¼ 1 z1 e½z (9) understood. The probable role played by each
parameter is summarized in Table 1. Also
where sgn is the signum function and zu is the contained in Table 1 are the sensitivity rankings
ultimate value of z given by (Ma et al. 2004) of the control parameters. As will
be explained in the next section, in system iden-
1=n tification, it is required to estimate only a subset
A
zu ¼ (10) of the 13 unspecified parameters.
nðb þ gÞ
The two functions z1(e) and z2(e) control the

progress of pinching and are written as System Identification
h i In analysis, the response of a system is sought if
z1 ðeÞ ¼ zs 1 eðp eÞ
(11) the system model and excitation are known. This
is sometimes termed a forward problem. In this
z 2 ð e Þ ¼ c þ dc e ð l þ z 1 Þ (12) interpretation, the inverse problem of finding a
system model given the excitation and response is
Six pinching parameters zs, q, p, c, dc, and l are called system identification. System identifica-
therefore present. Altogether there are thirteen tion in the time domain involves the determina-
loop parameters of hysteresis: A, a, b, g, n, dn, tion of unspecified parameters of an assumed
d, zs, q, p, c, dc, and l. This generalized model system model. This can always be formulated
of hysteresis possesses all the important features as an optimization problem. In the present
observed in real structures, which include context, suppose the differential model of
strength degradation, stiffness degradations, and hysteresis is adopted and a set of measured
pinching of the successive hysteresis loops. excitation–response data from cyclic perfor-
Compared with other parametric models, a mance tests of an inelastic structure is given.
differential model of hysteresis has many advan- How can the loop parameters of hysteresis be
tages (Foliente 1995; Song and Der Kiureghian estimated from the measured data? For each
2006). The primary one is its ability to generate a choice of the parameters, the response of the
large variety of realistic hysteresis loops. Another degrading structure subjected to the given exci-
advantage is the coupling of the equation of tation can be obtained by numerical simulation.
motion (1) to either Eqs. 3 or 4 to form an overall The calculated response data can then be com-
differential system. This greatly facilitates any pared to the measured data to see if there are large
theoretical and numerical manipulations. errors. Obviously, the assumed loop parameters
Seismic Response Prediction of Degrading Structures, Table 1 Parameters of the generalized differential model
of hysteresis
Local sensitivity ranking Global sensitivity ranking
Parameter Description (highest = 1) (highest = 1)
a Ratio of linear to nonlinear 1 2
response
A Basic hysteresis shape control Not varied Not varied
b Basic hysteresis shape control 5 4
g Basic hysteresis shape control 6 5
n Sharpness of yield 8 7
dn Strength degradation 12 9
d Stiffness degradation 4 8
zs Measure of total slip 2 1
q Pinching initiation 9 6
p Pinching slope 3 10
c Pinching magnitude 7 3
dc Pinching rate 11 12
l Pinching severity 10 11
provide a good fit if the errors are small. Thus, et al. 1998). In the few studies that involved
this amounts to estimating the 13 control param- degradation (Furukawa and Yagawa 1997; Sues
eters of the differential model of hysteresis when et al. 1988; Zhang et al. 2002), either a restricted
a load–displacement trace is given. The optimi- Bouc–Wen model containing less than thirteen
zation problem can be stated as the determination parameters or a restricted identification algorithm
of the parameter vector was used. For example, Zhang et al. (2002) used
gradient-based local-search algorithms to esti-

p ¼ A, a, b, g, n, dn , d , zs , q, p, c, dc , l (13) mate some of the hysteretic control parameters
for degrading structures. However, the general-
such that the objective function ized differential model of hysteresis is highly
nonlinear and any gradient-based method tends
1X N 2 to be trapped near local minima and therefore
gð p Þ ¼ x tj x^ tj jp (14)
N j¼1 fails to converge. A robust and efficient identifi-
cation algorithm is needed to estimate all thirteen
is minimized. In the above expression, tj is a unspecified parameters of the differential model
sequence of time instants and x(tj) is the given of hysteresis (Ma et al. 2006). S
system displacement at tj, where j ¼ 1, 2, , N.
On the other hand, x^ tj jp is the system displace- Reduction of Parameters
ment at tj calculated from Eqs. 1 and 4 when the In order to streamline the identification of the
parameter vector is equal to p. The chosen objec- control parameters of differential hysteresis, the
tive function is simply the mean-square error in generalized Bouc–Wen model has been
the displacement. Minimization of the objective reexamined (Ma et al. 2004). Two significant
function is subjected to the constraint that all issues have been uncovered. First, it was discov-
parameters in p with the exception of g are pos- ered that the unspecified parameters of the differ-
itive (Ma et al. 2004). ential model are functionally dependent. One of
Early studies of parametric identification of the 13 control parameters can be eliminated
hysteresis typically employed the non-degrading through suitable transformations in the parameter
classical differential model containing only five space. The number of unspecified parameters can
parameters (Kyprianou et al. 2001; Ni thus be reduced from 13 to 12 without any loss of
generality. Elimination of a parameter will appre- Nonlinear Optimization Algorithms

ciably accelerate the convergence of any identi- The goal of an optimization problem is to find a
fication algorithm. As explained by Ma vector p
in the search space S so that certain
et al. (2004), a convenient way to eliminate one quality criterion is satisfied, namely, the error
parameter is to map the parameter A into norm g(p) in Eq. 14 is minimized. The vector
1. Henceforth, A ¼ 1 will be assumed, and there p* is a solution to the minimization problem if
remain twelve loop parameters in the differential gðp
Þ is a global minimum in S. For the
model of hysteresis. constrained nonlinear optimization problem
The second issue uncovered is the existence of associated with the identification of differential
insensitive parameters in the generalized hysteresis, the error surface defined by the objec-
Bouc–Wen differential model: the variations of tive function and constraints can exhibit many
three or four control parameters in the generalized local minima or can even be multimodal. For
model do not appreciably alter the computed hys- this reason, different solution techniques will
teretic evolution of the system. Through local and have dramatically different performance. The
global sensitivity analyses (Iman and Helton 1991; primary consideration in evaluating an optimiza-
Ma et al. 2004), the twelve remaining parameters tion algorithm may be convergence speed or the
are ranked in order of decreasing sensitivity, as minimum error achieved. Secondary consider-
shown in Table 1. The method of global analysis ation may be consistency, robustness, computa-
employed is a probabilistic method recently tional efficiency, or tracking capabilities.
expounded by Sobol (1993). It can account for Solution techniques are classified into local or
the mutual interactions of the twelve parameters. population search. Basic local-search methods
On an overall basis, dn, dc, and l are the least rely on iterative descent toward a minimum in
sensitive parameters. In the identification of hys- which the direction of descent depends either on
teresis, these parameters are likely to slow down the gradient ∇gðpÞ or the Hessian matrix ∇2 gðpÞ of
numerical convergence. One extreme measure to the objective function. There are drawbacks of
deal with this problem is to set the insensitive most gradient-based techniques, which include
parameters to constants in the beginning; at any difficulties in calculating the gradient and fre-
rate they need not be estimated with high precision. quent trapping of the iterates near a local mini-
It is decided in this investigation that a two-stage mum (Bertsekas 1999). In view of that, Zhang
procedure will be adopted to streamline the identi- et al. (2002) used a non-gradient method
fication process. In the first stage, the three least based upon the Nelder and Mead simplex method
sensitive parameters dn, dc, and l are fixed, and a to minimize the objective function g(p) in Eq. 14.
crude value of the nine-parameter vector The simplex method explores the search space
S either by reflecting, contracting, or expanding
p1 ¼ a, b, g, n, d , zs , q, p, c (15) away from the worst vertex or shrinking toward
the best vertex. An appropriate sequence of such
is first estimated by minimization of the objective movements converges to the nearest local mini-
function in Eq. (14). This crude value of p1, mum. This downhill simplex method requires
together with the fixed values of dn, dc, and l, is only function evaluations and not derivatives.
then used as seeds in a second-stage identification However, even this method does not have desir-
to estimate the optimal value of the twelve- able convergence properties.
parameter vector Population-based methods, also known as
evolutionary computations, search the entire
p2 ¼ a, b, g, n, dn , d , zs , q, p, c, dc , l (16) solution space S by maintaining a group of can-
didate vectors. Evolutionary techniques are
In the identification of hysteresis, this two-stage inspired by natural evolution and adaptation
procedure economizes on both computer core with the essence of survival of the fittest. During
memory and computing time. the iterative process, new candidate vectors are
generated from existing ones by means of varia- Basic mechanisms of differential evolution can
tion and selection such that the set of candidates be described as follows:
would move toward increasingly favorable Initialization. In the beginning, the popula-
regions of the search space. Variation is achieved tion can be initialized with random values chosen
by first combining several candidate vectors to within boundaries specified by constraints if there
form a new one and then performing a random is no prior knowledge about the system. Alterna-
modification of the components of the resulting tively, a uniform probability distribution of the
vector. Selection is introduced by comparing the initial parameters may be assumed.
fitness of the candidate vectors; the smaller the Mutation. Suppose all P vectors in generation
objective function g(p), the fitter the vector p. G have been constructed. Denote these vectors by
The fitter vectors will be selected to survive and vi,G, where i ¼ 1, 2, . . . , P and every vector has
become members of a new generation. Popular dimension K. For each population member vi,G, a
evolutionary techniques include genetic algo- mutated vector w is generated by adding the
rithms (Goldberg 1989), evolution strategies weighted difference between two population
(Schwefel 1995), and differential evolution members to a third member such that
(Kyprianou et al. 2001; Lampinen and Storn
2004). Instead of performing sequential search
w ¼ vr 3 , G þ F vr 1 , G vr 2 , G (17)
as in the von Neumann architecture, evolutionary
techniques employ massively parallel schemes
with many computational elements connected In the above expression, 0 < r 1 , r 2 , r 3 P are
by links of various weights. As a consequence, three different integers chosen randomly, and
these techniques tend to be relatively robust 0 < F 1 is a preselected real-valued scaling
because the risk of being trapped near a local factor. A common variation of mutation is to
minimum is significantly reduced. The trade-off use the fittest vector vbest,G of the current gener-
is that evolutionary techniques are relatively ation in place of vr3 , G in Eq. 17.
slow compared with gradient methods. Genetic Crossover. In order to increase the diversity
algorithms require conversion of floating-point of the population, the operation of crossover
values into bit strings and are less efficient between vi,G and w is performed. First, indepen-
for problems in which the parameters are real dent uniform random numbers z1, z2, . . ., zK are
valued. For most floating-point problems, generated. With a preselected crossover constant
evolution strategies are more complicated and 0 CR 1, a trial vector u ¼ ½u1 , u2 , , uK T is
computationally involved than differential evolu- constructed by means of
tion. Differential evolution emerges as the best
evolutionary method for use in the identification wj if zj < CR
uj ¼ vi, G j otherwise (18)
of hysteretic parameters of the generalized S
Bouc–Wen model.
The constant CR determines whether each com-
Differential Evolution ponent of the vector u will originate from the
Differential evolution is a relatively new and mutated vector w or from the current vector
efficient approach for minimizing real-valued vi,G. In essence, CR controls the probability that
nonlinear and non-differentiable continuous- a crossover would occur for each component of
space functions (Price and Storn 1997). Let K be the trial vector u.
the dimension of the parameter vector p in Eq. 13. Selection. The selection process is performed
Suppose each generation in differential evolution by a one-to-one competition in which the fitness
consists of a population of P vectors. New vectors of a vector is determined by its objective value,
are generated in successive generations through which is calculated by Eq. 14 with a fourth-order
certain mechanisms such that the vectors are Runge–Kutta method for numerical integration.
getting closer and closer to the optimal value. The trial vector u is then compared to the
Generation G Generation G +1
current vector
(a) v1,G u
v1,G
v1,G
v2,G
(c) mutated vector trial vector (f)
w (e) u
v3,G (b) – F + g(u) < g(v 1,G
)?
crossover
(d) Yes
v4,G
No
(a) pick a current vector

vi,G (b) obtain the difference of two randomly chosen vectors
(c) multiply the difference by F
(d) add a third vector
vP,G (e) perform crossover
(f) decide to use u or not for next generation
Seismic Response Prediction of Degrading Structures, Fig. 2 Basic mechanisms of differential evolution
vector vi,G. If the vector u yields a smaller objec- Based upon the exposition in this section, a
tive value, it replaces vi,G as vi, Gþ1 in generation robust algorithm using the latest theory of differ-
G þ 1. Otherwise, the original vector vi,G will be ential evolution is constructed for the identifica-
kept. The process of mutation, crossover, and tion of differential hysteresis. As explained
selection of the differential evolution algorithm before, a two-stage procedure is adopted whereby
is summarized in Fig. 2. The process is repeated a crude value of the vector p1 in Eq. 15 is first
P times to construct all vectors of generation obtained before the optimal value of p2 in Eq. 16
G þ 1. As evolution progresses, new generations is computed. In simulations reported in this entry,
of vector members become fitter and fitter as the the population size P is set to 100, while both the
objective values decrease. scaling factor F and the crossover constant CR
In the differential evolution algorithm, three are set to 0.5. The computing time is dependent
control parameters P, F, and CR must be on the amount of input data used for identifica-
preselected. Lampinen and Storn (2004) indi- tion. The runtime can be reduced appreciably if
cated that convergence of differential evolution some of the insensitive parameters are fixed and
is not particularly dependent upon the choice of parallel computing is utilized.
these control parameters. It suffices to choose
F and CR as multiples of 0.1 and P as a multiple
of 10. Finally, a convergence criterion is needed Prediction of Performance
to determine if the evolution process has con-
verged. A common procedure is to set a desired With a robust and relatively efficient algorithm
objective value or to establish a threshold based devised in section “System Identification,” a dif-
on the relative fitness between the best and the ferential model of hysteresis of any degrading
worst vectors in each generation. Another system may be identified from a given
method is to require a minimum percentage of load–displacement trace. Suppose a hysteretic
improvement in fitness in order to continue the model is generated with a given load–displacement
iteration process. One may also limit the number trace. It will be shown that, under fairly broad
of generations or the computing time or both. conditions, the model predicts the response of the
Each convergence criterion has its advantages same structure when driven by other cyclic loads.
and drawbacks. Thus, a combination of conver- This can be carried out experimentally by building
gence criteria is often used. a number of identically configured structural joints
and subjecting them to different cyclic loads. One

set of load–displacement trace is chosen to gen-
erate a hysteretic model of the joint by system
identification. Subsequently, this identified
model is used to compute the displacements of
the joint under other cyclic loads. The computed
displacements will be compared with experimen-
tal measurements. A reasonable match will
validate the identified model and will demon-
strate its prediction capability. In order to reduce
the front-end costs of this investigation, only
wood joints are used in the experiments. It must
be emphasized that steel and concrete structures
may be used as easily. Although a limited set of
experimental data will be presented in this entry,
many different structural joints have been built
and tested to support any observations made
herein.
Test Setup
Cyclic tests of degrading structures are
Seismic Response Prediction of Degrading Struc-
performed in the Forest Products Laboratory tures, Fig. 3 Reaction frame test system with a specimen
located at the Richmond Field Station of the
University of California, Berkeley. Twenty iden-
tical specimens of a simple T-connection
consisting of two wood members joined by ply- Common seismic testing protocols are
wood gusset plates, as shown in Fig. 3, are employed to drive the T-connections. It is
constructed. The two wood members are made claimed that such protocols would simulate dam-
out of laminated veneer lumber (LVL), which is a age accumulation and degradation of the struc-
highly predictable engineered wood product with tural joint under real-life loads. Three of the
relatively small material variability. Each LVL cyclic excitations used in the experiments are
member is a uniform beam with a square cross shown in Figs. 4, 5, and 6. Excitation A is pro-
section of 3.5 3.5 in. and an elastic modulus of posed by the Consortium of Universities for
2.0 106 psi. The LVL lumber is sawn to con- Research in Earthquake Engineering (CUREE), S
sistent sizes and is virtually free from warping developed exclusively for wood testing
and splitting. The two LVL members are (Krawinkler et al. 2001). The CUREE protocol
connected on both sides by plywood gusset plates is designed to model ordinary ground motion
of 0.5 in. thickness. Six 2-in. metal screws are typical of most far-field locations. It is character-
arranged in three rows to fasten the plates to the ized by a pattern of a primary cycle followed by a
vertical beam. The two plates are also connected number of trailing cycles that are 75 % of the
securely to the horizontal beam with four metal previous primary cycle. Each primary cycle,
screws and two retrofit bolts. Cyclic loads are which has higher amplitude, initiates a new
applied parallel to the lower beam at a height level of damage to the structure. Damage then
36 in. above the lower beam. The arrangement accumulates and changes the structural response
is such that the T-connection will fail through in the trailing cycles. In addition, the number of
detachment of vertical LVL member from the reversed cycles is limited in order to induce fail-
gusset plates. ure modes resembling those in actual structures.
Excitation A
80
60
Displacement (mm)
40
20
0
0 100 200 300 400 500 600 700 800 900 1000
–20
–40
–60
–80 Time (sec)
Seismic Response Prediction of Degrading Structures, Fig. 4 Excitation of CUREE
Excitation B
60
40
Displacement (mm)
20
0
0 100 200 300 400 500 600 700 800 900 1000
–20
–40
–60 Time (sec)
Seismic Response Prediction of Degrading Structures, Fig. 5 Excitation of Forintek Canada Corporation
Excitation C– Reference
60
40
Displacement (mm)
20
0
0 100 200 300 400 500 600 700 800 900 1000
–20
–40
–60 Time (sec)
Seismic Response Prediction of Degrading Structures, Fig. 6 Excitation of Forest Products Laboratory
Excitation B is devised by Forintek Canada Cor- groups, each composed of three identical cycles.
poration (FCC), a national wood products The amplitude of each group is taken as a per-
research institute in Canada. Compared with centage of the nominal yield slip and is increased
excitation A, equal-amplitude primary and stepwise, with inter-spaced decreasing cycles
trailing cycles are used in this excitation. until the specimen fails. Excitation C was devel-
The FCC protocol consists of a sequence of cyclic oped at the Forest Products Laboratory for the
Seismic Response 6000

Prediction of Degrading
Structures,
Fig. 7 Hysteresis loops 4000
generated by excitation C in
500 s
2000
Load (N)
0
–40 –30 –20 –10 0 10 20 30 40
–2000
–4000
–6000
Displacement (mm)
experiments reported herein. This protocol fea- algorithm devised in the last section. Recall that
tures trailing cycles that are decreasing in ampli- A ¼ 1, and there remain twelve hysteretic param-
tude, which are different from the equal- eters to be estimated of the gusset-plate connec-
amplitude trailing cycles in the CUREE and tion. Once these twelve parameters are identified,
FCC excitations. In addition, excitation C does a hysteretic model of the T-connection is
not have the series of low-amplitude cycles fea- obtained. This hysteretic model can now be
tured in the starting interval of excitation A. used to compute the displacement history of the
gusset joint associated with excitation C. As illus-
Model Identification and Validation trated in Fig. 8, the computed displacement his-
Each specimen of the identically configured tory closely matches the experimental
T-connections is driven by a cyclic excitation measurements. This validates the identified hys-
and 1,000 s of load–displacement data are teretic model over the first 500 s.
recorded. Using any of these load–displacement A different hysteretic model of the
traces, a model of hysteretic evolution can be T-connection is generated when the duration
identified. Two issues will now be examined. of input data changes. In Fig. 9, the
(1) Can a hysteretic model identified with a load–displacement trace associated with excita-
given cyclic load predict the future structural tion C for 0 < t < 1, 000 s is shown. Compared S
response if the same cyclic load continues beyond with Fig. 7, pinching is more prominent after the
the duration used for identification? (2) Can a first 500 s. Thus, the twelve control parameters
hysteretic model identified with a given cyclic estimated with input data of Fig. 9 are different
load predict the structural response due to a dif- from those associated with Fig. 7. The new hys-
ferent cyclic load? teretic model must be validated again for a dura-
In Fig. 7, the load–displacement trace associ- tion of 1,000 s because the dominant
ated with excitation C for the first 500 s is shown. characteristics of degradation are different
As reflected in the hysteresis trace, the gusset between the two intervals 0 < t < 500 s and
joint is degrading in both strength and stiffness. 0 < t < 1, 000 s . A hysteretic model identified
There is also some pinching of the hysteretic with input data over 0 < t < 500 s will not pro-
loops, which is probably caused by slipping of duce a displacement history closely matching the
the connecting screws. The load–displacement experimental data over 500 < t < 1, 000 s . In
trace of Fig. 7 is used as input to the identification short, system identification cannot produce
50 predicted
measured
40
Displacement (mm)
30
20
10
0
0 50 100 150 200 250 300 350 400 450 500
–10
–20
–30
–40
Time (sec)
Seismic Response Prediction of Degrading Structures, Fig. 8 Comparison of computed and measured displace-
ments under excitation C
Seismic Response 8000

Structures,
Fig. 9 Hysteresis loops 6000
generated by excitation C in
1,000 s 4000
2000
Load (N)
0
–60 –40 –20 0 20 40 60
–2000
–4000
–6000
–8000
Displacement (mm)
features and trends not already contained in directly as input for identification. It can be argued
the input data. If the duration of input used for that such input data are corrupted by Gaussian
identification already contains all features of noise, and a more accurate model can be identified
hysteretic evolution, then a hysteretic model if noise is suppressed with a filter. To this end, the
identified with a given excitation can accurately input was first taken through a median filter and
predict the future response when the same exci- then a least-squares low-pass filter. Such a filtering
tation continues beyond the duration used for process is known to be effective in smoothing out
identification. This basically answers the first random fluctuations caused by Gaussian noise
question brought up at the beginning of this (Papoulis and Pillai 2002; Zhang et al. 2002). How-
section. ever, it has been found that a better match than
In identifying a hysteretic model of the what is shown in Fig. 8 is not achieved with the
T-connection, the experimental load–displacement noise-filtered data. It appears that differential evo-
trace associated with excitation C has been used lution is not sensitive to a moderate level of
80 predicted
measured
60
Displacement (mm)
40
20
0
0 200 400 600 800 1000
–20
–40
–60
–80 Time (sec)
Seismic Response Prediction of Degrading Structures, Fig. 10 Predicted and measured displacements under
excitation A in 1,000 s
80 predicted
measured
60
Displacement (mm)
40
20
0
0 200 400 600 800 1000
–20
–40
–60 Time (sec)
Seismic Response Prediction of Degrading Structures, Fig. 11 Predicted and measured displacements under
excitation B in 1,000 s
Gaussian noise. Subsequently, noise-filtering is not computed (or predicted) displacement can be
used on any of the load–displacement traces. compared with the measured displacement.
Can a hysteretic model obtained by identifica- As shown in Fig. 10, there is reasonably good
tion using a given excitation predict the structural match. A similar observation can be made with S
response due to a different cyclic load? This is the other cyclic loads. As illustrated in Fig. 11, there
second question brought up earlier. Using 1,000 s is also reasonably good match between the com-
of the load–displacement trace associated with puted (or predicted) and experimental data when
excitation C, a hysteretic model of the excitation B of Fig. 5 is used. This demonstrates
T-connection has been identified. Keep in mind that a hysteretic model identified with a given
that other cyclic loads do not play any part in the excitation may be used for response prediction
identification of hysteresis. The identified model under different excitations. However, the accu-
is then used to compute the displacement of the racy of prediction tends to diminish as time
gusset joint under excitation A of Fig. 4. How- increases.
ever, load–displacement data associated with
excitation A have already been collected experi- Requirements for Accurate Prediction
mentally using another specimen of the identi- As demonstrated in previous sections, a hyster-
cally configured gusset joints. Thus, the etic model obtained by identification can be used
Seismic Response
Structures,
Fig. 12 Failure
configuration of
T-connection under
excitation C over 1,000 s
to predict future performance of a system. Large

errors, however, can arise in many cases. For
example, the prediction of response associated
with excitation A is no longer satisfactory after
900 s, as shown in Fig. 10. Perhaps this is not
surprising. As time progresses, damage accumu-
lates and the T-connection begins to fail. There is,
however, a divergence in the failure configura-
tion due to different excitations.
As shown in Fig. 12, a typical failure config-
uration under excitation C over 1,000 s is that the
top two screws in the vertical member are appre-
ciably loosened due to wood crushing. Disassem-
bly of the gusset joint reveals that both screws
are broken into two pieces under excitation C.
If 1,000 s of the load–displacement trace associ-
ated with excitation C is used for identification, a
hysteretic model is generated that anticipates the
breaking of the top two screws. Under excitation
A, however, it has been observed that all four
screws in the top two rows are broken over
1,000 s, as shown in Fig. 13. If a model identified
with excitation C is used for response prediction
when the structure is driven by excitation A, it
fails to predict that all four screws in the top
two rows would be broken over 1,000 s. This
Seismic Response Prediction of Degrading Struc-
explains the discrepancy between the predicted tures, Fig. 13 Failure configuration of T-connection
and measured displacements under excitation under excitation A over 1,000 s
A in Fig. 10. If the failure mechanism can be cyclic load continues beyond the duration
controlled, the accuracy in response prediction used for identification.
will be increased. 3. A hysteretic model identified with a given
The observations made in experimenting with cyclic load may be used to predict the
the elementary T-connection can be extended to nonlinear response under different cyclic
complex assemblies or structures. A complex loads. As damage accumulates, the precision
structure consists of many different elementary of predicted response decreases, and the reli-
connections. Since each elementary connection ability of prediction ultimately depends on the
may have several failure configurations, the total- degree of similarity between failure configu-
ity of failure configurations is very large for a rations under different cyclic loads.
complex structure. If structural degradation can
be controlled so as to result in a failure configu- In the absence of a fundamental theory of
ration identical to what is contained in the input degradation, the response prediction of degrading
data used for identification, then nonlinear structures is indeed a challenging task. System
degrading response of a complex structure can identification of hysteretic evolution provides a
be predicted. Otherwise, prediction is only reli- brute-force procedure for prediction that has the
able up to a point at which the input data and the potential of allowing a closer representation of
existing complex structure progress into different reality. The research reported herein addresses
failure configurations. A study of the influence of various aspects of an identification methodology
failure configurations on the reliability of for predicting the performance of real-life deteri-
response prediction will be worthwhile in a sub- orating structures well beyond their linear ranges.
sequent course of investigation. Among other things, it is hoped that this entry
would point to directions along which further
research efforts should be made. This entry
Summary could also contribute to the development of a
general formulation of the mechanisms of degra-
A basic objective of this entry is to advance the dation in the long term.
methodology for predicting the performance of
nonlinear deteriorating structures. Using the gen-
eralized Bouc–Wen differential model, it is
shown how system identification can be utilized Cross-References
to predict the response of a degrading structure
well beyond its linear range. Important observa- ▶ Nonlinear System Identification: Particle-
tions reported in the paper are summarized in the Based Methods
following statements. ▶ Online Response Estimation in Structural
S
Dynamics
1. A robust identification algorithm based upon ▶ System and Damage Identification of Civil
the generalized Bouc–Wen model and the the- Structures
ory of differential evolution can be used to ▶ Vibration-Based Damage Identification: The
generate practical models of hysteresis of Z24 Bridge Benchmark
degrading structures. Differential evolution is
not sensitive to a moderate level of input References
noise.
2. If the duration of input used for identification Baber TT, Noori MN (1986) Modeling general hysteresis
already contains all features of hysteretic evo- behavior and random vibration application. ASME
J Vib Acoust Stress Reliab Des 108:411–420
lution, then a hysteretic model identified with Baber TT, Wen YK (1981) Random vibration of hysteretic
a given cyclic load can accurately predict the degrading systems. ASCE J Eng Mech
future structural response when the same 107(6):1069–1087
3004 Seismic Risk Assessment, Cascading Effects
Bertsekas DP (1999) Nonlinear programming, 2nd edn. Wen YK (1976) Method for random vibration of hyster-
Athena Scientific, Belmont etic systems. ASCE J Eng Mech 102(2):249–263
Bouc R (1967) Forced vibration of mechanical systems Zhang H, Foliente GC, Yang Y, Ma F (2002) Parameter
with hysteresis. In: Proceedings of 4th conference on identification of inelastic structures under dynamic
nonlinear oscillations, Prague, Czechoslovakia, p 315 loads. Earthq Eng Struct Dyn 31(5):1113–1130
Foliente GC (1995) Hysteretic modeling of wood joints
and structural systems. ASCE J Struct Eng
121:1013–1022
Furukawa T, Yagawa G (1997) Inelastic constitutive
parameter identification using an evolutionary algo-
rithm with continuous individuals. Int J Numer Seismic Risk Assessment, Cascading
Methods Eng 40:1071–1090 Effects
Goldberg DE (1989) Genetic algorithms in search, opti-
mization & machine learning. Addison-Wesley,
Reading Paolo Gasparini and
Iman RL, Helton JC (1991) The repeatability of uncer- Alexander Garcia-Aristizabal
tainty and sensitivity analyses for complex probabilis- Center for the Analysis and Monitoring of
tic risk assessments. Risk Anal 11:591–606
Kojic M, Bathe KJ (2005) Inelastic analysis of solids and
Environmental Risk (AMRA), Naples, Italy
structures. Springer, Berlin
Krasnoselskii MA, Pokrovskii AV (1989) Systems with
hysteresis. Springer, Berlin Synonyms
Krawinkler H, Parisi F, Ibarra L, Ayoub A, Medina
R (2001) Development of a testing protocol for
woodframe structures. CUREE-Caltech woodframe Domino effects; Short-term hazard and risk
project report W-02, Consortium of Universities for assessment; Time-dependent risk; Triggering
Research in Earthquake Engineering, Richmond effects
Kyprianou A, Worden K, Panet M (2001) Identification of
hysteretic systems using the differential evolution
algorithm. J Sound Vib 248:289–314
Lampinen J, Storn R (2004) Differential evolution. In: Introduction
Onwubolu GC, Babu BV (eds) New optimization tech-
niques in engineering. Springer, Berlin, pp 125–166
Ma F, Zhang H, Bockstedte A, Foliente GC, Paevere
Risk is conveniently assessed as a function of the
P (2004) Parameter analysis of the differential model probability that a certain event will occur and of
of hysteresis. ASME J App Mech 71:342–349 the extent of the damage caused to humans, envi-
Ma F, Ng CH, Ajavakom N (2006) On system identifica- ronment, and objects. Conceptually, it is the
tion and response prediction of degrading structures.
Struct Control Health Monit 13(1):347–364
result of the operation
Ni YQ, Ko JM, Wong CW (1998) Identification of
nonlinear hysteretic isolators from periodic vibration
tests. J Sound Vib 217:737–756 Risk ¼ Hazard Damage ¼ Hazard
(1)
Papoulis A, Pillai SU (2002) Probability, random vari- Vunerability Value at Risk
ables, and stochastic processes, 4th edn. McGraw-
Hill, New York
Price K, Storn R (1997) Differential evolution. Dr Dobb’s
where the Hazard can be understood as the prob-
J 22:18–24
Roberts JB, Spanos PD (1990) Random vibration and ability that a given adverse event of a given
statistical linearization. Wiley, New York intensity (hereinafter referred to as intensity mea-
Schwefel HP (1995) Evolution and optimum seeking. sure, IM) will occur in a certain area within a
Wiley, New York
defined time interval; the Vulnerability, a concept
Sobol IM (1993) Sensitivity estimates for nonlinear math-
ematical models. Math Model Comput Exp 1:407–414 interpreted and applied in various ways, here is
Song J, Der Kiureghian A (2006) Generalized Bouc-Wen considered in its physical dimension as the prob-
model for highly asymmetric hysteresis. ASCE J Eng able damage to an element at risk given a level of
Mech 132(6):610–618
intensity (IM) of the hazard; and the Value at Risk
Sues RH, Mau ST, Wen YK (1988) System identification
of degrading hysteretic restoring forces. ASCE J Eng term is considered as a measure of the total poten-
Mech 114:833–846 tial loss due to an adverse event in a given area.
Seismic Risk Assessment, Cascading Effects 3005
Note that both Hazard and Vulnerability are gen- Multi-risk as a Framework for the
erally defined in probabilistic terms, while Value Assessment of Cascading Effects
at Risk has some defined units of measure
(monetary cost, number of human lives, etc.). The transition from single- to multi-risk assess-
Seismic risk assessment methods have shown ment represents a process in which both the
an important development over the last 40 years complexity and the data requirements are signif-
reaching an outstanding level of acceptance by icantly increased (Fig. 1a). This transition implies
the scientific community as a valid framework for a shift from a “hazard-centered” perspective that
sound risk mitigation and decision-making. In characterizes the single-risk assessment, to a
general the seismic risk assessment uses seismic “territorial-centered” one. In fact, from a multi-
hazard information combined with fragilities of risk perspective, the first element to be defined is
structures and facilities in order to estimate the target area of interest, which is in general
the probabilities of damages and to measure terms the piece of territory composed of elements
expectancies of losses using a predefined metric at risk that are vulnerable in different ways to
of loss. various sources of hazard.
Different past disasters have highlighted the The evolution from the single-risk to the
fact that natural or man-made events can trigger multi-risk analysis consists of three different
other events, leading to a significant increase of and complementary elements: the single-risk
fatalities and damages. As a consequence, there is assessment (SR), the multi-hazard risk assess-
a growing demand from risk managers to implement (MHR), and the multi-risk assessment
ment multi-type hazard and risk assessments as (MR).
well as assessing cascading effects. Conse- Figure 1 represents the transition from the SR
quently, an increasing number of scientists study- to the MR assessment (Fig. 1a). The hazard-
ing natural hazards and risk assessment are centered perspective of the SR is represented in
turning their interest towards the multi-hazard Fig. 1b; in this case, the first element of the
risk and multi-risk assessment practice. The tran- analysis is generally the hazard source followed
sition from the single- to the multi-risk assess- by the definition of the impact area and the
ment represents a process in which both the assessment of the potential effects. The change
complexity and the data requirements are signif- towards a territorial perspective is represented in
icantly increased. This transition implies a shift Fig. 1c. In the MHR, different independent haz-
from the “hazard-centered” perspective that char- ard sources affecting a given common area of
acterizes the single-risk assessment to a interest are considered; in this case, the first ele-
“territorial-centered” one. ment of the analysis is the definition of the piece
The cascading events are one of the funda- of territory which is the target area for the analy-
mental concepts of the multi-risk assessment. It sis; the target area includes the elements at risk S
is worth noting that a different terminology is exposed to adverse events. Note that this change
used in the practice of risk evaluation when in perspective is common to both the MHR and
addressing the problem of chains of events. In the MR assessments. Once the target area has
particular, the term “domino effect” is generally been identified, the next element of the analysis
used in the field of industrial accidents, whereas is the identification of the potential hazard
the term “cascading effects” is mainly used in sources that may cause harm to the set of exposed
ecology and in natural hazards in the context of elements in the target area.
the multi-risk assessment. The objective of this Figure 1d illustrates how each element in this
entry is to discuss the role of cascading effects transition is a subset in the higher levels of the
triggered by earthquakes and its importance for a sequence. In fact, a set of SR analyses compose
holistic view of the seismic risk assessment an MHR analysis, and when the MHR assessment
process. is complemented with the analysis of interactions
Seismic Risk Assessment, Cascading Effects, Fig. 1 Representation of the transition from the single- to the multi-
risk assessment
among events, then the MR level is reached. This 5. Vulnerability and exposure assessment for
sequence is of course valid if the SR and MHR each scenario, taking into account the vulner-
analyses are done coherently in a multi-risk ability of combined hazards
framework, following some basic (and often 6. Loss estimation and multi-risk assessment
non-simple) elements of harmonization.
The multi-risk assessment (MR) is then a gen- In this framework, a set of scenarios consider-
eralization of the MHR assessment in the sense ing adverse events from different sources are
that it integrates also the possible interactions or identified along with the potential interactions.
cascading effects (Fig. 1c, d); therefore, the main Each chain of adverse events is defined in a
objective of the MR assessment level is to assess series/parallel sequence of happenings and is
the effects of possible interactions or cascading generally represented in an “event tree.” Each
effects among the different hazardous events. branch of the event tree is quantified by a proba-
Considering long- and short-term assessments, bilistic analysis considering the sequence of the
Marzocchi et al. (2012) identify the following events and the vulnerability and the exposed
main steps of the MR assessment procedure: values of the specified targets. This concept is
illustrated in Fig. 2, where the possible kinds of
1. Definition of the space-time window for the interactions are identified using discontinuous
risk assessment and the metric for evaluating red lines. For example, a sequence defined for
the risks independent hazards interacting at the vulnerabil-
2. Identification of the risks impending on the ity (or damage) level is represented by the red
selected area discontinuous line in the sequence at the left side
3. Identification of selected hazard scenarios of Fig. 2. Likewise, an example of interaction at
covering all possible intensities and relevant the hazard level is presented with source 3 trig-
hazard interactions gering source 2 (see the path defined with the red
4. Probabilistic assessment of each scenario discontinuous line at the right in Fig. 2). Note that
Seismic Risk Assessment, Cascading Effects, Fig. 2 Multi-risk framework considering possible interactions at the
hazard and the vulnerability levels (Reproduced from Marzocchi et al. 2012, with permission from Springer ®)
in this case, the interaction at the vulnerability Within this context of interactions and trigger-
level may or may not be present. Details about the ing mechanisms, earthquake-triggered events
kinds of interactions are more deeply presented in have drawn much attention in the last decades
the following sections of this entry. because of the severe damages that the triggered
events may cause. Sometimes, the losses caused possible kinds of interactions can be defined.
by the secondary events may exceed the losses Following the framework presented in
caused by the initial, triggering earthquake itself. Marzocchi et al. (2012), Garcia-Aristizabal
For example, in a number of earthquake- et al. (2013), and Garcia-Aristizabal et al.
triggered landslides, the losses caused by the (forthcoming-2015), we can consider the follow-
triggered mass movements have been compara- ing two major sets of interactions: (1) interactions
ble (if not higher) to the losses caused by the at the hazard level and (2) interactions at the
direct impact of ground shaking. An outstanding vulnerability level. This generic taxonomy of
case is the Huascaran rock-debris avalanche trig- possible interactions is qualitatively described
gered by the M7.7 Peru earthquake in 1970 (e.g., in the following paragraphs, and a more quanti-
Plafker et al. 1971), in which the impact of the tative definition is presented in the last section of
avalanche counts for more than 50 % of the total this entry.
casualties.
The kind of events that can be triggered and Interactions at the Hazard Level
the possible interactions that can be analyzed In this case the interaction problem is understood
vary in a wide spectrum of events including nat- as the assessment of possible chains of adverse
ural (as other earthquakes, landslides, tsunamis, events in which the occurrence of a given initial
volcanic unrest and eruptions, etc.) and “triggering” event entails a modification of the
nonnatural (as Natech) events. A non-exhaustive probability of occurrence of a secondary event.
description of known examples is presented in the Even if this typology of problem can be assessed
section “Examples of Cascading Effects Caused on a long-term basis, its utility can be highlighted
by Earthquakes.” in short-term problems. It can be seen that the
typology of interactions that can be grouped
under this name are in fact phenomena in which
Cascading Effects: An Overview the physical process of interest is a pure trigger-
ing mechanism in which a first occurring event
From the risk assessment point of view, the cas- produces a perturbation that, when acting over a
cades of events occurring after an earthquake given system, may bring it to unstable conditions
may represent an important source of loss ampli- forcing it to find a new equilibrium. Reaching this
fications. In this context, we consider as “loss new equilibrium may imply the occurrence of an
amplification” all the losses not caused directly event that in this case may be said as triggered by
by the earthquake (as the triggering event) but the initial triggering one.
caused by other secondary events occurring as a The link between the intensity of the trigger-
consequence of the initiating triggering event. ing event (e.g., the ground shaking caused by an
Therefore, the loss amplification can be a factor earthquake) and the intensity of the triggered
resulting from comparing the total losses caused event (e.g., a volume of mass moving in a slope)
by the chain of events triggered by an earthquake is governed by complex physical mechanisms
with the direct losses caused by the initial earth- that are intrinsically related to the specific trig-
quake itself. gering and triggered events. This fact and the
The cascading effects result as a consequence ubiquitous random effects that may affect these
of interactions generated by cause/effect relation- processes make of probabilistic approaches the
ships among different phenomena. The nature of most promising way for the quantitative charac-
these interactions may be described by a wide set terization of such interactions. Therefore,
of phenomenological relationships, and it makes assessing interactions at the hazard level in prob-
difficult to set a generalized procedure for the abilistic terms implies to quantify the conditional
quantification of cascading effects. For this rea- probabilities of occurrence of the triggered event,
son and to simplify the setting of the problem for given the occurrence of the triggering event (with
quantitative purposes, a generic taxonomy of the given intensity).
Interactions at the Vulnerability Level the conditional probabilities of having certain

In this case the problem fundamentally intends to damage state given the simultaneous or cumula-
assess the effects that the simultaneous occur- tive solicitation caused by different combinations
rence of two or more events (not necessarily of intensities of the different events considered.
linked) may have on the response of a given Note that this information can be represented in
typology of exposed elements. In this case, the different ways, according with the nature of the
process of interest is the additive effect of differ- acting events. For instance, if the acting events
ent events acting on the same elements. More are not causing cumulated damages in the same
specifically, this kind of interaction is referred elements (e.g., seismic loads from one side and
to the case in which the occurrence of one event the loading caused by snow accumulation or
(the first one occurring in time) may alter flooding water on the other), then the conditional
the response of the exposed elements against probabilities define a multidimensional fragility
another event (that may be of the same kind as surface. Conversely, if the different acting events
the former but also a different kind of hazard). produce cumulated damages in the same ele-
In this case, it is assumed that two or more events ments (i.e., when the first event pre-damages a
act on a system (e.g., a structure) and that the given element, as, e.g., two earthquakes acting
additive or cumulated effect produces a change in one after the other in a short time interval), the
the response of the system with respect to the conditional probabilities are generally referred as
conditions before the occurrence of the former time-dependent or damage-dependent fragilities
event. In practice, it implies that two or more (or vulnerabilities).
hazardous events act simultaneously or in a
short-time window (in general short enough that
the system cannot be repaired) on the same Examples of Cascading Effects Caused
structure. by Earthquakes
The cases that can be grouped under this kind
of interaction are of different nature; in general, Many examples of cascading effects caused by
the physical processes of interest are those related earthquakes can be cited as, for example, the
with the response of the system (in this case, an well-known cases of tsunamis triggered by
exposed element) to the loads caused by different earthquakes or earthquake-triggered landslides.
events taking in account their additive or cumu- As described in the previous section, there may
lated effects. Note that the different events may be different typologies of interactions, and the
be of the same nature (i.e., same kind of hazards, physical processes involved in these interactions
as, e.g., two earthquakes shaking the same struc- are extremely case dependent. In this section, we
tures in a short-time window) or coming from describe some outstanding cases of interactions
different kinds of phenomena (e.g., the shaking (mainly at the hazard level) that highlight the S
caused by an earthquake as the first event, potential loss amplifications that these triggering
followed by loads caused by river flooding or effects may produce.
strong wind). Note that the different events caus-
ing additive loads can be themselves the result of Earthquake-Earthquake Triggering
a common triggering event or may be indepen- Earthquake mainshocks cause aftershocks to
dent events, and this relationship is important for occur, which in turn activate their own subse-
the quantitative analyses. quent aftershock sequences, resulting in a cas-
As in the previous case, the quantitative cade of triggering that may increase the damage
assessment of this kind of interactions relies on caused to the exposed elements with respect to
probabilistic analyses. It may be understood as a the effects directly produced by the mainshock
multidimensional development of fragility func- alone. For this reason, contrary to the normal
tions for the different typologies of exposed ele- practice in seismic hazard and risk assessment,
ments. In general, it involves the determination of the cascade of seismic events triggered during a
seismic sequence must be of interest, at least in Considering in particular the trigger models
the short term. (Vere-Jones and Davies 1966), it is assumed
A cascade of earthquakes triggering earth- that the conditional probability of a shock occur-
quakes causes the seismicity to develop in com- ring at a time t after a given triggering mainshock
plex patterns (see, e.g., Marsan and Lengliné is proportional to a decay function l(t). In gen-
2008). Different physical mechanisms have eral, it considers that the triggering events occur
been proposed to study mainly the direct trigger- by following a Poisson process with a constant
ing problem; these include static stress changes rate, while the number of triggered events (i.e.,
(i.e., the difference in the stress field just before the group of shocks initiated by a single trigger-
an earthquake to shortly after the seismic waves ing event) decays by following a prescribed func-
have decayed, see, e.g., Hill et al. 2002; Manga tion. Regarding the nature of l(t), Vere-Jones and
and Brodsky 2006), quasi-static stress changes Davies (1966) considered in detail two functions:
(e.g., associated with slow viscous relaxation of
the lower crust and upper mantle beneath the • An exponential decay, l(t) = pept, ( p > 0)
epicenter of a large earthquake, see, e.g., Hill • An inverse power-law decay, l(t) = pc p/
et al. 2002), dynamic stress changes (i.e., induced (c + t)p + 1, (c > 0)
by the seismic waves from a large earthquake,
see, e.g., Hill et al. 2002; Manga and Brodsky (where the model parameters arise from the
2006), and fluid flow (e.g., migrating pore fluids Omori formula). Vere-Jones and Davies (1966)
and pore pressure changes as a mechanism con- suggested that the inverse power-law type of
trolling the timing of aftershocks, see, e.g., Harris decay function gives a better fit than the negative
1998 and references therein). The distinction exponential. Considering a restricted form of the
between static and dynamic stress transfer mech- trigger model, Ogata (1988) proposed the
anisms breaks down at distances less than a fault epidemic-type aftershock sequences (ETAS)
length from a given earthquake. That region, the model. ETAS is a stochastic point process in
near field, is generally the spatial domain of which each earthquake has some magnitude-
aftershock occurrences (Hill et al. 2002). dependent ability to trigger its own Omori
The evident complexity of the triggering prob- law-type aftershocks. The total occurrence rate
lem explains the objective difficulty to efficiently can be described, in time, as the superposition of
discriminate between direct and indirect trigger- a background uncorrelated seismicity m0 and the
ing in the observed data, limiting the possibility events triggered by another earthquake:
of implementing pure deterministic approaches
X
for the analysis of the events in a sequence. lðtÞ ¼ m0 þ li ðtÞ (2)
Therefore, the main efforts, as found in literature, i:ti <t
attempt to identify and separate earthquakes
between mainshocks and aftershocks, with the where li(t) is the rate of aftershocks induced by
most sophisticated methods employing stochastic an event occurred at time ti with magnitude Mi,
approaches. Our discussion in this section is defined as (Ogata 1988):
restricted to the description of stochastic model-
ing of aftershock sequences. k
li ðtÞ ¼ eaðMi Mc Þ (3)
The statistical properties of the aftershock ðc þ t ti Þp
sequences have long been of interest to study
the earthquake generation processes. It has for t > ti. The parameter k measures the produc-
resulted in the development of a number of tivity of the aftershock activity; a defines the
point process models (a type of random process relation between triggering capability and mag-
for which any realization consists of a set of nitude Mi of a triggering event; c measures the
isolated points in time or space) to describe the incompleteness of the catalog in the earliest part
standard activity of earthquake sequences. of each cluster; p controls the temporal decay of
Seismic Risk Other causes,

Assessment, Cascading 5.5%
Effects, Fig. 3 Percentage Erosion, 11.9%
of landslide events in Italy Ground water
classified by the triggering variation, 1.4%
mechanism (Source of Anthropic effects,
the data: CNR-GNDCI – 5.1%
database of areas affected
by landslides and floods in Earthquakes, 2.6%
Italy)
Precipitation and
infiltration; 73.5%
triggered events; and Mc is the completeness vulnerability level by using damage-dependent

magnitude of the data catalog. A space-time gen- fragilities.
eralization of ETAS was later introduced by
Ogata (1998). The ETAS model is widely used Earthquakes as Triggering Mechanism of
in different ways, as, for example, to describe Landslides
short-term seismicity, in the forecasting of after- Strong earthquakes may cause widespread land-
shock sequences (useful in short-term hazard slides and other ground failure phenomena as, for
studies) and also as a declustering method to example, liquefaction. A comprehensive histori-
separate the background seismicity from the cal review of outstanding past examples of
triggered one. earthquake-triggered landslides can be found in
From the perspective of considering cascading Keefer (2002) and references therein. In many
effects in the seismic risk assessment, the stochastic cases the loss of lives and properties caused by
description of the seismic sequences (in particular the landslides exceed the losses caused by the
by the ETAS model) has important consequences in earthquake itself (see, e.g., Chen et al. 2012).
probabilistic seismic hazard assessments, since it The occurrence of landslides can be associated S
permits to take into account a more realistic behav- with different triggering mechanisms as intense
ior of the seismicity to calculate aftershock hazard precipitation, erosion, ground shacking caused by
curves, which is of fundamental importance for earthquakes, groundwater variations, nonnatural
short-term hazard assessments. effects of anthropic activity, etc. For example, let
It is worth noting here that considering the us consider the database from the “National
sequences of earthquakes in a chain of triggering Group for Prevention of Hydrological Hazards”
effects can be regarded as an interaction at the of areas affected by landslides in Italy
hazard level (according to the taxonomy provided (CNR-GNDCI, http://avi.gndci.cnr.it); the most
in this entry). Nevertheless, it may require also frequent kind of events triggering landslides in
taking into account the cumulative effect of the Italy are (see Fig. 3): the precipitation and infil-
different earthquakes on the exposed elements tration (73.5 %), erosion (11.9 %), anthropic
(e.g., the physical structures), which requires to activity (5.1 %), earthquakes (2.6 %), and
consider also a possible interaction at the groundwater variation (1.4 %).
Taking as reference the available landslide the block displacement ends when the block-base
catalog in Italy, about a 3 % of the total landslide relative velocity equals zero. Displacement steps
occurrences in this area can be associated with are summed over the entire duration of the
earthquake-triggering mechanisms (Fig. 3). It is acceleration-time history to obtain the total dis-
worth noting that this example is representative placement. The method has undergone several
for the Italian area, and therefore the proportion modifications and improvements, and several
of events for the different triggering phenomena relations between seismic ground-motion param-
may considerably change from one region to the eters and computed landslide displacements have
other in function of different factors as, for exam- been proposed. A particularly interesting appli-
ple, the amount of precipitation, topography and cation for quantitative hazard assessment of
land use, size and location of earthquakes, etc. earthquake-triggered landslides is presented in
From the cascading effects assessment per- Miles and Ho (1999), in which hazard analyses
spective, the occurrence of earthquakes in are performed combining Newmark’s approach
landslide-prone areas may increase the likelihood and Boore’s stochastic ground-motion simula-
of landslide occurrence due to different physical tions. Likewise, stochastic methods for probabi-
mechanisms, as, for example, the inertial forces listic earthquake-triggered landslide hazard
during the ground shaking itself or the rapid assessment have been also proposed. For exam-
increase in pore water pressure induced by the ple, Jibson et al. (2000) performed a comparative
cyclic stresses and subsequent reduction of the analysis of the displacements modeled using the
soil shear strength. Furthermore, earthquakes Newmark’s permanent-deformation analysis and
may reduce the soil strength for longer periods, the landslide observations after the 1994
causing the affected area to be more susceptible Northridge, California earthquake and used this
to post-earthquake rainfall-triggered landslides information to construct a probability curve relat-
until the soil has regained its original strength. ing predicted displacement to probability of fail-
Several models have been proposed for ure. Likewise, Refice and Capolongo (2002)
assessing the stability of slopes during earth- presented a procedure that takes into account
quakes; an interesting overview of methods is uncertainties and the spatial variability of geo-
provided in Jibson (2011). These methods are logical, geotechnical, geomorphological, and
conveniently grouped into (1) pseudo-static ana- seismological parameters in the Newmark slope
lyses, (2) stress-deformation analyses, and stability model and applied it for probabilistic
(3) permanent displacement analyses. One of earthquake-triggered landslide hazard
the most popular models in the latter group is assessment.
that proposed by Newmark (1965), where the
landslide is modeled as a rigid block sliding on Earthquakes Triggering Volcanic Eruptions
an inclined surface. Movement of the block Deformation of the surrounding rocks is required
occurs when the sum of the static and dynamic for the magma movement before volcanic erup-
driving forces exceed the shear resistance of the tions; therefore, small to moderate local earth-
block. Permanent deformation occurs when the quakes (usually called volcano-tectonic
induced accelerations exceed some critical accel- seismicity) normally precede and accompany
eration. The sliding block model assumes (1) a eruptive activity. However, volcanic unrest epi-
rigid-plastic behavior of the boundary between sodes and volcanic eruptions have been observed
the surface and the sliding block, (2) the existence within days after the occurrence of big earth-
of a well-defined slip surface, (3) negligible loss quakes occurring outside the volcanic area. Evi-
in shear resistance during shaking, and (4) perma- dences of tectonic earthquakes preceding
nent displacement occurs only if dynamic volcanic activity can be found in different tec-
stresses exceed shear resistance. A displacement tonic setting and on different time and distance
can take place then during a cyclic load whenever scales (see, e.g., Eggert and Walter 2009;
the shear stress exceeds the base resistance, and Gasparini 2013). Well-known cases of this
possible triggering effect may include, among plausibly explain the volcanic eruption triggering
many others, the 1990 M7.8 Luzon earthquake by large earthquakes, being the most widely
and the reawakening of Mount Pinatubo in 1991 discussed those associated with transient
(e.g., Bautista et al. 1996); the 1960 M9.5 Chile (or dynamic), viscoelastic (or quasi-static), and
earthquake, which was followed about 38 h later permanent (or static) stress and strain changes.
by an eruption in the Puyehue-Cordón Caulle In fact, earthquakes can stress magmatic systems
volcano (e.g., Barrientos 1994); Mount Vesuvius either through static stresses (e.g., the offset of
having the largest eruptions shortly after earth- the fault generates a permanent deformation in
quakes in the Apennines (Nostro et al. 1998); and the crust), viscoelastic stresses (e.g., a change in
eruptions of different volcanoes in Kamchatka the stress state in the elastic crust associated with
(some of them not considered active) following slow, viscous relaxation of the lower crust and
earthquakes in different periods (e.g., Karymsky upper mantle beneath the epicenter of a large
and Dzenzur volcanoes erupting after an M8.5 earthquake), or dynamic stresses from the seis-
earthquake in 1923, the Karpinsky volcano and mic waves (Hill et al. 2002; Manga and
the Tao-Rusyr Caldera erupting few days after an Brodsky 2006).
M9.0 earthquake, and the Karymsky and Dynamic stresses are proportional to the seis-
Akademia Nauk volcanoes erupting few days mic wave amplitude, which is influenced by dif-
after a M7.1 earthquake, Walter 2007 and refer- ferent factors as directivity, radiation pattern, and
ences therein). crustal structure. However, the significant differ-
Approximately 0.4 % of explosive volcanic ence in dependence on distance is a robust feature
eruptions occur within a few days of large, distant distinguishing static and dynamic stresses. The
earthquakes, which is a value much greater than difference in decay rate helps to identify which
expected by chance (Manga and Brodsky 2006; stresses may be important for triggering erup-
Linde and Sacks 1998). All the available exam- tions. Static stress changes die off rapidly with
ples are suggestive of a triggering connection, but distance (as 1/r3, with r the distance from the
quantifying that connection is not straightfor- epicenter); therefore, their influence is limited to
ward. Linde and Sacks (1998) analyzed the his- distances of a few fault lengths (near field).
torical record of large earthquakes and explosive Quasi-static stress changes die off more slowly
eruptions and found that within a day or two of with distance (as 1/r or 1/r2), similarly to the
large earthquakes, there are many more eruptions dynamic stress changes that die off as 1/r for
within a range of about 750 km than would oth- surface waves and 1/r2 for body waves (e.g.,
erwise be expected. Therefore, they concluded Hill et al. 2002; Manga and Brodsky 2006).
that eruptions occur in the vicinity of a large To provide a comparative example, Table 1
earthquake more often than would be expected lists the order of magnitude of earthquake stresses
by chance. Additionally, it is well known that S
volcanoes separated by hundreds of kilometers Seismic Risk Assessment, Cascading Effects,
frequently erupt in unison; the characteristics of Table 1 Static and dynamic stress changes of an M8.0
such eruption pairs are also consistent with the earthquake at different distances from the source (From
Manga and Brodsky 2006)
hypothesis that the second eruption is also trig-
gered by the earthquakes associated with the first M8.0
earthquake Distance from the source
(Linde and Sacks 1998). There is also strong
Stress (MPa)
evidence that large earthquakes can affect the
102 km 103 km 104 km
unrest condition of volcanoes producing transient 1
Static 1.0 10 1.0 104 1.0 107
enhancement of microseismic and hydrothermal stress
activities as well as ground deformation (see changes
Gasparini 2013). Dynamic 3.0 6.0 102 1.0 103
As for the earthquake-earthquake interactions, stress
several mechanisms have been proposed to changes
Seismic Risk Assessment, Cascading Effects, process, examples of the models proposed
Table 2 Order of magnitude of other sources of forcing include (Hill et al. 2002):
(From Manga and Brodsky 2006)
Phenomena Stress MPa • Excitation of bubbles in a gas-saturated fluid
Solid earth tides 103 by passing seismic waves, which acts to
Ocean tides 102 increase fluid pressure through adjective over-
Hydrological loading 103 to 101
pressure and/or rectified diffusion (see, e.g.,
Glacier loading 101 to 102
Hill et al. 2002; Manga and Brodsky 2006).
• Hydraulic surge, a model in which the
at different distances from the source, and Table 2 seismogenic crust is separated from the under-
lists other external stresses that have been also lying plastic crust and an embedded magma
proposed to influence volcanoes (example from body by an impermeable transition zone; a
Manga and Brodsky 2006). In the near field, both hydraulic surge occurs when the large
static and dynamic stresses may exhibit signifi- dynamic strains associated with seismic
cant values; conversely, in the far field only the waves from a large, distant earthquake disrupt
dynamic stresses are larger than other potential the impermeable seal.
triggering stresses proposed; for instance, for dis- • Rupturing blockages in confined aquifers, in
tances greater than 103 km, the static stress which passing seismic waves disrupt block-
changes are at least one order of magnitude ages in fractures formed by precipitates in a
lower than the other forcing sources presented confined aquifer allowing local adjustments in
in Table 2. fluid pressure within the aquifer.
Several models have been proposed to explain • Sinking crystal plume model, in which rela-
the different triggering mechanisms. Here we tively dense, loosely held masses of crystals
summarize the main examples presented in Hill can accumulate on the ceiling and walls of a
et al. 2002 and in Manga and Brodsky 2006 (and crystallizing magma body and might be
references therein) for the static and dynamic dislodged by passing seismic waves. As crys-
stress triggering. The reader is invited to consult tal-rich plumes sink, less dense, crystal-poor
these references for more detailed information. magma will be forced upward initiating a con-
Static stress triggering: Static stress changes vective cycle. Volatiles in the ascending
can be a factor triggering volcanic eruptions at plume will form bubbles and a pressure
intermediate distances (few hundred kilometers). increase through advective overpressure.
Explanations for this kind of interaction are based • Relaxing magma body: A magma body may
on pressure changes in a magma body induced by reach a partially crystallized state by either
the isotropic, compressional component of the (1) gradual cooling of a previously molten
stress field in the vicinity of the volcano. volume or (2) partial melting of a previously
Increased compressional stress in the crust sur- solid volume. The solid matrix will partially
rounding an already fully charged magma cham- support accumulating tectonic stresses in the
ber may squeeze magma upward. Likewise, a region. Strong seismic waves may disrupt
decrease in compressional stress can promote the solid matrix thereby redistributing stress
vesiculation, bubble formation, and unclamping to the surrounding crust as the magma body
of conduits above the magma body (Hill deforms through viscous relaxation.
et al. 2002).
Dynamic stress interactions: Several models Different observations worldwide indicate
have been proposed to explain the phenomenon then that the triggering effect between earth-
of dynamic triggering by distant earthquakes but quakes and volcanic eruptions may exist under
the physical processes remain a matter of discus- certain conditions. Stress changes, in particular,
sion. Considering in particular those mechanisms seem to be a plausible mechanism capable of
in which fluids are active in the triggering triggering volcanic eruptions in a wide range of
distances and times. Given the complexity of the (because of damages in industrial facilities) can
problem and of the interactions, until now the be followed by other triggered hazards as, for
problem is generally treated statistically. New example, the release of hazardous materials,
observations in currently well-monitored volca- explosions, and fires.
noes, statistical significance of the correlations, A recent outstanding example of cascading
and physics of the triggering mechanisms are effects leading to industrial accidents triggered
important research topics that need to be by earthquakes is the case of the Tohoku earth-
addressed in order to better constrain this kind quake, Japan. On March 11, 2011, the Tohoku‐
of hazard interaction. Oki earthquake generated a tsunami 130 km off
the coast of Miyagi Prefecture, northeast Japan.
Earthquakes Triggering Industrial Accidents The tsunami first reached the Japanese mainland
Damages in industrial plants are known to have about 20 min after the earthquake and ultimately
triggered a number of severe accidents caused affected a 2,000 km stretch of Japan’s Pacific
mainly by the loss of containment of hazardous coast (e.g., Mori et al. 2011). The tsunami caused
substances. Industrial accidents can be triggered nuclear accidents, primarily the damages at three
by a number of causes, the most common of reactors in the Fukushima Daiichi Nuclear Power
which are internal system failures and/or human Plant complex. Accounting for all of the factors
errors (Rasmussen 1995). However, industrial relevant during the Tohoku disaster, Khazai
facilities may also be subject to accidents trig- et al. (2011) analyzed the cascading effects
gered by external factors as natural events. Ana- produced by the main earthquake and both the
lyzing different databases of industrial accidents tsunami and the nuclear power plant incidents
in Europe and the USA, Kraussmann et al. (2011) and produced a causal map depicting the direct
outlined that between 2 % and 5 % of the and indirect interactions between the various
recorded accidents were triggered by natural events.
hazards (so-called Natech accidents), being
this probably a low bound because of a
reporting bias towards accidents with severe Considering Cascading Effects in the
consequences. Seismic Risk Assessment
From the perspective of cascading effects,
earthquakes are one of the possible natural haz- Taking as reference the general concepts
ards triggering Natech accidents. Earthquakes described in Eq. 1, in more quantitative terms,
may cause damages in industrial facilities by the risk is often quantified using an expression of the
direct impact caused by the ground shaking and form (e.g., Cornell and Krawinkler 2000;
also by secondary effects as liquefaction. Consid- Marzocchi et al. 2012):
ering the results presented by Kraussmann S
et al. (2011), pipework, pipelines, and tanks for Z Z
the atmospheric storage of chemicals appear to be lð‘Þ ¼ Fð‘jDÞdGðDjI m Þ dlðI m Þ (4)
particularly vulnerable to excitation and damage D Im |fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflffl{zfflfflffl}
by the earthquake forces. In the analyzed cases, damgae & loss hazard
about 27 % of the accidents presented only struc-

tural damages, whereas about 73 % of the cases where l(l) represents a rate of loss, F(l|D) is
exhibited damage with release of hazardous the probability that a given loss is reached
materials. These results seem to suggest that given the damage state D, G(D|Im) is the vulner-
chemical equipment is much more likely to suffer ability (or fragility) term, l(Im) is the hazard,
damage with loss of containment than not measuring the exceedance rates of given intensity
(Kraussmann et al. 2011). Therefore, scenarios measures (Im) characterizing the hazard, l is the
of cascading effects with earthquakes as trigger- loss in one specific metric, and D is a given
ing mechanism causing Natech accidents damage state.
Assessing interactions, or cascading effects, in combining both kinds of approaches is applied

a multi-hazard problem consists on the identifi- in order to ensure an exhaustive exploration of
cation of the possible interactions that are likely scenarios. A wide description of practical
to happen and that may result in an amplification methods for scenario identification and struc-
of the expected losses in a given area of interest. turing can be found, for example, in
This process may be performed in the following Haimes (2009).
steps:
Taxonomy of Interactions
1. Identification and structuring of the plausible Taking as reference the works of Marzocchi
scenarios of cascading effects et al. (2012), Garcia-Aristizabal et al. (2013),
2. Identification of the kind of interactions Selva (2013), and Garcia-Aristizabal et al.
according to the proposed taxonomy (forthcoming-2015), it is possible to define a gen-
3. Quantification of the scenarios of interest eralized taxonomy of interactions considering
two possible kinds of interactions, namely,
Assuming that the value of the exposed ele- (1) interactions at the hazard level, in which the
ments remains constant in a given time window occurrence of a given initial triggering event
of interest, the amplification of the losses may entails a modification of the probability of occur-
be caused by an increase in the occurrence rence of a secondary event and (2) interaction at
probability of the hazard or by an increase in the vulnerability (or damage) level, in which the
the probability of damage by a change in the main interest is to assess the effects that the
fragility of the exposed elements. In some cases occurrence of one event (the first one occurring
both elements may be considered in a single in time) may have on the response of the exposed
scenario. In the first case, the analysis of inter- elements against another event (that may be of
actions updates the hazard term in Eq. 4; in the the same kind as the former but also a different
second case, the damage term in Eq. 4 is kind of hazard). Implicitly, a combination of both
updated. kinds of interactions is another possible case;
therefore, in the discussion of the interactions at
Identification and Structuring Scenarios of the vulnerability level, both dependent and inde-
Cascading Effects pendent hazards can be considered as possible
An important initial step towards the assessment cases.
of cascading effects is the identification and
structuring of the possible scenarios. The term Triggering Effects: Interactions at the Hazard Level
“scenario” is used in a wide range of fields and In this case, the interaction problem is understood
so different interpretations can be found in prac- as the assessment of possible “chains” of adverse
tical applications. In general, a scenario may be events in which the occurrence of a given initial
considered as a plausible and consistent represen- triggering event entails a modification of the
tation of an event or series of actions and events. occurrence probability of a secondary event.
It must be plausible because it must fall within the Even if this typology of problems can be assessed
limits of what might conceivably happen and in a long-term basis, their utility can be
must be consistent in the sense that the combined highlighted in short-term assessments, in which
logic used to construct a scenario must not have specific scenarios conditioned to the occurrence
built-in inconsistencies. of given specific events can be assessed and com-
Different strategies can be adopted in order pared among them.
to exhaustively identify a complete set of sce- Let us first consider two different threatening
narios. In general, they are based on different events, whose occurrences are E1 and E2. In gen-
complementary approaches ranging from eral, the probability of E2 occurring (generically
event-tree-like to fault-tree-like strategies. In hazard H2) can be written as (Marzocchi
many applications, an adaptive method et al. 2012):
H2 ¼ pðE2 Þ For j = 1, 2, 3, . . ., n, for the n (exhaustive and

mutually exclusive) classes of IM defined for the
¼ pðE2 jE1 ÞpðE1 Þ þ p E2 jE1 p E1 (5)
triggering event. IM1 is the intensity measure
defined for the triggering event, and IM2 is the
where, p represents a probability or a distribution intensity measure defined for the triggered event.
of probability and E1 means that the event E1 does Note that Eq. 7 is a generalization of Eq. 5. A key
not occur. Equation 5 represents the total proba- element in this formalization is p(IMi,j 2 ), which
bility of E2 occurring considering generically the we call the interaction term. When i > 1 and j >
occurrence or nonoccurrence of E1. However, the
2 ) is represented by a i j matrix that
1, p(IM i,j
quantification of risk generally requires the haz- defines a probability surface, as shown in Fig. 4.
ard assessed in function of a specific intensity When i = 1 or j = 1, p(IMi,j 2 ) is a row or column
measure (IM) characterizing the hazard, and vector, and in the hypothetical limit case in which
therefore Eq. 5 can be generalized in order to the problem does not consider the intensity mea-
consider the specific contribution of different sures of both the triggered and triggering events,
IM values. Therefore, the occurrence of the trig- the problem is simplified as the case described by
gered event 2 (E2) given the occurrence of the Eq. 5 (i.e., considering the “occurrence” and
triggering event 1 (E1) representing this kind of “nonoccurrence” of the event 1). Considering,
interaction can be assessed as (Garcia-Aristizabal for example, the earthquake-triggered landslide
et al. 2013): case, the earthquake is represented by E1 and the
X triggered landslides are the E2; IM1 can represent
p IMi2 ¼ p IMi2, j (6) the ground-motion intensity and IM2 the volume
j of the mass sliding. Therefore, p(IMi2|IMj1) rep-
resents the probability of exceeding a given slid-
that explicitly conditioned to E1 can be written as ing mass volume given a certain ground shaking
(e.g., Garcia-Aristizabal et al. 2013; Selva 2013; intensity, and p(IMi2) represents the total prob-
Garcia-Aristizabal et al. forthcoming-2015): ability of exceeding a given sliding mass volume
X considering the full range of possible ground
p IMi2 ¼ p IMi2 jIMj1 p IMj1 (7) shaking intensities taken into account from the
j triggering earthquake hazard.
Seismic Risk
Assessment, Cascading
Effects, Fig. 4 Probability
surface represented by S
p(IM2i,j ), for i > 1 and
j>1
Additive Effects of Different Events: Interactions at considering any value of the IM of the events
the Vulnerability Level 1 and 2, we need to consider two cases, namely,
This perspective of the cascading effect problem (1) when events 1 and 2 are independent and
fundamentally intends to assess the effects that (2) when there is dependence between events
the simultaneous occurrence of two or more 1 and 2.
events (not necessarily linked among them) may
have for the final risk assessment. In this case, the Independent Events In the simpler case, we
action of different hazards is considered and com- consider the two hazards as independent events,
bined at the vulnerability (or damage) level, and in the sense that the occurrence of one does not
the main interest is to assess the effects that the change the probability of the other occurring. In
occurrence of one event (the first one occurring in that case, the probability that a given damage
time) may have on the response of the exposed state is reached for all the possible values of
elements against another event (that may be of IM1 and IM2 can be defined as (e.g., Garcia-
the same kind as the former but also a different Aristizabal et al. 2013):
kind of hazard). Examples of this kind of interac-
XXh i
tions have been presented in literature, for exam- p ðD k Þ ¼ P Dk jIMi1 , IMj2 p IMi1 p IMj2
ple, in Lee and Rosowsky (2006), Zuccaro i j
et al. (2008), Marzocchi et al. (2012), Garcia- (9)
Aristizabal et al. (2013), and Selva (2013).
In practice, the idea in this case is to quantify which is an expression coherent with the frame-
how the expected damages in the target area work defined by Lee and Rosowsky (2006) ana-
(caused by a given hazard) can be modified if lyzing the effects of the combined seismic and
another hazardous event acts on the exposed ele- snow loads on wood structures. In this case, the
ments simultaneously or in a short-time window fragility function P(Dk|IMi1, IMj2) will represent a
(in general short enough that the system cannot probability surface as that shown in Fig. 5. The
be repaired). In the case of two hazards having example presented in Fig. 5 shows the combined
additive load effects (i.e., they act simultaneously effect of seismic (in pga) and volcanic ash
over the exposed element), the fragility function (in Kpa) loads acting simultaneously on roof
will depend on the intensities IM1 and IM2 of the structures in Naples, Italy. The data used to gen-
two hazards, and then it will represent a fragility erate this figure was derived from the work
surface. The probability that a given damage state presented in Zuccaro et al. (2008).
(Dk) is reached given the occurrence of the i-th
value of IM1 and the j-th value of IM2 can be Dependent Events If the occurrence of one
represented as (for details see Lee and Rosowsky
event (E1) does affect the probability of the
2006; Garcia-Aristizabal et al. 2013; Selva other occurring (E2), then the events are depen-
2013): dent. This is the case described in the interactions
at the hazard level. In this case, the term
XXh i p(IMj2 \ IMi1) = p(IMj2|IMi1)p(IMi1) and then the
pðDk Þ ¼ P Dk jIMi1 \IMj2 p IMi1 \IMj2 probability that a given damage state Dk is
i j
reached for the case of dependent events can be
(8) defined as:
XXh i
The conditional probability P(Dk|IMi1 \ IM2j ), pðDk Þ ¼ P Dk jIMj2 , IMi1 p IMj2 jIMi1 p IMi1
hereinafter referred to as P(Dk|IMi1, IMj2) for sim- i j
plicity, is the probability that the damage state Dk (10)

is reached at given levels of loads IM1 and IM2
(due to hazard events 1 and 2, respectively) acting In this case, P(Dk|IMj2, IMi1) is again a fragility
simultaneously. Therefore, to calculate p(Dk) surface as that represented in Fig. 5, and the
Seismic Risk
Assessment, Cascading
Effects, Fig. 5 Fragility
surface for the case of
seismic and volcanic ash
acting simultaneously on
the roofs in Naples, Italy
(Derived from the data in
Zuccaro et al. 2008). Note
that when the ash load tends
to zero, the fragility
function converges towards
the pure seismic fragility
hazard term takes into account the possible may just result in a modification of the response
dependence between events 1 and 2. For exam- of the exposed element (e.g., a structure) to
ple, considering again the earthquake-triggered another typology of hazard without causing a
landslide case, assessing the “collapse” damage permanent deformation or damage that remains
state would require the fragility surface P(Dk| in time. Examples of this case can be snow or
IMj2, IMi1) for assessing the probability of collapse volcanic ash deposited over a roof; this extra
given the combined effect of the ground shaking weight in the structure may cause it to behave
intensity (IM1) and, for example, the depth of the differently if it is shaken by the ground motion
deposit caused by the mass movement (IM2). On caused by an earthquake (compared with the nor-
the other hand, the hazard term in this case would mal structure’s behavior). However, in many
be calculated as shown in the example presented cases this interaction may imply that the first
for the interactions at the hazard level, in which acting event produces permanent deformations
the probability of exceeding a given IM2 value is in the structure (i.e., causing some damages)
assessed considering the full range of likely implying some modification to its response to
intensities of the triggering ground motions new solicitations and the need to develop
caused by the earthquakes. damage-dependent fragility functions. S
The most frequent development of damage-
Damage-Dependent Vulnerabilities in Cascading dependent fragility functions is found for seismic
Effects loads (i.e., to assess the structure’s response to
The damage-dependent vulnerabilities (or fragil- new earthquakes after it has been damaged by the
ities) can be considered as a special case of the past earthquakes). Different works tackling the
interaction at the vulnerability level. In fact, the problem of the response of structures
development of fragility surfaces, as presented in pre-damaged by earthquakes can be found in
the previous section for that kind of interactions, literature and mainly associated with the assess-
is intended for general cases in which the additive ment of post-mainshock time-dependent risk.
effect of different loads may act simultaneously Examples can be found in (Bazzurro et al. 2004;
over the exposed element, which does not neces- Yeo and Cornell 2005; Ryu et al. 2011; Luco
sarily imply a pre-damage of the element. For et al. 2011) and references therein. In general,
example, the action of the first occurring hazard the fragility curves used for the post-mainshock
risk assessment accounts for any damage caused without considering the effects of possible chains
by the mainshock. This typically increases the of events triggered by earthquakes. From the risk
probabilities of collapse for the considered assessment point of view, the cascades of events
range of potential (future) IM values, and the occurring after an earthquake may represent a
amount of increase depends of course on the non-negligible source of loss amplifications. For
extent of the mainshock damage, which is com- quantitative purposes, we can consider the cas-
monly discretized into damage states (DS) as, for cading effects analyzing two major sets of inter-
example, none, slight, moderate, extensive, and actions: (1) interactions at the hazard level and
complete (Luco et al. 2011). Following Ryu (2) interactions at the vulnerability level. In this
et al. (2011, and references therein), the fragility entry we present and discuss different examples
for a mainshock-damaged structure can be com- of scenarios of cascading effects triggered by
puted as: earthquakes.
pðDa > d a jIMa ¼ ima , Dm ¼ d m Þ

ð Cross-References
PðDa > d a jIMa ¼ ima , EDPm ¼ edpm Þ
¼
f ðEDPm ¼ edpm jDm ¼ d m Þdedpm ▶ Analytic Fragility and Limit States [P(EDP|
(11) IM)]: Nonlinear Dynamic Procedures
▶ Analytic Fragility and Limit States [P(EDP|
where Da represents the post-aftershock damage IM)]: Nonlinear Static Procedures
state, Dm represents the post-mainshock damage ▶ Early Earthquake Warning (EEW) System:
state, EDPm, the engineering demand parameter Overview
(e.g., drift), represents the mainshock building ▶ Earthquake Mechanism and Seafloor
response, and IMa represents the ground-motion Deformation for Tsunami Generation
intensity of an aftershock (Ryu et al. 2011). The ▶ Earthquake Mechanisms and Stress Field
first term in Eq. 11 can be computed using the ▶ Earthquake Swarms
fragility for the intact building or that for a ▶ Economic Impact of Seismic Events:
mainshock-damaged building whose mainshock Modeling
response (EDPm) is edpm. The second term in ▶ Empirical Fragility
Eq. 11 can be computed using the assumed dis- ▶ Probabilistic Seismic Hazard Models
tribution of mainshock response given post- ▶ Seismic Fragility Analysis
mainshock damage state (for details see Ryu ▶ Seismic Monitoring of Volcanoes
et al. 2011; Luco et al. 2011).
References
Summary
Barrientos SE (1994) Large thrust earthquakes and
volcanic eruptions. Pure Appl Geophys 142(1):
Different past disasters have highlighted the fact
225–237
that natural or man-made events can trigger other Bautista BC, Bautista MLP, Stein RS, Barcelona ES,
events, leading to a non-negligible increase of Punongbayan RS, Laguerta EP, Rasdas AR,
fatalities and damages. As a consequence, there Ambubuyog G, Amin EQ (1996) Relationship of
regional and local structures to Mount Pinatubo activ-
is an increasing interest on implementing multi- ity. In: Fire and mud; eruptions and lahars of Mount
type hazard and risk assessments as well as Pinatubo, Philippines. Philippine Institute of Volca-
assessing cascading effects, which are one of nology and Seismology/University of Washington
the fundamental concepts of the multi-risk Press United States, Quezon City/Philippines
Bazzurro P, Cornell CA, Menun C, Motahari M (2004)
assessment. As for most of the quantitative risk
Guidelines for seismic assessment of damaged build-
assessment for natural hazards, the seismic risk ings. In: Proceedings of the 13th world conference on
assessment is generally carried out separately earthquake engineering, Vancouver
Chen XL, Zhou Q, Ran H, Dong R (2012) Earthquake- and earthquake loading. Struct Saf 28(3):289–303.
triggered landslides in southwest China. Nat Hazards doi:10.1016/j.strusafe.2005.08.002
Earth Syst Sci 12:351–363. doi:10.5194/nhess-12- Linde AT, Sacks IS (1998) Triggering of volcanic erup-
351-2012 tions. Nature 395:888–890. doi:10.1038/27650
Cornell CA, Krawinkler H (2000) Progress and challenges Luco N, Gerstenberger MC, Uma SR, Ryu H, Liel AB,
in seismic performance assessment. PEER Center Raghunandan M (2011) A methodology for post-
News 3(2), URL http://peer.berkeley.edu/news/ mainshock probabilistic assessment of building
2000spring/index.html collapse risk. In: Proceedings of the ninth pacific con-
Eggert S, Walter TR (2009) Volcanic activity before ference on earthquake engineering building an
and after large tectonic earthquakes: observations earthquake-resilient society, Auckland
and statistical significance. Tectonophysics Manga M, Brodsky E (2006) Seismic Triggering of
471:14–26 Eruptions in the Far Field: Volcanoes and Geysers.
Garcia-Aristizabal A, Marzocchi W, Di Ruocco A (2013) Annual Review of Earth and Planetary Sciences
Probabilistic framework for multi-hazard assessment. 34:263–291. doi:10.1146/annurev.earth.34.031405.
Technical report D3.4, MATRIX project (New meth- 125125
odologies for multi-hazard and multi-risk assessment Marsan D, Lengliné O (2008) Extending earthquakes’
methods for Europe), grant no. 265138. URL http:// reach through cascading. Science 319(5866):
matrix.gpi.kit.edu/Deliverables.php 1076–1079. doi:10.1126/science.1148783
Garcia-Aristizabal A, Gasparini P, Uhinga G (forthcom- Marzocchi W, Garcia-Aristizabal A, Gasparini P,
ing-2015) Multi-risk assessment as a tool for decision- Mastellone ML, Di Ruocco A (2012) Basic principles
making. In: Pauleit S, Jorgensen G, Kabisch S, of multi-risk assessment: a case study in Italy. Nat
Gasparini P, Fohlmeister S, Simonis I, Yeshitela K, Hazards 62(2):551–573. doi:10.1007/s11069-012-
Coly A, Lindley S, Kombe WJ (eds) Future Cities, vol 0092-x
4 Climate change and urban vulnerability in Africa: a Miles SB, Ho CL (1999) Rigorous landslide hazard zona-
multidisciplinary approach. Springer Netherlands, tion using Newmark’s method and stochastic ground
ISBN 978-3-319- 03984-8 motion simulation. Soil Dyn Earthquake Eng
Gasparini P (2013) The physics of volcanoes since the 18:305–323
1960s. Rivista Nuovo Cimento 36:461–500. Mori N, Takahashi T, Yasuda T, Yanagisawa H (2011)
doi:10.1393/ncr/i2013-10093-0 Survey of 2011 Tohoku earthquake tsunami inunda-
Haimes Y (2009) Risk modeling, assessment, and man- tion and run‐up. Geophys Res Lett 38:L00G14.
agement, 3rd edn. Wiley, Hoboken doi:10.1029/2011GL049210
Harris RA (1998) Introduction to a special section: Newmark NM (1965) Effects of earthquakes on dams and
stress triggers, stress shadows, and implications embankments. Geotechnique 15(2):139–160
for seismic hazards. J Geophys Res 103: Nostro C, Stein RS, Cocco M, Belardinelli ME,
24,347–24,358 Marzocchi W (1998) Two-way coupling between
Hill DP, Pollitz F, Newhall C (2002) Earthquake-volcano Vesuvius eruptions and southern Apennine earth-
interactions. Physics Today 55(11):41–47 quakes, Italy, by elastic stress transfer. J Geophys
Jibson RW (2011) Methods for assessing the stability of Res B Solid Earth 103(B10):24487–24504
slopes during earthquakes – a retrospective. Eng Geol Ogata Y (1988) Statistical models for earthquake occur-
122:43–50 rences and residual analysis for point processes. J Am
Jibson RW, Harp EL, Michael JA (2000) A method for Stat Assoc 83:9–27
producing digital probabilistic seismic landslide haz- Ogata Y (1998) Space-time point-process models for
ard maps. Eng Geol 58(3–4):271–289. doi:10.1016/ earthquake occurrences. Ann Inst Stat Math
S0013-7952(00)00039-9 50:379–402 S
Keefer DK (2002) Investigating landslides caused by Plafker G, Ericksen GE, Concha JF (1971) Geological
earthquakes – a historical review. Surv Geophys aspects of the 31 May 1970, Peru earthquake. Bull
23:473–510 Seis Soc Am 61:543–578
Khazai B, Daniell JE, Wenzel F (2011) The March Rasmussen K (1995) Natural events and accidents with
2011 Japan earthquake: analysis of losses, impacts, and hazardous materials. J Hazard Mater 40:43–54
implications for the understanding of risks posed by Refice A, Capolongo D (2002) Probabilistic modeling of
extreme events. TECHNIKFOLGENABSCHÄTZUNG, uncertainties in earthquake-induced landslide hazard
Theorie und Praxis. N3, 20 Dec 2011. S. 22–33. http:// assessment. Comput Geosci 28(6):735–749.
www.itas.fzk.de/tatup/113/khua11a.htm doi:10.1016/S0098-3004(01)00104-2
Kraussmann E, Renni E, Campedel M, Cozzani V (2011) Ryu H, Luco N, Uma SR, Liel AB (2011) Developing
Industrial accidents triggered by earthquakes, floods fragilities for mainshock-damaged structures
and lightning: lessons learned from a database analy- through incremental dynamic analysis. In: Proceed-
sis. Nat Hazards. doi:10.1007/s11069-011-9754-3 ings of the ninth pacific conference on earthquake
Lee KH, Rosowsky DV (2006) Fragility analysis of engineering building an earthquake-resilient society,
woodframe buildings considering combined snow Auckland
3022 Seismic Robustness Analysis of Nuclear Power Plants
Selva J (2013) Long-term multi-risk assessment: statisti- early release (PSA level 2) is quantified and com-
cal treatment of interaction among risks. Nat Hazards pared to target values that are judged as low
67(2):701–722
Vere-Jones D, Davies RB (1966) A statistical survey of enough.
earthquakes in the main seismic region of New
Zealand. N Z J Geol Geophys 9(3):251–284. Consideration of Earthquakes in Nuclear
doi:10.1080/00288306.1966.10422815 Safety
Walter TR (2007) How a tectonic earthquake may wake
up volcanoes: stress transfer during the 1996 earthqua- Earthquakes are external hazards on which the
ke–eruption sequence at the Karymsky Volcanic duty holder of a nuclear facility, e.g., a nuclear
Group, Kamchatka. Earth Planet Sci Lett power plant, has no or only little control.
264(3–4):347–359 Effects to be expected at a site from an earth-
Yeo GL, Cornell CA (2005) Stochastic characterization
and decision based under time-dependent aftershock quake especially include vibrations induced in
risk in performance-based earthquake engineering. structures, systems, or components (SSC)
Technical report 2005/13. Pacific Earthquake Engi- through the civil structures of the plant. These
neering Research Center (PEER) vibrations could affect the plant safety functions
Zuccaro G, Cacace F, Spence R, Baxter P (2008) Impact
of explosive eruption scenarios at Vesuvius. directly, e.g., when the induced seismic loads
J Volcanol Geotherm Res 178(3):416–453 would exceed the capacity of safety relevant
equipment items. Indirect failure modes such as
mechanical interaction, dropped loads, release of
hazardous substances, seismic-induced fire or
flooding, impairment of operator access, or
Seismic Robustness Analysis of unavailability of evacuation and access routes
Nuclear Power Plants may also be of concern (see IAEA 2003a, §2.11).
Current design practice for a nuclear facility
Manuel Pellissetti and Ulrich Klapp ensures an adequate seismic design from
AREVA GmbH, Erlangen, Germany a deterministic perspective.
A safe shutdown earthquake (SSE) is selected
based on the results of a seismic hazard analysis
Synonyms of the site. The SSE represents a site-specific
extreme seismic event. The attribute “extreme”
Seismic margin assessment; Seismic PRA; Seis- is justified because the SSE is defined so conser-
mic probabilistic risk analysis; Seismic probabi- vatively that it can be expected with a very high
listic safety analysis; Seismic PSA; SMA level of confidence that it will not be exceeded
during the operational lifetime of the plant. SSCs
necessary to transfer the plant into a safe shut-
Introduction down state are identified and designed to remain
functional under SSE loading. Similarly, SSCs
Nuclear Safety whose seismic-induced failure could lead to
Commercial nuclear power plants (NPP) are interaction effects on safety equipment are iden-
industrial facilities that represent a significant tified and designed for integrity or stability as
hazard potential. Consequently, the pertinent required by safety analysis. The demonstration
safety requirements are particularly stringent. that the design requirements are met consists
Nuclear safety is assured by deterministic and in mechanical analysis or shaking table testing
probabilistic analysis. In deterministic analysis, of the selected systems, structures, and compo-
the safety systems are shown to withstand design nents (SSC), subject to vibrations induced by
basis accidents (DBAs), typically using conser- the SSE.
vative assumptions. In probabilistic safety analy- The described provisions ensure that even in
sis (PSA), the frequency of accident progressions the case of an extreme seismic event, represented
leading to core damage (PSA level 1) and large by the SSE, the main safety functions (reactivity
Seismic Robustness Analysis of Nuclear Power Plants 3023
control, residual heat removal, confinement of interest in studies quantifying the actual
radioactive substances) are met. margin of failure (“Seismic Margin
Assessment” – SMA) or providing an insight
Motivation for Evaluating Seismic into the risk associated to earthquakes (“Seismic
Robustness Probabilistic Safety Analysis” – SPSA, often also
What was described above is – in very simplified referred to as “Seismic Probabilistic Risk Assess-
terms – the procedure for ensuring a safe initial ment” – SPRA).
design of an NPP, as far as seismic loads are
concerned. Besides that, nuclear safety is also Seismic Hazard Assessment
concerned with ensuring continuously that it is Before introducing the two main approaches
safe to operate the plant, e.g., by means of peri- for assessing the robustness of NPPs, it is neces-
odic safety evaluations or if new insights are sary to briefly introduce the notion of seismic
gained. hazard.
Indeed, during the plant’s lifetime, new infor- The assessment of the site-specific seismic
mation becomes available which may challenge hazard – i.e., the characterization of the ground
the initial design basis of the plant. This new motions to be “expected” at a particular site – is
information may especially be of seismological a critical ingredient of both the (initial) design
nature (e.g., newly discovered seismogenic struc- against seismic loads and periodic safety reviews
tures, newly installed seismological networks, during the plant’s lifetime.
new paleoseismological evidence or actual Seismic events are unpredictable (stochastic)
occurrence of damaging earthquakes in the site as far as three of their most basic characteristics
vicinity). However, there may also be new tech- are concerned: timing, i.e., when they occur;
nical findings such as the indication that certain magnitude; and location of the epicenter. The
SSC may be more vulnerable than initially unpredictability of the latter two properties
expected (see IAEA 2009, §2.10). implies that the ground motion characteristics
In consequence the nuclear industry and reg- are also unpredictable. As in many fields of sci-
ulatory authorities are frequently faced with the ence and engineering, the difficulties associated
question whether these changes can be accom- with the unpredictability can be alleviated – to
modated within the seismic capacity of the orig- some extent – by adopting a probabilistic
inal design or whether modifications are approach: indeed the unpredictability is confined
necessary to maintain an adequate level of safety to individual realizations of the stochastic phe-
(see ASCE 2000, p. 41). nomenon, whereas ensemble properties
When assessing the safety of a nuclear facility, (statistics) are now predictable.
the plant performance in seismic events exceed- For the purpose of NPP safety, the results of
ing the ground motion characteristics of the SSE a seismic hazard assessment are essentially pairs S
(typically associated with mean return of data. Each pair consists of:
periods 10.000 years, see below) is also of
interest. See, for example, §2.39 in IAEA • One data item characterizing ground motion
(2003a): “Seismic design should be carried out (such as the peak ground acceleration, or
[. . .] to provide margins for seismic events that a response spectrum)
are beyond the design basis and to prevent poten- • One data item characterizing how rare it is to
tial small deviations in plant parameters from observe an event leading to such a ground
giving rise to severely abnormal plant behaviour motion (e.g., return period or annual probabil-
(‘cliff edge’ effects). [. . .]” ity of exceedance)
Last but not least, the severe accidents which
occurred at Japan’s Fukushima-Daiichi plants in The latter data item is typically governing the
response to the Great Tohoku Earthquake of specification of seismic loads. The annual proba-
March 2011 have stimulated an increased bility of exceedance is prescribed at an
adequately low level, which varies from country a core damage is evaluated. Core damage is
to country; typical values of this probability are mainly defined as the failure to meet the core
10-4/a and 10-5/a. The corresponding ground cooling criteria and the associated overheating
motion then represents the SSE. of the fuel rod cladding.
The need to characterize events with By taking into account the frequency of
extremely low probabilities of exceedance intro- occurrence of the characteristic ground
duces an obvious data problem, as there cannot motions – expressed by the so-called hazard
be – by definition – direct observations within the curve – and the corresponding core damage prob-
range of interest. Extensive efforts – which are far abilities, one obtains a concise measure of the
beyond the scope of this essay – have been ongo- seismic risk in the form of the seismic-induced
ing for decades in order to predict extreme seis- core damage frequency. In this quantity, the
mic events, mainly by modeling the processes robustness is implicitly contained. Indeed, for
from the source to the origin, i.e., the source a given hazard curve, a plant with a larger margin
mechanisms, the propagation (attenuation), and will have a smaller core damage frequency than
the site response. For an overview and further a plant with a smaller margin.
reading on seismic hazard refer, for example, to Conversely, the opposite can be said about the
Chen and Scawthorne (2002). The level of uncer- methods as far as the notion of risk is concerned:
tainty associated with these models – both due to in SMA the risk is contained implicitly, whereas
limited observations for calibrating the model the result of seismic PSA is an explicit expression
and due to the existence of concurrent of risk.
models – remains very high. In both approaches – SMA and seismic
This large degree of uncertainty associated PSA – the focus is on the plant-level robustness,
with seismic hazard estimates is also one of the i.e., it is essential that the safeguard systems of
main reasons for the significant interest in the the plant be able to fulfill their safety function,
seismic robustness. even if individual equipment items may have
failed. The analysis of the plant-level robust-
Two Approaches for Quantifying Seismic ness takes into account the multiple lines of
Robustness defense and the safety concepts – such as
There exist essentially two approaches for redundancy and diversity – implemented in
quantifying the robustness of NPPs to seismic- the engineered safeguards. Ultimately, how-
induced ground motions, namely, seismic margin ever, the plant-level robustness can be traced
assessment (SMA) and seismic probabilistic back to the seismic margin of individual sys-
safety analysis (seismic PSA). It should be tems, structures, and components (SSC).
noted that in the United States, the term “proba- Hence, before discussing seismic robustness
bilistic risk analysis” (PRA) is used instead in terms of plant-level safety functions, the
of PSA. concept of seismic margin is introduced for
In SMA the robustness is determined explic- individual SSC.
itly in terms of the margin between the design
acceleration, used for determining the seismic
loads on structures and equipment in the design Seismic Margin of Systems, Structures,
calculations, and the plant capacity, i.e., the limit and Components
load that the plant can at most sustain and hence
still be brought to a safe state and maintained in Load Versus Capacity
that state. Considering the deterministic approach applied
In contrast, in seismic PSA the margin is not for the seismic design of nuclear facilities,
quantified explicitly. Rather, for a comprehensive a straightforward approach to quantify seismic
range of characteristic ground motions of increas- margins would be to compare (for each important
ing severity, the corresponding probability of SSC) the load experienced by the SSC in case of
Seismic Robustness Seismic Load Load Capacity

Analysis of Nuclear at SSE (Onset of Failure)
Power Plants,
Fig. 1 Margin definition in
terms of seismic load
Margin
Load
SSE with the load capacity, i.e., with the maxi- peak spectral accelerations or the intensity on the
mum load the SSC can still handle without failing Medvedev–Sponheuer–Karnik (MSK) scale. Fre-
to meet its requirements (stability, integrity, or quently, the use of PGA as the representative
operability) (Fig. 1). parameter for expressing seismic margin and
However, a statement like “the mechanical seismic fragility is criticized, on the grounds
analysis of the control rod drive mechanism that it is a poor indicator of structural damage.
reveals a seismic margin of x kN” would not be Nevertheless, the PGA has gained the broadest
a very comprehensive expression of the existing acceptance as the characteristic margin parame-
margin, because it looks only at the quantities in ter, probably due to its simple physical interpre-
the final step of the analysis chain that needs to be tation that can be grasped also by decision makers
undergone in the seismic design of typical NPP not familiar with more sophisticated notions of
equipment. Indeed, for the standard equipment earthquake engineering or seismology (refer to
item, the mechanical analysis is preceded – at IAEA 2003b, p. 6).
least – by the analysis of the response of the It should be noted that there are ways for
floor on which the equipment is mounted. compensating the limitation of PGA as a unique
The above statement is hence an incomplete descriptor of a seismic event, either on the hazard
estimate of the margin, because it neglects addi- definition – relying on notions such as the inten-
tional margins accumulated in the previous links sity or cumulative absolute velocity (CAV; refer
of the analysis chain, such as the dynamic analy- to EPRI (2006), Campbell and Bozorgnia (2011))
sis of the civil structures. in the hazard curve development – or on the
Furthermore, it would not be a suitable basis structural response (!spectral shape factor in
for comparing the margins of different compo- the fragility definition below). S
nents relative to each other. For these reasons, it
has proved useful to define the margin in terms of Variability Due to Randomness and
one characteristic ground motion parameter. Uncertainty
The load to which an SSC is subjected during an
Expressing Margins in Terms of Ground earthquake of a certain PGA is clearly nondeter-
Motion Parameters ministic: indeed, there are an infinite number of
Typically an SSE is defined by free-field ground possible ground motions showing the same PGA,
motion response spectra. Seismic margin above each of which would induce different loads. This
the SSE is then most commonly quantified in inherently aleatory characteristic of seismic
terms of one parameter representative of the ground motions is denoted as randomness.
ground response spectra, the peak ground accel- Moreover, the majority of safety-related SSC
eration (PGA) being the most frequently used is not directly subjected to ground motion but to
parameter. Alternative parameters are, e.g., the vibrations induced through the plant’s civil
Seismic Robustness Load Demand Load Capacity

Analysis of Nuclear
Power Plants,
Probability Density
Fig. 2 Nondeterministic
character of load demand
and capacity
Load at a given PGA
structures. Thus, there is nondeterminism – both More details on the HCLPF concept, including
inherent (randomness) and epistemic the approaches for its quantification in
(uncertainty) – associated to effects like soil- practical problems, are presented in section
structure interaction, damping, load combination, “Sources of Seismic Margin” below.
etc. Finally, also the load capacity is nondeter-
ministic, considering, e.g., that material proper- Sources of Seismic Margin
ties cannot be fully controlled.
The nondeterministic character of both load In principle seismic margin could be introduced on
demand and seismic capacity is typically modeled a deterministic basis specifying design basis loads
by random variables, as illustrated in Fig. 2. exceeding those determined for the SSE or intro-
Seismic failure will occur when the load ducing penalizing stress utilization criteria. How-
demand D exceeds the capacity C; the probability ever, an overly conservative approach would
of failure PF is then induce significant extra cost in the entire life cycle
of the plant (equipment design and procurement,
P F ¼ P ðD > C Þ civil construction, maintenance of heavy equip-
ment items, decommissioning, etc.) and possibly
In view of the clearly nondeterministic nature detrimental effects on SSC performance in normal
of load demand and capacity, a conceptually plant operation. In practice there are already vari-
sound definition of margin must be formulated ous conservatisms in the seismic design of SSC.
in a probabilistic framework. A concept that has These conservatisms can be assessed and quanti-
found widespread acceptance for this purpose is fied when performing an SMA or an SPSA to show
the definition of a HCLPF (pronounced as “hay- that adequate margin is available. Seismic margin
clipf”) capacity. It is an implicit definition: if especially results from the following sources:
a component has a HCLPF capacity of x, then
the probability of failing due to a demand of value – Definition of the safe shutdown earthquake:
x is low (5 %), with a high confidence (95 %). For SSE ground motion response spectra are
the reasons stated in the previous section, the enveloping design spectra, i.e., the spectra
parameter in terms of which the HCLPF capacity have usually been subject of smoothing and
is typically defined is the PGA. broadening. In particular this introduces
Having captured the nondeterministic nature of a substantial artificial increase of energy con-
the quantities governing failure, by means of the tent of the excitation.
HCLPF capacity, it is now straightforward to – Conservatisms in the determination of the
quantify margin, e.g., as the distance between the seismic demand:
HCLPF capacity and the load demand prescribed Earthquakes induce oscillations of a chain
in the design of the NPP (i.e., the SSE). of different oscillators at a site: starting from
Seismic Robustness
Analysis of Nuclear
Power Plants,
Fig. 3 Earthquakes and
their impact on civil
structures and equipment
(Translated from Sadegh-
Azar and Hartmann 2011)
Site soil
conditions Incoming
waves(S, P)
Soil layers
Rock tion
paga
pro
ve
Wa
Hypocenter
the soil, the base plate, and the upper floors of

buildings up to smaller pieces of equipment Subsystem Excitation Response
installed in instrument racks or connected to Ground + Ground response Floor response
flexible piping (Fig. 3). These different oscil- buildings spectrum and spectra
lating subsystems show more or less signifi- corresponding (secondary
time histories spectra) and
cant interaction effects; for example, the corresponding
heavy weight of the reactor building has time histories
a damping effect on ground oscillations. Sim- Large Floor response Tertiary spectra S
ilarly heavy equipment installed inside the equipment spectra and
containment has a damping effect on the (e.g., primary (secondary corresponding
circuit), civil spectra) and time histories
vibrations of the supporting floors. A realistic substructures corresponding
description of this complex oscillation behav- time histories
ior would therefore require modeling the Equipment Tertiary spectra Quaternary
entire system and considering also nonlinear (e.g., pipes) and spectra and
corresponding corresponding
effects such as plastic deformation.
time histories time histories
Since such a global analysis is usually not Built-in Quaternary Loads
practicable, the overall system is normally split components spectra and
into several subsystems analyzed separately. (e.g., valve) corresponding
The following example illustrates this approach: time histories
Such an approach induces several sources – Application of safety factors following the
of conservatism: relevant codes.
– Excitation characteristics and physical – Use of conservative material properties
parameters considered for the different (typically 95 % fractiles are used). In case
models are chosen conservatively to ensure of civil structures, the time-dependent con-
that the conclusions made are valid crete strengthening is normally also
without high uncertainty. The resulting neglected.
level of conservatism may be moderate – Neglecting plastic deformation capability.
considering one model only. However, – Finally, seismic margin results from stress
combining several models in a calculation utilization < 1, i.e., if the resulting stresses
chain, these conservatisms are multiplied are smaller than the code allowable
and result typically in high factors of stresses. This is particular true when seis-
safety. mic loads are not the governing loads for
– Especially with respect to input character- an SSC.
istics, bounding assumptions are typically
made. For example, many relevant codes
require smoothing, broadening, or even
Quantification of Seismic Margin:
increasing of input spectra. Similarly, the
Fragility Curves and HCLPF Values
combination of load cases which can phys-
ically not occur simultaneously is often
Capacity and Fragility
applied for the sake of simplification.
In order to quantify the abovementioned sources
Finally, the orthogonal vibration compo-
of conservatism and derive an expression of the
nents are typically combined in
overall margin, the concept of seismic fragility as
a conservative manner.
introduced by Kennedy and Ravindra (1984) has
– Interaction effects and in particular
proven useful.
damping effects between the subsystems
Seismic fragility is defined as a conditional
are neglected. The resulting conserva-
failure probability, where the condition is
tisms are especially important if the sub-
represented by a seismic event with a given
structure is excited with a frequency
value of the characteristic ground motion param-
near its resonance frequency, i.e., when
eter, denoted by a. In the sequel it will be
the resulting vibrations are most
assumed that this ground motion parameter is
important.
given by the peak ground acceleration, i.e., the
– Realistic anchorage characteristics
zero period acceleration in the ground response
allowing small displacements - signifi-
spectrum. In this case the fragility is defined as
cantly influencing the vibration behavior
follows:
of the anchored system (shifting of natural
frequencies) and thus avoiding resonant
situations - are neglected. Fr ðaÞ ¼ P½failure j PGA ¼ a
– Other nonlinear effects such as plastic
deformation and friction are typically also The definition of fragility is flexible in the
not modeled, and the corresponding energy sense that the meaning of the term failure is
dissipation is neglected or considered with arbitrary. In the context of seismic PSA, it relates
simplified approaches only. mainly to components of safety-relevant systems
– Conservatisms in the determination of the or to structures in which such components are
resistance to seismic loads: installed; the entire set of these components and
These conservatisms result especially structures is abbreviated as SSC (systems, struc-
from: tures, and components). Quantitatively, failure
0 !1
occurs if one of the parameters representing the a
loads experienced by the analyzed SSC exceeds Bln ^ C
B A C
its limit state (e.g., the tensile stress in Fr ðaÞ ¼ FB
B b
C
C
a reinforcement bar exceeds the tensile strength @ R A
of the utilized steel). The definition of the limit
states for an SSC requires the identification of the
relevant failure modes. where F is the standard normal cumulative dis-
It is useful to express the fragility in terms of tribution function.
a random variable representing the capacity, In the second step of the fragility analysis, the
denoted by A, and defined in terms of the same assumption that eU = 1 is relaxed. Instead, eU is
ground motion parameter used for the fragility assumed to be log-normally distributed with log-
definition. Obviously, failure occurs if the capac- arithmic standard deviation bU.
ity is lower than the peak ground acceleration of Rewriting the capacity as
the assumed seismic event. The definition of the
^ ^
fragility can then be extended as follows:
A ¼ A eR eU ¼ AeU eR
^ ^
Fr ðaÞ ¼ P½failure j PGA ¼ a ¼ P½A < a and replacing A with A eU in the cumulative
¼ FA ðaÞ
probability density leads to
0 !1
The right portion of the above equation a
emphasizes that the fragility is equivalent to the Bln ^ C
B AeU C
cumulative density function of the capacity Fr ðaÞ ¼ FB
B b
C
C
A. This implies that defining the random capacity @ R A
A is equivalent to defining the fragility.
Log-Normal Capacity Model Due to the presence of eU, the fragility at

The most widely used model for the seismic a given PGA a itself turns also into a random
capacity is given by variable. It can be shown that for a given
probability level Q, the fractile of this random
^
A ¼ A eR eU variable is
^ 0 ! 1
where A is the median of the capacity, while eR a 1
and eU are log-normally distributed with unit Bln ^ þ bU F ðQÞC
B A C
median and logarithmic standard deviations of Fr ða, QÞ ¼ FB
B
C
C S
@ bR A
bR and bU, respectively. The random variables
eR and eU model the variability due to randomness
and due to uncertainty, respectively. In order to
preserve the distinction between these two Interpreting Q as a subjective probability, or
different sources of variability, the variables eR equivalently as a confidence level, one may use
and eU are introduced in two distinct steps of the the above equation to obtain the fragility curves
analysis. To begin with, it is assumed that associated with specific confidence levels. This is
only the variability to randomness exists, demonstrated in^Fig. 4 for the fragility model with
whereas there is no variability due to uncertainty, the parameters A ¼ 0:5 g, bR = 0.3, and bU = 0.4.
i.e., eU =1. The median (50 % confidence level) fragility
At this stage, the fragility is then (due to the curve is identical to the one obtained by
properties of the log-normal distribution of eR) neglecting uncertainty (i.e., bU = 0).
Seismic Robustness 1.0

Analysis of Nuclear
Power Plants, 0.9
Fig. 4 Example of
0.8
fragility curve in linear
scale 0.7
0.6
F [ −]
0.5
0.4
0.3
0.2
0.1
0.0
0 0.2 0.4 0.6 0.8 1
a [g]
Mean 5% 50 % 95 %
The mean fragility curve is obtained by aver- curves is based on separation of variables, as
aging over the random fragility FA(a) for each described in detail in Kennedy and Ravindra
value of the ground motion parameter a, i.e., the (1984) and in EPRI (1994).
averaging is done along the vertical direction. The basic concept underlying this approach is
It can be shown that the mean fragility has the the following expression for the capacity A as
following simple form a product of the (deterministic) value of the
!1 PGA, aSSE, adopted in the seismic design of
0
a the SSC under consideration, and the scaling
Bln ^ C qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi factor F:
B A C
FA ð a Þ ¼ F B
B b
C
C bC ¼ bR 2 þ bU 2
@ C A A ¼ aSSE F
The scaling factor F may hence be viewed as

The above model is called the composite fra- the maximum scalar, by which the design ground
gility model, because it considers random vari- motion can be multiplied (“scaled”) without pro-
ability and variability due to uncertainty in ducing failure. Alternatively, the factor F may
a compact form. It is used for point estimates however also be viewed as a safety factor, con-
(best estimate) of the seismic risk. sidering that it is the ratio between the actual
capacity A and the acceleration used in the
Methods for Deriving Fragility Curves of SSC design, aSSE:
For the quantification of seismic fragility, one can
generally distinguish between approaches relying
A
on the assessment of conservatisms in existing F¼
seismic verification studies and approaches aSSE
requiring new dynamic analyses.
For cost reasons, where possible, preference is It should be noted that the scaling (safety)
given to the former. Within this category, the factor F is a random variable, since the capacity
most widely used approach for deriving fragility A is also a random variable.
The essence of the separation of variables FRS ¼ FSA FGMI Fd FM FMC FEC FSSI
approach is to break down the scaling (safety)
factor F into a product of “partial” safety factors. where:
This factorization process extends over two
levels, since some of these partial safety factors FSA = spectral shape factor representing bias and
are yet again factorized. variability in ground motion and associated
For structures, the factor of safety F can be ground response spectra
modeled as the product of three random FGMI = ground motion incoherence factor that
variables: accounts for the fact that a traveling seismic
wave does not excite a large foundation
F ¼ FS Fm FRS uniformly
Fd = damping factor representing bias and vari-
where FS is the strength factor, Fm is the inelastic ability in response due to the difference
energy absorption factor, and FRS is the structural between actual damping and damping
response factor. assumed in design
The strength factor, FS, represents the ratio of FM = modeling factor accounting for limitations
ultimate strength (or strength at loss of function) regarding the resolution and calibration of the
to the stress calculated for aSSE. In calculating the model adopted for representing the structural
value of FS, the non-seismic portion of the total mechanics
load acting on the structure is subtracted from the FMC = mode combination factor accounting for
strength as follows: bias and variability in the estimated response
due to the method used in combining dynamic
S PN modes of response
FS ¼
PT PN FEC = earthquake component combination factor
accounting for bias and variability in response
where S is the strength of the structural element due to the method used in combining earth-
for the specific failure mode, PN is the normal quake components
operating load (i.e., dead load, operating temper- FSSI = factor that accounts for the effect of soil-
ature load, etc.), and PT is the total load on the structure interaction
structure (i.e., sum of the seismic load – due to
aSSE – and the normal operating load). Similarly, for equipment and other compo-
The inelastic energy absorption factor nents, the overall factor of safety is composed
(ductility factor), Fm, accounts for the fact that of a capacity factor, FC; a structure response
an earthquake represents a limited energy source, factor, FRS; and an equipment response (relative
and many structures or equipment items are capa- to the structure) factor, FRE. Thus, S
ble of absorbing substantial amounts of energy
beyond yield without loss of function. F ¼ FC FRE FRS
The structural response factor, FRS, is based on
the recognition of the fact that in the design The individual safety factors Fi are assumed to
analyses, structural response is computed using follow a log-normal distribution. Their character-
^
specific (often conservative) deterministic istic parameters are their median values Fi –
response parameters for the structure. Since representing the (often conservative) bias – and
many of these parameters are nondeterministic the logarithmic standard deviations bi,R and bi,U,
(often with wide variability), the actual response representing the variability due to aleatory and
may differ substantially from the calculated epistemic nondeterminism, respectively. The
response for a given peak ground acceleration. most widely used method for estimating these
FRS is modeled as a product of factors influencing characteristic parameters is the approximate
the response bias and variability: second moment procedure.
^
The overall median safety factor F may be Finally, the logarithmic standard deviations
found by scaling the ground response spectrum so of the individual safety factors are combined
long, until the limit state corresponding to the using square root of sum of squares (SRSS), in
analyzed failure mode is reached. All input vari- order to obtain the corresponding variability
ables must be set to their median values in this parameter of the global safety factor, bR and bU,
process for the scaling factor to correspond to the respectively.
median safety factor. Alternative, more refined methods can be used
Alternatively, the sources of conservatism of for deriving fragility curves, if necessary. For
the design calculations – compared to median- instance, the simplifying assumptions adopted
centered calculations – may be identified individ- in the separation of variables method can be
ually for each partial safety factor. For each of the relaxed by applying Monte Carlo simulation to
response factors, the response quantity of interest the structural dynamic analysis. In this method,
(e.g., stress) is calculated (or estimated) anew, but all parameters assumed to contribute significant
only with the corresponding input variables set to nondeterminism to the overall capacity are sam-
their median values. The ratio of this newly com- pled simultaneously, and the model used for
puted response with the one from the design structural dynamic analysis is perturbed corre-
calculations gives the corresponding median par- spondingly. Repeating this exercise for several
tial safety factor. For the median capacity factor, levels of the PGA leads to a discrete set of points
the margin between the median strength and the of the fragility curve, which can be interpolated
stress obtained in the design calculation is quan- or extrapolated as needed.
tified. The overall median safety factor is then
simply given by the product of the individual The HCLPF Capacity
safety factors. As already indicated in earlier sections, the
In order to estimate the logarithmic standard HCLPF value or HCLPF capacity is a crucial
deviations bi,R and bi,U, it is necessary to perform concept in the context of seismic robustness of
an additional calculation of the safety factor, in NPPs. It is defined – implicitly – as the value of
which all model variables are median centered, the PGA for which there is a high confidence
with the exception of the input variables (95 %) that the probability of failure does not
corresponding to the analyzed partial safety fac- exceed 5 % (i.e., it is low).
tor. The latter ones are instead perturbed by Based on the above definition, the HCLPF
a multiple k of their standard deviation. The capacity can be related to the median capacity
input variables should be perturbed to the side and the variability parameters as follows:
which leads to a smaller safety factor (since we
are more interested in quantifying the effect of
HCLPF ¼ Am expð1:65ðbR þ bU ÞÞ
variables in the more unfavorable cases).
The resulting partial safety factor is denoted
here as Fksi . where 1.65 is the 5 % fractile of the standard
The variability parameter (either randomness normal distribution.
or uncertainty) of the analyzed safety factor is The HCLPF capacity can be approximated by
then the 1 % fractile of the composite mean fragility
curve. This definition of the HCLPF in the com-

1 Fksi
posite model is conservative, because it is
bi ¼ ln ^ guaranteed to be equal or smaller than the
jkj F
HCLPFs of all corresponding non-composite fra-
gility models, i.e., those in which bR and bU
For several safety factors it is common practice to are kept separately (for a specific bC, there
use generic variability parameters, as suggested are infinitely many possible combinations of bR
in EPRI (1994). and bU).
Seismic Robustness 1.000

Analysis of Nuclear
Power Plants,
Fig. 5 Example of
fragility curve in
logarithmic scale
0.100
F [ −]
0.010
0.001
0.1 1
HCLPF Am
a [g]
Mean 5% 50 % 95 %
The HCLPF capacity has then the following a semi-probabilistic approach has been proposed.
simple form: It is referred to as “conservative deterministic
failure margin” (CDFM) method; compare
HCLPF ¼ Am expð2:33bC Þ EPRI (1991). The CDFM method can be briefly
summarized by the following four steps:
where 2.33 is the 1 % fractile of the standard
normal distribution. 1. For the targeted HCLPF capacity (e.g., the
The above-described relationship of the Review Level Earthquake in an SMA), the
HCLPF value with the fragility curve is exempli- floor responses (demand) are computed
fied in Fig. 5 for an SSC with a median capacity using the 84 % fractile (i.e., an 84 % probability
of 0.5 g and variability parameters bR and bU of non-exceedance) of the ground response
equal to 0.3 and 0.4, respectively. spectrum. Damping is to be assumed S
conservatively.
Simplified HCLPF Estimation Using the CDFM 2. The strength parameters are evaluated at the
Method 2 % fractile, i.e., the probability of exceedance
The previous section described the formal defini- of the selected strength is at the 98 % proba-
tion of the HCLPF capacity based on the bility level.
corresponding fragility curve. It is clear that 3. The inelastic energy absorption factor, Fm,
once a full-scope fragility analysis has been should be estimated conservatively, i.e., at
performed, the HCLPF capacity can also be eas- a 84 % probability of exceedance level.
ily obtained. 4. If, based on the above, the following inequal-
Within SMA it is, however, not necessary to ity is satisfied, then the targeted HCLPF
have a full description of the fragility curve; the capacity has been demonstrated:
HCLPF capacity is sufficient. As a simplified
alternative to estimate the HCLPF value, Demand=Capacity Fm
Analysis of Seismic Robustness at Plant Success-Path-Based Approach

Level This approach is also referred to as the EPRI
(Electric Power Research Institute)-approach,
Seismic Margin Assessment (SMA) because it is best described in EPRI NP-6041.
The basic concept underlying this approach is
Goals and Approaches that of a success path, indicating the transition
The main goal of an SMA is to identify the from a disturbance to the normal plant
seismic capacity of the plant, i.e., the maximum operation – caused by a design-exceeding
level of seismic ground motion for which the earthquake – to a safe state, in which the funda-
plant can still be safely shut down. In addition, mental safety functions (reactivity control, resid-
it is the goal of any SMA to identify the weaker ual heat removal, and radioactivity confinement)
components, i.e., the components limiting the are durably ensured.
seismic capacity of the plant. An example of a success path is shown for
There are essentially two approaches that have a pressurized water reactor (PWR) in Fig. 6.
gained widespread acceptance and that will be It is important to note that a reasonable
described in more detail in the sequel: a success- success-path definition should take into account
path-based approach (“EPRI method”) and the seismic design of the plants’ safety systems.
a fault-tree-based approach (“NRC method”). The preferred success paths should be those along
Stated in simple terms, the difference of the which the safety functions are fulfilled by seismi-
approaches is the following: cally designed systems.
The success-path approach identifies the com- The success-path definition should make sure
ponents necessary to avoid a core damage; the that the most likely scenarios of plant disturbance
plant capacity is that of the weakest of these after a design-exceeding seismic event are covered.
components. At least two success paths – leading to a durable
The fault-tree approach identifies the combi- safe state of the plant – need to be defined, and one
nations (“minimum cutsets,” abbreviated MCS) of the two success paths should include the mitiga-
of component failures sufficient to cause a core tion of a small break loss of coolant accident
damage. For each of these MCS, the capacity is (LOCA), unless even a small break LOCA can be
that of the strongest of these components. The excluded, e.g., due to a particularly high quality of
plant capacity is then equal to the minimum the supports of small-bore primary piping such as
capacity of any of the MCS capacities. measurement lines.
Having stated the differences between the two The above mentioned Fig. 6 describes the suc-
approaches, this introductory section deserves to cess path at the level of the safety functions. The
be concluded by noting a fundamental, unifying goal is to ensure that all the safety functions along
feature of the two approaches, namely, that they the success path have a sufficient margin over the
are essentially deterministic. This property SSE. This requirement obviously needs to be cas-
becomes particularly obvious when comparing caded down to the components making up the
the plant-level HCLPF capacity with a target systems fulfilling the safety functions. For exam-
value of the PGA. For instance the US Nuclear ple, in order to ensure steam generator (SG) feed
Regulatory Commission requires 1.67 times the following the earthquake, it is necessary for the
PGA of the design basis earthquake; refer to emergency feedwater pump to remain operational.
NUREG (1993). In the European Utility Require- The requirement of sufficient margin does
ments, a minimum capacity of 1.4 times the PGA not only apply to the so-called “frontline” systems,
of the SSE is recommended; refer to EUR orga- i.e., the fluid systems specifically performing
nisation (2001). The result of this comparison is the respective safety function, but also to the
a binary and hence deterministic expression, so-called support systems, such as electrical
indicating either sufficient margin or lack thereof, power supply, heating/ventilation/air-conditioning
but no intermediate answers. (HVAC), and instrumentation and control (I&C).
Reactor
Essential Secondary
Reactor Coolant
Support Side Heat Long Term Heat Removal
Shutdown System
Systems Removal
Inventory
Integrity of
Fast Control the Reactor Long Term
Rod Insertion Coolant Secondary Side
(SCRAM) Pressure Heat Removal
Boundary
Emergency Steam
Power Generator
Seismic Supply, Feed
Success
Event Safety &
Automation Steam
Systems etc. Release
Fast Injection Reactor

Emergency Secondary Primary
of Borated Coolant
Core Side Side Heat
Water into System
Cooling Cooldown Removal
the Core Boration
Seismic Robustness Analysis of Nuclear Power Plants, Fig. 6 Example of a success path to a safe shutdown state
for a pressurized water reactor
The result of cascading the requirement for The result of the screening process is a, typi-
sufficient margin to the individual components cally large, set of components which are
consists in a list of components which are needed judged – based on a structured and standardized
to fulfill the safety functions credited in the walkdown procedure – to have a generic HCLPF
success-path definition. This list is referred to as capacity equal or larger than the RLE. For the
Seismic Equipment List (SEL) or also Safe Shut- remaining components, individual HCLPF
down Equipment List (SSEL). values need to be derived.
The metric used to measure the margin is the The fact that the walkdown-based screening
HCLPF value introduced in earlier chapters. provides only a lower bound of the generic
Since the success-path-based approach is meant HCLPF capacity, instead of a sharp value, implies
to be a predominantly deterministic approach, the that the RLE should be set high enough, so that
CDFM method is preferred. some components for which a non-generic
Considering that – as a rough estimate – the HCLPF is evaluated are below the RLE. Only in
size of the SEL is in the order of 103 components, this way, a sharp plant HCLPF can be determined,
it should become clear that it is not feasible to whereas otherwise one obtains only a lower
perform dedicated HLCLF calculations for all bound, because the generic HCLPFs can only be S
components of the SEL. Instead, a walkdown- stated as being equal or higher than the RLE.
based screening process is to be implemented. For those components for which an individual
Since this screening process is common to all HCLPF is to be evaluated, the responses due to
approaches for seismic robustness and risk the RLE are needed. These may be either
assessment, it is discussed in the final chapter of obtained by performing new floor response spec-
this essay on practical issues. tra and equipment response calculations based on
A critical step preceding the screening process the RLE or by scaling existing responses to the
is the definition of the so-called Review Level ground motion of the SSE.
Earthquake (RLE). It consists in a site-specific Once both generic and individual HCLPF
characterization of a seismic ground motion capacities have been evaluated, the overall capac-
(typically a response spectrum) that is more ity of the plant results as the lowest HCLPF of the
severe than the ground motion considered in the considered equipment. This implies that both
design of the plant (i.e., the SSE). success paths must be sufficiently robust.
As a concluding remark, the deterministic the highest risk relevance, because their proba-
nature of the success-path-based approach is bility is higher than non-minimum cutsets.
emphasized, both at the individual component The components screened out on the basis of
level (CDFM used for HCLPF estimation) and the walkdown observations are not included in
at the system modeling level (no systematic gen- the system modeling, i.e., only components
eration of minimum cutsets; see the following which cannot be generically stated as having
section). a HCLPF capacity equal or higher than the RLE
are modeled in the fault trees. Individual HCLPF
Fault-Tree-Based Approach capacities of these components are to be esti-
This approach is also referred to as the NRC mated, either based on a fragility analysis or at
approach, because it has been proposed and least based on the CDFM method.
developed for the NRC in NUREG (1985, 1986). Performing a cutset analysis for the event trees
In the fault-tree-based approach, several judged as relevant for the SMA (typically at least
aspects are unchanged with respect to the a seismic-induced transient, such as loss of
success-path-based approach, such as the off-site power, and a small break LOCA) results
walkdown-based screening process. The main in a list of minimum cutsets to be further
difference compared to the success-path-based processed as follows.
approach is that a fault-tree-based system analy- For each of the cutsets, a characteristic
sis is relied upon in order to determine minimum HCLPF capacity is obtained by taking the max-
cutsets, i.e., combinations of failures leading to imum HCLPF capacities of the component fail-
core damage. The attribute “minimum” indicates ure forming the cutset. From the resulting list of
that no individual failure event can be removed cutset HCLPFs, the minimum is then taken, in
from the set without altering its property of lead- order to obtain the fault-tree-based plant HCLPF
ing to core damage. The minimum cutsets have capacity. Fig. 7 is meant to illustrate the process.
1,000
0,100
F [ -]
0,010
0,001
HCLPF Am 1
0,1
Seismic MCS MAX-Operation Seismic capacity of MCS
SEISi ∩ SEISj MAX (HCLPFSEISi , HCLPF

FSEIS HCLPFMCS1
Sj )
MIN -Operation
SEISi ∩ SEISk MAX (HCLPFSEIS , HCLPFSEIS ) HCLPFMCS2

i k
SEISi ∩ SEISj ∩ SEISk … …
… … …
MIN ( HCLPFMCS1 , HCLPFMCS2 ,…)
Plant HCLPF capacity
Seismic Robustness Analysis of Nuclear Power Plants, Fig. 7 Min-Max operation in fault-tree-based SMA
Individual seismic-induced failure events are applies to the event “seismic-induced core dam-
denoted by SEISi, SEISj, SEISk, etc. age,” which may or may not occur in case of
If the fault-tree-based plant HCLPF capacity a seismic event of a given PGA.
is higher than the RLE, then the plant HCLPF can The event “seismic-induced core damage” is
only be judged as being at least equal to the RLE, a central notion of both seismic PSA and SMA
since the components screened out on the basis of and hence a unifying aspect. Clearly, since both
the walkdown observations were not included in methodologies (seismic PSA and SMA) aim at
the system modeling. analyzing the plant-level resistance to seismic
If a PSA model is available, then the effort for events, one is – in general – not concerned with
performing the system modeling is greatly individual failures, but only with combinations of
reduced, since the seismic-induced failures of failure leading to core damage.
the non-screened components merely need to be However, there is a significant difference: in
added to the existing fault trees. A major benefit the seismic PSA, it is assumed that the event
of reusing an existing PSA model is also the “seismic-induced core damage” can – at least
possibility to analyze mixed cutsets, i.e., combi- theoretically – occur due to seismic events of
nations of seismic-induced and non-seismic- any intensity (and hence PGA).
induced failures, such as random component fail- This is where the concept of seismic hazard at
ures or human (operator) errors. the NPP site comes into the picture: it is the char-
As a final remark it is emphasized that – even acterization of the seismic ground motions “to be
though the probabilities of the cutsets are not used expected” at the site (see section “Motivation for
in the quantification of the plant HCLPF Evaluating Seismic Robustness”). For the purpose
capacity – the fault-tree-based approach to SMA of seismic PSA, the most important characterization
resembles a probabilistic approach, because is afforded by the so-called hazard curve, in which
cutsets are generated, i.e., a systematic exploration the probability of observing a seismic event exceed-
of the combinations of individual component failing a PGA equal to a during a predefined time
ures is performed. Nevertheless, the subsequent interval (typically one year) is expressed as
steps of the procedure and – in particular – the a function of a. Thus, the hazard curve is
final result are deterministic, since a single num- a monotonously decreasing function H(a).
ber (the HCLPF capacity) is used to characterize In Fig. 8, a set of hazard curves is shown, each
the capacity of each individual SSC. curve being associated with a probability level
(50 %, 5 %, and 95 %). This characterization of
Seismic PSA the site hazard in terms of a population of hazard
curves, rather than a single hazard curve, reflects the
Basic Notions uncertainties in the hazard estimation procedures.
While sharing numerous aspects with SMA, seis- Recall that for each level a of the PGA, core S
mic probabilistic safety analysis (PSA) is distinct damage may or may not occur and denote the
from its deterministic counterpart (SMA) in sev- corresponding probability by FCDF A (a). The prob-
eral respects. The most prominent difference ability that a seismic event leading to core dam-
is – of course – due to the attribute “probabilis- age occurs during the time interval to which the
tic,” which implies nondeterminacy for all the hazard curve refers (typically one year) can be
possible states of the system (i.e., the model of shown to be:
the NPP). Indeed, for a seismic event of a given
PGA, a component may fail or may not fail, and ð
1
dH
both possibilities are assigned a probability. This CDF ¼ A ð aÞ
FCDF da
da
probability is called fragility and has been intro- a¼0
duced at length in section “Quantification of Seis-
mic Margin: Fragility Curves and HCLPF The acronym CDF denotes core damage fre-
Values.” The same property of nondeterminacy quency, which is – in quantitative terms – almost
Seismic Robustness 1.00E-02

Analysis of Nuclear
Power Plants,
Frequency of exceedance [1/yr]

Fig. 8 Hazard curves 1.00E-03
associated with different
confidence levels
1.00E-04
1.00E-05
median
1.00E-06 5 % fractile
95 % fractile
1.00E-07
1.00E-02 1.00E-01 1.00E+00 1.00E+01
PGA (normalized)
identical to the annual probability of core ð

1
ðiÞ ðiÞ dH
damage. CDF ¼ FA ð aÞ da
da
In the practical implementation of seismic a¼0
PSA, the evaluation of the CDF is however
more involved. The reason is that the condi- where F(i)
A (a) is the probability of the MCSi, con-
tional probability of core damage FCDF A (a) ditional on a given PGA a, and H(a) is the fre-
cannot by obtained in a closed form, because quency of occurrence of a seismic event with
of the enormous number of component failure a PGA equal or higher than a.
combinations potentially leading to core The probability of the MCSi is
damage.
This is where the specific application, seismic n ði Þ mðiÞ
ðiÞ ðiÞ ðiÞ
PSA, becomes reminiscent of its more general FA ðaÞ ¼ ∏ fj ðaÞ ∏ pk
j¼1 k¼1
“parent,” namely, level 1 PSA for internal events.
In level 1 PSA risk-relevant accident progression
sequences are modeled by event trees, and the where f (i)j (a) is the jth fragility appearing
potential combinations of component failures in MCSi and p(i) k is the failure probability of
leading to unavailability of safety systems called the kth non-seismic basic event in MCSi. The
upon during the accident progression sequences integers n(i) and m(i) are the number of fragilities
are modeled by fault trees. and the number of non-seismic basic events in
These event trees and fault trees represent MCSi.
a (very complex) Boolean expression of the In practice the above integral is approximated
event “core damage.” by
A well-established procedure for evaluating
X
n
the probability of this event is to express the CDFðiÞ f al12 ðDH l Þ
(complex) Boolean expression as a union of l¼1
(much simpler) intersections of component fail-
ures; these intersections are called “minimum where a0 and an are the limits of the risk-relevant
cutsets” (MCS). PGA range and the remaining ai’s are
For each minimum cutset, MCSi, the contribu- selected at convenient points, e.g., the evaluations
tion to the core damage frequency is of the inverse hazard function at the values
101, 102, etc.The corresponding mid-interval conservative with respect to the probability of
fragilities f al12 are logarithmically interpolated, the top event, because correlation is detrimental
(i.e., the failure probability increases) for inter-
section of events and beneficial for unions of
f al12 ¼ 10ðlog10 f ðal1 Þþlog10 f ðal ÞÞ=2
events (Fig. 9).
An alternative approach, described in NUREG
The probability of occurrence associated with (1990), consists in modeling the dependencies
each interval is DHl = H(al) H(al1). The total between seismic failures in terms of the correla-
CDF is then obtained by evaluating the union of tion coefficients between the variables character-
the individual MCS. izing the seismic demand and capacity of the
SSCs. The adopted model makes full use of the
Correlation Between Seismic Failures possible (positive) range of correlation coeffi-
The issue of mutual dependence (or alternatively, cients (between 0 and 1).
correlation) between seismic-induced failures The separation of the correlation coefficients
relates to the question whether the seismic- of demand and capacity reflects the difference in
induced failure of a component, e.g., an emer- their respective cause: the correlation of the
gency diesel generator (EDG), is more probable, demand variables is due to “similarity” of the
under the condition that another component (e.g., excitation experienced by distinct components
the EDG of another safety train) has experienced (e.g., due to spatial proximity), whereas the cor-
seismic-induced failure. Several studies indicate relation of the capacity variables is due to “sim-
that in seismic PSA, the effect of correlations on ilarity” in terms of the physical properties of the
the seismic-induced core damage frequency components (e.g., due to the fact that they have
(CDF) may be very significant; refer to Pellissetti been manufactured by the same supplier). The
and Klapp (2011) and to references 1 through word “similarity” is placed between quotes
4 therein. because it is to be understood in the statistical
The most straightforward and – in sense.
general – most conservative approach consists In this approach failure is defined in terms of
in assuming full correlation between identical the following limit-state function:
and “seismically similar” components of differ-
ent, redundant safety trains, i.e., components
R
connected by “AND”-logic. Therefore, with X ¼ ln
S
this model the redundancy represented by
the different trains of the plant is not credited,
where R and S represent the response and the
as far as seismic-induced failures are concerned.
strength for a given failure criterion. Failure
Furthermore, for components of a single S
takes place if
train combined under the same “OR” gate in the
fault tree, the failures are assumed to be
independent. R>S , X>0
In this approach failures are thus assumed to
be either fully correlated or totally uncorrelated; It is convenient to model both R and S by
it is hence a binary model (black/white, zero/one) log-normal random variables with the parameters
which cannot accommodate gradually varying mR, bR and mS, bS. According to the definition of
levels of correlation. In fact, for redundant com- log-normal variables, the respective companion
ponents connected by “AND”-logic, this model normal variables are then CR = ln R (with mean
negates the existence of distinct failure events for mR and standard deviation bR) and CS = ln S (with
the individual redundant trains and will hence mean mS and standard deviation bS). In view of its
also be referred to as “deterministic correlation definition, X is then normally distributed and its
model.” This binary, deterministic model is mean and standard deviation are
independent Fragilities under AND - Logic
LHP LHQ
LHP LHQ
F '(a) = F(a)
dependent Fragilities under OR - Logic
LHP LHQ
RIS RIS
Pump Valve
F '(a) > F(a)
Seismic Robustness Analysis of Nuclear Power Plants, Fig. 9 Schematic representation (Venn diagrams) of
dependence between individual seismic-induced failures
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
the companion normal variables of Ri and Sj, i.e.,
mX ¼ mR mS , bX ¼ bR 2 þ bS 2
ln (Ri) and ln (Si). The advantage of calibrating
the correlation model in terms of the companion
The probability of failure is then given by normal variables is that the meaningful range
of the correlation coefficient is always [1, 1].
mX m This is not the case for log-normal random
P½ X > 0 ¼ 1 F ¼F X
bX bX variables. It should be noted that according to
its definition, the correlation coefficient rX1 X2
where F is the cumulative probability density depends on both the correlation coefficients and
function of the standard normal distribution. the standard deviations; this is necessary to pre-
Considering two individual components, one vent a pair of variables with a high correlation but
may then define the random variables R1, S1 and with very small standard deviation from introduc-
R2, S2, respectively, and X1, X2 according to the ing a high overall correlation. With the above
above definition of X. Assuming no cross- definitions, the joint probability of failure of the
correlation between Ri and Sj, after some arith- two components Q1\2 is defined in terms of the
metics, the correlation coefficient for X1, X2 may following two-dimensional integral:
be expressed as
Q1\2 ¼ P½ðX1 > 0Þ ^ ðX2 > 0Þ
1 ðð
rX1 X2 ¼ bR1 bR2 rR1 R2 þ bS1 bS2 rS1 S2
bX1 bX2 ¼ fX1 X2 ðx1 , x2 Þdx1 dx2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 0x1 , x2 1
where bXi ¼ bRi 2 þ bSi 2 .
It should be noted that in the above expression, In the above equation, fX1 X2 is the bivariate
the correlation coefficients rR1 R2 and rS1 S2 relate to normal probability density function (PDF):
1
fX1 X2 ðx1 , x2 Þ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 rX 1 X 2 2
2pbX1 bX2
2 2
1 x1 mX1 x2 mX2 x1 mX1 x2 mX2
exp þ 2rX1 X2
2 1 rX 1 X 2 2 bX1 2 bX2 2 bX 1 b X 2
Accordingly, the probability of simultaneous two times the PGA of the DBE; refer to
failure of n components has the form Kl€ugel (2007).
In the time-dependent model described in
Q1\...\n ¼ P½ðX1 > 0Þ ^ . . . ^ ðXn > 0Þ NUREG (1987), the seismic-induced human
ð ð error probability (HEP) is expressed as
¼ ... fX ðxÞdx a function of the time t between the earthquake
0x1 , ..., xn 1 and the action:
The application of the above correlation model in HEPðtÞ ¼ HEPt¼1 min ta
the context of fault-tree-based seismic
PSA – using general purpose PSA software – is In the ramp model presented by Yokobayashi
exemplified in Pellissetti and Klapp (2011). et al. (1998), the operator performance is
modeled by a linear relationship between the
Operator Actions seismic-induced HEP and the PGA until
Earthquakes not only may impact the reliability a maximum ground acceleration level L is
of safety systems but also influence operator reached.
performance: Finally, it is worth mentioning a model based
on the CAV (cumulative absolute velocity) value
• The workload for the shift team may be higher presented by Kl€ugel (2007). No adverse effect on
due to a failure of operational systems. operator performance is expected if the CAV is
• Spurious signals, e.g., caused by relay chatter, below 0.16 gs. In case of stronger earthquakes,
may make appropriate decision-making more
difficult. HEP ¼ HEPBaseline þ HEPGF, direct þ HEPShock
• Accessibility of buildings may be degraded
(e.g., due to a failure of lighting or automatic HEPBaseline is the HEP used in the PSA for
locking systems) and may hence lead to internal events. HEPGF,direct describes structural
smaller grace times for the necessary actions. failure of equipment (e.g., instruments, main con-
• Finally, earthquakes may have a direct impact trol room boards) as well as accessibility restric- S
on human behavior (shocks). tions leading to a guaranteed failure (GF) of the
considered action. Finally, HEPShock models
Past seismic PSA experiences have shown a failure of all operator actions due to operator
that the choice of probabilities for seismic- shock, depending on the CAV: shock-induced
induced operator actions can have a significant failure is assumed with probability one if the
impact on the overall PSA result. Thus, a short CAV is greater than 0.16 gs and strong-motion
overview of seismic human reliability models duration exceeds a threshold value; otherwise,
discussed in the literature is presented in the the probability of shock-induced failure is zero.
following.
According to the shock model, it is conserva- Uncertainty Analysis
tively assumed that all human actions fail if the In the earlier sections on seismic hazard analysis
peak ground acceleration exceeds a certain limit. and – in greater detail – on fragility analysis, two
For instance, Swiss regulations define this limit at sources of nondeterminism have been introduced
for the parameters relevant in a seismic PSA, i.e., subfigure below. In this subfigure, too, for
the frequency of occurrence of a specific PGA a defined value of the PGA the uncertainty in
(hazard) and the capacity of a component or the fragility is schematically highlighted by
structure (again in terms of the PGA). a blue, double-sided vertical arrow.
These two sources of nondeterminism are ran- Finally, in the right portion of the figure, it is
dom variability and (epistemic) uncertainty. indicated how the uncertainties in the seismic
Random variability is the nondeterminism due hazard and in the fragility propagate – across
to the inherently unpredictable nature of earth- the convolution integral introduced
quakes and, consequently, of the earthquake- previously – to the core damage frequency.
induced vibrations of the ground, the NPP civil The result of this uncertainty analysis – or
structures, and the NPP equipment. The degree of uncertainty propagation – is then a distribution
scatter induced by random variability is termed as function of the core damage frequency, as shown
irreducible, since additional collection of data in the lower right subfigure. The uncertainty in
can improve its statistical representation, but not the core damage frequency is here emphasized by
the scatter itself (expressed for instance by the a blue, double-sided horizontal arrow (Fig. 10).
variance or standard deviation).
Uncertainty, on the other hand, denotes the
nondeterminism due to limited accuracy of Practical Issues
models used to characterize the structural
response. This lack of accuracy can be caused Seismic Equipment List
by inherent limitations of the model when The assembly of the so-called Seismic Equip-
representing the phenomena it is meant to capture ment List (SEL) is a major task in both SMA
(including issues such as model resolution). Also, and seismic PSA. To begin with, this list includes
it can be caused by limited data availability for all those systems, structures, and components
model calibration. Another type of (SSC) which contribute to bringing the plant to
nondeterminism described by the term “uncer- a safe shutdown state after a seismic event (this is
tainty” is the subjective variability associated why in EPRI (1991), this list is referred to as Safe
with the process of selecting a model from a set Shutdown Equipment List (SSEL)).
of concurrent models involving different experts. For success-path-based SMA, EPRI (1991)
This source of uncertainty is particularly pro- provides detailed guidance on how to assemble
nounced in probabilistic seismic hazard analysis. the SEL. For fault-tree-based SMA and seismic
The uncertainty in the seismic hazard is typi- PSA, it is a common situation that a PSA model
cally characterized by deriving different hazard already exists for the level 1 PSA of internal
curves for different confidence levels. In the fol- events. The basic events of this level 1 PSA
lowing figure this is shown in the upper left model represent a good starting point for the
subfigure; in this case the confidence levels are SEL in this case; however, the list is to be
expressed by multiples of the standard deviation. extended by including passive components (e.g.,
For a defined value of the PGA, the uncertainty is tanks) and structures, which are typically not
schematically highlighted by the blue, modeled in level 1 PSA for internal events.
double-sided vertical arrow.
As discussed in section “Log-Normal Capac- Walkdown-Assisted Use of Generic Data and
ity Model,” the uncertainty associated with the Screening
seismic capacity of SSC is characterized by the The number of components in the initial SEL
uncertainty variability parameter bU. Similarly as (in the order of magnitude of 103) prevents
for the hazard curve, the uncertainty expressed by a consistent degree of detail to be applied in the
bU can be shown graphically in terms of the analysis of all of them. It is hence a common
fragility curves associated with the various con- feature of SMA and seismic PSA that the initial
fidence levels, as shown in the lower left SEL has to undergo a more or less extensive
1.00E-04
Annual Prob. of Exceedance
-3σ
1.00E-05
-2σ
-1σ
1.00E-06 µ
+1σ ∞
dH
CDF = ∫ F (a ) da da
+2σ CDF
1.00E-07 A
+3σ a =0
1.00E-08
PGA
1
0.9
Probability of non-exceedance
1.0 0.8
Prob. of non-exceedance
0.9
0.7
0.8
0.7 0.6
0.6 0.5
0.5 0.4
0.4
0.3
0.3
0.2 0.2
0.1 0.1
0.0
0
0 0.2 0.4 0.6 0.8 1 1.0E-09 1.0E-08 1.0E-07 1.0E-06
PGA Core Damage Frequency Expected Value
EV (Best estimate) 5% 50 % 95 %
Seismic Robustness Analysis of Nuclear Power Plants, Fig. 10 Schematic illustration of uncertainty quantification
in seismic PSA
screening process in order to reduce the number plant is designed – without experiencing the deg-
of equipment items on the list. The main purpose radation of equipment to a point that the plant
of doing so is to enable the analysts to focus the cannot be brought and maintained in a safe shut-
majority of their subsequent efforts on equipment down state.
with relatively low capacity and consequently By necessity, seismic robustness assessment
with the highest risk relevance. focuses on the systems, structures, and compo-
The screening process takes its legitimacy nents (SSC) that are essential to the goal of a safe
from extensive earthquake experience accumu- shutdown in the wake of an extreme seismic
lated over the years. A crucial ingredient of the event. For these SSC the actual capacity – or S
process is the execution of a seismic walkdown a lower bound thereof – is evaluated; in order to
during which it is verified, whether the equipment ensure a common measure of this capacity for all
in the experience database is representative of the SSC, it is quantified in terms of a characteristic
components encountered in the plant under ground motion parameter, typically the peak
investigation. ground acceleration.
Based on the above, evaluating the
seismic robustness requires both system
Summary analysis – accounting for the response of a plant
to a seismic-induced accident from a process
Seismic robustness of nuclear power plants engineering point of view – and structural analy-
(NPP) denotes the ability of the plant to withstand sis of the essential SSC.
severe seismic ground motion – possibly even There are two approaches that have been
beyond the seismic loads against which the found to be viable for fulfilling the above
requirements, namely, seismic margin assess- Electric Power Research Institute (EPRI) (1994) Method-
ment (SMA) and seismic probabilistic safety ology for developing seismic fragilities. Report
TR-103959, Palo Alto
analysis (seismic PSA; in the United States, the Electric Power Research Institute (EPRI), U.S. Depart-
term seismic PRA is preferred, with “risk” ment of Energy (2006) Program on technology inno-
replacing “safety”). vation: use of cumulative absolute velocity (CAV) in
The main difference between the two methods determining effects of small magnitude earthquakes on
seismic hazard analyses. Report 1014099, Palo Alto/
is that SMA is essentially a deterministic Germantown
approach, leading to an estimate of the seismic EUR Organisation (2001) European utility requirements
robustness of the plant in terms of a single num- for LWR nuclear power plants
ber, namely, the plant-level capacity. International Atomic Agency (IAEA) (2003a) Seismic
design and qualification for nuclear power plants.
In contrast, seismic PSA is an explicitly prob- Safety guide NS-G-16, Vienna
abilistic approach, analyzing those combinations International Atomic Agency (IAEA) (2003b) Seismic
of component failures that prevent the plant from evaluation of existing nuclear power plants. Safety
being transferred to a safe shutdown state, thus series 28, Vienna
International Atomic Agency (IAEA) (2009) Evaluation
resulting in a damage to the reactor core. of seismic safety for existing nuclear installations.
Compounding the probabilities of these combi- Safety guide no. NS-G-2.13, Vienna
nations with the frequency of occurrence of seis- Kennedy RP, Ravindra MK (1984) Seismic fragilities
mic events of various intensity results in an for nuclear power plant risk studies. Nucl Eng Des
79:47–68
estimate of the seismic-induced core damage fre- Kl€ugel J (2007) An improved methodology for the evalu-
quency. A small value of the latter indicates ation of human error probabilities in a seismic PSA.
a large seismic robustness of the plant. In: Transactions, SMiRT 19, Toronto
Pellissetti MF, Klapp U (2011) Integration of
correlation models for seismic failures into fault tree
based seismic PSA. In: Transactions, SMiRT 21, New
Cross-References Delhi
Sadegh-Azar H, Hartmann H-G (2011) Grundlagen der
▶ Analytic Fragility and Limit States [P(EDP| seismischen Auslegung von Kernkraftwerken und
Einfluss der Boden-Bauwerk Wechselwirkung.
IM)]: Nonlinear Dynamic Procedures
Bauingenieur (D-A-CH-Mitteilungsblatt)
▶ Code-Based Design: Seismic Isolation of U.S. Nuclear Regulatory Commission (NUREG)
Buildings (1985) An approach to the quantification of seismic
▶ Seismic Risk Assessment, Cascading Effects margins in nuclear power plants. NUREG/CR-4334,
UCID-20444, Washington, DC
▶ Site Response for Seismic Hazard Assessment
U.S. Nuclear Regulatory Commission (NUREG). (1986).
▶ Structural Reliability Estimation for Seismic Recommendations to the Nuclear regulatory commis-
Loading sion on trial guidelines for seismic margin reviews of
nuclear power Plants. NUREG/CR-4482, UCID-
20579, Washington, DC
U.S. Nuclear Regulatory Commission (NUREG) (1987)
References Relay Chatter and operator response after a large
earthquake. NUREG/CR-4910, Washington, DC
American Society of Civil Engineers (ASCE) (2000) Seis- U.S. Nuclear Regulatory Commission (NUREG)
mic analysis of safety-related nuclear structures and (1990) Procedures for the external event core damage
commentary. ASCE 4-98, Reston frequency analyses for NUREG-1150. NUREG/CR-
Campbell KW, Bozorgnia Y (2011) Evaluation of the use 4840, SAND88-3102, Washington, DC
of cumulative absolute velocity (CAV) in the shut- U.S. Nuclear Regulatory Commission (NUREG). (1993).
down of nuclear power plants after an earthquake. In: Subject: policy, technical, and licensing issues
Transactions, SMiRT 21, New Delhi pertaining to Evolutionary and Advanced Light-
Chen W-F, Scawthorn C (eds) (2002) Earthquake engi- Water Reactor (ALWR) designs. SECY 93–087,
neering handbook. CRC Press, Boca Raton Washington, DC
Electric Power Research Institute (EPRI) (1991) Yokobayashi M et al (1998) Consideration of the effect of
A methodology for assessment of nuclear power human error in a seismic PSA. In: Proceedings of the
plant seismic margin (revision 1). Report EPRI 4th international conference on probabilistic safety
NP-6041-SL, Palo Alto assessment and management, Berlin
Seismic Sources from Landslides and Glaciers 3045
processes as far as possible at the current state

Seismic Sources from Landslides and of knowledge. The potential usage of this seismic
Glaciers monitoring data for hazard estimation is also
addressed.
Ewald Br€ uckl1, Daniel Binder2 and Stefan Mertl3
1
Department of Geodesy and Geoinformation,
TU Wien, Vienna, Austria A New Class of Earthquakes
2
Zentralanstalt f€
ur Meteorologie und
Geodynamik, Vienna, Austria Seismic Signals from the 1980 Mount
3
Mertl Research GmbH, Vienna, Austria St. Helens Eruption and Landslide
Mount St. Helens (46.2 N, 122.2 W) has been
the most active volcano in the Cascade Range
Synonyms during the Holocene. Its most impressive active
phase in historical times started with a series of
Earthquakes generated by landslides and gla- small earthquakes in March 1980. During the first
ciers; Seismic activity of landslides and glaciers half of the following May, seismic activity
increased dramatically, and a bulge over 100 m
grew outward at the mountain’s north flank. The
Introduction cataclysmic Plinian-style eruption, VEI
5, occurred on May 18, 1980. Within 15–20 s of
With few exceptions, the discovery and analysis a magnitude 5.1 earthquake, the volcano’s bulge
of global earthquakes with landslides and glaciers and summit slid away in a huge landslide
as sources did not begin until the twenty-first (Fig. 1a). Volcanic explosions interacted with
century. This may come as a surprise considering the landslide process (Fig. 1b). The landslide,
that the magnitudes of such earthquakes have comprising a volume of 2.8 km3, rushed first
reached and even exceeded M = 5. In this down the slope to the north and bent later to the
essay, we describe the specific characteristics of west-northwest along the North Fork Toutle
these seismic signals and their source locations. River with a speed briefly exceeding
We describe the source mechanisms and discuss 200 km/h. The landslide covered an area of nearly
the information that can be derived from inter- 60 km2, with the most distal deposits traveling
pretation of the data. In the case of landslides, the about 25 km (Fig. 1c). The height of Mount
source process represents a severe hazard, and the St. Helens was reduced from about 2950 to
same may be true for glaciers. 2550 m (Brantley and Myers 2000).
Although seismic signals from landslides and Seismic signals from the May 18, 1980, erup-
glaciers are observed at intermediate scales, we tion and landslide of Mount St. Helens have been S
will jump from magnitudes of M ~ 5 to M < 0. recorded by seismic stations at epicentral dis-
The deployment of seismic stations on the target tances from 23 up to 133 (Fig. 2a). The
landslide or glacier is necessary to observe such seismograms show strong surface waves in the
weak signals. Systematic research on these period range of 100–260 s corresponding to
microearthquakes also did not start until the Ms = 5.9. The spectra fall off very rapidly at
twenty-first century. We offer an overview on periods shorter than 75 s. This makes these
landslide and glacier processes relevant to the recordings significantly different from typical
generation of microearthquakes. Using selected tectonic earthquake spectra (Fig. 2b). A cutoff
case studies, we present the diversity of seismic period of ~5 s would be expected for a tectonic
signals generated by landslide and glacier earthquake of similar magnitude. The
sources and emphasize the close relations to geo- low-frequency content of the Mount St. Helens
detic and hydrometeorological data. We relate recordings suggests a very slow source process.
the seismic data to landslide and glacier Kanamori and Given (1982) analyzed the
3046 Seismic Sources from Landslides and Glaciers
Seismic Sources from Landslides and Glaciers, deposit area (Modified after Voight 1981; Moore and
Fig. 1 The Mount St. Helens eruption and landslide Albee 1981, http://pubs.usgs.gov/gip/msh/mudflows.
18 May 1980; (a) the bulge at the north flank; (b) devel- html)
opment of the landslide during eruption; (c) landslide
radiation pattern of the Love and Rayleigh waves The Single-Force Seismic Source
and concluded that a subhorizontal, about Earthquakes of the new class produce long-
southward-directed single force is an adequate period surface waves detectable on a global
source mechanism and describes the radiation scale with wavelengths at least ten times the
pattern much better than a conventional extent of the source area. The spatial migration
DC. The earthquake attributed to the Mount of the source during the event cannot be resolved
St. Helens landslide resembled the first example by long period data, therefore we consider a sin-
of a new class of global earthquakes. gle force F(t) varying in time but fixed in space as
Seismic Sources from Mt. St. Helens May 18, 1980, M=5.9 Mammoth Lakes May 25, 1980, M=6.2
Landslides and Glaciers,
Fig. 2 Recordings from D=70.5° a D=77.34° b
seismic observatories in MAJ R1 MAJ R1
Japan (MAJ), Norway V V
(KON), and Australia R1
(NWA), 30 s high-cut R1
R R
filtered of (a) Mount.
St. Helens eruption and
G1 G1
landslide, (b) Mammoth T T
Lakes earthquake
(Modified after Kanamori
and Given 1982); V, R, and D=67.4° D=73.9°
T are the vertical, radial and R1 R1
KON KON
transverse components, R1 V V
and G1 mark the first
Rayleigh and Love phases R1 R1
R R
G1 G1
T T
D=133.3° D=134.6°
NWA R1 R1
V NWA
V
R1 R1
R R
G1
G1
T T
0 30 min
the seismic source. We refer to F(t) as force We have I(T) = v(T) = 0, where T is the
history. F(t) is the reaction force to the accelera- duration of the landslide. This condition
tion or deceleration b(t) times a mass M on the imposes a constraint on the derivation of the
Earth’s surface (Eq. 1). force history F(t). Kanamori and Given (1982)
found that the data of the Mount St. Helens erup-
FðtÞ ¼ bðtÞM (1) tion can be fitted by a sinusoidal F(t) with a period
of 240 s.
The acceleration is caused by the drop of the
S
frictional force at a sliding plane from static equi- CSF: Centroid Single Force
librium to its value during sliding. The excess The displacement multiplied by the landslide
gravitational forces result in the acceleration. mass at time t follows from an integration of
When the slope of the landslide trajectory I(t) (Eq. 2). The total displacement vector
reduces, frictional forces exceed the driving grav- D times M is given by Eq. 3.
itational forces, and the landslide decelerates.
The momentum I(t) and velocity of the landslide ðt
v(t) follow from integration of Eq. 1 over time DM ¼ IðtÞdt (3)
0
(Eq. 2).
ðt The product of D and M can be derived from
IðtÞ ¼ vðtÞM ¼ FðsÞds (2) seismic data. Satellite images or other topo-
0 graphic information may also be used to break
Seismic Sources from

Fig. 3 Recordings of the
Alaska glacial earthquake
and a M 4.2 earthquake
(Modified after Ekström
et al. 2003)
down the product into its factors D and M. Equa- of the new-class earthquake to a nearby tectonic
tions 1, 2, and 3 are valid for any surface mass earthquake of similar magnitude and may explain
accelerated or decelerated by either gravitational, why the new class of earthquakes has so long
inertial, or frictional forces, stress relief, or vol- been undiscovered. CSF inversion was applied
canic processes. Kawakatsu (1989) showed that to the events in Greenland and Antarctica by
D M represents the spatiotemporal centroid sin- Tsai and Ekström (2007) and Nettles and
gle force vector CSF = D M, the analogue of Ekström (2010), resulting in an improved loca-
the CMT for moment tensor seismic sources. He tion with an uncertainty of approximately 20 km.
also provided the formalism for normal-mode These events were clustered at or near eight fast-
inversion. moving outlet glaciers with speeds >800 m/year
(Fig. 4). An antisymmetric boxcar function (25 s
of constant acceleration before the centroid focal
Global Glacial Earthquakes time followed by 25 s of constant deceleration)
was assumed in order to derive reactive forces
A search for earthquakes of the new class in transmitted to the Earth’s crust by the displace-
Arctic and Antarctic areas in data provided by ment of a hypothetical mass. The magnitudes of
global networks was carried out by Ekström centroid single force vectors were in the range
et al. (2003) for the period 1999–2001 and 10 1012 CSF 200 1012 m kg, and the magni-
extended to the period 1993–2008 by Tsai and tudes were 4.6 Ms 5.1. Most CSF vectors
Ekström (2007) and Nettles and Ekström (2010). point in the opposite direction of the general flow
About 243 earthquakes were found in Greenland, of the glaciers.
14 in Antarctica, and 1 in Alaska. The one in The frequency of glacial earthquakes in
Alaska (Fig. 3) exemplarily shows the difference Greenland is subject to seasonal variations and

Fig. 4 Glacial earthquakes
in Greenland (Modified
after Tsai and Ekström
(2007) and Nettles and
Ekström (2010))
synchronous with maximum glacier flow veloci- front; force direction was parallel to the rift prop-
ties in July–August and a corresponding mini- agation direction. At Vanderford and Ninnis gla- S
mum in January and February (Nettles ciers, CSF and local ice flow directions coincided
et al. 2008). A power law does not properly with only one exception. The magnitudes of the
describe the magnitude/CSF – frequency rela- glacier earthquakes in Antarctica were
tion. It seems that each outlet glacier has its 4.2 Ms 4.9.
typical earthquake, which depends on glacier
size, hydrological conditions, and the calving
rate (Tsai and Ekström 2007). Global Landslide Earthquakes
Chen et al. (2011) detected 13 more new-class
glacial earthquakes in Antarctica with epicenters Ekström and Stark (2013) applied the detection
at or near the Ronne Ice Shelf, Ninnis Glacier, and location method developed for the glacial
and Vanderford Glacier (Fig. 5). At the Ronne Ice earthquakes to known large landslides or areas
Shelf, the direction of the CSF vector was normal where a potential for giant landslides can be
to the ice flow direction and parallel to the ice assumed. The landslide force history F(t) was

Fig. 5 Glacial earthquakes
in Antarctica (Modified
after Nettles and Ekström
(2010) and Chen
et al. (2011))
discretized into a sequence of overlapping isos- correlates with the landslide mass M and the
celes triangles with a half-width of 10–15 s. magnitude Ms (Fig. 7).
Momentum, displacement, and CSF were derived
from F(t). Satellite images and other topographic
information were used to separate the factors Mechanisms of Giant-Single Force
D and M out of the product CSF. Figure 6 Sources
shows, as an example, the Hunza-Attabad land-
slide in North Pakistan with a total mass of Landslide: Rotating Slider Block Model
140 109 kg. Ekström and Stark (2013) analyzed Planar slider block models have been proven to
seismic recordings of 29 giant landslides in the support understanding of earthquake processes
Aleutian and Coast Ranges, the Rocky Moun- considerably. When considering the application
tains, the Andes, the Alpine – Himalayan orog- of a planar slider block to earthquake processes,
eny, Taiwan, New Guinea, and Antarctica, the slider block is loaded by tectonic strain, and
with the Mount St. Helens 1980 landslide after strain release the slider block comes to rest.
included. The magnitudes of these events range For landslides, the situation is different; on an
from Ms = 4.6–5.6. The CSF can be calculated inclined planar sliding plane the slider block
from the data provided by Ekström and Stark accelerates constantly and the travel distance is
(2013). CSF or log10(CSF) significantly not confined. A rotational slider block overcomes
a KBL-IU ABKT–II
BHZ LHZ–00
KBL-IU ABKT–II
LONG TRAN
UOSS–II
KBL-IU LHZ–00
TRAN
ARU–II
KZA–KN LHZ–00
LHZ
GNI–IU
AAK–II LHZ–00
LHZ–00 KIV–II
LHZ–00
AAK–II
LONG KIV–II
TRAN
AAK–II
TRAN ANTO–IU
LHZ–00
EKS2–KN ANTO–IU
LHZ TRAN
−200 −100 0 100 200 300 400 500 600 0 200 400 600 800 1000 1200
seconds seconds
b 0.3
0.2
Force [1012 N]
0.1
0.0
−0.1 North
East
−0.2 Up
0 10 20 30 40 50 60 70
Time [s]
Seismic Sources from Landslides and Glaciers, src and dpo are centroids of the landslide mass M before
Fig. 6 Hunza-Attabad landslide 04 January 2010 and after the slide, D the trajectory derived from satellite
(Modified after Ekström and Stark 2013); (a) observed image; given CSF = M D = 14*1013 kg m the landslide
and synthetic traces; (b) landslide force history; (c) source mass is M ~ 14*1010 kg; note the damming of the Hunza
(red outline) and deposition area (blue) of the landslide; river
Seismic Sources from Landslides and Glaciers, Fig. 7 Relations between landslide data (Ekström and Stark 2013);
(a) Mass – CSF, (b) Ms – log10(CSF)
compensated for by the inertial force M d2a/

dt2 R (Eq. 5). We only look at one half oscillation
of the slider block.
M g sin ða0Þ M g cos ða0Þm0 ¼ 0 (4)
M g sin ðaÞ M gð cos aÞm

¼ M R d2 a=dt2 (5)
We consider a first-order approximation for

small amplitudes (sin(a) ~ a, cos(a) ~ 1) and
travel distance D ~ R (a a0). Introduction of
Seismic Sources from Landslides and Glaciers,
Fig. 8 Rotational slider block model (for description of these approximations and subtraction of Eq. 4
symbols see section “Landslide: Rotating Slider Block from Eq. 5 gives
Model”)
ðR=gÞd2 a=dt2 þ ða a0Þ ¼ m0 m ¼ Dm (6)
this problem and confines the travel distance
(e.g., Br€
uckl and Parotidis 2005). The solution of the inhomogeneous differen-
Figure 8 shows the geometry of a rotational tial equation for the initial conditions a = a0 and
slider block. The dynamic system of a landslide is d/dt(a) = 0 is given by Eqs. 7 and 8.
approximated by a mathematical pendulum with
the length R and the mass M concentrated to a aðtÞ ¼ a0 þ Dmð1 cos ðo0tÞÞ (7)
point. The position of the mass is defined by the
angle a times R. Before the landslide happens at o0 ¼ 2p=T0 ¼ sqrtðg=RÞ (8)
a = a0, the driving gravitational force M g sin
(a) is in equilibrium with the frictional force The position angles a of the slider block at
m0 M g cos(a), which always acts in opposition to start time t0, stop time t1 = T0/2, and the time of
the direction of movement (Eq. 4). A decrease of maximum velocity tm = T0/4 are a0,
the friction coefficient from m0 to m initiates the a1 = a0 + 2 Dm, and am = (a0 + a1)/2 = m =
landslide. The excess driving force is atan(CSFz/CSFh). The travel distance D and
maximum magnitude of the landslide force previously calved icebergs away from the ice
(positive for t = t0, negative for t = t1) are front out to sea. This toppling of large icebergs
given by Eqs. 9 and 10. has been considered a candidate for a seismic
source of global scale (Tsai et al. 2008).
D ¼ CSF=M ¼ Rða1 a0Þ ¼ 2 RDm Figure 10 illustrates the geometry, kinemat-
ics, and acting forces during the toppling of an
¼ 2Dm gðT0=2pÞ2 (9)
iceberg, leaving it in unstable equilibrium after
calving from the front of an ice stream. The
Fmax ¼ M g Dm ¼ g CSF=ð2RÞ calving front may or may not be grounded. The

side lengths of the cross section are H (the
¼ p2 CSF= 2ðT0=2Þ2 (10)
height before toppling) and L. The width of
the iceberg is W. With H > L, the iceberg is
The half-period T0/2 of the rotational slider initially in an unstable position. Toppling of
block is the duration of the slide or the landslide the iceberg implies not only a rotation
force history T. We introduce a calibration factor of ~90 but also acceleration off the glacier
k < 1 (Eq. 11) to account for the observation of front. This inertial force causes a reaction
the landslide force history F(t) not starting and force at the contact area of the iceberg with
ending with the extreme value predicted by the the glacier, which is then transmitted via the
slider block model. glacier ice to the glacier bed and represents the
main part of the glacial seismic source. A first-
T0=2 ¼ k T (11) order approximation of CSF:
A calibration factor of k = 0.68 brings Fmax,

CSF MðH LÞ=2, with M
calculated in Eq. 10, into a nearly perfect 1:1
¼ H L W di (12)
relation with the Fmax derived from the inverted
force–time function (Ekström and Stark 2013;
Fig. 9a). Figure 9b shows the dependence of am The density of ice is di and that of seawater dw.
and a0 on CSF. Over the range of CSF The characteristic time scale of toppling can be
(approximately three decades), the average slid- estimated from the period of an iceberg
ing angle reduces from ~25 to ~10 . There is interpreted as a mathematical pendulum:
also a systematic increase of ratio (a0–am) to
a0 with CSF from ~28 % to ~55 %.
T 2p sqrtðdi H=ððdw di ÞgÞ (13)
Calving Glacier: Toppling of an Iceberg Equations 12 and 13 do not take into consid- S
The large outlet glaciers of Antarctica and Green- eration the remaining kinetic energy the iceberg
land lose mass mainly by calving. One of the has after reaching its stable position at the hori-
best-monitored sites where calving takes place zontal travel distance (H-L)/2. Furthermore, the
is the front of the Jakobshavn Glacier near friction between the iceberg and the glacier ice
Ilulissat, West Greenland. On 28 May 2008, a front along with the interaction of the iceberg
gigantic sequence of calving events was filmed with seawater and the mélange of smaller ice-
by a team of Chasing Ice (http://earthsky.org/ bergs and fragments in front of the calving front
earth/video-largest-glacier-calving-ever-caught- is not taken into account. However, assuming the
on-film). An area of the size of Manhattan calved very reasonable dimensions H = 800 m,
successively over a time span of 75 min. One can L = 100 m, and W = 400 m, CSF ~ 1013 kg m
observe a great number of slabs of about 1000 m (Eq. 12), T ~ 150 s (Eq. 13), and Ms = 4.8
in height toppling from unstable to stable equi- according to the relation between log10(CSF)
librium, thus pushing the iceberg mélange of and Ms (Fig. 7b).
Seismic Sources from Landslides and Glaciers, Fig. 9 Relations derived from the rotational slider block model; (a)
maximum force observed – maximum force calculated; (b) a, a0 – log10(CSF)
Microearthquakes Generated by phases of varying levels of activity. A deeper

Landslides understanding of landslide processes and evolu-
tion is of fundamental importance for early warn-
The dynamic collapse of a slope within several ing and hazard estimation. Seismic activity
tens of seconds is only representative of how a connected to these processes can be observed by
landslide can end in the most extreme cases. local seismic networks deployed on the landslide
However, the majority of landslides do not mass. We term microearthquakes generated by
develop into dynamic and catastrophic failure landslides LMEs (landslide microearthquakes)
but reach more or less stable final states after in the following.

L a
Fig. 10 Toppling
(capsizing) of an iceberg
acting as single force
seismic source; (a)
kinematics and acting force D
(toppling may also occur by H
counterclockwise rotation);
(b) toppling of an iceberg F
during the 28 May 2008
calving event at
Jakobshavn glacier glacie
r bed
Green
land
glacier be
d
Antarctica
b
∼ 100 m
∼ 1000 m
Landslide Processes and Evolution physical states are depicted in Fig. 11a in a Mohr
We assume brittle rheology of the rock material diagram by the three Mohr’s circles (M1, M2,
constituting a landslide as a precondition for seis- M3) and Mohr-Coulomb failure criteria
mic activity. However, under some circum- representing the initial and degraded strengths S
stances, the landslide mass may behave as of the rock mass forming the slope. The Mohr’s
ductile, and even a brittle deformation will not circles symbolize the states of stress, which are
necessarily act as a seismic source. In the follow- relevant for the stability of the slope. The initial
ing, we present a simplified scheme of landslide rock strength is given by a Mohr-Coulomb failure
processes and possible evolutions to dynamic criterion for the shear stress, with C0 and f0 the
failure or stabilization (Fig. 11). This schema initial cohesion and the initial friction angle.
should provide a backbone for the interpretation Increasing pore pressure p shifts the Mohr’s cir-
of the seismic data provided by the case studies cles to the left and nearer to the initial rock
presented in sections “Pre- and Postfailure Pro- strength. In an extreme case of high pore pres-
cesses at Steep Slopes” to “Flow”. sure, a dynamic rupture could be induced; how-
Slopes of a given inclination and height can ever, this is not the only situation that reduces the
only exist if the stability is guaranteed by the slope stability. Subcritical crack growth (Br€uckl
initial strength of the rock material. The possible and Parotidis 2005) supported by deep chemical
Seismic Sources from Landslides and Glaciers, Fig. 11 Landslide processes and evolution; (a) reduction of rock
mass strength; (b) temporal evolution of displacement
or thermomechanical weathering (Gischig quasistationary creep, eventually interrupted by

et al. 2011) may progressively damage the rock episodic phases of high creep or sliding veloci-
material and degrade the rock strength. Ulti- ties, follows. However, we cannot exclude that
mately, cohesion (C0) is completely lost and the episodic high activity is the precursor for a later
friction angle diminished considerably from bifurcation between quasistationary creep and
f0 to f. This failure criterion of the degraded dynamic failure, as indicated in Fig. 11b by a
rock strength intersects Mohr’s circle M1 and is second grey dot. The longer the quasistationary
tangential to M2, which means that in the first phase lasts, the more the driving forces and the
case a dynamic failure has already occurred. In radius of the Mohr’s circle reduce, so that in the
the second case, the slope is in an equilibrium final state of stress represented by M3 the slope
state of driving (gravitational) and resistive stabilizes. Decrease of the pore pressure p by
(frictional) forces. Creep of the whole slope improved drainage of the slope due to increased
may reduce the inclination and therefore the driv- permeability of the rock mass may support
ing forces, and this situation is symbolized in stabilization.
Mohr’s circle M3. It is well below the failure
criterion of the degraded rock. The slope comes Rockfall
to an ultimate stabilization by this process. Rockfall is a phenomenon occurring on slopes
The temporal evolution of the deformation or where, at least locally, the inclination equals or
displacement of a landslide mass is schematically exceeds that of talus of the same rock material.
shown in Fig. 11b. The initial phase corresponds It may be the consequence of weathering but also
to the progressive damage from initial to an indication of increased activity in a landslide.
degraded rock strength (Fig. 11a), where the The seismic signature of rockfall may be a
whole rock mass of the landslide is degraded sequence of many irregular impulsive signals
and deformed. However, deformation concen- moving down the slope. Figure 13 shows seismic
trates successively to one basal or several main recordings from the tumbling of a medium-sized
sliding zones or surfaces. Toward the end of the boulder (~2 m diameter) at the Gradenbach site
initial phase, the further development of the slope (Fig. 12), images of the boulder itself and the site
comes to a bifurcation. In the case where Mohr’s of heaviest impact, and the seismic location of
circle M1 represents the critical state of stress in that place (Br€uckl et al. 2013).
the slope, the initial phase leads directly to a As a second example, the seismic signature of
dynamic slope failure. In the case of Mohr’s a rockfall at Steinlehnen (Fig. 12) is documented
circle M2-type critical state of stress, a phase of in Fig. 14. This rockfall was more of a small rock
Seismic Sources from Landslides and Glaciers, Fig. 12 Monitoring sites in the European Alps
avalanche, which left behind a path resolved by about 1 h before the collapse. Frequency of
ground-based InSAR monitoring. The whole detected seismic signals and the seismic energy
event lasted around 30 s, and location applied to obey inverse Omori’s laws. The data suggests
several time windows allowed for the tracing of that a warning based on increased activity could
the path of this rockfall. There is fair agreement be given ~10 min before the collapse and a reli-
between both tracing methods (Weginger 2012). able prediction of the rupture time about 1 min in
advance, provided data acquisition and continu-
Pre- and Postfailure Processes at Steep ous updating of calculations are done in real time.
Slopes The village of Randa, Matter Valley, Swiss
We denote “steep slopes” as slopes that are Alps (Figs. 12, 15), was threatened by rock- and
steeper than the initial angle of friction or the icefalls several times in its history. We anticipate
inclination of talus built up by the same rock a look from rock to ice and icefalls because of the
material. The rock material of these slopes similar phenomenology behind these processes in
needs cohesion to balance the gravitational rock and ice. Icefalls were released by the S
forces. In the case of cohesion dropping to zero Weisshorn hanging glacier five times in the last
because of progressive damage (Fig. 11a), the 35 years. The break-offs in 1973 and 2005 were
slope will immediately transit from this initial monitored by geodetic and seismic methods. The
phase to dynamic failure (Fig. 11b). Identification velocity of the glacier front developed according
and monitoring of precursors of this kind of to an inverse Omori law (Faillettaz et al. 2011)
dynamic rupture would be of great scientific and toward the instant of rupture. Log-periodic oscil-
practical importance. lations appeared superimposed on the general
The retreat of cliffs along the British and trend. Starting 30 days before rupture, three
French Channel Coasts by sea erosion–driven phases characterized by different cumulative
rockfalls is an ongoing process. Amitrano size-frequency distribution of the signal energies
et al. (2005) report seismic monitoring of a cliff and interevent waiting-time distribution behavior
collapse at Mesnil-Val on the Normandy coast, were identified and a correlation of seismic activ-
France. Precursory seismic activity was recorded ity with the log-periodic oscillations established.
2 sec
a b
6
8500 PDF
seismic station
c d
8000 0.95
7500 0.9
7000 5 0.85
4 1
[m]
6500
6 3 0.8
6000
2
0.75
5500
0.7
5000
0.65
4000 4500 5000 5500 6000 6500 7000
[m]
Seismic Sources from Landslides and Glaciers, local network; (c) main impact; (d) location by the seismic
Fig. 13 Rockfall at the Gradenbach landslide (Br€uckl data (normalized PDF) and actual position of the impact
et al. 2013); (a) main boulder; (b) recordings by the (star)
These phases represent the development from year 1991 with ~10–20 mm/year (2004–2008).
random crack generation over crack concentra- A seismic monitoring campaign was started in
tion at a fracture zone to the final rupture. The January 2002 and continued until July 2004
findings could be the basis of an efficient in-time (Spillmann et al. 2007), where a monitoring net-
warning. work was deployed above the scarp (Fig. 15).
From the detour to ice and a hanging glacier, Two types of waveforms classified as LMEs
we revert to rock and landslides. On 18 April and were observed: 94 single events comparable
9 May 1991, two successive rockslides took place to the recordings of small detonations and
near Randa (Fig. 15). The failure volumes 129 multiple events with a more complex
involved about 20 and 10 106 m3 massive para- source time function. Frequency content of the
and orthogneisses (Eberhardt et al. 2004). The signals reached up to 100 Hz for the surface
average slopes of the failure planes were roughly sensors and up to 200 Hz for the borehole sensors.
60 and 50 (Fig. 16). The rock masses dropped Most of the seismic activity was concentrated
down 200 and 300 m and produced Ms 4.9 and on a volume of low seismic velocities
4.8 earthquakes detected by Ekström and Stark (<= 1500 m/s) near the scarp, but also deeper-
(2013). Deformation of the slope above the actual reaching zones of increased seismic activity
scarp continued after the rockfall sequence of the were identified, possibly outlining the scarp of a
GB-SAR raw data image (time in CET)

a 13.06.10 22:26:31 - 13.06.10 22:32:52, t2 = 1000 b SL 03
3
1400
DB-SAR: differential displacement [mm]

2
1200 SL 01
1
SL03
Y [m]
1000
0
SL01
800
−1 SL 02
SL02
600
−2
400 SL0* seismic station

−3
−800 −600 −400 −200 0 200 400 600 800 60 sec
X [m]
Seismic Sources from Landslides and Glaciers, image shows the trace of a rockfall (black arrow); yellow
Fig. 14 Rockfall at the Steinlehnen landslide arrow represents the path of the rockfall from the seismic
(Weginger 2012); (a) differential ground based InSAR data shown in (b)
Seismic Sources from Landslides and Glaciers, Fig. 15 View from east to the glacier break-off and rockslide sites
near Randa; location of the profile in Fig. 16 is shown by a dashed red line
w E
2500
Scarp (May 9, 1991)

iss
2250 a gne
Par
Scarp (April 18, 1991) PDF
−3
eiss 10
h ogn −4
Or t 10
−5
2000 10
e iss
agn −6
Par
10
−7
10
1750
s
eis )
t h ogn neiss
Or a G
nd current topography
(Ra
1500
topography before
rockslide
200 m
Seismic Sources from Landslides and Glaciers, LMEs is shown within the detail marked by black rectan-
Fig. 16 Cross section through the Randa rockslide site gle (Modified after Spillmann et al. 2007); the location of
(Modified after Eberhardt et al. 2004); cumulative PDF of a possible next scarp is marked by red dots
future rock slide (Fig. 16). Hydrological and band slumping (Goodman and Kieffer 2000) can
meteorological conditions had no impact on the be summarized under DSGSD. The initiation of
seismic activity and the geodetically observed DSGSD at alpine slopes mostly dates back to the
displacement rates. Most magnitudes were in retreat of the ice age glaciers, when the lateral
the range 2.0 < Mw < 0.5. It is questionable support of the valley flank by the ice pressure was
if the observed seismic activity and geodetically lost. This conception implies a long initial phase
observed displacements are postfailure relaxation of mass movement, during which progressive
processes or if they belong to the initial phase of a damage degraded the rock strength almost to its
third, larger rock slide at Randa. residual value (see Fig. 11a; Br€uckl and Parotidis
2005). DSGSDs may be in a quasistationary
DSGSD: Deep-Seated Gravitational Slope creep phase or may exhibit episodic high defor-
Deformation mation rates in parts of the slope or its entirety
Deep-seated gravitational slope deformation interrupting longer phases of little activity
(DSGSD) is a phenomenon frequently observed (Fig. 11b). There is currently no general answer
in the hard rock slopes of alpine regions. Typi- as to whether these processes eventually lead to
cally, the deformation comprises the whole flank stabilization or a transition to dynamic rupture.
of a valley, and the maximum thickness of the In the following, we consider the seismic sig-
deformed rock mass exceeds 100 m. The mecha- nals recorded at three DSGSD examples in crys-
nisms of DSGSD are manifold, and deformation talline rocks of the European Alps: Séchilienne
styles such as “Bergzerreißung,” “Talzuschub,” (Helmstetter and Garambois 2010), Steinlehnen
sagging or Sackung (Zischinsky 1969), and kink (Weginger 2012), and Gradenbach (Br€uckl

Fig. 17 View from south
to the Séchilienne DSGSD
and the active zone (area
colored in red); the
locations of the 2 seismic
micro-arrays RUI and THE
are marked by symbols
et al. 2013; Mertl 2015). The locations of these high-velocity area or the head scarp. It is also
landslides are shown in Fig. 12; Figs. 17, 18, and possible that some hypocenters are located at or
19 supply information about the morphology, the near a basal shear zone. Rockfall and LME fre-
seismic monitoring network, and partly the geol- quencies precede accelerations of the active zone
ogy and structure of these landslides. The average by 4–5 days. Some but not all rainfall events
slopes of these DSGSDs vary between 28 trigger increased rockfall and seismic activity
(Gradenbach) and 35 (Steinlehnen). (Helmstetter and Garambois 2010).
At the Séchilienne DSGSD, a presently very Seismic monitoring campaigns at
active zone is located on top of the steeper lower Steinlehnen and Gradenbach had the opportu-
part of the slope (marked in Fig. 17). The velocity nity to observe episodic phases of high landslide
of this active zone reached 1.4 m/year in 2008. velocities interrupting longer periods of
Until April 2009, rockfalls were the most fre- quasistationary creep. During June 2010, the S
quent seismic sources related to the dynamics Steinlehnen DSGSD was monitored by a
recorded at the Séchillienne DSGSD. The cumu- ground-based InSAR system installed opposite
lative frequency of >3000 events obeys a power the Steinlehnen DSGSD. An acceleration phase
law distribution over 2.5 decades of the ampli- of the active slab was observed with total dis-
tude range with an exponent corresponding to a placements reaching up to 0.9 m. Up to five
b-value of b = 1.1. seismic monitoring stations were operating at
Figure 20 shows typical waveforms of LMEs the Steinlehnen slope during this period
at the Séchilienne DSGSD. One type corresponds (Fig. 18a). Besides the rockfall (see Fig. 14),
to impulsive sources similar to small explosions, Weginger (2012) identified three classes (n, ln,
while another event type shows mostly emergent i in Fig. 21a) of LMEs that had their sources
onsets and represents source-time functions within the area of the accelerated slab. This
lasting ~5 s, either relatively broadband or low classification was based on characteristic prop-
frequency. The LMEs cluster either in the erties of the seismic signals (e.g., duration,

Fig. 18 Steinlehnen
DSGSD (Weginger 2012);
(a) view from east to the
slope with contour of the
active zone; triangles mark
the locations of the seismic
monitoring stations, the
extent of the slab
accelerating during the
observation period is
colored in ochre; (b)
geological cross section
A-B (Modified after
Zangerl et al. 2007)
frequency spectral characteristic, amplitude, 21 June, and rockfall ceases completely thereaf-
envelopes, cross correlations, polarization). ter. The cumulative frequencies of the events n,
Figure 21b shows velocity and displacement of ln, i increase more or less continuously with the
the highly active slab for the period of highest increments during the high-velocity
10–28 June 2010. The high-velocity phase starts phase. No extraordinary rainfall was reported
on 20 June and lasts until 25 June. The cumula- to precede or accompany the period of high
tive frequencies for LME types n, ln, and i rise velocities of the active slab, and no external
from 13 June onward, and some rockfalls also trigger of this phase is known so far. The incip-
occur (Fig. 21c). A sudden increase of the ient seismic and rockfall activity on 13 June was
cumulative rockfall frequency accompanies the most likely the only precursor of the high-
onset of the high-velocity phase between 20 and velocity phase in the following week.
Seismic Sources from Landslides and Glaciers, of the basal sliding surface or zone was determined by
Fig. 19 Gradenbach DSGSD (Br€ uckl et al. 2013): (a) drilling (boreholes E2 and GB) and refraction seismic
Orthophoto with contour of the active zone and seismic measurements (c) displacement history derived from ter- S
stations GB01–GB07; the main scarp, the highly fractures restrial geodetic measurements 1969–1991, photogram-
zone in the uppermost area of the mass movement, and the metric evaluation of the aerial photos 1962 and 1996,
erosional zone at the toe of the mass movement are well and GPS since 1999
visible in the orthophoto; (b) cross section A-B; the depth
The episodic high–sliding velocity phase at for the data from the Steinlehnen site (Mertl
the Gradenbach DSGSD occurred during spring 2015). The cumulative frequencies of these
of 2009. In contrast to the Steinlehnen event, the events behave distinctly. Event types A and
whole landslide became active due to infiltration D start with the onset of snow melt (Fig. 22b–d)
of melting snow and precipitation (Br€uckl and high infiltration. The tremor type was only
et al. 2013). Five different LME types (AA, A, observed at that time. Thereafter, a quiet phase
B, D, tremor; Fig. 22a) besides rockfall (rf) were that lasted over 1 month followed. The beginning
identified using criteria comparable to those used of the episodic phase of high creep velocity was
a 100
[Hz]
50
0
33 34 35 36 37 10 20 30 40 50
b 100
50
0
29 30 31 32 33 10 20 30 40 50
c 100
50
0
204 205 206 207 208 170 180 190 200 210 220 230
d 100
50
0
135 136 137 138 139 110 120 130 140 150 160
[sec] [sec]
Seismic Sources from Landslides and Glaciers, after Helmstetter and Garambois 2010); (a) impulsive
Fig. 20 Typical waveforms and spectrograms from LME; (b) small explosion; (c) emergent and broad band
LMEs recorded at the Séchilienne DSGSD (Modified LME; (d) emergent and low frequency LME
accompanied by type B microearthquakes and landslide mass could increase in velocity (Mertl
some rockfall. The frequency of type A and 2015).
D events rose 1–2 weeks later and lasted for the
rest of the episodic high-creep phase. Magnitudes Flow
of the LMEs range from 2.8 to 0.9 with a A flow is a spatially continuous movement in
b-value of b = 1.7. which surfaces of shear are short lived, closely
Though the geodetically observed movements spaced, and not usually preserved. Whereas in
covered the whole landslide mass, the spatial DSGSD the original structure of the rock mass
distribution of the seismic activity during the is partly preserved, in a flow it is completely
episodic high-creep phase is concentrated to lim- destroyed and mixed up. The ratio of mean thick-
ited areas (Fig. 23). Event types A, D, and tremor, ness to length of a flow is <0.01 and significantly
which coincide with the onset of snow melt and lower than for DSGSDs. When considering the
high infiltration, are most likely very shallow scheme of landslide processes and evolution
events. The waveforms and coincidence in time presented in Fig. 11a, it is evident that the initial
with the onset of high creep velocities may be rock (or soil) strength has been completely
indications that the hypocenters of B-type LMEs reduced to its residual values. Changing
are at the level of the basal sliding zone. Asper- pore pressure may keep the flow either active
ities must have been overcome before the whole or temporarily inactive (Mohr’s circles M2 and
Seismic Sources from Landslides and Glaciers, (irregular sequence of impulses over a period of several
Fig. 21 LMEs observed at the Steinlehnen DSGSD, seconds); (b) displacement and velocity; (c) development
10–28 June 2010 (Weginger 2012); (a) typical waveforms of cumulative frequency of event types n, ln, i, and rf
and spectrograms of LME types n (emerging signal, few (rockfall)
seconds length), ln (irregular sequence of n-type signals), i
M3 in Fig. 11a). Flow here represents quakes.” They have an impulsive onset, last
quasistationary creep including episodic high 2–5 s, and cover a wide frequency band
activity, depending on infiltration and the pore (10–80 Hz). Later-arriving phases have lower
pressure (Fig. 11b). The distribution of velocities frequencies (Fig. 25a). The epicenters are con-
in space and time within the landslide mass jus- centrated around the high-velocity area of the
tifies its macroscopic description as a ductile landslide. Magnitudes between 3.2 M
medium. However, as the following two exam- 1.3 were detected following a power law distri-
ples will prove, brittle deformations or interac- bution with a b-value of b = 0.84. The hypocen-
tions between pore fluid and the rock/soil mass ters were probably located within the uppermost
also exist and resemble seismic sources. few meters and not at the basal sliding zone or S
The Super-Sauce landslide is a flow-type surface. The highest magnitudes of slide quakes
mass movement situated in the Barcelonnette followed few hours after a strong rainfall event;
basin in the southern French Alps (Fig. 12). It the temporal distribution was otherwise uniform.
developed from heterogeneous soft Jurassic black Single miniarrays also detected very high-
marks (Fig. 24). The average slope of the flow is frequency events (5–150 Hz) with durations of
~23 . The landslide triggered around 1960, ruling 2–20 s (Fig. 25b). These LMEs were concen-
out a causal connection to the retreat of ice age trated around a stable crest within the high-
glaciers. From 1996 to 2007, the velocities covered velocity zone of the flow and related to fissure
the range of 2–30 mm/day. development in the uppermost layer of the flow.
Seismic monitoring at the Super-Sauce land- Laboratory tests proved that the flow material can
slide was carried out between 14 and 24 July have brittle rheology.
2008 (Walter et al. 2012). Besides rockfalls, The second example of a flow is the Slumgul-
34 LMEs were located and labeled as “slide lion landslide (Fig. 26). The landslide occurs in
Seismic Sources from Landslides and Glaciers, but significantly higher frequency content and shorter
Fig. 22 LMEs observed at the Gradenbach DSGSD, duration than B), tremor (emergent low frequency signal,
March–October 2009 (Br€ uckl et al. 2013; Mertl 2015); duration >1 min, superimposed by irregular spikes), rf
(a) typical waveforms and spectrograms of LME types (rockfall); (b) displacement and velocity; (c) snow cover
AA (impulsive onset with short lower frequency coda), water equivalent and precipitation; (d) development of
A (like AA but with long coda), B (emergent onset, low cumulative frequency of event types
frequency, typical duration 20–30 s), D (emergent onset

Fig. 23 Spatial
distribution of seismic
activity March–October
2009 at the Gradenbach
DSGSD; colored contour
plot visualizes the
frequency of possible
epicenters with normalized
PDF > 0.7, event types A,
AA, D; blue area marks
area of event type
B epicenters (Br€uckl
et al. 2013; Mertl 2015)
Seismic Sources from Landslides and Glaciers, Fig. 24 Super-Sauce flow, red dotted line outlines landslide area
(Walter et al. 2012)
Tertiary volcanic rocks and comprises sandy, which moved over an older, inactive lower part
silty clay with scattered patches of boulder in its path. The average slope of the active part is
debris, clay, and pond and stream sediments. 13 . The Slumgullion moves primarily by sliding
The landslide consists of an active upper part, along discrete bounding faults. Additionally,
Seismic Sources from Landslides and Glaciers, Fig. 25 Typical waveforms and sonograms of LMEs observed at
Super-Sauce debris/mud flow (Modified after Walter et al. 2012); (a) “slide quakes”; (b) “fissure development”
differential internal processes take place at faults diminishes toward 125 Hz. High amplitudes of
and fractures within the landslide body. tremor envelopes last <10s and follow irregu-
During 18–26 August 2009, a seismic larly in intervals of several tens of seconds. Har-
monitoring campaign was carried out at the monic slide quakes last ~2 s, their fundamental
Slumgullion landslide (Gomberg et al. 2011). frequency is ~12 Hz, and no harmonics are visi-
Geodetic and hydrometeorological monitoring ble throughout the whole duration. This LME
supplemented the campaign. During the monitor- type occurs more frequently during daytime,
ing campaign, the displacement trend was and an increase in sliding velocity is also
~10 mm/day. One stronger period of rainfall observed during the day. Tremor and harmonic
occurred during the observation period. Several slide quakes most probably originate at the left
seismic signals were detected and related to side-bounding strike-slip faults of the landslide.
LMEs: “tremor” and “harmonic slide quakes” The stick–slip behavior of the side-bounding
(Fig. 27a, b). The dominant energy of tremor is strike-slip faults is explained by a transient dilat-
distributed broadly above 30–50 Hz and ant strengthening.
Seismic Sources from Landslides and Glaciers, Fig. 26 Slumgullion landslide (37 590 3000 N, 107 150 2500 W),
San Juan Mountains, Colorado, USA (Gomberg et al. 2011)
a b A03
P05 A05
P03 A07
O11 A11
O07 B02
B06
O04 C03
N10 C05
N05 C07
N03 D02
D06
M11 E03
M05 E05
M03 E07
L06 E02
L03 E06
G03
K11 G05
K05 G07
J10 H02
J06 H06
J02 I03
I05
S
I07 I07
H08 J02
G09 J04
G05 J06
J08
F08 K03
F02 K05
E07 K07
E03 L02
D06 L04
M03
D02 M05
C05 M07
B08 N02
B04 N04
O03
A11 O05
A07 P02
A03 P04
0 20 40 60 80 100 120 140 0 1 2 3
Seconds from 08/25/09 07:48:00 Seconds from 08/25/2009 17:10:46.73
Seismic Sources from Landslides and Glaciers, Fig. 27 Typical LME waveforms observed at Slumgullion flow
(Modified after Gomberg et al. 2011); (a) envelopes of tremor; (b) harmonic slide quake
Microearthquakes Generated by bed (Fig. 28b). For temperate glaciers, a thin

Glaciers coat of water between ice and the bedrock is
assumed, and sliding occurs over the thin viscous
Ductile and Brittle Deformations of Glaciers water layer (Fig. 28c). Sliding at the glacier bed is
In contrast to landslides, which are intrinsic tran- controlled by obstacles, which are overcome
sient phenomena, glaciers or ice sheets can reach either by enhanced flow due to stress concentra-
a long-lasting stationary state in steady climatic tion at the obstacles or regelation. All processes
conditions. Snow is transformed to ice and added related to glacier flow described so far are either
to the glacier in the accumulation area and ductile (internal deformation of the ice, viscous
removed in the ablation area. Glacier flow trans- flow of the water coat at the glacier bed, enhanced
ports ice from the accumulation area to the abla- flow over obstacles) or correspond to another
tion area, thereby keeping the total mass as well steady-state process (regelation) and cannot pro-
as the thickness of the glacier constant under duce earthquakes.
stationary conditions. The glacier flow is driven However, the rheology of ice changes signifi-
by gravitational forces, controlled by the rheol- cantly in the case of the hydrostatic pressure
ogy of the ice and physical conditions that allow component p of the state of stress becoming neg-
for sliding of the glacier ice over its bed (Cuffey ative. In this case, the ice behaves in a brittle
and Paterson 2010). manner and may radiate seismic energy. First,
We consider an infinite slab of ice with con- we consider tensile crack nucleation and propa-
stant thickness H resting on a plane glacier bed gation without the influence of water. Glacier
with the slope a. Shear stress parallel to the gla- thickness adapts to varying slopes of the glacier
cier bed is proportional to sin(a) and increases bed by keeping the basal shear stress approxi-
linearly with the overburden (Fig. 28a). mately constant. Therefore, the thickness is
According to Glen’s law, the corresponding inversely proportional to sin(a). Neglecting accu-
shear strain rate is proportional to the nth power mulation or ablation over a relatively short dis-
of the shear stress, with n ~ 3. tance of an increasing slope, the velocity must
Integration yields the velocity profile, which is increase to keep the flux of ice constant (Fig. 29).
an (n + 1)th-order parabola. For cold glaciers, the Longitudinal tensile stress must develop to
ice temperature is below melting point at the extend the glacier ice moving from the gentle
base, and the ice may be frozen to the glacier to the steeper slope. Surface crevasses develop
Seismic Sources from Landslides and Glaciers, cold glacier – no basal sliding; (c) velocity profile of a
Fig. 28 Planar slab of ice resting on a sloping plane; (a) temperate glacier – basal sliding
shear stress parallel to surface; (b) velocity profile of a
Seismic Sources from Landslides and Glaciers, Fig. 29 Crevasse formation due to thickness adaption to increasing
slope; stars mark possible locations of GMEs
to the depth at which the hydrostatic pressure (Alley et al. 1987). The mechanical strength of
equals the longitudinal tensile stress component. glacial till obeys the Mohr-Coulomb failure cri-
On alpine glaciers, surface crevasse depths sel- terion and may be comparable to the masses of
dom exceed 25–30 m. After reaching the zone of flow-type landslides treated in section “Flow.”
constant steeper slope, the tensile stress compo- Although steady shear deformation may domi-
nent disappears and the crevasses close by ductile nate, experience from seismic monitoring on
deformation. landslides shows that stick–slip deformation
Water significantly influences the mechanics may also occur. At the glacier beds of
of landslides and glaciers. The fundamental dif- polythermal and temperate glaciers, there may
ference between landslides and glaciers with also be asperities or “sticky spots,” which induce
respect to the effect of water is that buoyancy is stick–slip basal sliding under the otherwise
lower than hydrostatic pressure in landslides and nonseismic deformation conditions at the
higher in glaciers due to the comparative densi- glacier bed. S
ties of water (~1000 kg m3), rock In the following, we term seismic activity
(~2500 kg m3), and ice (~900 kg m3). There- connected to the various processes of glacier
fore, a water column of sufficient height can flow and observed by local networks GMEs
introduce a tensile stress regime and lead to (glacial microearthquakes).
crack development and propagation at all glacier
depths. Possible scenarios are shown in Fig. 30. Glacial Lake Outburst Floods at an Alpine
Thus far, we have only considered hard- Glacier
bedded glaciers where the glacier bed consists As an example of seismic activity of an alpine
of hard rock and is not deformed by the glacier glacier, we review observations made at the
flow. However, it is well known that fast-moving polythermal Gornergletscher, Monte Rosa Mas-
outlet glaciers in Alaska, Antarctica, and Green- sif, Switzerland (Fig. 12). At the confluence of
land rest on a basal layer of soft glacial till, which Gornergletscher and Grenzgletscher, there is the
makes up a considerable part of the basal sliding ice-dammed Gorner lake (Fig. 31). The lake
Seismic Sources from Landslides and Glaciers, influx from active moulins or basal channels (b) closure of
Fig. 30 Hydraulically induced fracturing in a glacier internal and basal crevasses due to decreasing water pres-
(Modified after Walter 2009; Walter et al. 2010); stars sure; the collapse of a basal cavity generated during the
mark possible locations of GMEs; (a) opening of internal high water pressure phase acts as seismic source
and basal crevasses due to high water pressure; meltwater
Seismic Sources from Landslides and Glaciers, SURF A, SURF B, and INT GME clusters agree with the
Fig. 31 GMEs observed at the Gornergletscher before orientation of crevasses; dotted line outlines approxi-
and during the GLOFs 2004 and 2006 (Walter 2009); (a) mately the Gorner lake at the maximum water levels;
location of the GME clusters; fault plane strike of the (b) focal depths of the different GME clusters
drains during annual glacial lake outburst floods the 2004 and 2006 field seasons, about 35000 and
(GLOFs), which were monitored in the years 50000 events were detected. Generally, diurnal
2004 and 2006 with seismic and geodetic net- variations in the seismicity were observed, with
works (Walter 2009; Walter et al. 2010). During the peak in the early afternoon and much lower
GME activity during the nighttime and morning In an analogy to waveforms of volcanic earth-
hours. GME clusters were recorded at three dif- quakes, the low-frequency tremor GMEs were
ferent depth levels: SURF_A and SURF_B at the explained as a fluid-driven crack model with
depth of surface crevasses, INT at ~100 m depth, a resonant water-filled cavity as the seismic
and BAS just above the glacier bed (Fig. 31b). source. Hybrid GMEs were interpreted as
The SURF_A, SURF_B, and INT GMEs have hydrofracturing followed by the rush of water
magnitudes between 2.3 Mw 1.6; the into the new opening.
BAS GMEs reach Mw = 0.7. With the excep- Relative maxima of the daily rate of
tion of BAS GMEs, fault planes are steeply dip- low-frequency events precede the corresponding
ping, and strike directions of the interpreted fault maxima of the high-frequency events by 1–6 days
planes follow the general trend of surface cre- (Fig. 32c). The interpretation of this pattern is
vasses (Fig. 31a). The radiation pattern of the that the peaks of the low-frequency events reflect
SURF_A and INT LMEs can best be explained the subglacial rush in of water leading to a
by the generation of tensile cracks, but the decoupling of the glacier from its bed and
SURF_B events more likely result from a DC consequently enhanced flow velocities. The
mechanism. The BAS GMEs occur on enhanced glacier motion is accompanied by
subhorizontal fault planes and are well explained an increase of brittle failures. Velocity data
by tensile cracks. The source mechanisms of the would be essential to further develop this
SURF_A, INT, and BAS cluster correspond to interpretation.
the processes schematically outlined in Figs. 29
and 30. The BAS events occurred mainly during Stick–slip Movement of an Antarctic Ice
otherwise quiescent night and early morning Stream
hours. This observation supports the idea of a Bindschadler et al. (2003) reported on major
collapsing basal cavity due to the reduced melt- West Antarctic ice streams discharging in sudden
water access and basal water pressure (Fig. 30b). periods of rapid motion, in particular Whillans
The occurrence of DC-source mechanisms with Ice Stream (WIS) (Fig. 33a), which has an area of
the SURF_B GMEs may be due to the particular 200 km 100 km with an average ice thickness
dynamic situation near the Gorner lake and the of ~600 m. During December 2010 and January
GLOF. 2011 and again during December 2011, Winberry
et al. (2013) deployed a GPS and seismic network
Arctic Glacier Prone to Surges on the WIS in order to monitor the slip nucleation
The temperate Bering Glacier flows from the and the subsequent period of rapid motion
St. Elias mountain range to its terminus on the (Fig. 33b). Figure 33c shows the displacements
south-central coast of Alaska and has a history of for a time span of ~1 h, covering the end of a
dramatic surges. Seismic monitoring was carried preceding interevent period, the nucleation S
out between the equilibrium line altitude and the phase, the rupture, the main-slip phase, and the
terminus (Fig. 32a) and covered the early melting transition to the following interevent period.
period from 20 April until 19 June 2007 (West Clear evidence for the nucleation phase is
et al. 2010). Yet another surge started in 2008 shown only at one GPS station (station 1),
(Burgess et al. 2012). where the rupture occurs within ~5 min at all
Around 160,000 events were detected during stations according to the rupture propagation
the observation period. Location of the GMEs speed of >500 m/s. The main-slip phase lasts
was not attempted. The waveforms of the GMEs ~30 min and terminates almost simultaneously
(Fig. 32b) show a bimodal distribution and range at all stations.
from high-frequency signals (20–35 Hz) to GMEs recorded at the individual stations cor-
low-frequency signals (6–15 Hz). The high- relate well in time and frequency domain with the
frequency GMEs have impulsive source func- ice stream velocities (Fig. 34). At station 1, seis-
tions and were interpreted as brittle failures. micity increases at the beginning of the
Seismic Sources from Landslides and Glaciers, stations (red triangles); (b) waveforms; (c) temporal evo-
Fig. 32 Seismic monitoring at the Bering glacier lution of daily rates of high and low frequency event
(Modified after West et al. 2010); (a) location of seismic
nucleation phase but only begins at the other families persist in subsequent high-slip-rate
stations around the rupture time. Two different phases. This observation is an indication of the
styles of GMEs or microseismicity can be distin- temporal stability of subsurface asperities
guished. One type (Fig. 34a) is built by a representing the source regions of the individual
sequence of fewer than a hundred to several thou- events. Other locations show in general a similar
sand individual impulsive events. The wide dis- behavior but are characterized by lower frequen-
tances between the seismic stations did not allow cies, indicating larger source areas. The lack of
for a precise location; however, the relative microseismic activity or GMEs at some locations
timing of P and S wave arrivals and the lack of may be explained by the presence of more con-
crevasses is consistent with source locations at or tinuous fault gouge and higher water pressure at
near the glacier bed. The individual events can be the glacier bed providing basal lubrication.
grouped into families with a high degree of sim- The second style of microseismicity is the
ilarity between each other. The same event emergent tremor (Fig. 34b). It is continuous and
a b c
1
2
0.4
0.3
Displacement (m)
3
Rupture
0.2
Interevent Period
Interevent Nucleation Main-Slip
Period Phase Phase 0.1
−10 0 10 20 30 40 50 60
Minutes Since Slip Initiation
Seismic Sources from Landslides and Glaciers, locations of monitoring-stations; (c) displacements versus
Fig. 33 Geodetic and seismic monitoring of an episodic time since slip initiation; numbers and colours assign
high flow velocity phase at the Whillan Ice Stream (WIS); displacement data to stations in (b) (Modified after
(a) location map; (b) zoom of the monitoring area with Winberry et al. 2013)
cannot be resolved into individual events. The characteristics of these earthquakes are the
overall spectral power of this tremor correlates depletion of short-period energy, the elon-
well with the glacial flow velocities. Seismic gated source time functions, and their radia-
energy during the tremor is concentrated in the tion patterns, which can be explained by a
horizontal components. A source, radiating shear single surface force better than by seismic
energy from near the glacier bed, is consistent moments. Efficient detection and location
with this observation. The spectra reveal coherent methods were developed and force histories
gliding spectral lines, some of which are over- F(t) along with the centroid single force
tones of fundamental frequencies. Periodic (CSF) derived. The CSF is the product of
stick–slip failure of small asperities could be the the total moving mass M and the displacement
source mechanism of these seismic signals. D of the centroid. It represents a major con-
Winberry et al. (2013) argue that the highly straint on the whole source process.
repeatable WIS system represents an excellent Originally, the new-class global earthquakes
natural laboratory which could provide insight with sources on glaciers were interpreted as S
into friction and sliding complementary to labo- stick–slip flow behavior of large portions of out-
ratory or other field studies (e.g., Gomberg let glaciers in Greenland and Antarctica. How-
et al. 2011). ever, the postulated movements of the ice masses
could not to date be confirmed geodetically. An
alternative source mechanism which generates
sufficient CSF and provides the appropriate dura-
Summary tion of the source function is the capsizing
(or toppling) of large icebergs at the front of the
A new class of global earthquakes (Ms > 4.5) outlet glaciers during major calving events. The
was discovered and related to giant and cata- basic mechanics of this process are well under-
strophic landslides or phenomena occurring stood, and correlations with glacier flow, sea-
near the front of fast-moving outlet glaciers sonal variations of iceberg production, and
in Greenland and Antarctica. The unique glacier retreat support the explanation of the
Seismic Sources from Landslides and Glaciers, Fig. 34 Waveforms observed at stations 1–4 at WIS; (a) impulsive,
high frequency waveforms; (b) low frequency (Modified after Winberry et al. 2013)
majority of these earthquakes coming about deployed on or very near to the landslide mass
through this source process. are necessary to capture these signals. A rich
Related to landslide-generated global earth- variety of waveforms can be observed and clas-
quakes, additional information from satellite sified by attributes derived from the frequency
images or local reconnaissance helps to separate content, the duration, or the shape of the enve-
CSF into mass and total displacement. The incli- lope. Currently, no definite assignment of the
nation of the CSF vector corresponds to the different waveforms to landslide types and pro-
Fahrböschung; the duration of the landslide cesses can be established. Impulsive events have
force history constrains the sliding velocity. been observed on slopes with steep or medium
Application of a rotational slider block model inclinations (>20 ); tremor or harmonic LMEs
allows, e.g., for estimates of the reduction of the dominate the seismic activity of gently sloping
friction coefficient from static to dynamic flows. Most seismic activity on landslides is
conditions. rather shallow. However, the location of a few
Microearthquakes related to the processes in events at the basal sliding zone cannot be
landslide masses (LMEs) that evolve in a more excluded. Rockfall and LME frequency correlate
steady form or those in a post- or prefailure state frequently with precipitation and precede
have magnitudes of M < 0. Seismic networks geodetically observed acceleration. An increase
in seismic activity preceding acceleration with- quantification by seismic methods may support
out discernible external triggers has also been mitigation measures and warnings against subse-
observed. The general pattern of rockfall and quent hazards like damming of the valley (see
LME activity suggests that the DSGSD- and Fig. 6c) or increased density of icebergs. LMEs
flow-type landslides are near their yield stress have been proven to indicate a dangerous accel-
not only at the basal sliding zone but also in eration of a landslide, usually earlier than the
their entire moving mass. Further progress in geodetically observable displacement. GMEs
the understanding of LMEs would be supported precede glacier break-off and glacial lake out-
by denser seismic networks including borehole burst floods (GLOFs). The results from basic
seismometers as a standard. Long-term seismic research on LMEs and GMEs lead to a recom-
monitoring covering seasonal variations and epi- mendation of seismic monitoring as an essential
sodic high-velocity phases combined with com- component of an integrated early warning system
plementary monitoring (e.g., geodetic, for landslides and glacial hazards.
meteorological, hydrological, geomorphologic)
are preconditions for testing refined hypotheses
about LMEs and their relation to landslide pro- Cross-References
cesses and evolution.
Monitoring of glacier flow-related microearth- ▶ Earthquake Location
quakes (GMEs) by seismic networks deployed on ▶ Earthquake Mechanism Description and
glaciers and supplemented by geodetic and mete- Inversion
orological devices provides a deeper insight into ▶ Mechanisms of Earthquakes in Iceland
the stress distribution within the glacier ice, gla- ▶ Non-Double-Couple Earthquakes
cier hydraulics, and processes at the glacier bed ▶ Rockfall Seismicity Accompanying Dome-
or within a basal layer of soft sediments. The Building Eruptions
latter aspect can be exemplarily studied at the ▶ Volcanic Eruptions, Real-Time Forecasting of
Whillans Ice Stream (WIS) in Antarctica. The
ice stream exhibits a stick–slip behavior, con-
trolled by tides with a CSF corresponding to References
Ms = 5.6. The duration of this process is above
Alley RB, Blankenship DD, Bentley CR, Rooney ST
the seismically observable range, but GMEs gen- (1987) Till beneath Ice Stream B. Till deformation:
erated by this process yield unique information evidence and implications. J Geophys Res
about the spatial distribution and mechanical 92(B9):8921–8929
behavior of asperities or “sticky spots” in the Amitrano D, Grasso JR, Senfaute G (2005) Seismic pre-
cursory patterns before a cliff collapse and critical
basal layer of soft sediments. Research in point Phenomena. Geophys Res Lett 32, L08314.
this field does not only support the understanding doi:10.1029/2004GL022270 S
of glacier sliding but addresses generally the Bindschadler RA, King MA, Alley EB,
friction behavior of fault gouges, e.g., the basal Anandakrishnan S, Padman L (2003) Tidally con-
trolled stick–slip discharge of a West Antarctic ice
sliding zones of landslides or active tectonic stream. Science 301(5636):1087–1089. doi:10.1126/
faults. science.1087231
Finally, we touch on how the potential moni- Brantley SR, Myers B (2000) Mount St. Helens – from the
toring and analyzing of landslide- and glacier- 1980 eruption to 2000. U.S. Geological Survey Fact
Sheet, FS-036-00, 2 p. http://pubs.er.usgs.gov/publica-
generated earthquakes contribute to prediction tion/fs03600
and early warning of catastrophic events. Global Br€
uckl E, Parotidis M (2005) Prediction of slope instabil-
landslide or glacial earthquakes always represent ities due to deep-seated gravitational creep. Nat Haz-
the finite state of a development and may mes- ards Earth Syst Sci 5:155–172. SRef-ID: 1684-9981/
nhess/2005-5-155
sage a disaster. Prediction and early warning is Br€
uckl E, Brunner FK, Lang E, Mertl S, M€ uller M,
not provided by the detection of these earth- Stary U (2013) The Gradenbach observatory-monitoring
quakes; however, instantaneous location and deep-seated gravitational slope deformation by geodetic,
hydrological, and seismological methods. Landslides. Nettles M, Ekström G (2010) Glacial earthquakes in
doi:10.1007/s10346-013-0417-1 Greenland and Antarctica. Annu Rev Earth Planet
Burgess EW, Forster RR, Larsen CF, Braun M (2012) Sci 38:467–491. doi:10.1146/annurev-earth-040809-
Surge dynamics on Bering Glacier, Alaska, in 152414
2008–2011. Cryosphere 6:1251–1262. doi:10.5194/ Nettles M, Larsen TB, Elósegui P, Hamilton GS,
tc-6-1251-2012 Stearns LA, Ahlstrøm AP, Davis JL, Andersen ML,
Chen X, Shearer PM, Walter F, Fricker HA (2011) Seven- de Juan J, Khan SA, Stenseng L, Ekström G,
teen Antarctic seismic events detected by global sur- Forsberg R (2008) Step-wise changes in glacier
face waves and a possible link to calving events from flow speed coincide with calving and glacial
satellite images. J Geophys Res 116, B06311. earthquakes at Helheim Glacier, Greenland.
doi:10.1029/2011JB008262 Geophys Res Lett 35, L24503. doi:10.1029/
Cuffey KM, Paterson WSB (2010) The physics of gla- 2008GL036127
ciers, 4th Edition. Academic Press, ISBN Spillmann T, Maurer H, Green AG, Heincke B, Willenberg H,
978012333694614, p 704 Husen S (2007) Microseismic investigation of an unstable
Eberhardt E, Stead D, Coggan JS (2004) Numerical anal- mountain slope in the Swiss Alps. J Geophys Res
ysis of initiation and progressive failure in natural rock 112, B07301. doi:10.1029/2006JB004723
slopes – the 1991 Randa rockslide. Int J Rock Mech Tsai VC, Ekström G (2007) Analysis of glacial earth-
Min Sci 41:69–87 quakes. J Geophys Res 113:F03S22. doi:10.1029/
Ekström G, Stark CP (2013) Simple scaling of 2006JF000596
catastrophic landslide dynamics. Science Tsai VC, Rice JR, Fahnestock M (2008) Possible
339(6126):1416–1419. doi:10.1126/science.1232887 mechanisms for glacial earthquakes. J Geophys Res
Ekström G, Nettles M, Abers GA (2003) Glacial earthquakes. 113, F03014. doi:10.1029/2007JF000944
Science 302:22–624. doi:10.1126/science.1088057 Voight B (1981) Time scale for the first moments of
Faillettaz J, Funk M, Sornette D (2011) Icequakes coupled the May 18 eruption. In: Lipman P, Mullineaux
with surface displacements for predicting glacier D (eds) The 1980 eruptions of Mount St. Helens
break-off. J Glaciol 57(203):453–460 Washington. Geological Survey Professional Paper
Gischig V, Moore JR, Evans KF, Amann F, Loew S (2011) 1250, pp 68–86
Thermomechanical forcing of deep rock Walter F (2009) Seismic activity on Gornergletscher dur-
slope deformation: 2. The Randa rock slope ing Gornersee outburst floods. PhD thesis, DISS.
instability. J Geophys Res 116:F04011. doi:10.1029/ ETHNo 18184, ETH Z€ urich, p 135
2011JF002007 Walter F, Dreger DS, Clinton JF, Deichmann N,
Gomberg J, Schulz W, Bodin P, Kean J (2011) Seismic Funk M (2010) Evidence for near-horizontal tensile
and geodetic signatures of fault slip at the Slumgullion faulting at the base of Gornergletscher, a Swiss Alpine
Landslide Natural Laboratory. J Geophys Res Glacier. Bull Seismol Soc Am 100(2):458–472.
116, B09404. doi:10.1029/2011JB008304 doi:10.1785/0120090083
Goodman RE, Kieffer DS (2000) Behaviour of rock in Walter M, Arnhardt C, Joswig M (2012) Seismic moni-
slopes. J Geotech Geoenviron 128(8):675–684 toring of rockfalls, slide quakes, and fissure develop-
Helmstetter A, Garambois S (2010) Seismic monitoring of ment at the Super-Sauze mudslide, French Alps. Eng
Séchilienne rockslide (French Alps): analysis of seismic Geol 128:12–22
signals and their correlation with rainfalls. J Geophys Weginger S (2012) Detektion und Lokalisierung
Res 115, F03016. doi:10.1029/2009JF001532 seismischer Signale zur Überwachung der
Kanamori H, Given JW (1982) Analysis of long‐period Massenbewegung Steinlehnen. Master thesis, Vienna
seismic waves excited by the May 18, 1980, eruption University of Technology, p 96
of Mount St. Helens – a terrestrial monopole? West ME, Larsen C, Truffer M, O’Neel S, LeBlanc
J Geophys Res 87:5422–5432. doi:10.1029/ L (2010) Glacier microseismicity. Geology
JB087iB07p05422. ISSN: 0148–0227 38(4):319–322. doi:10.1130/G30606.1
Kawakatsu H (1989) Centroid single force inversion of Winberry JP, Anandakrishnan S, Wiens DA, Alley RB
seismic waves generated by landslides. J Geophys Res (2013) Nucleation and seismic Tremor associated
94:12,363–12,374. doi:10.1029/89JB01118. ISSN: with the glacial earthquakes of Whillans Ice Stream,
0148–0227 Antartica. Geophys Res Lett 40:312–315.
Mertl S (2015) Characterization of local seismic events on doi:10.1002/grl.50130
the deep-seated events gravitational slobe deformation Zangerl C, Eberhardt E, Schönlaub H, Anegg J (2007)
Gradenbach. PhD thesis, Vienna University of Tech- Deformation behavior of deep-seated rockslides in
nology, p 176 crystalline rock. In: Rock mechanics: proceedings of
Moore J, Albee W (1981) Topographical and structural the 1st Canada-US rock mechanics symposium,
changes, March-July 1980 – photogrammetric data. In: Vancouver, 27–31 May 2007. doi:10.1201/
Lipman P, Mullineaux D (eds) The 1980 eruptions of NOE0415444019-c112
Mount St. Helens Washington. Geological Survey Pro- Zischinsky U (1969) Über Sackungen. Rock Mech
fessional Paper 1250, pp 123–134 1:30–52
Seismic Strengthening Strategies for Existing (Code-Deficient) Ordinary Structures 3079
instead of a practically unfeasible (due to

Seismic Strengthening Strategies for cost or disturbance of use) high target, the
Existing (Code-Deficient) Ordinary adoption of a lower, yet feasible, target is
Structures more pragmatic. This option is typically not
available in the case of upgrading against per-
Kosmas-Athanasios Stylianidis manent loads.
Civil Engineering Department, Aristotle Intervention: The term structural intervention
University of Thessaloniki, Thessaloniki, Greece implies any operation that results in the
desired modification of existing mechanical
characteristics of a member or a structure and
Synonyms has as a consequence the modification of its
response.
Intervention procedures to upgrade existing ordi- Repair: The term implies the process of interven-
nary structures which are designed and executed tion to a structure damaged by any cause that
according to older codes reinstates the mechanical characteristics of its
structural members to their pre-damage level
and restores its original structural capacity.
Introduction Strengthening: The term implies the process of
intervention to a structure with or without
Scope of the Entry damage, which increases the capacity and/or
This entry presents basic principles, criteria, and ductility of a member or the entire structure to
procedures in regard to the conceptual a level higher than that prescribed in the orig-
design of seismic strengthening of existing ordi- inal design.
nary structures and its treatment in current codes. Strategy: It is a high-level plan to achieve certain
In this context, certain code provisions constitute objectives, usually under uncertain conditions.
a significant part of the present entry. Technique: It is a procedure to be implemented to
Although it can be read by nonexperts, this complete a task (see also Wikipedia 2014).
entry is mainly addressed to structural engineers Code Deficient: The term refers to buildings
involved in the design of structures and interven- designed according to older seismic codes or
tions in buildings. without any code. It is typically the case that
contemporary codes introduce significantly
Notes on Terminology Used higher demands than older codes and this is
The terms used in the entry are standard and the main reason that upgrading is needed.
familiar to engineers; nevertheless some clarifi- Ordinary Structures: This term includes mostly
cations are in order: above surface R/C building structures. Under- S
ground, special structures (bridges, chimneys,
Seismic: The entry is restricted to seismic etc.) and historical buildings fall beyond the
upgrading only. Certain aspects of the meth- scope of this entry. The same principles can be
odology may also apply to upgrading against applied to structures made of different struc-
other actions. However, it is pointed out that tural materials, such as steel, masonry, and
the main characteristic of a seismic strength- wood, after suitable adaptation.
ening strategy is the fact that an earthquake is
an accidental load. In this sense, it is possible Sources
to define target performance levels, dependent The entry is primarily based on two sources, the
on the level of the design earthquake, which in European Standard EN 1998–3 (2005) and the
general can be related to the accepted risk. Greek Code of Structural Interventions (2012),
This probability leads to a compromise which standardize the existing knowledge on
between “desirable” and “feasible,” i.e., upgrading of ordinary structures.
3080 Seismic Strengthening Strategies for Existing (Code-Deficient) Ordinary Structures
Assessment the selected redesign objective, the result

of the assessment might lead to a decision
General for intervention (repair and/or retrofit).
The assessment of an existing structure is not (ii) In case that intervention is required, the
strictly mandatory, but it is valuable, since it structure is assessed at its pristine condi-
reveals the weaknesses of the structural system tion, i.e., assuming that damage will be
and indicates the strengthening measures to be repaired. Depending on the selected
taken. In this sense, the assessment is integrated in redesign objective, the result of this
the design strategy and is strongly recommended. assessment will lead to the decision for
The assessment of existing structures follows repair only or for repair and retrofit.
the steps below:
Collection of Data
• Collection of data (investigation of structural The collection of the data required for the assess-
history) ment should be governed by the following
• Analysis (see Section Methods of Analysis for principles:
Assessment and Redesign of the present entry)
• Verification against limit states (see Section (a) The data required for the assessment of the
Compliance Criteria for Assessment and bearing capacity of existing structures should
Redesign of the present entry) be cross-verified and properly calibrated,
wherever possible.
Scope (b) It is recommended that the program of field
(a) The purpose of the assessment of an existing and laboratory investigations be designed,
structure is the evaluation of its available executed, and supervised by the designer of
bearing capacity and the verification that the the assessment, according to the specific
minimum mandatory requirements imposed design requirements.
by the existing codes are met. (c) The reliability of the data collected should be
(b) To estimate the available bearing capacity of properly taken into account in the assessment
the structure, the data from the structural of the existing structure and the development
history survey should be taken into account of the intervention strategy.
(see Greek Code of Structural Interventions
(2012), Chap. 3, and European Standard EN Data Reliability Level (DRL)
1998–3 (2005), Chap. 3).
(c) The designer should prescribe and supervise General
a series of investigating works in order to (a) The reliability level of data (DRL) related to
document and justify the assumptions on actions or resistances indicates the adequacy
which the assessment will be based. of the information regarding the existing
(d) The process of assessment differs depending building and is taken into account in the
on the existence or otherwise of damage in assessment and redesign. In existing struc-
the building assessed. tures, the numerical values of the data
(e) In case of no damage, the result of the assess- required for the assessment and redesign are
ment, depending on the selected redesign objec- usually characterized by a higher uncertainty
tive, will dictate the decision for potential retrofit. than in the case of new structures.
(f) In the presence of damage, due to any past (b) DRL is not necessarily the same for the entire
actions (prescribed or not by the standards building. Individual DRLs for the various
applicable up to then), the assessment process subcategories of information can be deter-
includes two phases: mined. For the selection of methods of anal-
(i) First, the structure is assessed as it is, tak- ysis, the most unfavorable among the
ing damage into account. Depending on individual DRLs are used.
(c) The concept of DRL is also applied in In certain cases of high uncertainty (and if
assessing the completeness of the survey of it is considered that the influence of the mag-
the structure and infill walls, especially in nitude of the corresponding action is signifi-
case of hidden elements. The effects of uncer- cant), consideration of two “reasonably
tainties can be taken into account in actions or extreme” representative values (Sk, min and
resistances depending on the case (e.g., Sk, max) is recommended.
uncertainty in the thickness of the flooring (b) The appropriate safety factors gm are selected
of the slab will be taken into account in according to the data for existing materials
actions; uncertainty in the thickness of the (see also Greek Code of Structural Interven-
slab itself will be considered mainly in the tions 2012, Chap. 4). TMaterial data are
resistances). the dimensions of members and strengths of
concrete and reinforcing steel, as well as
Description of DRLs the actual reinforcement detailing, anchor-
(a) According to the European Standard EN ing, starter bars etc. that determine the
1998–3 (2005), three knowledge levels of resistances.
data are adopted: (KL) 1–3 (limited, normal,
full). According to the Greek Code of Struc- Assessment Principles
tural Interventions (2012), Chap. 3, these
levels correspond to high, satisfactory, and General
tolerable data reliability level. Assessment of existing structures follows the
(b) Regarding the self-weight, the characteristic principles listed below:
value considered must be the most unfavor-
able value that is consistent with the geome- (a) When the existing structural system is
try of the structure and/or applies for such expected to contribute to the redesigned
structures. structural system solely by resisting vertical
(c) Regarding the resistances, their values can be loads, its assessment may be performed based
determined from the dimensions, reinforce- on simple, yet conservative, methods.
ment, and material characteristics that lead to In this case, the accuracy of the assessment
the justification of prior behavior of the struc- method used should be adjusted to the desired
ture. So, for example, a strength value that goal. For instance, an approximate, yet con-
corresponds to the ultimate resistance of servative, assessment method is sufficient to
a cross section under the existing acting demonstrate the adequacy of the existing
loads can be used. Similarly, dimensions of structural system against vertical loads.
inaccessible foundations can be estimated When the existing structural system is appar-
with the assumption that they correspond to ently inadequate and is expected to be fully S
the ultimate soil bearing capacity under act- demolished, its assessment is not necessary.
ing loads, etc. (b) However, when the existing structural system
is expected to contribute to the redesigned
Impact of DRL on the Assessment and Redesign structural system by resisting both vertical
Depending on the reliability of the data: and seismic loads, it should be assessed
based on the principles below. For the assess-
(a) The appropriate safety factors gf for certain ment of the structure against vertical loads, it
actions with uncertain values are selected is possible to use the methods prescribed by
(see Greek Code of Structural Interventions Eurocode 2 (EN 1992-1-1 2004), appropri-
2012, Chap. 4). Such may be the case for the ately adapted:
representative values of some indirect actions (i) The assessment is made using analyti-
(e.g., soil pressure) and the weight of inac- cal methods (see Section Methods of
cessible infill walls or coating/plastering. Analysis for Assessment and Redesign
of the present entry). Especially in the combinations of actions assumed to

case of structures without damage, for have been applied in the past, etc.
which the approved design study (vi) For analysis, limit state control, verifi-
(which has been applied) is available, cation of the adopted behavior factor,
the assessment could be based on it. control of the imposed displacements,
(ii) The numerical models to be used for and local ductility indices, Sections
the assessment may represent the entire Methods of Analysis for Assessment
structure or individual members. Dif- and Redesign and Compliance Criteria
ferent numerical models may be used, for Assessment and Redesign of the
depending on the type of the imposed present entry are applied. In the case
actions. In general, the types of numer- of masonry walls, paragraph Consider-
ical models should be selected consis- ation of Masonry Infill Walls is applied.
tently with the calculation methods to (vii) In many cases, a quick assessment of the
be applied. Advanced methods of anal- loss of bearing capacity of a damaged or
ysis should be combined with detailed deteriorated structure may be useful
numerical models. and/or necessary. This estimate can be
(iii) It is recommended that the accuracy of made based on the intensity and extent
the methods used be compatible with of damage, as derived according to well-
the accuracy of the data. High-quality established (refined or approximate)
numerical analysis is reliable only if it methods.
is based on equally reliable data.
(iv) The use of empirical–analytical or purely Consideration of Masonry Infill Walls
empirical methods is allowed only in (a) It is not permitted to consider masonry infill
cases covered by relevant special provi- walls as part of the system that carries
sions issued by the competent public non-seismic actions (such as gravity loads).
authority. Such specific provisions can To calculate the internal forces of the struc-
be issued, provided that they refer to ture due to non-seismic actions, numerical
a building stock with common, known, models are used that do not include masonry
features and that they are always pre- infill walls or do not impose stresses (induced
ceded by a proper investigation. by vertical loads) to them.
(v) In cases of structures that have already (b) It is recommended to consider masonry infill
suffered damage or deterioration, the walls as part of the seismic action resisting
applied assessment method must be system.
able to interpret, as a rough approxima- (c) It is mandatory to consider masonry infill
tion, both the type and the location of walls as part of the system resisting
this significant damage. In structures of seismic actions, when this decision has
high importance, where damage has an adverse effect on the results obtained
been identified, parametric analyses for the structural system at a global or local
may be required to achieve the inter- level.
pretation of damage based on their type The assessment of whether the influence
and location. of infill walls is favorable or unfavorable has
The possible interpretation of dam- to be made by the designer; however, the
age, in terms of type and location, con- difficulty of the assessment has to be noted,
stitutes an acceptance criterion for the particularly in the case that analysis data
analytical methods used. Possible param- and calculations are not available. As a result,
eters may include non-visible geometri- the above assessment will be on the safety
cal data, mechanical characteristics that side, if the masonry infills are introduced in
have not been investigated, random advance to the numerical analysis models.
Redesign (d) Preliminary estimate is made of the ductility

class that the structure will fall into after the
General intervention, or (in case of application of
The redesign of existing structures follows these inelastic static analysis) preliminary estimate
steps: is made of either the amplitude of the target
displacement or the tolerable rotation angle
• Conceptual and preliminary design of all structural members after intervention.
• Analysis (see Section Methods of Analysis for
Assessment and Redesign of the present entry)
• Verification against limit states (see Section Assessment and Redesign Objectives
Compliance Criteria for Assessment and
Redesign of the present entry) General
(a) In order to satisfy broader socioeconomic
Conceptual and Preliminary Design needs, various “performance levels” (target
(a) According to the criteria and the types of behaviors) are stipulated under relevant pre-
structural interventions, to be presented in scribed design earthquakes.
Section Selection Criteria and Types of (b) The objectives of the assessment or redesign
Structural Interventions, an intervention (Table 1; see Greek Code of Structural Inter-
strategy is devised and the type and extent ventions 2012, Chap. 2.2) are combinations
of interventions is decided. of both a performance level and a seismic
Decisions on the appropriate strategy and action, given an “acceptable probability of
the subsequent type of interventions for each exceedance within the technical life cycle of
case should be made on the basis of all infor- the building” (design earthquake). The Euro-
mation obtained during the assessment stage pean Standard EN 1998–3 (2005) specifies
of the existing structure. The perception of that these levels should be defined in the
the overall behavior of the building and the National Annex (for each country).
identification of its weaknesses, such as inad- (c) In the present entry, objectives that refer
equate strength or stiffness or ductility, the solely to the structural system are prescribed.
unfavorable structural system, individual In contrast, no objectives are set for the
weaknesses, etc., should be dominant in the nonstructural system.
decision-making process. The relevant fundamental requirements, set
Regardless of the analysis method that by Chap. 2.1 of European Standard EN 1998–3
will be eventually adopted for the redesigned (2005), are satisfied through Table 1. In the
structure, inelastic static analysis may pro- case of two reassessment objectives, the pos-
vide substantial assistance in identifying sible pairs are B1 and A2 or C1 and B2. S
these weaknesses. Furthermore, with the aid The term “structural system” is used here
of the above method, it is feasible to prelim- in the classical sense and corresponds to the
inarily select the characteristics of the types system bearing vertical loads. Accordingly,
of intervention that will be prioritized. the term “nonstructural system” corresponds
(b) In all cases this selection should be justified to the system that does not participate in bear-
(compared with other possible options), ing vertical loads. It is noted that the above
while the anticipated post-intervention conditions are not associated with the terms
behavior of the building should also be “primary” and “secondary” structural ele-
described qualitatively. ments that are used in subsequent paragraphs.
(c) A preliminary estimate of the dimensions and (d) The objectives of the assessment or redesign
strength of the materials used and the modi- are not necessarily identical. The objectives
fied stiffness of the structural elements under- of redesign may be higher than those of the
going intervention should be made. assessment.
Seismic Strengthening Strategies for Existing (Code- It is noted that according to Table 1,
Deficient) Ordinary Structures, Table 1 Assessment design objective B1 is the objective normally
or redesign objectives of the structure
set for new structures.
Probability of Performance level The adoption of an assessment or redesign
exceedance of
objective associated with a probability of
seismic action
within a exceedance of the seismic action of 50 %
conventional Damage Significant Near will generally lead to more frequent, more
life cycle of limitation damage collapse extensive, and more severe damage com-
50 years (%) (DL) (SD) (NC)
pared to a corresponding objective associated
10 A1 1 C1
with a probability of exceedance of seismic
50 A2 2 C2
action equal to 10 %.
The probability of exceedance of 50 %
(e) The minimum acceptable assessment or rede- (maximum tolerable) in 50 years corresponds
sign objectives for the structural system of to an average return period of about 70 years,
existing buildings are defined ad hoc by the while a probability of exceedance of 10 % in
public authority. In special cases, the public 50 years corresponds to an average return
authority may designate additional objectives period of approximately 475 years.
for assessment or redesign of the nonbearing
system as well. In this case, the same author- Structural Performance Levels
ity also defines the criteria for meeting the The performance levels of the structure are
respective objectives. defined as follows [see also Chap. 2.1 of Euro-
(f) In any case, the reassessment objective pean Standard EN 1998–3 (2005)]:
(assessment or redesign) is chosen by the
project owner, provided that it is equal to or (a) “Damage Limitation” (A) is a condition
higher than the above minimum acceptable wherein the structure is only lightly damaged,
objectives. In defining these objectives, the with reinforced concrete elements prevented
following criteria (among others) should be from significant yielding and retaining
taken into account: their strength and stiffness properties.
• Importance of the building (e.g., tempo- Nonstructural components, such as partitions
rary structure, ordinary residential house, and infills, may show distributed cracking, but
area of public gathering, area of crisis the damage could be economically repaired.
management, high-risk facility) Permanent drifts are negligible. The structure
• Available financial resources during the does not need any repair measures. No building
given period operation is interrupted during and after the
(g) The owner of the project or the public author- design earthquake, with the possible exception
ity has to define the time frame within which of minor importance functions. A few hairline
the relevant interventions will be conducted, cracks may occur in the structure.
wherever required. (b) “Significant Damage” (B) is a condition
(h) A nominal life cycle equal to the conven- wherein the structure is significantly dam-
tional lifetime of 50 years is generally aged, with some residual lateral strength and
accepted, regardless of the estimated “actual” stiffness, and vertical elements are capable of
remaining life of the building. An exception sustaining vertical loads. Nonstructural com-
to this rule is permitted only under special ponents are damaged, although partitions and
circumstances where the remaining lifetime infills have not failed out of plane. Moderate
is fully guaranteed, based on the judgment permanent drifts are present. The structure
and approval of the public authority; in such can sustain aftershocks of moderate intensity.
a case, the seismic actions are modified The damage to the structure, expected to
accordingly. occur during the design earthquake, is
repairable but possibly uneconomic, without to the importance and age of the building
causing loss or serious injury of people. under consideration.
(c) “Near Collapse” (C) is a condition wherein the • The available quality of the work (it is
structure is heavily damaged, with low resid- very important that intervention measures
ual lateral strength and stiffness, although ver- be compatible with available resources
tical elements are still capable of sustaining and available quality of work).
vertical loads. Most nonstructural components • The availability of an adequate quality
have collapsed. Large permanent drifts are control system.
present. The structure is near collapse and • The use of the building (possible conse-
would probably not survive another earth- quences of the intervention works to the
quake, even of moderate intensity. The dam- use of the building).
age to the structure is in general unrepairable. • The design, from an aesthetic point of
Injuries of certain individuals due to structural view (the intervention scheme may vary
damage or falling elements of the nonbearing between a fully invisible intervention and
structure or other objects are not excluded. a deliberately distinctive set of new or
The term “unrepairable damage” refers to added members).
serious or severe damage, for which strength- • The conservation of the architectural iden-
ening (and not just repair) or replacement or tity and integrity of historical buildings
substitution of the component or the entire and the consideration of the degree of
structure is required. reversibility of the interventions.
• The duration of works.
(c) The selection of the type, technique, scale,
General Principles for Pre- and Post- and timing of the intervention should be
earthquake Intervention Decisions based on technical criteria related to the
observed current state of the building, as
Selection Criteria and Types of Structural well as to a provision to maximize the ability
Interventions of the structure to absorb seismic energy
(a) Based on the conclusions drawn during the (ductility) after the intervention.
assessment of the structure, and the nature, Such technical criteria include the
extent, and intensity of the damage or deterio- following:
ration (if any), intervention-related decisions • All identified serious deficiencies must be
are made, aiming to (a) meet the basic require- remedied.
ments of the seismic code, (b) minimize the • All identified serious damage (and deteri-
cost, and (c) serve the needs of the society. oration) in primary structural members
(b) The selection of the type of structural inter- must be repaired properly. S
vention should be made primarily on the • In the case of highly irregular buildings
basis of general cost- and time-related (mainly in terms of distribution of their
criteria, the availability of the resources overstrength, both in plan and in height of
required, architectural or other needs, etc. In the building), structural regularity should be
this selection, the financial (or other) value of improved to the maximum possible extent.
the structure should also be taken into con- • All resistance requirements in critical
sideration, both prior and subsequent to the regions of primary structural members
intervention. (i.e., the required resistance and plastic
Such general criteria include the deformation capacity) must be satisfied
following: after the intervention (distinguishing
• The cost, both initial and long term (i.e., between primary and secondary members).
the cost of maintenance and possible • Where possible, the increase of local duc-
future damage or deterioration), compared tility in critical regions should be pursued.
Particular provision should be taken, to the Some types of interventions in struc-

largest extent possible, so that the local tural elements associated with specific
repair and strengthening do not diversely strengthening strategies of technical
affect the available ductility within the nature are listed below:
critical region. • The enhancement of strength and stiff-
• In special cases, the durability of both new ness is alternatively achieved by selective
and original structural members and the or large-scale strengthening of structural
potential acceleration of the deterioration members or by the addition of new ele-
should be taken into consideration. ments that can resist either partially or
fully the seismic actions (e.g., reinforced
Types of Intervention and Their concrete shear walls, steel trusses, infill
Consequences walls, etc.). In this case, particular atten-
(a) Based on the foregoing criteria and the results tion should be given to the design of
of the assessment of the structure, appropriate the foundation due to the increase of
types of intervention should be ad hoc both the structural mass and the seismic
selected for individual structural members loads.
or the entire building and the nonstructural • The enhancement of post-elastic deforma-
system (if required), always taking into tion capacity (ductility) is achieved by
account the side effect of the interventions improving the confinement of existing
on the foundations. This selection is part of members, e.g., with external connectors,
an intervention strategy, which aims to strips of steel or fiber-reinforced
improve the seismic behavior of the building polymers, etc.
by modifying or controlling the basic param- • The remedy of critical deficiencies refers to
eters that affect its seismic behavior. In order alleviating those features that lead to unfa-
to achieve a reduction of seismic risk, strate- vorable seismic behavior. Indicatively:
gies of technical or managerial nature, or – Modification of the structural system
combination of the two, can be adopted. (abolition of certain expansion joints,
A number of technical and managerial replacement of sensitive members,
strategies are indicatively given herein: alteration actions aiming at a more reg-
Technical strategies: ular and ductile system)
• Enhancement of the building strength – Addition of special links to connect the
• Enhancement of the building stiffness brittle masonry and surrounding mem-
• Enhancement of the deformation capacity ber, whenever this is permitted by the
of the structural members strength of masonry
• Reduction of seismic demand – Local or global modification of mem-
Managerial strategies: bers with or without damage
• Limitation or change of use of the – Full replacement of inadequate mem-
building. bers or members that have suffered
• Partial or global demolition (i.e., of extensive damage
a number of storeys). – Redistribution of demand (e.g., through
• “Rigid body” transfer of the entire struc- external prestressing)
ture to another location. • The reduction of seismic demand is
• Decision for “no intervention”. In such achieved by reducing the mass of the
a case, a reduction of the nominal life structure and modifying the structural
cycle of the structure can be accepted, system toward a favorable shift of
under the condition that upon expiry of the fundamental period of the structure
this period, demolition of the structure is [e.g., through seismic isolation systems
guaranteed. or absorption of seismic energy; see
Chap. 10 of European Standard EN deterioration, as well as on the available data

1998–1 (2004)]. regarding the cross sections and strength of
(b) In cases where, for the redesign objective set, its structural members.
the seismic behavior of nonbearing structural Whenever possible, it is recommended to
members might endanger the lives of the calibrate such methods through comparison
occupants (or third persons) or might have with the behavior of buildings that have
consequences to stored goods, measures already been studied with the particular
should be taken to repair or strengthen the methods.
particular members. (c) Where appropriate, additional partial factors
In such cases, local or global collapse gSd will be applied to account for the uncer-
should be prevented by: tainties related to the numerical analysis
• Appropriate links to the structural mem- models.
bers or by taking supportive measures to
prevent possible fallout of parts of those Consideration of Masonry Infill Walls
members Consideration of the masonry infill walls in the
• The improvement of the mechanical char- redesigned structure can be carried out, subject to
acteristics of nonstructural members the conditions of paragraph Consideration of
(c) The potential impact of repairs and strength- Masonry Infill Walls
ening of nonstructural members should be As part of the redesign process, it is desirable
taken into account. to make every effort to mitigate the potential
(d) The side effects of all structural interventions deficiencies imposed by the masonry infills.
on the local and global capacity of the build- Addition or upgrading of masonry infills can be
ing to absorb seismic energy should be taken used for the improvement and strengthening of
into account. existing buildings, subject to conditions of para-
The enhancement of strength usually graph Consideration of Masonry Infill Walls
leads to a reduction in ductility, unless spe-
cial measures are taken (e.g., in reinforced Methods of Analysis
concrete elements, the increase of the tensile For the assessment and redesign of a building,
reinforcement should be in principle accom- one of the following analysis methods may be
panied by a sufficient increase of the com- used. The field of application of each analysis
pression reinforcement and of confinement). method depends on satisfying a number of con-
ditions, primarily related to regularity (see Greek
Code of Structural Interventions 2012, Chap. 5):
Methods of Analysis for Assessment
and Redesign (a) Elastic (equivalent) static analysis with S
global (q) or local (m) behavior
General (or ductility) factors, regardless of the data
(a) The action effects and/or the required plastic reliability level.
rotations of all structural members of the (b) Elastic dynamic analysis with global (q) or
building under the design earthquake and local (m) behavior (or ductility) factors,
other combinations of actions are derived regardless of the data reliability level.
using appropriate analysis methods. (c) Inelastic static analysis. In this case, it is
To determine the internal forces and dis- recommended to ensure, as a minimum,
placements, it is permissible to ignore prox- a “satisfactory” data reliability level.
imity to other buildings. (d) Inelastic dynamic (response history) analy-
(b) The selection of the appropriate method of sis. In this case, it is again recommended to
analysis should be based on the importance ensure, as a minimum, a “satisfactory” data
of the building and its potential damage or reliability level.
(e) In special cases, solely for the assessment of loaded due to lateral displacements of the build-
existing buildings, it is permitted to estimate ing, can be distinguished during assessment or
the demand approximately, without detailed redesign into “principal” (or “primary”) and
analysis involving a finite element model of “secondary.”
the entire building. In general, the structural members or substruc-
(f) Apart from the above analytical methods, tures that contribute to the strength and stability
solely for the assessment of existing build- of a building under seismic loading will be char-
ings, in special cases and for specific objec- acterized as principal. The remaining structural
tives, it is possible to use empirical methods. elements or substructures will be characterized as
(g) It is permitted to apply the elastic methods secondary. See also related Chap. 4.2.2 of Euro-
provided that all the following conditions pean Standard EN 1998–1 (2004).
apply: The main consequence of classifying a struc-
(i) The failure index (l) of each primary tural member (or substructure) as a secondary is
member is in general lower than 2.5. that for these members, different performance
The adopted threshold value of the criteria apply, i.e., they are permitted to undergo
failure index (l) generally implies that larger displacements and suffer higher damage
the available strength of each primary compared to the primary elements.
structural member is at least 40 % of the In cases where damage limitation after the
demand resulting from an elastic seis- earthquake has been set as the assessment or
mic analysis without reducing the seis- redesign objective, the above distinction between
mic action, i.e., for q = 1. primary and secondary data is not permitted.
(ii) The average failure index (lk ) in each For the masonry infill walls, which do not
storey does not exceed 1.50 times the carry vertical loads (see paragraph Consideration
average failure index of the storey of Masonry Infill Walls), the distinction between
above or below. primary and secondary members does not apply.
It is deemed that the average failure Where those members are considered as part of
index (lk ) identifies the regularity in the the seismic action resisting system, they are
resistance (strength) along the building addressed and verified separately.
height, whereas its adopted threshold
value ensures that no weak, in flexure
and/or shear, intermediate storey exists. Compliance Criteria for Assessment
(iii) The failure index (l) of each primary and Redesign
structural member that is located on one
side of the building, for a given direc- General
tion of seismic action, does not exceed (a) Compliance with the selected performance
1.50 times the average failure index (l) level for assessment and redesign is achieved
of a primary member that is located on by adoption of the seismic action, method of
any other side of the same storey. analysis, verification, and detailing proce-
It is deemed that with this provision, dures, appropriate for the different structural
issues of torsionally sensitive storeys materials (concrete, steel, masonry) within
are addressed. the scope of the code used.
(b) Except when using the q-factor approach,
Principal (or Primary) and Secondary compliance is checked by making use of the
Structural Members seismic action as defined in Table 1.
The individual parts of the structure of a building (c) For the verification of structural elements,
and the individual structural elements (members) a distinction is made between “ductile” and
affecting the stiffness and demand distribution “brittle” ones. Except when using the q-factor
within the building, or the members that are approach, the former is verified by checking
that demands do not exceed the corresponding (b) The partial safety factors of the existing and
capacities in terms of deformations. The latter added materials should take into account the
is verified by checking that demands do not geometrical uncertainties, the dispersion of
exceed the corresponding capacities in terms material properties, the relevant information
of strengths (see Greek Code of Structural available on site, as well as any uncertainties
Interventions (2012), Chap. 9). due to the nature of works and the difficulties
(d) Alternatively, a q-factor approach may be of effective quality control (see Greek Code of
used, where use is made of a seismic action Structural Interventions 2012, Chaps. 4, 7–9).
(response spectrum) reduced by a q-factor. In (c) Where appropriate, additional factors gRd are
safety verifications all structural elements are applied to account for the uncertainties aris-
verified by checking that demands due to the ing from the numerical modeling of the resis-
reduced seismic action do not exceed the tance in critical (or noncritical) regions.
corresponding capacities in terms of strength. (d) In cases of structural interventions against
(e) For the estimation of the capacities of ductile seismic actions, the damage limitation verifi-
or brittle elements, where these will be com- cation is made in accordance with appropri-
pared with demands for safety verifications, ate provisions (see Greek Code of Structural
mean value properties of the existing mate- Interventions 2012, Chap. 9). The damage
rials are used as directly obtained from in situ limitation verification generally includes the
tests and from additional sources of informa- primary and secondary structural members,
tion. Nominal properties are used for new or infills, and appendages.
added materials.
(f) Some of the existing structural elements can Verification of the Adopted Behavior Factor
be designated as “secondary seismic” (see After the verifications above, it is required to
Chapter Principal (or Primary) and Second- approximately reevaluate the predefined behav-
ary Structural Members), in accordance with ior factor for the repaired–strengthened building,
the definitions in European Standard EN taking into account all the criteria favoring
1998–1 (2004), Chap. 4.2.2 (1)P, (2) and energy absorption, such as:
(3). “Secondary seismic” elements have to
be verified according to the same compliance (a) The sequence of failure of horizontal and
criteria as primary seismic ones, but using vertical structural members
less conservative estimates of their capacity (b) The type of failure in critical regions of each
than for elements considered as “primary structural member (i.e., the ratio of the ulti-
seismic.” mate shear force to the effective shear at the
time of flexural failure, as imposed by capac-
Safety Verifications ity design) S
(a) The available resistance in the critical regions (c) The local available ductility in critical
of all structural members (i.e., the resistance regions
quantities and/or the tolerable plastic rota- (d) The available secondary resistance mecha-
tions) should be calculated on the basis of nisms at large relative displacements
rational numerical models, which are widely (e) The potential consequences of the brittleness
accepted by the international scientific com- of a limited number of structural members on
munity, especially in terms of force transfer the ductility of the entire structure
between existing and added materials or
members (see Greek Code of Structural Inter- It is pointed out that in existing structures the
ventions (2012), Chap. 6 for the numerical requirements of capacity design, limitation of the
models, Chap. 7 for the determination of the axial force, local confinement, etc. have not been
behavior of structural members, and Chap. 8 in general met. The implication of this is the
for the design of the interventions). difficulty in assessing a global behavior factor.
3090 Seismic Strengthening Strategies for Heritage Structures
Summary ▶ Retrofitting and Strengthening of Structures:

Basic Principles of Structural Interventions
This entry deals with the seismic strengthening ▶ Seismic Analysis of Steel–Concrete
strategies for existing (code-deficient) ordinary Composite Buildings: Numerical Modeling
structures. Reference to two regulatory texts that ▶ Strengthening Techniques: Code-Deficient
are compatible with each other, namely, the R/C Buildings
European Standard EN 1998–3 (2005) and the
Greek Code of Structural Interventions (2012),
has been selected as a basis for the presentation of References
the issue. These texts condense existing knowl-
edge concerning the upgrading of ordinary struc- Greek Earthquake Planning and Protection
Organisation – EPPO (2012) Greek code of structural
tures and at the same time determine the
interventions – harmonized text http://ecpfe.oasp.gr/
procedure of the normative treatment of the issue. sites/default/files/files/%CE%9A%CE%91%CE%9D%
First, clarifications on basic terms are given. CE%95%CE%A0%CE%95_EN2013_FINAL.pdf
Then, the general principles that rule the two European Standard EN 1992-1-1 (2004) Design of con-
crete structures – general rules and rules for buildings
aspects on which the issue of upgrading is European Standard EN 1998–1 (2004) Design of struc-
based, i.e., assessment and redesign, are ana- tures for earthquake resistance – general rules, seismic
lyzed. Subsequently, the assessment and redesign actions and rules for buildings
objectives, which in general differ from those of European Standard EN 1998–3 (2005) Design of struc-
tures for earthquake resistance – assessment and
the design of new structures, are presented.
retrofitting of buildings
The possibility of a variety of redesign objec- Wikipedia, the free Encyclopedia (2014) Terms “strategy”
tives is introduced. High targets are proposed for and “technique”
a small, yet critical in terms of safety, group of
buildings. On the contrary, there is a possibility,
under certain conditions, to adopt lower targets
for common existing buildings, in comparison to
the targets for new buildings, with a view to Seismic Strengthening Strategies for
a realistic treatment of the need for upgrading. Heritage Structures
Furthermore, the general principles for pre-
and post-earthquake intervention decisions are Dina D’Ayala1 and Sara Paganoni2
1
set, the methods of analysis and the conditions Department of Civil, Environmental and
of their application are briefly presented, and, Geomatic Engineering, University College
finally, the compliance criteria for assessment London, London, UK
2
and redesign are described. Ziegert|Roswag|Seiler Architekten Ingenieure,
Berlin, Germany
Cross-References
Introduction
▶ Assessment of Existing Structures Using
Inelastic Static Analysis The global seismic behavior of historic masonry
▶ Behavior Factor and Ductility buildings is highly influenced by the integrity of
▶ Equivalent Static Analysis of Structures the connections among vertical and horizontal
Subjected to Seismic Actions structural elements, to ensure the so-called box
▶ Performance-Based Design Procedure for behavior. This, providing the transfer of inertial
Structures with Magneto-Rheological and dynamic actions from elements working in
Dampers flexure out-of-plane to elements working in
▶ Reinforced Concrete Structures in in-plane shear, leads to a global response best
Earthquake-Resistant Construction suited to the strength capacity of the constitutive
Seismic Strengthening Strategies for Heritage Structures 3091
Seismic Strengthening LEVEL OF KNOWLEDGE

Strategies for Heritage
Structures,
Fig. 1 Correlation SEISMIC INTENSITY MEASURE STRUCTURAL ANALYSIS
between knowledge level, Peak Ground Acceleration Simplified models/limit analysis
analytical tools, risk Linear spectra Elastic FE. models /static analysis
representation, and
interventions. credits: this Nonlinear spectra Non linear FE. models /pushover
publication D’Ayala Accelerogrammes Non linear FE. models/ time history
Paganoni 2014
RISK INTERVENTIONS
Ratio of demand/capacity Strength
Performance based criteria, drift Ductility
Life cycle analysis Energy dissipation
materials and hence enhanced performance and reinforcement techniques, with the addition of
lower damage level. Notwithstanding the impor- innovative materials and a deeper insight in the
tance of connections’ integrity, analytical checks laws governing the dynamics of structures.
of existing connecting elements, or design of new The choice of the most suitable strengthening
elements to strengthen existing connections, are system for a given building, however, is deter-
generally based on qualitative rules or simplified mined not only by the set of constraints of the
overall checks, rather than rigorous analytical specific project but, very importantly, by the
approach. The “Guidelines for Earthquake Resis- framework used for its assessment. The EC8
tant Non-Engineered Construction” were first part 3 or ASCE41-06, dealing with assessment,
published by the International Association for repair, and strengthening, introduce the concept
Earthquake Engineering (IAEE) in 1986 specifi- of knowledge level as the determining factor for
cally with the objective of improving the seismic the choice of alternative assessment procedures
safety of non-engineered housing constructions. involving diverse levels of resources. The dia-
A wide range of techniques and products for gram in Fig. 1 shows that there is a strict correla-
the seismic strengthening of heritage buildings tion not just between the representation of the
are reported in the scientific literature and used seismic input and the type of analysis that can
in current practice to ensure the enhancement of be carried out with it but also and most impor-
existing connections. Heritage buildings require tantly between the input and the way risk is quan-
far more attention, especially when dealing with tified and measured and between such measures S
issues such as the compatibility between the and the principle on which possible strengthening
chemical and mechanical properties of the interventions operate. Hence the level of knowl-
strengthening system and the parent material. edge and availability of data to carry out the
Many strengthening techniques, after an initial assessment becomes critical in determining and
success and a strong commercial promotion, fully designing the most appropriate strengthen-
underperformed and showed unexpected drawing solution.
backs when put to the test of real seismic loading Moreover, although European and national
outside the controlled conditions of the labora- codes (e.g., Italian Ministry of Cultural Heritage
tory environment. On the other hand, some and Activities 2006) suggest the use of various
existing strengthening systems can provide systems for the strengthening of connections, for
highly flexible applications and meet the example, ring beams, no detailed reference is
expected requirements in terms of performance; made to specific procedures for the dimensioning
some of these systems in fact draw on traditional and checks of such elements. The only indication
in this sense can be found in Section 6.1 Retrofit The last 15 years have seen a steep increase in
Design Procedure for existing building of the use of new technologies for strengthening
Eurocode 8 (EN 1998-3:2005), which states that heritage buildings. Such fast development, as
the design process should cover: well as the high level of expertise and financial
resources required for their application, often
1. Selection of techniques and/or materials, as results in lack of standardization. Innovative
well as of type and layout of intervention. technologies have not been extensively applied
2. Preliminary sizing of additional structural and validated in real-life situations yet, and the
parts. retrofit of a complex, precious building by means
3. Preliminary calculation of stiffness of of unconventional systems is a difficult task that
strengthened elements. goes beyond the standard conservation practice.
4. Analysis of strengthened structure by linear or In fact, looking at the current scientific literature,
nonlinear analysis. The typology of analysis is it is clear that many projects of restoration and
chosen depending on the level of knowledge upgrade of monumental buildings are carried out
regarding the geometry detailing and mate- by research organizations within the framework
rials of the structure. of specific projects or by large enterprises that
5. Safety verifications for existing, modified, and specialize in the production and design of
new structural elements carried out by strengthening devices.
checking that the demand at three different On the other hand, the lack of appropriate
limit states – damage limitation, significant standards and procedures can be blamed for the
damage, and near collapse – is lower than the incorrect application of some innovative
structural capacity. strengthening techniques, notwithstanding recog-
nized effectiveness and other benefits. One
In the safety verifications, mean values of exception to this situation is the case of fiber-
mechanical properties of existing materials reinforced polymers for which guidelines and
derived from in situ tests and other sources of standards have been produced for selected coun-
information, like available documentation or rel- tries (see, for instance, CNR-DT 200/04,
evant sources, shall be used, taking into account CNR-DT 200R/13).
the confidence factors (CFs) specified in 3.5 of In the following an overview of methods to
Eurocode 8 (EN 1998-3:2005). Conversely, for structurally strengthen the masonry elements of
new materials, nominal properties shall be used historic buildings is given, pointing out the
without modification by confidence factor. The advantages and pitfalls of each system and the
code also states that in case the structural system, fundamental physical parameters and design pro-
comprising both existing and new structural ele- cess required. Available design procedures are
ments, can be made to fulfill the requirements of reviewed and referenced, when available, to the
EN1998-1-2004, the verifications may be carried writers’ knowledge. The systems reviewed are
out in accordance with the provisions therein. applicable mainly to stone or brick masonry
This last sentence indicates that for systems structures with timber floors and roof or with
such as reinforced concrete (RC) ring beams or vaulted structures. The concepts underlying
corner confinement, reference can be made to the these methods are also usually applied to the
specifications for RC members in the relevant strengthening of earth structures, although some
sections of EC8 and other Eurocodes. However, of the details of the specific applications might
this leaves open the problem of quantifying the differ due to the specific characteristics of the
interaction between original and new structural parent material.
elements, and the assessment of the global seis- The effective strengthening of a heritage
mic performance of the strengthened structure building also relies on a correct redistribution of
will still be affected by a large number of the inertia forces among vertical elements. For
uncertainties. this to occur, the floor structures need to be acting
as a whole and ensure diaphragm action. While In the following we review:

there is a wealth of research and methods to
assess and strengthen historic and traditional • Stiffening of the wall system by corner and
floor structures to ensure they develop such T-junctions confinement of wall panels
action, these are not currently included in this achieved by inserting vertical concrete col-
chapter. umns or steel meshes, fiber-reinforced hori-
Base isolation is also increasingly becoming a zontal strips, and polypropylene meshes
proposed option for the retrofitting of historic • Stiffening of the wall and floor system and
buildings; however, actual implementations are connection to the horizontal structures by
still very rare and very expensive. This technique ring beams
is also not treated in this current edition. • Connection of the walls and floor system by
In this current edition, only methods suggested anchorage systems
in the Eurocodes or methods related to recent • Improvement of the performance of the wall
technological developments are included. Many system by including energy absorbers and dis-
traditional “vernacular” methods of strengthen- sipating devices to connect the walls
ing historic building exist in earthquake-prone
countries and some have been studied in detail. The design of effective connections between
These are not addressed here either. various structural elements of a masonry struc-
Strengthening systems are reviewed tures is a critical step for the achievement of a
depending on the type of enhancement achieved, good structural response in case of seismic
i.e., whether there is an increase in strength, a events. Current codes provide for carefully
control of deformation and displacement, or a designed and detailed connections; for instance,
dissipation of energy. the Eurocode states that: “Floor systems and
the roof should be provided with in-plane
stiffness and resistance and with effective
Connections Between Vertical and connection to the vertical structural systems
Horizontal Macroelements (EN 1998-1:2004).”
This rule is also applicable to interventions
The Eurocode 8 states that to improve connection on existing structures as the Eurocode indeed
between intersecting walls, use should be made states that: “The connection between the floors
of cross-bonded bricks or stones. The connection and walls shall be provided by steel ties or
can be made more effective in different ways (EN reinforced concrete ring beams (EN 1998-
1998-3:2005): 1:2004)” and “Specifically for masonry struc-
tures: non-ductile lintels should be replaced,
I. Through construction of a reinforced con- inadequate connections between floor and walls S
crete belt should be improved, out-of-plane horizontal
II. By addition of steel plates or meshes in the thrusts against walls should be eliminated (EN
bed joints 1998-3:2005).”
III. Through insertion of inclined steel bars in Moreover, in relation to the connections
holes drilled in the masonry and grouting between walls and floors, EN 1998-3:2005 states
thereafter that: if existing tie-beams . . . are damaged, they
IV. Through post-tensioning should be repaired or rebuilt. If there are no
tie-beams in the original building structure, such
The addition of steel ties, along or transversely beams should be added.
to the walls, external or within holes drilled in the Improvement of connections between vertical
walls, is an efficient means of connecting walls and horizontal structures can be achieved through
and improving the overall behavior of masonry a variety of methods, which are described in the
buildings (EN 1998-3:2005). following.
Seismic Strengthening
Structures,
Fig. 2 Example of column
tie (From http://www.
know2do.org/Retrofit/
BrickMasonry/
RSBrickMasonryP.htm)
Confinement at Walls Junctions: Strength For the construction of the confinement ele-
Enhancement ments, all the bricks in the intervention area are
removed; the concrete of the existing tie beams is
Column Ties also removed to allow for the connection between
By confining plain masonry walls with vertical existing and new reinforcement. Vertical rebars
elements placed at all corners and wall intersec- and stirrups are placed and concrete cast (Fig. 2).
tions, as well as along the vertical frame of Tie columns should be as thick as the wall
large openings, the seismic performance of a where they are located, although in many cases
masonry building is improved as a result of the of repair of existing buildings, they are of reduced
enhanced integrity of the structural system dimensions due to on-site constraints. Further-
(Tomaževič,1999). This effectively changes an more, they may sometimes be replaced by sets
unreinforced masonry structure in a confined of reinforcing bars placed in holes drilled in the
masonry structure. The disruption is significant masonry and connected to the surrounding sub-
and can affect very substantially the heritage stratum by stirrups.
value of the building. Moreover, such a technique The confinement prevents disintegration and
is suitable only when horizontal RC tie beams in improves ductility and energy dissipation of
the bearing walls at floor level (see this IS section URM buildings but has limited effect on the
“Confinement at Wall to Floor Junctions: Ring ultimate load resistance (Chuxian et al. 1997).
Beams”) and stiff, monolithic floor diaphragms However, the real confinement effect mainly
are in place; otherwise, the effect of the vertical depends on the relative stiffness between the
confinement is scarce, or null, as shown by anal- masonry wall and the surrounding resulting con-
ysis conducted by Karantoni and Fardis (1992). crete frame. Before cracking, the confinement
Creation of corner confinement by column ties is effect can in general be neglected (Chuxian
not recommended for stonework masonry et al. 1997; Karantoni and Fardis 1992). For
(Tomaževič 1999). very squat URM walls (geometrical aspect ratio
of 0.33 and double fixed boundary conditions), Confinement by Fiber-Reinforced Plastic Strips
the confinement increased the cracking load by a The application of fiber-reinforced polymer
factor of 1.27 and the ultimate lateral capacity by (FRP) material to the strengthening of masonry
a factor of 1.2 (Chuxian et al. 1997). For walls structures has to an extent mimicked application
with higher aspect ratio, the confinement to RC structures. As a result, there are numerous
increased the ultimate capacity by a factor of examples in the literature relating to the confine-
1.5. In addition, the confinement improved the ment of masonry columns (e.g., Di Ludovico
lateral deformations and energy dissipation by et al. 2010; Corradi et al. 2007). In the case of
more than 50 %. junctions between walls, horizontal strips of
The dimensioning procedure for column ties FRPs bonded at various levels along a masonry
can be derived from the prescriptions specific to panel and anchored to the side walls can be used
confined masonry structures. The process can be to restore corner connections, thus preventing
summarized as follows: overturning of façade walls. Optimal application
is achieved when the whole perimeter of the
• Calculation of the appropriate combination of structure is confined. Application of such tech-
static and seismic loads acting on the struc- niques entails the removal of existing plaster and
tural elements (EN 1991-1-1:2002 and EN other superficial finish as a strong bond of the
1998-1:2004) FRP strips to the masonry is necessary to ensure
• Dimensioning and verification of the column effectiveness of the system. The strips also need
ties for both static and seismic loading follow- to be protected from UV radiations to prevent
ing the prescriptions provided for confined material deterioration. Their implementation
masonry structures (EN 1996-1-1:2005 and might cause substantial loss of heritage value,
EN 1998-1:2004) and it is not recommended when valuable finishes
are in place.
Special care should be taken in realizing the For the confinement to be effective, strips
connection between the column ties and the hor- should be laid both horizontally and vertically.
izontal structures in order to ensure a monolithic Tests by Hamoush et al. (2001) on walls strength-
behavior of the resulting RC frame. ened with vertical and horizontal strips showed
A much cheaper solution to confining masonry the effectiveness of the FRP systems as out-of-
walls is achieved with Welded Wire Mesh plane flexural strengthening elements, while
(WWM). This consists in deploying a steel Tumialan et al. (2001) have investigated the
mesh on the two sides of a wall and around the potential of near surface-mounted GFRP rods
corners, connecting it with bolts and then con- embedded into epoxy-based paste in the bed
crete over, by either shotcreting or other forms of joints as shear enhancement. Cyclic testing of
plastering. The anchorage of the mesh to the walls strengthened with vertical and horizontal S
foundation and horizontal structures is funda- GFRP or CFRP strips was performed by Marcari
mental to achieve proper action transfer during et al. (2003) and by Krevaikas and Triantafillou
shaking. Although less destructive than the con- (2006). Those tests have highlighted a general
struction of column ties, the system seriously decrease in strength and ductility of the confined
affects the breathability of the walls and may members with the increase of the aspect ratio of
cause deterioration due to moisture entrapment. the confined cross section. This is due to a reduc-
The thickness of this reinforced plaster may vary tion in ultimate strain at failure when increasing
between 25 and 100 mm depending on the regu- the aspect ratio. Glass fibers are more effective
larity of the walls underneath. It cannot be than carbon fibers, owing to lower stiffness.
applied in the presence of historic plasters or However, its effectiveness in connections
other valuable finishes. Its effectiveness highly between orthogonal walls has not been tested to
depends on the relative stiffness between the the authors’ knowledge. Issues of debonding
original wall and the added layers. should be thoroughly investigated on a case by
Strategies for Heritage Nd
Structures, Fig. 3 Static
scheme for the calculation
of the tensile capacity of
FRP strengthening to Fd
prevent overturning of
front wall (CNR-DT 200/
04) Qd
h h*
Pd
case basis as failure mode and extent will be h*: distance between level where FRPs are
highly affected by the integrity and mechanical bonded and bottom hinge
characteristics of the parent material external Af: FRP area
strata in relation to the composite characteristics. 2. Cheking the rip-off of FRP stripes from side
FRP confinement is dimensioned by verifying: walls:

1. The tensile capacity of the composite material Fd < 2Fpd ¼ 2 Af f pd (3)
2. The capacity of the anchorage area
where
In detail this is done by (CNR-DT 200/04): Fd is calculated as above.
Fpd is maximum anchorage capacity of
1. Checking that: FRP to one of the two side walls.
fpd is design debonding strength of FRPs.
Fd 2FRd ¼ 2 Af f fd (1) Af is FRP area.
where The second check is generally more demand-

Fd is the design action applied to the FRP ing than the first, but it is not needed when full
stripes by the front wall undergoing confinement of the structure is achieved by wrap-
overturning caused by seismic equivalent ping FRP stripes around the whole perimeter of
action Qd; this is calculated by: the building and a suitable anchorage length is
ensured (CNR-DT 200/04). Anchorage length
1 should be at least 300 mm; otherwise, mechanical
Fd ¼ ð Q h N d t Pd t Þ (2)
2h d fixings can be used; however, the guideline does
not specify suitable types and the procedure for
Dimensions and weight involved in the sizing and positioning the fixings.
overturning mechanism are shown in Fig. 3: Additional calculations of the strengthened
FRd: design tensile capacity of the FRP structure are necessary to ensure that the
strengthening strengthened structure will be able to withstand
ffd: design ultimate strength of the FRP other actions, such as the increased in-plane
strengthening shear loading, now transferred from the walls
Seismic Strengthening Strategies for Heritage Structures, Fig. 4 Corner confinement of masonry walls by use of
SRP strips anchored with SRP connectors (Photo Paolo Casadei 2008, product FIDSTEEL)
Seismic Strengthening Strategies for Heritage connection, and (c) retrofitted wall before application of
Structures, Fig. 5 Procedure for installation of PP final layer of mortar (Mayorca and Meguro 2004)
meshing: (a) PP-band mesh, (b) detail of top/bottom
orthogonal to the walls parallel to the action. the retrofitting technique is simple and suitable
This should be done according to the specific for local builders. PP meshing was developed by
indications of codes dealing with masonry struc- Meguro Lab, Tokyo University, and has had
tures in general. application in Kashmir and Nepal.
Of course in order to prevent damage in the The retrofitting installation procedure is as
historic building, the capacity of the strengthen- follows (Mayorca and Meguro 2004):
ing system needs to be less than the shearing
capacity of any masonry course in the façade 1. The PP-bands are arranged in a mesh and
wall. Given the absence of ductility in the fiber connected at their crossing points (Fig. 5a).
composite, this system also presents the funda- 2. Two steel rods are placed at the edges of the
mental drawback of brittle failure, modest ductil- mesh; these bars are used to anchor the mesh at S
ity, and relying on the damaging process of the the foundation and at the top edge of the wall
masonry for energy dissipation. An improved (Fig. 5b).
version in this respect is provided by SRP strips, 3. The walls are cleaned and, if possible, the
strips reinforced with steel, which show a greater paint is removed. Any loose brick is removed
ductility (see Fig. 4). and replaced.
4. Six millimeter diameter holes are drilled
Confinement by Polypropylene (PP) Mesh through the wall at approximately
PP meshing uses common packaging straps 250–300 mm distance. The holes are cleaned
(PP-bands) to form a mesh, which is then used with water spray or air.
to encase masonry walls, preventing both col- 5. The meshes are installed on both sides of the
lapse and the fall of debris during earthquakes. wall and wrapped around the corners and wall
PP-bands are commonly used for packaging and edges. An overlapping length of approxi-
are therefore cheap and readily available while mately 300 mm is needed.
6. Wire is passed through the holes and used 1976 earthquake onwards. The construction of
to connect the meshes on both wall sides. RC ring beams involves:
In order to prevent the wires from cutting the
PP-band mesh, a plastic element is placed • Consolidating the top part of the existing
between the band and the wire. Connectors masonry by injections
are placed in proximity of the wall intersec- • Drilling vertical holes and inserting steel rods
tions and of the wall edges. for the vertical connection between the
7. The top and bottom edges of the mesh are masonry and new element
glued to the foundation and top of the wall • Building the formwork and placing the rein-
by epoxy resin. The epoxy is used to connect forcement of the beam
the bars and the wall and it is not directly • Casting the concrete
applied to the mesh. The bands, which are
rolled around the bars, transfer their load Such system improves the seismic capacity of
through friction. structures by:
8. The overlapping parts of mesh are glued
together so as to ensure the continuity of the • Distributing the vertical loads
strengthening (Fig. 5c). • Transferring the horizontal loads from the
9. A layer of mortar is laid on top of the mesh to floors to the bearing walls
protect it from UV radiation and rain and also • Connecting the bearing walls so as to create a
to provide further bond. boxlike behavior and prevent out-of-plane
failure
Shear tests on walls strengthened by PP mesh
(Mayorca and Meguro 2004) showed that However, during a seismic event, the high
although the strengthening does not increase the stiffness of RC beams may induce out-of-plane
peak strength, it contributes to improving the bending of the portions of wall between the
structural performance after crack occurrence. restrained floors (Borri et al. 2009) with conse-
The strengthened walls exhibit larger post-peak quent bulging of walls and activation of out-of-
strength, while the mesh helps to spread the diag- plane damage mechanisms. Furthermore, such
onal cracks over a wide region. The presence of technique is often time-consuming, not cost-
connectors and the mortar layer, ensuring bond to effective, and adds mass to the structure, which
the masonry, are critical to the performance of increases the earthquake-induced inertia forces
the wall. and consequently requires strengthening at the
basis of the walls.
Confinement at Wall-to-Floor Junctions: Ring All these collateral effects represent major
Beams drawbacks for heritage buildings due to the
A ring beam is a structural element built on top of large deformability and scarce cohesion of his-
the masonry structure to the purpose of: toric masonry as opposed to the high stiffness and
weight of reinforced concrete. Moreover, the
• Creating a continuous connection between the construction of a ring beam has a high impact
roof structure and walls to better distribute the from the aesthetic point of view and may require
vertical loads the removal of large portions of the original mate-
• Improving the connection between orthogonal rial of the walls and of part of the structure of
walls and the three-dimensional behavior of the roof.
structure Indeed, several surveys carried out in Euro-
pean historic centers after earthquakes (e.g.,
Reinforced concrete (RC) ring beams have Spence and D’Ayala 1999; D’Ayala and
been widely used for the retrofit of masonry Paganoni 2011) highlighted how RC beams with
structures from the aftermath of the Friuli, Italy poor connections to the underlying material
negatively affected the buildings rather than • Improve the connection between orthogonal
improving its structural response. Damage gen- walls so as to enhance the three-dimensional
erally consisted of a shear failure at the interface behavior of structure.
between roof ring beam and wall, with more or
less extensive damage to the masonry and even- The advantage of the technique in respect to
tual collapse of walls. RC ring beams is that stiffness and mass are
Nevertheless, the use of RC ring beam is closer to the original masonry. A RM ring beam
allowed for the connection between walls and has also good vertical deformability, and there-
roof structure (Italian Ministry of Cultural Heri- fore it spreads the vertical load to the masonry
tage and Activities 2006), providing that the beneath. Furthermore, it has a lower aesthetic
dimensions of the ring beam itself do not cause impact, as the additional element can match the
an excessive increase of mass. However, con- appearance of the original masonry.
versely to steel ring beam, the use of RC beams RM beams are realized by:
for the strengthening of the connection between
walls and intermediate floor structures is advised • Building two wallets of brickwork or stone-
against (Ministero deri Beni Architettonici e work on top of the existing wall, leaving a
Culturali, Italia, 2006), as it is far too invasive cavity between the two
and deeply affects the performance of the struc- • Placing reinforcement bars within the cavity
tural system. • Casting the concrete within the cavity, as to
The 2007 edition of the California Historical create a beam
Building Code (2007), instead, generally recom-
mends the use of tie beams but does not specify Steel plate and transversal bricks connect the
the use of reinforced concrete. outer wallets.
In the scientific literature, references to pro- In some cases, consolidation of the masonry
cedures for the sizing of reinforced concrete ring under the ring beam by injections is needed to
beams can be hardly found, and it is highly likely ensure a sufficient shear capacity and similar
that in common practice a large majority of addi- stiffness.
tional concrete elements are simply dimensioned Like in the case of RC ring beams, examples
by the rule of thumb, using standard geometry of the sizing of RM ring beams can be hardly
and amount of reinforcement of ring beams in RC found in the scientific literature.
frames. A systematic dimensioning procedure for RM
An improvement on RC ring beams are ring beams could be derived from the prescrip-
reinforced masonry ring beams, combining good tions specific to RM structures. The process can
quality brickwork (generally made of solid be detailed as follows:
bricks) to steel or composite reinforcement S
bonded together by a binder-like grout • Calculation of the appropriate combination of
(Ministero deri Beni Architettonici e Culturali, static and seismic loads acting on the struc-
Italia, 2007). tural elements (EN 1991-1-1:2002 and EN
In the case of steel reinforcement, the brick- 1998-1:2004).
work of the ring beam is laid leaving an internal • Dimensioning and verification of the beam for
cavity where the reinforcement bars and stirrups both static and seismic loading following the
are positioned before casting the mortar-crete. prescriptions provided for reinforced masonry
The creation of a reinforced masonry ring structures (EN 1996-1-1:2005 and EN 1998-
beam aims to: 1:2004). This is done because a RM ring beam
is in fact a bearing element and should therefore
• Provide a continuous connection between the be calculated for vertical loading as beams are.
roof structure and walls to better distribute the Furthermore, the RM ring beams should be
vertical loads. dimensioned for tensile load as the confining
Seismic Strengthening Strategies for Heritage Struc-

tures, Fig. 6 Example of steel cord used in SRG (Borri
et al. 2007)
tures, Fig. 7 Assembly of LATLAM ring beam (Borri
action of the beam consists indeed in resisting et al. 2007)
the thrust generated by walls that separate
under horizontal actions; this is done by ignor- similarly to glue-laminated timber beams. The
ing the cross-sectional area of concrete and assembly acts as a single structural element and
masonry and sizing the steel reinforcement so combines the compressive strength of masonry
that it can bear the whole in-plane action. units with the good tensile properties of composite
• Dimensioning and checks of metallic connec- materials, avoiding the problem of the increase of
tors for both static and seismic loading mass typical of RC ring beams. A simplified model
(EN 1993-1:2005). for the evaluation of the ultimate strength of a
• Dimensioning and checks of connection of the LATLAM section under bending stresses is avail-
metallic fasteners to the concrete element of able in Borri et al. (2007). The collapse is assumed
the beam for both static and seismic loading to occur by crushing of masonry.
(EN 1992-1-1:2004 – Section 8 Detailing of
Reinforcement; DD CEN/TS 1992-4-1:2009). Strengthening by Way of Ties: Displacement
• Dimensioning and checks of connection of the Control
metallic fasteners to the masonry of walls for The insertion of ties to connect walls to walls,
both static and seismic loading. walls to floor, or walls to vaults is a traditional
• Dimensioning and checks of connection of the system, still in large use nowadays.
metallic fasteners fixed to the timber elements Post earthquake observation, experimental
of the roof for both static and seismic loading evidence and computational results, all show
(EN 1995-1-1:2004 – Section 8 Connections that crossties installed at the intersection of per-
with Metal Fasteners). pendicular sets of walls are able to prevent the
overturning of whole façades without interfering
Borri et al. (2009) propose a system combining with the original structural layout (Tomaževič
FRPs and steel-reinforced grout (SRG), the latter 1999) (D’Ayala and Yeomans 2004; D’Ayala
made of high strength steel wires forming cords and Paganoni 2011).
(Fig. 6) embedded in a cementitious grout. The For the ties to be effective, proper anchorage
ring-beam system, called LATLAM, is obtained within or against the masonry is necessary. This
by overlapping several layers of bricks and lami- might be achieved, either by using end plates or
nates embedded within a polymeric matrix or a by using grouted ends. The tie is usually a passive
cementitious grout (Fig. 7). By superposition of element which becomes active when cracks open
blocks and composite sheets, it is possible to between orthogonal walls or timber beams tend to
achieve a structural element of any size and length, slide off their seat.
A Consolidamento del muro in prossimità del capochiave

S
B
S/2
8308
400
50
24
40
C C
400 8 50
SEZIONE B-B
B
400
50
24
15
20 40
50
S
S/2
104010
20 100 C
5 A SEZIONE A-A SEZIONE C-C
25 cm 0 50
0 10
Seismic Strengthening Strategies for Heritage Structures, Fig. 8 Examples of metallic ties with end plates
(Giuffrè 1993)
The dimensioning procedure of metallic ties might be important to proceed to a local consol-
connecting timber joist to walls can be derived idation of the masonry by use of mortar grout
from the prescriptions specific to steel structures. before implementing the tie. The major pitfall
The process can be detailed as follows: of crossties is the possibility of pullout damage
at the head of the anchorage due to the different
(a) Calculation of the appropriate combination of deformabilities of anchor and masonry. The use
static and seismic loads acting on the struc- of high-ductility systems recommended by codes
tural elements (EN 1991-1-1:2002 and EN would overcome this issue, thus achieving the
1998-1:2004). objective of protecting culturally valuable fin-
(b) Dimensioning and checks of the ties follow- ishes and preserving life and safety.
ing the prescriptions provided for steel struc-
tures (EN 1993-1-1:2005). Ties with End Plates S
(c) Dimensioning and checks of connection The dimensioning process of metallic ties with
between metallic elements and timber ele- end plates (Fig. 8) consists in:
ments of the horizontal structure (EN 1995-
1:2004). • Sizing the cross section of the metallic rod to
(d) Dimensioning and checks of connection of resist the axial load deriving from the pulling
the metallic elements to masonry. Depending action of the portion of wall constrained by the
on the type of connection – by binder or by tie in case of out-of-plane action. The mini-
mechanical locking – checks should be mum diameter of the rod is calculated by
performed in a different way as discussed in (Tomaževič 1999):
the next two subsections.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
H u, seg 4 1
In the case of ties connecting masonry to Dmin ¼ (4)
n p fy
masonry point c is not relevant. However, it
where
Hu,seg: ultimate seismic resistance of a crit-
ical part of a building,
n: number steel ties,
fy: yield stress of steel.
The horizontal load that activates the
overturning mechanism in a portion of wall,
Hu,seg, needs to be equilibrated by the action of
crossties and can be calculated by limit analy-
sis. The position of hinges and the collapse
load factor, for the portion of wall involved
in the out-of-plane mechanism, are calculated
so as to satisfy rotational equilibrium and the
distribution of stress assumed in the masonry
section at collapse as shown in Fig. 9. In
deciding the static scheme for the calculation
of the mechanism, the type of connections
should be considered as they influence the
constraints of the ideal beam that represents
the wall: for instance, a wall with no positive
connections to the floor structures can be
modeled as a cantilever, while the positive
effect of well-connected horizontal structures
should be accounted for by using a simply
supported beam scheme. tures, Fig. 9 Static scheme of multistory wall panel for
• Sizing the end plate to prevent punching fail- determination of hinge position and collapse load multi-
ure. This is done by verifying that the tensile plier (D’Ayala, Speranza, 2003)
strength of the parent material is sufficient
to withstand the 45 component of the axial
load acting on the tie. The rupture surface to expansion anchors, which are designed to intro-
consider for the calculation of the acting stress duce a radial pressure between the anchor and the
is a conical/pyramid surface with sides hole. Finally bond systems rely on the use of a
inclined at 45 in respect to the plain of the binding agent, which is either injected or released
wall (Fig. 10). through a capsule system (Fig. 12).
A fairly common commercial product consists
Ties Without End Plates of metallic profiles shaped as a coil; such profiles
Anchors without end plates (Fig. 11) consist of a can be dry screwed in the masonry, thus mainly
metallic profile that is embedded in the masonry relying on a mechanical and frictional mechanism,
and can transmit loads to the substratum by or injected with resins/grout, in which case they
mechanical locking, friction, bond, or a combi- work through the bond established between
nation of these three (Eligehausen et al. 2006), masonry and binder. Another popular system,
rather than use of an element such as a plate, a increasingly used in conservation, is formed by
key, or a peg. Mechanical locking can be steel sections provided with a fabric sleeve; pro-
obtained, for instance, by undercut, namely, files are installed in holes and then grout/resin is
shaping the end part of the hole and introducing injected in the sleeve, so that the sleeve molds
an anchor with an end shape so as to result larger itself to the spaces and voids in the wall and creates
than the rest of the shaft. Friction systems, a system combining mechanical locking and bond.
instead, consist of torsion or displacement- A further advantage of this latter system is the
a b
S S
45
45
°
°
S
a
45
d d 2S+d
°
1 a
S
S.
tn
dq
(2
)
S
S.
S.
m
m
dq
dq
(2
(2
)
)
S S S
Seismic Strengthening Strategies for Heritage the end plate in case of: (a) rectangular key or (b) circular
Structures, Fig. 10 Rupture surface for the calculation end plate (http://postterremoto.altervista.org/Dati/i/i.
of tensile strength acting on parent material and sizing of html, last accessed 7th March 2011)
Seismic Strengthening Strategies for Heritage Structures, Fig. 11 Grouted anchor injection scheme (Gigla,
Wenzel 2000)
Seismic Strengthening Strategies for Heritage Structures, Fig. 12 Post-installed anchors: (a) systems for the
transmission of load to substratum. (b) Typologies of binding materials for bond anchors (Eligehausen et al. 2006)
control over the diffusion of the binder within the material, dry/wet diamond rotary drilling
parent material, while the grout can expand to the rather than percussive drilling is
cavity around the ties, particularly useful for rub- recommended.
ble double leaf masonry. • Removing all cores from the bore hole and
A variety of materials, ranging from mineral checking the depth. Removing dust and
binders to polymers, can be used as binder. How- debris.
ever, materials based on polymers, such as epoxy • Placing anchors and injecting the binder.
or polyester resins, which are often used for con-
crete structures, should undergo careful assess- The anchors utilized are usually stainless steel,
ment before use in historical masonry, due to threaded rods, or special prestressing bars. The
possible issues regarding the mechanical and diameter of the hole drilled to accommodate the
physical compatibility. Much more common for anchor is a function of the strength differential
the repair of historical buildings are mineral between parent material and grout and this and
binder systems based on cement or hydraulic the steel itself. The bond of the anchor inside the
lime with the addition of admixtures and fillers masonry unit depends on the type and properties
or aggregate. To inject bore holes, usually pure of the grouting material and on the type of
water/binder systems are used with typical w/b masonry block units.
values of 0.8–1.0. However, the w/b ratio has to The failure mechanisms of injection anchors
be adjusted according to the volume to be injected in masonry are discussed in Gigla (2004, 2010),
and to the moisture content of the substrate. based on an extensive test program of over
Post-installed anchors are a very common 500 pullout tests. The studies consider five differ-
method for the strengthening of historic masonry ent failure modes including failure mechanisms
since they allow for a wide range of applications, related to the (a) poor strength of the injected
from restoring the through-thickness cohesion of grout, (b) exceeded tensile strength of surround-
multilayered masonry to the increase of tensile ing parent material, (c) bond failure between the
and shear capacity of panels; furthermore, they outer surface of the grout and the parent material,
can be used at the joints between structural ele- (d) a combination of b and c, and (e) failure of the
ments either in the form of short connectors or as steel or grout in tension (under-designed).
longitudinal elements. These latter have the same Recommendations for the design of injection
function as metallic crossties, but, rather than anchors are given in Gigla (2004). The design
relying on an end plate for the connection to the value for the bond strength is computed as
wall, pullout loads are transmitted by a shear follows:
mechanism – for friction and bond anchors – or
through mechanical locking within the masonry !
FJ f 2G, c
itself. X A, d ¼ þ X B, W (5)
gm 500
The diameter of the anchors is chosen as a
function of the expected pullout force applied,
the tensile strength of the masonry, and the XA,d: Design value of bond strength, as a function
bond strength of the grout to masonry interface. of the compressive strength of grout and indepen-
Bonded anchors can be passive or active, i.e., dent of bond length.
prestressed or posttensioned. Passive anchors Required minimum bond length:
are used for up to 4 m in length; prestressed Lb = 150 mm inside monolithic stone,
anchors can be used up to 35 m in length (Gigla Lb = 190 mm in bed or head joints of brick, and
and Wenzel 2000). Lb = 430 mm in bed and head joints of
The installation of anchors is carried out by: blockwork.
fG,c: Compressive strength of grout. Minimum
• Drilling holes in the parent material; due to the value: fG,c = 16.6 N/mm2. Maximum value
weakness and preciousness of the parent covered: fG,c = 38.7 N/mm2. A minimum
bending tensile strength of fG,ct = fG,c/8 = hS: minimal distance to edge of surrounding
2.0 N/mm2 is required. masonry
fG,ct: Flexural strength of grout obtained from Lb: bond length
standard. tan(j): tangent of angle of friction for force
FJ: Reduction factor for bond in bed or head transmission between anchor’s grout and bore-
joints, FJ = 0.5. hole, about 50 in water-absorptive material and
XB,W: Term indicating the increase of bond about 60 in non-water-absorptive material
strength inside water-absorptive stone material. gM: partial factor of safety for stone tensile
XB,W = 0.15 N/mm2. strength; recommendation, 1,5
gM: Partial factor for property, recommended: While these values are appropriate for an axial
gM = 1.35 pullout test, they do not apply to failures of
The term XB,W considers the water absorption of anchors loaded in shear or bending as a result of
the masonry substrate. Pullout studies revealed relative movements between the structural ele-
that materials with higher water absorption capac- ments that they reconnect. Furthermore, as the
ity yielded higher pullout strength than materials family of dowel anchors also includes short fix-
with low or no water absorption capacity. ings such as pins and nail-like fixings, failures
Furthermore, the anchor capacity can be connected to loads transmitted from other ele-
designed by first computing Ra,d which refers to ments and in proximity to edges or openings
the bond capacity between grout and rod. The should also be analyzed, these being highly rele-
equation considers the ratio of bed and head vant to various typologies of strengthening, such
joints of the borehole surface across bond length as confinement or end plates, where the use of
and limits RA,d to bond inside full stone sections fixings is required but rarely regulated.
(AB/AG,d = 1 completely in stone; AB/AG,d < 1 in Whereas the dimensioning procedure of
joint): bonded anchors in concrete has been extensively
studied and commented (Eligehausen et al. 2006)
AB and European guidelines do exist (EOTA TR029
RA, d ¼ XA, d AA, d (6)
AG, d 2010; EOTA TR045 2013; DD CEN/TS 1992-4-
1:2009), anchors in masonry lack specifically
with RA,d, design capacity of the injected anchor; dedicated codes or recommendations. Modes of
AB, portion of the cylindrical surface of injected failure and hence the procedure for dimensioning
mortar grout made of stones or units; AG,d, total could be partly derived from DD CEN/TS 1992-
cylindrical surface of injected mortar grout; and 4-1:2009. Possible modes of failure due to axial
AA,d, interface of tensile element and injected load may be bond between metal and binder or
mortar plug (cylindrical surface of steel bar), between binder and parent material; cone pullout
calculated with nominal bar diameter and bond of parent material, due either to the failure of S
length. The design strength of the anchor shall be bond in the mortar joints or to tensile failure of
smaller than the tensile capacity of the surround- masonry units; and failure of steel, although this
ing masonry so as to cause failure by yielding of is highly unlikely unless the metallic section is
the metallic element and prevent brittle failure of severely under-designed. Possible modes of fail-
the masonry. This threshold can be computed as ure due to loading transversal to the anchor axis
may be crushing failure of either the binder or the
1, 9 f B, t Lb p dB ðhS 2 dB 2 parent material, edge failure of masonry when the
F (7)
gM tan ðjÞ ðdB 2 þ hS 2 load is applied in proximity of openings and
corners, pryout failure of parent material, or fail-
where ure of steel for combined bending and shear
F: anchor tensile force action. Each of these failures is considered by
fB,t: tensile strength of surrounding stone the guidelines and the elements of the anchorage
dB: borehole diameter assembly dimensioned accordingly.
Other factors considered in the guidelines are The selection of the type, technique, extent and
the performance of anchors when undergoing urgency of the intervention shall be based on the
structural information collected during the assess-
fatigue loading and the dimensioning criteria for ment of the building. The following aspects should
seismic loading. For anchors embedded in con- be taken into account:
crete, ductile failure is achieved by controlling [..]
the dimensioning of steel so as to avoid the fragile d) Increase in the local ductility supply should
be pursued where required;
failure of the substratum, and this concept should e) The increase in strength after the intervention
be extended to masonry structures. However, pre- should not reduce the available global ductility.
scriptions in EOTA TR029, EOTA TR045, and
DD CEN/TS 1992-4-1:2009 rely on a series of In the field of cultural heritage, such tech-
parameters, such as bond strength or minimum niques must aim to enhance the seismic perfor-
distance to edge, that should be either provided mance of structures as well as limit the aesthetic
by the producer of the anchors or derived exper- impact on the building and protect elements that,
imentally. In case of anchors for masonry, due to despite not being relevant to the structural behav-
the lack of standardization as well as the variabil- ior, have cultural and historic value.
ity of parent materials and masonry fabrics, not Energy control systems can either provide the
every producer is able to provide all the required structure with additional dissipation capacity
information and extensive tests covering the full and/or reduce the amount of ground input energy
range of parameters needed are missing from the transferred to the structure. However, base isola-
scientific literature. tion systems require heavy interventions on the
bearing structure (base cut, new foundation struc-
Energy Dissipation Systems ture, etc.) and are therefore rarely suitable for the
Systems like crossties have been and are still retrofit of heritage buildings.
commonly applied in rehabilitation practice Energy dissipation devices can instead be
throughout Europe for providing connection at placed in a number of key locations within the
the joints of perpendicular sets of masonry walls structure where relative displacements between
where out-of-plane damage is most likely to members are expected to occur; thus, from
occur. Nonetheless pullout damage at the head the point of view of intrusiveness, they do not
of the anchorage and increased in-plane diagonal differ from many other standard techniques.
cracking may affect valuable finishes and pre- A suitable position for the installation of dissi-
cious frescoes. pative devices is at the wall-to-floor interface of
The current codes encourage the use of these masonry buildings, where relative motion can
conventional stiffness-based systems (EN 1998 occur without impairing the global structural
Eurocode 8; Italian Ministry of Cultural Heritage integrity.
and Activities 2006) because innovative tech-
niques drawing on performance-based principles Energy Absorbers
(Priestley 2000), despite their effectiveness in Benedetti (2004) developed a series of energy-
new structures, rarely meet the requirements of absorbing devices for existing masonry buildings
reversibility and low impact required for historic drawing on experimental results that had showed
structures. that the more energy is absorbed through damage
As widely experienced in many applications by the noncritical elements of the structure, the
carried out in recent years, an effective alterna- later global failure occurs. The main require-
tive to conventional strengthening techniques is ments set for the devices were:
the use of passive energy control and dissipation
techniques. • Activation of the device for very low level of
Energy dissipation systems indeed comply damage and cracking
with the concept expressed in EC 8 (EN 1998- • Sensitivity to both in-plane and out-of-plane
3:2005) that: movements
Seismic Strengthening Strategies for Heritage Structures, Fig. 13 RAG energy absorber: (a) prototype
(Benedetti 2004); (b) setup in series with metallic crosstie installed on a masonry specimen (Benedetti 2004)
• Limitation of forces transmitted to the parent Shock Transmission Units

material at the fixings of the dissipative STUs, or viscous dampers, consist of a hollow
devices, so as to avoid localized damage and cylinder filled with fluid, this typically being sil-
detachment of the devices icone based. As the damper piston rod and piston
head receive an impact, the fluid is forced to flow
The RAG energy absorber (Fig. 13a) consists through orifices either around or through the pis-
of four arms hinged at their end so as to form a ton head. The resulting differential in pressure
square element. Hinges are made of lead cylin- across the piston head can produce very large
ders press-fitted into the arms. When two (or four) forces that resist the relative motion of the
of the corners of the device, which are connected damper, while the input energy is dissipated in
to the masonry wall, displace relatively to each form of heat due to friction between the piston
other as a consequence of damage in the parent head and fluid particles flowing at high velocities.
material, the hinges deform plastically in torsion Conversely, since this type of devices is velocity
(in-plane action of the device) or bending (out-of- dependent, slow movements such as thermal
plane action of the device), thanks to the low yield expansions are allowed.
strength of lead. Conversely, the arms are made of The viscous devices are characterized by a
a stiffer material so as to remain in the elastic nominal strength: for high load rates and for S
range. Besides a full characterization of the input forces below their capacity, they show
device, RAG energy absorbers were also tested high stiffness; above it they feature a perfect
inserted in full-scale masonry specimens on shak- plastic behavior, with dissipation of hysteretic
ing table in an out-of-plane configuration energy in case of cyclic loads. The main draw-
(Fig. 13b), where the devices were used in series back of this type of devices is the risk of leakage
with traditional metallic crossties to connect of the fluid, which involves the need for regular
two parallel walls. The holes that can be observed inspections and further costs involved with the
in the arms of the RAG device aim at reducing maintenance of the dissipative system. Mandara
the weight of the element and avoiding a delay and Mazzolani (2001) investigated two different
in the activation of the yielding mechanism. methods for the dimensioning of viscous devices.
A pretensioning element was positioned in series In the Plastic Threshold Approach (PTA),
with the anchor to further ensure the lack of any devices are conceived and sized to limit the mag-
initial deformation in the crosstie. nitude of force transmitted across connected
members to a maximum value, determined These characteristics make SMAs particularly

according to the design resistance of structural suitable for use in seismic dampers, especially
elements involved. Beyond this threshold, the considering that some alloys, such as
hysteretic energy dissipation takes place, while nitinol, have very good corrosion-resistance
below the threshold the behavior of structural characteristics.
members is virtually rigid, which ensures the Shape Memory Alloy Devices (SMADs),
maximum degree of redundancy to the structure which consist of metallic ties and groups of
under serviceability load conditions. According wires of SMAs, were developed within the frame-
to the Optimal Viscous Approach (OVA), the work of the ISTECH project (Indirli et al. 2001)
interaction between connected members is con- and patented by Fip Industriale, Padova, Italy.
trolled by the viscous properties of the devices, Their application has featured in high-profile pro-
which are dimensioned so as to minimize the jects such as the repair and strengthening of the
magnitude of the force acting on them indepen- Upper Basilica of St. Francis in Assisi, San
dently from its value. Contrary to PTA, the con- Feliciano cathedral in Foligno, and San Serafino
nection between elements is never fully rigid, so Church in Montegranaro, Italy. In the Upper
that energy can be dissipated under moderate Basilica of St. Francis, 47 SMADs were used to
intensity earthquakes too. connect the roof to the tympanum of the transept;
the SMADs were connected on one end to a
Shape Memory Alloys threaded bar attached to an anchorage plate in
Shape Memory Alloys (SMAs) are metallic mate- the façade and on the other end to a plate bolted
rials that show special thermomechanical proper- to a counter plate embedded in the RC rib that
ties due to a reversible transformation between two was cast to stiff the existing concrete roof (dating
crystalline configurations, austenite and martens- back to the 1950s). SMADs were three-plateau
ite, without degradation of the crystal structure. self-balanced devices of three different sizes,
For some SMAs, the phase transformation is with design capacity ranging from 17 to 52 kN
temperature-correlated, while for others, such as and maximum allowable displacement between
nitinol (NiTi SMA), the phase change can be 8 and 25 mm.
stress induced at room temperature.
The stress–strain curve, measured during a Dissipative Anchor Devices
monotonic tension test on a NiTi SMA wire, Given the difficulty of ensuring ductile failures of
shows: anchors embedded in masonry, D’Ayala and
Paganoni (2014) in collaboration with CINTEC
(a) An almost linear portion corresponding to the International developed two prototypes of dissi-
elastic deformation of the material in its aus- pative anchor devices to address the problem of
tenitic phase. out-of-plane mechanisms.
(b) A loading plateau due to the transformation The devices are conceived to be inserted at the
in martensite. joint between perpendicular walls, as part of lon-
(c) A new elastic phase that initiates after the gitudinal steel anchors grouted within the thick-
complete transition to the martensitic phase ness of the walls. Devices are designed to work in
(whose upper boundary is called maximum tension, similarly to traditional crossties, which
superelastic strain). experience axial loading when one wall panel
(d) A final plateau, related to the true yielding of tilts outwards as a consequence of ground accel-
the alloy, which ends with failure. eration. This type of installation ensures a low
(e) Stress removal causes a reverse phase transfor- impact on the aesthetics of the building as it does
mation, where the material goes back to the not affect the finishes.
austenitic phase, stable for lower load’s values. One device relies on the plastic behavior of
(f) Strains are almost completely recovered. This steel and consists of a metallic element designed
property is known as superelasticity. to achieve lower capacity than the standard tie to
Structures,
Fig. 14 Dissipative
anchor devices: (a)
hysteretic and (b) friction
based (D’Ayala, Paganoni
2014)
which it is connected (Fig. 14a). The second high stiffness and fail at the interface between
device (Fig. 14b) is made of a set of small plates; grout and parent material or in the masonry
bolts are used to apply a perpendicular pressure units. In the case of friction devices, it was
creating friction and adjusting the level of slip observed that, due to tolerances of the pieces
load beyond which relative sliding occurs. Spe- composing the assembly, a locking phenomenon
cially designed stops ensure that the sliding occurred for the higher levels of perpendicular
motion remains in the desired range. pressures; consequently, load-deflection curves
While the metallic profiles improve the box- feature additional stiffness in respect to the
like behavior of the building, contributing to an flat branch typical of a friction mechanism.
increase in stiffness that improves the structural Such behavior will need to be corrected so that
response to low excitations, the devices allow the device can provide homogeneous perfor-
small relative displacements between orthogonal mance over the required range of perpendicular
sets of walls; additionally, for higher horizontal forces.
loads, they dissipate part of the energy input into The dimensioning procedure is based on the
the structure so that problems of localized dam- idea that a strengthening system can be divided
age can be avoided. Therefore, the design focuses into subcomponents to which one type of failure
on control of displacements and reduction of controlled by a single parameter can be associ-
accelerations and stress concentration. ated. These parameters can be used in the formu- S
The two dissipative devices have been vali- lae prescribed by design codes to calculate the
dated by cyclic pseudo-static and dynamic tests capacities of components and hence determine
on the isolated devices and by pullout tests on the hierarchy of failure. The parameter values
the devices anchored in series to a conventional can be obtained via a number of sources: pro-
post-installed anchor embedded in low-shear ducers’ specifications, recommended or limit
capacity masonry (D’Ayala and Paganoni values provided by codes, and, if any, laboratory
2014). and/or on-site tests.
Experimental results confirmed that both the Table 1 exemplifies the procedure for the cal-
yielding and frictional elements are able to pro- culation of the tensile capacity of the dissipative
vide the anchorage with ductility and prevent anchoring devices in series with a grouted metal-
damage to occur in either the grouted sleeve or lic anchor. Maximum capacity is reached if one
the masonry, while traditional anchors, which of the components fails or when the dissipative
were tested for comparison purposes, display a device is activated. Further checks can also be
Seismic Strengthening Strategies for Heritage Structures, Table 1 Design of dissipative anchors: param-
eters that identify the tensile capacity, value range achieved during experimental validation, and range prescribed/
recommended by design codes (From D’Ayala and Paganoni 2014)
performed in terms of displacement and ductility, Capacities are calculated using information in the
as discussed in the following. relevant row in Table 1. It is important to notice
The demand in terms of tensile capacity of that while certain values in Table 1 are generally
metallic anchors is calculated as: applicable, others are not: the standardization of
steel production ensures highly repeatable perfor-
FD ¼ M a i (8) mance, whereas bond strength, for instance,
largely varies depending on the substratum.
Therefore, it is advisable that tests be performed
where
each time; if this cannot be done, lower limit
M: mass of structure that bears on the ith values provided by the code and reported in
anchor of the strengthening system. M depends
Table 1 can be used.
on the geometry and construction arrangement of
Table 1 does not include the connections’
the building, including horizontal structures, and anchor/dissipative devices and the stops of the
on the layout of the set of anchors to be designed.
friction device. However, these components
ai: horizontal acceleration experienced by the
should also be dimensioned to resist FDNC as
mass M. An estimate of the natural period of the also required by EN 15129:2009.
system can be used to determine the correct
The dissipative elements of the devices, either
design spectral ordinate and the distribution of
hysteretic or frictional, are designed to be acti-
amplification over the height of the structure. An vated at the threshold of damage limitation, when
accurate estimate of the acceleration at the height
cracks start opening and allow for the dissipative
of the anchor can be obtained either by using the
elements to become active. When the hysteretic
first natural modal shape or the procedure pro- devices enter the plastic field, or the friction
posed by Miranda and Taghavi (2005).
devices start sliding, the connection between
The reference acceleration is calculated as
wall panels is still maintained, but the pullout of
function of the three limit states defined in EN
the head of the anchorage is prevented and drift
1998-3:2004, so that the design demand is:
controlled.
Hence, from the point of view of force design,
• FDNC: near collapse, calculated for a seismic
depending on which dissipative device is used,
action with a probability of exceedance of 2 %
the following requirements apply:
in 50 years
• FDSD: significant damage, calculated for a
seismic action with probability of exceedance Fyield ¼ FDD or
(10)
of 10 % in 50 years F== ¼ FDD
• FDDL: damage limitation, calculated for a seis-
mic action with probability of exceedance of
where Fyield is the yielding capacity of the hys- S
20 % in 50 years teretic device and F// is the slip load of the fric-
tional device.
All the subcomponents of the strength-only
The dissipative devices should also comply
portion of the anchor assembly are brittle or, in with requirements for interstorey drift of build-
the case of the grouted steel elements, are not
ings undergoing seismic action. The chosen value
supposed to experience large deformations;
of maximum allowable drift, dr = 0.003, for
therefore, they are dimensioned in terms of damage limitation is taken from OPCM (2005)
strength for near collapse limit state, according
which provides drift limits for damage limitation
to EN 1998-3:2004. The minimum capacity in the in walls of masonry buildings. This limit is also in
assembly must be: line with the expected drift stated in FEMA
356 (BSSC 2000) for unreinforced masonry
Min Fsteel , Fa=b bond , Fb=p bond , Fmasonry > FDU buildings at the limit state of immediate
(9) occupancy.
For historic buildings, the allowable drift is Summary

increased by using the reduction factor n = 0.4,
which is taken from EN 1998:2004 and accounts In the last decades, a number of technical solu-
for the fact that devices are designed to be tions for the improvement of structural connec-
used in heritage structures, which fall in tions have been developed in response to the
the importance category III of Eurocode 8 (EN increasing demand for strengthening systems
1998:2004). specifically designed for the protection of heri-
It is therefore tage assets from earthquake-induced damage.
Notwithstanding recommendations from seis-
mic standards and conservation codes of practice,
dr ¼ De < 0:003 h=v (11)
and the availability of alternative technical solu-
tions on the market, much emphasis when choos-
where ing strengthening solutions of connections is still
De: the elongation of the device before yield- placed on their capacity to enhance the overall
ing, in case of the hysteretic device, or before strength of the system, rather than improve its
activation of the friction mechanism, in case of ductility or ability to dissipate energy. When con-
the frictional device, nections between vertical elements, and between
h: interstorey height, or vertical distance of vertical and horizontal elements, as well as vaults
installation of anchors. A standard distance of and domes have sufficient capacity, vertical and
3 m has been assumed in the calculations, but horizontal loads are better distributed, out-of-
anchors might need to be spaced more closely plane mechanisms of damage are prevented,
along the height of the wall to prevent substratum and, hence, failure of floors and roofs can be
failures. avoided, substantially reducing the probability
The first threshold of the hysteretic device, of collapse and fatalities in an earthquake.
identified at 0.5 % elongation of the dissipative Historically, builders mastered the art of care-
element, coincides with 46 % of the hysteretic ful detailing of joints from a process of trial and
device load capacity. The second threshold, at error; today’s engineers possess the necessary
5 % elongation and 72 % maximum load, should insight into physical and mechanical laws
instead be used to verify the capacity of the dis- governing the dynamics of structures to control
sipative device for the limit state of significant the process of structural upgrading at the design
damage, so that dissipation of energy is ensured level. Thus, when connections, in existing build-
during low-to-medium seismic excitations. ings, lack adequate capacity, further elements can
Beyond this limit, the device has a further margin be added to the original structure to achieve the
of safety given by buckling, meaning that the desired performance.
device can reach the limit of near collapse with The review of modern techniques to enhance
damage, but not complete failure, and it could connections presented in the previous section
still be substituted, as it has been proved by highlights aspects specific to each applications
experimental testing and in response to the as well as advantages and pitfalls typical of each
requirements of EN 1998:2004. system. A large majority of the discussed
In the case of a frictional device, the drift limit methods are applicable both at the local and
is ensured by default because: before activation global scale for repair and upgrade of single
of the friction mechanism, deformations are neg- structural elements or of whole structures.
ligible and, beyond activation of sliding, the In respect to the past, strengthening systems
device displacement is limited by the assembly for structural connections rely on a more accurate
stops. The device can therefore perform for all design, modern structural systems, and innova-
limit states, as long as the connections and stops tive and more durable materials, for example,
in the assembly are designed to resist up to the stainless steel or titanium, which substituted
state of near collapse. iron in crossties, with considerable advantages
Seismic Strengthening Strategies for Heritage Structures, Table 2 Performance parameters typical of each
strengthening technique
Typology of strengthening Performance parameters
Ring beams Flexural, tensile, and shear capacity of beam (kNm/kN)
Capacity of fixings between ring beam and parent material (crushing or pryout
of parent material, shearing off of fixings, shearing of parent material, etc.)
(kN)
Horizontal sliding shear capacity of the masonry underneath the level of the
fixings of the ring beam into the parent material ()
Wall plates Flexural, tensile, and shear capacity of wall plates (kNm/kN)
Capacity of fixings between wall plates and parent material (crushing or pryout
(kN)
Horizontal sliding shear capacity of the masonry underneath the level of the
fixings of the ring beam into the parent material ()
Steel connectors for floor/roof Tensile capacity (kN) as minimum depending on
structures
Anchoring with end plates Combined tensile and bending strength of connector/tie (N/mm2)
Anchoring without end plates and Tensile capacity of fixings in timber elements (kN)
with grouting
Nailing Bond strength (for bonded systems) connector/binder (N/mm2)
Bond strength (for bonded systems) binder/parent material (N/mm2)
Tensile capacity of fixings (frictional or mechanical fixings) in parent
material (kN)
Tensile strength of parent material for cone pullout or punching shear
(N/mm2)
Reduction of tensile capacity due to distance to edge (%)
Reduction of tensile capacity due to distance between anchors, nails, and
connectors (%)
Compressive strength of parent material under action of end plate (N/mm2)
Flexural capacity of end plate (kN)
Shear capacity (kN) as minimum depending on
Shear strength of anchor (N/mm2)
Tensile strength of anchor for combined axial and bending load (N/mm2)
Bearing capacity of parent material (N/mm2)
Bearing capacity (for bonded system) of binder (N/mm2)
Shear capacity of fixings in timber elements (kN)
Reduction of shear capacity due to distance to edge (%)
Reduction of shear capacity due to distance between anchors (%) S
Corners Column ties Flexural, tensile, and shear capacity of columns (kNm/kN)
confinement Capacity of fixings between ring beam and parent material (crushing or pryout
(kN)
FRPs Tensile capacity of FRPs (kN)
Bond strength between FRP/parent material (N/mm2) and the tensile and
compressive capacity of the parent material (N/mm2)
Polymer mesh, PP Tensile capacity (kN) as minimum depending on
mesh, steel mesh Tensile strength of mesh (N/mm2)
Tensile strength of fixings (N/mm2)
Bond strength fixing/binder (N/mm2)
Bond strength binder/parent material (N/mm2)
Shear strength of parent material for cone pullout (N/mm2)
(continued)
Seismic Strengthening Strategies for Heritage Structures, Table 2 (continued)

Shear strength of parent material at the interface strengthened/
unstrengthened material
Shear strength of fixings (N/mm2)
Bearing capacity of binder (N/mm2)
Bearing capacity of parent material (N/mm2)
Energy Yielding anchor Yielding load of device (kN)
dissipation devices Allowable displacement after yielding (mm)
systems Dissipated energy (kJ)
Change in material properties during cyclic loading (e.g., hardening of metal,
progressive crushing of parent material)
Buckling of device
Ultimate tensile capacity (kN) as minimum depending on
Tensile strength of device (N/mm2)
Other checks same as traditional anchors
Shear strength of device (N/mm2)
Other checks same as traditional anchors
Frictional anchor Load for activation of frictional mechanism (kN)
devices Ultimate tensile load of device (kN)
Ultimate allowable displacement for sliding (mm)
Dissipated energy (kJ)
deterioration of frictional surfaces, effect of thermal expansion, presence of
debris on frictional surfaces, progressive crushing of parent material)
Tensile capacity of device (kN)
Other checks as traditional anchors
Checks as per traditional anchors
Energy RAG energy Axial yielding load that leads to yielding of lead pins (kN)
dissipation absorbers Allowable displacement after yielding (mm)
system Dissipated energy (kJ)
progressive crushing of parent material)
Tensile strength of anchor (N/mm2)
Tensile strength of RAG device
Tensile strength of parent material to punching failure (N/mm2)
Compressive strength of parent material under action of end plate (N/mm2)
Flexural capacity of end plate (kN)
RETE energy Yielding load of the lead wire (kN)
absorbers Ultimate tensile load of device (kN)
Allowable displacement after yielding (mm)
Dissipated energy (kJ)
Change in material properties during cyclic loading (e.g., hardening of
metal, progressive crushing of parent material)
(continued)
Seismic Strengthening Strategies for Heritage Structures, Table 2 (continued)

Shock 1st threshold – thermal deformations
Transmission Units Allowable displacement (mm)
(STUs) and viscous Transmitted load for low-rate loading (kN)
dampers
2nd threshold – low seismic excitation, connections expected to perform as
pinned connections
Allowable displacements (mm)
Capacity (kN)
Capacity of connectors between STUs and parent material at 1st threshold
(kN)
3rd threshold – higher seismic excitation, connections expected to have
hysteretic response
Allowable displacement (mm)
Transmitted load (kN)
Dissipated energy for hysteretic response (kJ)
Shape Memory 1st threshold – thermal deformations
Alloy Devices Allowable displacement (mm)
(SMADs) Transmitted load for low-rate loading (kN)
2nd threshold – low seismic excitation, connections expected to perform as
pinned connections
Allowable displacements (mm)
Capacity (kN)
Capacity of connectors between STUs and parent material (e.g., pullout or
tensile failure of fixings) (kN)
3rd threshold – higher seismic excitation, connections expected to have
hysteretic response
Allowable displacement (mm)
Transmitted load (kN)
Dissipated energy for hysteretic response (kJ)
ith threshold: same parameters as per 2nd and 3rd threshold repeated for the
higher plateaus of Shape Memory Alloys
in terms of issues related to material deterioration while remaining economically feasible, espe-
(e.g., expansion due to corrosion, failures due to cially for vulnerable communities in developing S
reduced resisting section). countries.
Some of the techniques described have under- The most important shift for future implemen-
gone revision after observation of their perfor- tation is the increased awareness and technolog-
mance in the aftermath of major earthquakes ical research for economic solutions for ductility
proved that they are unsuited to certain applica- and energy dissipation-based devices, which can
tions. For instance, RC ring beams need to be allow the improvement of performance of entire
especially well connected to bearing walls; oth- building stock in historic centers of earthquake-
erwise, they tend to create shear failure in walls prone countries.
and collapse; furthermore, it has been shown that The performance of each system can be veri-
they are not a viable solution in case of multileaf fied by determining the parameters and carrying
masonry not well bonded. out the checks summarized in Table 2. The
Recent strengthening solutions are evolving to parameters listed refer, when available, to dimen-
include issues of environmental compatibility sioning procedures prescribed by codes and
guidelines and otherwise from the available tech- CNR-DT 200 R1/2013. Guide for the Design and Con-
nical literature included in this review. Some of struction of Externally Bonded FRP Systems for
Strengthening Existing Structures, ROMA – CNR
the systems in the table are not codified; there- October 10th 2013 – release of May 15th 2014
fore, the dimensioning procedure is not standard- Corradi M, Grazini A, Borri A (2007) Confinement of
ized or sometimes not reported for intellectual brick masonry columns with CFRP materials. Compos
property reasons; in these cases parameters are Sci Technol 67:1772–1783
D’Ayala D, Paganoni S (2011) Assessment and analysis of
suggested by the authors on the basis of their damage in L’Aquila historic city centre after 6th April
expertise and of the available information on the 2009. Bull Earthq Eng 9(1):81–104
behavior of the various devices. D’Ayala D, Paganoni S (2014) Protocol for testing and
The major advantage of Table 2 is that param- design of dissipative devices for out-of-plane damage
Struct Build 167(1):26–40
eters have been homogenized, using a common D’Ayala D, Speranza E (2003) Definition of Collapse
nomenclature, to facilitate the comparison of var- Mechanisms and Seismic Vulnerability of Historic
ious techniques. Masonry Buildings. Earthquake Spectra 19
(3):479–509. doi:10.1193/1.1599896
D’Ayala D, Yeomans D (2004) Assessing the seismic
Cross-References vulnerability of late Ottoman buildings in Istanbul.
In: IV international seminar on structural analysis of
historical constructions – SAHC04, Padua
▶ Passive Control Techniques for Retrofitting of DD CEN/TS 1992-4-1:2009. Design of fastenings for use
Existing Structures in concrete
▶ Retrofitting and Strengthening of Structures: Di Ludovico M, D’Ambra C, Prota A, Manfredi G (2010)
FRP confinement of tuff and clay brick columns,
Basic Principles of Structural Interventions
experimental study and assessment of analytical
▶ Strengthening Techniques: Masonry and models. J Compos Constr 14(5):583–596
Heritage Structures Eligehausen R, Mall R, Silva JF (2006) Anchorage in
concrete construction. Ernst Sohn, Berlin
EN 1991-1:2002. Eurocode 1- Actions on structures
EN 1992-1:2004. Eurocode 2- Design of concrete
References structures
EN 1993-1:2005. Eurocode 3- Design of steel structures
Benedetti D (2004) Increasing available ductility in EN 1995-1:2004 + A1:2008. Eurocode 5- Design of tim-
masonry buildings via energy absorbers. Shaking ber structures
table tests. Eur Earthq Eng 18(3):3–16 EN 1996-12005. Eurocode 6- Design of masonry
Borri A, Castori G, Grazini A, Giannantoni A (2007) structures
Performance of masonry elements strengthened with EN 1998-1:2004. Eurocode 8- Design of structure for
steel reinforced grout. In: Fiber-reinforced polymer earthquake resistance. Part 1: General rules, seismic
reinforcement for concrete structures – FRPRCS-8, actions and rules for buildings
Patras, 16–18 July EN 1998-3:2005. Eurocode 8- Design of structure for
Borri A, Castori G, Grazini A (2009) Retrofitting of earthquake resistance. Part 3: Assessment and
masonry building with reinforced masonry ring- retrofitting of buildings
beam. Constr Build Mater 23(5):1892–1901 EN 15129:2009: Anti-seismic devices
Building Seismic Safety Council (BSSC) EOTA, European Organisation for Technical Assessments
(2000) Prestandard and commentary for the seismic (2010) TR 029, Design of bonded anchors, Sept 2010.
rehabilitation of buildings, FEMA, 356. Federal Emer- Downloaded at www.eota.eu. Accessed 8 Jan 2014
gency Management Agency, Washington, DC EOTA, European Organisation for Technical Assessments
California Historical Building Code (2007) California (2013) TR 045, Design of metal anchors for use in
code of regulations. Title 24, part 8. California Build- concrete under seismic actions. Downloaded at www.
ing Standards Commission, International Code Coun- eota.eu. Accessed 8 Jan 2014
cil, Washington, DC Gigla B (2004) Bond strength of injection anchors as
Casadei P, Agneloni E (2008) Advance composites appli- supplementary reinforcement inside historic masonry.
cations on historical structures in Italy: Case studies In: XIII international brick and block masonry confer-
and future developments 6th International Conference ence, Amsterdam, pp 1–10
on Structural Analysis of Historical Construction Gigla B (2010). Comparison of failure of injection anchors
Chuxian S, Guiqiu L, Wenchao W (1997) The design of as supplementary reinforcement inside masonry and
brick masonry structure with concrete column. In: XI concrete. In: VIII international masonry conference,
brick and block masonry conference, Shanghai Dresden, pp 2019–2028
Seismic Tomography of Volcanoes 3117
Gigla B, Wenzel F (2000) Design recommendations for

injection anchors as supplementary reinforcement of Seismic Tomography of Volcanoes
historic masonry. In: XII international brick and block
masonry conference, Madrid, pp 691–706
Giuffrè A (1993) Sicurezza e conservazione dei centri Ivan Koulakov1,2 and Nikolay Shapiro3
1
storici: il caso di Ortigia. Laterza, Roma-Bari Trofimuk Institute of Petroleum Geology and
Hamoush SA, McGinley MW, Mlakar P, Scott D, Murray Geophysics, SB RAS, Novosibirsk, Russia
K (2001) Out-of-plane strengthening of masonry walls 2
with reinforced composites. J Compos Constr Novosibirsk State University, Novosibirsk,
5(3):139–145 Russia
3
Indirli M, Castellano MG, Clemente P, Martelli A (2001) Institut de Physique du Globe de Paris
Demo-application of shape memory alloy devices, the Laboratoire de Sismologie, Paris, France
rehabilitation of S. Giorgio Church Bell-Tower. In: VI
international symposium on smart structures and
materials – SPIE 2001, Newport Beach
Ministero deri Beni Architettonici e Culturali, Italia, Synonyms
Guidelines for evaluation and mitigation of seismic
risk to cultural heritage : (2007) Cangemi Editore
S.p.A. Ambient noise tomography; Body-wave tomog-
Karantoni F, Fardis M (1992) Effectiveness of seismic raphy; Magma sources; Seismic observations on
strengthening techniques for masonry buildings. volcanoes; Seismic properties of crust; Seismic
ASCE 118(7):1884–1902 tomography
Krevaikas TD, Triantafillou T (2006) Masonry confine-
ment with fiber-reinforced polymers. ASCE J Compos
Constr 9(2):128–135
Mandara A, Mazzolani FM (2001) Energy dissipation
devices in seismic up-grading of monumental
Introduction
buildings. In: III international seminar on structural
analysis of historical constructions – SAHC01, Volcano tomography is a branch of geophysics
Guimarães oriented to studying the deep structures beneath
Marcari G, Manfredi G, Pecce M (2003) Experimental
volcanoes by means of seismic tomography. Seis-
behaviour of masonry panels strengthened with FRP
sheets. In: Tan KH (ed) 6th international conference on mic tomography is a method for reconstruction
fibre-reinforced plastics for reinforced concrete struc- of continuous distribution of seismic parameters
tures, World Scientific, Singapore, pp 1209–1218 in 1D, 2D, 3D, or 4D (space and time) using the
Mayorca P, Meguro K (2004) Proposal of an efficient
characteristics of seismic waves traveling
technique for retrofitting unreinforced masonry dwell-
ings. In: XIII world conference on earthquake engi- between sources and receivers. Seismic parame-
neering, Vancouver ters to be found in tomographic inversion are in
Miranda E, Taghavi S (2005) Approximate floor acceler- most cases velocities of P and S seismic waves
ation demands in multistory buildings. I: formulation.
(P and S velocities). For volcanoes, one of the key
J Struct Eng 131:203–211
OPCM (2005) No. 3431, May 3, 2005. Official Bulletin parameters appears to be the Vp/Vs ratio which
no. 107, May 10, 2005 (in Italian). Gazzetta Ufficiale – can be used to evaluate the content of fluids and S
Serie Generale n. 107 melts. Besides the velocity distributions, seismic
Priestley MJN (2000) Performance based seismic design.
tomography may provide the information on the
In: XII world conference of earthquake engineering,
Auckland anisotropy of seismic parameters which helps
Spence R, D’Ayala D (1999) Damage assessment and studying regional stresses and space-oriented
analysis of the 1997 Umbria-Marche earthquakes. geological structures. In some tomography
Struct Eng Int 9(3):229–233
studies, the target parameter might be the attenu-
Tomazevic M (1999) Earthquake-resistant design of
masonry buildings. Imperial College Press, London ation of P or S waves which may also give impor-
Tumialan G, Huang P-C, Nanni A, Silva P (2001) tant information on magma sources beneath
Strengthening of masonry walls by FRP structural volcanoes.
repointing. In: Burgoyne CJ (ed) 5th international
At all stages of the human history, understand-
conference on fibre-reinforced plastics for reinforced
concrete structures, Thomas Telford, Cambridge, ing the mechanisms causing volcano eruptions
pp 1033–1042 was one of the most intriguing problems.
3118 Seismic Tomography of Volcanoes
Until relatively recent time, scientists had only Seismic Sources

indirect ways to explore processes beneath vol- Sources used for seismic imaging of volcanoes
canoes, mainly by making analogies with geo- can be divided in three categories: human-made
logical signatures of ancient magmatic systems. sources, earthquakes, and ambient seismic noise.
During the last decades, the advancement of Fully controlled artificial sources are expected to
geophysical methods made possible direct provide the highest accuracy of tomographic
observations of deep structures beneath pres- images. At the same time, their implementation
ently active volcanoes which give precious on volcanoes is very sophisticated and expensive.
information about general mechanism of work- Another shortcoming of artificial sources is that
ing the volcanic systems. Among all geophysical they cannot be produced at depth. In regions with
methods, seismic tomography is one of the most well-distributed volcanic or local tectonic seis-
powerful and effective approaches to look at micity, earthquakes may give a distribution of
great depths below the volcanoes. Using power- sources plausible for high-resolution seismic
ful artificial and natural sources of seismic signal tomography. However, many volcanoes do not
makes it possible getting the information from generate sustained seismicity. In this case, the
depths where the main magma sources are recently developed methods based on correla-
located. tions of ambient seismic noise become suitable
for imaging volcanic edifices.
Observation Schemes Human-Made Seismic Sources

Human-made seismic sources include explo-
The configurations and characteristics of obser- sions, vibrators, air-gun shots, and weight
vation systems in seismology depend on the prop- (or mass) drops and are usually called active
erties and sizes of the target objects, as well as on sources. There are some onshore experiments
the type of seismic sources. In case of volcano which use chemical explosions of sufficiently
seismology, the parameters of the target volume large magnitude to generate seismic signal prop-
depend on the type of volcanoes. For example, agating to the distances of dozens kilometers and
the subduction-related volcanoes usually have depths of a few kilometers. The problem of such
complex multilevel structures of feeding system explosions is that they are fairly expensive and
which may cover the depths down to thus their number is strongly limited. Further-
100–200 km. In the case of the hot-spot volca- more, chemical explosions are not ecologically
noes, the initial source of magmatism is related to friendly and they are prohibited in most volcanic
mantle plumes which may propagate throughout areas. In most cases, the surveys based on such
the mantle; however, the magma reservoirs and sources are used for 2D profiling (Fig. 1c).
the related seismicity are fairly shallow in this An alternative method of active-source
case (only the first kilometers of depth). When onshore generation might be using vibrators
designing an experiment for tomography studies which are more ecologically safe compared to
of volcanoes, one should clearly figure out the explosions. However, for the purposes of volcano
target properties and define the configuration of tomography, the vibrators should be sufficiently
the network according to this. In reality, there are powerful and heavy to enable the required prop-
many limitations, such as insufficient amount of agation of seismic rays. In reality, the transporta-
instruments and difficulties in access to some tion of powerful vibrators appears to be not
areas around a volcano that make impossible possible in most volcanic areas; therefore, they
deploying an ideal network. In the following sec- are not widely used in practice.
tion, the major components of the observation In case if a volcano is located close to the
systems are described, and some recommenda- seashore, it is possible to use underwater air
tions to optimize the designing of the seismic guns as sources of seismic signal (Fig. 2a). The
networks are given. advantage is that air-gun shots are rather cheap
Seismic Tomography of Volcanoes, Fig. 1 Observation schemes used in volcano tomography. Blue triangles depict
seismic stations; red dots are sources; black lines are seismic rays
and sufficiently powerful. They can be frequently The major advantage of the active sources is
generated along the ship cruise that gives a huge that their parameters are known. The problem is
data base with a clear technology of processing. that they are located close to the surface, and
The problem is that these sources are limited by seismic signal propagate along the nearly hori-
the shore line and not suitable for most volcanoes. zontal ray paths. Another problem, which is
Seismic Tomography of
Volcanoes,
Fig. 2 Schemes for active-
source experiments using
air-gun shots (a) and for
ambient noise tomography
(b). Blue triangles depict
seismic stations; red dots
are sources
mostly actual for the air-gun shots, is that they known and must be determined simultaneously
generate very weak S waves which cannot be with velocity models. This leads to a coupled
picked and used for tomography studies. problem with significant trade-off between veloc-
ity and source parameters. However, an impor-
Earthquakes tant advantage of natural sources compared to the
Volcanic and tectonic earthquakes located in the artificial sources is that they are located at some
volcanic area (including some events located out- depths which allow illuminating the target object
side the perimeter of the network) are regularly from below at different directions. For many
used as sources for seismic tomography of volca- cases, the amount of earthquakes is much larger
noes (Fig. 1b). The major problem with earth- than that of the artificial explosions. In addition,
quakes is that their location parameters earthquakes in most cases produce clear P and
(coordinates and origin times) are not exactly S waves which can be used for exploring P and
S velocities. In turn, both P and S velocities give of Broadband Seismometry”). It should be noted
much more constraints on the petrophysical state that for the body-wave tomography based on
of rocks below volcanoes than just a single local seismicity such broad frequency capacity
P model. is not really required: the signal from local earth-
quakes is clearly seen at frequencies of more than
Ambient Seismic Noise 1Hz.
Ambient seismic noise is the permanent motion Seismic recorder is a tool which digitizes the
of the Earth surface that is not related to earth- analogue electrical signal coming from the sensor
quakes or specific controlled sources (see also and records it onto a flash or another kind of
“▶ Seismic Noise”). When averaged over long memory. In each recorder, there should be also
time series, this wavefield can be approximately a GPS receiver which synchronizes the absolute
considered as produced by randomly and homo- time. In some cases, sensors and recorders are
geneously distributed sources (Fig. 2b). Cross- combined in one unit.
correlation between records of such random All recorders and most broadband sensors
wavefields at two stations yields the impulsive consume the electric power, and an important
response of the media (Green’s function) part of the seismic station is power supply. It
between these two points (e.g., Gouédard can include chemical rechargeable or single-used
et al. 2008). In the present form, every recorder batteries, solar panels, or wind generators. Using
can be converted into a “virtual” source recorded each type of power source is determined by the
at all other recorders resulting in N(N-1)/2 conditions of real experiments. For example, in
source-receiver pairs for a network of Kamchatkan volcanoes, solar panels cannot be
N stations, as shown in Fig. 2b. Shortcomings of used due to large amount of snow during the
noise-based approach are (1) dependence on most part of the year. Furthermore, there is a
noise sources distribution that in many cases is risk of destruction of any part located outside
not perfectly homogeneous possibly preventing the ground by wild animals or vandals. In this
successful reconstruction of impulsive responses case, an only possibility to conserve the station is
for a part of station pairs; (2) need of long con- to place the station deep into the ground together
tinuous records (months or years); and (3) the fact with batteries.
that reconstructed virtual sources are located at
the surface and are dominated by surface waves. Some Hints on the Network Design
Design of seismological networks for performing
Receivers tomography studies has some particular features.
Networks for active and passive surveys on vol- Here some hints which should be taken into
canoes use generally the same instruments as for account while deploying seismic networks are
similar seismological studies in other regions. given. S
Tomography inversion needs numerous stations
in a network (more than ten) which are usually • In case of using refracted rays from active
deployed temporarily. Such networks consist of (artificial) sources, when both sources and
portable seismic stations which are composed of receivers are located near the surface, a typical
three major components: sensor, recorder, and depth of seismic ray propagation is 1/10–1/15
power supply. In some cases, there is also a of the source-receiver distance (Fig. 1c). Thus,
device for telemetric transition of signal. to study the depths of 10 km beneath the
Seismic sensor is a kind of “microphone” volcano, the sources and receivers should be
which converts the ground movements into elec- located in the opposite sizes of the volcano at
tric signal. Modern sensors which are used for distances of 50–70 km.
volcano surveys are three-component broadband • The spacing between stations normally has the
instruments which allow recording oscillations of same order as the minimum size of resolved
more than 30–40 s period (see also “▶ Principles anomalies in the tomographic inversion.
Further decrease or increase of the resolution waves are produced by point sources. Along this
depends on the distribution of events. If there way, seismic rays accumulate the information on
are a very few events, densification of the the inner structure of the Earth. The purpose of
network would hardly bring to the resolution seismic tomography is to decipher this informa-
increase. In contrary, in case of a large amount tion and to reconstruct the structure inside the
of sources evenly distributed throughout the Earth. Most simple and most popular way for
study area, the resolved patterns can be many practical applications, including volcano
smaller than the station spacing. tomography, is the kinematic approach based on
• For passive sources (local earthquakes), the using travel times of seismic rays. Travel times of
depth of seismic model resolved from tomo- seismic rays can be represented as an integral
graphic inversion cannot be larger than the along the ray paths
maximum depths of sources.
ð
• For uncontrolled sources (earthquakes), a crit- dS
T¼ (1)
ical parameter for tomography is a number of V ðx, y, zÞ
picks per events. If this number is equal to g
4, one can theoretically locate the sources but

cannot get any information about the velocity where g is the ray path and V is seismic velocity.
model. In practice, the data are noisy, and to Here, an infinite frequency of the wave is pre-
get a stable location of source, more than four sumed which allows presenting the ray as a
picks are required. The information on the curved line. Actually, this is often not true in
velocity structure can be extracted only if the seismology where the wavelengths are in some
number of picks is more than 10. The higher cases compatible with the size of the target
this value is, the more velocity parameters can objects. For these cases, there are methods of
be retrieved and the higher resolution model finite-frequency tomography which are based on
will be obtained. “fat” rays representing a banana-shaped area hav-
• Using out-of-network events (within reason- ing a rather complex sensitivity patterns (e.g.,
able limits) can appear to be useful for tomog- Husen and Kissling 2001). Although there are
raphy as was shown by Koulakov (2009b). some successful applications of this method in
Although the location accuracy for such global or in teleseismic tomography (Montelli
events is fairly low, they improve the ray cov- et al. 2004), it was not yet widely implemented
erage and help in increasing the spatial reso- for volcano tomography. Up to today, most of the
lution beneath the network area. practical applications for studying the volcanoes
• In all cases, prior the deployment of the net- are based on the ray-based kinematic approach.
work, it is useful to simulate a series of syn- In general, the velocity field in Eq. 1 is a
thetic models with possible configurations of function which is represented by infinite number
network, sources, and target objects that are of mathematical points. In practice, to solve the
expected in the real case. This can help to inverse problem, i.e., to derive velocity distribu-
optimize the distribution of seismic stations tion based on given travel times, the velocity field
in the real experiment. should be parameterized with a finite number of
parameters. There are different methods of
parameterization, such as using regular or irreg-
Seismic Tomography Methods Based on ular cells, nodes with linear interpolation, har-
Body Waves monic functions with unknown coefficients, etc.
Another problem of solving the inverse prob-
Basic Principle of Body-Wave Seismic lem based on the representation in Eq. 1 is that
Tomography this problem is nonlinear. It means that one
In the ray-based approach which is used in most should find the velocity distribution based on
body-wave tomography algorithms, seismic the ray paths which are strongly affected by the
unknown velocity distribution. The direct solu- neighboring parameters (nodes or cells) that
tion of this nonlinear problem is almost never allows controlling the smoothness of the solution.
used for practical applications. Actually, in most The quality of the solution and optimal values
cases, this problem is reduced to the linear repre- of the inversion parameters are estimated based
sentation. It is assumed that the unknown velocity on synthetic modeling which should simulate the
distribution is close to the given reference model, conditions of a real experiment as close as possi-
and in this case, the problem in Eq. 1 is reduced to ble. The synthetic travel times are computed in an
the linear integral artificial model and perturbed by a realistic noise.
Then, the model parameters are “forgotten,” and
ð
Dv dS the reconstruction procedure is performed identi-
Dt ¼ (2)
V 20 ðx, y, zÞ cally as in the case of observed data processing.
g0
This step is especially important for the volcano
tomography where the distributions of sources
where Dv is unknown velocity anomalies, Dt is and events are often far from optimal.
time residuals, V02 is velocity in the reference There are different seismic tomography
model, and g0 is the ray path in the given refer- schemes which are used for different purposes
ence model. Then, the continuous representation in geosciences. Many of them are used for study-
in Eq. 2 can be reduced into the discrete system of ing the volcano feeding systems on scales from
linear equations: the entire mantle (thousands kilometers) to the
X uppermost layers (hundreds meters) (e.g.,
Dti ¼ Aij Dvj (3) Koulakov 2013). Here, only some examples of
j tomography studies on scales from first to dozens
kilometers are given. They provide the informa-
where Dv is the unknown value of jth velocity tion about magma sources in the crust and upper-
parameter and Aij is the first-derivative matrix most mantle that are directly responsible for
which represents an effect of a unit velocity eruptions of active volcanoes.
change of jth parameter on travel time ith ray.
In the case of cell parameterization, each element Teleseismic Tomography
of this matrix is just the length of jth ray in ith Seismic tomography is a relatively young method
cell. In the case of uncontrolled sources, this which started actively developed from the late
matrix also includes the elements for relocation seventies after the pioneer work by Aki
of sources (four parameters, dx, dy, dz, and dt, for et al. (1977). The method proposed in this study
each source). In practical tomography studies, is called teleseismic tomography, and it uses the
where this matrix may appear to be very large, relative time delays from distant (teleseismic)
the system in Eq. 3 is solved using the iterative earthquakes recorded by seismic stations located S
LSQR algorithm (Paige and Saunders 1982) in the study area (Fig. 1a). There were several
which proceeds only nonzero elements in the attempts of using this method for studying the
matrix and performs the calculations line by line volcanoes. For example, Ellsworth and Koyanagi
without keeping the entire matrix in the memory. (1977) have implemented the teleseismic scheme
The direct solving of the system as it stands in for studying the crustal structure beneath Kilauea
Eq. 3 normally fails because of strong instability volcano in Hawaii. Three years later, Sharp
of the solution due to noise in the data. To avoid et al. (1980) have published the tomographic
this problem, the inversion should be regularized model for Mt. Etna. Among later studies, one
using some additional terms. The most popular can mention the work by Stauber et al. (1988)
way of regularization is the amplitude damping who presented tomographic models of Newberry
which is realized by adding a diagonal matrix Volcano, Oregon, based on teleseismic delays.
block with zero data vector. Another way is It should be noted that during the last decades,
damping a gradient between all pairs of the teleseismic tomography is almost not used
anymore for studies of crustal structures beneath deployment of a large number of seismic stations
volcanoes because of several reasons. The first (for more than several months) which is required
problem is that nearly vertical seismic rays in the to accumulate sufficient information on the seis-
study area cause poor vertical resolution which micity in the volcano area. This needs solving
leads to strong vertical smearing of the retrieved some logistical issues related to providing
anomalies. Another problem, which seems to be power supply and hiding the equipment.
more serious, is related to the fact that the The first LET algorithm was developed and
teleseismic waves are usually of low frequency implemented to real seismological data by Thurber
(around 1 Hz and lower). Thus, the large wave- (1983). One year later, the same algorithm was
lengths strongly limit the resolution capacity of used for studying the crustal structure beneath
this method. Theoretically, it cannot provide Kilauea volcano (Thurber 1984). Among the later
robust images of anomalies smaller than studies based on this approach, one can single out
8–15 km. Magma reservoirs are usually expected the studies of Hengill-Grensdalur volcanic com-
to be smaller in size, and thus, they are hardly plex in Iceland (Foulger and Toomey 1989) and of
resolvable by teleseismic tomography. Mount St. Helens (Lees 1992). There are more
than ten different LET studies of Mt. Etna which
Local Earthquake Tomography appears to be one of the best studied volcanoes in
Another method, which is called local earth- the world (e.g., Villasegnor et al. 1998; Chiarabba
quake tomography (LET), uses the data of et al. 2000; Patane et al. 2006). Note that the work
arrival times of P and S waves from earthquakes by Patane et al. (2006) was a pioneer study
located within the study area or slightly outside presenting the 4D tomography model which
(Fig. 1b). Because of using the uncontrolled revealed temporal variations of seismic structure
sources, the solution in this case is reduced to beneath the volcano having possible relations to
the coupled problem of simultaneous determina- the eruption activity (see also “▶ Seismic Moni-
tion of source parameters and velocity models. toring of Volcanoes”). During the last decades, the
Active volcanoes usually produce a lot of seis- LET methods have become the most popular for
micity related to processes in magmatic reser- studying crustal structure beneath volcanoes.
voirs and tectonic displacements in the crust. Nowadays, there are dozens of successful studies
Therefore, deploying stations for the LET exper- of volcanoes in different parts of the world. In the
iment is much cheaper than organizing active- final part of this paper, two examples of recent
source works with the use of explosions or studies of Mt. Spurr in Alaska (Koulakov
air-gun shots. Furthermore, the earthquakes in et al. 2013a) and of Klyuchevskoy volcano group
passive source schemes are distributed at some in Kamchatka (Koulakov et al. 2011, 2013b) will
depths that enable much more favorable observa- be presented. Both of these studies were performed
tion system than in the case of active-source using the LET code LOTOS (Koulakov 2009a).
studies. In addition, seismic records from earth-
quakes usually provide the data on P and S waves Studies with Human-Made Sources
that allows getting more information on Human-made sources or active sources are used
petrophysical state of rocks in volcanoes than in many seismic studies of volcanoes. The first
just a single P-velocity model which is usually experiments of three-dimensional active-source
resulted from the active-source studies. The prob- investigations of volcanoes were made in two
lem of LET studies is an uncontrolled distribution relict calderas located in the Cascade Range,
of sources which is not always optimal to get western USA. One of them is the potentially
optimal ray coverage in the target areas. An active Newberry shield volcano with the caldera
important difficulty of LET is the uncertainty of of 8 km diameter. In the active seismic exper-
the source locations and the trade-off between iment, Achaer et al. (1988) used three chemical
source and velocity parameters. Another techni- explosions and a dense receiver network covering
cal problem of LET is the necessity of fairly long the caldera. They obtained a rather clear
low-velocity anomaly at about 3 km depth which Antarctic which was studied by Zandomeneghi
was interpreted as a magma chamber. Another et al. (2009). This island has a circular shape
study was made in the same year in the Medicine and is suitable for performing air-gun shots
Lake Caldera by Evans and Zucca (1988). They inside and outside the circle. Using onshore
deployed 120 seismometers and made eight and offshore seismometers made this observa-
chemical explosions which were recorded by all tion system suitable for high-resolution tomog-
stations. The tomography inversion of travel time raphy studies. The inversion of travel time data
data revealed a low-velocity pattern beneath the revealed the seismic structures down to 2 km
caldera which was also interpreted as a magma which appeared to be linked with the main
source. geological structures. The circular rim of the
Performing 3D experiments using onshore caldera is associated with high-velocity linea-
explosions appeared too expensive and not pro- ments. Inside the caldera, a low-velocity anom-
viding sufficient spatial resolution with the avail- aly is observed.
able number of explosions. Therefore, many • A network with onshore and offshore stations
onshore experiments were mostly organized and 4,414 air-gun shots were used by Paulato
along one or several profiles (Fig. 1c). For exam- et al. (2010) to build a tomographic model
ple, Zollo et al. (1996) used four chemical explo- beneath Montserrat Island in Lesser Antilles.
sions on a profile passing through the summit of They found that the location of the Island is
the Vesuvius volcano (Italy). Their P-velocity associated with high P velocity. Especially
model reveals a rather clear high-velocity anom- sharp positive anomaly is located beneath the
aly beneath the volcano. A similar experiment, Soufriere volcano.
but with five explosions on two crossing profiles, • The high-resolution 3D structure beneath the
was performed by Aoki et al. (2009) for the Tenerife Island was studied by Garcia-Yeguas
Asama volcano in Japan. Similarly, as in the et al. (2012) based on a large amount of
previous case, they found a clear high-velocity air-gun shots and fairly dense onshore seismic
body beneath the volcano. network consisting of 140 stations. The main
It should be noted that logistically the experi- volcanic structure of the Las Cañadas-Teide-
ments based on chemical explosions are very Pico Viejo Complex (CTPVC) is character-
difficult. Because of high expenses, the number ized by a high P-wave velocity body, similar
of shots is strictly limited that makes it problem- to many other stratovolcanoes. Furthermore,
atic to illuminate the target objects from different reduced P-wave velocities are found in a small
directions. Furthermore, the 2D approximations confined region in CTPVC and are more likely
are often too rough for the volcanic areas where related to hydrothermal alteration, as indi-
the expected structures are strongly 3D. In addi- cated by the existence of fumaroles.
tion, the explosion type sources are not ecologi- S
cally safe, and in most volcanic areas, they are
prohibited now. Imaging and Monitoring of Volcanoes
Good solutions for areas which are close to the Based on Ambient Seismic Noise
seashore are the underwater air-gun shots which
can be used to generate high-quality and suffi- General Principle, Data Processing, and
ciently powerful seismic signal and can surround Inversion Approaches
the target area (Fig. 2a). Nowadays, this scheme New methods for probing the Earth’s interior
is widely used in experiments in many volcanic using noise records only emerged during the last
areas. Among many different studies, three decade (e.g., Campillo et al. 2011). Origin of
examples are selected here: seismic noise strongly depends on the considered
spectral range (see also “▶ Seismic Noise”). At
• A very interesting site for seismic investiga- high frequencies (>1Hz), the noise is strongly
tions is the volcanic Deception Island in dominated by local sources that are often related
to strong winds or the anthropogenic activity. Using surface waves extracted from correla-
Seismic noise at intermediate periods (between tions of seismic noise for imaging of the Earth’s
1 and 20 s) that is most often used for studies of interior is called ambient noise surface-wave
volcanoes is dominated by natural sources. In tomography. This approach generally consists of
particular, the two main picks in the seismic three steps (e.g., Ritzwoller et al. 2011). First,
noise spectra in this so-called microseismic noise cross-correlations between every pair of sta-
band (1–20 s) are related to forcing from oceanic tions in the array are computed, and interstation
gravity waves. The interaction between these phase or group travel times as a function of period
oceanic waves and the solid Earth is governed are measured from the resulting waveforms. Sec-
by a complex nonlinear mechanism (Longuet- ond, two-dimensional maps of the speed of Ray-
Higgins 1950), and the noise excitation depends leigh or Love waves as a function of frequency are
on many factors such as the intensity of the oce- produced. At the last step, these maps are inverted
anic waves but also the intensity of their interfer- for a 3D distribution of shear speeds.
ences as well as the seafloor topography. Overall,
the generation of seismic noise is expected to be Applications of Ambient Noise Seismic
strongly modulated by strong oceanic storms and, Tomography (ANST) to Studies of Volcanoes
therefore, to have a clear seasonal and After recent emergence of the ambient noise
nonrandom pattern. However, when averaged surface-wave tomography (e.g., Shapiro
over long time series, distribution of sources of et al. 2005), it became clear that this method
microseismic noise can be approximated as ran- might be very useful for studying structures
dom and homogeneous. In this case, cross- within volcanoes because of its relatively low
correlation of noise records between two stations coast (especially comparing with active-source
yields the impulsive response (Green’s function) studies) and because of good coverage of the
between these two points (e.g., Gouédard shallow part of edifices compared with
et al. 2008). In the case of a uniform spatial earthquake-based body-wave imaging. First
distribution of noise sources, the cross- “volcanic” application of ANST was performed
correlation of noise records converges to the at Piton de la Fournaise located at La Réunion
complete Green’s function of the medium, Island (Brenguier et al. 2007). This hot-spot vol-
including all reflection, scattering, and propaga- cano is very active with on average more than one
tion modes. However, in the case of the Earth, eruption per year during the past two centuries.
most of ambient seismic noise is generated by The noise-based tomography was based on cor-
atmospheric and oceanic forcing at the surface. relations of 18 months of records by 21 short-
Therefore, the surface-wave part of the Green’s period vertical-component stations of the perma-
function is most easily extracted from the noise nent volcano monitoring network. Rayleigh wave
cross-correlations. group velocities could be measured between
Seismic surface waves, in contrast with body 2 and 5 s of period. Inversion of these dispersion
waves, are waves that propagate in a waveguide curves resulted in a shear velocity model
near the Earth’s surface. There are two types of extending between the surface and the depth of
surface waves: Rayleigh (vertically polarized 1 km below sea level (while average elevation of
waves) and Love (horizontally polarized waves the volcano is 2 km). The main result of this
transverse to the direction of motion). Their depth study was imaging of an intrusive high-velocity
sensitivity depends on frequency, with a fair body below the main crater moving westward
approximation being about a third of a wave- from the surface to 1 km below sea level. This
length. As a result, the surface waves are strongly body is located above the shallow magma reser-
dispersive with longer periods (wavelengths) voir and coincides with the region through which
propagating faster than shorter periods because the magma ascends during the summit eruptions.
of the general increase of seismic speed with Following the pioneering study of Brenguier
depth in the Earth. et al. (2007), the ANST was applied to image
subduction zone volcanoes: Okmok in Aleutian distribution of the wavefield sources, while prop-
Islands (Masterlark et al. 2010), Lake Toba in erties of a real seismic noise may deviate from
Sumatra (Stankiewicz et al. 2010), and Asama this simplified model. This implies that applica-
in Japan (Nagaoka et al. 2012). The first study tions of the noise-based imaging or monitoring
used 12 stations from a permanent monitoring should be accompanied by studies of properties
network, and the second study used 40 stations of recorded seismic noise to characterize the
from a temporary deployment, while the third degree of its spatial and temporal inhomogeneity
study used 81 sensors combined from permanent and to determine the minimal time required for
and temporary networks. In all these studies, the good convergence of noise cross-correlations.
authors reported very similar results: imaging of Development of advanced methods for the
low-velocity anomalies corresponding to magma preprocessing of the noise time series in order to
reservoirs at depths below 4 km. improve the quality of the extracted signals and to
decrease the convergence time is a very active
Noise-Based Volcano Monitoring research area.
One of the advantages of using continuous seis- Up to date, the noise-based seismic tomogra-
mic noise records to characterize the earth mate- phy was based on surface waves that are most
rials is that a measurement can easily be repeated easily extracted from noise cross-correlations but
(see also “▶ Seismic Monitoring of Volcanoes”). have a limited resolution in deep layers. Most
First application of this idea to volcano monitor- recent results demonstrate that, when working
ing was done by Sens-Schönfelder and Wegler with large and dense arrays of seismometers, it
(2006) who used the repetitive waveforms of is possible to indentify body waves in correlation
seismic noise cross-correlations to track changes waveforms and to measure their travel times
in the subsurface material properties on mount (e.g., Boué et al. 2013; Lin et al. 2013). This
Merapi in Indonesia during 2 years. This study gives a hope that noise-based body-wave tomog-
was based on high-frequency (>1 Hz) waves and raphy of volcanoes could be developed in the
mainly revealed seasonal variations caused by future.
changes of hydrological conditions in a very shal-
low layer. Brenguier et al. (2008) used noise
cross-correlations at intermediate periods Examples of Tomography Studies
(between 1 and 10 s) to detect decreases in seis- of Volcanoes
mic velocity a few weeks before eruptions that
suggests preeruptive inflation of the volcanic Tomography of Mt. Spurr
edifice of the Piton de la Fournaise. Mount Spurr is an active predominantly andesitic
A comprehensive review of recent advancements volcano in Alaska located at the northeastern end
in the noise-based volcano monitoring is proof the Aleutian arc. Before a single eruption in S
vided by Brenguier et al. (2011) (see also 1953, it was considered as a dormant volcano. In
“▶ Tracking Changes in Volcanic Systems with 1992, a series of three explosive eruptive events
Seismic Interferometry”). occurred on June 27, August 18, and September
16–17 in a small composite cone forming the
Current Limitations and Prospective Crater Peak on the south flank of Mount Spurr.
Application of seismic noise cross-correlation The structure beneath Mt. Spurr in a period of
methods for studies of volcanoes is a rapidly time nearly corresponding to this eruption was
growing field. However, these methods have studied by Power et al. (1998) using the data
some important limitations that are mainly collected from 1991 to 1993 by 7–11 seismic
related to the properties of the seismic noise stations of the Alaska Volcano Observatory
recorded at the Earth surface. So far, the basic (AVO).
theoretical foundation of the cross-correlation The most recent episode of Mt. Spurr unrest
method assumes a random and homogeneous occurred in 2004–2006 which accompanied with
strong seismicity and fumarolic activity. How- The combination of high Vp and low Vs suggests
ever, this episode did not result at any magma the presence of liquid fluids and/or melts and the
eruption, and thus, it was called as “lost erup- rock composition corresponding to the low
tion.” For 1 month in June 2005 corresponding to crustal or even mantle levels. The existence of
the middle stage of this unrest, the AVO deployed very low values of Vp/Vs above the NW anomaly
a fairly dense seismic network consisting of might be an indicative for aggregate transition
26 broadband three-component stations which occurred between the top of the plume and the
operated in addition to 11 short-period stations surface. It is important that the NW finger loca-
of the permanent network. During this period, tion generally coincides with the presence of
512 crustal earthquakes were located in the area fumaroles and active volatile flux during the
of Mt. Spurr using P and S picks recorded by most recent eruption in 2004–2005. It can be
portable and permanent stations. The final cata- proposed that this liquid fluids brought by the
log contained 5,960 P- and 4,973 S-wave arrival NW conduit are transformed to gases at shallow
times with an average of 20 picks per event. depths which can escape to the surface without
These data were used to perform the tomographic producing any stresses. This explains lack of any
inversion presented in Koulakov et al. (2013a). explosive activity in this location during the
The result of tomographic inversion including 2004–2005 unrest of Mt. Spurr (Fig. 3).
the 3D distributions of P and S velocities, Vp/Vs Another conduit located below the Crater
ratio, and accurate source locations is presented Peak seems to be blocked at 20 km depth.
in Fig. 3. The horizontal section at 5 km depth This fact is supported by the occurrence of deep
demonstrates a prominent anomaly of high seismicity just above this anomaly. In this
Vp/Vs ratio which reaches the value of 1.9 blocked channel, the accumulation of high pres-
located beneath the northern rim of the volcano. sure is more probable which makes it more dan-
This pattern coincides with the distribution of gerous for a potential explosive eruption. Note
shallow seismicity. In the deeper section at that the explosive eruption of Mt. Spurr in the SE
22 km depth, this anomaly is located at the rim in 1992 occurred just above this anomaly.
same place, but it appears to be less intensive
than another zone of high Vp/Vs ratio which is Tomography of the Klyuchevskoy Volcano
collocated with a cluster of deep crustal earth- Group
quakes. In vertical section, these two anomalies The Klyuchevskoy volcano group is located in
look like two fingers of high Vp/Vs ratio sepa- Kamchatka, in the far east of Russia. It includes
rated by 8 km of lateral distance. One of the two 13 active and dormant volcanoes of various types
anomalies nearly reaches sea level beneath the and compositions located in a relatively small
northern border of the caldera, but is separated area of approximately 100 60 km. The main
from the surface with a zone of low Vp/Vs ratio. volcano of the group, Klyuchevskoy Sopka, has
The top of the second anomaly appears to be the elevation of 4,800 m, and it is the highest
much deeper and may reach a depth of 20 km. volcano in continental Eurasia. Another volcano
The earthquakes that occurred during the 3- of the group, Bezymianny, located at approxi-
month span show distinct clusters near the tops mately 10 km distance from the Klyuchevskoy
of each of these fingers which support the active volcano is an explosive-type andesitic volcano.
behavior of these structures. It is important that One of the world’s largest explosive eruptions in
for both these finger-shaped anomalies, the high the twentieth century occurred here in 1957. The
Vp/Vs correspond to higher P and lower Tolbachik complex is situated in the southwest-
S velocities. ern part of the Klyuchevskoy group, and it con-
The observed finger-like features are sugges- sists of several shield and stratovolcanoes of
tive of pathways through the crust and possibly predominantly basaltic compositions. Plosky
correspond to complexes of conduits that trans- Tolbachik, which is currently active, is a typical
port liquid volatiles and/or melts from the mantle. Hawaiian-type volcano with fissure eruptions and
Seismic Tomography of Volcanoes, Fig. 3 Distribution the interpretation of the results. Blue waved arrows indi-
of the Vp/Vs ratio in two horizontal and one vertical cate hypothetical migration of fluids and their escape to
section. Dotted line depicts the limits of the caldera rim. the atmosphere in 2004–2006 unrest (Reproduced from
Triangles indicate the locations of stations. Yellow dots Koulakov et al. 2013a, JRL)
mark the source locations. Bottom-right plot represents
calderas. The existence of so many different vol- reveal at least two levels of magma sources in the
canoes located close to each other can most likely crust: one is located at a depth of 10–12 km and
be explained by complex processes of another is just below the Klyuchevskoy volcano
fractioning, mixing, and melting in magmatic close to the surface. This may explain the diver-
reservoirs located under the volcanic group at sity of the volcanic styles in the group. The dis-
different depth levels. tribution of earthquakes appears to link different
During the last several decades, the volcano- levels of magma storage that may trace the paths
related seismicity in the Klyuchevskoy group has of magma migration.
been monitored by the Kamchatkan Branch of Time variations of seismic structure beneath
Geophysical Survey. After 1999, a network the Klyuchevskoy group in a time period from
consisting of 17 digital telemetric stations was 1999 to 2009 were studied by Koulakov et al.
installed in the region of the Klyuchevskoy (2013b) and presented in Fig. 4. A common fea-
group. From 1999 to 2009, it provided the infor- ture which is seen in all yearly windows is the
mation on more than 80 thousand events and the anomaly with an extremely high Vp/Vs ratio
corresponding half million of P and S arrival located below 25 km depth, which was previ-
times. The tomographic inversions were ously mentioned for the year 2004 and interpreted
performed based on selected yearly subsets. The as a mantle conduit feeding all volcanoes of the
temporal variations of seismic structure are espe- group. For the crust, this study demonstrates con-
cially interesting in the context of the volcanic siderable temporal changes of seismic structures.
activity during the considered time. In particular, In particular, prior the 2005 eruptions of the
in 2005, there were simultaneous eruptions of the Klyuchevskoy and Bezymianny volcanoes, the
Klyuchevskoy and Bezymianny volcanoes that seismic structures in the crust look generally sim-
give a possibility to explore the changes of the ilar and contain two anomalies of high Vp/Vs
crust related to the volcanic activity at stages of ratio beneath the Klyuchevskoy volcano at a
preparation, activation, and relaxation (see also depth of 10–12 km and close to the surface. As
“▶ Seismic Monitoring of Volcanoes”). mentioned above, these anomalies are interpreted
The results of tomographic inversion as two levels of magma sources. In 2005, the
corresponding to the year 2004 just before the average value of the Vp/Vs ratio increases, and
activation phase were presented in a separate it may be explained by a considerable increase of
study by Koulakov et al. (2011). Similar struc- the fluid and melt content in the crust. In this year,
tures are observed in the precedent 2003 year a strong anomaly of high Vp/Vs ratio is observed
(Koulakov 2013). The key feature of these beneath the Bezymianny volcano, and it appears
models is a strong anomaly of a high Vp/Vs to link the volcano with mantle sources. In other
ratio, reaching values of 2.2, located below years, this anomaly was not observed. The shal-
25 km depth. Within this anomaly, the P velocity low anomaly of high Vp/Vs anomaly, which was
appears to be higher than average, whereas the previously observed beneath the Klyuchevskoy
S velocity is strongly low. This combination of volcano, disappears in 2005. The deeper anomaly
seismic anomalies is a possible indicator for a at 10–12 km depth is still presented. In the years
mantle conduit beneath the Klyuchevskoy group after the eruption, 2006 and 2007, the average
with rocks brought from deep levels and high Vp/Vs ratio lowers, and all the patterns at
content of fluids and melts. It was proposed that 10–12 km depth that were interpreted as interme-
this conduit is the original magma source that diate magma sources in the crust are not observed
feeds all of the volcanoes in the Klyuchevskoy anymore. Three years later, in 2008, a similar
group. anomaly at the same location starts reappearing
The tomography result corresponding to 2004 again.
reveals complex contrasted structures in the crust The observed variations of the seismic prop-
which can be associated with magma sources. erties are probably related to changes in the stress
The distributions of high values of Vp/Vs ratio and deformation fields, migration of fluids, and
Seismic Tomography of Volcanoes
Seismic Tomography of Volcanoes, Fig. 4 Cross sections of Vp/Vs from tomo- less than 0.5 km from the profile, and triangles mark the locations of the volcanoes near
graphic inversions for each year from 2001 to 2008 for a profile passing through the profile. The eruptions of Klyuchevskoy and Bezymianny volcanoes in 2005 are
Klyuchevskoy (KLU) and Bezymianny (BEZ) volcanoes. In areas of high Vp/Vs highlighted. The main phases and structural elements, which are discussed in the text,
ratios, yellow lines show contours of Vp/Vs = 2.0, 2.1, and 2.2. Dots show earthquakes are indicated (Reproduced from Koulakov 2013, JVGR)
3131
S
aggregate transformations in the crust. The man- During the last years, there appear more and
tle conduit, which is visible as an anomaly of high more dense networks on different volcanoes of
Vp/Vs ratio below 25 km depth at all time the world designed for passive and active exper-
periods, should have strong mechanical, thermal, iments. There are also new algorithms actively
and/or chemical influence upon the bottom of the developed during the last years, such as scatter-
crust and cause strong seismicity. These stresses ing tomography or full waveform inversion,
lead to the origin of fractures and upward pene- which might appear very effective for studying
tration of the fluids to the crust. Reaching inter- volcanoes. This gives a hope that increasing
mediate levels, these fluids lead to melting of the amount of data and implementing new algo-
overheated rocks. Due to decompression, some of rithms will lead in the nearest time to a real
the fluids are transformed to gases that increase breakthrough in understanding the causes of vol-
the pressure and cause additional fractures. These canic activity.
avalanche-type processes finally lead to the erup-
tions. During the eruption, most of the fluids
Acknowledgments The contribution of Ivan Koulakov is
escape to the atmosphere, and it causes a decrease supported by the Russian Scientific Foundation (grant #14-
in the melt content in the intermediate magma 17-00430).
sources in the post-eruption period. Several years
later, a new portion of fluids comes from the
mantle conduit, and melting of rocks in interme- References
diate reservoirs starts again. An important con-
clusion that can be drawn from the 4D Achauer U, Evans JR, Stauber DA (1988) High-resolution
seismic tomography of compressional wave velocity
tomography study is that the magmatic sources
structure at Newberry Volcano, Oregon Cascade
appear to be dynamic systems with rapidly vari- Range. J Geophys Res Solid Earth (1978–2012)
able properties. According to this concept, the 93(B9):10135–10147
magma sources are composed of sponge-type Aki K, Christoffersson A, Husebye ES (1977) Determina-
tion of the three-dimensional seismic structure of the
materials with the capacity to change the propor-
lithosphere. J Geophys Res 82(2):277–296
tion of the molten partition depending on the Aoki Y, Takeo M, Aoyama H, Fujimatsu J, Matsumoto S,
presence of fluids. Propagation of cracks and Miyamachi H, Yamawaki T (2009) P-wave velocity
fluids may occur rather rapidly, and this may structure beneath Asama Volcano, Japan, inferred
from active source seismic experiment. J Volcanol
abruptly change the properties of the “sponge”
Geotherm Res 187(3):272–277
material over a period of years or even months or Boué P, Poli P, Campillo M, Pedersen H, Briand X, Roux
weeks. P (2013) Teleseismic correlations of ambient seismic
noise for deep global imaging of the. Earth Geophys
J Int 194(2):844–848. doi:10.1093/gji/ggt160
Brenguier F, Shapiro NM, Campillo M, Nercessian A,
Summary Ferrazzini V (2007) 3-D surface wave tomography of
the Piton de la Fournaise volcano using seismic noise
This paper presents an overview with the descrip- correlations. Geophys Res Lett 34:L02305.
doi:10.1029/2006GL028586
tion of several seismic tomography schemes
Brenguier F, Shapiro N, Campillo M, Ferrazzini V,
which are used for studying volcanoes. It Duputel Z, Coutant O, Nercessian A (2008) Toward
describes several examples of practical works forecasting volcanic eruptions using seismic noise. Nat
including two featured case studies of Mt. Spurr Geosci 1(2):126–130
Brenguier F, Clarke D, Aoki Y, Shapiro NM, Campillo M,
and Klyuchevskoy volcano group. It was demon-
Ferrazzini V (2011) Monitoring volcanoes using seis-
strated that seismic tomography gives rich mate- mic noise correlations. C R Geosci 343:633–638.
rial which allows better understanding of the doi:10.1016/j.crte.2010.12.010
behavior of magmatic systems. A variety of the Campillo M, Roux P, Shapiro NM (2011) Correlations
of seismic ambient noise to image and to monitor
tomographic solutions show that there is no uni-
the Solid Earth. In: Gupta HK (ed) Encyclopedia of
form structure: each volcano has some particular solid earth geophysics. Springer, Dordrecht,
features and appears to be unique. pp 1230–1235
Chiarabba C, Amato A, Boschi E, Barberi F (2000) Recent Lees JM (1992) The magma system of Mount St. Helens:
seismicity and tomographic modeling of the Mount non-linear high-resolution P-wave tomography.
Etna plumbing system. J Geophys Res Solid Earth J Volcanol Geotherm Res 53(1):103–116
(1978–2012) 105(B5):10923–10938 Lin FC, Li D, Clayton RW, Hollis D (2013) High-
Ellsworth WL, Koyanagi RY (1977) Three-dimensional resolution 3D shallow crustal structure in Long
crust and mantle structure of Kilauea Volcano, Hawaii. Beach, California: application of ambient noise
J Geophys Res 82(33):5379–5394 tomography on a dense seismic array. Geophysics
Evans JR, Zucca JJ (1988) Active highresolution seis- 78(4):Q45–Q56. doi:10.1190/geo2012-0453.1
mic tomography of compressional wave velocity and Longuet-Higgins MS (1950) A theory of the origin of
attenuation structure at Medicine Lake Volcano, microseisms. R Soc Lond Philos Trans Ser A 243:1–35
Northern California Cascade Range. J Geophys Res Masterlark T, Haney M, Dickinson H, Fournier T, Searcy
Solid Earth (1978–2012) 93(B12):15016–15036 C (2010) Rheologic and structural controls on the
Foulger GR, Toomey DR (1989) Structure and evolution deformation of Okmok volcano, Alaska: FEMs,
of the Hengill-Grensdalur Volcanic Complex, Iceland: InSAR, and ambient noise tomography. J Geophys
geology, geophysics, and seismic tomography. Res 115:B02409. doi:10.1029/2009JB006324
J Geophys Res Solid Earth (1978–2012) 94(B12): Montelli R, Nolet G, Dahlen FA, Masters G, Engdahl ER,
17511–17522 Hung SH (2004) Finite-frequency tomography reveals
Garcı́a-Yeguas A, Koulakov I, Ibáñez JM, Rietbrock A a variety of plumes in the mantle. Science 303(5656):
(2012) High resolution 3D P wave velocity structure 338–343
beneath Tenerife Island (Canary Islands, Spain) based Nagaoka Y, Nishida K, Aoki Y, Takeo M, Ohminato
on tomographic inversion of active-source data. T (2012) Seismic imaging of magma chamber beneath
J Geophys Res 117, B09309, doi:10.1029/2011 an active volcano. Earth Planet Sci Lett 333–334:1–8.
JB008970 doi:10.1016/j.epsl.2012.03.034
Gouédard P, Stehly L, Brenguier F, Campillo M, Colin de Paige CC, Saunders MA (1982) LSQR: an algorithm for
Verdière Y, Larose E, Margerin L, Roux P, Sanchez- sparse linear equations and sparse least squares. ACM
Sesma FJ, Shapiro NM, Weaver RL (2008) Cross- Trans Math Soft 8:43–71
correlation of random fields: mathematical approach Patanè D, Barberi G, Cocina O, De Gori P, Chiarabba C
and applications. Geophys Prospect 56:375–393 (2006) Time-resolved seismic tomography detects
Husen S, Kissling E (2001) Local earthquake tomography magma intrusions at Mount Etna. Science 313(5788):
between rays and waves: fat ray tomography. Phys 821–823
Earth Planet In 123(2):127–147 Paulatto M, Minshull TA, Baptie B, Dean S, Hammond
Koulakov I (2009a) LOTOS code for local earthquake JOS, Henstock T, Voight B (2010) Upper crustal struc-
tomographic inversion. Benchmarks for testing tomo- ture of an active volcano from refraction/reflection
graphic algorithms. Bull Seismol Soc Am tomography, Montserrat, Lesser Antilles. Geophys
99(1):194–214. doi:10.1785/0120080013 J Int 180(2):685–696
Koulakov I (2009b) Out-of-network events can be of great Power JA, Villasenor A, Benz HM (1998) Seismic image
importance for improving results of local earthquake of the Mount Spurr magmatic system. Bull Volcanol
tomography. Bull Seismol Soc Am 99(4):2556–2563. 60(1):27–37
doi:10.1785/0120080365 Ritzwoller MH, Lin FC, Shen W (2011) Ambient noise
Koulakov I (2013) Studying deep sources of volcanism tomography with a large seismic array. Compte
using multiscale seismic tomography. J Volcanol Rendus Geosci. doi:10.1016/j.crte.2011.03.007
Geotherm Res 257:205–226. doi:10.1016/j. Sens-Schönfelder C, Wegler U (2006) Passive image
jvolgeores.2013.03.012 interferometry and seasonal variations of seismic
Koulakov I, Gordeev EI, Dobretsov NL, Vernikovsky VA, velocities at Merapi Volcano, Indonesia. Geophys S
Senyukov S, Jakovlev A (2011) Feeding volcanoes of Res Lett 33:L21302
the Klyuchevskoy group from the results of local Shapiro NM, Campillo M, Stehly L, Ritzwoller M (2005)
earthquake tomography. Geophys Res Lett 38: High resolution surface wave tomography from ambi-
L09305. doi:10.1029/2011GL046957 ent seismic noise. Science 307:1615–1618
Koulakov I, West M, Izbekov P (2013a) Fluid ascent Sharp ADL, Davis PM, Gray F (1980) A low velocity zone
during the 2004–2005 unrest at Mt. Spurr inferred beneath Mount Etna and magma storage. Nature
from seismic tomography. Geophys Res Lett. 287:587–591
doi:10.1002/grl.50674. Stankiewicz J, Ryberg T, Haberland C, Fauzi, Natawidjaja
Koulakov I, Gordeev EI, Dobretsov NL, Vernikovsky VA, D (2010) Lake Toba volcano magma chamber imaged
Senyukov S, Jakovlev A, Jaxybulatov K (2013b) by ambient seismic noise tomography. Geophys Res
Rapid changes in magma storage beneath the Lett 37:L17306. doi:10.1029/2010GL044211
Klyuchevskoy group of volcanoes inferred from Stauber DA, Green SM, Iyer HM (1988) Three-
time-dependent seismic tomography. J Volcanol Geo- dimensional P velocity structure of the crust below
thermal Res (available online). doi:10.1016/j. Newberry Volcano, Oregon. J Geophy Res Solid
jvolgeores.2012.10.014 Earth (1978–2012) 93(B9):10095–10107
3134 Seismic Vulnerability Assessment: Lifelines
Thurber CH (1983) Earthquake locations and three- They constitute by themselves a set of critical
dimensional crustal structure in the Coyote Lake facilities for the proper and safe functioning of
area, central California. J Geophys Res Solid Earth
(1978–2012) 88(B10):8226–8236 the societies. In case of a strong earthquake
Thurber CH (1984) Seismic detection of the summit motion, their physical damages and the conse-
magma complex of Kilauea volcano, Hawaii. Science quent disruption of the services they provide
223(4632):165–167 may contribute seriously to the global economic
Villasenor A, Benz HM, Filippi L, De Luca G, Scarpa R,
Patanè G, Vinciguerra S (1998) Three-dimensional loss. At the same time, their reparation cost may
P-wave velocity structure of Mt. Etna, Italy. Geophys be very high, reaching in several cases 10 % or
Res Lett 25(11):1975–1978 15 % of the initial construction cost of the whole
Zandomeneghi D, Barclay AH, Almendros J, Ibáñez JM, system to which they belong.
Wilcock WSD, Ben-Zvi T (2009) Crustal structure of
deception island volcano from p-wave seismic tomog- The term lifeline is ambiguous. Lifelines may
raphy: tectonic and volcanic implications. J Geophys be distinguished in two major categories:
Res 114:B06310 (i) Utility systems including potable water, natu-
Zollo AEA, Gasparini P, Virieux J, Le Meur H, De ral gas, oil, electric power distribution, wastewa-
Natale G, Biella G, Vilardo G (1996) Seismic evidence
for a low-velocity zone in the upper crust beneath ter, or communication systems and
Mount Vesuvius. Science 274(5287):592–594 (ii) transportation systems comprising roadways,
railways, airport, and port facilities. Sometimes
the terms infrastructures or critical facilities are
used instead, at least for some of them.
Seismic Vulnerability Assessment: Compared to buildings, lifeline systems pre-
Lifelines sent three distinctive features: (a) spatial variabil-
ity and topology and exposure to different
Kyriazis Pitilakis and Sotiris Argyroudis geological and geotechnical hazards, (b) wide
Department of Civil Engineering, Aristotle variety of component typologies and material
University, Thessaloniki, Greece used, and (c) specific functionality requirements,
which make them highly hierarchical networks.
For practical reasons they are generally catego-
Synonyms rized in links (e.g., pipelines, roads) and nodes
(e.g., tanks, power substations). Moreover, for
Fragility curves; Infrastructures; Seismic risk their vulnerability assessment, an important
assessment; Transportation networks (railway, parameter is the presence of synergies between
roadway, harbor, airports); Utilities (gas, oil, components within the same system (intra-
water, wastewater, energy supply) dependencies) or between different systems
(interdependencies). Due to these synergies, the
physical damage of a component of a system
Introduction interacting with another one may affect seriously
the second system’s performance and functional-
Main Features of Lifelines ity. Therefore, it is essential to define the taxon-
Lifelines refer to the complex system and net- omy of each interacting lifeline system that
work assets of connected components, usually describes the individual components of each sys-
interacting with other components and systems, tem and their role in the network as well as the
which are performing vital functions that are way other systems affect its performance.
essential to sustain the life and the growth of a Due to their spatial extent, lifeline systems are
community, such as producing, transporting, and exposed to variable seismic ground motions,
distributing goods or services. Their global value often presenting important incoherency (shaking
for the society and economy is permanently effects), and to geotechnical hazards expressed in
increased in our modern, technologically terms of permanent ground deformations,
advanced, highly demanding, and fragile world. resulted from fault crossing, landslides, and
Seismic Vulnerability Assessment: Lifelines 3135
liquefaction (e.g., lateral spreading, settlements, individual infrastructure and part of the network
buoyancy effects). Various parameters may be is variable and at the same time an essential factor
used to describe and characterize the strong in the seismic risk management. Finally, the syn-
ground motion and the induced phenomena. The ergies that are presented in a road network and
selection of appropriate parameter (i.e., intensity can affect its overall functionality in case of
measure), which is efficiently correlated with the earthquake include the supply of electric power
response and the expected damages of each for the signaling and lighting and the proximity to
exposed elements, depends on the technical and buildings and utility networks (e.g., water or gas
response characteristics of each element. pipes) that may induce the interruption of road
As an example, a motorway network presents traffic.
the following features: It includes multiple ele-
ments at risk like bridges, overpasses, inter- Experience from Past Earthquakes
changes, tunnels, culverts, embankments, cuts, Several devastating earthquakes occurred during
slopes, retaining walls, signage and markings, the last century, causing extensive losses in terms
electrical systems, rest areas, and buildings of fatalities, injuries, damages, and disruptions.
(e.g., tolls, maintenance, traffic management). Knowledge, experience, and lessons learned
These structures are usually built with different from past earthquakes have significantly contrib-
design methods, construction techniques, and uted to disaster risk reduction efforts across the
materials. The seismic design requirements, and globe.
therefore their vulnerability, depend on the type The 1971 San Fernando earthquake in USA
and the location of the structure, the seismicity, (ML6.6), an event that caused catastrophic dam-
and the tectonic features of the area. In particular, age to almost all types of lifelines, motivated the
roadways are crossing different geological and evolution of the lifeline earthquake engineering
topographical conditions such as valleys, rivers, as well as many new efforts to better understand
canals, gulfs, or mountainous regions. Obviously, the causes of these failures and identify ways to
the network components can be exposed to vari- mitigate future earthquake damage and disrup-
ous seismic ground motions in terms of ampli- tion (e.g., the establishment of the Technical
tude, frequency, or duration, as well as to Council of Lifeline Earthquake Engineering/
different geotechnical hazards (subsidence, liq- TCLEE, in ASCE). Other strong and damaging
uefaction, landslides, faults, etc.). In this context, earthquakes that contributed to the progress of
the design of a long bridge crossing an active lifeline earthquake engineering are the 1978
seismic fault is much more demanding than a Miyagiken-oki (Japan, Ms7.7), the 1985
single span bridge in the same area. On the Michoacan (Mexico, Mw8.3), the 1989 Loma
other hand, nowadays, the design level (and sub- Prieta (USA, Mw6.9), the 1990 Luzon (the Phil-
sequently the resistance to earthquakes) is higher ippines, Ms7.8), the 1993 Kushiro-Οki (Japan, S
than of 30 years ago, making old structures more Mw7.6), the 1993 Hokkaido–Nansei-Oki (Japan,
vulnerable, introducing a new factor, the aging or Mw7.7), the 1994 Northridge (USA, Mw6.7), the
time effects. Furthermore, the hierarchy 1995 Hyogoken–Nanbu (Kobe) (Japan, Mw6.8),
(importance) of roads depends on their traffic the 1999 Chi-Chi (Taiwan, Mw7.6), the 1999
capacity, i.e., highways and major arterials and Kocaeli (Turkey, Mw7.4), and the 2001 El Salva-
secondary and local urban roads. The importance dor (Mw7.6) events. More recent earthquakes
of the road segments, serving different part of the which caused serious damages and losses to util-
urban fabric, depends also on the land use and the ity systems and transportation infrastructures,
density of the population. The functional role of even in developed countries, include the 2004
roads is related to the specific served areas such Niigata ken Chuetsu (Mw6.8) and 2007 Niigata-
as industrial or residential areas, hospital facili- Chuetsu Oki (Mw6.6) earthquakes in Japan, the
ties, and transport hubs (e.g., airports and major 2008 Wenchuan (China, Mw7.9), the
harbors). Therefore, the importance of each 2009 L’Aquila (Italy, Mw6.3), the 2010 Maule
(Chile, Mw8.8), and recently in 2011 the Christ- water that flowed into gas pipelines. A large
church (New Zealand, Mw6.3) and the mega 2011 area was also burned down in Kobe due to the
Tohoku (Japan, Mw9.0) earthquake and the tsu- disruption and the inefficiency of the fire-fighting
nami associated with this earthquake. network. Considering that modern societies and
The experience from major earthquakes built environment are relying on the good perfor-
proved that most lifeline elements are vulnerable, mance of their interconnected infrastructures, it is
and the mid- and long-term effects, mainly in clear that an efficient rescue policy and an opti-
terms of financial losses, may be very important. mum recovery strategy after a strong earthquake
A typical example of long-term effects is the port require the evaluation of the interactions between
of Kobe in Japan, one of the largest container different lifeline systems.
cargo ports in the world, which suffered major In addition to ground shaking, ground failures
damages, mostly due to liquefaction-induced (i.e., landslides, liquefaction, later spreading,
phenomena, as a result of the 1995 Mw6.8 fault ruptures), if they occur, may produce even
Hyogoken–Nanbu earthquake. A year after more severe damages and losses to lifelines and
almost all port infrastructures have been infrastructures. As an example, the 2008 Mw7.9
reconstructed, but it is remarkable that in 1998, Wenchuan earthquake in China triggered more
3 years after the disaster, cargo traffic remained at than 15,000 landslides of various types within an
roughly half of pre-disaster levels. For the same area of 50,000 km2 and causing more than 20,000
earthquake, the direct repair cost for the Hanshin fatalities representing one-quarter of the total
Expressway was $4.6 billion (NCEER 1995), fatalities. The landslides produced extensive
while the total cost was actually much higher if damage to housing settlements, irrigation chan-
the $3.4 million of daily income from tolls and nels, highways, bridges, and other infrastruc-
the losses from business interruption and traffic tures. The city of Wenchuan and many other
delays was taken into account. In the 1994 Mw6.7 towns were isolated from the rest of the country,
Northridge earthquake, the repair cost for dam- and the rescue and relief efforts were greatly
aged bridges was evaluated at $190 million, affected due to blocking and damages in road
almost the 10 % of the total cost of the whole network (Tang et al. 2011). During the 2010
transportation system ($1.8 billion, Basöz and Mw7.1 followed by the 2011 Mw6.3 Christchurch
Kiremidjian 1998). earthquake sequences in New Zealand, the
Interactions between systems may have a sig- ground and slope failures, mainly due to lique-
nificant impact exceeding direct consequences. faction, produced extensive damages to the
Loss of power or communication can severely water, electricity, and road networks rendering
impact the emergency response, especially many of them inoperable (O’Rourke
when the disruptions concern critical facilities, et al. 2012). On the contrary the gas system
such as hospitals, command centers, etc. Loss of performed rather well. A large number of slope
power may have severe indirect effects due to the failures, landslides, and rock falls occurred dur-
synergies between the lifelines and the depen- ing the 2004 Mw6.8 Niigata ken Chuetsu earth-
dence of all networks on the power supply sys- quake on steep natural slopes and artificial cuts in
tem. In 1989 Mw6.9 Loma Prieta earthquake, the the area, resulting in major damage to road and
damages that occurred in the water network railway infrastructures (cuts, embankments,
greatly affected the fire-fighting capacity in the pavements, tunnels, abutments, retaining walls),
area of San Francisco, Oakland, and Berkeley. as well as to other lifeline components (water
Representative cases of interactions have been pipes, electric power poles). Similar damages
also reported after the 1994 Mw6.7 Northridge and disruptions, related to geotechnical failures,
and 1995 Mw6.8 Hyogoken–Nanbu earthquakes, were recorded after the subsequent 2007 Mw6.6
when the reparation of the water and gas net- Niigata-Chuetsu Oki earthquake in Japan and in
works have been delayed due to traffic conges- several recent events in Europe (e.g., 2003 Mw6.4
tion, street blocking, damaged buildings, and Lefkas earthquake in Greece, the 2009
Seismic Vulnerability
Assessment: Lifelines,
Fig. 1 Damages to road,
telecommunication, and
gas system in 1994
Northridge (Mw6.7)
earthquake (Photo Credit:
US Geological Survey)
Fig. 2 Damaged
equipment in Adapazari
substation in 1999 Kocaeli
(Mw7.4) earthquake (Photo
Credit: Kandilli
Observatory and
Earthquake Research
Institute/KOERI)
Mw6.3 L’Aquila earthquake in Italy, the 1999 The poor performance of elements at risk in a
Mw7.4 Kocaeli earthquake in Turkey) and world- network is likely to make the whole system much
wide. In Figs. 1, 2, 3, 4, 5, 6, 7, and 8, some more vulnerable and susceptible to strong
representative damages for utility and transporta- earthquakes.
tion networks are shown. Finally, the short- and long-term economic S
Several utility systems and networks are now impact due to earthquake damages of lifelines
outdated and lack proper seismic design and pro- can be significant. In a recent study by CATDAT
tection. For example, there are parts of the water (Daniell and Vervaeck 2013), the disaggregation
and wastewater systems in several cities built of direct economic losses of 61 selected major
more than a century ago. Hence, their material earthquakes from 1900 to 2012 showed that
properties and strength are seriously affected around 30 % of the direct losses came from infra-
from aging effects and are consequently more structure losses in transport, communications,
vulnerable and exposed to earthquake risk. Cur- pipelines, energy supply systems, etc.
rently, most of urban infrastructures are not
designed to sustain severe seismic hazard and Lifeline Earthquake Engineering Research
complex systemic threats, and it is difficult to In the last two decades, several important projects
anticipate what may be the complex socioeco- and research efforts related to the vulnerability
nomic losses due to potential strong earthquakes. and risk assessment of different lifelines at
Seismic Vulnerability Assessment: Lifelines,

Fig. 4 Damages to road embankments in 2007 Niigata-
Chuetsu Oki (Mw6.6) earthquake (Photo Credit: US Geo-
logical Survey)
Seismic Vulnerability Assessment: Lifelines,

methods, screening procedures and additional
Fig. 3 Lateral deformation of train tracks at the Arahama tools, retrofit technologies, design criteria, and
station due to surface waves in 2007 Niigata-Chuetsu Oki other approaches to reduce seismic vulnerability
(Mw6.6) earthquake (Photo Credit: US Geological of existing and future highway infrastructure.
Survey)
REDARS (Risks due to Earthquake DAmage to
Roadway Systems) is a public-domain software
regional or urban scale have been undertaken. package developed under this MCEER project
They resulted in the development of relevant that accounts for how earthquake damage affects
methods and guidelines. A short reference to the post-event traffic flows and travel times and esti-
most important contributions is following. mates losses from these travel-time and traffic-
flow impacts (http://mceer.buffalo.edu/research/
United States redars/).
In 1992, the Multidisciplinary Center for Earth- In 1996, the PEER Lifelines Program (http://
quake Engineering Research (MCEER) initiated peer.berkeley.edu/lifelines) was initiated, under
the Highway Project, funded by the Federal the coordination of the Pacific Earthquake Engi-
Highway Administration (FHWA). The project neering Research Center, funded by the Califor-
uniquely examined the impact of earthquakes on nia Department of Transportation and the Pacific
the highway system as an integrated network, Gas and Electric Company. The aim was to
rather than a collection of individual road seg- improve the seismic safety and reliability of life-
ments, bridges, embankments, tunnels, etc. The line systems with particular focus on the charac-
purpose of the project was to ensure the usability terization of ground motions, local soil and site
of highways following earthquakes, by improv- response, and the performance of bridge struc-
ing performance of all interconnected compo- tures and electric substation equipment.
nents. Overall goals were to deepen the In 1997, the Federal Emergency Management
understanding of the seismic hazard impact to Agency (FEMA) released the first edition of
highways and to improve and develop analysis HAZUS methodology and software that
Fig. 5 Collapse of a
section of the Nishinomiya
in 1995 (Mw6.8)
Hyogoken–Nanbu
earthquake (Photo by
K. Pitilakis)
Fig. 6 Damage to Trans-
European Motorway due to
fault rupture in 1999
Kocaeli (Mw7.4)
earthquake (Photo Credit:
Kandilli Observatory and
Earthquake Research
Institute/Photo by
D. Kalafat)
Fig. 7 Blocking of
Pan-American Highway
due to landslide in 2001 El
Salvador (Mw7.6)
earthquake (Photo Credit: S
US Geological Survey/
Photo by E.L. Harp)
estimates potential building and infrastructure It was released in 2004 (www.fema.gov/hazus).

losses from earthquakes, built using GIS technol- HAZUS-MH loss estimates reflect the state-of-
ogy. The current version is HAZUS-MH V2.0, the-art scientific and engineering knowledge and
including also losses from floods and hurricanes. can be used to inform decision makers at all
Seismic Vulnerability Assessment: Lifelines, Fig. 8 Failure of port infrastructures in 1995 (Mw6.8)
Hyogoken–Nanbu earthquake (Photo by K. Pitilakis)
levels of government by providing a reasonable stakeholders, owners/operators, and recognizant

basis for developing mitigation, emergency pre- regulatory bodies was established to facilitate use
paredness, response, and recovery plans and pol- of developed guidelines (http://www.americanlife-
icies. The main merit of the HAZUS platform is linesalliance.com/).
that it provided for the first time an unparalleled ERGO (the new name for the Multi-Hazard
set of fragility models for all the main compo- Assessment, Response, and Planning software
nents and systems of the built environment. How- previously known as mHARP and MAEviz)
ever, it must be recognized that many of these started somewhat later than HAZUS and was
models have been derived solely on expert judg- the product of the research efforts carried out at
ment and the level of knowledge at that time. One the Mid-America Earthquake Center in collabora-
effect of the sheer size of the HAZUS framework tion with the National Center for Supercomputing
and its set of tools is that it established itself very Applications (NCSA). It is an open-source soft-
soon as the reference for all studies in the sector ware and incorporates many of the design concepts
worldwide. and capabilities motivated by NCSA efforts to
In 1998, FEMA and the American Society of develop “cyberenvironments” that span scientific
Civil Engineers entered into a cooperative agree- disciplines and that can rapidly evolve to incorpo-
ment to establish the American Lifelines Alliance rate new research results (Elnashai et al. 2008).
(ALA). ALA’s objectives were to facilitate the An important aspect of ERGO is its extensibility,
creation, adoption, and implementation of design both in terms of analysis/features modules and
and retrofit guidelines and other national consen- of visualization/representation (GIS) modules.
sus documents to improve the performance of The framework has been designed to implement
utility and transportation systems (electric the consequence-based risk management (CRM)
power, telecommunication, water, wastewater, paradigm supported by the NCSA and MAE
oil, natural gas, rail, and shipping ports) in natural center.
hazard events, including earthquakes. ALA has
not attempted to focus efforts on all lifeline systems Europe
and components but has instead chosen to place Until now, no coordinated action for the devel-
priority on specific topics relevant to selected life- opment of a comprehensive methodology and
line systems where a need for improved hazard software tool at European level is made. How-
mitigation practices is identified. The primary ever, important efforts have been undertaken
function was to bridge the gap between hazard through research projects mainly funded by Euro-
mitigation practices for buildings and lifeline sys- pean Union. Brief description of the most impor-
tems; therefore a close collaboration with lifeline tant is following.
RISK-UE (2001–2004) project involved the vulnerability functions to assess the vulnerability
assessment of earthquake scenarios based on the of physical assets. The increasing impact due
analysis of the global impact of one or more to interdependencies and intra-dependencies
plausible earthquakes at city scale, within a Euro- between different components and among differ-
pean context. The primary aim of these scenarios ent interacting systems is treated in a comprehen-
was to increase awareness within the decision- sive way, providing specifications for each
making centers of a city. The developed method- network and infrastructure. The proposed socio-
ology for creating earthquake scenarios focused economic model integrates social vulnerability
on the distinctive features of European cities with into the physical systems modeling approaches
regard to current and historical buildings and providing to decision makers with a dynamic
lifelines, as well as on their functional and social platform to capture post disaster emergency
organization, in order to identify weak points issues like shelter demand and health impact
within the urban system. The approach was decisions (Pitilakis et al. 2014a, b).
applied to 7 European cities: Barcelona, Bitola,
Bucharest, Catania, Nice, Sofia, and Thessaloniki Other Efforts Worldwide
(Mouroux and Le Brun 2006). The Central American Probabilistic Risk Assess-
LESS-LOSS (2003–2007) was a European ment (CAPRA) platform was developed in
Integrated Project focusing on risk mitigation for partnership with Central American governments,
earthquakes and landslides that relied on the active the support of the Central American
participation of 46 partners from both academia Coordination Centre for Disaster Prevention
and industry. The project addressed research issues (CEPREDENAC), the Inter-American Develop-
on seismic engineering, earthquake risk and ment Bank (IDB), the International Strategy of
impact assessment, landslides monitoring, map- United Nations for Disaster Reduction
ping and management strategies, improved disas- (UN-ISDR), and the World Bank. CAPRA is an
ter preparedness and mitigation of geotechnical information platform to enhance decision-
hazards, development of advanced methods for making in risk management across various sec-
risk assessment, methods of appraising environ- tors, such as emergency management, territorial
mental quality, and relevant pre-normative planning, public investment, and the financial
research. Separate sections were devoted on earth- sectors. Through the application of probabilistic
quake disaster scenario predictions and loss risk assessment principles to threats like hurri-
modeling for urban areas and infrastructures with canes, earthquakes, volcanic activity, floods, tsu-
emphasis on water and gas systems (Faccioli namis, and landslides, CAPRA allows to measure
2007). and compare different types of risks and to
SYNER-G (2009–2013, www.syner-g.eu) is develop sector-specific applications for risk man-
the most recent European funded research project agement (Cardona et al. 2012; http://www. S
that developed an integrated, general methodol- ecapra.org/). It is not directly designed for lifeline
ogy and a comprehensive framework for the sys- vulnerability and risk assessment, but it can be
temic vulnerability and risk analysis of buildings, used for this purpose as well.
lifelines, and infrastructures. The proposed meth-
odology and tools encompass in an integrated Key Elements of Seismic Risk Assessment
fashion all aspects in the chain, from hazard to Risk assessment is the process used to determine
the vulnerability assessment of components and risk management priorities by evaluating and
systems and to the socioeconomic impacts of an comparing the level of risk to specific standards,
earthquake, accounting for most relevant uncer- either defined by a code or a set of target risk
tainties within an efficient quantitative simula- levels and criteria. The estimation of losses, both
tion scheme and modeling interactions between material and immaterial, is the ultimate target of
the multiple component systems. It systemati- the risk assessment. Then, the risk management
cally integrates the most advanced fragility or procedure should define pre-seismic (retrofitting
and strengthening actions), coseismic, and post- taxonomy/typology definitions of all elements at
seismic strategies and policies. The risk assess- risk and systems are the necessary initial steps to
ment of lifelines and infrastructures follows the describe the elements and networks exposed to
general scheme of seismic risk assessment: seismic risk; at the same time, the development of
detailed inventories is the most expensive task in
terms of financial cost and time. The estimation of
½seismic risk
¼½seismic hazard
½vulnerability
seismic hazard and site-specific seismic ground

½exposure elements at risk
motion characteristics provides appropriate seis-
mic intensity measures (IM), which will be used
The complexity of elements at risk, their variabil- later in the vulnerability and loss assessment
ity from one place and one country to another, modeling. Vulnerability is the propensity of dam-
and, till recently, the lack of well-validated damage of an element (building, bridge, pipeline,
age and loss data from strong earthquakes make roadway, oil tank, etc.) or a network of specific
the vulnerability assessment of each particular elements (e.g., water or gas system) to a given
component and of the network as a whole a chal- seismic intensity defined in the seismic hazard
lenging task. Adding to that the spatial extent of analysis. It is commonly assessed based on
lifelines, the synergies between different sys- fragility models, which estimate the probability
tems, and the several sources of uncertainties of damage for a given seismic intensity. Fragility
that are inherent in the various aspects (e.g., functions are constructed with respect to the
typology, damage states), models (e.g., seismic taxonomy and typological characteristics of
hazard, spatial correlation, fragility curves), and each element at risk. The expected damages and
finally tools and methods for estimation of losses, losses of the entire network are estimated through
it is obvious that the risk assessment of lifelines is a systemic analysis, which takes into account the
indeed a very complex and challenging issue. physical damages of the various parts and seg-
Therefore, multidisciplinary task and combined ments of the system, and in some cases, the intra-
efforts are really needed to reach reliable and dependencies between different components
comprehensive estimates. within the system and the interdependencies
This difficulty is amplified due to lack of well- between systems, in order to estimate the service-
documented data and the partial or poor knowl- ability or flow capacity of the damaged network.
edge of the geometric and other features of the The results are commonly provided in terms of
systems; in some cases it is also due to the confi- performance indicators (PI), which are describing
dentiality of data (i.e., communication or oil net- the network functional or the expected economic
works). The situation however has been or socioeconomic losses. Efficient risk manage-
improved the last decade. Important strong earth- ment is based on the evaluation of the resulting
quakes provided valuable good quality data, performance indicator values.
while the public awareness and the reported
huge direct and indirect losses associated to life- Seismic Hazard for Distributed Networks
line damages drew the attention of the scientific Lifelines are in most cases spatially distributed
community, the governmental authorities, and systems. Therefore, seismic hazard should meet
the insurance sector. Moreover, the development the specific features of each system considering
of geographical information systems (GIS) as also its spatial variability. The vulnerability anal-
well as remote sensing technologies offered an ysis and risk assessment must be evaluated
excellent platform for the implementation of effi- according to the precise typological characteris-
cient and innovative techniques to screen, cap- tics of the components and networks, taking also
ture, store, manipulate, analyze, manage, and into account the models used to describe vulner-
present geographical data. ability, usually in terms of fragility curves and
Few terms employed in this chapter are relationships. Moreover, due to the spatial extent
described in the following. Inventory and of utility and transportation networks, the seismic
hazard should describe the spatial variability of intended to relate the degree of deformation and
ground motion considering local soil conditions, the probability of the geotechnical hazard occur-
topographic effects, and geotechnical hazards; ring to the intensity of the ground motion
site-specific seismic hazard analyses are always (Weatherill et al. 2014). However, it should be
necessary. Traditional seismic hazard analysis, noted that most theoretical and empirical models
while effective in translating the hazard into a relating PGD to strong shaking require a level of
probabilistic formulation, is limited in the extent geotechnical detail that may be impractical to
to which it can incorporate spatial coherency of obtain for a spatially distributed set of sites.
the form needed for estimation of loss to spatially HAZUS methodology (FEMA 2003) provides a
distributed portfolios. The extension of seismic rather simple “baseline” model that can be
risk analysis to multiple systems of spatially dis- implemented in the widest variety of applications.
tributed infrastructures presents new challenges When the scenario type or probabilistic site-
in the characterization of the seismic hazard specific seismic hazard is available, the damage
input, particularly with respect to the spatial cor- assessment of a lifeline system is evaluated using
relation structure of the ground motion residuals appropriate fragility or vulnerability functions
(i.e., residuals from empirical intensity models) relating the probability of damages to seismic
that form the basis for the systemic risk analysis. intensity measures. The data and parameters
A general procedure entitled “Shakefield” has needed to perform this analysis are shortly
been recently established in the frame of described in the following.
SYNER-G project (Pitilakis et al. 2014b), which
allows for the generation of samples of ground Taxonomy, Inventory, and Typology
motion fields both for single-scenario-type events The key assumption of the vulnerability assess-
and for stochastically generated sets of events ment of lifelines is that the structures and com-
needed for probabilistic seismic risk analysis ponents of the systems, having similar structural
(Weatherill et al. 2014). For a spatially distrib- characteristics (e.g., a bridge of a given typol-
uted infrastructure of vulnerable elements, the ogy), being in similar geotechnical conditions
spatial correlation of the ground motion fields (e.g., a segment of a pipeline crossing the same
for different measures of the ground motion soil conditions and exposed to the same geotech-
intensity is incorporated into the simulation pro- nical hazards), are expected to perform in the
cedure. This is extended further to consider spa- same way for a given seismic excitation. Within
tial cross-correlation between different measures this context, expected damages are directly
of ground motion intensity. related to the structural properties of the elements
The consideration of hazard from permanent at risk. Taxonomy and typology are thus funda-
ground deformation (PGD) is essential in model- mental descriptors of a system that are derived
ing the seismic risk to lifeline systems. For pipe- from the inventory of each element and system. S
lines and similar systems with linear elements, Geometry, material properties, morphological
fragility models are generally given in terms of features, age, seismic design level, anchorage of
PGD, as they are mostly vulnerable to permanent the equipment, soil conditions, and foundation
displacement of the ground rather than transient details are among usual typology descriptors/
shaking. Four primary causes of permanent parameters. Buildings, bridges, pipelines (gas,
ground displacements are commonly considered: fuel, water, wastewater), tunnels, road embank-
liquefaction-induced lateral spread, liquefaction- ments, power substations, harbor facilities, and
induced settlement, slope displacement, and road and railway networks have their own spe-
coseismic fault rupture. In addition to strong cific set of typologies and different taxonomy.
shaking, another transient effect that poses a The taxonomy of any lifeline network is thus
potential risk to lifeline systems is the transient an essential step for identifying, characterizing,
ground strain (PGS). Several models are avail- and classifying all types of lifeline elements
able for the estimation of PGD; some of them are according to their specific typology and their
distinctive geometric, structural, and functional comprehensive taxonomy presently available in

features. It allows the classification of the net- Europe is produced in SYNER-G project
work elements in an ordered classification system (Table 1, Pitilakis et al. 2014a). Detailed classi-
adequate for the seismic risk assessment of each fications are also provided in HAZUS (FEMA
component and the network (or system) as a 2003) and ALA (2001).
whole. The taxonomy is specific for each lifeline Each component is further classified
system and it needs a detailed typology definition according to its specific features regarding its
of all elements at risk. For that, an adequate seismic behavior. For instance, the main typo-
inventory process is essential. However, several logical features of gas pipelines include the
difficulties arise in the collection and archiving of material type, material strength, diameter, wall
the data related to the absence of well-organized thickness and smoothness of coating, type of
archives in public and often in private organiza- connection and joints, as well as the nominal
tions managing the systems, the oldness of the design and actual flow. Among them, material
networks (e.g., water network), and the high cost and the connection types (joints) together with
to draw up complete inventories. To this respect the soil type are the most critical parameters for
remote sensing techniques and GIS offer a useful their seismic response. Table 2 outlines common
and indispensable instrument and platform to types of pipelines based on the above classifica-
enhance with relatively low cost the collection tion criteria.
of the data and implement any inventory inquires. Another example is storage tanks, which are
Different types of satellite data can be used to usually categorized according to their material
extract information and parameters needed to type (steel or reinforced concrete), construction
compile or update inventories for seismic risk type (at grade or elevated), anchorage (anchored
assessment. Examples include the width of or unanchored), roof type and capacity, shape
roads, the geometry or material of buildings, factor (height-on-diameter ratio), and amount of
building aggregates, infrastructures like bridges content in the tank (full, half full, empty). Among
and channels, land use characteristics, and others. them, material type, construction type, and
The inventory of a specific structure in a anchorage are considered as the most important.
region and the capability to create classes of Table 3 presents the classification of tanks in
structural types (e.g., with respect to material, USA according to ALA (2001).
geometry, and design code level) are among the For a systematic and detailed description of all
main challenges when carrying out a general classes and subclasses of all elements at risk in
seismic risk assessment, for example, at a city most lifeline systems and networks in the Euro-
or region scale, where it is practically impossible pean context, the reader should refer to Pitilakis
to perform this assessment separately for each et al. (2014a).
single structure. It is necessary to classify all
elements at risk, in “as much as possible” homo- Damage Assessment
geneous classes and subclasses presenting more- The damage patterns and mechanisms are differ-
or-less similar response characteristics to ent for each lifeline system and element
ground shaking. Thus, the derivation of appro- (component), and they are strongly depending
priate fragility curves for any type of structure on the typology of each structure (i.e., materials,
depends entirely on the creation of a reasonable geometry, structural types). For example, the
taxonomy that is able to classify the different damage mechanisms of pipelines are generally
kinds of structures and infrastructures in any classified in the form of breaks or leaks; material
system exposed to seismic hazard. Several type (i.e., brittle or ductile) and joints (i.e., flex-
uncertainties are inherent to each taxonomy sys- ible or rigid) are also factors affecting the seismic
tem, which are inevitably propagated through response and damages of pipelines. In case of
the generic fragility (vulnerability) curves to tunnels, damage patterns include lining cracks
the final loss estimation. The most coherent and (longitudinal or transverse) and spalling, wall
Seismic Vulnerability Assessment: Lifelines, Table 1 SYNER-G infrastructure taxonomy (Pitilakis et al. 2014a)
System Component (and subcomponents)
BDG: buildings Force-resisting mechanism (FRM1): moment-resisting frame structural wall, flat slab,
bearing walls, precast, confined masonry
FRM material (FRMM1): concrete, masonry
Plan (P): regular, irregular
Elevation (E): regular/irregular geometry
Cladding (C): regular infill vertically, irregular infill vertically, bare
Detailing (D): ductile, non-ductile, with tie rods/beams, without tie rods/beams
Floor system (FS): rigid, flexible
Roof system (RS): peaked, flat, gable end walls
Height level (HL): Low rise, mid-rise, high rise, tall
Code level (CL): none, low, moderate, high
EPN: electric power EPN01: electric power grid
network EPN02: generation plant
EPN03: substation
EPN04: distribution circuits
EPN05–09: substation macro-components (autotransformer line; line without transformer;
bar-connecting line; bars; cluster)
EPN10–23: substation micro-components (circuit breaker; lightning arrester or discharger;
horizontal disconnect switch or horizontal sectionalizing switch; vertical disconnect switch
or vertical sectionalizing switch; transformer or autotransformer; current transformer;
voltage transformer; box or control house; power supply to protection system; coil support;
bar support or pothead; regulator; bus; capacitor tank)
EPN24: transmission or distribution line
GAS: natural gas GAS01: production and gathering facility (onshore, offshore)
system GAS02: treatment plant
GAS03: storage tank
GAS04: station (compression; metering/pressure reduction; regulator; metering)
GAS05: pipeline
GAS06: SCADA
OIL: oil system OIL01: production and gathering facility (onshore, offshore)
OIL02: refinery
OIL03: storage tank farm
OIL04: pumping plant
OIL05: pipeline
OIL06: SCADA
WSS: water-supply WSS01: source (springs, rivers, natural lakes, impounding reservoirs, shallow or deep S
network wells)
WSS02: treatment plant
WSS03: pumping station
WSS04: storage tank
WSS05: pipe
WSS06: tunnel
WSS07: canal
WSS08: SCADA system
WWN: wastewater WWN01: wastewater treatment plant
network WWN02: pumping (lift) station
WWN03: pipe
WWN04: tunnel
WWN05: SCADA system
(continued)
Seismic Vulnerability Assessment: Lifelines, Table 1 (continued)

System Component (and subcomponents)
RDN: road network RDN01: bridge (material, type of deck, deck structural system, pier to deck connection,
type of pier to deck connection; type of section of the pier, spans, type of connection to the
abutments, skew, bridge configuration, foundation type, seismic design level)
RDN02: tunnel
RDN03: embankment (road on)
RDN04: trench (road in)
RDN05: unstable slope (road on or running along)
RDN06: road pavement (ground failure)
RDN07: bridge abutment
RWN: railway RWN01: bridge
network RWN02: tunnel
RWN03: embankment (track on)
RWN04: trench (track in a)
RWN05: unstable slope (track on or running along)
RWN06: track
RDN07: bridge abutment
RWN08: station
HBR: harbor HBR01: waterfront components (gravity retaining structures; sheet pile wharves; piers;
breakwaters mooring and breasting dolphins)
HBR02: earthen embankments (hydraulic fills and native soil material)
HBR03: cargo handling and storage components (cranes, tanks, etc.)
HBR04: buildings (sheds, warehouse, offices, etc.)
HBR05: liquid fuel system (as per the OIL system)
FFS: fire-fighting FFS01: fire-fighter station
system FFS02: pumping station
FFS03: storage tank
FFS04: fire hydrant
FFS05: pipe
deformation, bending and buckling of reinforcing Therefore, classification of damage and the
bars, obstruction of the opening, or pavement subsequent definition of specific damage states
cracks. In subway stations damages to columns are important in the vulnerability assessment as
may also occur. Damages to quay walls are the seismic intensity is correlated to the expected
related to excessive lateral pressures or decrease damage level through the fragility or vulnerabil-
of shear strength of the foundation soil that can ity functions. Again, the form of the fragility
cause sliding, deformation, and tilting of the functions depends on the typology of the element
walls. Settlements and lateral movement of the at risk. For common structures (e.g., buildings,
backfill materials and cracking of apron pave- bridges) and other not extended elements (e.g.,
ments can be also induced. Damage to electric cranes, tanks, substations), the fragility curves
power substations includes failures of the various describe the response and damage level of partic-
subcomponents (e.g., disconnect switches, circuit ular subcomponents (e.g., columns, transformers)
breakers, transformers) or damage of the build- or of the entire structure. For linear elements of
ing. Earth structures such as highway and railway extended networks such as gas pipelines, the fra-
embankments can spread laterally and settle, gility functions describe the number of expected
resulting in opening of cracks in the road pave- damages along a certain length (i.e., per km).
ment or displacement of the railway tracks. The Examples and further details are given in the
list of possible damage patterns is unlimited. next sections.
Seismic Vulnerability Assessment: Lifelines, Seismic Vulnerability Assessment: Lifelines,

Table 2 Common types of materials and connections Table 3 Typology of tanks (ALA 2001)
for buried pipelines
Unanchored redwood tank (5 104–5 105 gal)
Connection Unanchored post-tensioned circular concrete tank
Material type type (>1 106 gal)
Asbestos cement (AC) Arc welded Unanchored steel tank with integral shell roof
Cast iron (CI) Bell and spigot (1 105–2 106 gal)
Ductile iron (DI) Cemented Unanchored steel tank with wood roof (1 105–2
Concrete (C) Riveted 106 gal)
Polyvinyl chloride (PVC) Rubber gasket Anchored steel tank with integral steel roof
Welded steel (WS) Gas welded (1 105–2 106 gal)
Medium-density polyethylene Unanchored steel tank with integral steel roof
(MDPE) (>2 106 gal)
High-density polyethylene (HDPE) Anchored steel tank with wood roof (>2 106 gal)
Anchored reinforced (or prestressed) concrete tank
(5 104–1 106 gal)
Elevated steel tank with no seismic design
Damage States Elevated steel tank with nominal seismic design
In seismic risk assessment, the performance Open-cut reservoir
levels of a structure, for example, a reinforced Fiberglass tanks
concrete building, belonging to a specific class
(Pitilakis et al. 2014a) can be defined through
damage thresholds called limit states. A limit corresponding functionality (e.g., power avail-
state defines a boundary between two different ability for electric power substations, number of
damage conditions often referred to as damage available traffic lanes for roads, flow or pressure
states. The thresholds are related to functionality level for water system). These correlations pro-
and serviceability levels while they are usually vide quantitative measures of the component’s
defined based on engineering judgment and com- performance and can be applied for the definition
mon sense. They are also strongly depending on of specific performance indicators (PIs), which
the model applied for the analysis and derivation are introduced in the systemic analysis of each
of the fragility functions. Uncertainties related to network. Therefore, the comparison of a demand
this stage of the analysis are normally referred as with a capacity quantity, the consequence of a
epistemic uncertainties (see Pinto 2014). Differ- mitigation action, or the accumulated conse-
ent damage criteria have been proposed quences of all damages (usually referred as the
depending on the typologies of elements at risk “impact”) can be evaluated.
and the approach used for the derivation of fra- Methods for deriving fragility curves gener-
gility curves. The most common way to define ally describe damages on a discrete damage scale. S
earthquake consequences is a classification in In the empirical procedures, the scale is used in
terms of the following damage states: no damage, survey efforts to produce post-earthquake dam-
slight/minor, moderate, extensive, and complete. age statistics, and sometimes it is rather subjec-
This qualitative approach requires an agreement tive (i.e., there is often a discrepancy between the
on the meaning and the content of each damage damage levels that any two different inspectors
state. The number of damage states is variable would assign for the same incident). In analytical
and is related with the functionality of the com- procedures the scale is related to limit state of
ponents and/or the repair duration and cost. In selected mechanical properties that are described
this way the total losses of the system (economic by appropriate indices, such as the displacement
and functional) can be estimated. In particular, capacity or the storey drift in the case of buildings
physical damages are related to the expected ser- or simple drift in pier bridges. For other elements
viceability level of the component (i.e., fully or at risk, the definition of the performance levels or
partial operational or inoperative) and the the limit states may be more vague and follows
other criteria related, for example, in the case of parameters of the ground motion have been also
pipelines, to the limit strength characteristics of used for different structures composing a lifeline
the material used in each typology. The definition system. More recently, it is proposed to use two
and consequently the selection of the damage descriptors instead of one, and hence the fragility
thresholds, i.e., limit states, are among the main, curve is transformed in fragility surfaces (Seyedi
yet, unavoidable sources of uncertainties. et al. 2010; Douglas et al. 2014).
The selection of the adequate intensity parame-
Intensity Measures ter is also related to the approach that is followed
An important issue related to the fragility curve for the derivation of fragility curves and the typol-
construction and implicitly to the risk assessment ogy of elements at risk. The identification of the
is the selection of an appropriate earthquake proper IM is determined from different constraints,
intensity measure (IM) that characterizes the which are first of all related to the adopted hazard
strong ground motion that best correlates with model, but also to the element at risk under consid-
the response of each element, for example, build- eration and the availability of data and fragility
ing, pipeline, or harbor facilities like cranes. Sev- functions for all different exposed assets. Empirical
eral measures of the intensity of ground motion fragility functions are usually expressed in terms of
(IMs) have been developed. Each intensity mea- the macroseismic intensity defined according to
sure may describe different characteristics of the different macroseismic scales (EMS, MCS, and
motion, some of which may be more adverse for MM). Analytical or hybrid fragility functions are,
the structure or the system under consideration. on the contrary, related to instrumental IMs, which
The use of a particular IM in seismic risk analysis are related to parameters of the ground motion
should be guided by the extent to which the (PGA, PGV, PGD) or of the structural response
measure corresponds to damage to the compo- (spectral acceleration Sa or spectral displacement
nents of a system. Optimum intensity measures Sd, for a given value of the period of vibration T).
are defined in terms of practicality, effectiveness, When the vulnerability of elements due to ground
efficiency, sufficiency, robustness, and comput- failure is examined (i.e., liquefaction, fault rupture,
ability (Mackie and Stojadinovic 2003). landslides), permanent ground deformation (PGD)
In general, IMs are grouped in two general is the most appropriate IM.
classes: empirical intensity measures and instru-
mental intensity measures. With regard to the Vulnerability Assessment and Fragility
empirical IMs, different macroseismic intensity Curves
scales could be used to identify the observed The fundamental tool in seismic risk assessment
effects of ground shaking over a limited area. of lifeline components is the fragility curves
Instrumental IMs are, by far, more accurate and which describe the probability that a structure
representative of the seismic intensity character- will reach or exceed a certain damage state for a
istics and the severity of ground shaking. For given ground motion intensity. An extensive
example, for bridges the best descriptor is a spec- review of available fragility functions and state-
tral response value at a specific period (i.e., T = of-the-art methods for vulnerability assessment
1.0 s). For other lifeline components, it may be of buildings and lifeline components can be
the peak ground acceleration (e.g., buildings, found in Pitilakis et al. (2014a). Fragility curves
tanks, electric power substations), peak ground are usually represented by two-parameter
velocity (e.g., pipelines), or even the permanent (median and log-standard deviation) cumulative
ground deformations (e.g., pipes, embankments, lognormal distributions. Several approaches are
roadways, railways). The correlation between used to establish fragility functions. They are
damages of specific elements at risk and intensity grouped in the following five categories:
measures is not simple and never unique. Several
other descriptors like peak ground strain, Arias – Empirical fragility curves, based on post-
intensity, cumulative absolute velocity, and other earthquake surveys and observations of actual
Fig. 9 Empirical fragility
curves for power grids
made up of substations of
different voltages based on
data from US west coast
earthquakes (Dueñas-
Osorio et al. 2007)
damage. They are specific to particular sites The main reason is the considerable improve-
and seismotectonic, geological, and geotech- ment of computational tools, methods, and
nical conditions, as well as the properties of skills, which allows comprehensive paramet-
the damaged structures. Consequently, the use ric studies covering most common typologies
of these functions in different regions is to be undertaken. Moreover, several of the
always questionable (Figs. 9 and 10). associated uncertainties, e.g., material proper-
ties, are better controlled.
In case of pipelines, the empirical fragility – Judgmental or expert elicitation fragility
functions relate the repair rates (RR) expressed curves, using questionnaires by which the
as repairs/km with the peak ground velocity experts are queried on the probability of a com-
(PGV) or permanent ground deformation (PGD) ponent being in a certain damage state for a
(Fig. 11). The curves may be further adapted to given intensity (Fig. 14). They are versatile
the material properties and geometry of the pipe- and relatively fast to establish, but their reliabil-
lines and soil conditions. In the last two decades, ity is questionable because of their dependence
the increased density of high-quality strong on the experiences of the experts consulted.
ground motion records in different soil condi- – Hybrid fragility curves, which combine any of
tions, in combination with new technologies the above-mentioned techniques in order to
such as geographical information systems (GIS) compensate for their respective drawbacks. S
and remote sensing technologies (e.g., LiDAR) – Fragility curves based on a fault-tree analysis,
capable to measure more accurately ground where complex components (e.g., substations,
movements, contributed significantly to the pumping plants, hospitals) are broken down
development and verification of such relation- into subcomponents and the global fragility
ships (O’Rourke et al. 2012). is obtained based on the relationships between
the subcomponents and their individual fragil-
– Analytical fragility curves, based on numeri- ities (Fig. 15).
cal simulations of structural models under
increasing earthquake loads (Figs. 12 and An example of damage estimation for “on-
13). Analytical methods, validated with ground unanchored steel tanks” due to ground
large-scale experimental data and observa- shaking and “pipelines” due to ground failure is
tions from recent strong earthquakes, have given in Table 4. In the first case, the exceedance
become more popular in recent years. probabilities of each damage state are estimated
curves for on-ground steel
tanks subjected to ground
shaking (ALA 2001)
based on the fragility curves in Fig. 10, and insufficient data availability and the need to
then the punctual probabilities of each damage account for their spatial and temporal aspects of
state are obtained. In the second case, the repairs complex supply–demand operation (Satumtira
per kilometer are estimated for a given value and Duenas-Osorio 2010). The various
of PGD based on the curves in Fig. 11. The approaches available in the literature depend on
estimated repairs should be modified according the simulation method, modeling objectives,
to the length of the pipe segment under study. scale of analysis, availability of input data, and
end-user type or needs. A comprehensive catego-
Systemic Analysis rization scheme describing different levels is pro-
The majority of the methodologies that have been vided in Modaressi et al. (2014):
developed worldwide for the seismic risk assess-
ment of lifelines and infrastructures refer to sin- – Vulnerability analysis: This level considers
gle systems, without considering interactions, only the potential physical damages of the
cascading failures, and complex impacts. The components of the systems, with no consider-
study of interdependent infrastructures is chal- ation of functionality of either the elements or
lenging due to heterogeneous quality and the whole system.
functions for common
pipeline typologies
provided by ALA (2001):
(a) pipelines subjected to
wave propagation (PGV),
(b) pipelines subjected to
permanent ground
deformation (PGD)
– Connectivity analysis: This level describes the estimate capacitive flows from sources (e.g.,
probability of the demand nodes to be generators, reservoirs) to sinks (i.e., distribu-
connected to functioning supply nodes tion nodes), based on the damages sustained S
through undamaged paths. In this approach by the network components (from total
the damaged components are removed from destruction to slight damages reducing the
the network and the adjacency matrix is capacity).
updated accordingly, thus pointing out the – Fault-tree analysis: This level of analysis con-
nodes or areas that are disconnected from the cerns critical infrastructures, where multiple
rest of the system. This qualitative approach is conditions are necessary for the systems to
used for all utility networks (water, electricity, ensure its task. This type of approach aims to
gas) and the road transportation system. evaluate the remaining operating capacity
– Capacitive analysis: This level describes the (residual operation capacity) of objects such
ability of the system to provide to the users the as health-care facilities. The system is broken
required functionality through a quantitative down into structural, nonstructural, or human
approach. For utility networks, graph algo- components, each one of them being
rithms and flow equations can be used to connected with logic operators.
Fig. 12 Analytical
fragility curves for alluvial
bored tunnels due to ground
shaking, classified to
ground type B, C, and
D according to Eurocode
8 (Argyroudis and Pitilakis
2012)
Fig. 13 Analytical
fragility curves for gravity
waterfront structures due to
ground shaking, classified
according to the wall height
(H) and the soil foundation
conditions (Vs values)
(Kakderi and Pitilakis
2010)
Fig. 14 Expert judgment S
fragility curves for railway
tracks subjected to
permanent ground
deformation (Argyroudis
and Kaynia 2014)
Fig. 15 Fault-tree based
fragility curves for water
treatment plant with
anchored components
subjected to ground
shaking (Pitilakis
et al. 2014a)
Seismic Vulnerability Assessment: Lifelines, Table 4 Example of damage estimation for steel tanks and pipelines
On-ground unanchored steel tanks subjected to ground shaking (PGA = 0.4 g)
Damage state No damage Minor Moderate Extensive Complete
Probability of exceedance – 0.919 0.571 0.240 0.108
Probability of occurrence 0.081 0.348 0.331 0.132 0.108
(=1–0.919) (=0.919–0.571) (=0.571–0.240) (=0.240–0.108)
Pipelines subjected to ground failure (PGD = 40 cm)
Material diameter Welded steel – large Welded steel – small PVC – small
Repairs/km 1.26 5.86 6.70
The performance of each network (e.g., utility curves for specific components that comprise a
or accessibility losses) is commonly measured set of subcomponents (e.g., health-care facilities,
through appropriate performance indicators water treatment plants).
(PIs), which, if combined with direct losses The importance of the interconnection
from physical damages, can yield a first partial between different systems is a more recent acqui-
estimate of the overall socioeconomic impact of sition that targets two or, rarely, more systems
an earthquake. Performance indicators, at the (Pitilakis et al. 2014b). Several classifications
component or the system level, depend on the have been proposed to categorize the types of
type of analysis that is performed. Connectivity interactions. The most common are physical,
analysis gives access to indices such as the con- demand, and geographic interactions (Rinaldi
nectivity loss (measure of the reduction of the et al. 2001). Physical interaction describes phys-
number of possible paths from sources to sinks). ical reliance on material flow from one infrastruc-
Capacitive modeling yields more elaborate per- tural system to another as, for example, the
formance indicators at the distribution nodes supply of power to various network facilities by
(e.g., head ratio for water system, voltage ratio electric generators. Demand interactions corre-
for electric buses) or for the whole system (e.g., spond to a supply–demand from a given compo-
system serviceability index comparing the cus- nent to another system. An example is the number
tomer demand satisfaction before and after the of casualties that should be treated by health-care
seismic event). The fault-tree analysis method is facilities after an earthquake. Finally, geographic
generally used for the derivation of fragility interactions describe the way that a local
environmental event affects components across The distribution of performance losses is

multiple infrastructural systems due to physical shown in Fig. 16c as the mean annual frequency
proximity. For instance, the collapse of buildings (MAF) of exceedance for the system serviceabil-
in city centers can induce the blockage of adja- ity index (SSI). SSI provides a global scalar mea-
cent roads due the debris accumulation. sure of the system performance. It is defined as
A comprehensive methodological framework the ratio of the sum of the real power delivered
for the assessment of physical as well as socio- from load buses after an earthquake to that before
economic seismic vulnerability and risk of build- the earthquake. For each bus inside the substa-
ings, lifelines, and infrastructures at urban and tions, voltage ratio (VR) is defined as the ratio of
regional level considering inter-element and the voltage magnitude in the seismically dam-
intra-systems dependencies has been developed aged network to the reference value for
in SYNER-G project. The reader is referred to non-seismic, normal conditions. Figure 16b dis-
Pitilakis et al. (2014b) for more details and appli- plays a contour map of the expected values of
cations of the proposed framework. VR, averaged on the whole simulation for each
demand node. It can be seen that the reduction in
Seismic Risk Assessment of Lifelines: voltage due to induced damage is less than the
Examples tolerated threshold of 10 %, allowing the power
Representative examples of seismic risk assess- demand delivery everywhere in the island.
ment studies for lifelines are given in the follow-
ing, for different scales of analysis. Water Network of Thessaloniki, Greece (City Level)
A connectivity analysis is performed for the main
Medium- to High-Voltage Electric Power Network water system (WSS) of Thessaloniki in Greece
of the Sicily Region, Southern Italy (Regional considering the interaction of electric power net-
Scale) work (EPN) with pumping stations (Pitilakis
A power flow analysis is performed for the elec- et al. 2014b). If a pump serving a source node is
tric power network or Sicily, which is composed not fed by the reference EPN node due to damage
of 181 nodes and 220 transmission lines in EPN substation, then the pump itself is consid-
(Cavalieri et al. 2014). The nodes, i.e., the ered out of service and the relative WSS node is
buses, are subdivided into 175 demand or load removed from the system. A probabilistic evalu-
nodes and six supply nodes, five of which are ation of the performance of networks is carried
power plants and one is the balance node out by means of Monte Carlo simulation by sam-
(or slack bus), which is coinciding with the gen- pling seismic events for five seismic zones with
eration node providing the highest power. The Mmin = 5.5 and Mmax = 7.5. Pipeline damage is
load nodes (two for transmission/distribution evaluated for each simulation considering both
and one for distribution substations) deliver wave propagation and ground failure due to liq- S
power to users. In total, 390 municipalities are uefaction. The network is analyzed for each sam-
served by the network. All transmission lines are pled event and the results are aggregated all over
overhead lines and considered as non-vulnerable the sampled events, in order to numerically
elements. They are classified into high- (HV), obtain the marginal distribution of performance
medium- (MV), and low-voltage (LV) lines losses (Fig. 17). The interaction can be important;
(Fig. 16a). The vulnerable elements are the com- as an example the water connectivity loss is
ponents within substations, called micro- increased from 1 to 1.8 % for l = 0.001
components. A probabilistic evaluation of the (corresponding to mean return period T = 1,000
performance of network is carried out by means years) when the connections of water pumping
of Monte Carlo simulation by sampling seismic stations to EPN are included in the analysis.
events for 18 faults taken from the Italian DISS Figure 18 shows the level of correlation
database employing the truncated Gutenberg and between the water connectivity loss and damages
Richter recurrence model for the source activity. in pipes as well as the nonfunctional EPN
Fig. 16 Seismic risk
assessment of electric
power network in Sicily,
Italy (Source: Cavalieri
et al. 2013). (a)
Transmission lines,
classified by voltage. (b)
Contour map of expected
values of voltage ratio
(VR). (c) Mean annual
frequency (l) curve for
electric power system
serviceability index (SSI)
Fig. 17 Mean annual
water connectivity loss
with and without
interaction with electric
power network (EPN)
Seismic Vulnerability Assessment: Lifelines, Fig. 18 Correlation of damaged pipes and nonfunctional EPN
transmission stations to water network connectivity
Fig. 19 Mean annual
road network connectivity
loss with and without
interaction with collapsed
buildings
substations supplying the water pumping sta- Road closures are estimated due to soil liquefac-
tions. The most correlated pipes are concentrated tion and bridge damages. Moreover, the interac-
along the coastline where liquefaction suscepti- tion with collapsed buildings that can induce road
bility is high and therefore damages due to per- blockages is considered (Pitilakis et al. 2014b).
manent ground displacement are expected. A probabilistic evaluation of the network’s per-
A higher level of correlation is obtained for the formance is carried similarly to the one described
EPN transmission substations. The highest value in the previous example.
of 80 % is attributed to a component in the south- The interaction with building collapses can be
east part of the city, where several pumping sta- important especially for return periods higher
tions (connected to EPN) are located. than 500 years (l = 0.002). As an example the
connectivity loss is increased from 20 % to 33 %
Road Network of Thessaloniki, Greece (City Level) for l = 0.001 (T = 1,000 years) when the building
The main network of the urban area is considered collapses are included in the analysis (Fig. 19).
in this case study, together with the ring road and Figures 20 and 21 show the level of correlation
the main exits of the city where the majority of between the connectivity loss and the distribution
bridges and overpasses are located. In particular, of blockages due to building collapses and dam-
594 nodes and 674 edges are included in the age in bridges and road pavements, respectively.
simulation. The nodes are subdivided into Relatively higher correlation factors are found for
15 external nodes, 127 traffic analysis zone edges blocked by building collapse, demonstrat-
(TAZ) centroids, and 452 simple intersections. ing the importance of this failure mechanism in
Edges are assumed to be the only vulnerable the analysis. A few road segments near the coast-
components in the network. They are classified line which are subjected to ground failure due to
into road pavements and bridges, with fragility liquefaction are also highly correlated to the net-
models expressed in terms of permanent ground work connectivity. The high risk of failure for
deformation (PGD) due to liquefaction and bridges is attributed to their typology character-
peak ground acceleration (PGA) for ground istics (old, simple span bridges) and the high
shaking, respectively (Pitilakis et al. 2014b). values of PGA.
Seismic Vulnerability Assessment: Lifelines, Fig. 20 Correlation of edges blocked by buildings’ collapse to road
network connectivity
Port System of Thessaloniki, Greece (Infrastructure Mmax = 7.5. The performance of the port is
Level) described through the total cargo or containers
The port covers an area of 1,550,000 m2 and handled in a predefined time frame per terminal
trades approximately 16,000,000 t of cargo annu- and for the whole port system (Pitilakis
ally, having a capacity of 370,000 containers and et al. 2014b).
six piers with 6,500 m length. In this case study, Figures 22 and 23 show the level of correlation S
waterfront structures, cargo handling equipment, between the total cargo handled per day (TCaH)
power supply system, roadway system, and and the distribution of damages in cranes and
buildings are examined. In particular, waterfront non-functionality of electric power distribution
structures of a total 6.5 km length, 48 crane substations, respectively. In this way the most
nodes, and two terminals (one container and one critical components can be defined in relation
bulk cargo) are considered. The interactions with their contribution to the performance loss
accounted for in the analysis are the supply of of the system. All cranes have medium (40–70 %)
EPN to cranes and the road closures due to build- to high (over 70 %) levels of correlation, indicat-
ing collapses. A probabilistic evaluation of the ing their great importance to the functionality of
performance of networks is carried out by means the overall port system. A higher level of corre-
of Monte Carlo simulation by sampling seismic lation is estimated for the EPN distribution sub-
events for five seismic zones affecting the city of stations, with 40 % of the components having
Thessaloniki and the harbor with Mmin = 5.5 and values greater than 70 %.
Seismic Vulnerability Assessment: Lifelines, Fig. 21 Correlation of broken edges (bridges due to ground shaking or
road segments due to liquefaction) to road network connectivity
Seismic Vulnerability Assessment: Lifelines, Fig. 22 Correlation of damaged cranes to port performance (PI =
TCaH)
Seismic Vulnerability Assessment: Lifelines, Fig. 23 Correlation of nonfunctional electric power distribution
substations to port performance (PI = TCaH)
NETWORK INVENTORY SEISMICHAZARD

/ TAXONOMY Probabilistic / Scenarios
TYPOLOGY Ground shaking Geotechnical hazards

of components
- Site effects - Landslides
- Spatial correlation - Liquefaction
- Cross correlation - Fault rupture
Damage FRAGILITY
States FUNCTIONS Intensity Measures (IM)
Performance VULNERABILITY & DAMAGE

Indicators assessment
S
NETWORK/ SYSTEMIC DIRECT LOSSES
ANALYSIS & DISRUPTIONS
IMPACT/ INDIRECT LOSSES

INTERACTIONS
Components/Systems
DECISION SUPPORT
Mitigation / Restoration Strategies
Seismic Vulnerability Assessment: Lifelines, Fig. 24 General layout for the seismic risk assessment of lifelines
Summary provided in terms of performance indicators that

describe the network functional losses or the
Lifelines are spatially distributed systems that expected economic or socioeconomic losses, pro-
provide essential services to any modern society. vide the means for an efficient mitigation or
They play also an important role for emergency recovery planning. Obviously, several sources
response and restoration in the aftermath of disas- of uncertainties are inherent in the various defini-
trous earthquakes; in general they are vital for the tions, models, tools, and methods for estimation
resilience of society. They are often grouped to of losses, which make the risk assessment of
transportation and utility systems and comprise lifelines a complex and challenging topic.
multiple components, which are exposed to dif-
ferent ground shaking effects and geotechnical
Cross-References
hazards. Research efforts and studies undertaken
the last 20 years, and after numerous devastating ▶ Damage to Infrastructure: Modeling
earthquakes that caused extensive losses and dis-
▶ Earthquake Risk Mitigation of Lifelines and
ruptions to lifelines and infrastructures, contrib-
Critical Facilities
uted to important improvement of knowledge and ▶ Empirical Fragility
expertise in the seismic vulnerability and risk
▶ Liquefaction: Performance of Building
assessment of lifelines. The main objective of
Foundation Systems
these efforts has been the development of meth- ▶ Probabilistic Seismic Hazard Models
odologies and tools for the estimation of probable
▶ Seismic Actions Due to Near-Fault Ground
direct and indirect losses (physical, performance,
Motion
economic, social) due to future earthquakes in ▶ Seismic Fragility Analysis
order to develop efficient emergency response
▶ Seismic Loss Assessment
and mitigation strategies. A general layout for
▶ Seismic Risk Assessment, Cascading Effects
the seismic risk assessment of lifelines is outlined ▶ Seismic Vulnerability Assessment: Reinforced
in Fig. 24. Inventory and taxonomy/typology def-
Concrete Structures
initions of all elements at risk and systems are the
▶ Spatial Variability of Ground Motion: Seismic
necessary initial steps to describe the elements Analysis
and networks exposed to seismic risk. The esti-
▶ Strengthening Techniques: Bridges
mation of site-specific seismic ground motion and
▶ Uncertainty Theories: Overview
other geotechnical hazards and the selection–es-
timation of appropriate seismic intensity mea-
sures (IM) is the basis of the seismic hazard
References
analysis. Vulnerability is the expected response
of an element or a network to a given seismic American Lifelines Alliance [ALA] (2001) Seismic fra-
intensity. It is commonly assessed based on fragility formulations for water systems. Part 1 – guide-
gility models, which estimate the probability of line. ASCE-FEMA, Washington, DC, 104 pp
Argyroudis S, Kaynia AM (2014) Fragility functions of
exceeding certain damage states for given seismic
highway and railway infrastructure. In: Pitilakis K,
intensity. Fragility functions are adapted with Crowley H, Kaynia AM (eds) SYNER-G: typology
respect to the typological characteristics of each definition and fragility functions for physical elements
element at risk. The expected damage and loss of at seismic risk, vol 27, Geotechnical, geological and
earthquake engineering. Springer, Dordrecht.
the entire network are estimated through appro- doi:10.1007/978-94-007-7872-6_10
priate systemic analysis, which takes into account Argyroudis S, Pitilakis K (2012) Seismic fragility curves
the physical damage and the relations and inter- of shallow tunnels in alluvial deposits. Soil Dyn Earthq
actions between the different components and Eng 35:1–12
Basöz N, Kiremidjian AS (1998) Evaluation of bridge
systems. The interdependencies among different
damage data from the Loma Prieta and Northridge,
systems may considerably increase the overall California earthquake. Technical report MCEER-98-
impact. The results, which are commonly 0004. State University of New York, Buffalo
Cardona OD, Ordaz MG, Reinoso E, Yamı́n LE, Barbar lifeline systems and critical facilities. Methodology
AH (2012) CAPRA – comprehensive approach to and applications. Springer, Dordrecht. ISBN 978-94-
probabilistic risk assessment: international initiative 017-8834-2
for risk management effectiveness. In: Proceedings Mouroux P, Le Brun B (2006) Risk-UE project: an
of the 15th world conference of earthquake engineer- advanced approach to earthquake risk scenarios with
ing, Lisbon, 24–28 Sept 2012 application to different European towns. In: Oliveira
Cavalieri F et al (2013) Application and validation study CS, Roca A, Goula X (eds) Assessing and managing
to an electric power network in Italy. In: Pitilakis K, earthquake risk. Springer, Netherlands, pp 479–508.
Argyroudis S (eds) Systemic seismic vulnerability and doi:10.1007/978-1-4020-3608-8_23
loss assessment: validation studies SYNER-G. Refer- NCEER (1995) The Hanshin-Awaji earthquake of January
ence report 6, Publications Office of the European 17, 1995: performance of lifelines, technical report
Union, doi: 10.2788/16706 NCEER-95-0015 (ed: Shinozuka M), State University
Cavalieri F, Franchin P, Pinto PE (2014) Application to of New York, Buffalo
selected transportation and electric networks in Italy. O’Rourke TD, Jeon SS, Toprak S, Cubrinovski M, Jung
In: Pitilakis K et al (eds) SYNER-G: systemic seismic JK (2012) Underground lifeline system performance
vulnerability and risk assessment of complex urban, during the Canterbury earthquake sequence. In: Pro-
utility, lifeline systems and critical facilities. Method- ceedings of the 15th world conference on earthquake
ology and applications. Springer, Dordrecht. ISBN engineering, Lisbon
978-94-017-8834-2 Pinto P (2014) Modeling and propagation of uncertainties.
Daniell J, Vervaeck A (2013) CATDAT Damaging earth- In: Pitilakis K, Crowley H, Kaynia AM (eds) SYNER-
quakes in 2012 – the year in review. http://earthquake- G: typology definition and fragility functions for phys-
report.com/2013/01/07/damaging-earthquakes-2012- ical elements at seismic risk, vol 27, Geotechnical,
database-report-the-year-in-review/ geological and earthquake engineering. Springer, Dor-
Douglas J, Seyedi DM, Ulrich T, Modaressi H, Foerster E, drecht. doi:10.1007/978-94-007-7872-6_2
Pitilakis K, Pitilakis D, Karatzetzou A, Gazetas G, Pitilakis K, Crowley E, Kaynia A (eds) (2014a) SYNER-
Garini E, Loli M (2014) Evaluation of seismic hazard G: typology definition and fragility functions for phys-
for the assessment of historical elements at risk: ical elements at seismic risk, vol 27, Geotechnical,
description of input and selection of intensity mea- geological and earthquake engineering. Springer, Hei-
sures. Bull Earthquake Eng. doi:10.1007/s10518-014- delberg. ISBN 978-94-007-7872-6
9606-0 Pitilakis K, Franchin P, Khazai B, Wenzel H (eds) (2014b)
Dueñas-Osorio L, Craig JI, Goodno BJ (2007) Seismic SYNER-G: systemic seismic vulnerability and risk
response of critical interdependent net-works. Earthq assessment of complex urban, utility, lifeline systems
Eng Struct 36(2):285–306 and critical facilities. Methodology and applications,
Elnashai A, Hampton S, Lee JS, McLaren T, Myers JD, Geotechnical, geological and earthquake engineering.
Navarro C, Spencer B, Tolbert N (2008) Architectural Springer, Heidelberg. ISBN 978-94-017-8834-2
overview of MAEviz-HAZTURK. J Earthq Eng Rinaldi SM, Peerenboom JP, Kelly TK (2001) Identifying,
12(S2):92–99. doi:10.1080/13632460802013610 understanding, and analyzing critical infrastructure
Faccioli E (ed) (2007) Prediction of ground motion and loss interdependencies. IEEE Contr Syst Mag 21(6):11–25
scenarios for selected infrastructure systems in Euro- Satumtira G, Duenas-Osorio L (2010) Synthesis of model-
pean urban environments: LESSLOSS report No. 2007/ ing and simulation methods on critical infrastructure
08, ISBN: 978-88-6198-012-9. IUSS Press, Pavia interdependencies research. In: Gopalakrishnan K,
Federal Emergency Management Agency [FEMA] Peeta S (eds) Sustainable and resilient critical infra-
(2003) Multi hazard loss estimation methodology: structure systems. Springer, Berlin/Heidelberg.
earthquake model – HAZUS-MH MR3 technical man- doi:10.1007/978-3-642-11405-2 S
ual. FEMA, Washington, DC Seyedi DM, Gehl P, Douglas J, Davenne L, Mezher N,
Kakderi K, Pitilakis K (2010) Seismic analysis and fragil- Ghavamian S (2010) Development of seismic fragility
ity curves of gravity waterfront structures. In: Fifth surfaces for reinforced concrete buildings by means of
international conference on recent advances in geo- nonlinear time-history analysis. Earthq Eng Struct
technical. Earthquak Engineering and Soil Dynamics 39:91–108
and Symposium in Honour of Prof. I. M. Idriss, 6.04a, Tang C, Zhu J, Qi X (2011) Landslide hazard assessment
San Diego, CA, May 24–29 of the 2008 Wenchuan earthquake: a case study in
Mackie K, Stojadinovic B (2003) Seismic demands for Beichuan. Can Geotech J 48:128–145
performance-based design of bridges. PEER Report Weatherill G, Esposito S, Iervolino I, Franchin P, Cavalieri
2003/16. Pacific Earthquake Engineering Research F (2014) Framework for seismic hazard analysis of
Center, University of California, Berkeley spatially distributed systems. In: Pitilakis K et al (eds)
Modaressi H, Desramaut N, Gehl P (2014) Specification SYNER-G: systemic seismic vulnerability and risk
of the vulnerability of physical systems. In: Pitilakis assessment of complex urban, utility, lifeline systems
K et al (eds) SYNER-G: systemic seismic vulnerabil- and critical facilities. Methodology and applications.
ity and risk assessment of complex urban, utility, Springer, Dordrecht. ISBN 978-94-017-8834-2
3164 Seismic Vulnerability Assessment: Masonry Structures
technical resources available and of the dimen-

Seismic Vulnerability Assessment: sions and building density of the area under
Masonry Structures investigation, as well as the diversity within the
building stock.
Dina D’Ayala and Viviana Novelli Seismic vulnerability analysis of masonry
Department of Civil, Environmental and structures has traditionally been carried out
Geomatic Engineering, University College using “empirical” (or statistical) methods. These
London, London, UK can be divided in categorization methods, which
classify buildings into typologies characterized
by propensity to damage, or inspection and rating
Introduction methods, wherein scores are attributed to each
significant characteristic of the buildings thought
The assessment of the vulnerability of the build- to have a bearing on its overall vulnerability
ing stock of an urban center is an essential pre- (UNDP/UNIDO 1985). Several of these methods
requisite to its seismic risk assessment. The have been developed in the past 30 years, and the
other two ingredients are the expected hazard more robust and commonly used are reviewed in
over given return periods and the distribution section “Empirical Methods.”
and values of the assets constituting the building Recent dramatic increase in computational
stock. All three elements of the seismic risk powers and consequent improvement and refine-
assessment are affected by uncertainties of ale- ment in numerical modelling of relatively com-
atory nature, related to the spatial variability of plex structures, using either static or dynamic
the parameters involved in the assessment, and approaches (e.g., Fajfar and Gašperšič 1996;
epistemic nature, related to the limited capacity Vamvatsikos and Cornell 2005), has facilitated
of the models used to capture all aspects of the greater exploitation of analytical approaches for
seismic behavior of buildings and to describe vulnerability assessment purposes. Analytical
them in simple terms, suitable for this type of methods need experimental validation of the
analysis. Hence it should always be kept in mind various parameters used to define the vulnerabil-
that the computation of a risk level is highly ity, and so far relatively little experimental work
probabilistic and that to accurately represent on the behavior of masonry structures has been
the risk, the expected values should always be undertaken by the international community,
accompanied by a measure of the associated albeit a wide variety of existing unreinforced
dispersion. A very preliminary estimate of the masonry typologies, both historic and modern,
seismic capacity of the local building stock can exists. The framework within which analytical
be obtained by consulting requirements included methods can be applied and their limitation are
in seismic standards and code of practices in presented in section “Analytical Methods.”
force at the time of construction of such build- An attempt to exploit positive aspects of the
ings. This information together with a temporal two classes of methods above is made with the
and spatial record of the growth of the urban so-called hybrid methods. These combine numer-
center can provide a definition of classes of ical input/output from analytical models with sta-
buildings assumed to have different capacities. tistical and probabilistic data to define exposure
This information can be obtained by looking at and vulnerability distribution. This allows to
past and present cadastral maps with ages of reduce the analytical burden while grounding
buildings and knowing the historical develop- results in a geographical context. Applicability
ment and enforcement of codes at the site. In and reliability of such approaches are constrained
general however for a correct assessment of the by the ability to define certain mechanical and
seismic risk, a more detailed inventory and clas- structural characteristics in numerical terms and
sification should be considered, the extent of by the need to define a common approach for
which is a function of the economic and treating the various sources of uncertainty
Seismic Vulnerability Assessment: Masonry Structures 3165
associated with vulnerability, exposure, and haz- historic seismicity records (Askan and Yucemen
ard. This is further discussed in section “Hybrid 2010; Zobin et al. 2010). Notwithstanding its
Methods.” popularity, the DPM has major limitations: dis-
The review and discussion of existing methods crete definition of damage levels/states and
sets out the rationale for a combined method, dependence on a specific seismic and architec-
called FaMIVE (Failure Mechanism Identifica- tural context. Hence, it may not be applicable to
tion and Vulnerability Evaluation), and devel- different geographic locations, in the absence of
oped by the first author and her co-researchers, direct damage data.
in the past decade. The hypotheses and analytical Some of these limitations are overcome by the
approach to derive capacity curves and fragility Vulnerability Index Method (VIM) (Benedetti
functions are presented in detail in section et al. 1988), based on the vulnerability index Iv.
“Development of Analytical Methods: FaMIVE.” The vulnerability index is a summation of
Results obtained from the FaMIVE method are weighted parameters, each associated with a con-
presented in section “Damage Thresholds and structional or structural feature of the typology,
Performance Indicators” in terms of damage which affects its vulnerability. In this way the
limit states and drift and compared with the rec- definition of vulnerability relates only to the
ommendation of Eurocode 8 (EC8) (CEN 2005) mechanical characteristics of the building, while
and experimental evidence to discuss issues of damage data from past earthquakes are used to
validation and calibration. calibrate the vulnerability functions, by relating
Iv to observed global damage levels for buildings
of the same typology and hence extending appli-
Vulnerability Approaches cability in regions having experienced the same
level of macroseismic intensity or peak ground
Empirical Methods acceleration (PGA).
The damage probability matrix (DPM) method The most substantial improvement of the VIM
has been the most common vulnerability assess- over the DPM is that the former provides a contin-
ment method used since the 1970s. Based on field uous vulnerability function, while the latter uses
observations, it expresses in a discrete form the discrete vulnerability classes expressed in terms of
conditional probability of obtaining a damage expected damage. This means that possible inter-
level Dj, due to a ground motion of macro inten- ventions to shift the vulnerability level of a struc-
sity Ii, tural typology are readily quantifiable with Iv,
while they are only recorded in the DPM if the
Dj intervention results in a shift of damage class.
P¼ D¼ :
Ii Moreover, the correlation of VIM with an instru-
mental measure of seismic action allows for appli- S
The method relies on the wealth of observed cation to and comparison between different seismic
damage data available from past earthquakes zones, even though in the last decade the reliability
and the correlation of these, on the one hand, of correlating PGA and damage has been put into
with assigned shaking macroseismic intensity at question by various authors (Elenas 2000; Wu
the site of observation and, on the other, with et al. 2003). Giovinazzi and Lagomarsino (2004)
construction materials and methods in different have proposed a direct correlation of the two
geographical and seismic regions. First proposed methods by introducing a definition of damage
by Whitman et al. (1973) after the San Fernando states and DPM as function of the Iv using the
earthquake of 1971, specific applications of this definition of the EMS-98 macroseismic scale
method to masonry structures are numerous (Gr€unthal 1998). This approach has been further
(Braga et al. 1982; Corsanego and Petrini 1990; developed in the European RISK-UE project for
Di Pasquale et al. 2005) and still widely used in larger number of building typologies and vulnera-
developing countries and regions with extensive bility classes (Lagomarsino and Giovinazzi 2006).
The EMS-98 scale (Gr€unthal 1998) is in itself a vulnerability assessment algorithms with direct
DPM approach, where vulnerability classes are physical meaning. In the development of analyt-
defined across different construction typologies. ical fragility functions for masonry structures,
The uncertainty inherent in the attribution of a two methods can be identified: (i) correlation
typology to a class and of its response to a seismic between damage index and damage thresholds
event in terms of damage level is dealt with by and (ii) correlation between acceleration/dis-
assuming a central value and a fuzzy set member- placement capacity curves and spectral demand
ship, so that both vulnerability and damage can be curves, following the HAZUS99 (FEMA 2001)
expressed in descriptive (few, many, most) rather or N2 method (Fajfar and Gašperšič 1996). These
than numerical terms. A method to compute total two procedures, in line with the performance-
probability of damage of a given typology spread- based design for new built, define the seismic
ing over more vulnerability classes is proposed in assessment of existing building stock and predic-
Jaiswal et al. (2011). tion of losses. The required analysis steps are
A first attempt at deriving generic fragility (i) classification of buildings by typology and
curves was made by Spence et al. (1992) with seismic design, (ii) definition of damage states,
the Parameterless Scale Intensity (PSI), aimed at (iii) assignment of capacity curves, (iv) definition
providing fragility curves for different building of demand spectra (associated with return periods
typologies and at quantifying benefits from and performance targets), and (v) evaluation of
upgrading and strengthening interventions of building response in terms of performance points.
the building stock. The advantage of this The two procedures, applied to large sets of
approach is that damage grades and fragility buildings, allow to derive fragility curves and
curves are independent of macroseismic inten- damage scenarios for given sites and as a function
sity scales, as they are defined relative to each of ground motion parameters which are linked
other, on the assumption that for each building with hazard levels. The damage levels are
type, it is possible to define the level of the scale directly related to the demand parameter, which
corresponding to the median of the fragility is expressed in terms of displacement or drift. As
curve for level of damage D3 (structural dam- the fragility curves are developed on the basis of
age). Limitations of applicability are however lognormal distribution, once the typologies have
related to the extensive need for observations been defined, it is not necessary to have a detailed
to calibrate each fragility curve and to the diffi- knowledge of the building stock. All required are
culty to correlate observed or expected damage the parameters defining the capacity curve for
into construction and structural characteristics in each typology, various damage thresholds, and
a deterministic way. the number of buildings belonging to each typol-
ogy. Once capacity curves are developed, then
Analytical Methods the range of behavior and variability of fragility
Analytical methods present the advantage of within a building class or stock can be analyzed
framing the problem of seismic vulnerability of by parametric analysis, providing important
masonry structures in structural engineering insight for retrofitting (see D’Ayala 2005). Ana-
terms, defining a direct relationship among con- lytical approaches to define seismic vulnerability
struction characteristics, structural response to of masonry buildings are becoming more and
seismic action, and damage effects. more popular, as improved engineering knowl-
The development of attenuation equations for edge on the behavior of masonry structures
specific seismic regions and corresponding deri- increases confidence on the reliability of such
vation of seismic hazard maps in terms of spectral models. In the past decade, a relatively significant
ordinates, as opposed to macroseismic intensity number of procedures aimed at defining reliable
or PGA, has given impetuous to the development analytical vulnerability functions for masonry
of analytical methods. These methods tend to structures in urban context have been proposed
feature more detailed and transparent (Lang and Bachmann 2004; Erberik 2008; Borzi
et al. 2008). Although they share similar concep- et al. (2008) have applied it for estimating direct
tual hypotheses, they differ substantially in losses for masonry buildings in Thessaloniki
modelling/numerical complexity and treatment (Greece). This procedure is very useful when
of uncertainties. Two main limitations are com- the damage data collection of an area of interest
mon to all: limited geographic applicability and with a specific intensity is only partially avail-
limited number of failure mechanisms. Specifi- able. However, the use of different data sources
cally, for the latter, most of the analytical models derived from different procedures, for which a
only consider in-plane or frame-like behavior, direct cross-correlation is not readily available,
disregarding overturning and out-of-plane mech- might not necessarily result in reduced uncer-
anisms, often occurring at lower levels of shaking tainty of the output.
and hence substantially affecting the seismic vul- An application of this approach to masonry
nerability in an urban context. Analytical vulner- structure within the Risk-UE project shows the
ability assessment methods for existing masonry importance of knowledge of the building stock
buildings feature in many risk assessment inte- for reliable seismic vulnerability assessment, as
grated systems; however, their choices of capac- increased level of uncertainties is compounded
ity curves are predetermined. Among nonlinear both on capacity and demand curves
analysis methods, the collapse-mechanism or (Lagomarsino and Giovinazzi 2006). Application
limit-equilibrium method presents the advantage made using code prescription in terms of capacity
of requiring few input parameters and allowing curves provides a good first approximation but
consideration of different failure modes. Proce- needs extensive in situ calibration as shown by
dures using this approach are based on collapse Erberick (2008).
load multipliers which identify the occurrence A summary of the methods reviewed above is
of different possible mechanisms for given typol- presented in Table 1. While this is by no means
ogy and structural characteristics. Bernardini et al. an exhaustive list of the many applications of
(1990) developed a numerical routine (VULNUS), seismic vulnerability assessment for masonry
combined with a fuzzy set theory, in recognition of structures available in literature, it concentrates
the limited knowledge associated with many of the on procedures that have been specifically devel-
parameters. Only one in-plane and one simple oped for vulnerability assessment at territorial
out-of-plane mechanism were initially considered scale. For this reason procedures aimed at the
as possible failures. D’Ayala and Speranza (2003) assessment of single buildings are not included.
further enhanced this approach by developing the The choice of the most suitable procedure is
FaMIVE method, a mechanical approach based on highly dependent on the resources available for
a suite of 12 possible failure mechanisms directly the data collection, the computational expertise
correlated with in situ observed damage. The available, and ultimately the scale and aim of
FaMIVE algorithm produces vulnerability the study. Empirical procedures can be used for S
functions for different building typologies and fairly large-scale studies to define damage sce-
quantifies the effect of strengthening and repair narios; however, if the purpose of the study is to
intervention on reduction of vulnerability. The identify within a district or urban center specific
method has been applied in several locations buildings in need of strengthening, so as to
worldwide and recently in Abruzzo, Italy, follow- increase their seismic capacity, then a suitable
ing the 2009 L’Aquila earthquake (D’Ayala and analytical procedure should be preferred.
Paganoni 2011).
Hybrid Methods Development of Analytical Methods:

Hybrid methods combine post-earthquake dam- FaMIVE
age statistics with analytically derived nonlinear
behavior using pushover analysis (e.g., Kappos The seismic vulnerability assessment of
et al. 1998; Barbat et al. 1996). Recently, Kappos unreinforced masonry historic buildings can be
3168
Seismic Vulnerability Assessment: Masonry Structures, Table 1 Summary of procedures for the seismic vulnerability assessment of masonry structures
Assessment
Method References Data requirement Data collection type Approach Demand input Output
VULNUS Bernardini et al. (1990) Geometry, material On-site Analytical Mechanism PGA, Lateral force, lateral
parameters, observation or method response drift, vulnerability index
structural details systematic spectrum
survey
FaMIVE D’Ayala and Speranza Geometry, material On-site Analytical Capacity PGA, Lateral force, lateral
(2003); D’Ayala (2005) parameters, observation or spectrum-based response drift, damage scale,
structural details systematic method, spectrum vulnerability index
survey mechanism
SP- Borzi et al. (2008) Structural Random Analytical Pushover-based PGA, Lateral force, lateral drift
BELA description generation method response
spectrum
CSBM RISK-UE Kappos Structural Systematic Analytical Capacity PGA, Lateral force, lateral drift
et al. (2008) description survey and spectrum-based response
random method spectrum
generation
VIM Benedetti et al. (1988); Typological On-site Empirical Vulnerability Macroseismic Phenomenological
Lagomarsino and description observation index method intensity, PGA damage scale
Giovinazzi (2006)
DPM Whitman et al. (1973) Typological On-site Empirical Damage Macroseismic Phenomenological
description observation probability matrix intensity damage scale
AeDES Baggio et al. (2009) Typological On-site Empirical Vulnerability Macroseismic Phenomenological
description observation index method intensity damage scale
PSI Spence et al. (1992) Typological Exposure Empirical, Damage Macroseismic Phenomenological
description database statistical probability intensity damage scale
function
Hybrid Kappos et al. (1998); Geometry, material On-site Empirical/ Hybrid methods Macroseismic Damage scale, lateral
Methods Barbat et al. (1996) parameters observation analytical, intensity forces
statistical basis
Seismic Vulnerability Assessment: Masonry Structures
Seismic Vulnerability Combined Mechanisms

Assessment: Masonry
Structures,
Fig. 1 Mechanisms for
computation of limit lateral
capacity of masonry
façades
B1: façade B2: façade C: overturning with F: overturning
overturning with overturning with diagonal cracks constrained by
one side walls two side walls involving corners rings beams or ties
In plane Mechanisms
H1: diagonal H2: diagonal M1: soft storey M2: soft storey
cracks mainly cracks mainly due to shear due to bending
in piers in spandrel
Out of Plane Mechanism
A: façade D: façade E: façade G: façade

overturning with overturning with overturning with overturning with
vertical cracks diagonal crack crack at spandrels diagonal cracks
performed with the Failure Mechanisms Identifi- over either part or the whole façade. As shown in
cation and Vulnerability Evaluation (FaMIVE) the flowchart of Fig. 2, the program FaMIVE first
analytical method (D’Ayala and Speranza 2003; calculates the collapse load factor for each façade
D’Ayala 2005). The FaMIVE method uses a in a building and then, taking into account geomet-
nonlinear pseudo-static structural analysis with ric and structural characteristics and constraints,
a degrading pushover curve to estimate the per- identifies the one which is most likely to occur
formance points by way of a variant of the N2 considering the combination of the largest portion S
method (Fajfar and Gašperšič 1996), included in mobilized with the lowest collapse load factor at
EC8 part 3 (CEN 2005). It yields as output col- building level.
lapse load multipliers which identify the occur- The FaMIVE algorithm produces vulnerabil-
rence of possible different mechanisms for a ity functions in terms of ultimate lateral capacity
given masonry construction typology, given cer- for different building typologies and quantifies
tain structural characteristics. the effect of strengthening and repair intervention
It is based on computing the collapse load fac- on reduction of vulnerability. In its current ver-
tor for 12 different failure modes, shown in Fig. 1. sion, it also computes capacity curves and perfor-
Each mode of failure corresponds to different con- mance points and outputs fragility curves for
straints and conditions between the façade and the different seismic scenarios in terms of intermedi-
rest of the structure; hence a collapse mechanism ate and ultimate displacements or ultimate accel-
can be univocally defined and its collapse load eration. Within the FaMIVE database, capacity
factor computed for each mechanism developing curves and fragility functions are available for
North
t
es st
W
South Ea
FaMIVE
Data Collection
West South East North
Analysis
West South East North
A .. M A .. M A .. M A .. M
Min Min Min Min
Equivalent Shear Capacity Mechanism
Non-Linear
SDOF
Fb
Meff
eff Keff
Meffüg(t)
Teff
v for a selected SITE

DL SD NC C
Spectral Response Acceleration
Lateral capacity
Capacity Curve
Mean STD Mean STD Mean STD Mean STD SD NC

vU
Sa (a/g)
vY
Logn. Distr Logn. Distr Logn. Distr Logn. Distr
DL
C
Y sd U c
Period T
Lateral displacement
Performance point
SD NC
Sa (a/g)
DL
C
Sd (cm)
Seismic Vulnerability Assessment: Masonry Structures, Fig. 2 Flowchart setting out the rationale of the FaMIVE
procedure
various unreinforced masonry typologies, from Et I eff Et Aeff

K eff ¼ k1 þ k2 (1)
adobe to concrete blocks, for a number of refer- H 3eff H eff
ence typologies studied at sites in Italy, Spain,
Slovenia, Turkey, Nepal, India, Iran, and Iraq. where Heff is the height of the portion involved in
The procedure has been validated against the the mechanism; Et is the estimated modulus of
EMS-98 vulnerability classes and recently used the masonry as it can be obtained from experi-
to produce capacity curves and fragility curves mental literature for different masonry typolo-
for use in the USGS PAGER environment. gies; Ieff and Aeff are the second moment of area
The mechanism’s characteristics are used to and the cross-sectional area, respectively, calcu-
derive an equivalent nonlinear single degree- lated taking into account extent and position of
of-freedom capacity curve to be compared to a openings and variation of thickness over height;
spectrum demand curve and eventually define and k1and k2 are constants which can assume
performance points as illustrated in the flow- values equal or greater than 0, depending on
chart in Fig. 2. edge constraints and whether shear and/or flex-
In order to derive fragility curves, limit state ural stiffness are relevant for the specific mecha-
performance criteria correlated to damage states nism. For instance, for in-plane mechanisms, the
need to be defined. This step is fraught with uncer- inclusion of shear stiffness in Eq. 1 will depend
tainties, as very limited consolidated evidence on the slenderness of the pier and the relative
exists to perform such correlation over a wide stiffness between piers and spandrels of the
range of building typologies and shaking levels. façade being assessed.
While robust database of damage states exist The tributary mass Oeff is calculated following
in literature, no attempt has been so far made the same approach, and it includes the portion
to record permanent drift and corresponding of the elevation activated by the mechanisms
ground shaking in a consistent way so as to provide plus the mass of the horizontal structures
empirical evidence for capacity curves. As an involved in the mechanism:
alternative, a number of authors have worked on
correlating performance indicators and damage Oeff ¼ V eff dm þ Of þ Or (2)
indicators on experimentally obtained capacity
curves, by way of shaking table tests or pushover
tests. where Veff is the solid volume of the portion of
wall involved in the mechanism, dm is the density
of the masonry Of, Or are the masses of the
Definition of Capacity Curve horizontal structures involved in the mechanism.
Capacity curves can be derived for each façade Effective mass and effective stiffness are used to
on the basis of the following steps. The first step calculate a natural period Teff, which character-
S
is to calculate the lateral effective stiffness for izes an equivalent single degree-of-freedom
each wall and its tributary mass. As the floor and (SDoF) oscillator:
roof structures in traditional masonry buildings
sffiffiffiffiffiffiffiffi
have relatively modest in-plane stiffness, it is Oeff
assumed that redistribution of inertia forces T eff ¼ 2p (3)
K eff
among walls will be ineffective and torsional
effects are not considered. The effective stiff-
ness of each external wall is calculated on the The mass is applied at the height of the center of
basis of the type of mechanism attained, gravity of the collapsing portion with respect to
the geometry of the wall and layout of opening, the ground, and a linear acceleration distribution
the constraints to other walls and floors, and over the wall height is assumed. The elastic limit
the portion of other walls involved in the acceleration Say is also computed depending on
mechanism: the failure mechanism identified; for failure
mechanisms involving flexural strain limit and brickwork in mud mortar, while the upper
(assuming no tensile capacity in the material), bound can be used for massive stone, brickwork
for instance, Say will be the value of lateral set in lime or cement mortar, and concrete
acceleration that, combined with gravitational blockwork. The near collapse condition is deter-
load resultant, will cause a triangular distribu- mined by the displacement Sdu identified by
tion of compression stresses at the base the condition of loss of vertical equilibrium
of the overturning portion, just before the onset which, for overturning mechanisms, can be com-
of cracking: puted as a lateral displacement at the top or
for in-plane mechanism by the loss of overlap
tb of two units in successive courses as quantifiable
Say ¼ g with corresponding displacement
6h0 with one of the two alternative expressions
Say in Eq. 7:
Sde ¼ 2 T 2eff
4p
(4) Sdu tb =3 or Sdu lu =2 (7)
where tb is the effective thickness of the wall at

the base of the overturning portion, h0 is the where tb is the thickness at the base of the
height to the ground of the center of mass of the overturning portion and lu is the typical length
overturning portion, and Teff is the natural period of units forming the wall. In the case of in-plane
of the equivalent single degree-of-freedom mechanism, the geometric parameter used for the
(SDoF) oscillator. The maximum lateral capacity elastic limit is, rather than the wall thickness, the
Sau is defined as width of the slender pier.
The threshold points identified by Eqs. 4, 5, 6,
lc and 7 can be associated to corresponding states of
Sau ¼ g (5) damage. Specifically DL, damage limitation, cor-
a1
responds to the elastic lateral capacity threshold
(Sde, Say) defined by Eq. 4; SD, significant dam-
where lc is the collapse load factor of the mech-
age, corresponds to the peak capacity threshold
anism chosen, calculated by FaMIVE, and a1 is
(Sds, Sau) defined by Eqs. 5 and 6; and NC, near
the proportion of total mass participating to the
collapse, corresponds to incipient or partial col-
mechanism. This is calculated as the ratio of the
lapse threshold (Sdu, Sau) defined by either condi-
mass of the façade and sides or internal walls
tions of Eq. 7.
and floor involved in the mechanism Oeff to the
The procedure’s approach allows a direct
total mass of the involved macroelements (walls,
analysis of the influence of different parameters
portion of floors, and portion of roof supported
on the resulting capacity curves, whether these
by the wall). The displacement corresponding
are geometrical, mechanical, or structural. By
to the damage threshold of structural damage,
way of example, Figs. 3 and 4 show a comparison
Sds is
of average capacity curves obtained by grouping
the results using different criteria for the same
3Sdy Sds 6Sdy (6) sample of buildings in L’Aquila. In Fig. 3 the
average curves represent the behavior according
as suggested by (Tomazevic 2007). The range in to the classes of mechanisms as per Fig. 1. In
Eq. 6 is useful to characterize masonry fabric of Fig. 4 the capacity curves are obtained by con-
diverse levels of regularity and its integrity at sidering different structural typologies, as classi-
ultimate conditions, with the lower bound better fied by the WHE-PAGER project (Jaiswal
describing the behavior of adobe, rubblestone, et al. 2011) and shown in Table 2. It can be seen
Seismic Vulnerability 0.3

Assessment: Masonry
Structures, Combined
Fig. 3 Average capacity 0.25 IN Plane

curves for sample grouped OUT of Plane
by collapse-mechanism
Lateral Capacity (g)

0.2
classes
0.15
0.1
0.05
0
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035
Drift

Assessment: Masonry
Structures, UFB3
Fig. 4 Average capacity 0.25 UFB5
curves for sample grouped DS2
by structural typology
Lateral Capacity (g)
DS4
0.2 RS3
0.15
0.1
0.05
0
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
Drift
that the correlation between mode of failure and the building stock. The lack of such knowledge S
structural typology is qualitatively good but not determines a non-negligible epistemic error.
univocal, and the grouping affects both ultimate In Figs. 3 and 4 the capacity curves show a
lateral capacity and drift at all damage states. fourth point, which represents the collapse of the
Such results bring in sharper focus the façade corresponding to a lateral displacement Sdc
limitation and inaccuracy of using idealized and a lateral acceleration Sac as computed in Eq. 8:
models and average curves without adequately
considering the inherent aleatoric variation asso- Sau
Sdc tb =2 or Sdc lu , Sac ¼ (8)
ciated with any given site where the assessment is 2
conducted and the importance of a detailed
knowledge of the local construction characteris- Such relationship are estimated on the basis of
tics when sampling the building representative of experimental work available in literature.
Seismic Vulnerability Assessment: Masonry Struc- curve derived above with the demand spectra
tures, Table 2 Structural typologies classification for different return periods in relation to the
according to PAGER (Jaiswal et al. 2011)
performance criteria considered. Two broadly
Load-bearing PAGER equivalent approaches for the derivation of
material code Description
the nonlinear demand spectra exist: the N2
Stone RS3 Local fieldstones with
masonry lime mortar method (Fajfar 1999) included in the EC8 and
RS4 Local field stones with the capacity spectrum method (CSP) (FEMA
cement mortar, vaulted 2001). The two methods differ essentially in
brick roof, and floors the way the nonlinear demand spectrum is
DS2 Rectangular cut stone arrived at: the N2 method uses a reduction
masonry block with lime
factor R, function of the structure expected duc-
mortar
DS3 Rectangular cut stone
tility m, while the CSP uses a fictitious damping
masonry block with factor derived from the hysteresis loop of
cement mortar the structure. In the following the N2 method
DS4 Rectangular cut stone will be used to illustrate the derivation of
masonry block with performance points.
reinforced concrete floors
and roof The original version of the N2 method was
MS Massive stone masonry in derived for elastic-perfectly plastic SDoF oscil-
lime or cement mortar lator, and the relationship between the reduction
Brickwork UFB1 Unreinforced brick factor R and the ductility m, used to determine the
or masonry in mud mortar nonlinear response spectrum, is dependent on the
blockwork without timber posts
bilinear shape of the capacity curve. Alternative
UFB3 Unreinforced brick
masonry in lime mortar.
relationships between the reduction factor, the
Timber flooring ductility, and the natural period of the system,
UFB4 Unreinforced fired brick R-m-T, have been developed to determine the
masonry, cement mortar. inelastic displacement ratios and the inelastic
Timber flooring response spectra for structures characterized by
UFB5 Unreinforced fired brick
quadri-linear backbone curves, characterized by
masonry, cement mortar,
but with reinforced a plateau of reduced capacity and further ductil-
concrete floor and roof ity, after a negative slope branch (see Miranda
slabs 2001; Dolsek and Fajfar 2004; Vamvatsikos and
UCB Unreinforced concrete Cornell 2005).
block masonry with lime
or cement mortar
In the case of the capacity curves derived with
the FaMIVE procedure however, as mentioned
above, the behavior beyond the near collapse
performance point is fraught with uncertainty
and hence the descending branch of the curve is
Damage Thresholds and Performance ignored, and the simplified relationship for bilin-
Indicators ear curves is used.
In the previous section capacity curves have To calculate the coordinates of the perfor-
been determined as polylines characterized by mance point in the displacement-acceleration
four notable points which represent as many space, the intersection of the capacity curve
damage thresholds. A method for assessing the with the nonlinear demand spectrum for an
overall behavior by use of a global performance appropriate level of ductility m can be determined
indicator is the computation of the performance as shown in Eq. 8, given the value of ultimate
point. In order to calculate the performance lateral capacity of the equivalent SDoF oscillator,
point, it is necessary to intersect the capacity Sau:
if T < T c
T 2c ðSae ðT Þ Sa ðT ÞÞ2 gm
if Sau Sa ðT Þ ) Sd ðmÞ ¼ 2
2 S ðT Þ
ð m 1Þ 4p a
2
T 2
ð S ð T Þ S Þ gm
if Sa ðT c Þ < Sau < Sa ðT Þ ) Sd ðmÞ ¼ c
ae au
2
2
ð m 1Þ 4p Sau
2
gT ðSae ðT ÞÞ
2
if Sau Sa ðT c Þ ) Sd ðmÞ ¼ c 2 (9)
4p mSau
if T T c
gT 2 ðSae ðT ÞÞ2
if Sau Sa ðT Þ ) Sd ðmÞ ¼ c 2
4p mSa ðT Þ
gT 2 ðSae ðT ÞÞ2
if Sau < Sa ðT Þ ) Sd ðmÞ ¼ c 2
4p mSau
where two different formulations are provided for Figures 6 and 7 show the set of fragility curves
values of ultimate lateral capacity Sau greater or obtained for the damage limit states of damage
smaller than the nonlinear spectral acceleration limitation and significant damage as computed
Sa(TC) associated with the corner period TC mark- for the two Italian sites of Nocera Umbra and
ing the transition from constant acceleration to Serravalle considering separately the three types
constant velocity section of the parent elastic spec- of structural behavior. Once a structural typology
trum. Figure 5 shows an example of calculation of has been assigned, the values of the mechanical
performance points for median capacity curves for characteristics are the same across the two sam-
some different types of mechanisms (see Fig. 1). ples, while the structural details are accounted for
directly in the three classes of mechanisms.
Derivation of Fragility Functions Hence the variability observed between samples
Once capacity curves and limit state points are in each chart can be related directly to geometric
identified on them, fragility curves for different differences and masonry fabric, i.e., to the local
limit states are obtained by computing the median aspects of the construction practice and architec-
and standard deviation values of the performance tural layout. Curves on the left of the diagrams
point displacements for each index building in a relate to corresponding capacity curves with a
given set of buildings surveyed on-site or created stiffer elastic branch in the case of damage limi-
through randomization of the input parameters. tation and with lower ductility in the case of
The median and standard deviation are used to significant damage. However the distribution
derive equivalent lognormal distributions. To this does not bare consistency across the three classes
end the median displacement for each limit state of mechanism for the two sites. S
can be calculated as:
1 X
n
S^dsi ¼ emi with mi ¼ lnSjdsi (10) Conducting Seismic Vulnerability
n j¼1 Assessment
and the corresponding standard deviation as: As seen in the previous section, an analytical
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi method based on limit state analysis and mecha-
bdsi ¼ emi þ2si esi 1 with si
1 2 2
nism approach has the advantage of employing
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uX n 2 specific structural models to assess the seismic
u j
u lnSjdsi lnSdsi behavior of masonry buildings of different typol-
t j¼1
¼ (11) ogies, represented through capacity curves and
n performance points, while at the same time

Assessment: Masonry LS EC8 soil type B
Structures, 1.2 NLS EC8 soil type B, m=2.5
Fig. 5 Example of Mean CC COMB
Mean CC IP
calculation of performance 1
Acceleration (%g)
Mean CC OOP
points for median capacity Performance point
curves for types of 0.8
mechanisms triggered
(COMB combined 0.6
mechanism, IP in-plane
mechanism, OOP out-of- 0.4
plane mechanism; see also
Fig. 1) 0.2
0
0 5 10 15 20 25 30 35 40 45
Displacement (cm)
Seismic Vulnerability LS Damage Limitation

Assessment: Masonry 1
Serravalle Combined
Structures, 0.9
Serravalle In-Plane
Fig. 6 Fragility curves for 0.8 Serravalle Out-Of-Plane
limit state of damage 0.7 Nocera Umbra Combined
limitation for the three 0.6 Nocera Umbra In-Plane
classes of collapse 0.5 Nocera Umbra Out-Of-Plane
mechanisms for two 0.4
different Italian sites
0.3
0.2
0.1
0
0.0001 0.001 0.01 0.1 1 10 100
Displacement (cm)
Seismic Vulnerability LS Significant Damage

Assessment: Masonry 1 Serravalle Combined
Structures, 0.9 Serravalle In-Plane
Fig. 7 Fragility curves for 0.8 Serravalle Out-Of-Plane
limit state of significant 0.7 Nocera Umbra Combined
damage for the three classes Nocera Umbra In-Plane
0.6
Nocera Umbra Out-Of-Plane
of collapse mechanisms for 0.5
two different Italian sites 0.4
0.3
0.2
0.1
0
0.0001 0.001 0.01 0.1 1 10 100
Displacement (cm)
being applicable to relatively large numbers of when a group of heritage structures at territorial
buildings. Hence procedures of this type, if prop- scale are considered (D’Ayala and Ansal 2012).
erly developed, are particularly suitable for the A robust approach for seismic vulnerability
seismic assessment of building aggregates in his- assessment at territorial scale needs to address a
torical centers (Novelli and D’Ayala 2014) or number of issues, mainly related to a correct
Seismic Vulnerability a 1
Assessment: Masonry 0.9
% of damaged buildings
Structures, 0.8
Fig. 8 Fragility curves for
0.7
two different samples – (a)
30 buildings and (b) 0.6
200 buildings – represented 0.5
as actual distribution and 0.4
lognormal distribution 0.3
(Eqs. 10 and 11) 0.2
0.1
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6
Collapse load factor (g)
as surveyed good connections ties

as surveyed LGN good connections LGN ties LGN
b 100
90
80
70
60
50
Cumulative Distribution
40
Lognormal Distribution
30
20
10
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Collapse Load Factor (g)
representation of the existing exposure so as to • Identification of typologies and construction

obtain meaningful fragility and vulnerability details
functions, while explicitly accounting for the • Size and heterogeneity of sample to be
uncertainties related to modelling limitation, the surveyed
inherent randomness of the sample, and the ran- • Data collection S
domness of the response. Given a territorial dis- • Data analysis
tribution, uncertainties are also present about the • Safety and conservation requirements
seismic action, relating to record to record, atten- • Seismic hazard
uation, soil characteristics, etc. Hence specific • Definition of performance points
attention and due consideration should be given • Definition of fragility functions
to the following aspects of a process leading to a • Rehabilitation decisions
seismic vulnerability assessment at territorial
scale and identification of rehabilitation A detailed discussion of each step is beyond the
strategies. scope of this document. However some observa-
Such process can be articulated in 10 steps: tions in relation to specific applications of FaMIVE
can help in highlighting some of the issues.
• Consideration on the urban fabric and In relation to the sample size and heterogene-
study area ity, for instance, Fig. 8a, b shows the resulting
distributions for two different cases. The first superior to KL1, limited, but not quite equal to
sample included 30 buildings, which were con- KL2, normal. For this level of knowledge, a
sidered relatively homogenous in terms of struc- static nonlinear analysis, such as the limit state
tural typology and materials but differing in mechanism approach, leading to a capacity
geometry and quality of connections. The second curve is deemed appropriate. Hence according
one is made up of more than 200 buildings of to the recommended values, the confidence fac-
varied characteristics. As it can be seen, the tor CF should be in the range 1.2 to 1.35
assumption of a lognormal model to describe depending on how closer the actual knowledge
the fragility cumulative distribution is more reli- can be considered to the reference level identi-
able as the size and heterogeneity of the sample fied by KL2. The confidence factor is then used
increases. in EC8 to reduce the capacity values as obtained
Figure 8a also shows that results are highly from the assessment.
affected by data collection and data analysis: the The FaMIVE procedure uses a measure of
three cumulative distributions are obtained for uncertainty of the input data to determine the
the same sample of buildings when the presence uncertainty of the output. Depending on whether
of anchors or good wall connection is in turn data, in each section of the data collection form,
ignored or considered. Because the collection has been collected and measured directly
and interpretation of this type of data is very on-site, or collected on-site and confirmed by
sensitive to the expertise of the surveyor and the existing drawings or photograph, or collected
analyst, results can be highly affected by either from photographic evidence only, three levels
incorrect interpretation of this information on the of uncertainty are considered, as high, medium,
basis of the survey or use of a modelling tool that and low, respectively, to which three confidence
cannot properly simulate the behavior of the ranges of the value given for a parameter can be
buildings when these constraints are in place. considered corresponding to 10 % variation,
The requirement of the codes and the limita- 20 % variation, and 30 % variation. The param-
tions imposed by conservation principles also eter value attributed during the survey is consid-
affect the way in which the fragility functions ered central to the confidence range so that the
can be derived and used. interval of existence of each parameter is defined
The reliability of the results obtained in the as m 5 %, m 10 %, and m 15 %, depending
previous section can be considered within the on highest or lowest uncertainty. The uncer-
framework set out in the EC 8 (CEN 2005), tainty applied to the output parameters, specifi-
whereby the reliability associated to the results cally lateral acceleration and limit states’
of a seismic assessment of a structure is displacement, is calculated as a weighted aver-
expressed as a function of the level of knowledge age of the uncertainty of each section of the data
and quantified by means of the confidence fac- form, with minimum 5 % confidence range to
tor. Hence this can be considered a measure of maximum 15 % confidence range.
the epistemic uncertainty. EC 8 (CEN 2005) In Fig. 9 the cumulative distribution for drift at
recognizes three levels of knowledge (limited, the structural damage state threshold is compared
normal, and full) and three fields of knowledge with the curve obtained assuming a confidence
(geometry, construction details, and materials). factor of 1.25 in agreement with EC8 knowledge
As data used in the FaMIVE approach are col- level and with the confidence boundaries com-
lected by on-site visual inspection with some puted by FaMIVE according to the procedure
measurement and in situ accurate observation outlined above.
of construction details, while only very limited The major limitation of using confidence fac-
in situ nondestructive test on materials are tors in the way suggested by EC8 is the possible
performed and material characteristics are oth- underestimation of the overall capacity of the
erwise assigned based on literature or surveyor sample and hence an appropriate representation
experience, then the level of knowledge is of epistemic uncertainty.
Seismic Vulnerability 100

Assessment: Masonry
90
Structures,
Fig. 9 Cumulative 80
distribution for drift at 70
structural damage distribution
threshold: comparison with 60
curve obtained for an EC8 50
distribution /CF =1.25
confidence factor of 1.25 40
and with FaMIVE
confidence boundary 30 distribution reliability
low bound
20
distribution reliability
10 upper bound
0
0.1 1 10
drift (%)
FaMIVE can also be used to foresee the shift behavior. This approach has serious limitations
in seismic performance of the building stock of a and it is nonconservative, as it overlooks out-of-
historic center if certain interventions are plane behavior which might occur for structures
implemented to improve the seismic behavior. with insufficient out-of-plane wall capacity.
Among possible interventions, the improvement Nevertheless, recommendations of the EC
of cohesion of the masonry through grouting of 8 (CEN 2005) only consider basic in-plane
multi-leaf walls with lime-based grout and the behavior, either controlled by flexure or con-
introduction of ties connecting walls together trolled by shear, and set limit drifts for this
are seen as meeting the conservation principles performance criterion. The application of the
of like for like, minimum intervention, and, in FaMIVE procedure to a number of cases in
the case of ties, reversibility. Figure 10a, b regions of different seismicities shows the rele-
shows the change in the median capacity curves vance of construction details and the importance
when these interventions are considered: it can of considering separately different failure
be seen that the response is different depending mechanisms to identify realistic risks of failure.
on the typology; in general both interventions A procedure for the computation of the perfor-
cause an increase in lateral capacity. In some mance point given performance criteria and
cases also a change in ductility and/or in the hazard return period is introduced to show the
behavior up to structural damage as the interven- variability of performance points for different
tion causes a change in the prevalent failure failure behaviors and how they affect probabil- S
mechanisms. ity of failure computed as cumulative. The ran-
dom uncertainty associated with this type of
analysis is presented in terms of fragility curves
Summary that can be modelled as lognormal distribution.
In the same way, the epistemic uncertainty
Traditionally seismic vulnerability of historic associated with the modelling procedure is also
buildings has been evaluated using empirical evaluated using the same approach. Results
methods based either on damage observation show that the reliability approach set out in
or construction data. In the last decade, analyt- FaMIVE correctly accounts for epistemic
ical methods for assessment of masonry struc- uncertainty and it is of the same order of mag-
tures have increased in number and quality. nitude as the confidence factor suggested by
However, most of them assume only in-plane EC8, although this only accounts for negative
failure behavior reducing the walls to frame-like effects.
Assessment: Masonry
Structures,
Fig. 10 Comparison
between the capacity
curves for (DS2), (DS4),
(UFB3), and (RS4) for
(a) the condition of
cohesion, c = 0 and
c = 0.25 N/mm2, and
(b) with the addition of
ties to the condition of
cohesion c = 0.25 N/mm2
Cross-References ▶ Seismic Vulnerability Assessment: Reinforced

Concrete Structures
▶ Analytic Fragility and Limit States [P(EDP| ▶ Strengthening Techniques: Masonry and
IM)]: Nonlinear Static Procedures Heritage Structures
▶ Empirical Fragility
▶ Masonry Structures: Overview
▶ Performance-Based Design Procedure for
Structures with Magneto-Rheological References
Dampers
Askan A, Yucemen MS (2010) Probabilistic methods for
▶ Seismic Analysis of Masonry Buildings: the estimation of potential seismic damage: applica-
Numerical Modeling tion to reinforced concrete buildings in Turkey. Struct
▶ Seismic Loss Assessment Saf 32(4):262–271
Baggio C, Bernardini A, Colozza R, Corazza L, Della Fajfar P (1999) Capacity spectrum method based on
Bella M, Di Pasquale G, Dolce M, Goretti A, inelastic demand spectra. Earthq Eng Struct Dyn
Martinelli A (2009) Manuale per la compilazione 28(9):979–993
della scheda di 1 livello di rilevamento danno, pronto Fajfar P, Gašperšič P (1996) The N2 method for the
intervento e agibilità per edifici ordinari seismic damage analysis of R.C. buildings. Earthq
nell’emergenza post-sismica (AeDES). Editrice Eng Struct Dyn 25(1):31–46
Italiani nel Mondo srl, Roma Federal Emergency Management Agency (FEMA)
Barbat AH, Yépez Moya F, Canas JA (1996) Damage (2001) HAZUS99 technical manual. Service Release
scenarios simulation for seismic risk assessment in 2. FEMA, Washington, DC
urban zones. Earthq Spectra 12(3):371–394 Giovinazzi S, Lagomarsino S (2004) A macroseismic
Benedetti D, Benzoni G, Parisi MA (1988) Seismic vul- method for the vulnerability assessment of buildings.
nerability and risk evaluation for old urban nuclei. In: Proceedings of the 13th world conference on earth-
Earthq Eng Struct Dyn 16:183–201 quake engineering, Vancouver, Paper No. 896
Bernardini A, Gori R, Modena C (1990) An application of Gr€unthal G (ed) (1998) Cahiers du Centre Européen de
coupled analytical models and experiential knowledge Géodynamique et de Séismologie: volume 15 – Euro-
for seismic vulnerability analyses of masonry buildings. pean macroseismic scale 1998. European Center for
In: Koridze A (ed) Engineering aspects of earthquake Geodynamics and Seismology, Luxembourg
phenomena, vol 3. Omega Scientific, Oxon, pp 161–180 Jaiswal K, Wald D, D’Ayala D (2011) Developing empir-
Borzi B, Crowley H, Pinho R (2008) Simplified pushover- ical collapse fragility functions for global building
based earthquake loss assessment (SP-BELA) method types. Earthq Spectra 27:775
for masonry buildings. Int J Architect Herit Kappos AJ, Stylianidis KC, Pitilakis K (1998) Develop-
2(4):353–376 ment of seismic risk scenarios based on a hybrid
Braga F, Dolce M, Liberatore D (1982) A statistical study method of vulnerability assessment. Nat Hazards
on damaged buildings and ensuing review of the MSK 17(2):177–192
76 SCALE. In: Proceedings of the 7th European Kappos AJ, Panagopoulos G, Penelis GG (2008) Develop-
conference on earthquake engineering, Athens ment of a seismic damage and loss scenario for con-
CEN – EN 1998-3 (2005) Eurocode 8: design of structures temporary and historical buildings in Thessaloniki,
for earthquake resistance, Part 3: strengthening and Greece. Soil Dyn Earthq Eng 28(10–11):836–850
repair of buildings. Comitee’ Europeen de Lagomarsino S, Giovinazzi S (2006) Macroseismic and
Normalisation, Brussels mechanical models for the vulnerability and damage
Corsanego A, Petrini V (1990) Seismic vulnerability of assessment of current buildings. Bull Earthq Eng
buildings – work in progress. In: Proceedings of the 4(4):415–443
workshop II on seismic risk vulnerability and risk Lang K, Bachmann H (2004) On the seismic vulnerability
assessment, Trieste, pp 577–598 of existing buildings: a case study of the city of Basel.
D’Ayala D (2005) Force and displacement based vulner- Earthq Spectra 20(1):43–66
ability assessment for traditional buildings. Bull Miranda E (2001) Estimation of inelastic deformation
Earthq Eng 3:235–265, Springer demands of SDOF systems. J Struct Eng (ASCE)
D’Ayala D, Ansal A (2012) Nonlinear push over assess- 127(9):1005–1012
ment of heritage buildings in Istanbul to define seismic Novelli V, D’Ayala DF (2014) The PERPETUATE pro-
risk. Bull Earthq Eng 10(1):285–306 cedure for the seismic vulnerability assessment at ter-
D’Ayala D, Paganoni S (2011) Assessment and analysis of ritorial scale. Application to the Casbah of Algiers.
damage in L’Aquila historic city centre after 6th April Bull Earthq Eng 12:1–24
2009. Bull Earthq Eng 9(1):81–104 Spence R, Coburn AW, Pomonis A (1992) Correlation of
D’Ayala D, Speranza E (2003) Definition of collapse ground motion with building damage: the definition of S
mechanisms and seismic vulnerability of historic a new damage-based seismic intensity scale. In: Pro-
masonry buildings. Earthq Spectra 19:479–509 ceedings of the tenth world conference on earthquake
Di Pasquale G, Orsini G, Romeo RW (2005) New devel- engineering, Madrid, vol 1, pp 551–556
opments in seismic risk assessment in Italy. Bull Tomaževič M (2007) Damage as a measure for
Earthq Eng 3(1):101–128 earthquake-resistant design of masonry structures:
Dolsek M, Fajfar P (2004) Inelastic spectra for infilled Slovenian experience. Can J Civil Eng
reinforced concrete frames. Earthq Eng Struct Dyn 34(11):1403–1412
33:1395–1416 UNDP/UNIDO Project RER/79/015 (1985) Post-
Elenas A (2000) Correlation between seismic acceleration earthquake damage evaluation and strength assess-
parameters and overall structural damage indices of ment of buildings under seismic condition,
buildings. Soil Dyn Earthq Eng 20(1–4):93–100 vol 4. UNDP, Vienna
Erberik MA (2008) Generation of fragility curves for Vamvatsikos D, Cornell C (2005) Direct estimation of
Turkish masonry buildings considering in-plane fail- seismic demand and capacity of multidegree-of-
ure modes. Earthq Eng Struct Dyn 37:387–405 freedom systems through incremental dynamic
3182 Seismic Vulnerability Assessment: Reinforced Concrete Structures
analysis of single degree of freedom approximation. (ii) vulnerability, and (iii) exposure, and they
J Struct Eng 131(4):589–599 are based on the total probability theorem.
Whitman RV, Reed JW, Hong ST (1973) Earthquake
damage probability matrices. In: Proceedings of the Available seismic vulnerability methods,
5th world conference on earthquake engineering, i.e., empirical, analytical, or hybrid (e.g., Calvi
Rome, Italy, pp. 2531–2540 et al. 2006), developed in the last 30 years,
Wu Y-M, Teng T-l et al (2003) Relationship between peak differ because of the nature of tools and data
ground acceleration, peak ground velocity, and inten-
sity in Taiwan. Bull Seismol Soc Am 93(1):386–396 employed. In the following only analytical
Zobin VM, Cruz-Bravo AA, Ventura-Ramırez F (2010) approach to seismic vulnerability is considered,
Microzonation of seismic risk in a low-rise Latin given the significance that such approach had in
American city, based on the macroseismic evaluation the last years. Analytical vulnerability methods
of the vulnerability of residential buildings: Colima
city, Mexico. Nat Hazard Earthq Syst Sci stand at the basis of current worldwide codes and
10:1347–1358 guidelines for seismic design and assessment of
structures.
Therefore, next-generation codes (e.g., FEMA
P-58-1 2012) are proceeding toward an explicit
quantification of seismic risk through the
Seismic Vulnerability Assessment: Performance-Based Earthquake Engineering
Reinforced Concrete Structures (PBEE) framework. Section “Performance Earth-
quake Engineering Framework” of this entry
Flavia De Luca1 and Gerardo M. Verderame2 describes how in PBEE seismic risk assessment
1
Department of Civil Engineering, University of is deconstructed through the total probability the-
Bristol, Bristol, UK orem. PBEE’s deconstruction of seismic risk in
2
Department of Structures for Engineering and hazard, structural, damage, and loss analysis
Architecture (DiSt), University of Naples emphasizes how structural and damage analyses
Federico II, Naples, Italy are strictly related to vulnerability assessment in
strict sense.
Notwithstanding the fact that PBEE is a gen-
Synonyms eralized methodology that can be applied for any
kind of structure, the case of reinforced concrete
Axial load failure; Damage measure; Flexure; (RC) structures is specifically addressed in the
Nonlinear modeling; PBEE; RC structures; Shear following. As far as the relationship between
intensity measure and engineering demand
parameter arises in PBEE, it is the phase in
Introduction which it is necessary to focus on the nonlinear
behavior of RC structural elements (section
Seismic vulnerability can be defined as the “Behavior of RC Elements and Non Structural
degree of loss to a given element at risk (e.g., Elements”), in order to approach analytical
buildings) resulting from the occurrence of an modeling (section “Analytical Modeling of RC
earthquake event (Coburn and Spence 2002). Structures”), and qualify and quantify compo-
The development of methodologies for seismic nents damage measures (section “Damage Mea-
vulnerability assessment is an essential tool for sures”). A review of experimental and analytical
seismic risk management and for prioritizing results collected in the last decades is provided
pre-earthquake strengthening of the built herein. It aims at the quantification of the behav-
environment. ior of RC structures and, in turn, of RC elements.
Seismic risk can be approached through Most significant results on vulnerability assess-
different methodologies. Most of them aim ment of RC structures are collected together, and
to deconstruct the problem into the typical ele- they allow emphasizing future challenges
ments defining any kind of risk: (i) hazard, and needs (section “Future Challenges and
Seismic Vulnerability Assessment: Reinforced Concrete Structures 3183
Concluding Remarks”) that scientific community this PBEE consolidated framework, thus aiming
and practitioners should take on in the next years. at the final challenge to contribute effectively to
The focus in the following is the quantification the reduction of losses and the improvement of
of nonlinear parameters of RC elements aimed at safety (Cornell and Krawinkler 2000).
a reliable modeling of their behavior. Emphasis is The basis of performance assessment can
also given to consolidated and more recent liter- be summarized in a single equation (see Eq. 1).
ature results that have been included in codes and This equation suggests a generic structure for
guidelines. coordinating, combining, and assessing the
many considerations implicit in performance-
based seismic assessment.
Performance Earthquake Engineering
Framework ððð
lðDV Þ ¼ j GhDVjDMidG hDMjEDPidG
“PBEE implies design, evaluation, construction,
hEDPjIMidlðIMÞj
monitoring the function and the maintenance of
engineered facilities whose performance under (1)
common and extreme loads respond to the
diverse needs and objectives of owners-users Based on total probability theorem, Eq. 1
and society. It is based on the premise that per- allows deconstructing the problem in four steps:
formance can be predicted and evaluated with (i) hazard analysis, (ii) structural analysis, (iii)
quantifiable confidence to make, together with damage analysis, and (iv) loss analysis. Each
the client, intelligent and informed trade-offs step carries out a specific generalized variable:
based on life cycle considerations rather than intensity measure (IM), engineering demand
construction costs alone” (Krawinkler and parameter (EDP), damage measure (DM), and
Miranda 2004). Guidelines and codes, since the decision variable (DV). The key issue of PBEE
1990s, have partially implemented, in their gen- methodology is to identify and quantify DV of
eral framework, the basic concepts of PBEE in primary interest to the decision makers with con-
various forms, resulting in the attempt to tie sideration to all important uncertainties. DVs
design criteria to a performance level, usually have been defined in terms of different quantities,
that of collapse prevention. The Pacific Earth- such as repair costs, downtime, and casualty
quake Engineering Research (PEER) Center has rates.
focused for several years on the development of First member of Eq. 1 is a probabilistic descrip-
procedures, knowledge, and tools for a compre- tion of the DV, such as the mean annual frequency
hensive seismic performance assessment of of the DV exceeding a specified value. l(DV)
buildings and bridges. The efforts made have might be the mean annual frequency of the direct S
resulted in a general framework that is now dollar loss (repair cost) exceeding 50 % of the
shared by the earthquake engineering community replacement cost of the facility. The terms dG
all over the world. Different publications (EDP|IM), dG(DM|EDP), and G(DV|DM) or
describe the basic steps of PBEE methodology their derivatives on the right side of Eq. 1 are
(e.g., Cornell and Krawinkler 2000; Krawinkler conditional probabilities relating one quantity to
2002; Moehle 2003; Porter 2003; Deierlein another, while dl(IM) is the derivative of the haz-
et al. 2003; Krawinkler and Miranda 2004; ard curve, relating ground motion intensity mea-
Moehle and Deierlein 2004), and a comprehen- sure to its mean annual frequency of exceedance.
sive document for a punctual methodological Aimed at facilitating probability calculation,
description of last enhancements of PBEE is but mostly aimed at compartmentalizing
available in FEMA P-58-1 document (2012). discipline-specific knowledge, the above PBEE
Sharing a common vision and approach allows framework should choose intermediate variables
placing further enhancements and progresses in (EDP and DM), so that conditional probabilities
are independent of one another (Deierlein Notwithstanding the fact that hazard analysis is
et al. 2003). A brief intro to each step of PBEE a common step for each structural typology to
is provided herein. It is a general methodological which PBEE methodology is applied, it is worth
framework description that does not refer yet noting that it is also the step in which most of the
specifically to RC structures. variability of the whole problem is considered. The
so-called record-to-record variability is the source
Hazard Analysis of the most significant variability in risk evaluation
Hazard analysis allows describing earthquake (e.g., Goulet et al. 2007). On the other hand, such
hazard in a probabilistic manner. The result of source of variability is also a common aspect for
this phase is a hazard curve, which shows the each structural typology, and it is not an aspect that
variation of the selected IM versus mean annual strictly characterizes RC more than other structural
frequency of exceedance (MAF). Probabilistic typologies (e.g., steel, masonry, etc.).
seismic hazard analysis, for a specific site, passes
through four main steps: (i) characterization of Structural Analysis
the seismic sources (e.g., identified faults or geo- Structural analysis is the phase of PBEE in which
graphical areas), in which it is generally assumed structural simulations are performed for the eval-
that the occurrence of an event at one source does uation of EDPs at given IMs. At this step, a
not affect the occurrence of events at other structural model is built up for the estimation of
sources; (ii) characterization of magnitude distri- EDPs, such as internal member forces and local
bution, generally obtained through a truncated or global deformations, including structural col-
exponential distribution or through characteristic lapse. The computational model is built up taking
magnitude distribution estimated for specific into account uncertainties in model parameters.
faults; (iii) ground motion estimation evaluated A proper nonlinear model of a structure is the
through prediction equation of the considered IM; result of a reliable analytical model of the ele-
and (iv) probability analysis for the evaluation of ments of the structure considered and of the anal-
the hazard curve (see also McGuire 2004). Haz- ysis of experimental behavior of the elements.
ard analysis can be performed with different The choice of EDPs depends on the perfor-
intensity measures, provided the appropriate pre- mance target and on the type of system of inter-
diction equation. Notwithstanding the fact that est. The general approaches for the evaluation of
advanced intensity measures have been consid- dG(EDP|IM) is to perform incremental dynamic
ered in literature (e.g., Tothong and Luco 2007) analyses or other kinds of nonlinear analyses; see
and prediction equations for these IMs are avail- also FEMA P-58-1 (2012).
able (e.g., Tothong and Cornell 2006; De Luca
et al. 2014a), still the employment of peak ground Damage Analysis
acceleration (PGA) and, more commonly, of Damage analysis is the third step of PBEE meth-
spectral acceleration at the fundamental period odology. It relates the EDPs to DMs. The DMs
of the structure, Sa(T), represents the basic include quantitative descriptions of damage to
practice. structural elements, nonstructural elements, and
Hazard analysis is generally performed for the contents. This quantification must be in sufficient
evaluation of a target spectrum finally aimed at detail to enable subsequent quantification of the
ground motion selection, if the assessment is necessary repairs, disruption of function, and
made through nonlinear dynamic analysis. It is safety hazards (Moehle and Deierlein 2004).
worth to note that according to FEMA P-58-1 The EDPs considered in structural analysis are
(2012), hazard analysis for vulnerability assess- the input to a set of fragility functions that model
ment and loss assessment can be performed with the probability of various level of physical dam-
three different approaches according to the final age, conditioned on structural response. Physical
aim of the study: intensity-based, scenario-based, damage is described at a detailed level, defined
and time-based. relative to particular repair efforts required to
restore the component to its undamaged state. elements can show; so, in the following, behavior
The current approach is a component-based of RC elements is described considering the case
approach (e.g., Aslani 2005; FEMA P-58-1 of existing elements that represent the most gen-
2012). The study of experimental behavior of eral situation in vulnerability assessment prob-
components is still a key issue for reliable evalu- lems. This section is primarily focused on the
ation of damage measures. vulnerability of RC frame structures; thus the
behavior of columns and beam–column joints is
Loss Analysis specifically addressed. At the end of this section,
The last step of a PBEE loss assessment is the behavior of masonry infills is also considered.
probabilistic estimation of performances through Masonry infills are considered because of the
DVs. DVs represent the outcome parameters of significant effect they can have on vulnerability
PBEE and allow a transformation of engineering of RC frame structures. In fact, they provide on
evaluations in terms of variables of interest for one hand the strength and stiffness increase and
stakeholders (e.g., dollars, deaths, downtime); on the other hand, the local interaction between
see Porter (2003). Different studies are available infill and RC frame.
in literature for the evaluation of direct and indi-
rect losses (e.g., Mitrani-Reiser 2007), and results RC Members
of such studies are now the basis of the loss The specification of nonlinear structural component
assessment framework provided by FEMA models in the form of monotonic backbone curves
P-58-1 (2012). It is worth noting that loss analysis and hysteretic rules for RC structures is one of the
works on repair costs and downtime that, in turn, key targets for the earthquake engineering commu-
are strictly related to the social and economical nity worldwide. Monotonic backbone curves and
environment in which they are evaluated. hysteretic rules should be able to capture the exper-
imental behavior of RC elements. Any RC element
component model is defined as function of specific
Behavior of RC Elements and parameters, such as shear span ratio, materials’
Nonstructural Elements strengths, longitudinal and transversal reinforce-
ment ratios, and axial load ratio (e.g., CEB 1996;
The basis of a reliable vulnerability analysis for Elwood 2004; Elwood and Moehle 2005a, b; Zhu
any structural material passes through a solid et al. 2007; Haselton et al. 2008; Biskinis and Fardis
modeling approach. PBEE assessment framework 2010a, b). Specific combinations of the above
naturally points to nonlinear dynamic analysis as ruling parameters end up to the different definitions
the reference structural analysis tool for RC struc- of wall, beam, or column. Furthermore, the struc-
tures. Thus, it is necessary to consider reliable tural difference between walls, beams, and columns
nonlinear hysteretic models of all components of is also made by the structural typology in which S
RC structures. Structural modeling of RC ele- the above prismatic RC members are included,
ments is based on the observation and interpreta- and by the structural role they play with respect to
tion of experimental tests and post-earthquake loads they are subjected to (e.g., gravity loads,
evidence from which it is possible to carry out earthquake loads, etc.).
nonlinear models of element behavior. It is impor- In particular a prismatic RC member that
tant to state that a significant difference has to be carries gravity loads through shear and bending
made between elements with and without seismic and that is designed under the assumption of
detailing. In fact, existing elements are susceptible uniaxial load with zero axial force is referred as
of various modes of failure, while capacity design a beam in RC frame structures. Again, a prismatic
in seismic detailed elements prevents brittle RC member in which the ratio between the two
modes of failure. nondominant dimensions (i.e., width and depth of
Seismic vulnerability assessment has to the cross section) is higher than four is typically
account for all typical modes of failure that RC defined as wall. This latter geometrical
assumption implies that a “wall” resists lateral especially for inelastic cyclic loading for
forces mainly in one direction, and it can be non-capacity-designed elements. Flexural
designed for such unidirectional resistance by response can be affected by the interaction with
assigning flexural resistance to the two far ends shear also in linear elastic phase, when plastic
of the section and shear resistance to the web shear capacity of the element is not attained. Con-
(Fardis 2009). The classification of an RC pris- versely, in the case of inelastic response, shear
matic elements as “column” is the more general capacity can decrease after yielding as a result of
possible; it covers the widest range of parameters increasing ductility demand in the element. In fact,
ruling RC component modeling (e.g., variable damage caused by cyclic degradation in flexural
axial load ratio, biaxial load, variable reinforce- response decreases post-cracking shear capacity
ment ratios). Therefore, in the following, specific mechanisms (e.g., aggregate interlock, loss of
attention is given to modeling issues related to anchorage of transverse reinforcement, etc.)
RC columns rather than beams or walls. resulting in a decrease of shear capacity.
Notwithstanding the fact that columns are In Fig. 1, two experimental tests on rectangu-
characterized by biaxial loading and by the wid- lar RC columns from PEER database by Berry
est variety of ruling parameters, experimental et al. (2004) are shown. The first column, Fig. 1a,
data more commonly available address the case is characterized by flexural-dominated behavior,
of typical columns in RC frame structures. More while the second column, Fig. 1b, is character-
specifically, most test specimens have rectangu- ized by post-yielding interaction between flexure
lar cross section, symmetric reinforcement, and and shear.
nonzero axial load, and rarely they are character- The above observations allow indentifying
ized by biaxial loading. The latter is the reason three different failure modes for columns
why most reliable nonlinear structural compo- resulting from flexure–shear interaction, as
nent models are conceived for columns under shown, schematically, in Fig. 2. The analytical
uniaxial loading. It is worth noting that recent classification of different modes of failure can be
experimental and modeling efforts have been made only considering degrading shear capacity
made for the characterization of biaxial compo- models which account for shear capacity degra-
nent models (e.g., Bousias 1993; Bousias dation after yielding.
et al. 2002; Di Ludovico et al. 2013), and some
guidelines now provide recommendations for 1. Shear failure mode, S – initial shear capacity
their employments. On the other hand, biaxial (V0), i.e., not affected by cyclic degradation, is
modeling efforts are still in a preliminary phase, lower than the plastic shear capacity (Vp), and
and most common PBEE nonlinear analysis flexural response is modified by a preemptive
applications are performed discarding biaxial shear failure in the elastic response phase of
nonlinear component models. Thus, in the fol- the element. Deformation capacity (i.e., drift,
lowing, the only uniaxial behavior of elements chord rotation) is very limited. Subsequent
is described. response of the element is characterized by a
strictly degrading behavior with significant
Failure Mode Classification of RC Elements shear strength reduction and a consequently
Columns are generally subjected to the contempo- loss of axial load-carrying capacity; see
rary presence of axial load, shear forces, and bend- Fig. 2a.
ing moment. The interaction between those 2. Flexure–shear failure mode, FS – V0 is higher
internal forces strictly affects the response of than Vp, and flexural response is characterized
structural elements and consequently their mode by an inelastic phase. However, if degraded
of failure. Interaction effects, i.e., the influence of shear strength (Vd), the lower red dotted
shear forces on flexural deformations and that of branch in Fig. 1, is lower than Vp, the inelastic
normal stress resultants on shear deformations, are flexural response is modified by a shear failure
important. In general, they cannot be neglected, that occurs at an intermediate shear strength
a 300 b 400
Experimental Results Experimental Results
300
200
200
Shear Force (kN)
Shear Force (kN)

100
100
0 0
−100
−100
−200
−200
−300
−300 −400
−0.1 −0.05 0 0.05 0.1 −0.1 −0.05 0 0.05 0.1
Column Drift (displacement/height) Column Drift (displacement/height)
Seismic Vulnerability Assessment: Reinforced Con- and Paultre (2000) and (b) Pujol (2002) (Adapted from
crete Structures, Fig. 1 Experimental results for Berry Haselton et al. 2008)
et al. (2004) database. Experimental tests by (a) Legeron
a b
V V
V0 shear capacity curve

Vp flexural response Vp flexural response
Vy Vy Vd
V0
shear capacity curve
Vcr Vd Vcr
δ s δy δa δf δ δy δs δa δf
c V0
V
shear capacity curve
Vd
Vp
flexural response
Vy
S
Vcr
δy δf δ
Seismic Vulnerability Assessment: Reinforced Con- moment: (a) shear failure mode, S; (b) flexure–shear fail-
crete Structures, Fig. 2 Collapse modes of RC ele- ure mode, FS; (c) flexural failure mode, F
ments subjected to axial load, shear, and bending
between V0 and Vd. The corresponding defor- degrading up to the loss of axial load-carrying
mation capacity is higher with respect to that capacity; see Figs. 1b and 2b.
of case (1). The behavior of the element 3. Flexural failure mode, F – Vd is higher than Vp,
succeeding this kind of shear failure is strictly and flexural response is not affected by
interaction with shear. Thus, the element is yielding steel strength (fy), and section depth (d),
characterized by inelastic deformation domi- computed as distance from the extreme compres-
nated by flexure and its typical damage, such sion fiber to centroid of tension reinforcement.
as local buckling of longitudinal reinforce- This model takes into account shear strength deg-
ment and concrete spalling. This damage typ- radation after yielding through a degradation
ically corresponds to the ultimate flexural coefficient (k); k is defined to be equal to 1.0 for
capacity (yf). The softening branch of shear displacement ductility less than 2, to be equal to
force–displacement envelope ends up to the 0.7 for displacement ductility exceeding 6, and to
loss of vertical load-carrying capacity; see vary linearly for intermediate displacement duc-
Figs. 1a and 2c. This is the failure mode typ- tilities. The degradation coefficient k is applied to
ical of capacity-designed elements in which both Vc and Vs, under the assumption that con-
shear–flexure hierarchy is controlled during crete component degrades due to increased crack-
the design process (see also De Luca and ing and degradation of the aggregate interlock
Verderame 2013). mechanism, while steel component likewise
degrades due to hoop opening and bond degrada-
The classification of failure mode is made on tion. This shear capacity model is adopted in
the basis of the ratio between plastic shear capac- ASCE/SEI 41 guidelines (2007).
ity Vp and degrading shear strength Vn. Several It is worth noting that the classification of
models have been developed to represent the column failure modes based only on shear
degradation of shear strength with increasing strength and Vp can be still not adequate. Other
inelastic deformations (Watanabe and Ichinose column parameters may also influence the
1992; Aschheim and Moehle 1992; Priestley observed failure mode (Zhu et al. 2007). These
et al. 1994; Biskinis et al. 2004; Sezen and considerations are reflected by ASCE/SEI 41 –
Moehle 2004). supplement 1 provisions (Elwood et al. 2007) as
In particular, the shear-degrading model pro- follows: the expected failure is first evaluated
posed by Sezen and Moehle (2004) estimates based on the ratio between Vp and the shear
column shear strength as the summation of capacity, and then, when classifying a column
shear carried by concrete (Vc) and shear carried as flexure-controlled, further parameters are con-
by transverse reinforcement through a 45 truss sidered, namely, the transverse reinforcement
model (Vs). This shear strength model is cali- ratio, assuming 0.002 as the lower limit, as in
brated on 51 experimental column tests, and it Zhu et al. (2007), and the tie spacing-to-section
has the form shown in Eq. 2. depth ratio, assuming 0.5 as the upper limit.
Characteristic points of shear force–dis-
V n ¼ k V o ¼ k ðV c þ V w Þ placement envelope (i.e., moment–chord rota-
" pffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
0:5 f c tion), for each failure mode, can be identified
P
¼k 1þ pffiffiffiffi schematically through the representation shown
LV =d 0:5 f c Ag
in Fig. 2. Monotonic and cyclic deformation
Asw f yw d capacities to be attributed to each characteristic
0:8Ag þ
s point can be found in literature. It is worth to note
(2) that if in the elastic phase the distinction between
monotonic and cyclic deformation capacity can
Shear strength contribution carried by con- be unnecessary (pre-yielding phase), such dis-
crete depends on concrete compressive strength, tinction becomes significant in the inelastic
(fc); shear span, (LV); section depth, (d); axial phase (post-yielding), especially if those defor-
load, (P); and gross area section, (Ag), while mation capacities are employed in the analytical
shear strength contribution carried by reinforce- modeling of the element.
ments depends on transverse reinforcement area Monotonic deformation capacities are employed
(Asw) within a spacing (s) in the loading direction, for the definition of the force–deformation envelope
a b
Effective Force
Effective Force
Force-Displacement Force-Displacement
Envelope Envelope
F0.004
F0.004 Ffirstyield
0.8Fmax
Ffirstyield
Effective Stiffness Effective Stiffness
Δy Displacement Δy Displacement
Seismic Vulnerability Assessment: Reinforced Con- (a) yielding columns and (b) columns that did not yield
crete Structures, Fig. 3 Definition of yield displace- (From Elwood and Eberhard 2009)
ment and effective stiffness from test data for
of the element. Successively, it is necessary to on the assumption that yielding drift is the sum of
assign cyclic degrading rules. So, for example, a three different components: a flexural compo-
specific degradation percentage of peak force nent, a shear component, and a fixed end rotation
(conventional failure) is attained at a deformation component due to longitudinal bar slip; see Eq. 3.
capacity lower with respect to that considered on Alternatively yield drift can be evaluated indi-
the monotonic envelope. When cyclic degrading rectly from the estimation of yield stiffness
rules are not modeled, the conventional failure has (Elwood and Eberhard 2009; Biskinis and Fardis
to be fixed at a displacement capacity that implicitly 2010a).
accounts for it (i.e., a value on the monotonic enve-
lope that corresponds to the lower capacity dy ¼ dy, flex þ dy, shear þ dy, slip (3)
degraded by cyclic effects). It is worth noting that
the latter solution is obviously an approximated Yield drift obtained through Eq. 3 is generally
way for the estimation of a reliable deformation evaluated as the corner for an ideal envelope
capacity. Cyclic degradation depends on the load approximating a bilinear force deformation
path, and it cannot be accurately captured analyti- response of the element and not as the drift
cally with monotonic models. Still, for some spe- corresponding to the first yielding in the rein- S
cific failure modes, the monotonic degraded forcement steel (or equivalently in the concrete)
solution can be more straightforward with respect at the first section of the element. Figure 3 shows
to the calibration of fulfilling cyclic models. the procedure by Elwood and Eberhard (2009)
for the definition of yielding displacement on the
Yielding Displacement in RC Members envelope of measured lateral load–displacement
For the estimation of a reliable inelastic backbone relationship corrected for P-delta. Elwood and
of an RC element, the estimation of the yielding Eberhard’s procedure covers both the cases in
point is quite relevant. In literature, different which yield force is attained in the envelope
empirical or hybrid models for the estimation of (Fig. 3a) and cases in which the calculated
yielding drift capacity based on experimental yield force is not attained (Fig. 3b); in fact, it
observations are available. Direct estimations of provides an estimate of effective stiffness also in
yielding drift (Elwood and Eberhard 2009; the case of columns that do not yield (see
Biskinis and Fardis 2010a) are generally based Fig. 3b).
Eurocode 8 part 3 (CEN 2005) recommends a

three-component model for the estimation of
yielding drift, while ASCE/SEI 41 (2007) pro-
vides an effective stiffness to be computed as
function of the axial load to which the element
is subjected. Experimental databases on which
the above formulations are calibrated are often
characterized by modern code-conforming ele-
ments with proper seismic detailing. In the case
of nonconforming elements, such formulations
have still shown a fair agreement; emphasizing
that seismic detailing does not affect strictly
Seismic Vulnerability Assessment: Reinforced Con-
yielding capacity (Biskinis and Fardis 2010a). It crete Structures, Fig. 4 Conventional criterion for the
is worth noting that nonconforming definition evaluation of ultimate drift capacity independently of the
changes according to the specific code consid- failure mode observed
ered; on the other hand, this definition is gener-
ally referred to lack of detailing in transversal
reinforcements, smooth bar presence, and lack
shear- and flexure–shear-dominated failures.
of confinement (e.g., no 135 hooks, poor stirrup
Classification is made on the basis of three col-
spacing, etc.). Analogously, database of
umn parameters: ratio of plastic shear demand to
nonconforming elements with smooth bars have
strength ratio (Vp /Vn), the aspect ratio (a/d), and
shown a fair agreement with empirical formula-
the transverse reinforcement ratio (r00 ). The two
tions based on conforming elements with ribbed
sub-databases are employed for the estimation of
bars (e.g., Ricci et al. 2013).
the median ultimate drift capacity (20 % degra-
dation of maximum shear strength) in the two
Ultimate Drift in RC Members
cases, with characterization of logarithmic stan-
Other characteristic point of RC member
dard deviations. Equations 4 and 5 show the
force–displacement response is the ultimate drift
median estimates for Zone S and Zone
capacity for each failure mode considered. Ulti-
F ultimate drift capacity, respectively. Zone
mate drift capacity is generally evaluated at the
S ultimate drift is function of hoop spacing to
drift characterizing a 20 % loss of the maximum
depth ratio (s/d), aspect ratio (a/d), and normal-
shear strength (FIB 2003) attained in the element;
ized axial force (n). Zone F ultimate drift depends
see Fig. 4. According to the failure mode, differ-
on longitudinal reinforcement ratio (rl), mechan-
ent empirical ultimate drift capacity formulations
ical transversal reinforcement ratio (r00 fyt/fc), s/d,
are available in the literature. Most experimental
and n.
databases are characterized by cyclic tests and
few monotonic tests. Thus, most of those empir- s a
ical formulations account implicitly for cyclic ds ¼ 2:02r00 0:025 þ 0:013 0:031n (4)
d d
degradation.
Zhu et al. (2007) provide an ultimate drift r00 fyt
capacity model for two collapse modes. They df ¼ 0:049 þ 0:716rl þ 0:120
fc (5)
considere a database of 125 nonconforming col- s
umns. Tests are divided in two different 0:042 0:070n
d
sub-databases according to a binary classification
approach: Zone S columns (shear-dominated fail- An alternative solution for the characteriza-
ures) and Zone F columns (flexural-dominated tion of ultimate drift capacity for flexure–shear
failures). Zone F is composed of 85 tests, while mode of failure (i.e., after yielding) is provided
Zone S is composed of 40 tests that include both by Elwood and Moehle (2005a) based on a
a 1.2 b 1.5
Ks Mc Non-Deteriorated Backbone
Normalized Moment (M/My)
Normalized Moment (M/My)

1 My 1
θcaptot
0.8 0.5
θcapPI θpc
0.6 0
Ke
0.4 −0.5
Kc
0.2 −1
θy θcap
0 −1.5
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 −8 −6 −4 −2 0 2 4 6 8
Chord Rotation (radians) Chord Rotation (radians)
Seismic Vulnerability Assessment: Reinforced Concrete Structures, Fig. 5 (a) Monotonic and (b) cyclic
behavior of component model used in the calibration study by Haselton et al. (2008) (From Haselton et al. 2008)

database of 51 lightly transversal-reinforced col- maxð0:01; o0 Þ 0:225
umns (Sezen and Moehle 2004); see Eq. 6. This df ¼ 0:016 ð0:3 Þn
fc
maxð0:01; oÞ
ultimate drift depends on the maximum nominal
(8)
shear stress (u), r00 , and n. LV 0:35 arwf fyw 100r
min 9; 25 c 1:25 d
h
3 1 u 1 1
ds ¼ þ 4r00 pffi n (6) Considering a database of 48 cyclic tests,
100 40 fc 40 100
Biskinis and Fardis (2010b) provide a correction
For flexural-dominated behavior, an empirical factor to Eq. 8 that accounts for poor detailing
ultimate drift capacity is provided by Biskinis (i.e., older columns), and it is equal to (1/1.20).
and Fardis (2010b). It represents the evolution For substandard elements with smooth bars
of the first formulation provided by Panagiotakos (based on 31 tests), the same authors provide
and Fardis (2001). This formulation is based on a an additional correction factor, equal to 0.95, to
database of 1,352 uniaxial columns with good be applied to Eq. 8 (see also Verderame et al.
detailing and continuous bars (299 monotonic 2010).
and 1,053 cyclic tests). According to the loading
of the test, Fardis and his coauthors provide two Flexural Component Model
empirical formulations for ultimate drift capac- For flexural-dominated behavior, Haselton
ity, one for monotonic loading (see Eq. 7) and one et al. (2008) provide a complete calibration
for cyclic loading (see Eq. 8). Biskinis and based on Berry et al. (2004) database for the S
Fardis’ formulations depend on n, fc, shear span specification of a complete nonlinear structural
ratio (Lv/h), longitudinal mechanical reinforce- component model based on the beam–column
ment ratio in compression and tension (o0 and element model developed by Ibarra et al.
o, respectively), confinement effectiveness fac- (2005). Figure 5 shows the monotonic and cyclic
tor (a), transversal reinforcement ratio (rw or r00 ), behavior of the component model employed by
transversal reinforcement yielding strength (fyw Haselton et al. (2008). Parameters to be consid-
or fyt), and diagonal reinforcement ratio (rd). ered for the employment of Haselton et al. (2008)
model are (1) My, yielding moment; (2) yy, yield-
ing rotation; (3) ycap, monotonic chord rotation at
maxð0:01; o0 Þ 0:225
df ¼ 0:028 ð0:3n Þ fc onset of strength loss (i.e., capping); (4) Ks, hard-
maxð0:01; oÞ

(7) ening stiffness; (5) ypc or Kc, post-capping stiff-
LV 0:35 arwf fyw ness; (6) l, normalized hysteretic energy
min 9; 25 c 1:25 100rd
h degradation capacity (i.e., cyclic degradation
a 300 b 400
Experimental Results Experimental Results
Haselton model 300 Haselton model
200
200
Shear Force (kN)

Shear Force (kN)
100
100
0 0
−100
−100
−200
−200
−300
−300 −400
−0.1 −0.05 0 0.05 0.1 −0.1 −0.05 0 0.05 0.1
Column Drift (displacement/height) Column Drift (displacement/height)
Seismic Vulnerability Assessment: Reinforced Con- characterized by (a) flexural failure mode and (b)
crete Structures, Fig. 6 Example of Haselton et al.’s flexure–shear failure mode (Adapted from Haselton
calibration for test specimens 193 and 212 (see Fig. 1) et al. 2008)
parameter); and (7) c, exponent term to model attainment of ultimate deformation, a loss of ver-
rate of deterioration (i.e., cyclic parameter). This tical carrying capacity. In particular, Elwood
model accounts for cyclic degradation directly and Moehle (2005b), on the basis of 12 tests
through the modeling of hysteretic behavior. It characterized by flexure–shear mode of
is significantly different with respect to the other failure, observed that the drift ratio at axial
models considered and also more complete. On load failure ðda Þ of a shear-damaged column is
the other hand, drift capacity thresholds that inversely proportional to the magnitude of the
implicitly account for cyclic degradation cannot axial load and directly proportional to the amount
be evaluated on the monotonic backbone proof transverse reinforcement. Based on the classi-
vided by this model. In fact, the monotonic back- cal shear friction model, Elwood and Moehle
bone by Haselton without the degrading (2005b) propose a model for the drift at axial
parameters l and c is incomplete and failure for shear-damaged columns. The axial
nonconservative for RC elements capacity eval- load on a shear-damaged column is assumed to
uation through static methods. be supported by a combination of compression of
It is worth to note that Haselton et al. make the longitudinal reinforcement and force transfer
a calibration effort for most of columns of through shear friction on an idealized shear fail-
Berry et al. database considering both columns ure plane (see Fig. 7).
that failed in flexural mode and columns charac- The effective coefficient of friction from the
terized by flexure–shear mode. In Fig. 6, an classical shear friction equation is related to the
example of the calibration made by Haselton drift ratio at axial failure using the results from
et al. (2008) is shown for the two experimental 12 full-scale pseudo-static column tests. The
test specimens provided in Fig. 1. The compar- effective coefficient of friction along the critical
ison of the two examples of calibration empha- shear failure plane can be calculated using equi-
sizes that this component model is very well librium and subsequently related to the drift ratio
fitted for columns characterized by flexural at axial failure. Elwood and Moehle found that
behavior and less accurate for flexure–shear- the effective coefficient of friction calculated by
dominated columns. ignoring the contribution of longitudinal rein-
forcement provided good agreement with the
Loss of Vertical Carrying Capacity experimental data. According to the above obser-
Some of the experimental tests characterized by vations, the empirical formulation provided in
shear-dominated failure have shown, after the Eq. 9 is proposed for the evaluation of drift at
P based on the same model, but the most recent

study (i.e., Eq. 10 by Zhu et al.) is based on a
M
V larger database, and it also provides a coefficient
of variation equal to 0.35 for the median estimate
of Eq. 10.
dc
da ¼ 0:184expð1:45mÞ (10)
P
1
θ Ast fyt dc =s
Vd m¼ (11)
P 1
Astfyt tan y
Ast fyt dc =s tan y
Ps
s
Beam–Column Joints
N Vsf
The performance of beam–column joints has
Astfyt been identified as a critical issue in the seismic
resistance of RC moment-resisting frames
Vd
(RC MRF). In RC MRF structures under severe
ground motions, beam–column connections are
Ps subjected to moment reversals across the joint
due to the adjacent beams and columns. As a
Seismic Vulnerability Assessment: Reinforced Con-
crete Structures, Fig. 7 Free body diagram of column result, the joint regions undergo significant hori-
after shear failure (From Elwood and Moehle 2005b) zontal (and vertical) shear forces whose magni-
tudes are much larger than those in the adjacent
members (CEB 1996); see Fig. 8.
loss of axial load-carrying capacity, where y is As a result, beam–column joints are suscepti-
equal to 65 , dc is the depth of the column core ble to shear failure which generally involves a
(centerline to centerline of transverse reinforce- brittle process. Such brittle shear failure must be
ment), and Ast and s are area and spacing of avoided through appropriate design to ensure
transverse reinforcement, respectively. ductile response of the frame. Despite the impor-
tance of shear design of RC beam–column joints,
4 1 þ tan ðyÞ2 the research community has not yet developed a
da ¼
(9)
100 s commonly accepted approach for the determina-
tan ðyÞ þ P
Ast fyt dc tan ðyÞ tion of the shear strength of RC beam–column
joints, probably due to the complexity of joint S
Elwood and Moehle’s model is successively behavior.
recalibrated by Zhu et al. (2007). Zhu et al. Many reinforced concrete (RC) buildings
employed a database of 28 column specimens. constructed without transverse steel shear rein-
All tests are unidirectional pseudo-static tests and forcement in the beam–column joint region still
were terminated after loss of axial load capacity. widely exist in seismically active regions, since
All the considered columns experienced flexural the transverse reinforcement requirements for the
yielding prior to shear failure. Zhu et al. provide design of beam–column joints were not
Eq. 10 for the evaluation of drift ratio at loss of addressed in earlier code provisions. Such
vertical load-carrying capacity (axial failure), in unreinforced joints are considered vulnerable to
which m is the effective coefficient of friction brittle shear failure under earthquake shaking due
evaluated according to Eq. 11. The same ruling to insufficient shear reinforcement in the joint
parameters are assumed with respect to Eq. 9, and region. These problems have been highlighted,
y is equal, again, to 65 . Both Eqs. 9 and 10 are in recent past, by the damage observed after
Vc Mc Vc
Vb
Mb
Vjh
Vjv
Vb
Seismic Vulnerability Assessment: Reinforced Con- shear, respectively. Vjh and Vjv are horizontal and vertical
crete Structures, Fig. 8 Shear forces within the joint. joint shears (Adapted from Fardis 2009)
Mc, Vc, Mb, and Vb are column and beam moment and
Seismic Vulnerability Assessment: Reinforced Concrete Structures, Table 1 gn for joint shear strength
evaluation
gn ðMPaÞ0:5
Interior joint Interior joint Exterior joint Exterior joint Knee joint with or
with transverse without transverse with transverse without transverse without transverse
r00 beams beams beams beams beams
<0.003 1.0 0.8 0.7 0.5 0.3
0.003 1.7 1.2 1.2 1.0 0.7
devastating earthquakes in different countries; Therefore, current code provisions and available
moreover, many tests have proven the poor seis- analytical models may be inappropriate to predict
mic performance of unreinforced joints, espe- the shear strength of unreinforced joints; see also
cially of exterior joints. Park and Mosalam (2013) for details.
Design recommendations for beam–column Alternatively, ASCE/SEI 41 (2007) provides
joints in code provisions have been considerably recommendations for the shear strength and joint
revised since the ACI-ASCE Committee shear stress–strain relationship of unreinforced
352 (ACI 352R-02 2002) published its first seis- joints for seismic rehabilitation purposes based
mic design guidelines in 1976. Still, nominal on the pre-standard developed in FEMA
joint shear strengths proposed in current code 273 (FEMA 1997) and FEMA 356 (FEMA
provisions are only applicable to joints satisfying 2000). According to ASCE 41, nominal joint
the minimum requirements of transverse rein- shear strength (Vn) is defined according to
forcement in the joint region. Some analytical Eq. 12; it is a function of a coefficient gn, joint
models (Vollum 1998; Hwang and Lee 1999; width (bj), column
pffiffiffiffi depth (hc), and concrete ten-
Lowes and Altoontash 2003; Wong 2005) for sile strength ( fc )
predicting joint shear strength have been devel-
oped based on the strut-and-tie (SAT) idealiza- pffiffiffiffi
tion incorporating Mohr’s strain compatibility V n ¼ gn fc bj hc (12)
and softening concrete behavior, e.g., the modi-
fied compression field theory (MCFT) proposed The values of gn for joint shear strength cal-
by Vecchio and Collins (1986). However, culation are presented in Table 1. To investigate
LaFave and Shin (2005) indicated that the the relevance of the strength recommendations
strength prediction using the MCFT tends to in ASCE 41, Park and Mosalam (2012) collected
underestimate the shear strength of lightly 62 previous unreinforced exterior or corner joint
reinforced and that of unreinforced joints. test data. The evaluation of the joint shear
a b
l
l
Seismic Vulnerability Assessment: Reinforced Con- evaluation of the database against joint aspect ratio
crete Structures, Fig. 9 Effect of the joint aspect ratio: (Adapted from Park and Mosalam 2012)
(a) SAT idealizations for two joint aspect ratios and (b)
strength for the collected database reveals that recognized by other researchers (Wong 2005;
ASCE 41 may underestimate the shear strength Kim and LaFave 2007):
of unreinforced exterior joints. This parametric
study shows that the shear strength of 1. Joint failure prior to beam longitudinal rein-
unreinforced exterior joints is affected by the forcement yielding (referred to as J type
joint aspect ratio, defined as the ratio of beam failure)
to column cross-sectional heights. The effect of 2. Joint failure after beam longitudinal reinforce-
the joint aspect ratio can be explained by the ment yielding (referred to as BJ-type failure)
SAT idealization where a steeper diagonal strut
is developed in the unreinforced joint with a high The joint shear strength in most code provi-
aspect ratio; see Fig. 9. Consequently, this sions is pertinent to joint failure independent of
steeper diagonal strut results in less effective beam longitudinal bars yielding or not, whereas
shear resistance to equilibrate the horizontal the maximum joint shear stress for the BJ-type S
joint shear force. Hence, the shear strength of failure is directly determined from the joint shear
unreinforced exterior joints decreases with the demand dictated by the yielding of the beam
increase of the joint aspect ratio. Similar results longitudinal reinforcement. Conversely, experi-
are reported by Kim and LaFave (2007), Vollum mental results show that the joint shear strength
and Newman (1999), and Bakir and of unreinforced joints increases with the beam
Boduroğlu (2002). longitudinal reinforcement ratio. This evidence
Another key parameter, joint shear demand can be explained as follows: (i) increasing the
calculated from beam longitudinal bars and beam longitudinal reinforcement ratio leads to
frame geometry, has been suggested by Ander- the increase of the horizontal joint shear force
son et al. (2008) based on results of unreinforced without the yielding of the beam longitudinal
interior joint tests by Walker (2001) and Alire bars, i.e., larger horizontal joint shear force is
(2002). Depending on the joint shear demand, imposed with less deterioration of bond resis-
two types of joint failure have been commonly tance around the beam longitudinal bars in the
a 50
b 50
40
40
30 30
20
Lateral Load (kips)
20
Lateral Load (kips)

10 10
0 0
−10 −10
−20 −20
−30 −30
−40 −40
−50 −50
−10 −9 −8 −7 −6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6 7 8 9 10 −10 −9 −8 −7 −6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6 7 8 9 10
% Drift % Drift
Seismic Vulnerability Assessment: Reinforced Con- joint failure on the sub-assemblage; (b) effect of anchor-
crete Structures, Fig. 10 Experimental results of age failure (Adapted from Pantelides et al. 2002)
unreinforced exterior joint: (a) effect of beam–column
joint region; (ii) this more stable bond resistance transverse reinforcement clamping the crack and
produces a wider diagonal strut which can carry longitudinal reinforcement supporting axial force
the larger horizontal joint shear force. Some directly. Axial failure of a joint may be viewed
strength models use either ductility factor (Park similarly (Hassan 2011). After developing joint
2002; Hakuto et al. 2000) to predict the joint shear failure, the axial load will be supported by
shear strength of BJ type of joint failure. shear friction on the diagonal shear failure plane
A similar approach is adopted to predict the and the axial capacity of column reinforcing bars.
shear strength of columns (Sezen and Moehle Two axial capacity models designed for
2004). unconfined joints were developed. However,
In general, experimental results on this result is obtained based on a small dataset.
beam–column joints without transverse steel
shear reinforcement emphasize a strictly Infills
degrading behavior caused, on one hand, by the The practice of realizing RC frames with
shear behavior of the joint panels and, on the masonry infill panels is very common in Euro-
other hand, by the bond–slip behavior of longitu- pean countries. Masonry infill walls affect the
dinal reinforcements anchored in it; see Fig. 10. strength and stiffness of infilled RC frame struc-
Experimental research on the seismic perfor- tures. The general approach to discard their con-
mance of the beam–column joints (Walker tribution in design and assessment could be
2001; Alire 2002; Pantelides et al. 2002) has acceptable when proper capacity design rules
revealed that the joint shear stress–strain are employed, since code recommendations
response, typically, has a degrading envelope account implicitly for their contribution through
and a highly pinched hysteresis. Moreover, the proper acceptance criteria and drift limitations
common practice of terminating the beam’s bot- (CEN 2004; ASCE/SEI 41 2007; DM14/01/
tom reinforcement within the joints makes the 2008 2008). In addition, their contribution to
bottom reinforcement prone to pullout under a overall strength and stiffness is less significant
seismic excitation. Insufficient beam bottom bar in modern designed structures (e.g., Dolšek and
anchorage precludes the formation of bond Fajfar 2001). On the other hand, assessment of
stresses necessary to develop the yield stress in RC structures often involves substandard build-
the beam’s bottom reinforcement. ings in which the presence of masonry infill can
Finally, as described in section “Loss of Ver- strictly affect the structural behavior. Typical
tical Carrying Capacity,” axial failure of a shear- two-layer hollow clay brick infills employed in
damaged column can occur by sliding along an the Mediterranean area and very common in
inclined crack plane, with resistance provided by existing RC buildings cannot be neglected.
a 120.0 b 70.0
90.0 60.0 Specimen 9

Specimen 3
60.0
Lateral load (kips)
Lateral load (kips)

50.0
30.0
40.0
0.0
30.0
−30.0 Specimen 8 Specimen 1
20.0 Specimen 2
−60.0 Bare Frame
Weak Infill
−90.0 10.0 Strong Infill
−120.0 0.0
−1.6 −1.2 −0.8 −0.4 0.0 0.4 0.8 1.2 1.6 0.0 0.4 0.8 1.2 1.6 2.0
Lateral displacement (in.) Lateral displacement (in.)
Seismic Vulnerability Assessment: Reinforced Concrete Structures, Fig. 11 Example of (a) cyclic- and (b)
monotonically loaded specimens of RC-infilled frames (From Mehrabi et al. 1996)
Their contribution represents the first capacity have been proposed. Infill failure can be classi-
source to earthquake shaking in this kind of build- fied into five distinct modes (e.g., Shing and
ings (e.g., De Luca et al. 2014b; Manfredi Mehrabi 2002; Calvi et al. 2004; Asteris
et al. 2014), and, at the same time, they can lead et al. 2011). Infill failure graphical examples
to preemptive brittle failures due to local interac- according to Shing and Mehrabi (2002) are
tion between infill and RC frame (e.g., shown in Fig. 12.
Verderame et al. 2011).
Notwithstanding the above considerations, it – Sliding shear mode represents horizontal slid-
is worth noting that these elements are considered ing shear failure through bed joints of a
as nonstructural in the design process, and they masonry infill. This mode is associated with
are not subjected to the same acceptance and infill of weak mortar joints and a strong frame.
quality controls of RC elements. Thus their – Diagonal cracking mode is seen in the form of
behavior can be significantly affected by specific a crack across the compressed diagonal of the
executive practice and very different material infill panel and often takes place with simul-
properties. taneous initiation of the sliding shear mode.
Experimental investigations of the last This mode is associated with a weak frame or a
decades (e.g., Mehrabi et al. 1996; Mosalam frame with weak joints and strong members
1996; Colangelo 2003; Calvi et al. 2004, among infilled with a rather strong infill. S
others) emphasized a considerable reduction in – Diagonal compression mode represents the
the response of infilled frames under reversed crushing of the infill within its central region.
cyclic loading. This behavior is due to the rapid This mode is associated with a relatively slen-
degradation of stiffness, strength, and low energy der infill, where failure results from out-of-
dissipation capacity, resulting from the brittle and plane buckling of the infill.
sudden damage of the unreinforced masonry – Corner crushing mode represents the crushing
(URM) infill walls. In Fig. 11 an example of of the infill in at least one of its loaded corners.
cyclic (Fig. 11a) and monotonic (Fig. 11b) behav- This mode is usually associated with infilled
ior of masonry infill frames from the experimen- frames consisting of a weak masonry infill
tal campaign by Mehrabi et al. (1996) is shown. panel surrounded by a frame with weak joints
Based on both experimental and analytical and strong members.
results of the last decades, different failure – Frame failure mode is seen in the form of
mode classifications of masonry-infilled frames plastic hinges developing in the columns or
a b c
d e
Seismic Vulnerability Assessment: Reinforced Con- midheight crack, (b) diagonal crack, (c and d) corner
crete Structures, Fig. 12 Failure mode according to the crushing, (e) horizontal slip (Adapted from Calvi
classification made by Shing and Mehrabi (2002), (a) et al. 2004)
Seismic Vulnerability Assessment: Reinforced Concrete Structures, Fig. 13 (a) Strut model analogy of infilled
frame (From Asteris et al. 2011); (b) Fardis and Panagiotakos (1997) analytical model for infills (From Fardis 2009)
the beam–column connections. This mode is characterization of the strut can be made
associated with a weak frame or a frame with according to different formulations. In literature
weak joints and strong members infilled with a are available different overviews of the different
rather strong infill. analytical approaches (e.g., Crisafulli 1997;
Chrysostomou and Asteris 2012, among others).
Analytical modeling of infills can be carried Some analytical approaches are able to account
out according to different macro-modeling for all the failure modes considered above (e.g.,
approaches. Equivalent strut macro-model Bertoldi et al. 1993), other macro-models does
approach is the simplest way to model the global not account explicitly for all the five modes of
interaction between infill and RC elements; see failure but show a fair agreement with experi-
Fig. 13a; it is also explicitly suggested by codes mental results (e.g., Fardis and Panagiotakos
(e.g., ASCE/SEI 41 2007). On the other hand, the 1997); see Fig. 13b.
Seismic Vulnerability Assessment: Reinforced Concrete Structures, Fig. 14 Idealized models of RC elements
(Adapted from Deierlein et al. 2010)
Local interaction between infills and RC frame the definition of nonlinear structural component
can be captured by multiple strut macro-models models for RC buildings, in terms of M-y. It was
(e.g., Crisafulli 1997) or by single strut macro- emphasized that component models are generally
models with ends located at the edge of the column defined on the basis of experimental results, and
or of the beam (e.g., ASCE/SEI 41 2007). Not- often curvature is not the easiest parameter to be
withstanding the fact that it can be quite hard to employed for the description of all the modes of
distribute strength among the different struts of the failures characteristic of RC element. Thus, most
analytical model, still modeling local interaction commonly, PBEE applications employ lumped
allows to check if brittle failures are likely to occur plasticity models. On the other hand, it is possible
in the RC elements (e.g., Verderame et al. 2011). to find PBEE loss and vulnerability assessment
applications on RC structures that employ fiber
models for low-intensity levels (e.g., Mitrani-
Analytical Modeling of RC Structures Reiser 2007).
In the following, a brief description of both
Inelastic analytical structural component models analytical modeling solutions is provided.
can be differentiated by the way that plasticity is A more complete overview of modeling
distributed through the member cross sections approaches is provided in Deierlein
and along its length. Modeling approaches can et al. (2010). In Fig. 14, the comparison of differ-
be divided in three main groups: (i) lumped plas- ent idealized model types for simulating the S
ticity models, (ii) distributed plasticity models, inelastic response of elements is shown.
and (iii) finite element micro-modeling. The main Lumped plasticity models concentrate the
difference between lumped plasticity, distributed inelastic deformations at the end of the element,
plasticity, and finite element models resides in the such as through a rigid-plastic hinge or an inelas-
parameters employed for the definition of inelastic spring with hysteretic properties (Fig. 14a). By
tic behavior. In the first case, moment–rotation concentrating the plasticity in zero-length hinges
relationships are employed, while, in the other with moment–rotation model parameters, these
two options, moment–curvature or stress–strain elements have relatively condensed numerically
relationships are assumed, and it is necessary to efficient formulations.
pass through a numerical integration step. Distributed plasticity models are different
The analysis of experimental behavior of according to the way the element is discretized,
structural elements in section “Behavior of RC and numerical integration is carried out resulting
Elements and Nonstructural Elements” allowed in finite length hinge models (Fig. 14b) or fiber
Seismic Vulnerability Assessment: Reinforced Concrete Structures, Fig. 15 Shear spring in series model using
limit state material model (From Elwood 2004)
model (Fig. 14c). In the finite length hinge model, latter approach represents the most detailed way
cross sections in the inelastic hinge zones are to represent the elements; on the other hand, it is
characterized through either nonlinear computationally demanding, and it asks for the
moment–curvature relationships or explicit definition of numerous input parameters that, in
fiber-section integrations that enforce the turn, need to be calibrated.
assumption that plane sections remain plane. While distributed plasticity formulations cap-
The inelastic hinge length may be fixed or vari- ture variations of stress and strain through the
able, as determined from the moment–curvature section and along the member in more detail,
characteristics of the section together with the important local behaviors, such as strength deg-
concurrent moment gradient and axial force. radation due to local buckling of steel
The fiber formulation models distribute plasticity reinforcing bars or nonlinear interaction of flex-
by numerical integrations through the member ural and shear, are difficult to capture without
cross sections and along the member length. sophisticated and numerically intensive models.
In this latter case, stress–strain relationship Phenomenological concentrated hinge/spring
is defined and plane-sections-remain-plane models may be better suited to capture the
assumption is enforced; see Deierlein et al. nonlinear degrading response of members
(2010). Distributed fiber formulations do not gen- through calibration (see Fig. 15), using member
erally report plastic hinge rotations, but instead test data, on experimental moment rotations and
report strains in concrete cross-section fibers. The hysteresis curves (Deierlein et al. 2010);
calculated strain demands can be quite sensitive see Fig. 6. It is worth noting that even when
to the moment gradient, element length, integra- opting for lumped plasticity models, it is possi-
tion method, and strain hardening parameters on ble to refine the analytical modeling choice in
the calculated strain demands. These modeling different ways, and the accuracy of results is
approaches allow capturing the interaction consequently affected; see, for example, Yavari
between axial load and biaxial components of et al. (2009).
bending moment at section level, under cyclic Beam–column joint modeling is still at an
loading. early and less mature stage with respect to mem-
Finally, finite element models (Fig. 14d) rep- ber modeling (i.e., beam and columns), and prac-
resent a detailed micro-modeling solution in tical available solutions are to consider rotational
which material constitutive relationships are springs capable to describe the beam–column
assigned to each element. It is obvious that this joint behavior that is governed by shear and
a b c
2
for joint shear
5 1
2 2
6 1 3 1
4 1
for bond-slip
d e f
Seismic Vulnerability Assessment: Reinforced Con- (d) Lowes and Altoontash (2003), (e) Altoontash (2004),
crete Structures, Fig. 16 Existing beam–column joint and (f) Shin and LaFave (2004) (From Celik and
models: (a) Alath and Kunnath (1995), (b) Biddah and Ellingwood 2008)
Ghobarah (1999), (c) Youssef and Ghobarah (2001),
bond–slip deformation; see some modeling zero-length rotational springs located at

scheme examples in Fig. 16 as collected by beam–column joint interfaces (Fig. 16e). Shin
Celik and Ellingwood (2008). Alath and Kunnath and LaFave (2004) represented the joint by rigid
(1995) modeled the joint shear deformation with elements located at the edges of the panel zone and
a rotational spring model with degrading hyster- rotational springs embedded in one of the four
esis (Fig. 16a). Biddah and Ghobarah (1999) hinges linking adjacent rigid elements (Fig. 16f).
modeled the joint shear and bond–slip deforma-
tions with separate rotational springs (Fig. 16b). S
Youssef and Ghobarah (2001) proposed a joint Damage Measures
element in which two diagonal translational
springs connecting the opposite corners of the The nature and amount of structural damage
panel zone simulate the joint shear deformation, depends necessarily on the quality of the mate-
and 12 translational springs at the panel zone rials that compose structural and nonstructural
interfaced simulate all other modes of inelastic elements, and on the configuration and type of
behavior (Fig. 16c). Lowes and Altoontash structural systems. In the early years of modern
(2003) proposed a 4-node 12 of freedom joint earthquake engineering, damage definition was
element that explicitly represents three types of basically approached in qualitative terms (e.g.,
inelastic mechanisms of beam–column joints through the definition of probable localization
under reversed cyclic loading (Fig. 16d). Succes- of such damage in a structure). This kind of
sively Altoontash (2004) simplified the model approach relied fundamentally on the observation
by Lowes and Altoontash introducing a four of damaged structures after seismic events or it
was based on experimental tests. The first studies number of data, and they cannot be considered
for damage quantification of RC structural mem- fully reliable yet, since they need further numer-
bers date back to late 1980s (e.g., Park and Ang ical refinement.
1985; Fardis et al. 1992). The availability of Finally, in section “Code Limit States,” a brief
experimental data allowed more and more overview of code acceptance criteria for RC ele-
detailed approaches to damage analysis up to ments at different limit states is provided, and
the most recent PBEE applications. compared with the information provided in sec-
Most recent studies on damage analysis tion “Examples of Damage Analysis.” The two
employ component-based approaches (e.g., sections allow a comparison between next-
Aslani 2005; Mitrani-Reiser 2007). The PBEE generation performance-based codes (i.e., section
damage analysis overpasses the approach to “Examples of Damage Analysis”) with respect to
global estimate of damage implicit in previous current performance-based codes (i.e., section
damage scales (e.g., ATC-13 1985; Grunthal “Code Limit States”).
1998; NIBS 1999) and switch to a component-
based damage analysis that allows a more accu- Examples of Damage Analysis
rate definition of losses. On the other hand, Various damage indices or EDP thresholds are
component-based approach requires a significant used to quantify damage of RC structural mem-
amount of data, such as experimental data, earth- bers. As an example, Mitrani-Reiser (2007)
quake experience, expert opinion, and some com- employed for beam and columns’ damage states
bination of these. the fragility curves developed by Beck
For component-based approaches, it is neces- et al. (2002) and, in turn, based on Williams and
sary to define fragility functions for various dam- Sexsmith (1997) studies. Beck et al.’s fragility
age states for each structural and nonstructural functions use a deformation damage index (DDI)
component in typical RC frames. In literature, that can be evaluated from chord rotations
there are available different benchmark studies obtained in structural analyses. This approach
for damage assessment of RC buildings (e.g., allows the definition of four damage states for
Aslani 2005; Mitrani-Reiser 2007). FEMA RC members. On the other hand, Aslani (2005)
P-58-1 (2012) provides a list of fragility specifi- provides fragility functions for RC columns with
cations for structural components, nonstructural light transversal reinforcement in terms of IDR,
components, and contents. evaluated from experimental data, and based on
The functional form most commonly used for the component modeling provided in section
components’ fragility functions is the lognormal “Behavior of RC Elements and Non Structural
distribution that asks for the definition of median Elements.” For both approaches, four damage
and logarithmic standard deviation. Therefore, states are defined resulting in damage descrip-
for each component, it is necessary to define tions compared in Table 2. Notwithstanding the
median capacity and logarithmic standard devia- fact that Aslani refers specifically to lightly
tion (slog) in terms of EDP for each damage state. reinforced RC columns (i.e., non-ductile col-
In section “Examples of Damage Analysis”, umns), similar damage descriptions are provided
there is a brief overview of component-based in Table 2. In both approaches, the typical log-
damage analyses for RC structural components normal shape is adopted for the fragility curve, so
and infills according to different authors. This that for each damage state, a median EDP and a
overview is aimed at providing an example of slog are estimated.
damage evaluation made for the elements whose Aslani estimates median and slog for each
behavior was considered in section “Behavior of damage states, based on a database of 92 lightly
RC Elements and Non Structural Elements.” Not- reinforced columns. The preliminary result of
withstanding the fact that quantitative approaches this estimation leads to a significant dispersion
to component damage analysis rely on experi- for all damage states; so an additional parameter
mental databases, they are still based on a limited (a) is considered in the regression to reduce the
Seismic Vulnerability Assessment: Reinforced Concrete Structures, Table 2 Comparison of damage states for
columns
Williams and Sexsmith (1997) – ductile columns Aslani (2005) – non-ductile columns
Damage state Damage description Damage state Damage description
None None or small number of light
cracks, either flexural (90 ) or
shear (45 )
Light Widespread light cracking or a few DS1: light Visible cracks; crack widths smaller
cracks >1 mm or light shear cracking than 0.3 mm. Light repair to improve
cracks tending to flatten toward aesthetic appearance
30
Moderate Significant cracking, e.g., 90 DS2: severe Wider and deeper crack widths, more
cracks >2 mm; 45 cracking extensive compared to DS1
cracks >2 mm; 30 cracks >1 mm
Severe Very large flexure or shear cracks, DS3: shear Deterioration of shear capacity after
usually accompanied by limited failure yielding (that occurs at earlier stages of
spalling of cover concrete loading compared to ductile columns)
leads to shear failure
Collapse Very severe cracking and spalling DS4: axial Loss of vertical carrying capacity. This
of concrete; buckling, kinking, or failure damage state has possible disastrous
fracture of rebar consequences, if there is no possibility
to redistribute vertical load to other
members
a 1 b 1
DS1 DS1
DS2 DS2
0.8 DS3 0.8 DS3
DS4 DS4
P(DS=dsi⎪IDR)
P(DS=dsi⎪IDR)
0.6 0.6
0.4 0.4
0.2 0.2
α=86 α=15
0 0 S
0 0.04 0.08 0.12 0.16 0.2 0 0.04 0.08 0.12 0.16 0.2
IDR IDR
Seismic Vulnerability Assessment: Reinforced Concrete Structures, Fig. 17 Probability of being at each damage
state for non-ductile RC columns characterized by different a parameter
resulting slog of the fragilities and to adjust the columns is shown, and the effect of a is empha-
estimated median IDR at each damage states. The sized through the comparison of the curves in
parameter a = P/(Agr00 fc) is function of axial load Fig. 17a, b. It is worth noting that a affects strictly
(P), gross area section (Ag), concrete compressive shear failure and axial failure as it can be
strength (fc), and transversal reinforcement ratio expected by the observation of the parameters
(r00 ). In Fig. 17, the probability of being in a ruling equations shown in section “Behavior of
specific damage state for lightly reinforced RC Elements and Non Structural Elements.”
Seismic Vulnerability Assessment: Reinforced Con- terms of cracking, crushing, etc. and the feasibil-
crete Structures, Table 3 Damage state definition for ity of repair. As an example, Table 4 provides
interior and exterior beam–column joints according to
Aslani (2005) damage description at different damage states
according to different authors; see Colangelo
Damage state Damage description
(2013). Significant differences can be found in
DS1: sever cracking Wide and extensive cracking of
in the beam the beam damage descriptions and, as consequences, also
DS2: severe cracking Wide and extensive cracking of in the fragility function parameters estimated
in the column the column from each description (see also Colangelo 2013).
DS3: severe cracking Severe cracking at the Despite the significant differences that can be
in the joint beam–column joint found in literature formulations for component
DS4: joint spalling Spalling of the concrete cover fragilities of infills, it is worth noting that widely
of the beam–column joint
employed damage scales for RC structures that
DS5: loss of vertical The joint collapses under its
carrying capacity gravity load interpret global damage of buildings often refer
to infill damage. An example is the EMS 98 scale
(Grunthal 1998) in which the first three grades of
For beam–column connections, Aslani con- global damage are mainly characterized by dam-
siders different fragility curves for interior age description of masonry infills; see, for exam-
beam–column connections and exterior beam ple, De Luca et al. (2014b) and Manfredi
column connections. Damage state definitions et al. (2014), in which infill damage analysis is
are identical in the two cases, and they are employed as analytical proxy for damage classi-
shown in Table 3, while fragility functions, fication of the whole RC building. On the other
based on experimental data, can be really differ- hand, EMS 98 damage scale is not a component-
ent. The numerical estimation for joints is again based approach, but the single most damaged
affected by high uncertainties. element represents the damage classification of
Fragility functions for masonry infill panels the whole structure.
are not available in the damage and loss studies
cited up to this point. On the other hand, experi- Code Limit States
mental tests (e.g., Mehrabi et al. 1996; Colangelo Current PBEE approach is also the result of the
2003) and post-earthquake damage observations huge effort made in ATC-58 project. The purpose
(e.g., Ricci et al. 2011; De Luca et al. 2014b; of this project is to develop next-generation seis-
Manfredi et al. 2014) have shown that this kind mic design procedures that will provide a more
of elements can have a significant impact on both reliable means of predicting and designing per-
damage and loss analysis of RC structures, espe- formances of structures. On the other hand, PBEE
cially in the Mediterranean area, in which this has been already implemented in codes providing
kind of structural typology is quite common. target performance levels. As an example, FEMA
Notwithstanding the fact that damage observa- 356 (FEMA 2000) and successively ASCE/SEI
tions based on experimental tests are available 41 (2007) do not attempt to quantify the proba-
in literature and macro-modeling approaches bility of achieving a given performance level or
and component modeling backbones as well are to quantify losses; but, both FEMA 356 and
available in literature, still characteristic param- ASCE/SEI 41 address component-level and
eters for infill fragilities are characterized by sig- system-level damage states, and they relate dam-
nificant differences according to different authors age to life safety and post-earthquake operability.
(e.g., Gu and Lu 2005; Özcebe et al. 2012; The damage states given in FEMA 356 tend to be
Colangelo 2013). The point is that in the case of qualitative and open to multiple interpretations.
damage analysis of infills, authors do not even In Table 5, target performance levels according to
agree on damage states’ qualitative definitions. FEMA 356 and ASCE/SEI 41 are shown.
The damage degrees are distinguished on the ASCE/SEI 41 (2007) succeeded FEMA 356 as
basis of the description of physical damage in document for the seismic assessment of existing
Seismic Vulnerability Assessment: Reinforced Concrete Structures, Table 4 Damage state definition for
masonry infills according to Colangelo (2013)
Damage
description
Damage description (Colangelo Damage description (Gu and Lu (Özcebe
Damage state 2013) 2005) et al. 2012)
DS1 Onset of cracking in the bricks, Minor cracking and falling of plaster; Negligible cracks;
associated with the first noticeable only local repair needed with maximum
reduction of stiffness function maintained; maximum strength (i.e., base
strength shear)
DS2 Moderate cracks before attaining Continual diagonal cracking and Appreciable
the maximum strength (i.e., base flaking; repairable damage; 30 % damage;
shear) reduction of the maximum strength maximum stress
in the infill
DS3 Extensive cracks with tensile Loss of wall integrity; 70 % reduction Ultimate strain in
splitting and falling of the outer of the maximum strength the infill
layer of a few bricks; repairable
damage
DS4 So many broken bricks that repair
is unreasonable; reconstruction
needed
buildings; it provides a performance-based engi- according to Biskinis and Fardis (2010a). All
neering framework whereby deformation and these chord rotation capacities are intended as
force demands for different seismic hazards are member capacity thresholds if shear capacity
compared against deformation and force capaci- evaluated according to Biskinis et al. (2004) is
ties for various performance levels. Elwood not attained, and a preemptive brittle failure
et al. (2007) provided an update to concrete pro- (computed through a force-based approach)
visions in ASCE/SEI 41 given new data available does not occur before the attainment of such
in literature and considering that practitioners chord rotations.
observed that previous concrete provisions tend Regarding acceptance criteria for infills, codes
to err on the conservative side. In their update to (e.g., CEN 2004), in general, do not provide
concrete provisions, Elwood et al. provide specific drift limits for infill elements, since it is
modeling parameters and acceptance criteria not taken for granted the fact that infills are
modifications for columns, beam–column joints, explicitly modeled. On the other hand, immediate
etc. Regarding columns, the classification occupancy and operational acceptance criteria
described in section “RC Members” is consid- are evaluated to implicitly account for damage S
ered. As an example, for columns, drift limits at to masonry infills.
the different acceptance criteria change In FEMA 356 and ASCE/SEI 41, four
according to the classification of the expected nonstructural performance levels are classified
mode of failure. (see Table 5). The limit for exterior walls differs
European code for the assessment of RC ele- from the limit for heavy partitions (light parti-
ments (CEN 2005) provides, as collapse preven- tions may be drywall partitions with studs, for
tion acceptance criteria for RC members, the instance) for both occupancy and life safety per-
value of the ultimate rotation capacity according formance level. For the occupancy level for
to Biskinis and Fardis (2010b); as life safety heavy partitions, the limit is 0.5 %, the same as
acceptance criteria, 75 % of the ultimate rotation Eurocode 8 (CEN 2004), while for exterior walls,
capacity according to Biskinis and Fardis the limit is 1 %. For life safety level, the above
(2010b); and, as immediate occupancy accep- occupancy limits double. The Italian code
tance criteria, the value of yielding rotation (DM14/1/2008 2008) also prescribes 0.5 % for
Seismic Vulnerability Assessment: Reinforced Concrete Structures, Table 5 Damage control and building
performance level extracted from Table C1-2 in FEMA 356 (FEMA 2000) and ASCE/SEI 41 (2007)
Collapse prevention Immediate
level Life safety level occupancy level Operational level
Overall Severe Moderate Light Very light
damage
General Little residual stiffness Some residual strength No permanent drift. No permanent drift.
and strength, but load- and stiffness left in all Structure Structure
bearing columns and stories. Gravity-load- substantially retains substantially retains
walls function. Large bearing elements original strength and original strength and
permanent drifts. function. No out-of- stiffness. Minor stiffness. Minor
Some exits blocked. plane failure of walls cracking of facades, cracking of facades,
Infills and unbraced or tipping of parapets. partitions, and partitions, and
parapets failed or at Some permanent drift. ceilings as well as ceilings as well as
incipient failure. Damage to partitions. structural elements. structural elements.
Building is near Building may be Elevators can be All systems
collapse beyond economical restarted. Fire important to normal
repair protection operable operation are
functional
Nonstructural Extensive damage Falling hazards Equipment and Negligible damage
components mitigated but many contents are occurs. Power and
architectural, generally secure but other utilities are
mechanical, and may not operate due available, possibly
electrical systems are to mechanical failure from standby sources
damaged or lack of utilities
occupancy, while 0.5 % times 2/3 is prescribed to non-ductile RC elements (e.g., characterization
ensure operation for civil protection. Moreover, of axial load failure) should be better character-
with regard to existing buildings, in the commen- ized on the basis of higher number of experimen-
tary to the Italian code (CM 617 2009), both tal tests in order to better characterize both
limits are reduced if the analysis model includes modeling issues and damage analysis issues in
the infills. In such a case, the limits become the terms of median and logarithmic standard devia-
same as for the masonry buildings, which are tion. Same considerations can be made on
equal to 0.3 % and 0.2 % to ensure occupancy beam–column joints and masonry infills.
and operation, respectively. Finally, masonry infills represent a relevant
issue for all countries in the Mediterranean area
in which this kind of construction practice is very
Future Challenges and Concluding common. Despite the huge efforts made in the
Remarks last decades, the characterization of nonstructural
component modeling and damage is still chal-
The overview of the seismic assessment frame- lenging considering that they are not subjected
work for RC building provided in this entry has to the same design and control process to which
been organized according to the most recent state new RC elements are subjected to.
of the art and guidelines that provide a vision for Performance evaluation of RC structure
next-generation codes for RC structures. On the according to the methodological framework of
other hand, there are still different aspects on FEMA P-58-1 (2012) provides a significant
which research and, in turn, code standards enhancement toward a more reliable performance
should focus. evaluation for stakeholders and decision makers.
In particular, behavior of RC elements and Still, some procedures need to be codified in
their modeling still requires significant experi- simpler and more user-friendly tools to be
mental and numerical efforts. Behavior of employed by practitioners and professional
engineers in any structural performance evalua- ▶ Performance-Based Design Procedure for

tions as current and consolidated practice. Structures with Magneto-Rheological Dampers
▶ Seismic Collapse Assessment
▶ Seismic Reliability Assessment, Alternative
Summary Methods for
▶ Seismic Risk Assessment, Cascading Effects
The framework in which modern design and ▶ Site Response for Seismic Hazard Assessment
assessment of reinforced concrete structures are
placed is the well-known Performance-Based
References
Earthquake Engineering (PBEE) framework.
The most recent methodological organization of ACI 352R-02 (2002) Recommendations for design of
PBEE tools is represented by FEMA P-58-1 doc- beam-column connections in monolithic reinforced
ument. The brief analysis of the main steps of concrete structures. American Concrete Institute, Far-
PBEE provides the basis to introduce where, in mington Hills
Alath S, Kunnath SK (1995) Modeling inelastic shear
such framework, the specific structural deformations in rc beam-column joints. In: Engineer-
material – in this case RC – plays a role. In ing mechanics proceedings of 10th conference, Uni-
particular, when the relationship between inten- versity of Colorado at Boulder, Boulder, 21–24 May,
sity measure and engineering demand parameter vol 2. ASCE, New York, pp 822–825
Alire DA (2002) Seismic evaluation of existing
arises in PBEE, it is the phase in which it is unconfined RC beam–column joints. MSc thesis, Uni-
necessary to focus on the behavior of RC struc- versity of Washington
tural elements, in order to approach analytical Altoontash A (2004) Simulation and damage models for
modeling, and quantify components damage performance assessment of reinforced concrete beam-
column joints. PhD dissertation, Department of Civil
measures. and Environment Engineering, Stanford University,
Seismic vulnerability assessment has to Stanford
account for all the typical modes of failure that American Society of Civil Engineers (ASCE) (2007) Seis-
RC elements can show. Thus, behavior of RC mic rehabilitation of existing buildings, ASCE/SEI
41-06. American Society of Civil Engineers, Reston, VA
elements (e.g., beams, columns, beam–column Anderson M, Lehman D, Stanton J (2008) A cyclic shear
joints) is described considering the case of stress-strain model for joints without transverse rein-
existing elements representing the most general forcement. Eng Struct 30:941–954
situation for vulnerability assessment problems. ATC Applied Technology Council (1985) ATC-13, earth-
quake damage evaluation data for California. Applied
The behavior of masonry infills is also consid- Technology Council, Redwood City, CA, pp 492
ered, given the significant effect they can have on Aschheim AM, Moehle JP (1992) Shear strength and
vulnerability in terms of strength and stiffness deformability of RC bridge columns subjected to
increase, and in terms of occurrence of brittle inelastic displacements. Technical Report No
UCB/EERC 92/04, University of California at Berke-
failures caused by local interaction of infill and ley, Berkeley S
RC frame. Critical review of behavior of RC Aslani H (2005) Probabilistic earthquake loss estimation
structures and, in turn, of RC elements, allows and loss disaggregation in buildings. PhD thesis,
emphasizing future challenges and needs that the Stanford University
Asteris PG, Kakaletsis DJ, Chrysostomou CZ, Smyrou EE
scientific community and practitioners should (2011) Failure modes in infilled frames. Electron
take on in the next years. J Struct Eng 11(1):11–20
Bakir PG, Boduroğlu HM (2002) A new design equation
for predicting the joint shear strength of monotonically
Cross-References loaded exterior beam-column joints. Eng Struct
24:1105–1117
▶ Analytic Fragility and Limit States [P(EDP| Beck JL, Porter KA, Shaikhutdinov R, Au SK, Moroi T,
Tsukada Y, Masuda M (2002) Impact of seismic risk
IM)]: Nonlinear Dynamic Procedures
on lifetime property values, final report for CUREE-
▶ Analytic Fragility and Limit States [P(EDP| Kajima phase IV project. Consortium of Universities
IM)]: Nonlinear Static Procedures for Research in Earthquake Engineering, Richmond
Berry M, Parrish M, Eberhard M (2004) PEER structural CM 617 (2009) Istruzioni per l’applicazione delle nuove
performance database user’s manual. Pacific Engi- norme tecniche per le costruzioni. Ministero delle
neering Research Center, University of California, Infrastrutture e dei Trasporti, Rome (in Italian)
Berkeley, 38 pp. Available at http://nisee.berkeley. Coburn A, Spence R (2002) Earthquake protection,
edu/spd/ and http://maximus.ce.washington.edu/ 2nd edn. Wiley, Chichester
peera1/. Accessed 23 Feb 2014 Colangelo F (2003) Experimental evaluation of member-
Bertoldi SH, Decanini LD, Gavarini C (1993) Telai by-member models and damage indices for infilled
tamponati soggetti ad azioni sismiche, un modello frames. J Earthq Eng 7(1):25–50
semplificato: confronto sperimentale e numerico. In: Colangelo F (2013) Drift-sensitive non-structural damage
Proceeding of 6 Convegno Nazionale ANIDIS, to masonry infilled reinforced concrete frames
vol 2, 13–15 Perugia, Italy, 815–824 (in Italian) designed to Eurocode 8. Bull Earthq Eng.
Biddah A, Ghobarah A (1999) Modelling of shear defor- doi:10.1007/s10518-013-9503-y
mation and bond slip in reinforced concrete joints. Cornell A, Krawinkler H (2000) Progress and challenges
Struct Eng Mech 7(4):413–432 in seismic performance assessment. PEER News, Apr
Biskinis D, Fardis MN (2010a) Deformations at flexural Crisafulli FG (1997) Seismic behaviour of reinforced con-
yielding of members with continuous or lap-spliced crete structures with masonry infills. PhD thesis, Uni-
bars. Struct Concr 11(3):127–138 versity of Canterbury, Christchurch
Biskinis D, Fardis MN (2010b) Flexure-controlled ulti- De Luca F, Verderame GM (2013) A practice-oriented
mate deformations of members with continuous or approach for the assessment of brittle failures in
lap-spliced bars. Struct Concr 11(2):93–108 existing RC elements. Eng Struct 48:373–388
Biskinis DE, Roupakias GK, Fardis MN (2004) Degrada- De Luca F, Ameri G, Iervolino I, Bindi D (2014a) Toward
tion of shear strength of reinforced concrete members validation of simulated accelerograms via prediction
with inelastic cyclic displacement. ACI Struct equations for nonlinear SDOF response. Boll Geofis
J 101(6):773–783 Teor Appl 55(1):81–101
Bousias SN (1993) Experimental and analytical study of De Luca F, Verderame GM, Gómez-Martı́nez F, Pérez-
RC columns in cyclic biaxial bending with axial force. Garcı́a A (2014b) The structural role played by
Doctoral thesis, Civil Engineering Department, Uni- masonry infills on RC building performances after
versity of Patras, Patras the 2011 Lorca, Spain, earthquake. Bull Earthq Eng
Bousias SN, Panagiotakos TB, Fardis MN (2002) Model- 12(5):1999–2026
ling of RC members under cyclic biaxial flexure and Deierlein GG, Krawinkler H, Cornell CA
axial force. J Earthq Eng 6(3):213–238 (2003) A framework for performance-based earth-
Calvi GM, Bolognini D, Penna A (2004) Seismic perfor- quake engineering. In: Proceedings of 2003 pacific
mance of masonry-infilled RC frames–benefits of earthquake conference on earthquake engineering,
slight reinforcements. Invited lecture to “Sı́smica University of Canterbury, Christchurch
2004 – 6 Congresso Nacional de Sismologia e Deierlein GG, Reinhorn AM, Willford MR
Engenharia Sı́smica”, Guimarães, 14–16 Apr (2010) Nonlinear structural analysis for seismic
Calvi GM, Pinho R, Magenes G, Bommer JJ, Restrepo- design, NEHRP Seismic Design Technical Brief
Vélez LF, Crowley H (2006) Development of seismic No. 4, produced by the NEHRP Consultants Joint
vulnerability assessment methodologies over the past Venture, a partnership of the Applied Technology
30 years. ISET J Earthq Technol Paper No Council and the Consortium of Universities for
472 43(3):75–104 Research in Earthquake Engineering, for the National
CEB (1996) RC frames under cyclic loading. T. Telford, Institute of Standards and Technology, Gaithersburg,
London, for Comite Euro-international du beton, NIST GCR 10-917-5
Lausanne Di Ludovico M, Verderame G, Prota A, Manfredi G,
Celik OC, Ellingwood BR (2008) Modeling Cosenza E (2013) Experimental behavior of
beam–column joints in fragility assessment of gravity nonconforming RC columns with plain bars under
load designed reinforced concrete frames. J Earthq constant axial load and biaxial bending. J Struct Eng
Eng 12(3):357–381 139(6):897–914
CEN (2004) EN 1998-1 Eurocode 8: design of structures DM 14/1/2008(2008) Norme tecniche per le costruzioni.
for earthquake resistance-part 1: general rules, seismic Ministero delle Infrastrutture e dei Trasporti, Rome
actions and rules for buildings. European Committee (in Italian)
for Standardization, Brussels Dolšek M, Fajfar P (2001) Soft storey effects in uniformly
CEN (2005) Eurocode 8: design of structures for earth- infilled reinforced concrete frames. J Earthq Eng
quake resistance – part 3: assessment and retrofitting of 5(1):1–12
buildings. European Standard EN 1998-1:2005- Elwood K (2004) Modelling failures in existing reinforced
Comité Européen de Normaliation, Brussels concrete columns. Can J Civ Eng 31:846–859
Chrysostomou CZ, Asteris PG (2012) On the in-pane Elwood KJ, Eberhard M (2009) Effective stiffness of
properties and capacities of infilled frames. Eng Struct reinforced concrete columns. ACI Struct J 106(4):
41:385–402 476–484
Elwood K, Moehle JP (2005a) Drift capacity of reinforced predicting flexural response leading to global collapse
concrete columns with light transverse reinforcement. of RC frame buildings, PEER report 2007/03
Earthq Spectra 21(1):71–89 Hassan WM (2011) Analytical and experimental assess-
Elwood K, Moehle JP (2005b) Axial capacity model for ment of seismic vulnerability of beam-column joints
shear-damaged columns. ACI Struct without transverse reinforcement in concrete build-
J 106(S45):578–587 ings. PhD dissertation, University of California,
Elwood KJ, Matamoros AB, Wallace JW, Lehman DE, Berkeley
Heintz JA, Mitchell AD, Moore MA, Valley MT, Hwang S, Lee H (1999) Analytical model for predicting
Lowes LN, Comartin CD, Moehle JP (2007) Update shear strengths of exterior reinforced concrete beam-
to ASCE/SEI 41 concrete provisions. Earthq Spectra column joints for seismic resistance. ACI Struct
23(3):493–523 J 96(5):846–858
Fardis MN (2009) Seismic design, assessment and Ibarra LF, Medina RA, Krawinkler H (2005) Hysteretic
retrofitting of concrete buildings based on models that incorporate strength and stiffness deterio-
EN-Eurocode 8. Springer, Dordrecht ration. Earthq Eng Struct Dyn 34:1489–1511
Fardis MN, Panagiotakos TB (1997) Seismic design and Kim J, LaFave JM (2007) Key influence parameters for
response of bare and infilled reinforced concrete the joint shear behavior of reinforced concrete
buildings – part II: infilled structures. J Earthq Eng (RC) beam-column connections. Eng Struct
1(3):473–503 29:2523–2539
Fardis MN et al (1992) Damage measures and failure Krawinkler H (2002) A general approach to seismic per-
criteria during seismic response. In: Cooperative formance assessment. In: Proceedings of international
research program on the seismic response of conference on advances and new challenges in earth-
reinforced concrete structures. First phase, Final quake engineering research, ICANCEER, Hong Kong,
report, II, National Laboratory for Civil Engineering, 19–20 Aug
Lisbon Krawinkler H, Miranda E (2004) Performance-based
Federal Emergency Management Agency (FEMA) earthquake engineering, chapter 9. In: Bozorgnia Y,
(1997) FEMA 273, guidelines for the seismic rehabil- Bertero VV (eds) Earthquake engineering: from engi-
itation. Federal Emergency Management Agency, neering seismology to performance-based
Washington, DC engineering. CRC Press, Boca Raton
Federal Emergency Management Agency (FEMA) LaFave M, Shin M (2005) Discussion of ‘Modeling
(2000) FEMA 356, prestandard and commentary for reinforced-concrete beam-column joints subjected to
the seismic rehabilitation of buildings. Federal Emer- cyclic loading’, by Lowes, L. N. and Altoontash,
gency Management Agency, Washington, DC A. ASCE J Struct Eng 131(6):992–993
Fédération internationale du béton (FIB). Task Group 7.1 Legeron F, Paultre P (2000) Behavior of high-strength
(2003) Seismic assessment and retrofit of reinforced concrete columns under cyclic flexure and constant
concrete buildings: state-of-art report (vol 24). Inter- axial load. ACI Struct J 97(4):591–601
national Federation for Structural Concrete (ed). Lowes LN, Altoontash A (2003) Modeling reinforced-
FIB-Féd. Int. du Béton concrete beam-column joints subjected to cyclic load-
Federal Emergency Management Agency (FEMA) P-58-1 ing. ASCE J Struct Eng 129(12):1686–1697
(2012) Seismic performance assessment of buildings Manfredi G, Prota A, Verderame GM, Ricci P, De Luca
volume 1 – 558 methodology. Federal Emergency F (2014) 2012 Emilia earthquake, Italy: reinforced
Management Agency, Washington, DC concrete buildings response. Bull Earthq Eng
Goulet CA, Haselton CB, Mitrani-Reiser J, Beck JL, 12(5):2275–2298
Deierlein GG, Porter KA, Stewart JP (2007) Evaluation McGuire RK (2004) Seismic hazard and risk analysis.
of the seismic performance of a code-conforming Report MNO-10. Earthquake Engineering Research S
reinforced-concrete frame building – from seismic Institute Publication, Oakland
hazard to collapse safety and economic losses. Earthq Mehrabi AB, Shing PB, Schuller MP, Noland JL
Eng Struct Dyn 36(13):1973–1997 (1996) Experimental evaluation of masonry-infilled
Grunthal G (ed) (1998) European Macroseismic Scale RC frames. J Struct Eng ASCE 122(3):228–237
1998 (EMS-98), Cahiers du Centre Européen de Mitrani-Reiser J (2007) An ounce of prevention: probabi-
Géodynamique et de Séismologie, 15, Luxembourg listic loss estimation for performance-based earth-
Gu X, Lu Y (2005) A fuzzy–random analysis model for quake engineering. PhD dissertation, California
seismic performance of framed structures incorporat- Institute of Technology (Caltech), Pasadena
ing structural and non-structural damage. Earthq Eng Moehle JP (2003) A framework for performance-
Struct Dyn 34(10):1305–1321 based earthquake engineering. In: Proceedings of
Hakuto S, Park R, Tanaka H (2000) Seismic load tests on ATC-15-9 workshop on the improvement of building
interior and exterior beam-column joints with substan- structural design and construction practices, Maui,
dard reinforcing details. ACI Struct J 97(1):11–25 June
Haselton CB, Liel AB, Taylor Lange S, Deierlein GG Moehle JP, Deierlein GG (2004) A framework
(2008) Beam-column element model calibrated for for performance-based earthquake engineering.
In: Proceedings of 13th world conference on earth- Shin M, LaFave JM (2004) Testing and modeling for
quake engineering, Paper No 679, Vancouver cyclic joint shear deformations in rc beam-column
Mosalam KM (1996) Experimental and computational connections. In: Proceedings of the thirteenth world
strategies for the seismic behavior evaluation of conference on earthquake engineering, Vancouver,
frames with infill walls. PhD dissertations, Cornell 1–6 Aug 2004, Paper No 0301
University, Ithaca Shing PB, Mehrabi AB (2002) Behaviour and analysis of
National Institute of Building Science (NIBS) (1997, 1999 masonry-infilled frames. Prog Struct Eng Mater
and 2002) Earthquake loss estimation methodology. 4:320–331
HAZUS. Technical manuals, vols 1, 2, 3. Federal Tothong P, Cornell CA (2006) An empirical
Emergency Management Agency (FEMA), Washing- ground motion attenuation relation for inelastic spec-
ton, DC. http://www.fema.gov/hazus/ tral displacement. Bull Seismol Soc Am 96:
Özcebe S, Crowley H, Bal IE (2012) Distinction between 2146–2164
no and slight damage states for existing RC buildings Tothong P, Luco N (2007) Probabilistic seismic demand
using a displacement-based approach. In: Proceedings analysis using advanced ground motion intensity mea-
of the 15th world conference on earthquake engineer- sures. Earthq Eng Struct Dyn 36:1837–1860
ing, paper 5126, 24–28, Lisbon, Portugal Vecchio FJ, Collins MP (1986) The modified
Panagiotakos TB, Fardis MN (2001) Deformations of RC compression-field theory of reinforced concrete
members at yielding and ultimate. ACI Struct elements subjected to shear. ACI Struct J 83(2):
J 98(2):135–148 219–231
Pantelides CP, Hansen J, Nadauld J, Reaveley LD Verderame GM, Ricci P, Manfredi G, Cosenza E (2010)
(2002) Assessment of reinforced concrete building Ultimate chord rotation of RC columns with smooth
exterior joints with substandard details, PEER 2002/ bars: some considerations about EC8 prescriptions.
18, May 2002 Bull Earthq Eng 8(6):1351–1373
Park R (2002) A summary of results of simulated seismic Verderame GM, De Luca F, Ricci P, Manfredi G (2011)
load tests on RC beam–column joints, beam and col- Preliminary analysis of a soft-storey mechanism after
umns with substandard reinforcing details. J Earthq the 2009 L’Aquila earthquake. Earthq Eng Struct Dyn
Eng 6(2):147–174 40(8):925–944
Park YJ, Ang AMS (1985) Mechanistic seismic damage Vollum RL (1998) Design and analysis of exterior beam
model of reinforced concrete. ASCE J Struct Eng column connections. PhD dissertation, Imperial Col-
111:722–739 lege of Science Technology and Medicine-University
Park S, Mosalam KM (2012) Parameters for shear strength of London
prediction of exterior beam–column joints without Vollum RL, Newman JB (1999) Strut and tie models for
transverse reinforcement. Eng Struct 36:198–209 the analysis/design of external beam-column joints.
Park S, Mosalam KM (2013) Experimental investigation Mag Concr Res 51(6):415–425
of nonductile RC corner beam-column joints with floor Walker SG (2001) Seismic performance of existing RC
slabs. ASCE J Struct Eng 139(1):1–14 beam–column joints. MSc thesis, University of
Porter KA (2003) An overview of PEER’s performance- Washington
based earthquake engineering methodology. In: Con- Watanabe F, Ichinose T (1992) Strength and ductility of
ference on Applications of Statistics and Probability in RC members subjected to combined bending and
Civil Engineering (ICASP9), Civil Engineering Risk shear. In: Concrete shear in earthquake. Elsevier
and Reliability Association (CERRA), San Francisco, Applied Science, New York, pp 429–438
6–9 July Williams MS, Sexsmith RG (1997) Seismic assessment of
Priestley MJN, Verma R, Xiao Y (1994) Seismic shear concrete bridges using inelastic damage analysis. Eng
strength of reinforced concrete columns. J Struct Eng Struct 19(3):208–216
120(8):2310–2329 Wong HF (2005) Shear strength and seismic performance
Pujol S (2002) Drift capacity of reinforced concrete col- of non-seismically designed reinforced concrete
umns subjected to displacement reversals. PhD thesis, beam-column joints. PhD dissertation, Department of
Purdue University Civil Engineering, The Hong Kong University of Sci-
Ricci P, De Luca F, Verderame GM (2011) 6th April 2009 ence and Technology, August
L’Aquila earthquake, Italy: reinforced concrete build- Yavari S, Elwood KJ, Wu C (2009) Collapse of a
ing performance. Bull Earthq Eng 9(1):285–305 nonductile concrete frame: evaluation of analytical
Ricci P, Verderame G, Manfredi G (2013) ASCE/SEI models. Earthq Eng Struct Dyn 38(2):225–241
41 provisions on deformation capacity of older-type Youssef M, Ghobarah A (2001) Modelling of RC beam-
reinforced concrete columns with plain bars. ASCE column joints and structural walls. J Earthq Eng
J Struct Eng 139(12):04013014 5(1):93–111
Sezen H, Moehle JP (2004) Shear strength model for Zhu L, Elwood K, Haukaas T (2007) Classification and
lightly reinforced concrete columns. ASCE J Struct seismic safety evaluation of existing reinforced con-
Eng 130(11):1692–1703 crete columns. J Struct Eng 133(9):1316–1330
Seismometer Arrays 3211
the different array sites after applying appropriate

Seismometer Arrays phase delays. With these so-called “beamforming”
techniques, arrays show superior signal detection
Johannes Schweitzer capabilities with respect to single 3-component
NORSAR, Kjeller, Norway (3C) seismic stations. The second main purpose
to install a seismic array is the capability to esti-
mate the station-to-seismic source azimuth
Synonyms (backazimuth, BAZ) and the apparent velocity of
unknown seismic signals crossing the array.
Arrays; Beamforming; fk-analysis; Seismic Thirdly, a single seismic array often provides
arrays enough information about the incoming seismic
signals that an automatic algorithm can be used
to estimate a first (preliminary) solution of the
Introduction seismic source location.
Definition and Purpose History of Seismometer Arrays

In seismology, the term “array” (of seismometers) The history of seismometer array installations is
has been known for more than 50 years. Today this going back to the 1950s. At that time, the idea of
phrase is often used for any group of seismic installing arrays of sensors to improve the signal-
stations in the sense as defined in Macmillan’s to-noise ratio (SNR) of a seismic onset was
online dictionary, where an array is described as adopted from radio astronomy, radar, acoustics,
“a number of pieces of equipment of the same and sonar. Since then, classified arrays deployed
type, connected together to do a particular job” to monitor nuclear test activities at teleseismic
(Macmillan 2013). In this chapter, the term “seis- distances have been built worldwide. In 1958,
mometer arrays” is used with a more restricted “The Conference of Experts to study the methods
definition, as it was originally introduced to seis- of detecting violations of a possible agreement on
mology. A seismometer array can be defined as the suspension of nuclear tests” was held under
(Schweitzer et al. 2012): “A seismic array is a set the auspices of the United Nations in Geneva.
of seismometers deployed so that characteristics of This conference was followed up by several ini-
the seismic wavefield at a specified reference tiatives for improving the quality of seismic sta-
point, within or close to the array, can be inferred tions worldwide. Many of these classified arrays
by analyzing the waveforms recorded at the differ- became known in the 1990s and are today part of
ent sites. A seismic array differs from a local net- the International Monitoring System (IMS) for
work of seismic stations mainly by the techniques the Comprehensive Nuclear-Test-Ban Treaty
used for data analysis. Thus, in principle, Organization (CTBTO) (see e.g., Douglas 2002; S
a network of seismic stations can be used as an Dahlman et al. 2009) as primary or auxiliary
array, and data from an array can be analyzed as stations, e.g., AKASG (Malin, Ukraine), ASAR
data from a network. The size of an array is defined (Alice Springs, Australia), BRTR (Keskin, Tur-
by its aperture, which is the largest horizontal key), CMAR (Chiang Mai, Thailand), ESDC
distance between two sensors of the array. In prac- (Sonseca, Spain), ILAR (Eielson, Alaska), and
tice, the geometry and the number of seismometer KURK (Kurchatov, Kazakhstan). Many of these
sites of an array are determined by the intended arrays have quite diverse geometries (Fig. 1) and
scientific purpose and economic limits.” The main in some cases comprise different installations for
purpose to install a seismic array is threefold. At short-period (frequencies above 0.5 Hz) and
first, an array can be steered as an antenna to long-period (frequencies below 0.1 Hz) signals
amplify the signals of interest by stacking as, e.g., the Belbaşi array nearby the Keskin
(summing) the seismic wave energy recorded at array. Similar to arrays of seismometers, the
3212 Seismometer Arrays
Seismometer Arrays, Fig. 1 Seismic arrays equipped with which it is registered in the international registry of
with short-period or broadband seismic sensors in opera- seismic stations at the International Seismological Centre
tion in December 2013 as part of the International Moni- (http://www.isc.ac.uk/registries/) is provided for each
toring System (IMS) for the Comprehensive Nuclear- array. All array maps are plotted in the same scale
Test-Ban Treaty Organization (CTBTO). The code name (Courtesy of S. J. Gibbons, NORSAR)
IMS includes also arrays of infrasound sensors (known as UK array) is orthogonal linear or
and hydrophones, and similar data analysis tech- L-shaped, with apertures between 10 and 25 km.
niques are applied as for data from seismic arrays. Later, arrays of the same type were built in Aus-
To the best of knowledge, the first experimen- tralia (Warramunga), Brazil (Brasilia), and India
tal seismic array with more than four elements (Gauribidanur).
and openly available data was established in Feb- In the 1960s, arrays with very different aper-
ruary 1961 by the United Kingdom Atomic tures and geometries were tested, from small
Energy Agency (UKAEA) on the Salisbury circular ones with apertures of a few kilometers
Plain (UK), followed in December 1961 by the to huge arrays with apertures of up to 200 km.
Pole Mountain array (PMA, Wyoming, USA), in The largest arrays were the Large Aperture Seis-
June 1962 by the Eskdalemuir array (EKA, Scot- mic Array (LASA) in Montana (USA), opened in
land, UK), and in December 1963 by the Yellow- 1965 and in operation until 1978 with 525 seis-
knife array (YKA, Canada). This array type mometer sites, and the original Norwegian
Seismic Array (NORSAR) in southern Norway, characterization of seismic signals that was supe-
consisting of 132 sites over an aperture of approx- rior to that of single three-component
imately 100 km with altogether 198 seismome- (3C) stations. Today, many of the seismic stations
ters, which became fully operational in spring of the IMS for the CTBTO are arrays (e.g.,
1971. The original NORSAR array was reduced Dahlman et al. 2009).
in 1976 to seven subarrays and was assigned the During the last two decades, temporary, very
new code name NOA. small aperture arrays (apertures usually below
LASA, NORSAR, and the UKAEA arrays had 1 km) were used to investigate the distribution
narrowband short-period seismometers (for sig- of (mostly) S-wave velocities below and nearby
nal frequencies around 1 Hz) at all sites and the array installation. Knowing S-wave velocities
additional long-period seismometers (for signal of the uppermost layers is essential for seismic
periods around 20 s) at selected sites in their hazard mitigation; for details see, (e.g., Schweit-
original configurations. In Germany, a new zer et al. 2012).
array type was planned and installed in the Further details about different array configu-
1970s with an aperture of about 110 km. The rations can be found in the literature (e.g., Barber
Gräfenberg-Array (GRF) was installed on 1958; Haubrich 1968; Harjes and Henger 1973;
the limestone plateau of the Franconian Jura as Mykkeltveit et al. 1983). As examples of seismic
the world’s first seismometer array equipped arrays, Fig. 1 shows maps of all primary and
entirely with broadband sensors (for frequencies auxiliary IMS arrays in operation in December
between 0.01 and 8 Hz). Since then, the short- 2013; note the huge variety in geometries, num-
period and long-period sensors of many arrays ber of sites, and apertures.
were or will be in the near future exchanged
with broadband seismometers.
In the 1980s, the geometry of the so-called Basics of Array Seismology
regional arrays was developed. This is often
called a NORES-type array design and has array Basic Requirements for Seismometer Arrays
sites located on concentric rings (each with an The observed apparent-velocity range and the
odd number of sites) spaced at log-periodic inter- dominant frequency content of seismic signals
vals. It is now used for the design of most mod- are quite different for different types of signals
ern, small aperture arrays; only the number of (i.e., local, regional, or teleseismic observations,
rings and the aperture (diameter of the outermost body waves, or surface waves). Thus, the geom-
ring of sites) differ from installation to etry, aperture, and instrumentation of seismic
installation. arrays have to be adjusted with respect to the
Another approach to seismometer arrays was scientific requirements of an array installation.
developed in the 1990s. In parts of the world, the Proper analysis of array data depends on S
networks of seismometer stations became so a stable, high-precision relative timing of all
dense that data from these single station networks array elements. This is required because most of
could be combined and analyzed as data from the parametric information calculated using an
a seismic array. Examples are the J-array in array involves the measurement of (usually very
Japan, the German Regional Seismic Network, small) time differences (phase shifts) between the
the Californian array, the Kyrgyz Network seismic signals recorded at the different sensors.
(K-Net), and the ongoing USArray project with Most array-data analysis algorithms assume that
semi-temporary stations. Most of the known seismic energy crosses an array as a plane wave.
array techniques could be applied to analyze From the theory of signal processing, it is known
data from these and other networks. that signals can be constructively summed up as
Array installations demonstrated during the long as the time shift between the signals is
last decades of the twentieth century that seismic not larger than about ¼ of the dominant signal
arrays could facilitate detection and period T. This rule applies also to the case of
seismic signals observed with an array. Many lateral heterogeneities can have large effects.
different effects can influence the arrival time of To avoid this, many arrays are built on homo-
seismic onsets and thereby disturb the plane- geneous geological units, but this is not every-
wave approximation and should be taken in where possible, and uniform geology at the
account when analyzing array data: Earth’s surface does not guarantee that deeper
structures are similarly homogeneous. There-
• Seismic waves usually propagate with spheri- fore, it is necessary to investigate recorded
cal (in the case of body waves) or circular data from each seismic array for systematic
geometry (in the case of surface waves). effects which may disturb the plane-wave
Therefore, the aperture of an array should be approximation.
small enough that the plane-wave approxima- • There are many examples of how timing-
tion is still valid, i.e., the (theoretical) arrival- system errors at single array sites can produce
time differences with respect to the plane- difficulties for array-data analysis. However,
wave approximation should be so small that stable and correct relative timing can be
they do not influence the analysis results. achieved by installing at the array a central
• Since the fundamental work of Hans Benndorf clock, which distributes a common time signal
(1870–1953) and Emil Wiechert (1861–1921) to all digitizers at all array sites. Even in the
during the first decade of the twentieth cen- case that this central timing system fails and has
tury, it is known that in the case of a body some offset to absolute time, all array-analysis
wave, there is a direct relation between the algorithms can still be applied. So, whenever it
observed (apparent) propagation velocity of is possible, a central timing system should be
a seismic wave, the ray parameter (the deriv- considered for new array installations.
ative of the travel-time curve), and the seismic
velocity at the turning point of the observed The previously mentioned “beamforming”
seismic wave within the Earth. With an array techniques delay (phase shift) and stack the signals
of seismometers, this (apparent) velocity can from the different sensors. Due to constructive
be directly measured as long as all sensors are interference of the signals, the SNR is then
located in one horizontal plane. enhanced, whereas the (random and uncorrelated)
In the case that the sensors are not located background noise is suppressed. One can show
on one horizontal plane, the theoretical time that a seismic arraypcan
ffiffiffiffiffi theoretically improve the
shifts between the array sensors do not only SNR by a factor of M, where M is the number of
depend on the propagation direction and array elements. Figure 2 shows an example from
(apparent) velocity of the plane wave but the GERES array in Germany. The tiny onset of
also on the wavelength-dependent seismic a P phase, which was reflected at the Earth’s core,
velocities below the array sites. If travel-time was recorded with vertical sensors at 20 sites of the
effects due to array topography become larger array. Figure 2 shows these data in blue and on top
than about one fourth of the signal period, it the array beam in red. On many single traces the
becomes important for the accuracy of data signal is hardly discernible, but it is clearly visible
analysis results to take this effect in account. on the beam trace due to the drastically improvedpffiffiffiffiffi
However, analysis algorithms then become SNR. To achieve the theoretical factor of M ,
rather complicated, and if one plans for most array-data processing techniques require not
a new array, one should try to avoid this addi- only high clock stability of the recording systems
tional complexity and locate all seismometer at the different array sites and best knowledge
sites on one horizontal plane. about possible deviations from the plane-wave
• The seismic wavefield is always influenced by approach but also high signal coherency across
lateral heterogeneities in the Earth, which can the array.
disturb the plane-wave approximation due to Waveforms of seismic signals can be
velocity inhomogeneities. In particular local influenced by interference with wave energy
Seismometer Arrays, Fig. 2 Vertical seismograms about 1,070 km. The top trace (red) shows the array beam
(blue) with the onset of a tiny P onset reflected from the of the shown vertical records after applying “delay-and-
Earth’s core (PcP) and recorded at 03:29:29.5 on sum” processing. All data are Butterworth band-pass fil-
13 December 1990 with the GERES array after an earth- tered between 0.8 and 4 Hz and equally scaled
quake in the Tyrrhenian Sea at an epicentral distance of
scattered at heterogeneities along the entire ray the propagation direction and the apparent velocity
path, and as a result, waveforms observed at of a wave front can be measured – improves with
different seismometer sites may significantly dif- increasing array aperture. However, the apparent
fer. Since high signal coherency is required for velocity of a seismic onset changes with epicentral
seismic arrays, such waveform-altering effects distance and the shape of a seismic waveform can
should be minimized by avoiding array locations change drastically due to interference with other
close to known lateral inhomogeneous structures. seismic phases; in addition, signal coherency
This constitutes an additional reason to prefer diminishes with increasing sensor separation.
installing seismic arrays at one relatively homo- Therefore, when building an array, one has to
geneous geological unit, almost transparent to find a balance between coherency and theoretical S
seismic waves. Since such local site effects are slowness resolution.
signal frequency dependent, signal coherency Finally, some remarks about the instrumenta-
over the array also becomes frequency depen- tion of arrays. Traditionally, arrays were equipped
dent, and this may impose constraints on array with vertical sensors at all sites and only a few sites
geometry, spatial extent, and instrumentation. with additional horizontal sensors. This was
As mentioned earlier, seismometer arrays can mostly related to the additional costs for
provide estimates of the station-to-event azimuth more sensors, data transmission, and data storage.
(backazimuth, BAZ) and the apparent velocity of However, many studies have shown that arrays
seismic signals. These estimates are important equipped with more 3C sensors are superior in
both for event location purposes and for signal analyzing S-type onsets. Since costs for data trans-
identification and classification, e.g., as P, S, mission and storage have drastically dropped, one
local, regional, or teleseismic phases. The slow- should always consider installing 3C instruments
ness resolution of an array – i.e., how accurately at all sites for new array installations.
The Array-Transfer Function 2. The number of sites controls the quality of the
In signal-processing theory, the process of array as a wavenumber filter, i.e., its ability to
steering a seismic array to a specific target, i.e., suppress seismic energy crossing simulta-
optimizing the array beam for a given BAZ and neously the array with a different slowness
apparent velocity of a seismic wavefield, can be than the one on which the array is steered.
described as a linear filter, which allows only 3. The distances between the seismometers
signals with these characteristics to pass. In the define the position of the side lobes of the
case of an array, it is a two-dimensional filter, array-transfer function and the largest resolv-
which is defined in the frequency-wavenumber able wavenumber: the smaller the mean dis-
(fk) space. The wavenumber k of a signal is tance, the smaller the wavelength of
defined as k = 2pf/v or k = 2p/l, where f is the a resolvable seismic phase will be (for
frequency, l is the wavelength, and v is the appar- a given apparent velocity).
ent velocity of the signal. Filter-transfer functions 4. The geometry of the array defines the azimuth
are mathematical descriptions of the filter char- dependence of the aforementioned points.
acteristics (bandpass filter). The two-dimensional
array filter is described by the sharpness of its Figure 3 shows two examples of such array-
main lobe (maximum of passing signals) and the transfer functions, the ARCES array in northern
position and relative height of eventually existing Norway and the Yellowknife array in northern
side maxima (side lobes). The narrower the main Canada for a 1 Hz signal. The geometry and
lobe and the smaller are eventual side lobes, the aperture of these two arrays is included in
better the array’s performance as a filter for sig- Fig. 1. The ARCES shows no differences for
nals with very specific BAZ and apparent veloc- signals from different azimuths, and the side
ity. For more details on the theory of seismic lobes of the transfer function are far away
arrays, one may look in, e.g., Capon (1973), Aki (outside the plot) from the main lobe. However,
and Richards (1980), Buttkus (2000), Johnson and because of the small aperture of only 3 km, this
Dudgeon (2002), Schweitzer et al. (2012), and the array cannot distinguish between waves with
citations therein. By comparing array-transfer small wavenumber differences, as can be seen
functions, the quality of an array of seismometers in the relatively wide main lobe of the transfer
as a frequency-wavenumber filter can be discussed. function. In contrast, in the case of Yellowknife,
The amount of literature about general criteria used the main lobe is very narrow because of the much
to evaluate array-transfer functions for a given larger aperture of the array of about 25 km. This
array geometry is huge, and some suggestions results in a much higher resolution in measuring
for more details are, e.g., in Harjes and Henger small apparent-velocity differences. However,
(1973) or Schweitzer et al. (2012) and the refer- the array shows resolution differences for the
ences therein. However, some general rules about different azimuths: the many side lobes of the
transfer-function characteristics of seismometer transfer function along north-south and east-
arrays can be formulated as follows (from Harjes west oriented lines are caused by its cross-shaped
and Henger 1973; and Schweitzer et al. 2012): geometry and the relative large distances
between the single array sites.
1. The aperture a of an array defines the resolu- For further reading, reviews on array theory,
tion of the array for small wavenumbers k. The together with quite comprehensive citation lists,
larger the aperture is, the smaller the see, e.g., Douglas (2002), Rost and Thomas
wavenumbers that can be measured with the (2002, 2009), and Schweitzer et al. (2012).
array. The upper limit for the longest wave-
length l that can be resolved by array tech- Data Analysis Algorithms for Seismometer
niques is approximately similar to the aperture Arrays
of the array. The array behaves like a single One standard analysis technique for array data is
station for signals with l a. the earlier described beamforming algorithm,
Seismometer Arrays, Fig. 3 The array-transfer func- relative power (color coded) of the array response nor-
tions of the circular ARCES array (left) and of the cross- malized with its maximum in [db] (Modified from
shaped Yellowknife array (right) for a 1 Hz signal as Schweitzer et al. 2012)
also referred to as “delay-and-sum” processing. signal classification (MUSIC) algorithm

In seismic prospecting, “beamforming” is called (Schmidt 1986), and various adaptive algorithms
“stacking.” An extension of this concept is the for the estimation of spectral power density (e.g.,
so-called “double-beam” technique, in which Goldstein and Archuleta 1991). Later, fk-analysis
not only data of seismometer arrays are stacked was expanded to wider frequency bands and 3C
but also the array beams of different seismic data (Kværna and Doornbos 1986), the inversion
sources closely located to each other (Kr€uger for spherical waveforms (Almendros et al. 1999),
et al. 1993). and arrays with incoherent data (Gibbons
Another standard array-data analysis tech- et al. 2008, 2012).
nique is frequency-wavenumber analysis One can also directly measure signal-arrival
(fk-analysis). This method, originally developed times at the different array sites and invert the
for narrowband filtered one-component data, observed travel-time differences for the best
determines the slowness vector (i.e., BAZ and fitting plane wave. The (relative) arrival times
apparent velocity) for a single signal at a single can be either determined by an analyst or by
frequency. Fk-analysis can be performed either in automatic arrival-time-picking algorithms, as, S
the frequency domain or the time domain (see e.g., by correlation analysis (Cansi 1995).
Schweitzer et al. 2012). Fk-analysis applies sys- One widely applied technique to analyze data
tematically different phase shifts (or delay times) observed with an array is the VElocity SPectrum
to a selected time window of the different array- Analysis (VESPA) process (Davies et al. 1971),
site data and calculates the power of the seismic also called vespagram. This algorithm is very
wave. By searching for the maximum power, the similar to the time-domain version of the
algorithm estimates the corresponding BAZ and fk-analysis, but it measures the power of seismic
apparent velocity of the signal. A variety of signals from a constant BAZ with different appar-
fk-analysis algorithms have been developed, ent velocities as function of time to investigate
which mostly differ in the method applied to apparent-velocity changes for the different onsets
find this maximum: e.g., maximum-likelihood in a seismogram. Later, the vespagram concept
method (Levin 1964; Capon et al. 1967, 1973), was expanded by calculating the observed power
maximum-entropy method (Burg 1964), multiple from different azimuths for a constant, specific
apparent velocity. In this case, the vespagram is This technique became recently quite important
a very useful tool to investigate the BAZ of the in context with seismic risk and hazard studies.
seismic phases, their precursors, and in particular Since ambient noise mostly consists of surface-
the scattered energy of seismic waves arriving wave energy, one can use local dispersion-curve
after seismic onsets. data, measured with temporary, very small aper-
ture arrays, to invert for the near-surface
S-velocity structure; for further details see, e.g.,
Seismological Research with Schweitzer et al. (2012).
Seismometer Arrays Seismic arrays are also used to investigate the
nature and source regions of microseisms and to
There are numerous publications from authors locate and track volcanic tremor for analyzing
using data of seismometer arrays, which cannot complex seismic wavefield properties in volcanic
be discussed in detail herein. In the following, areas.
only some general topics are named to illustrate The capability of seismometer arrays to locate
the diversity of array-seismology applications. seismic events was already mentioned at the
Seismic array data have been used to detect beginning of this chapter. The data analysis of
and investigate all kinds of small amplitude arrays can be automated for signal detection and
phases: seismic onsets observed in front of seis- fk-analysis of the detected onsets. Then, BAZ
mic phases, which had traversed the Earth’s core observations can be used together with additional
(PKP), were often interpreted as regular seismic travel-time constrains to automatically group the
phases, which led to many, quite controversial onsets to events. The observed apparent veloci-
velocity models of the lower part of the outer ties can be used to classify the type of seismic
core of the Earth. Using array observations it onset, together with other signal characteristics,
could be shown that these precursors are scattered as, e.g., the dominant frequency. This way, P- and
energy from lateral heterogeneities in the Earth’s S-type onsets from local and regional events can
interior and not regular seismic phases. There be automatically identified with the use of small
exist numerous array-data studies of the global aperture, regional arrays and the associated seis-
and local structure of the lower mantle, the later- mic event can be located. A detailed description
ally heterogeneous lowermost part of the mantle of such an automatic regional event location
directly above the core-mantle boundary (known algorithm can be found in (Schweitzer
as D” region), the core-mantle boundary, and the et al. 2012).
inner-core boundary. Apparent-velocity observations from seismic
Also strong-motion instruments have been arrays of at least 10 km aperture can be directly
installed in array configurations at the Earth’s inverted to an epicentral distance for the first
surface, as well as in boreholes to study near- arriving P-type onsets from seismic events at
field effects of earthquakes, and arrays were teleseismic distances (i.e., from about 22 to
used to track the aftershock activity after larger about 100 epicentral distance). Knowing the
earthquakes or to monitor earthquake swarms. In epicentral distance, the observed BAZ can then
combination with cross-correlation techniques, be used to define the epicentral coordinates. This
this became a very powerful tool to lower the algorithm cannot be applied to local or regional
detection thresholds for seismic events in mon- events because measurable apparent velocities
itored areas. Similar is the quite recent approach of the P-wave onsets can no longer directly be
to track the earthquake slip process itself with inverted to epicentral distance. At distances
array data observed at teleseismic distances. For beyond ~90 , these derivatives become again
such studies data from seismic networks are very small for P waves, and in the Earth’s
combined to virtual arrays and analyzed. shadow zone (distance >100 ), the interpreta-
Seismic arrays have been used to measure tion of the different core-phase onsets is also
dispersion curves of surface-wave velocities. quite difficult and limits the location capabilities
of a seismic array. The described event location References

technique has been in use at least since the
1960s, and a quick look in the bulletins of Aki K, Richards PG (1980) Quantitative
seismology – theory and methods, vol II. Freeman
the International Seismological Centre shows
and Company, San Francisco, pp 609–625. ISBN
the huge amount of reported teleseismic event 0-7167-1059-5
locations made with, e.g., the Large Aperture Almendros J, Ibáñez JM, Alguacil G, Del Pezzo E (1999)
Seismic Array (LASA) in Montana, USA, the Array analysis using circular-wave-front geometry: an
application to locate the nearby seismo-volcanic
Yellowknife Array (YKA) in Northern Canada,
source. Geophys J Int 136:159–170
the Gräfenberg Array (GRF) in Bavaria, Ger- Barber NS (1958) Optimum arrays for direction finding.
many, or the large Norwegian Seismic Array N Z J Sci 1:35–51
(NOA) in southern Norway. Burg JP (1964) Three-dimensional filtering with an array
of seismometers. Geophysics 29:693–713
For further reading, more details on the
Buttkus B (2000) Spectral analysis and filter theory in
research topics mentioned herein and for many applied geophysics. Springer, Berlin, xv + 667
more array applications in seismology, together pp. ISBN 3-540-62674-3
with quite comprehensive citation lists, see, e.g., Cansi Y (1995) An automatic seismic event processing for
detection and location: P.M.C.C. method. Geophys
Gibbons et al. (2008, 2012), Rost and Thomas
Res Lett 22:1021–1024
(2002, 2009), and Schweitzer et al. (2012). Capon J (1973) Signal processing and frequency-
wavenumber spectrum analysis for a large aperture
seismic array. In: Bolt BA (ed) Methods in computa-
tional physics, 13 geophysics. Academic, New York,
Summary xiii + 473 pp. ISBN 0-12-460813-2, 1–59
Capon J, Greenfield RJ, Kolker KJ
A seismometer array is a set of seismic sensors (1967) Multidimensional maximum-likelihood
deployed so that characteristics of the seismic processing of a Large Aperture Seismic Array. Proc
IEEE 55(2):192–211
wavefield at a specified reference point can be
Dahlman O, Mykkeltveit S, Haak H (2009) Nuclear test
inferred by analyzing together the waveforms ban – converting political visions to reality. Springer,
recorded at the different array sites. Seismome- Dordrecht. doi:10.1007/978-1-4020-6885-0, xviii +
ter arrays have been shown to be superior to 277 pp. ISBN 978-1-4020-6883-6
Davies D, Kelly EJ, Filson JR (1971) The VESPA
single, 3C stations: they are able to lower the
process for the analysis of seismic signals. Nature
detection threshold for seismic onsets due to 232:8–13
beamforming; they are able to measure the Douglas A (2002) Seismometer arrays – their use in earth-
apparent velocity of seismic onsets and their quake and test ban seismology. In: Lee WHK,
Kanamori H, Jennings PC, Kisslinger C (eds) Hand-
propagation direction; they can be used to locate
book of earthquake and engineering seismology Part
seismic events in an automated data processing A. Academic, Amsterdam, xxiii + 933 pp. ISBN 0-12-
scheme; and they are very successful tools to 440652-1, 357–367
investigate the Earth’s interior on local, Gibbons SJ (2012) The applicability of incoherent array
processing to IMS seismic arrays. Pure Appl Geophys.
S
regional, or global scale.
doi:10.1007/s00024-012-0613-2
Gibbons SJ, Ringdal F, Kværna T (2008) Detection and
characterization of seismic phases using continuous
Cross-References spectral estimation on incoherent and partially coher-
ent arrays. Geophys J Int 172:405–421. doi:10.1111/
▶ Earthquake Location j.1365-246X.2007.03650.x
Goldstein P, Archuleta RJ (1991) Deterministic
▶ Earthquake Swarms
frequency-wavenumber methods and direct measure-
▶ Principles of Broadband Seismometry ments of rupture propagation during earthquakes using
▶ Recording Seismic Signals a dense array: theory and methods. J Geophys Res
▶ Seismic Event Detection 96:6173–6185
Harjes H-P, Henger M (1973) Array-Seismologie.
▶ Seismic Network and Data Quality
Z Geophys 39:865–905
▶ Seismic Noise Haubrich RA (1968) Array design. Bull Seismol Soc Am
▶ Time History Seismic Analysis 58:977–991
3220 Seismometer Self-Noise and Measuring Methods
Johnson DH, Dudgeon DE (2002) Array signal Introduction

processing: concepts and techniques, Prentice Hall
signal processing series. PTR Prentice Hall, Upper
Saddle River, XIII + 533 pp. ISBN 0-13-048513-6 Seismometer self-noise is usually not considered
Kr€uger F, Weber M, Scherbaum F, Schlittenhardt J when selecting and using seismic waveform data
(1993) Double beam analysis of anomalies in the in scientific research as it is typically assumed
core-mantle boundary region. Geophys Res Lett 20: that the self-noise is negligibly small compared
1475–1478
Kværna T, Doornbos DJ (1986) An integrated approach to to seismic signals. However, instrumental noise
slowness analysis with arrays and three-component is part of the noise in any seismic record, and in
stations. NORSAR Sci Rep, 2-85/86, 60–69 particular, at frequencies below a few mHz, the
Levin MJ (1964) Maximum-likelihood array processing. instrumental noise has a frequency-dependent
Lincoln Laboratory report, 31 Dec 1964
Macmillan (2013) Array. http://www.macmillan- character and may dominate the noise. When
dictionary.com/dictionary/british/array. Accessed seismic noise itself is considered as a carrier of
08 Dec 2013 information, as in seismic interferometry (e.g.,
Mykkeltveit S, Åstebøl K, Doornbos DJ, Husebye ES Chaput et al. 2012), it becomes extremely impor-
(1983) Seismic array configuration optimization. Bull
Seismol Soc Am 73:173–186 tant to estimate the contribution of instrumental
Rost S, Thomas C (2002) Array seismology: methods and noise to the recordings.
applications. Rev Geophys 40(3):1008. doi:10.1029/ Noise in seismic recordings, commonly
2000RG000100 called seismic background noise or ambient
Rost S, Thomas C (2009) Improving seismic resolution
through seismic arrays. Surv Geophys 30:271–299. Earth noise, usually refers to the sum of the
doi:10.1007/s10712-009-9070-6 individual noise sources in a seismic recording
Schmidt RO (1986) Multiple emitter location and signal in the absence of any earthquake signal. Site
parameter estimation. IEEE Trans Ant Prop noise (e.g., cultural sources, nearby tilt signals,
AP-34:276–280
Schweitzer J, Fyen F, Mykkeltveit S, Gibbons SJ, Pirli M, etc.) and noise introduced by the sensitivity of an
K€uhn D, Kværna T (2012) Seismic arrays. http:// instrument to non-seismic signals (e.g., temper-
ebooks.gfz-potsdam.de/pubman/item/escidoc 43213:7. ature and pressure variations, magnetic field
doi:10.2312/GFZ.NMSOP-2_ch9; 80 pp. In: Bormann changes, etc.) both contribute to the ambient
P (ed) New manual of seismological observatory
practice (NMSOP-2), 2nd (rev) edn. Deutsches seismic noise levels. The background noise
GeoForschungsZentrum GFZ, Potsdam. doi:10.2312/ ultimately defines a lower limit for the ability
GFZ.NMSOP-2. Accessed 08 Dec 2013 to detect and characterize various seismic
signals of interest. Background noise levels
have also been found to introduce a systematic
bias in arrival times because the amplitude of
Seismometer Self-Noise and the seismic phase must rise above the station’s
Measuring Methods noise levels (Rӧhm et al. 1999). The upper
limit of useful signals is governed by the clip
Adam T. Ringler1, Reinoud Sleeman2, level of the recording system (the point at
Charles Robert Hutt1 and Lind S. Gee1 which a recording system’s output is no longer
1
Albuquerque Seismological Laboratory, a linearly time-invariant representation of the
U.S. Geological Survey, Albuquerque, NM, USA input).
2
Seismology Division, Royal Netherlands Site noise can be reduced by careful site
Meteorological Institute (KNMI), De Bilt, selection (e.g., hard rock far from strong noise
Netherlands sources) and by emplacing instruments in good
vaults or boreholes. It is also possible to reduce
sensitivity to non-seismic signals by thermal
Synonyms insulation and appropriate shielding such as
pressure chambers (Hanka 2000). At quiet sites
Coherence analysis; Instrument noise; Seismom- with well-installed instrumentation, instrument
eter testing noise may be the dominant noise source
Seismometer Self-Noise and Measuring Methods 3221
(Berger et al. 2004); this is especially true for is possible to obtain an estimate of the self-noise by
long-period seismic data (>100 s period) on locking the mass of the instrument.
very broadband instruments (e.g., Streckeisen This is a general overview on the various
STS-1 seismometer). The interpretation of such methods currently used for estimating the instru-
data only makes sense if the instrumental noise ment self-noise when using one, two, or three
level is known. Also, research on noise levels in sensors. As there is no universally “best” method
seismic recordings, the effect of noise reduction for all types of instruments and test conditions,
by the installation technique, and the nature and some of the advantages and disadvantages of
contribution of different noise sources to the each method are discussed. Variants of these
recordings require knowledge of instrumental methods, such as rotating horizontal compo-
self-noise to rule out bias from the instrumenta- nents of Earth motions to maximize the coher-
tion self-noise. ence and correcting for misalignments, are also
A number of tests have been developed, under included. Finally, the test setup used for estimat-
the assumption that instrument self-noise is ing self-noise for the various methods is
approximately constant as a function of time discussed and examples of corresponding test
(e.g., Evans et al. 2010; Holcomb 1989, 1990; results are given.
Sleeman et al. 2006). Only recent studies have
started to look at the potential time dependence
of instrument self-noise (e.g., Sleeman and Methods
Melichar 2012). Understanding the self-noise of
a given piece of recording equipment helps sta- To discuss the various methods currently in use
tion operators to better identify sites that can take for estimating the self-noise of a seismometer, a
advantage of low-self-noise instruments. This mathematical framework common to all the
knowledge also allows a network operator to methods is developed. The system under test is
provide higher-quality data with limited assumed to be a linear time-invariant system
resources by making better use of their high- (LTI) making the system completely determined
quality instruments. Having a rough understand- by its impulse response (Scherbaum 2007).
ing of an instrument’s self-noise also gives a first-
order diagnostic for determining if the recording Basic Assumptions and Conventions
system, as installed, is performing satisfactorily. Let xi denote an input seismic signal, hi denote the
In order to estimate the self-noise of a seismom- seismometer’s impulse response, and ni denote
eter, it is necessary to remove non-instrumental the self-noise of instrument i. Then the instru-
noise signals (e.g., earthquakes and ambient ment’s output signal yi can be modeled, in the
Earth noise) from the data. This is often accom- time domain, as
plished by using coherence analysis techniques. In S
its simplest form, one can select a quiet time period y i ¼ hi ð x i þ ni Þ (1)
at a low-noise site and attribute all recorded noise
to the sensor (one-sensor method). However, it can where “*” denotes convolution (Holcomb 1990;
be difficult to find stations with sufficiently low Sleeman et al. 2006). It is possible to write this in
site noise to evaluate high-quality broadband the frequency domain as
seismometers. This is especially true at periods
between the primary and secondary microseism Y i ¼ H i ðXi þ N i Þ; (2)
(approximately 4–22 s period). In such cases,
more sophisticated techniques, where one removes where capital letters denote the Fourier trans-
coherent signals using colocated instruments, are forms of the corresponding lowercase letter time
required. This can be done using a second instru- domain terms of index i. With the assumption that
ment or two additional instruments (two- and the self-noise of two different instruments is inco-
three-instrument methods). For passive sensors, it herent, the term Nij = NiNj is zero for i 6¼ j. It is
Seismometer Self-Noise and Measuring Methods, depict the 95 % and 68 % percentile power spectral den-
Fig. 1 Self-noise estimates of the vertical component of sity (PSD) estimates. Finally, median (blue) and minimum
the STS-2 seismometer for the minimum noise (green) (black) self-noise estimates for the STS-2 are taken from
and the mode (red) of sensor noise measured at the Conrad the data at the Albuquerque Seismological Laboratory
Observatory, Austria. The light gray and dark gray bands (ASL)
also assumed that the self-noise and the input noise models derived from the 74 Global Seismo-
signal are incoherent, XiNj is zero for i 6¼ j. graphic Network (GSN) stations along with one
The cross power between instrument i and j is additional station (Peterson 1993). This NLNM
denoted by Pij so for i 6¼ j we can write the cross represents the lower envelope of 2,000 data
power as records and represents an approximation of the
lowest observed seismic noise levels, whereas the
Pij ¼ Hi Xi Hj Xj (3) NHNM represents the upper envelope of the
same 2,000 data records and represents an
where the bar denotes the complex conjugate. approximation of the highest observed seismic
The coherence between instruments i and j is noise levels (Fig. 1).
given by
2 Single-Sensor Method
Pij When testing instruments at locations where the
g ¼
2
(4)
Pii Pjj site noise is well below the self-noise of the
instrument, it is possible to attribute the power
Finally, it is assumed that all of the instruments in a given frequency band entirely to the instru-
in the test have common output units so that they ment’s self-noise (Fig. 2). This often occurs when
have similar ground motion units (e.g., m/s or testing strong-motion accelerometers in a quiet
m/s2) after removing the instrument response. vault or lower-grade sensor in almost any good
This assumption can generally be made true by site (Evans et al. 2010). In such cases the simple
using an “omega correction” (multiplying or relation is obtained:
dividing by o = 2pf, where f is the frequency)
in the frequency domain (Stearns 1975). Pii
Nii ¼ (5)
Noise levels are often compared to absolute Hi Hi
Earth noise models such as Peterson’s New
Low-Noise Model (NLNM) and New High- assuming that Xi Nii. Even when this assump-
Noise Model (NHNM), which are global seismic tion is not satisfied across the entire frequency
Seismometer Self-Noise
and Measuring Methods,
Fig. 2 Self-noise
estimates for a strong-
motion accelerometer using
the single-sensor method
(red), the Sleeman
(Sleeman et al. 2006) three-
sensor method (green), and
the Holcomb (Holcomb
1989) two-sensor method
(blue). For reference the
New Low-Noise Model
(NLNM) is included
(black)
band, it is often possible to get an initial estimate signals (same ground motion), Xi = Xj, it is
of the instrument’s self-noise in a specific fre- possible to derive the self-noise of instrument
quency band (e.g., outside the primary and sec- i from Eq. 2. Recalling the assumption that Nij =
ondary microseism bands for state-of-the-art 0 for i 6¼ j we have
accelerometers). The single-sensor method can
also be used to get an upper bound on the self- Pii Pij
Nii ¼ 2
: (6)
noise of an instrument. If this upper bound is jH i j H i Hj
obtained at a relatively quiet location, it might
characterize the self-noise sufficiently well to Under the assumption that the sensors both have
identify the suitability of an instrument at a sta- well-known responses, the instrument corrected
tion with potentially much higher site noise. This output, from the two sensors, should only differ in
method gives a first approximation of an instru- instrumental self-noise. By estimating the coher-
ment’s self-noise, which can be valuable when ent signal between the two records, the coherence S
testing time is limited or there are insufficient signal is removed resulting in the incoherent sig-
resources to use multiple sensors. nal which is attributed to the self-noise. It can be
seen from this estimate of the self-noise that it is
Two-Sensor Methods critical to have well-described transfer functions
Estimating self-noise using two sensors has been for both instruments i and j. Since the transfer
the traditional method for noise testing broad- functions are used in the calculation, errors in
band seismometers for many years (Holcomb the transfer function will produce errors in the
1989). As very few locations have background self-noise estimates of the instrument. This
noise levels below that of high-quality broad- method was originally proposed by Holcomb
band sensors across a wide frequency range, it (1989), who later characterized error sources
becomes important to remove the local back- (Holcomb 1990). In the latter work, Holcomb
ground noise (Fig. 3). Assuming that the also suggested alternative two-sensor methods,
colocated sensors are recording similar seismic under the assumptions that the two sensors
Seismometer Self-Noise and Measuring Methods, method (red), and the two-sensor Tasič (Tasič and Runovc
Fig. 3 Comparison of five different methods for estimat- 2012) method (cyan) (for the different methods, see text).
ing the self-noise using vertical broadband sensor data: the The cyan and red lines overlay at frequencies less than
one-sensor (direct power spectral density (PSD)) method approximately 0.01 Hz. For reference the New Low-Noise
(blue), the three-sensor Sleeman (Sleeman et al. 2006) Model (NLNM) is the solid black line. The elevated noise
method (green), the two-sensor Holcomb (Holcomb in the microseism band is caused by misalignment
1989) method (light gray), the two-sensor g2 Holcomb
under test have equal noise and a high signal-to- Three-Sensor Method
noise ratio (ratio of input signal to instrument Using three colocated sensors, it is possible to
self-noise). In this case, it is possible to derive estimate the self-noise of each instrument while
the following estimate for the self-noise of instru- minimizing errors in the estimate due to uncertainty
ment i: in the transfer functions (Sleeman et al. 2006). This
method has become the preferred approach for

estimating self-noise for broadband sensors, even
ð1 g2 Þ
Pii though it requires additional resources and setup; it
g2
Nii ¼ : (7) typically “sees through” site noise more deeply to
Hi Hj
extract lower estimates of instrument noise.
Using Eq. 2 and the assumption that
Using a related approach, Tasič and Runovc
Xi = Xj = Xk, one can estimate the self-noise of
(2012) develop a different method for estimating
instrument i as
the self-noise using two instruments. In their
approach the self-noise of instrument i is given by
Pik
Pii Pij
Pjk
Pii ð1 g2 Þ Nii ¼ : (9)
N ii ¼ (8) Hi Hl
Hi HJ
The term Pik/Pjk can be viewed as a “relative
where again they assume the signal-to-noise ratio transfer function” Hij between instruments i and j.
is large. The three-sensor method is related to the
Fig. 4 Self-noise estimate
of the Geotech GS-13
geophone using a
theoretical self-noise model
(blue) versus the estimated
low self-noise model of the
Geotech GS-13 (red) from
11 self-noise
measurements; the median
self-noise from these
11 tests is also shown
(green). The self-noise of
the digitizers used,
Quanterra Q330HRs, is
also shown after correction
with the response of the
Geotech GS-13 (cyan). For
reference the New
Low-Noise Model
(NLNM) is included
(black)
two-sensor method developed by Tasič and Such theoretical methods for estimating the self-
Runovc (2012) when one compares relative trans- noise of an instrument have been applied to a
fer functions. It is important to note that the trans- number of seismometers in conjunction with their
fer function of instrument i is only used to convert digitizers (Rodgers 1994). However, when such
the noise estimate to units of ground motion; thus, methods are used, a number of assumptions must
errors in the transfer functions do not propagate to be made (e.g., that the pendulum obeys the small-
the self-noise estimate before deconvolution with angle approximation and all electronic components
the individual instrument (Ringler et al. 2011). perfectly match their specified noise levels).
Because the three-sensor method is not as sensitive For seismometers that are not controlled by
to errors in the transfer functions, it extracts the feedback but having self-noise well below any
self-noise with a potentially higher accuracy than available test site’s ambient noise, such as geo-
the two-sensor method (Fig. 3). phones at high frequency (e.g., GeoTech GS-13),
it is possible to estimate the self-noise using a
Other Techniques “locked mass” test (Havskov and Alguacil 2004). S
When instrument designers are selecting compo- This test is performed by locking the mass of the
nents for building a seismometer, they often have instrument and recording the output of a single
noise estimates for each electronic component as instrument. This test provides a lower noise-
well as estimates of fundamental noise contribu- bound estimate but may not fully characterize
tions (e.g., Brownian noise). Through modeling the self-noise of the instrument because it lacks
these individual noise sources, it is possible to noise contributions from moving hinges and
estimate the total self-noise of a seismometer with- other mechanical sources.
out having a working prototype. A modeling exam-
ple for Brownian noise is described by Aki and
Richards (2002), using the physical parameters of Testing and Analysis
a simple gravitational pendulum. Such methods
often give a first approximation of the instrumental Characterizing the self-noise of a seismometer
noise to be aimed for during design (Fig. 4). requires attention to detail in the test setup,
Fig. 5 Three colocated
Geotech GS-13 short-
period seismometers setup
for test to be sensitive to
vertical ground motion. All
three are sitting on a granite
slab supported at three
points and are in the
Albuquerque
Seismological Laboratory
(ASL) underground vault.
On the back right is a
reference Streckeisen
STS-2 seismometer in a
steel bell jar (not used in the
test)
careful selection of data windows to avoid tran- also contribute to a decrease in signal coherency
sients, identification of potentially non-seismic and increase the estimate of the self-noise. Such
noise sources, and methods for selecting data to relative errors in misalignment can be seen by
analyze. In both the two- and three-sensor elevated incoherent noise-level estimates in the
methods, it is necessary to make sure all instru- microseism band (Fig. 6), so it is possible to
ments truly are recording the same ground reduce such misalignment errors by rotating the
motion. instrument outputs numerically to maximize
coherence (Tasič and Runovc 2012). Recent
Test Setup work by Gerner and Bokelmann (2013) shows
Both the two- and three-sensor methods assume that the leakage of microseism signals into the
that all instruments are recording the same self-noise can be eliminated successfully by
ground motion. In order for this to be true, it is numerical rotation of the three-component traces
critical that the instruments be colocated and and using the three-sensor method. Their study
co-aligned. Furthermore, since local tilt signals optimally aligns two RefTek 151-60A sensors
can vary widely over even a few inches, one must with a third one and shows that self-noise esti-
take care to insure that all three instruments are mates outside the microseism bands are not
measuring similar tilt signals so that they can be compromised by misalignment errors.
removed using coherence analysis. Locating all Finally, if one wants to isolate the self-noise of
instruments on a stiff slab (e.g., granite or gabbro) a seismometer, it is critical that the digitizer and
with a three-point suspension is usually sufficient other recording equipment have self-noise levels
for this purpose (Fig. 5). well below that of the sensor. This can be a
It is also important to isolate the instruments problem on low-gain instruments, such as
from non-seismic noise sources (e.g., tempera- strong-motion accelerometers. Verifying that
ture variations, pressure variations, locally the digitizer’s self-noise levels are below that of
induced tilt). These noise sources can increase the seismometer requires a terminated-input test
the self-noise estimate of the individual instru- of the digitizer, using a terminating resistor sim-
ments as they generally do not respond coher- ilar to the output impedance of the sensors to be
ently to such sources (Anderson et al. 2012). tested. When the digitizer’s noise level is not
Misalignment of the horizontal components can below the seismometer noise level for any of the
Seismometer Self-Noise and Measuring Methods, installed with the same orientation on the same baseplate
Fig. 6 Self-noise estimates from three horizontal very and the baseplate is supported at three points only. The
broadband Metrozet M-2166 seismometers (flat to veloc- relatively small increase in the self-noise in the secondary
ity from 0.0028 to10 Hz) using Sleeman (Sleeman microseism frequency band (at approximately
et al. 2006) three-sensor self-noise estimate (solid lines). 0.125–0.25 Hz) suggests all three instruments are well
Single-sensor estimates are in dotted lines. In order to aligned to one another
minimize incoherent tilt noise, all three instruments were
frequency range of interest, digitizer preamps occurrences of periodic pulsing in the data are
generally can be used for low-noise amplification not yet understood (though many have been
of the sensor signal to a level above the digitizer named “popcorn noise,” “spherics,” and so forth
noise level. and are tentatively explained in various ways).
Some of these noise sources could properly be
Data Selection considered either as part of the self-noise of the
Data selection is currently an important item in instrument or as transient external signals that S
the development of algorithms and techniques to would artificially elevate the self-noise if
estimate self-noise. This requirement makes it included (Sleeman and Melichar 2012). The dis-
necessary to estimate self-noise noise during cussion on data selection of time segments
only “quiet” time periods (e.g., during nighttime becomes even more controversial when limited
and periods with low pressure-induced tilt noise). amounts of data are available. By using multiple
No current agreement exists as to whether data quiet time segments, it is possible to understand
selection should be applied or not, or which the self-noise of an instrument in at least a best-
criteria should be used for data selection. In par- case scenario (Ringler and Hutt 2010). One such
ticular, when features in the data are not fully example of a best-case scenario, along with a
understood, the debate on data selection is ongo- self-noise estimate using the median of multiple
ing and not yet entirely resolved, though a best- tests, is shown in Fig. 4. This example also sug-
case result seems to require such minimization gests that even when using coherency analysis
(Ringler et al. 2011). For example, the techniques, it is important to do such tests in a
low-noise environment. The selection of time To directly compare results between noise tests,
segments is additionally controversial when limit is important that the tests be conducted using
ited amounts of test data exist, because the similar methods (e.g., settling time, installation
assumption that the self-noise of a seismometer methods, data selection, and spectral processing
is time invariant is only approximately true, as parameters).
seen in the relatively large gray bands in Fig. 1,
which depict the 95 % and 68 % of the self-noise
of the Streckeisen STS-2 seismometer as Expected Results and Caveats
recorded under very stable conditions at the Con-
rad Observatory (Vienna, Austria) over an A discussion on the expected self-noise results
entire year. for various types of seismometers now follows.
A synthetic-data experiment by Sleeman and This discussion is in no way exhaustive but sim-
Melichar (2012) shows that the three-sensor tech- ply discusses some of the more common situa-
nique can reliably extract instrumental noise even tions and what is generally expected for various
for high “seismic-signal-to-instrumental-noise test scenarios.
(SNR)” ratios, but is in practice limited to lower
ratios due to misalignments or mis-leveling Strong-Motion Accelerometers
between the sensors. As discussed earlier, small The noise levels of strong-motion accelerometers
alignment or leveling differences between the make estimating the self-noise of such sensors
sensors decrease the coherency between the easier than typical low-noise broadband sensors
recordings and thus affect the noise estimate. It (Fig. 2). Since the user community is generally
was found that an alignment error of 0.2 in any only interested in strong motions with frequen-
axis allows the technique to extract self-noise for cies from a few Hz to a few tens of seconds
SNR values up to 60 dB, which was confirmed in period, it is often possible to resolve the self-
their associated experiment using real data. As noise using relatively short testing intervals.
this order of misalignment between As noted earlier, one must be careful to make
corresponding axes in similar colocated sensors sure that the digitizer noise is below that of the
is in agreement with the precision of best- sensor. Since strong-motion accelerometers are
practices manufacturing, one should only use generally sampled at higher rates than broadband
data taken during “quiet” time periods (SNR sensors (100 sps), it is necessary to identify the
below 60 dB, e.g., during night and periods with self-noise to much higher frequencies than for
minimal pressure-induced tilt). Even during these broadband sensors (Cauzzi and Clinton 2013). At
times, microseism noise can make it difficult to these higher frequencies, it becomes more difficult
estimate the self-noise of broadband sensors in to isolate the instrument from potentially
the period bands of 4–8 s and 18–22 s for instru- non-seismic noise sources such as electrical
ments (Ringler et al. 2011). power grid frequencies (e.g., 50 Hz in Europe
and 60 Hz in North America) noise. At these
Analysis Parameters higher frequencies, the coherency between record-
As in any time series analysis, appropriate time ings from colocated sensors presumably
window lengths, number, and overlapping must decreases, but this decrease in coherency is typi-
be used to resolve the low frequencies of interest cally compensated by decreasing ambient noise
while minimizing variance. For very broadband and increasing instrument self-noise.
seismometers, it is often of interest to understand
the self-noise for periods up to several thousands Broadband Seismometers
of seconds, which corresponds to a total window After installing a broadband seismometer, it is
length of at least 8 h divided into overlapping important to let the instrument fully settle down
sub-windows, each of about an hour (e.g., from mechanical stresses that can build up during
Table 1 of Evans et al. 2010). transport and adjust to temperature before
Seismometer Self-Noise and Measuring Methods, insulation consists of thin layers of neoprene around the
Fig. 7 Colocated Streckeisen STS-2 sensors at the Con- sensors, which also reduces the noise contribution due to
rad Observatory, Austria. Sensor casings are co-aligned air convection around the sensors
by parallel grooves in the underlying glass plate. Thermal
attempting to estimate the self-noise. What con- systems. As the only assumptions are that all
stitutes an instrument that is “fully settled” is not instruments will have a common input signal
well understood and currently a topic of some and output units, it is also possible to estimate
controversy. After the instrument has been the self-noise of digitizers (a “terminated noise”
allowed to settle, one can estimate the self-noise test), infrasound sensors (Hart et al. 2007), tilt
by using long-running quiet time periods (Hutt meters, gravimeters, and rotational seismic
et al. 2010). Local changes in wind and pressure sensors.
can introduce incoherent elevated noise levels on In all cases, the general approach for estimat-
horizontal channels at long periods (e.g., >100 s ing the self-noise will be similar, though the
period) as well as short (e.g., <1 s period). details may differ (e.g., test setup logistics,
frequency band of interest, etc.). For a discus-
Estimates Using Other Data Selection sion on applying the three-sensor self-noise
Techniques technique to digitizers, the reader is referred to
Although the process of self-noise measuring is Kromer (2006) and Sleeman et al. (2006).
not yet mature, the methods and techniques are Details on applying self-noise techniques to
rapidly improving. As manufacturers’ abilities other geophysical instruments are beyond the
to reduce sensor noise levels improves, methods scope of this entry.
for measuring the self-noise will also need to
improve. A better understanding of the sensitiv-
ity of seismometers to non-seismic phenomena Summary S
and techniques to thermally isolate the sensors
(e.g., Fig. 7) or correct the data from these With the suite of various methods for testing
sources will be needed. Although there has seismometer self-noise, it is possible to get an
been initial work on this subject (e.g., Forbriger initial estimate of the self-noise of a seismometer
et al. 2010; Z€urn et al. 2007) with promising even in less-than-ideal testing situations. For
results, there is still much to be done before instruments with relatively high self-noise, sim-
these methods become widely used by the seis- ple (one-sensor) self-noise methods may prove
mological community. sufficient.
Using additional high-quality sensors and
Other Geophysical Instruments more sophisticated techniques (two- or three-
The above methods for estimating seismometer sensor methods) can help to eliminate common
and accelerometer self-noise can be applied more signals when there is concern of the local back-
generally to other instruments and recording ground noise being above the sensor’s self-noise.
This will almost always be the case when testing calculating self-noise spectra and operating ranges for
broadband seismometers, but will not necessarily seismographic inertial sensors and recorders. Seismol
Res Lett 81:640–646
be the case when testing accelerometers though at Forbriger T, Sidmer-Schnidrig R, Wielandt E, Hayman M,
least microseisms are above the noise floor of the Ackerley N (2010) Magnetic field background varia-
state-of-the-art accelerometers even at an inland tions can limit the resolution of seismic broad-band
continental site. sensors. Geophys J Int 183:303–312
Gerner A, Bokelmann G (2013) Instrument self-noise and
Equally important to the two- and three-sensor sensor misalignment. Adv Geosci 36:17–20
methods is that the common seismic signals be Hanka W (2000). Which parameters influence the very
coherent among all instruments in the test. As long period performance of a seismological station?
discussed, this can be done by installing the GEOFON Program. http://geofon.gfz-potsdam.de/
geofon/manual/welcome.html
instruments in a quiet vault on a granite slab or Hart D, Merchant B, Chael E (2007) Seismic and
other stiff material to increase the coherence infrasound sensor testing using three-channel coher-
among sensors. Care should also be taken to ence analysis. In: 29th monitoring research review:
reduce the influence on instruments under test ground-based nuclear explosion monitoring technolo-
gies. LA-UR-07-5613, 935–944
from local non-seismic noise sources such as Havskov J, Alguacil G (2004) Instrumentation in earth-
pressure or temperature variations and that ade- quake seismology. Springer, Dordrecht
quate thermal settling time be provided before Holcomb LG (1989) A direct method for calculating
testing. instrument noise levels in side-by-side seismometer
evaluations. U.S. Geological Survey Open File Report,
Any use of trade, product, or firm names is for 89–214, 34 pp
descriptive purposes only and does not imply Holcomb LG (1990) A numerical study of some potential
endorsement by the US government. sources of error in side-by-side seismometer evalua-
tions. U.S. Geological Survey Open File Report,
90–406, 41 pp
Cross-References Hutt CR, Evans JR, Followill F, Nigbor RL, Wielandt
E (2010) Guidelines for standardized testing of broad-
band seismometers and accelerometers. U.-
▶ Seismic Network and Data Quality S. Geological Survey Open File Report, 2009–1295,
▶ Seismic Noise 62 pp
▶ Sensors, Calibration of Kromer RP (2006) Technology report evaluation of the
Kinemetrics/Quanterra Q330HR Remote Seismic Sys-
tem for IRIS/GSN Q330HR/GainX1 Configuration,
32286.http://www.iris.edu/hq/files/programs/gsn/doc-
References uments/IRIS_Progress_Report2_Q330HR.pdf
Peterson J (1993) Observations and modeling of seismic
Aki K, Richards PG (2002) Quantitative seismology. Uni- background noise. U.S. Geological Survey Open File
versity Science, Sausalito Report, 93–322, 95 pp
Anderson KE, Anderson JF, Anthony RE, Chaput J, Ringler AT, Hutt CR (2010) Self-noise models of seismic
McMahon ND, Morton EA, Aster R (2012) A site instruments. Seismol Res Lett 81:972–983
comparison between shallow vault-deployed and Ringler AT, Hutt CR, Evans JR, Sandoval LD
direct burial broadband seismometers, IRIS 2012 (2011) A comparison of seismic instrument noise
Workshop Abstracts. http://www.iris.edu/hq/iris_ coherence analysis techniques. Bull Seismol Soc Am
workshop2012/scihi/WebPages/0070.html 101:558–567
Berger J, Davis P, Ekstrӧm G (2004) Ambient earth noise: Rodgers PW (1994) Self-noise spectra for 34 common
a survey of the global seismographic network. electromagnetic seismometer/preamplifier Pairs. Bull
J Geophys Res 109, B11307 Seismol Soc Am 84:222–228
Cauzzi C, Clinton J (2013) A high- and low-noise model Rӧhm AHE, Trampert J, Paulssen H, Snieder RK
for high-quality strong-motion accelerometer stations. (1999) Bias in reported seismic arrival times deduced
Earthq Spectra 29:85–102 from the ISC Bulletin. Geophys J Int 137:163–174
Chaput JA, Zandomeneghi D, Aster RC, Hunter K, Kyle Scherbaum F (2007) Of poles and zeros: fundamentals of
PR (2012) Imaging of Erebus volcano using body digital seismology, 2nd edn. Springer, Dordrecht
wave seismic interferometry of Strombolian eruption Sleeman R, Melichar P (2012) A PDF representation of
coda. Geophys Res Lett 39, L07304 the STS-2 self-noise obtained from one year of data
Evans JR, Followill F, Hutt CR, Kromer RP, Nigbor RL, recorded in the Conrad Observatory, Austria. Bull
Ringler AT, Steim JM, Wielandt E (2010) Method for Seismol Soc Am 102:587–597
Seismometer, Extended Response 3231
Sleeman R, van Wettum A, Trampert J (2006) Three- significant bulk and weight and an unwelcome
channel correlation analysis: a new technique to mea- increase in mechanical complexity.
sure instrumental noise of digitizers and seismic sen-
sors. Bull Seismol Soc Am 96:258–271 Thus, combining the ruggedness of the short-
Stearns SD (1975) Digital signal analysis. Hayden Book period geophone with the extended frequency
Company, Rochelle Park response of an LP seismometer has long been
Tasič I, Runovc F (2012) Seismometer self-noise estima- a goal for researchers and instrument makers
tion using a single reference instrument. J Seismol
16:183–194 alike.
Z€
urn W, Exß J, Steffen H, Kroner C, Jahr T, Westerhaus
M (2007) On reduction of long-period horizontal seis-
mic noise using local barometric pressure. Geophys Geophone Frequency Response
J Int 171:780–796
In its simplest form, a geophone consists of

a mass suspended by a spring. To pick up the
relative motion between the mass (which, due to
Seismometer, Extended Response inertia, tends to remain steady) and the frame
(which will follow the motion of the Earth’s
Dieter Stoll surface), a moving coil system is typically used.
Lennartz Electronic GmbH, T€ubingen, Germany A coil that is attached to the suspended mass
moves in a magnetic field provided by
a permanent magnet attached to the frame, thus
providing an electrical output voltage that is pro-
Synonyms
portional to the relative velocity between mass
and frame.
Feedback system; Inverse filtering; Negative
If the suspended mass were an ideal oscillator
impedance converter
(i.e., without any damping forces at play),
a geophone would not pick up Earth motion
very faithfully. In fact, a single impulse would
Introduction cause it to oscillate forever at its natural fre-
quency. A certain amount of damping is always
Until the advent of the broadband seismometer provided by mechanical damping (friction) and
(Wielandt and Streckeisen 1982), electrody- electromagnetic damping, with the latter being
namic geophones were used for picking up responsible for the major part of damping. Elec-
short-period (1 s and less) earth movement, and tromagnetic damping is a consequence of Lenz’s
LP (long period) seismometers were used for law (in a closed circuit moving inside a magnetic
anything with lower frequency than one second. field, a current is induced which in turn generates S
Owing to their bulk and weight, LP seismome- a magnetic field opposed to the original magnetic
ters are not practical for use in the field; they field, thus resulting in a force counteracting the
typically require a controlled observatory-type movement of the conductor), and is proportional
environment. Most types of geophones, on the to the sum of the coil resistance and the external
other hand, are by their very nature field worthy, load on the coil (the input resistance of the ampli-
compact, and cheap. This is because they are fier or recorder).
mass-produced for use in seismic exploration In a plain geophone (without any active elec-
where they are typically deployed in large num- tronics involved), the minimum amount of
bers (hundreds or thousands). The downside is damping is given by the open circuit damping
that their natural frequency is usually 4.5 Hz or (typically in the range 0.25–0.5, but some com-
more. Some geophones designed for scientific mercially available geophones go down to 0.02
use have a natural frequency of 1 Hz, but this (Rodgers 1993)), meaning there is no external
low-frequency limit comes at the expense of load connected. With no external load connected,
3232 Seismometer, Extended Response
Seismometer, Extended 101

Response, 0.075
0.20
Fig. 1 Geophone 0.707
frequency response for 1.0 (critical)
different damping factors 1
(normalized to be 1.0 for f
>> f0)
10−1
|H(f)|
10−2
10−3
10−4
10−2 10−1 1 101 102
f/f0
the circuit is open, no current flows through the factors, ranging from open circuit (very low
coil, and no additional damping is contributed by damping, strong resonance peak) to 1.0. Usually,
the coil. The other extreme is a shorted coil – in the desired value is 0.707 of critical damping
this case, the external load is 0 Ω, and only the (thick line). At this setting, the modulus of the
intrinsic resistance (inductive reactance, to be transfer function at the natural frequency is 0.707
more precise) limits the amount of current (0.5*√2).
flowing. Consequently, this setup provides the Since open circuit damping is lower than the
maximum possible amount of damping (this is desired value of 0.707, an external damping resis-
why classical geophones should be transported tor (often referred to as a “shunt”) is used to
only with the output coil(s) shorted – the high achieve higher damping. An inevitable conse-
damping will help inhibit extraneous mass quence of shunting the geophone coil is that its
movement). output decreases (because the external resistor
The modulus of the complex transfer function and the geophone coil resistance form a voltage
of an electrodynamic geophone with respect to divider). In the flat part of the transfer function,
ground velocity depends on damping as follows the geophone’s output is proportional to
(Bormann 2012, Chapter 5.2):

Rd
(2)
o2 Rd þ Rcoil
H V ðoÞ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2 ffi (1)
o0 o2 þ 4h2 o2 o20
where Rd is the damping (shunt) resistance, and
Rcoil is the coil resistance.
where h is the damping factor expressed as the This fact has been deliberately omitted from
ratio of actual to critical damping (when h = 1), o Fig. 1; all curves have been normalized to 1.0 for
is angular frequency, and o0 denotes the natural frequencies much higher than the natural
angular frequency. frequency.
Figure 1 shows the modulus of a geophone’s For frequencies << f0, the geophone’s output
transfer function to ground velocity over a wide rolls off at 12 dB/octave (proportional to f2). For
range of frequencies for different damping frequencies >> f0, the transfer function is flat.

Response,
Fig. 2 Schematic
operation of inverse filter, 103
new f0 = 0.1* original f0
inv
102 er
se
filt
er
tra
ns
101 fer
fun
cti
on
|H(f)|
1
t ion
nc
r fu
fe on
10−1 ns cti
g tra fun
r
u ltin sfe
res an
l tr
−2 ina
10 o rig
10−3
10−4 −2
10 10−1 1 101 102
f/f0
Methods of Frequency Response which, unfortunately, typically is found to be

Extension not less than a few Hz – a typical operation
amplifier such as the OP-27 has a 1/f corner
Extending the low-frequency response of a geo- frequency of 2.7 Hz (Analog Devices 2006).
phone without increasing its bulk and weight can Since the inverse filter method requires an increas-
be achieved with several methods. The most ing amount of amplification towards lower
common ones will be briefly outlined here. frequencies, intrinsic 1/f noise will be amplified
overproportionally. Further, maintaining thermal
stability of low-frequency circuits is not trivial.
Using an Inverse Filter Thus, signal-to-noise ratio for low frequencies
An electronic filter can be implemented to will be problematic. Also, the signal chain may
compensate the geophone’s roll-off at low fre- take a prohibitively long time to recover from even
quencies (Havskov and Alguacil 2004). In real a brief overload. Inverse filtering was applied in
life, the filter’s transfer function cannot increase the 1960s–1980s but, owing to its drawbacks and
indefinitely towards lower frequencies. A new the advent of superior methods, is no longer in S
fictitious natural frequency is defined by the practical use today.
point where the filter’s transfer function returns
to being flat. Figure 2 shows the schematic func-
tion of an inverse filter (Lippmann 1982). Positive Feedback
Intriguing as it may seem at first sight, this As outlined previously, the bare-bones version of
approach is plagued with many practical prob- a geophone consists of a coil suspended by one or
lems. Noise appearing in electronic circuits is not more springs, and moving in a magnetic field.
necessarily “white,” i.e., equally distributed over When motion occurs, a voltage is induced across
a wide frequency range. There is one particularly the coil’s electrical terminals. However, fancier
annoying type of noise called 1/f noise, which, as versions exist where a second coil is present. This
its name implies, means it gets increasingly stron- coil is not a movable coil but is present so as to be
ger the lower the frequency gets. This phenome- able to feed an external current through it which,
non occurs below a certain corner frequency again through induction, will exert a force on the
Seismometer, Extended 101 0.707

Response, 2.0
5.0
Fig. 3 Geophone 10.0
20.0
frequency response for 50.0
100.0
different(over) damping 1
factors
10−1
|H(f)|
10−2 al)
ion
por t
pro
on
ati
ler
ce
(ac
10−3 ~f
10−4
10−2 10−1 1 101 102

f/f0
suspended mass. This procedure is usually all; in other words, the mass is accelerated in
applied just for calibration (hence the name “cal- exactly the same way as the Earth, and the current
ibration coil”) but can also be applied for positive required to keep it steady with respect to the
feedback. frame is proportional to acceleration for frequen-
Hypothesizing the presence of such a coil, cies higher than the new corner frequency (it is
positive feedback could be applied by integrating not feasible to extend the bandwidth to arbitrarily
the output signal (thus rendering it proportional to low frequencies). Practical limitations of this
displacement) and feeding the integrated voltage method are two coils required (thus excluding
back through the feedback coil (Lippmann 1982). cheap mass-produced exploration geophones)
When applied with the correct polarity, the feed- and stability problems of feedback electronics
back signal will “nudge” the mass so that the due to inductive coupling of signal and feedback
original movement will be amplified, thus coils (Lippmann 1982).
counteracting the decay of sensitivity. The effect Broadband sensors also use negative feed-
is similar to using a softer spring, leading to back, but they use a capacitive transducer to
a smaller restoring force and, consequently, to pick up mass movement, thus avoiding the sta-
a lower natural frequency. Like the inverse filter bility issues. However, it is a nontrivial undertak-
method, positive feedback has practical stability ing to fit a capacitive pickup in an off-the-shelf
problems and is not in practical use nowadays. exploration geophone, so this method, while
promising, does not lend itself well to
Negative Feedback commercializing.
Rather than integrating the output signal, this
method differentiates it (thus rendering it propor-
tional to acceleration of the suspended mass) and Response Extension Using NIC (Negative
feeds it back through the second coil choosing the Impedance Converter)
polarity so that it will not amplify but attenuate
the mass movement. Ideally, the geophone mass Recalling Eq. 1, let us now extend the series of
would be damped so strongly that it does not damping factors in Fig. 1 towards higher values,
perform any movement relative to the frame at and we get this (Fig. 3).
It is evident that the higher the damping factor Practical Considerations

gets, the wider is the frequency range in which
the output is proportional to acceleration. Linearity
If a way can be found to achieve such high Conventional geophones are plagued with linear-
damping factors, the geophone will effectively ity issues on account of the inherent nonlinearity
be transformed into an accelerometer with of the suspension spring(s). Positive feedback
a wide frequency range, albeit at the expense of would magnify this type of problems, whereas
output. As can be seen from Fig. 3, output negative feedback (by overdamping) will actu-
decreases by several orders of magnitude in ally alleviate them. Since, in a first-order approx-
a certain frequency range. imation, the mass will remain steady, the
suspension springs will be flexed only very little.
Since spring nonlinearity increases with the
Achieving Overdamping
Neglecting the mechanical contribution to amount of flexing or elongation, this is a very
efficient means of suppressing spring-induced
damping (which is a reasonable simplification
nonlinearity.
for typical geophones if a very high damping is
required), we can limit our consideration to elec-
Amplification
trical damping which is proportional to the com-
Since, as seen in Fig. 3, output all across the
bined resistances of the generator coil and the
external shunt resistor (Havskov and Alguacil acceleration proportional passband is signifi-
cantly lower than the velocity proportional signal
2004):
of the non-overdamped geophone, additional
amplification is called for. However, it is not
CDR
h¼ (3) convenient to amplify the acceleration propor-
Rcoil þ Rshunt tional signal since this would create dynamic
range problems. Applying linear amplification
where CDR is the total resistance needed to get by, say, a factor of 1,000 would lead to impracti-
h = 1 (critical damping resistor). cally large amplitudes for the high frequencies. It
As we can see, the maximum amount of is a much more practical approach to apply
damping that can be achieved in a bare-bones amplification only after the acceleration propor-
geophone is when the signal coil is shorted tional signal has been converted back to velocity
(Rshunt = 0). In this case, only the intrinsic coil proportional (see next section).
resistance is responsible for damping. Loading
the coil with any external resistance at all will Converting the Output Signal Back to
only decrease damping. In order to arrive at Velocity
damping factors higher than short-circuit The obvious solution of simply integrating the S
damping, a negative resistance would have to be acceleration proportional signal back to velocity
used – clearly a feat that cannot be achieved using is not a very practical one. As can be seen in
passive electronics. An NIC (negative impedance Fig. 3, the low-frequency roll-off of the standard
converter) provides exactly that functionality damped seismometer is no longer seen in the
(Lippmann 1982; Ulmann 2005). In its simplest overdamped transfer function, but there is still
form, it is just an operational amplifier and two quite an amplitude gap to cover. The difference
resistors. in amplitude is largest around the original corner
A practical value for the negative resistance is frequency and decreases towards higher and
80 % of the coil resistance (Ulmann 2005). lower frequencies, but it is especially towards
Completely compensating the coil resistance lower frequencies that trouble is to be expected
with close to 100 %, negative resistance is not if one were to attempt a simple integration with-
practical for reasons of stability. out any frequency band limitation. As an

Response, Fig. 4 Design
of integrating/
differentiating filter
In
n
t
tio
eg
tia
ra
tio
n
re
n
ffe
Di
1
new f0 is 0.1*original f0
|H(f)|
10−1
e
h on
op
ge
d
pe
am
erd
f ov
to
10−2 tp
u
ou
nal
r tio
opo
pr
n
io old f0
rat
e le
c
10−3 ac
10−2 10−1 1 101 102
f/f0
example, let’s take a look at 0.01 f/f0. The output level comparable to the non-overdamped geo-
of the standard damping seismometer is down to phone, the filter must provide more amplifica-
104 of its plateau output, and the overdamped tion than shown in our schematic example. The
(acceleration proportional) output is down even exact amount of peak amplification depends on
further, already off scale. Bringing these low the ratio between the original and the new virtual
frequencies back to “flat” would lead to natural frequency.
extremely high amplification factors, leading to A filter with the desired characteristics
the same type of problems that were described (6 dB/octave roll-off on either side of a given
above in “Using an Inverse Filter.” Rather than frequency) is not difficult to implement – it is
simulating an extremely broadband seismome- basically a combination of a first-order high-
ter, a band-limited approach is a lot more prom- pass and a first-order low-pass filter, with both
ising. To this end, a filter is needed that corner frequencies being set to the desired new f0
integrates frequencies above its corner fre- and of course an amplification stage.
quency and differentiates those below. Figure 4
(after Lippmann 1982) schematically shows its
functional principle. With a frequency propor- Environmental Influences
tional rise for f < f0NEW, and a frequency pro- Except for a few broadband models, seismome-
portional decay for f > f0NEW, the result of ters will typically be used in field scenarios, i.e.,
multiplying the frequency proportional acceler- they have to work under adverse environmental
ation signal with the filter’s transfer function conditions, and will have to maintain their char-
will be exactly the transfer function of acteristics over a wide range of temperatures.
a geophone with a virtual natural frequency of The negative feedback principle outlined
f0NEW. However, in order to arrive at an output above basically amounts to measuring one small
physical entity (in our case, the current flowing latter part is close to negligible, compared to the
through the generator coil when the coil starts NIC’s contribution.
moving), then adding a counteracting entity of The only thing of relevance to the practi-
similar magnitude (in our case, a current that tioner, though, is the composite noise, expressed
will create a force to keep the coil steady). Effec- not in electrical but in seismological units, i.e.,
tively two similar entities are subtracted from in equivalent ground velocity or acceleration.
each other. Consequently, if one or both of the The following slightly edited diagram from
two entities are susceptible to changes in environ- Wielandt (1991) shows the equivalent RMS
mental parameters, the result of the subtraction velocity for two different types of Lennartz
will also be susceptible to changes. The coil resis- (2013) seismometers – the LE-3Dlite type
tance is a “real” resistor consisting of many wind- based on physical 4.5 Hz geophones and
ings of copper wire whose value will increase converted to 1 Hz seismometers and the
with temperature. The negative impedance is LE-3D/5 s type based on physical 2 Hz geo-
implemented by active circuitry which will not phones and converted to 0.2 Hz (5 s) seismom-
be immune to temperature changes but will cer- eters. For reference, the NLNM (New Low
tainly react differently from the purely ohmic coil Noise Model, Peterson (1993)) and the noise
resistance. The following example from Lipp- level at the BFO (Black Forest Observatory)
mann (1982) illustrates the problem, focusing site are also given. Note that the noise levels
only on the coil. have been calculated for frequency bins of con-
A coil resistance of 600 Ω at 20 C will change stant relative bandwidth. For the 1/3 octave
to 640 Ω at 40 C. If our negative resistance is bandwidth shown here, the relative bandwidth
520 Ω, the effective damping resistance will is 23.2 %. For example, the 10 Hz bin is 2.32 Hz
change from 80 Ω (20 C) to 120 Ω at wide, whereas the 50 Hz bin is 11.6 Hz wide.
40 C – a 50 % increase, leading to drastic Using constant relative bandwidth bins is best
changes in the sensor’s characteristics. suited for comparing and converting noise mea-
Clearly, temperature compensation needs to surements (Bormann 2012, Chapter 4.1) (Fig. 5).
be applied. Lippmann (1982) describes the mea- It turns out that both seismometer types are
sures and results. able to resolve ground noise at a very quiet site
like BFO over a wide range of frequencies. The
Electronic Noise vast majority of sites will exhibit much more
Electronic noise is the prime limiting factor for pronounced noise than BFO, thus rendering the
how much the response can be extended since it extended response seismometers useable over
defines the smallest detectable ground velocity their full frequency range.
amplitude. Electronic noise is generally fre-
quency dependent. As outlined previously in the S
“Inverse Filter” entry, below a certain threshold A Practical Implementation
frequency (which depends on the particular
active electronic element, e.g., an operational The photo shows a classical, purely mechanical
amplifier), noise tends to be proportional to seismometer next to an electronically enhanced
1/f. In other words, lower frequencies tend to seismometer built around a 4.5 Hz geophone. In
exhibit more noise. both cases, natural frequency is one hertz, and
In an extended response seismometer using both are single-component instruments (Fig. 6).
negative feedback and a subsequent integrating/
differentiating filter, electronic noise can be con-
sidered separately for the acceleration propor- Summary
tional part (i.e., the combination of geophone
and NIC) and for the filter/amplifier part. Lipp- Range-extended seismometers based on robust,
mann (1982) reveals that the contribution of the readily available, and comparatively cheap
Seismometer, Extended Response, Fig. 5 L2-xD Self Noise in 1/10 decade (1/3 octave)
Seismometer, Extended
Response,
Fig. 6 Classical Geophone
(left) vs. Range-Extended
Seismometer (right), both
of them single-component,
1 Hz units
exploration-grade geophones provide a viable range-extended seismometers do not contain

alternative to clumsy and delicate mechanical physical elements tuned to very low frequencies
seismometers for short- and intermediate-period and do not respond to very low frequencies. Con-
applications. Unlike typical broadband sensors, sequently they usually require no shielding from
Selection of Ground Motions for Response History Analysis 3239
pressure and temperature changes and provide

extended high-frequency response, typically up Selection of Ground Motions for
to and above 100 Hz. Inherent limitations such as Response History Analysis
electronic noise can be mitigated by proper
design and implementation. Anastasios G. Sextos
Division of Structural Engineering, Department
of Civil Engineering, Aristotle University of
Thessaloniki, Thessaloniki, Greece
Cross-References
▶ Principles of Broadband Seismometry

Synonyms
▶ Recording Seismic Signals
▶ Seismic Noise
Earthquake records; Ground motion; Record
▶ Seismometer Self-Noise and Measuring
scaling; Response history analysis
Methods
▶ Sensors, Calibration of
Introduction
References The evolution in computational power and the

parallel processing capabilities of modern engi-
Analog Devices (2006) OP27 datasheet, downloadable neering software make nowadays the use of com-
from http://www.analog.com/static/imported-files/
data_sheets/OP27.pdf
plicated structural analysis methods an attractive
Bormann P (ed) (2012) New manual of seismological alternative for the design and assessment of struc-
observatory practice (NMSOP-2), IASPEI. GFZ Ger- tures. In contrast to the past, when the elastic
man Research Centre for Geosciences, Potsdam. static analysis was almost exclusively used for
http://nmsop.gfz-potsdam.de
the seismic design of structures, the state of prac-
quake seismology. Springer, Dordrecht tice has progressively moved toward dynamic-
Lennartz electronic (2013) LE-xD seismometer family elastic, nonlinear-static (i.e., single mode or
(Document 990-0003), PDF document downloadable multi-modal “pushover”), and even nonlinear
from http://www.lennartz-electronic.de
response history analysis. The latter, capturing
Lippmann E (1982) Umwandlung eines elektrody-
namischen Seismometers in einen Beschleunigungs- more efficiently the hierarchy of failure mecha-
Aufnehmer und Entwicklung eines Filters zur Simula- nisms, the energy dissipation, the force redistri-
tion eines niederfrequenten Seismometers. Diploma bution among the structural members, and
thesis, Ludwig-Maximilians-Universität M€
unchen
contact issues (such as gap, impact, sliding, and
Peterson J (1993) Observations and modeling of seismic
background noise. US geological survey open file uplift), is deemed preferable in cases of signifi- S
report 93-322 cant material or geometrical nonlinearities and,
Rodgers P (1993) Maximizing the signal-to-noise ratio of as such, is used for the design of seismically
the electromagnetic seismometer: the optimum coil
isolated buildings and bridges or the assessment
resistance, amplifier characteristics, and circuit.
B Seismol Soc A 83:561–582 of existing structures with various degrees of
Ulmann B (2005) Overdamping geophones using negative damage. Elastic response history analysis is also
impedances. http://www.vaxman.de/publications/ extensively used, primarily for structures whose
teach_gp.pdf
response is dominated by higher modes (mostly
Wielandt E (1991) Messungen der kurzperiodischen
Bodenunruhe in Schiltach (BFO). Internal report, tall and irregular buildings and towers)
published in German only. PDF of original text and or structures of high importance that are typically
English translation available from the author of the designed to remain elastic even for long return-
present article
period earthquake intensities (i.e., industrial
Wielandt E, Streckeisen G (1982) The leaf-spring seis-
mometer: design and performance. B Seismol Soc facilities, power plants, dams, critical administra-
A 72:2349–2367 tive buildings, etc).
3240 Selection of Ground Motions for Response History Analysis
In all cases, the main task of the design pro- earthquake scenario of given magnitude, M,
cedures is to achieve more predictable and reli- source-to-site distance, R, and, in some
able levels of safety and operability against cases, faulting type or soil class (i.e., Vs,30) at
different levels of seismic intensity, a framework the site of interest.
known as performance-based design and assess- • Predict the Median or Mean of Structural
ment. Despite the above major advances made in Response from ground motions that are
terms of structural analysis, the reliability of the selected (or generated) to match a median tar-
analysis output and the subsequent structural per- get response spectrum for a given M-R pair,
formance prediction strongly depend on the deci- obtained from a ground motion prediction
sions made for the selection of the seismic input equation (or attenuation relationship). Disper-
which is used as ground excitation. Research has sion of structural response might also be of
shown that among all possible sources of uncer- interest; however, minimizing the standard
tainty stemming from structural and soil material error of response quantities is not explicitly
properties, the modeling approximations, and the envisaged.
design and analysis assumptions as well as the • Predict the Probability Distribution (mean
earthquake-induced ground motion, the latter has and dispersion) of Structural Response using
by far the highest effect on the variability ground motions that satisfy a given spectral
observed in the structural response (Elnashai acceleration at the fundamental period of the
and McClure 1996; Padgett and Desroches structure, Sa(T1), that has resulted from an
2007; Shome et al. 1998). Therefore, the selec- associated M and R (as well as fault mecha-
tion of a “reliable” suite of earthquake ground nism and soil class or Vs,30) scenario. Notably,
motions constitutes an important prerequisite for a priori knowledge of the structural properties
the reliability of the structural analysis procedure (i.e., its fundamental period T1) is
as a whole. a prerequisite.
Along these lines, numerous computational • Predict the Median or Mean of Structural
methods and tools have been developed for Response for a given set of Sa(T1), M, and R.
(a) selecting suites of earthquake records from In this case, the scenario refers to a Maximum
available strong ground motion record databases Considered Earthquake Sa(T1) and not a MCE
(b) generating synthetic and artificial ground response spectrum.
motions or (c) modifying existing ground motions
until they present desirable target characteristics. A fifth objective can also be distinguished to
minimize the structural response discrepancy by
considering a threshold confidence level for the
Objectives of Ground Motion Selection standard error of the response quantities,
additional to conventional spectral matching
Currently more than 40 methods exist for (Katsanos and Sextos 2013). This involves con-
selecting, modifying, and scaling earthquake sideration of structural analysis results within the
ground motions. These methods can be grouped GMSM process.
conceptually by their objective as follows based
on the Ground Motion Selection and Modifica-
tion (GMSM) Program of the Pacific Earthquake Methodologies for Ground Motion
Engineering Research (PEER) Center (Haselton Selection and Scaling
2009) in the following broad four categories:
Independently of the ultimate objective of the
• Predict the Probability Distribution (mean GMSM process, the various methodologies for
and dispersion) of Structural Response (i.e., developing earthquake ground motion sets for
engineering demand parameter, EDP) from (linear or nonlinear) dynamic analysis of struc-
ground motions that comply with an tures can also be classified based on the concept
of selection and the procedure of modulating or a target, Uniform Hazard Spectrum (UHS)
scaling ground motions. Naturally, method (Kramer 1996; American Society of Civil Engi-
objectives and selection and scaling procedures neers 2005; McGuire 2004), which is deter-
are not fully correlated. The following classes of mined from (a) a ground motion prediction
methods are listed below. A detailed review of relationship, (b) a seismic hazard assessment
most important methods is also performed else- for the site of interest, or (c) the seismic code
where (Katsanos et al. 2010): provisions. In this case, ground motions are
scaled with a single or individual scaling factors
Selection by M and R and Scaling to a Target so that their individual spectrum, or the mean of
Intensity Measure their response spectra closely match and in fact
The simplest GMSM method involves the forma- exceed, in terms of “shape,” the ordinates of the
tion of set (bin) of motions that satisfy, as close as target UHS, typically within a given range of
possible but without explicit constraints, prelim- periods around the fundamental period of the
inary magnitude, M, source-to-site distance, R, structure. When a uniform scaling factor is
and often seismotectonic or soil class criteria. It is sought, it may be determined through the fol-
recalled that the total number of sets Ntot that can lowing expression:
be formed, from m, potentially eligible, records

1
out of a larger group of s records, can be calcu- Saavg ðT i Þ
lated by the following factorial formula of the sf avg¼ min , i ¼ 1 to N
Satarget ðT i Þ
binomial coefficient:
(2)

s s! where Ti is the sample period and N is the size of

N tot ¼ ¼ (1)
m m!ðs mÞ! the sample into which the prescribed period range
is discretized.
Quite commonly, all eligible suites are ranked
Once eligible ground motions are selected, their
according to their “goodness-of-fit” to the
accelerations are multiplied by a scaling factor to
target spectrum, as quantified by the normalized
match a target intensity measure (IM), typically
root-mean-square-error, d, between the scaled
being the peak ground acceleration (PGA) or the
average Sa,avg, and the target spectrum Sa,trt
spectral acceleration at the fundamental period of
(Iervolino et al. 2010b):
the structure, Sa(T1). In the latter case, it is evi-
dent that all ground motion spectra will have sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ffi
identical ordinates at period T1 and different 1 XN Saavg ðT i Þ Satrt ðT i Þ 2
d¼ (3)
spectral accelerations in all other periods. More N i¼1 Satrt ðT i Þ
elaborative IMs have also been proposed, involv- S
ing the spectral shape and the structural charac- A wide variety of similar expressions have also
teristics. Such IMs are expected to result into been used in the literature. Further discussion on
a relatively more accurate prediction of the seis- the efficiency of various spectral matching indi-
mic demand (Baker and Cornell 2005; Luco and cators can be found elsewhere (Beyer and
Cornell 2007; Tothong and Luco 2007); however, Bommer 2007; Jayaram et al. 2011; Kottke and
the approximate definition of the seismic scenario Rathje 2008; Naeim et al. 2004).
is a significant limitation. In addition to spectral shape, these methods
may consider other earthquake, site, or ground
Selection and Scaling to a Target Uniform motion parameters in selecting ground motions.
Hazard Spectrum Finally, the mean or the maximum (depending on
In the light of the above criteria, it is also com- the number of ground motions within the set) of
mon to envision matching of the response spectra the response quantities is used as the design
of the selected acceleration time series with value.
Seismic Codes of Practice average of the fault-normal components is not

Most contemporary seismic codes and design less than the MCE response spectrum for the
recommendations, such as Eurocode 8 for build- period range of spectral matching (as described
ings (CEN 2004) and bridges (CEN 2005), below).
ASCE standards 7-10 (ASCE/SEI 2010) and
FEMA P-750 (FEMA 2009a) as well as various Spectral Matching A distinction is typically
national norms (New Zealand Standards, Italian made between two-dimensional and three-
Code, and Greek Seismic Code), describe rela- dimensional analysis, the latter involving selec-
tively similar procedures that are based on spec- tion and application of pairs of records for linear
tral matching for the selection of earthquake or nonlinear dynamic analysis. In principle, the
ground motions in the framework of dynamic procedure is similar among most important seis-
analysis of structures. In most cases, seismic mic codes and recommendations and involves the
motions can be represented by real, artificial, or following steps: (a) the 5 %-damped elastic spec-
simulated records, typically complying with the tra are derived for each component of the eligible
aforementioned preliminary criteria of earth- horizontal motions selected; (b) the mean of
quake magnitude, distance, the seismotectonic the individual (EC8-Part 1) or the square root of
environment, and the local soil conditions. Dif- the sum of squares (SRSS spectra) of the individ-
ferences among the codes can be summarized as ual elastic spectra ordinates is determined; (c) the
follows: mean spectrum of the ground motions is
compared with the target code-prescribed spec-
Duration Strong-motion duration is not explic- trum; and (d) records are scaled so that the spec-
itly considered as an additional selection param- tral ordinates of the mean spectrum (either from
eter in most documents inclusive of ASCE individual or SRSS spectra) exceed a lower
standards 7-10 (ASCE/SEI 2010) and the bound of the target spectral acceleration
Eurocodes. (Table 1) within a prescribed period range (also
depicted in Table 1).
Simulated and Artificial Ground Motions Use
of simulated ground motions is permitted in Scaling Scaling of the individual records toward
a number of seismic codes; however, it is spectral matching can be performed either with
restricted only to cases of inadequate number of the use of a uniform scaling factor (i.e.,
real accelerograms in others, i.e., FEMA P-750 EC8–Part2, ASCE 7-10) or a record-dependent,
and Eurocode 8 Part 2 for bridge design (CEN individual scaling factor (FEMA P-750). Other
2005). codes (i.e., EC8-Part1) do not provide specific
guidelines regarding the scaling of seismic
Near-Field Considerations Near-field effects records in order to establish the required compli-
are typically either ignored or considered in ance with the target design spectrum. New
a quantitative manner in terms of direction of Zealand Standards (Standards New Zealand
fault rupture and velocity pulses, an example (SNZ) 2004) on the contrary prescribe two dis-
being FEMA P-750 (FEMA 2009a). More tinct scale factors k1 and k2, the first minimizing
detailed distinction is made in FEMA P695 the difference between recorded and target
(FEMA 2009b) by forming distinct far-field and response spectrum in a least mean square scheme
near-field record sets. ASCE standards 7-10 pre- over the period range of interest, while the latter
scribe different design quantities for sites within ensures that the energy in the spectrum of the
3 miles (5 km) of the active fault that controls the strongest ground motion exceeds the energy of
hazard. In this case, each pair of ground motion the target spectrum.
components shall be rotated to the fault-normal
and fault-parallel directions of the causative fault. Design Quantities Structural analysis results
Next, it shall be appropriately scaled so that the are processed in a statistical manner, and the
Selection of Ground Motions for Response History Analysis, Table 1 Earthquake records selection and spectral
matching criteria prescribed in the seismic codes and guidelines studied here
Seismic Spectral Lower bound of
codes and Ensemble matching mean spectral Scaling
guidelines Selection criteria spectrum period range acceleration factor
EC8 Part 1 Seismotectonic Mean of 0.2–2.0 T1 0.90 Sa(T)design Not specified
(buildings) features, soil type individual spectra
EC8 Part 2 Source mechanism, Mean of SRSS 0.2–1.5 T1 1.30 Sa(T)design Uniform
(bridges) M, R spectra
FEMA P-750 Source mechanism, Mean of SRSS 0.2–1.5 T1 1.17 Sa(T)design Individual
M, R spectra
ASCE/SEI Source Mean of SRSS 0.2–1.5 T1 1.00 Sa(T)design Uniform
7-10 mechanism, M, R spectra
NZS 1170.5 Seismotectonic Individual record 0.4–1.3 T1 Matching as nearly k1 and k2
features, soil type scaling and as practicable factors
matching
Taiwan Source Mean of 0.2–1.5 T1 0.90 Sa(T)design Not specified
seismic code mechanism, M, R individual spectra
design quantities are defined as either the mean Selection and Scaling to a Target Conditional
(in case of seven records or more) or the maxi- Mean Spectrum
mum (in case of three records) of the engineering An alternative group of methods for ground
demand parameters of interest (EC8, FEMA motion selection and scaling is similar to the
P-750). Certain codes solely provide the option aforementioned spectral matching, but the Uni-
of obtaining the maximum response out of the form Hazard target Spectrum, the Conditional
three response history analyses (CPA 2011; Stan- Mean Spectrum (CMS) (Baker 2011), is used
dards New Zealand (SNZ) 2004). The most instead. The main reason behind this substitution
refined to date provision to derive seismic design (Kottke and Rathje 2008; Lin et al. 2013a; Wang
values is prescribed in ASCE 7-10 clearly 2010) is that the UHS is shown to be unsuitable to
distinguishing between (a) force response param- serve the main purpose of dynamic analysis,
eters that shall be multiplied for each ground which is to excite the structure under consider-
motion analyzed by a factor Ie /R, where Ie is ation with ground motions having a specified
the importance factor and R is the Response spectral acceleration at a given period. In fact,
Modification Coefficient, and (b) drift response UHS compatible ground motions are often asso-
quantities, which shall be multiplied by Cd /R, Cd ciated with large-amplitude spectral values in
being the deflection amplification factor. It is a wide range of periods. Furthermore, UHS can S
noted herein that the above methods refer to the hardly be considered as a spectrum of a single
design of new buildings according to modern earthquake event as it rather represents an enve-
seismic codes and do not apply to the probabilis- lope of spectra corresponding to different seismic
tic assessment of the nonlinear response of events and sources. On the contrary,
existing buildings (Jalayer and Cornell 2009). It a Conditional Mean Spectrum (CMS) represents
is also noted that despite the simplification and the expected (i.e., mean) response spectrum, con-
applicability of the above code-based procedures, ditioned on the occurrence of a target spectral
significant limitations also exist (Sextos acceleration at the period of interest, typically,
et al. 2010). though not exclusively, the fundamental period
More details on comparative seismic code of the structure T1.
provisions on ground motion selection and scal- Similarly to the UHS, the target CMS is cal-
ing can be found elsewhere (Hachem et al. culated for the scenario Sa(T1), magnitude, M,
2010). and source-to-site distance, R, together with
other preliminary criteria such as fault type or and does not change in case of record scaling. On
soil, if desirable. In simpler words, the CMS the other hand, it is also noticeable that for
represents the expected response spectrum for a given ground motion record, “e” is clearly
the defined ground motion scenario, which is a function of the period of interest and depends
based on a target Sa(T1) at a single structural on the particular ground motion prediction model
period, in contrast to the UHS, which represents used, since different attenuation relationships
equally rare Sa(T) values at many periods lead to different mean and standard deviation of
(including T1) simultaneously (Haselton 2009). ln{Sa(T)}. Therefore, it is important to ensure
To develop this new target spectrum, Probabilis- that the ground motion prediction model used to
tic Seismic Hazard Assessment can be used to compute e is the same model used in the ground
determine the spectral acceleration Sa(T1) that motion hazard assessment. This dependence of e
corresponds to the target probability of exceed- to attenuation relationships is perhaps the most
ance at the site of interest, denoted as Sa(T1)*. important drawback in the use of this parameter.
De-aggregation can then be used to estimate the Ground motion selection methods (Tothong
mean values of magnitude, source-to-site
dis- and Luco 2007) that use “e” as a proxy of the
tance, and “epsilon” (“e”) M, R, e that lead to CMS spectral shape ensure that the value of “e” at
an acceleration equal to Sa(T1)*. The definition of the fundamental period of the building, denoted
parameter “epsilon” is provided in the following as e(T1), is as close as possible to the target e(T1)
section. The way in which spectral matching to of the ground motion scenario. This practically
the target CMS is achieved is similar to the pro- implies that the record-to-record variability at the
cedures described above. fundamental period of the structure is reduced,
and in turn, the discrepancy in the structural
Selection and Scaling to a Proxy (“«”) of the response of the corresponding SDOF system is
Target Conditional Mean Spectrum also lower. Some concerns have been expressed
A more advanced approach category of methods regarding the appropriate choice of CMS and the
for ground motion selection and scaling to conditioning period, primarily related to the
a target Conditional Mean Spectrum utilizes the importance of higher modes of vibration of
parameter “epsilon” (“e”) as a proxy of the desir- MDOF systems and the anticipated period elon-
able CMS spectral shape (Baker and Cornell gation under strong ground motions (Katsanos
2006). It is recalled that “epsilon” was first et al. 2014). However, it has been shown that
defined by engineering seismologists as the num- risk-based assessments are relatively insensitive
ber of standard deviations by which a given spec- to the choice of conditioning period provided that
tral acceleration, expressed in logarithmic terms, the ground motions are carefully selected to
differs from the mean logarithmic spectral accel- ensure hazard consistency (Lin et al. 2013a).
eration provided by a ground motion prediction
(attenuation) equation. In other words, “e” is
derived by subtracting the predicted mean loga- Available Computational Tools and
rithmic spectral acceleration at given period T1 Databases
(ln{Sa(T1)}) from the corresponding value (ln
{Sa(T1)}) of the record under examination and Strong Ground Motion Databases
then dividing by the logarithmic standard devia- Among numerous strong ground motion data-
tion estimated by the attenuation relationship. In bases in Japan, Taiwan, and Europe (European
practical terms, this implies that a record with “e Strong Ground Motion database, www.isesd.hi.
of 1.5 at 0.5 s” has a spectral acceleration at is), the PEER-NGA Next Generation Attenuation
a period T1 = 0.5 s that is 1.5 standard deviations strong-motion database is a continuously devel-
higher than the predicted mean spectral value. oping project currently consisting of 3,551 pub-
An advantage of the “e” parameter is that it is licly available, three-component seismic records
determined with respect to the unscaled record (i.e., about 10,650 individual earthquake
acceleration time series) that have been recorded run at the background, a target level of dispersion
during 173 shallow crustal earthquakes from of structural response quantities.
active tectonic regions worldwide. The
corresponding seismic events, which have been
recorded primarily in California, range in magni-
Accepted Knowledge on Ground Motion
tude from 4.2 to 7.9 and cover epicentral dis-
Selection for Response History Analysis
tances in the range 0.2–600 km. Apart from the
Purposes
magnitude and the distance, the earthquake
database contains basic information about the
– Structural response is inherently probabilistic
seismic source including date and time of the
in nature due to the variability among ground
event, hypocenter location, faulting mechanism,
motions.
seismotectonic environment, and others. Detailed
– A major challenge in choosing the most appro-
data about 1,600 strong-motion stations are also
priate method for ground motion selection and
provided (i.e., site characterizations, surface
scaling is to understand the purpose of the
geology, shallow subsurface conditions, the loca-
dynamic analysis for which the ground motions
tion of the instrument inside the structure’s instal-
are sought. A major distinction is between
lation place). Furthermore, each acceleration
design and assessment purposes. Careful selec-
time history has been corrected for the response
tion of methods, intensity measures, and engi-
of the strong-motion instrument itself and filtered
neering demand parameters is necessary.
out the noise included while it can also be auto-
– An additional challenge, at least for a design
matically scaled online.
viewpoint, is to keep structural response
discrepancy low.
Software and Tools for Ground Motion
– Numerous GMSM methods exist in the litera-
Selection
ture with distinct advantages and drawbacks.
Given the above extensive repository of earth-
The direct comparison of these methods is not
quake records and the fact that the most common
always feasible.
earthquake record selection procedures involve
– Most GMSM methods adopted in contempo-
spectral matching, recent work evolved to
rary seismic codes are rather simplified com-
develop computational tools for quantifying
pared to the breadth of the existing methods
and/or optimizing spectrum compatibility to
available in the literature. Still, they require
a code prescribed or CMS (Youngs et al. 2007).
significant effort, deep understanding of the
REXEL (Iervolino et al. 2010a, 2011, 2012) was
physical problems and the parameters
the first, all-in-one, software introduced for this
involved as well as specialized computational
purpose and facilitates the search for suites of
tools in order to overpass their inherent
waveforms compatible to target spectra that are S
limitations.
either user defined or automatically generated
according to Eurocode 8 and the recently issued
Italian seismic code. An alternative web-based Summary
software for earthquake record selection is
SelEQ (Dias 2010), offering various filtering Linear and nonlinear dynamic analysis of struc-
options. More recently, the Integrated System tures is becoming increasingly popular in struc-
for Structural Analysis and Record Selection tural design and assessment practice. Along these
(ISSARS) software has been developed lines, the selection and scaling of the appropriate
(Katsanos and Sextos 2013), retrieving dynami- set of earthquake ground motions, required in the
cally ground motions from the PEER-NGA data- framework of dynamic analysis, has become of
base, to form suites of records that not only paramount importance due to the significant sen-
comply with specific criteria but also ensure, sitivity of structural response to the assumptions
through numerical analyses of the structure that made in forming the necessary set of earthquake
records. A large number of ground motion selec- Baker JW, Cornell CA (2005) A vector-valued ground
tion and modification (GMSM) methods have motion intensity measure consisting of spectral accel-
eration and epsilon. Earthq Eng Struct Dyn
been proposed in the literature; still though, the 34:1193–1217. doi:10.1002/eqe.474
major progress made during the last decade is not Baker JW, Cornell CA (2006) Spectral shape, epsilon and
yet reflected in modern seismic codes. Along record selection. Earthq Eng Struct Dyn
these lines, the present entry aims at briefly 35(9):1077–1095. doi:10.1002/eqe.571
Beyer K, Bommer JJ (2007) Selection and scaling of real
presenting, in the simplest possible terms but accelerograms for bi-directional loading: a review of
not simpler than necessary, the objectives and current practice and code provisions. J Earthq Eng
fundamental concepts of GMSM methods along 11:13–45. doi:10.1080/13632460701280013
with the current seismic code framework and the CEN (2004) European Standard EN 1998-1. Eurocode
8: design of structures for earthquake resistance,
computational tools developed to facilitate code- part 1: general rules, seismic actions and rules for
prescribed procedures. buildings, committee for standardization. Design,
vol 3. European Committee for Standardization,
Brussels
CEN (2005) European Standard EN 1998-2. Eurocode 8:
Cross-References design of structures for earthquake resistance – part
2 bridges, committee for standardization, vol 3. Euro-
▶ Assessment of Existing Structures Using pean Committee for Standardization, Brussels
Response History Analysis CPA (2011) Seismic design code and commentary for
buildings, construction and planning agency. Ministry
▶ Conditional Spectra of Interior Affair, Taipei (in Chinese), pp 4–51
▶ Engineering Characterization of Earthquake Dias J (2010) SelEQ : a web-based application for the
Ground Motions selection of earthquake ground motions for structural
▶ Nonlinear Dynamic Seismic Analysis analysis. In: The 14th European conference on earth-
quake engineering, Ohrid, 30 Aug–3 Sept
▶ Performance-Based Design Procedure for Elnashai AS, McClure DC (1996) Effect of modelling
Structures with Magneto-Rheological assumptions and input motion characteristics on seis-
Dampers mic design parameters of RC bridge piers. Earthq Eng
▶ Probabilistic Seismic Hazard Models Struct Dyn 25(5):435–463
FEMA (2009a) NEHRP recommended seismic provisions
▶ Response-Spectrum-Compatible Ground for new buildings and other structures, FHMA 750.
Motion Processes Building Seismic Safety Council, Washington, DC
▶ Seismic Collapse Assessment FEMA (2009b) Quantification of building seismic perfor-
▶ Spatial Variability of Ground Motion: Seismic mance factors, FEMA P695. Federal Emergency Man-
agement Agency, Washington, DC
Analysis Hachem MM, Mathias NJ, Wang YY, Fajfar P, Tsai K-C,
▶ Stochastic Ground Motion Simulation Ingham JM, . . . Francisco S (2010) An international
▶ Time History Seismic Analysis comparison of ground motion selection criteria
for seismic design. In: Codes in structural engineering,
developments and needs for international practice,
joint IABSE – fib conference. Dubrovnik, pp 237–250
References Haselton CB (2009) Evaluation of ground motion selec-
tion and modification methods: predicting median
American Society of Civil Engineers (2005) Seismic interstory drift response of buildings, PEER report,
design criteria for structures, systems, and components 2009/01
in nuclear facilities, Structural Engineering Institute, Iervolino I, Galasso C, Cosenza E (2010a) New features of
Working Group for Seismic Design Criteria for REXEL 2. 61 beta, a tool for automated record selec-
Nuclear Facilities, ASCE/SEI 43-05, Reston, VA, 81 tion. In: Bulletin of earthquake engineering,
pp vol 8. Ohrid, 30 Aug–3 Sept, pp 339–362.
ASCE/SEI (2010) Minimum design loads for buildings doi:10.1007/s10518-009-9146-1
and other structures: ASCE standard 7-10. American Iervolino I, Galasso C, Cosenza E (2010b) REXEL: com-
Society of Civil Engineers/Structural Engineering puter aided record selection for code-based seismic
Institute, Reston structural analysis. Bull Earthq Eng 8:339–362.
Baker JW (2011) The conditional mean spectrum: a tool for doi:10.1007/s10518-009-9146-1
ground motion selection. J Struct Eng 137(3):322–331. Iervolino I, Galasso C, Paolucci R, Pacor F (2011) Engi-
doi:10.1061/(ASCE)ST.1943-541X.0000215 neering ground motion record selection in the ITalian
Sensitivity of First-Excursion Probabilities for Nonlinear Stochastic Dynamical Systems 3247
ACcelerometric Archive. Bull Earthq Eng validation study based on observed damage of an
9(6):1761–1778. doi:10.1007/s10518-011-9300-4 irregular R/C building. Soil Dyn Earthq Eng 1–15.
Iervolino I, Galasso C, Chioccarelli E (2012) REXEL 3.3: doi:10.1016/j.soildyn.2010.10.009
closing the loop of computer aided record selection. In: Shome N, Cornell CA, Bazzurro P, Carballo EJ
The 15th world conference on earthquake engineering, (1998) Earthquakes, records and nonlinear responses.
vol 10, Lisbon Earthq Spectra 14(3):469–500
Jalayer F, Cornell CA (2009) Alternative nonlinear Standards New Zealand (SNZ) (2004) NZS
demand estimation methods for probability-based 1170.5:2004 – Structural design actions. Earthquake
seismic assessments. Earthq Eng Struct Dyn actions. Standards New Zealand, Wellington
38(8):951–972 Tothong P, Luco N (2007) Probabilistic seismic demand
Jayaram N, Lin T, Baker JW (2011) A computationally analysis using advanced ground motion intensity mea-
efficient ground-motion selection algorithm for sures. Earthq Eng Struct Dyn 36:1813–1835.
matching a target response spectrum mean and vari- doi:10.1002/eqe
ance. Earthq Spectra 27(3):797–815. doi:10.1193/ Wang G (2010) A ground motion selection and modifi-
1.3608002 cation method capturing response spectrum charac-
Katsanos EI, Sextos AG (2013) ISSARS: an integrated teristics and variability of scenario earthquakes. Soil
software environment for structure-specific earth- Dyn Earthq Eng 1–15. doi:10.1016/j.soildyn.2010.
quake ground motion selection. Adv Eng Softw 11.007
58:70–85. doi:10.1016/j.advengsoft.2013.01.003 Youngs RR, Power MS, Wang G, Makdisi F, Chin CC
Katsanos EI, Sextos AG, Manolis GD (2010) Selection of (2007) Design ground motion library (DGML) – tool
earthquake ground motion records: a state-of-the-art for selecting time history records for specific engineer-
review from a structural engineering perspective. Soil ing applications. In: Proceedings of SMIP07 seminar
Dyn Earthq Eng 30(4):157–169. doi:10.1016/j. on utilization of strong-motion data, Sacramento
soildyn.2009.10.005
Katsanos EI, Sextos AG, Elnashai AS (2014) Prediction of
inelastic response periods of buildings based on intensity
measures and analytical model parameters. Eng Struct
71:161–177. doi:10.1016/j.engstruct.2014.04.007
Kottke AR, Rathje EM (2008) A semi-automated proce- Sensitivity of First-Excursion
dure for selecting and scaling recorded earthquake Probabilities for Nonlinear
motions for dynamic analysis. Earthq Spectra
24(4):911–932
Stochastic Dynamical Systems
Kramer SL (1996) Geotechnical earthquake engineering.
Prentice Hall, Upper Saddle River, N.J., 653 pp Marcos Valdebenito and Hector A. Jensen
Lin T, Haselton CB, Baker JW (2013a) Conditional- Department of Civil Engineering, Santa Maria
spectrum-based ground motion selection. Part II:
intensity-based assessments and evaluation of alterna-
University, Casilla, Valparaiso, Chile
tive target spectra. Earthq Eng Struct Dyn.
doi:10.1002/eqe
Lin T, Haselton CB, Baker JW (2013b) Conditional- Synonyms
spectrum-based ground motion selection. Part I: haz-
ard consistency for risk-based assessments. Earthq
Eng Struct Dyn. doi:10.1002/eqe First Excursion; Stochastic Dynamics; Sensitiv-
Luco N, Cornell CA (2007) Structure-specific scalar ity; Simulation Methods S
intensity measures for near-source and ordinary earth-
quake ground motions. Earthq Spectra 23(2):357–392.
doi:10.1193/1.2723158
McGuire RK (2004) Seismic hazard and risk analysis, Introduction
Earthquake Engineering Research Institute, Berkeley,
240 pp Quantification of the performance of structural
Naeim F, Alimoradi A, Pezeshk S (2004) Selection and
scaling of ground motion time histories for structural
systems subject to dynamic loading is of para-
design using genetic algorithms. Earthq Spectra mount interest in several fields of engineering and
20(2):413–426 particularly in the case of earthquake engineer-
Padgett JE, Desroches R (2007) Sensitivity of seismic ing. Knowledge on the performance of a structure
response and fragility to parameter uncertainty.
J Struct Eng 133(12):1710–1718
during seismic events allows taking design deci-
Sextos AG, Katsanos EI, Manolis GD (2010) EC8-based sions that ensure its serviceability and safety
earthquake record selection procedure evaluation: throughout its life. Nonetheless, quantification
3248 Sensitivity of First-Excursion Probabilities for Nonlinear Stochastic Dynamical Systems
of performance is a challenging task as there is structural system. For example, determining

always uncertainty on future loadings that affect the variation in probability due to a change in
a structure during its lifetime. Structural reliabil- the size of a structural member can provide use-
ity has emerged as a discipline that allows ful information to increase the safety level or to
accounting for the unavoidable effects of uncer- identify the most influential design parameters.
tainty over performance. Thus, probability the- Nonetheless, estimation of the sensitivity of
ory is used to describe the uncertainty associated first-excursion probabilities for dynamical sys-
with different relevant parameters that affect tems is a challenging task as it comprises not
performance by means of random variables, only taking into account the uncertainty in input
random fields, and/or stochastic processes. parameters but also assessing how performance
In this manner, uncertainty is propagated from is affected due to variations in properties of
these input parameters to the responses of inter- a structure.
est such as displacements, accelerations, This contribution presents an approach for
forces, etc. assessing probability sensitivity of systems subject
A particularly useful way to measure the to stochastic excitation with emphasis on struc-
effects of uncertainty in the dynamic response tures whose response is nonlinear. The approach
of structural systems is the so-called first- combines state-of-the-art simulation strategies
excursion probability. This probability is widely with a series of approximation concepts. Salient
used in stochastic structural dynamics and mea- features of the approach are the capability of con-
sures the chances that one or more structural sidering problems involving a large number of
responses exceed a prescribed threshold level random variables (in the order of thousands), the
within the duration of a dynamical excitation possibility of estimating sensitivity with respect to
(Soong and Grigoriu 1993). First-excursion several variables simultaneously (scalability), and
probability estimation is particularly challeng- a high numerical efficiency achieved by integrat-
ing as characterization of uncertain loading usu- ing an advanced simulation algorithm with local
ally comprises stochastic processes whose approximations of the functions modeling the
discrete representation can involve hundreds or structural performance.
even thousands of random variables. Similarly,
the number of possible failure criteria involved
can be extremely large as well, i.e., there can be Formulation of the Problem
several responses of interest that must be con-
trolled at a large number of discrete time Structural Model
instants. Hence, several different techniques Let f(t) be a scalar representing loading acting
have been proposed in order to estimate first- over a structure during a time span t [0, T].
excursion probabilities. Among these, methods This load is modeled at discrete time instants of
based on simulation (such as the Monte Carlo analysis tk = (k 1)Dt, k = 1, . . . , nT where Dt is
method and its more advanced variants) have the time step and nT is the number of time points
been shown to be the most appropriate approach considered (clearly, Dt = T/(nT 1)). The load-
to compute these probabilities (Schuëller ing f(t) is uncertain and is characterized by means
et al. 2004). of a stochastic process using an appropriate rep-
Although first-excursion probability provides resentation (Schuëller 1997), e.g., Karhunen-
a most useful way to rationally account for the Loève (KL) expansion, polynomial chaos
effects of uncertainty on structures subject to (PC) expansion, etc. Thus, the loading can be
stochastic loading, it is certainly not the only represented as f(tk, z) where z is a vector of ran-
metric that should be taken into account when dom variables of dimension nz whose associated
designing a system. In fact, it is also of interest probability density function fz(z) depends on the
analyzing the sensitivity of the probability with characteristics of the stochastic process. Note that
respect to variations in the properties of the depending on the specific situation under
analysis, load could also be represented as a - engineering, a response of interest ri could be

vector-valued stochastic process (i.e., a vector the interstory drift displacement, which can be
whose entries are stochastic processes) instead calculated as a linear combination of the dis-
of a scalar. However, in order to simplify the placement vector, i.e., ri(t, y, z) = gTi xi(t, y, z),
presentation of this contribution and with no where g i is a vector of size n whose entries are
loss of generality, the vector-valued case is not 0 and 1. For design purposes, the responses of
explored further. interest ri, i = 1, . . . , nr are checked against
In addition to the stochastic loading f(t, z), allowable threshold levels r*i , i = 1, . . . , nr. In
consider a vector y Oy ℝny of dimension ny a deterministic design framework, the objective
grouping the design variables of the structural is verifying that these responses do not exceed
system. This vector can group those variables their prescribed thresholds in order to avoid unde-
that can be altered during the design process sirable situations (such as loss of serviceability or
(such as cross section of structural members, collapse). However, when uncertainties are
material properties, etc.). Moreover, consider a explicitly taken into account, the aforementioned
classically damped structural system represented condition cannot be always satisfied, i.e., there is
by an appropriate model (e.g., a finite element the chance that the responses surpass their pre-
model (Bathe 1996)) comprising a total of scribed thresholds thus leading to an undesirable
n degrees of freedom. Then, the differential equa- situation.
tion describing the response of the structure sub- In order to account for the effects of uncer-
ject to the stochastic excitation f(t, z) is (see, e.g., tainties and potential undesirable performance of
Chopra, 1995) a structure during its lifetime, reliability offers
the means for quantifying the level of safety
associated with a structural system. A criterion
xðt, y, zÞ þ CðyÞx_ ðt, y, zÞ þ KðyÞxðt, y, zÞ
MðyÞ€
widely used for characterizing safety of
þ fNL ðy, xðt, zÞ, x_ðt, y, zÞÞ ¼ gf ðt, zÞ
a structure is the first-excursion probability (see,
(1) e.g., Soong and Grigoriu, 1993). This probability
measures the chances that uncertain structural
where x is the displacement response vector of responses exceed in magnitude prescribed thresh-
dimension n; M, C, and K are the mass, olds within a specified time interval. That is, first-
damping, and stiffness matrices of dimension excursion probability measures the chances of
n n; fNL ðx, yðt, y, zÞ, x_ðt, y, zÞÞ is a vector of occurrence of the following event F (which is
dimension n representing the effect of nonlinear termed in the sequence as failure event):
forces of the structure due to, e.g., special
devices for energy dissipation, yielding, F ¼ DN ðy, zÞ 1 (2)
nonlinear behavior, etc.; and g is the vector of S
dimension n that couples the stochastic excita-
where DN(y, z) is the so-called normalized
tion f(t, z) with the degrees of freedom of the
demand (Au and Beck 2001) and is defined as
structure. Note that mass, damping, stiffness,
and nonlinear forces are a function of the vector
jr i ðt, y; zÞj
of design variables y. DN ðy, zÞ ¼ max max (3)
i¼1,..., nr t ½0, T
r i
First-Excursion Probability
The performance of the structural system in The normalized demand represents the maximum
view of the stochastic excitation is characterized of the quotient between the structural responses
by means of nr responses of interest ri(t, y, z), of interest and their corresponding threshold
i = 1, . . . , nr, t [0, T] measuring, e.g., displace- levels. Clearly, whenever DN(y, z) exceeds
ments, accelerations, stresses, etc. For example, 1, there is failure as the response exceeds its
in applications associated with earthquake maximum acceptable value. In this context, note
the word failure is not a synonym of collapse. been developed, allowing to estimate small fail-
It is intended to denote unacceptable system per- ure probabilities at affordable numerical costs
formance which can range from partial damage (Schuëller et al. 2004).
states (e.g., loss of serviceability) to collapse
depending on the specific application under Sensitivity of First-Excursion Probability
study. As shown in Eq. 4, the probability of failure is
The probability of occurrence of the failure a function of the design variable vector y. Such
event can be defined by means of the following dependence can be understood as follows: modi-
classical probability integral: fications on the design vector do affect the
ð response of the structure and, consequently, also
PF ðyÞ ¼ P½DN ðy, zÞ 1
¼ f z ðzÞdz affect the probability of exceeding the prescribed
DN ðy, zÞ1 thresholds. In consequence, for decision making
ð
and risk analysis, it is important to evaluate the
¼ I F ðy, zÞf z ðzÞdz probability of failure PF (y) and its sensitivity
z Oz
(4) with respect to changes in the design vector y.
That is, besides computing the value of the failure
probability, it is also of relevance estimating how
In the above equation, P[] denotes probability of
much the probability changes due to
occurrence of the argument between brackets,
a modification of the design variable vector.
PF (y) represents the probability of failure (i.e.,
A classical measure for sensitivity is calculating
probability of occurrence of the event F), and IF
the gradient of the quantity of interest. However,
(y, z) denotes the indicator function which is
within the context of nonlinear dynamics, the
equal to 1 in case the normalized demand is
estimation of such quantity may not be feasible
equal or larger than 1 and 0 otherwise.
as the gradient may not exist due to the
The evaluation of the failure probability inte-
nonsmooth normalized demand defined in Eq. 3
gral is a challenging task as it usually involves
(see, e.g., Kang et al. 2006). In order to circum-
a high number of dimensions (for applications of
vent this difficulty, an approximate representa-
practical interest, nz can be in the order of thou-
tion of the failure probabilities that is
sands) and the normalized demand function
differentiable is constructed, and then, the gradi-
DN(y, z) can be evaluated point-wise only through
ent of this approximation is estimated. Details on
(numerically demanding) FE analyses.
these approximations are discussed below.
A possible means for evaluating probability inte-
grals is applying simulation methods. Among
different available simulation methods, Monte
Carlo simulation (MCS) (Metropolis and Ulam Estimation of First-Excursion Probability
1949) is the most general technique. However, Sensitivity
MCS is numerically demanding for estimating
low failure probabilities (which are typical in From a mathematical viewpoint, estimating the
engineering applications). In order to circumvent gradient of the first-excursion probability implies
this issue, advanced simulation techniques have solving the following limit:

@PF ðyÞ PF y þ vðqÞDyq PF ðyÞ
¼ lim , q ¼ 1, . . . , ny (5)
@yq Dyq !0 Dyq
In Eq. 5, v(q) is a vector of dimension ny with all which is equal to one. Introducing the definition
entries equal to zero, except by the q-th entry, of failure probability (see Eq. 4) in Eq. 5 yields

@PF ðyÞ PF DN y þ vðqÞDyq , z 1 P½DN ðy, zÞ 1
¼ lim , q ¼ 1, . . . , ny (6)
@yq Dyq !0 Dyq
Note the limit (as well as the partial derivative) in the design variable vector, and aq, q = 1, . . . , ny
the above equation may not exist as the normal- are real, constant coefficients. The procedure to
ized demand function DN(y, z) may be nonsmooth determine these coefficients is described in sec-
with respect to both the design parameter vector y tion “Numerical Implementation.” The second
and the uncertain variable vector z. In order to approximation involves an approximate repre-
avoid this issue and still obtain a sensitivity mea- sentation of the probability that the normalized
sure of the first-excursion probability, approxi- demand exceeds a threshold level b, i.e.,
mate representations for the normalized demand
function and the excursion probability are intro-
P½DN ðy, zÞ b
ec0 þc1 ðb1Þ , b ½1 ϵ, 1 þ ϵ
duced. These approximations – which were pro-

posed in Jensen et al. (2009) and Valdebenito and (8)
Schuëller (2011) – are discussed in the following.
The first approximation comprises an approxi- where ϵ is a small constant and where c0, c1 are
mate representation of the normalized demand real coefficients. The issue of how to calculate
function, i.e., them is analyzed in section “Numerical Imple-
mentation.” A thorough discussion on the appli-
~ N ðy þ Dy, zÞ
DN ðy þ Dy, zÞ D cability of these two approximations can be found
in Jensen et al. (2009) and Valdebenito and
X
ny
¼ DN ðy, zÞ þ aq Dyq (7) Schuëller (2011).
q¼1 Using the approximations introduced in Eqs. 7
and 8, it can be shown (see Valdebenito and
~ N ðy, zÞ is the approximate normalized
where D Schuëller (2011)) that the sought gradient can
demand function, Dy is a certain perturbation of be approximated as follows:

@ P~F ðyÞ ~ N y þ vðqÞDyq , z 1 PðDN ðy, zÞ 1
P D
¼ lim (9)
@yq Dyq !0 Dyq
P½DN ðy, zÞ 1
aq Dyq
PðDN ðy, zÞ 1
¼ lim (10) S
Dyq !0 Dyq
eco c1 aq Dyq eco representation of the failure probability with

¼ lim (11) respect to yq, q = 1,. . ., ny and P^F ðyÞ is the
Dyq !0 Dyq
estimate for the failure probability in Eq. 4
which is calculated using an advanced simulation
¼ c1 aq eco (12) method. Thus, for estimating the gradient of the
probability, it is necessary to determine the prob-
¼ c1 aq P~F ðyÞ (13) ability P^F ðyÞ and the coefficients c1 and aq, q =
1,. . ., ny. A procedure for determining this prob-
In the above equations, @ P~F ðyÞ=@yq represents ability and these coefficients is discussed in the
the partial derivative of the approximate following section.
Numerical Implementation a rarer occurrence than the previous one, i.e., P

[Fi] < P[Fi–1]. Hence, a full run of SS does not
General Remarks only provide the probability of occurrence asso-
This section provides details on the procedure to ciated with a normalized threshold b = 1 (see
calculate the different coefficients and probabil- Eq. 4), but actually for a range of the normalized
ity involved in the estimation of the first- threshold levels (Au and Beck 2001).
excursion probability sensitivity according to
the approximate formula proposed in Eq. 13. Sensitivity of First-Excursion Probability with
First, it is explained how excursion probabilities Respect to Threshold
are estimated. Then, the sensitivity of the first- The approximation introduced in Eq. 8 allows esti-
excursion probability with respect to mating the probability that the normalized demand
a normalized threshold level is analyzed, and function DN(y, z) exceeds a prescribed threshold b,
finally, an approximate representation of the nor- i.e., it is sought to estimate the curve relating nor-
malized demand function is discussed. malized thresholds b and their corresponding first-
excursion probabilities. Note that this curve can be
First-Excursion Probability Estimations calculated as a byproduct of reliability analysis
As already pointed out, estimating a first- applying advanced simulation methods such as
excursion probability implies solving the integral SS. Thus, the coefficients c0 and c1 can be esti-
in Eq. 4. In this contribution, this probability is mated with no additional numerical efforts once
evaluated by means of the so-called advanced a reliability analysis has been carried out. In fact,
simulation methods (Schuëller et al. 2004). In when applying SS, the sought coefficients can be
particular, in this contribution, subset simulation calculated with the samples of the normalized
(SS) (Au and Beck 2001) is applied to estimate demand generated at the last stage of SS. The
first-excursion failure probabilities. main idea is generating the curve probability versus
In SS, the failure domain F (see Eq. 2) is normalized threshold in a discrete manner using
defined as a sequence of subsets samples already available. Then, the sought coeffi-
(or intermediate failure events) Fi, i = 1,. . .,m cients are calculated in a least square sense consid-
such that F1 F2 . . . Fm = F. Thus, the ering the analytic approximation of Eq. 8. Details
failure probability is cast as a product of condi- on how to implement this procedure are discussed
tional failure probabilities, i.e., in Valdebenito and Schuëller (2011).
m1
PF ¼ P½Fm
¼ P½F1
∏ P½Fiþ1 jFi
(14) Approximate Representation of Normalized
i¼1
Demand Function
where P [Fi+1| Fi] is the probability of occurrence The approximate representation in Eq. 7 suggests
of the event Fi+1 conditioned on the event Fi. In that changes in the normalized demand function
this way, a small failure probability is expressed DN(y, z) due to changes in the design variable
as the product of larger, conditional probabilities vector y can be explained through a linear relation.
which can be calculated using, e.g., Monte Carlo Such relation is clearly not captured exactly by
simulation (MCS). The practical implementation such a simple expression due to a number of rea-
of SS requires an efficient algorithm for generat- sons: higher-order terms involving y are ignored,
ing samples of the uncertain parameters (z) con- no interaction between y and z is considered, the
ditioned on an intermediate failure event, such as nonlinear nature of the normalized demand func-
the modified Metropolis algorithm (Au and Beck tion (see Eq. 3) is not captured appropriately,
2001). Details on the algorithmic implementation etc. However, in case the coefficients aq, q =
of SS can be found in Au and Beck (2001). 1,. . ., ny are chosen appropriately, it could be
A salient feature of SS is that it populates the expected that D ~ N ðy þ Dy, zÞ approximates
space of uncertain parameters by means of suc- DN(y + Dy, z) sufficiently well. That is, the coeffi-
cessive subsets Fi, i =1,. . ., m, each of which is of cients aq represent an average sensitivity on how
Sensitivity of First- NLD

Excursion Probabilities rigid panel
3.5 [m]
for Nonlinear Stochastic
NLD
Dynamical Systems,
Fig. 1 Example – 2-story 3.5 [m] rigid panel
RC frame structure
including nonlinear devices 5 [m] 5 [m] 5 [m] 5 [m] 5 [m]
(NLDs) gA(t )
the design variables affect the normalized demand. takes place whenever displacement of each floor
Numerical validation (see Jensen et al. 2009; of the building exceeds a prescribed threshold
Valdebenito and Schuëller 2011) has shown such within the duration of the stochastic ground accel-
an assumption is appropriate within the scope of eration. The design variables of the problem refer
the problems studied in this contribution. to the dimensions of the columns of the RC frame.
The procedure applied to calculate the coeffi- The RC frame possesses a Young’s modulus
cients aq, q = 1, ny is quite straightforward. Sam- equal to 2 1010 [N/m2]. Each of its floors
ples of the uncertain variable vector z are taken (of mass 1.5 105 [kg]) is supported by six
from the last stage of SS. It is expected that for columns of square cross section (side length
these samples the normalized demand function is equal to 0.5 [m]) and a height of 3.5 [m]. The
close to 1. Then, for each of these samples, the beams of the frame are rigid in the axial direction,
value of the normalized demand is reevaluated so each floor can be described by a single hori-
considering perturbed values of the design vari- zontal degree of freedom (DOF); thus, the model
able vector. Then, the sought coefficients are involves a total of two DOFs. It is assumed that
estimated in a least square sense considering the the columns and beams remain linear within the
sampled data and the analytic model of Eq. 7. For duration of the stochastic ground acceleration and
details on the implementation of this procedure, classical modal damping of 5 % for all modes is
the reader is referred to Jensen et al. (2009) and assumed.
Valdebenito and Schuëller (2011). In order to improve the safety of the frame,
two hysteretic NLDs are included. The restoring
force (FR,i()) associated with the i-th NLD is
Example described by the following model (Pradlwarter
and Schuëller 1993):
Description of the Problem
In order to illustrate the application of the proce-
FR, i ðtÞ ¼ kd Di ðtÞ q1i ðtÞ þ q2i ðtÞ , i ¼ 1, 2
dure for failure probability sensitivity estimation,
the following example is considered. It involves (15)
S
a two-story reinforced concrete (RC) frame
which includes nonlinear hysteretic devices where Di() is the relative displacement between
(NLDs). Figure 1 illustrates the elevation of the the (i, i – 1)-th floors, kd is the stiffness of the
model. The frame is excited by a horizontal NLD, and q1i () and q2i () denote the plastic elon-
ground acceleration of 15 [s] duration, which is gations of the NLD, which are governed by the
modeled as a stochastic process. The failure event following equations:
ti ðtÞ Dy

q_1i ðtÞ _ _
¼ Di ðtÞH Di ðtÞ H ti ðtÞ Dp þ H ti ðtÞ Dy H Dp ti ðtÞ , i ¼ 1, 2 (16)
Dp Dy
ti ðtÞ Dy

_ _ _
qi ðtÞ ¼ Di ðtÞH Di ðtÞ H ti ðtÞ Dp þ H ti ðtÞ Dy
2
H Dp þ ti ðtÞ , i ¼ 1, 2 (17)
Dp Dy
where ti(t) is an auxiliary variable defined as Sensitivity of First-Excursion Probabilities for

ti(t) = Di(t) q1i (t) + q2i (t), i = 1, 2, and Dy and Nonlinear Stochastic Dynamical Systems,
Dp are the yielding and plastic displacements, Table 1 Estimates of first-excursion probability sensitiv-
ity (cov: coefficient of variation)
respectively. The numerical values considered
for the parameters of the NLD are kd = 108 Proposed approach Finite differences
@ P~ F ðyÞ @ P~ F ðyÞ @ P~ F ðyÞ @ P~ F ðyÞ
[N/m], Dp = 6 103 [m], and Dy = 0.8Dp. @y1 @y2 @y1 @y2
The ground acceleration (gA(t)) is modeled as (cov) (cov)
a filtered white noise of 15 [s] duration. The 2.24 1.11 2.49 1.19
ground acceleration is calculated as gA(t) = aT (20.7)% (24.8)%
p(t); the vectors aT and p(t) are defined as
displacement. The threshold values are chosen
aT ¼< O21 , 2x1 O1 , O22 , 2x2 O2 > (18) equal to 0.2 % of the floor height for the interstory
drift displacements and 0.1 % of the frame height
0 1 for the roof displacement. The design variables
0 1 0 0
B O2 2x O C refer to the cross sections of the columns of the
B 1 1 0 0 C
p_ ðtÞ ¼ B 1
C pð t Þ RC frame, more specifically to the second
@ 0 0 0 1 A
moment of area of the cross section. These cross
O12
2x1 O1 O2 2x2 O2
2
sections are grouped into two design variables
0 1
0 linking the columns of the first and second floor,
B oðtÞeðtÞ C respectively.
B C
þB C
@ 0 A
Results
0
The sensitivity of the first-excursion probabilities
(19) is estimated using the approach described previ-
ously. In particular, SS is applied considering
where O1 = 15 [rad/s], x1 = 0.8, O2 = 0.3 [rad/s],
2,000 samples of the uncertain ground accelera-
and x2 = 0.995 are the filter parameters; o(t) tion at each simulation stage. The resulting first-
denotes a white noise signal; and e(t) is an enve-
excursion probability estimate is equal to
lope function:
P~F ¼ 103 . Hence, a total of 6,000 samples of
8 2 the uncertain variables are required in order to
< t =16 0½s
t < 4½s
estimate the sought probability. Recall that these
e ðt Þ ¼ 1 4½s
t < 10½s
(20)
: ðt10Þ2 samples are also used to estimate the coefficients
e 10½s
t 15½s
c0 and c1. In order to calibrate the approximate
model of Eq. 7, 200 perturbed designs are ana-
A time discretization step equal to Dt = 0.01 [s] is lyzed. Hence, the estimation of the probability
used to model the ground acceleration. Thus, the sensitivity demands only 200 extra structural ana-
discrete representation of the white noise signal is lyses. In order to illustrate the variability associ-
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
oðtk Þ ¼ 2pS=Dt zk, k ¼ 1, . . . , 1501, where ated with the approach for estimating first-
S = 103 [m2/s3] is the spectral density of the excursion probability sensitivity, a total of
white noise and zk, k = 1,. . ., 1501 are indepen- 100 independent runs were generated. The results
dent, identically distributed standard Gaussian in terms of the mean of these runs and their
variables. corresponding coefficient of variation are shown
The failure event is formulated as a first- in Table 1. In addition and in order to validate the
excursion problem during the time of analysis; results obtained in Table 1, the probability sensi-
the structural responses to be controlled are the tivity is estimated using a central finite difference
two interstory drift displacements and the roof estimator. In order to ensure the probability
Sensors, Calibration of 3255
estimates used in the finite difference scheme are References

sufficiently accurate, the average of 100 indepen-
dent runs is considered. The results associated Au S, Beck J (2001) Estimation of small failure probabil-
ities in high dimensions by subset simulation. Proba-
with the finite differences are presented in Table 1
bilist Eng Mech 16(4):263–277
as well. It can be observed that the results Bathe K (1996) Finite element procedures. Prentice Hall,
presented are in good agreement, indicating the New Jersey
approach for sensitivity estimation reported Chopra A (1995) Dynamics of structures: theory and
applications to earthquake engineering. Prentice Hall,
herein provides appropriate results.
Englewood Cliffs
Jensen H, Valdebenito M, Schuëller G, Kusanovic
D (2009) Reliability-based optimization of stochastic
Summary systems using line search. Comput Method Appl Mech
Eng 198(49–52):3915–3924
Kang B-S, Park G-J, Arora J (2006) A review of optimi-
This contribution has presented an approach for zation of structures subjected to transient loads. Struct
sensitivity analysis of first-excursion probabili- Multidiscip Optim 31(2):81–95
ties associated with nonlinear dynamical sys- Metropolis N, Ulam S (1949) The Monte Carlo method.
J Am Stat Assoc 44(247):335–341
tems. The basis of the approach is combining
Pradlwarter HJ, Schuëller GI (1993) Equivalent lineariza-
advanced simulation methods for probability tion – a suitable tool for analyzing MDOF systems.
estimation with a series of local approximations Probabilist Eng Mech 8(2):115–126
involving the normalized demand function. Schuëller G (1997) A state-of-the-art report on computa-
tional stochastic mechanics. Probabilist Eng Mech
A salient feature of the approach reported
12(4):197–321
herein is that it is numerically efficient. This is Schuëller G, Pradlwarter H, Koutsourelakis P (2004)
achieved as the proposed local approximations A critical appraisal of reliability estimation procedures
take advantage of the results already available for high dimensions. Probabilist Eng Mech
19(4):463–474
from probability estimation.
Soong T, Grigoriu M (1993) Random vibration of
The results presented in this contribution indi- mechanical and structural systems. Prentice Hall,
cate that the proposed approach for sensitivity is Englewood Cliffs
applicable for problems involving a large number Valdebenito M, Schuëller G (2011) Efficient strategies for
reliability-based optimization involving non linear,
of uncertain parameters; results presented else-
dynamical structures. Comput Struct 89(19–20):
where (see, e.g., Valdebenito and Schuëller 2011) 1797–1811
indicate that the approach is also capable of includ-
ing a considerable number of design variables.
Cross-References Sensors, Calibration of

S
▶ Reliability Estimation and Analysis Jens Havskov1 and Gerardo Alguacil2
▶ Robust Design Optimization for Earthquake 1
Department of Earth Science, University of
Loads Bergen, Bergen, Norway
▶ Stochastic Analysis of Nonlinear Systems 2
Instituto Andaluz de Geofı́sica, University of
▶ Structural Optimization Under Random Granada, Granada, Spain
Dynamic Seismic Excitation
▶ Structural Reliability Estimation for Seismic
Loading Synonyms
▶ Structural Seismic Reliability Analysis
▶ Subset Simulation Method for Rare Event Accelerometer; Response function; Seismome-
Estimation: An Introduction ter; Seismometer damping
3256 Sensors, Calibration of
Introduction basic parameters will be described, some tests

used to obtain them as well as methods for
The output of a seismic sensor, a seismometer or obtaining A(o) directly.
an accelerometer, is a time-varying voltage,
which is related to the ground motion by
a differential equation in the time domain or Seismic Sensors
by a transfer function in the frequency domain.
This transfer function or response function is Seismic sensors can be divided into two kinds:
characterized by a number of parameters, sensors measuring the ground velocity (velocity
which are assumed to be constant, at least in sensor or seismometer) and sensors measuring
the short term. the ground acceleration (accelerometers). A seis-
A seismic sensor will have calibration infor- mometer can be passive, meaning there is no
mation given by the manufacturer. These specifi- electronic parts or active with an electronic cir-
cations will be used to correct the seismic signal cuit integrated. All accelerometers for seismic
from the sensor to produce the true ground recording are active sensors. Piezoelectric accel-
motion. If, e.g., for a given harmonic ground erometers are passive (although they may have
velocity X(o), the output from the sensor is a conditioning circuit built-in) and are widely
Y(o), the amplitude response or transfer function used for structural vibrations monitoring, but sel-
A(o) is defined as the ratio A(o) = Y(o)/X(o), dom in seismology, due to their poor sensitivity at
where o is the frequency in radian/s. So, if A(o) low frequency and low dynamic range.
is known, the input (ground motion) can then be Passive Seismometers: A passive seismome-
calculated as ter consists of a swinging mass with a coil mov-
ing in magnetic field. It is also called an
XðoÞ ¼ Y ðoÞ=AðoÞ (1) electromagnetic sensor. The swinging system
has resonance frequency o0. An example is
The sensor will, in addition to changing the seen in Fig. 1.
amplitude of the ground signal, also change the When the mass is moving, the magnetic field
phase of the signal, so an additional function will will vary in the coil. An output voltage propor-
be needed for the phase response. The phase and tional to the velocity of the mass relative to the
amplitude response functions can be combined in ground will then be produced. The proportional-
one complex function (A(o) would then be com- ity constant is called the generator constant G and
plex); however, here, for simplicity, they will be has the unit V/ms1. But the relative motion of
treated separately. the mass depends on the frequency of the ground
The amplitude and phase response functions motion. Thus the sensor will not be equally sen-
are usually calculated from the sensor parameters sitive to ground motion for all frequencies. Qual-
but can also be obtained by direct measurements. itatively the response of the sensor can be
With time, the sensor might degrade or it understood as follows. If the ground moves with
might develop a fault so there is a need to be a very fast sinusoidal motion, the mass remains
able to calibrate the sensor. This means deter- stationary in an inertial frame, and thus the
mining the instrument calibration parameters ground sinusoidal velocity is measured directly.
(parametric calibration) or determining A(o) With the ground moving very slowly, the mass
directly for each frequency of interest would have time to follow the ground motion, so
(empirical calibration). there would be little relative motion and the gain
Modern sensors can be very complicated in would be low. At the resonance frequency, the
terms of electronics and mechanical construction mass could get a new push at the exact right time,
so it might not be possible to determine all rele- so the mass would move with a larger and larger
vant parameters; however, some parameters can amplitude, only limited by the damping of the
be obtained with simple tests. In this section, motion. The sensor amplitude response A(o),
through a resistor R in a negative feedback loop.

The polarity of the current is such that it opposes
any motion of the mass, and it will try to prevent
the mass from moving at all with respect to the
frame. A small permanent acceleration on the
mass will therefore result in a small permanent
current, and a large acceleration will need a large
current. The current is in fact proportional to the
ground acceleration, so the voltage over the resis-
tor gives a direct measure of acceleration. This is
how nearly all accelerometers work, and in prac-
tice the only constant of importance is the gener-
ator constant G in units of V/g, where g is the
gravity acceleration.
Sensors, Calibration of, Fig. 1 A model of an electro-
The FBA principle is now the heart of nearly
magnetic sensor. The mass m is also a magnet suspended all modern strong motion and broadband sen-
by a spring k. The coil resistance is Rg, the damping sors (BB, sensors recording in a large frequency
resistor is R, and the voltage output is Vout. The mass band like 0.01–50 Hz). By connecting an inte-
motion is damped by the current through the coil (fixed
to the frame) and the external damping resistor
grating circuit after the output, the sensor can
give out a voltage proportional to velocity like
for passive sensors. However, due to the
which is the output voltage of the sensor as mechanical-electrical qualities of the sensor,
a function of the input ground velocity, can be there is in practice a low frequency limit for
obtained as (e.g., Lay and Wallace 1995) the flat velocity response. For lower frequen-
cies, the amplitude response decreases propor-
tional to frequency squared. This means that
Go2 in practice, a BB sensor will have the equivalent
AðoÞ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2 ffi (2)
o0 o2 þ 4h2 o2 o20 of a free period and a damping, and its ampli-
tude response can be approximated by Eq. 2,
although a more exact model of its response
where h is the damping, o is the frequency, and should include additional parameters that
o0 is the natural frequency of the swinging sys- influence the high-frequency behavior. In prac-
tem. Figure 2 shows examples of A(o) with dif- tice Eq. 2 must then be multiplied with a func-
ferent damping. tion which often represents a filter, like a
It is seen that, as the damping decreases, the Butterworth filter that can be represented S
gain gets a larger and larger peak at the natural by a simple function of filter frequency and
frequency. The flat curve represents a damping of number of poles of the filter; see Havskov and
h = 0.707 and is the desired value. The damping Alguacil (2010). An estimate for the filter
is achieved by shunting the signal coil with parameters can be obtained by empirical cali-
a damping resistor and thereby draining energy bration; see later.
out of the swinging system. The sensor parameters to measure can there-
Active Sensors: The heart of the active sensor fore be summarized as instrument natural fre-
is a device measuring the ground acceleration, the quency, damping, and generator constant. With
so-called force balanced accelerometer (FBA) these parameters, the amplitude response func-
(Fig. 3). The FBA has a feedback coil, which tion Eq. 2 can be calculated.
can exert a force equal and opposite to the inertia Alternatively the amplitude response function
force due to the acceleration. The displacement can be determined directly by measuring the out-
transducer sends a current to this force coil put of the sensor with a controlled input.
AMPLITUDE RESPONSE FOR GROUND VELOCITY

104
h = 0.07
h = 0.28
h = 0.71
103
102
V/(m/s)
101
100
10−1
0,1 1 10 100
FREQUENCY Hz
Sensors, Calibration of, Fig. 2 Amplitude response to ground velocity of a typical passive sensor with a natural
frequency of 1 Hz for different damping values
Sensors, Calibration of,

Fig. 3 Simplified principle
behind the force balanced
accelerometer. The spring
displacement transducer
normally uses a capacitor
C, whose capacitance
varies with the
displacement of the mass. mass
Volt out ∼
A current, proportional to Force R
acceleration
the displacement coil
transducer output, will
force the mass to remain
stationary relative to the
frame (Figure from Displacement
transducer C
Havskov and Alguacil
2010)
Determining Sensor Parameters damping makes it swing at a lower frequency

o1. The relation is (Havskov and Alguacil 2010)
Natural Frequency
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
The frequency with which the seismometer mass
o1 ¼ o20 1 h2 (3)
is oscillating depends on the damping; more
100
z1 z3
h = 0.02
z2 z4
−100
100
a = 30 a = 19 h = 0.07
−100
100
h = 0.28
50
−50
0 5 10 15
TIME s
Sensors, Calibration of, Fig. 4 Free swinging of three z2, etc. (top trace). On the middle trace, the amplitudes of
typical 1 Hz seismometers with different open-circuit two maximums following each other are given in an
damping h. Traces have been generated synthetically. arbitrary scale (Figure modified from Havskov and
The decaying extremes (peak amplitudes) are labeled z1, Alguacil 2010)
The natural frequency is therefore best measured a calibration pulse from which the “apparent free
without any external damping. The free period period” and damping can be obtained; see section
can be determined in several ways depending on on calibration pulse later.
the type of sensor. If the open-circuit damping is high, say larger
Passive Sensor: Give the sensor an impulse or than 0.15 (corresponding to about 1 % change in
a step to make it swing, a small push or tilt will period), the measured free period should be
do. For a very few sensors, it is possible to see the corrected for damping.
mass swing, else the output signal has to be
observed on an oscilloscope or recorded. Mea- Damping
sure the time of several swings and calculate the The open-circuit damping hm is caused by the
average or measure the frequency from the friction in the mechanical system and can be S
recorded signal. Many short-period sensors determined from the signals from the free swing-
(natural frequency higher than 1 Hz) have too ing system as shown in Fig. 4. Using the extreme
high open-circuit damping (damping without amplitudes z1 and zn where z1 is extreme 1 and zn
a damping resistor) to observe more than a few is extreme n + 1 (so n = 1 for two following
swings of the signal (Fig. 4), and the signal must extremes (Fig. 4)), the open-circuit damping can
be recorded to measure the period. be determined as (Havskov and Alguacil 2010)
BB Sensor: The response of a BB sensor is
controlled mainly by its internal feedback loop. lnðz1 =zn Þ
The user can only measure its “apparent free hm ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (4)
period”; however, this is difficult since the n2 p2 þ ln2 ðz1 =zn Þ
damping is fixed at 0.7 so there will be almost
no swinging (see Fig. 9). Most BB sensors have Measuring z1 and z2, hm can now be determined
a calibration input, which can be used to produce directly. In Fig. 4, two following maxima (n = 2)
Sensors, Calibration of,

Fig. 5 Tilting an
accelerometer to determine m
generator constant. It is
assumed that the sensor
horizontal direction is East mg sin(a)
toward the east α
(Figure from Havskov and
Alguacil 2010) mg cos(a)
mg α
have approximate amplitudes 30 and 19, respec- When h2 = 0.707, the corresponding
tively. This gives a damping of RT2 = R0.707 is
lnð30=19Þ h1 hm
hm ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ 0:0725 (5) R0:707 ¼ RT1 (9)
0:707 hm
22 p2 þ ln2 ð30=19Þ
so the procedure to determine R0.707 is:
A sensor in operation should always have
a damping resistor connected in order to get the 1. Determine hm as described above.
desired total damping of 0.707 so as to get the 2. Connect a damping resistor RT1 of “suitable
most flat response. The additional damping size,” meaning that the overswing is substan-
caused by the electrical damping is called he, tially less than when measuring hm but large
and the corresponding total resistance of the coil enough to get an accurate determination of h1.
and damping resistor is RT. The damping resistor 3. Use Eq. 9 to determine R0.707.
cannot be determined experimentally as above,
since with total damping of 0.707, there would be For BB sensors or other active sensors, there is
too little overswing. It can be shown (Havskov no damping resistor, and the damping is preset
and Alguacil 2010) that the ratio between two from the factory. These sensors have a buffered
electrical damping constants is inversely related low-impedance voltage output, so an external
to the ratio of the damping resistances: load resistor would not influence the sensor at all.
he1
RT2 ¼ RT1 (6) Generator Constant
he2
In order to determine the generator constant exper-
imentally, a known input to the sensor must be
RT1 and RT2 are two different total damping resis-
used or the generator constant must be calculated
tances and he1 and he2 the corresponding damping
from other known measurable parameters.
coefficients caused by the resistances, respec-
Accelerometer: For a static ground accelera-
tively. In addition comes the open-circuit
tion, an accelerometer has an output proportional
damping, so, the total damping with RT1 is h1:
to this acceleration. An accelerometer cannot dis-
tinguish any difference between an inertial force
h1 ¼ he1 þ hm (7) due to ground acceleration and an equivalent
gravity force. By tilting the accelerometer, the
and writing Eq. 6 in terms of the total damping effective force on the three components can be
gives determined for different tilt angles and the sensi-
tivity determined; see Fig. 5.
he1 h1 hm The accelerometer is tilted an angle a. The
RT2 ¼ RT1 ¼ RT1 (8)
he2 h2 hm force in the horizontal direction is now mgsin(a),
Seismo-
Amplifier Filter
meter 1
Seismo-
Amplifier Filter
meter 2
Sensors, Calibration of, Fig. 6 Comparison of signals frequencies. The ratio of the output amplitudes (right)
recorded from two different seismometers with the same indicates the ratio of the sensor generator constants
recording equipment. The filters and amplifiers are iden- (Figure from Havskov and Alguacil 2010)
tical and filter out signals below the seismometers natural
while in the vertical direction it is mgcos(a). The All Sensors: A shaking table is a platform that
voltage output for the vertical component is Vz, moves with a controlled motion. Setting it up to
and for the horizontal component it is Vh. Con- move with a harmonic motion with maximum
sidering that in the horizontal position, the output amplitude A (velocity or acceleration) at a fre-
is supposed to be zero and that the vertical force quency where the response curve is flat, the output
has been decreased from mg to mgcos(a), the from the sensor will also be a harmonic function
generator constant can be calculated as with amplitude B. The generator constant is then
V z Vh G ¼ B=A (12)
Gz ¼ and Gh ¼ (10)
gð1 cos ðaÞÞ g sin ðaÞ
Unfortunately a shaking table is rarely avail-
able, so other methods must generally be used.
Vz will be negative since the vertical force is
The ground is always moving so if the motion
decreased, and Vh will be positive since the force
is known, it is equivalent to having a shaking
is in the east direction. This method is very sim-
table moving both horizontally and vertically.
ple and can determine the generator constant
The problem is then to determine this motion in
accurately, but will not give a dynamic calibra-
order to have an absolute measure. This is done
tion. The horizontal components can be tested for
by using a well-calibrated sensor. The experi-
symmetry by inclining in the opposite direction.
ment then consists of setting up the two sensors
In the case that the instrument output is not zero
close to each other and assuming that they are
in the horizontal position, this offset should be S
subject to the same ground motion (Fig. 6). The
adjusted before any measurement is made; alter-
data is filtered so as to use a frequency band
natively it may be subtracted from the output for
where both sensors are measuring the signal in
each measurement.
the flat part of their instrument response function,
Passive Velocity Seismometer: The generator
e.g., from 2 to 5 Hz using a BB sensor as a refer-
constant can be calculated from
ence and a 1 Hz sensor to be calibrated.
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi The generator constant can then be determined
G¼ 2mo0 he RT (11)
as
where m is the mass of the moving mass and he G ¼ GR B=A (13)
is the electrical damping corresponding to the
total resistance RT. In the previous section it has where GR is the generator constant of the refer-
been described how RT, he, and o0 can be ence sensor, A is the amplitude of the reference
determined. sensor, and B is the amplitude of the unknown
sensor. The amplitude can be picked for any part

of the two seismograms provided that they are
identified as the same “swing”; see Fig. 6. It is Sensor
here assumed that both sensors are of the same
type like comparing a velocity sensor to
a velocity senor or an accelerometer to an accel-
erometer. Note that many accelerometers are not
sensitive enough to be able to use this method, so
the tilt method should be used.
Calibration of Velocity Sensors by Stepwise
Motion: The main idea behind this method is to
(1) move the sensor to a known distance (like x
1 mm) in its direction of sensitivity, (2) record
the signal, and (3) correct the signal for the
known displacement frequency response Sensors, Calibration of, Fig. 7 Moving a sensor at
(response function relating sensor output to dis- a controlled distance vertically with a mechanical balance.
The ratio between the motion at x (where we can measure
placement) to get the displacement. Theoreti-
it) and on the balance table can be determined by placing
cally, the displacement is now calculated, which a mass m1 at x that will balance a mass m2 on the balance
can be compared to the actual displacement, from table. The ratio of the two motions is then m1/m2
which the generator constant can be calculated. (Figure from Havskov and Alguacil 2010)
The method is described in detail by Wielandt in
chapter “▶ MEMS Sensors for Measurement of It should be noted that some of the triaxial BB
Structure Seismic Response and Their Applica- seismometers have the three sensors arranged
tion” in NMSOP (Bormann 2002), which claims symmetrically, forming an orthogonal frame
that the method works well for broadband sensors with its active axes inclined 54.7 with respect
(accuracy down to 1 %) and even 10 Hz seis- to horizontal, so each sensor is sensitive to verti-
mometers. Although the method sounds simple, it cal motions. Their outputs are then electronically
is not trivial to correct for the instrument response combined to generate the conventional two hori-
down to DC, particularly for SP instruments. zontal and one vertical outputs. If the individual
Another problem is how to move the sensor, sensor outputs are available, they can be cali-
horizontally or vertically, at a controlled dis- brated with only the vertical step motion.
tance. A simple instrument that has been used
for vertical motion is a mechanical balance
(Fig. 7). Placing the sensor on the mechanical Measuring the Complete Response
balance table, the seismometer can be moved at Curve A(v)
a controlled distance as measured on the balance
arm vertical scale. In principle, it is enough to The alternative to obtain the individual sensor
make one displacement, but usually several are parameters and then calculating A(o) is to obtain
made to take average measurements. A(o) directly. This is an empirical calibration
A simple portable calibration table which and can only be done by providing a known input
operates on this principle is commercially avail- to the sensor X(o) and measuring the output
able from Lennartz Electronic (www.lennartz- Y(o), from which the response function can be
electronic.de). it moves the vertical axis a obtained as
known displacement. It also allows to tilt the
sensor a known angle, which is sensed as an AðoÞ ¼ Y ðoÞ= XðoÞ (14)
apparent step in acceleration along the horizontal
axis, thus permitting the calibration of horizontal Shaking Table: Again a shaking table is the
components as well. simplest way. The shaking table can move
SIGNAL SEISMIC
Output Input 1
GENERATOR SENSOR
Input 2
Sensors, Calibration of, Fig. 8 Setup for measuring the instrument response using the calibration input (Figure from
Havskov and Alguacil 2010)
vertically and horizontally in a controlled fashion where Kc is the calibration coil motor constant
so the ground input is exactly known. All that has (N/A) and Rc is the resistance of the calibration
to be done is to measure the output, as a function coil. For the active sensor
of frequency, and divide by the known input to
get the response function. It is not as simple as it fi ¼ K a V 0 (16)
sounds. Making a precise shaking table, particu-
larly for horizontal motion, without introducing
tilt, is complicated and expensive, so shaking where Ka is the calibration input sensitivity
tables are rarely available for the general user (N/V). With a sensor mass of m, the equivalent
and only used in special laboratories. ground acceleration amplitude is then fi/m, the
Sensor Calibration Input: Many sensors have equivalent ground velocity is fi/mo, and the
a calibration input. For passive seismometers, this equivalent ground displacement amplitude is fi/
is a calibration coil around the mass, and applying mo2. If the amplitude of the seismometer voltage
a current to the coil results in a force on the mass. output is Vs, then the velocity amplitude response
Most active sensors do not have a calibration coil, (V/ms1) for this frequency is
but an equivalent test can be made when a voltage
is applied to the feedback loop to produce a force V s mo
AðoÞ ¼ (17)
on the mass. In both cases it is possible to accel- fi
erate the moving mass by applying a voltage to
the calibration input. By using a sine wave, the and similarly for displacement response. By
sensors experience the equivalent of a ground varying the frequency and measuring both input
acceleration, and, by varying the frequency, the and output signals, a complete response curve can S
complete frequency-dependent amplitude be obtained.
response can be determined, since the input When the calibration coil is used to calibrate
X(o) is known and Y(o) is measured. This a passive sensor, an undesirable effect may be
method assumes that the calibration circuit is present: a spurious coupling between the calibra-
accurate. An experimental setup is seen in Fig. 8. tion coil and the signal coil. This coupling affects
A signal generator sends out a sine wave with mainly the higher frequencies and may be
a constant amplitude V0. This voltage is applied approximately corrected for by subtracting the
to the sensor calibration input and exerts a force signal output when the same current is injected
on the mass proportional to V0. For the velocity with the sensor mass locked (Sauter and Dorman
sensor with a calibration coil, the amplitude of the 1986; Steck and Prothero 1989).
force fi will be From Fig. 8, it is seen that the output signal not
only has been changed in amplitude but also has
fi ¼ K c V 0 =Rc (15) been delayed a little relative to the input signal; in
other words, there has been a phase shift. In this vice versa. Thus, by measuring the signals of the
example, the phase shift is positive; see definition known and unknown sensor, it is in principle
in Eq. 18. The complete frequency response of simple to determine the response function of the
the sensor therefore consists of both the ampli- unknown sensor. In order for the method to work,
tude response function and the phase response the sensor self-noise must be well below the
function F(o). Considering a general input har- signal levels generated by the ground motion.
monic waveform x(o,t) = X(o) cos (ot) at fre- That will clearly limit the usefulness of
quency o, the output can be written as the method at lower frequencies, where, e.g.,
for a geophone, the output signal level will be
yðo, tÞ ¼ XðoÞ AðoÞ cos ðot þ FðoÞÞ (18) low. So, contrary to normal sensor installation,
these tests should be made in a high ground
noise environment like the top floor of a building,
The phase shift is here defined as a quantity although some care is required to avoid
being added to the phase as seen above. Thus
air-coupled acoustic noise, which may affect
comparing Fig. 8 and Eq. 18, it is seen that the
both sensors in different ways (Pavlis and Vernon
phase shift is positive. 1994).
The phase response function can therefore
If the spectra Z1(o) and Z2(o) of the two
also be measured using this harmonic drive
sensors outputs are contaminated with instru-
method. It can similarly be obtained using mental noise (a noncausal output), a simple
a shaking table.
ratio Z2/Z1 may yield a biased or unreliable esti-
Knowing the complete response function
mate of the response. By using only the corre-
A(o), it is now possible to get an indication if lated part of the signals, i.e., the part due to the
the response function can be represented by Eq. 2
common input of both sensors, the ground noise
or there are additional filters to include.
in this case, a more reliable estimate, will be
Using the Ground Motion as a Shaking Table:
obtained. This may be done by using a known
The experiment consists of setting up two sensors
relation between input and output of linear and
close to each other and assuming that they are
causal systems: the input-output cross-spectrum
subject to the same ground motion. If the known
is the product of the transfer function and the
and unknown sensors have recorded signals of
input power spectrum (see, e.g., Ljung 1999).
the ground motion z1 and z2, respectively, and
Consider a linear system whose input is the out-
corresponding spectra are Z1(o) and Z2(o), the
put of seismometer 1 (Z1) and its output is the
input ground motion (displacement, velocity or
output of seismometer 2 (Z2). This system would
acceleration) is
have a transfer function A2/A1 = P21(o)/P11(o),
or Z2/Z1 if these two spectra were noise-free and
Z 1 ðo Þ
X1 ðoÞ ¼ (19) generated by the same input. Therefore, a more
A1 ðoÞ robust estimation of the unknown response A2 is
obtained using
where A1 is the response function of the known
sensor 1. The unknown response function A2 is P21 ðoÞ
then A2 ð o Þ ¼ A 1 ð o Þ (21)
P11 ðoÞ
Z2 ðoÞ where P11(o) represents the power spectrum of

A2 ðoÞ ¼ A1 ðoÞ (20)
Z1 ðoÞ the output of sensor 1 and P21(o) is the cross-
spectrum between the outputs of sensors 1 and
Note that if A1 is in displacement, then A2 will 2. The advantage of this equation is that, ideally,
also be in displacement and similarly if A1 is in it cancels out the contributions of uncorrelated
velocity or acceleration. It is, e.g., possible to use components (i.e., self-noise) of the input and
an accelerometer to calibrate a seismometer or output signals. This method has been further
Amplitude spectrum
Calibration pulse
Sensors, Calibration of, Fig. 9 Calibration pulse from spectrum of the calibration pulse. The sensor natural fre-
a velocity sensor. Left: The recorded calibration pulse due quency o0 is indicated. The damping is 0.707
to an input step in acceleration. Right: The log-log (Figure from Havskov and Alguacil 2010)
improved using three channels of data, resulting 1/√2 = 0.707 or the frequency for which the
in more robust estimate of the relative response low-frequency and high-frequency asymptotes
function (Sleeman et al. 2006). However, small intersect. It is then in principle simple to get the
misalignments between the three sensors may frequency response for the sensor by just doing
prevent the background noise to be fully canceled spectral analysis of the calibration pulse; how-
out, and a leak of such noise is then evaluated as ever, ground and electronic noise might make it
sensor self-noise. Gerner and Bokelmann (2013) inaccurate. In practice the signal-to-noise ratio is
suggest a technique to numerically fix this effect improved by summing several calibration pulses
by searching for the optimal alignment. with their onsets well aligned. The background
Calibration Pulse: It is common practice to noise is decreased in this way, while the signal
generate a calibration pulse by applying a step due to the calibration pulses is increased.
current into the calibration coil of a passive sen-
sor. For a BB sensor an equivalent test can be
made when a voltage is applied to the feedback Summary
loop to produce a force on the mass. As the
applied force is proportional to the coil current Calibration of seismic sensor can be done in
or the voltage, this input signal is equivalent to many ways. The calibration can determine the
a ground step in acceleration (see also section frequency response of a sensor by determining
above). The output pulse can be recorded the sensor parameters natural frequency,
(Fig. 9), and this pulse can be used in several damping, and generator constant from which the
ways to obtain sensor parameters. frequency response can be calculated. The alter-
From signal theory (e.g., Scherbaum 2007) it native is to experimentally determine the com-
is known that the Fourier transform of the impulse plete frequency response of the sensor using, e.g.,
response of a linear system is the frequency a shaking table or the sensor calibration input. In S
response function. However, what is generated most cases it will be possible for the general user
here is a step. If the amplitude spectrum of the to get a good idea about the response function
calibration pulse is C(o), the response function using one of the methods described in this paper.
will then be A(o) = oC(o). This corresponds to
the response function for acceleration since the
input was an acceleration step so the velocity Cross-References
response for the velocity sensor is then o2C(o).
Thus, multiplying the spectrum in Fig. 9 with o2 ▶ Passive Seismometers
will give the velocity response for h = 0.707 seen ▶ Principles of Broadband Seismometry
in Fig. 2. It is also possible to obtain the approx- ▶ Recording Seismic Signals
imate natural frequency o0 as the frequency ▶ Seismic Accelerometers
where the spectral amplitude (Fig. 9) has ▶ Seismometer Self-Noise and Measuring
decreased from the flat level by a factor Methods
3266 Site Response for Seismic Hazard Assessment
References deep wave propagation from the fault to the bed-

rock, and the local soil conditions. The first two
Bormann P (ed) (2002) IASPEI New manual of seismo- phenomena are commonly studied by seismolo-
logical observatory practice (NMSOP). GeoForch-
gists, geologists, and geophysicists, while the
ungsZentrum Potsdam, Potsdam
Gerner A, Bokelmann G (2013) Instrument self-noise and third one falls in the geotechnical earthquake
sensor misalignment. Adv Geosci 36:17–20. engineering field, being strongly related to the
doi:10.5194/adgeo-36-17-2013 mechanical behavior of soils subjected to
dynamic loading. The term “site effects” refers
quake seismology. Springer, Dordrecht, p 358
Lay T, Wallace TC (1995) Modern global seismology. to the overall set of modifications of the bedrock
Academic, San Diego, p 521. ISBN 0-12-732870-X motion, in terms of amplitude, frequency content,
Ljung L (1999) System identification: theory for the user, and duration, during its (almost) vertical propa-
2nd edn. Prentice Hall PTR, Englewood Cliffs, p 609
gation through shallow geological deposits.
Pavlis GL, Vernon FL (1994) Calibration of seismometers
using ground noise. Bull Seismol Soc Am 84:1243–1255 Despite the traveling path of the propagating
Sauter AW, Dorman LM (1986) Instrument calibration of waves within the surface soils is often less than
ocean bottom seismographs. Mar Geophys Res 100 m, local site conditions can produce signifi-
8:265–275
cant ground motion amplification, slope instabil-
Scherbaum F (2007) Of poles and zeros, fundamentals of
digital seismology, revised 2nd edn. Springer, Dor- ity, excessive building settlements (i.e.,
drecht, p 271 subsidence), and liquefaction in loose and satu-
Sleeman R, van Wettum A, Trampert J (2006) Three- rated granular deposits, thus playing a crucial role
channel correlation analysis: a new technique to mea-
in building and infrastructure damage (Fig. 1).
sure instrumental noise of digitizers and seismic sen-
sors. Bull Seismol Soc Am 96(1):258–271 Ground response numerical analyses are, there-
Steck L, Prothero WA (1989) Seismic calibration using the fore, used by geotechnical engineers to predict the
simplex algorithm. Bull Seismol Soc Am 79:1618–1628 free-field motion, to determine permanent soil
deformations (leading to subsidence and liquefac-
tion), and to evaluate the risk of instability of earth
structures (e.g., earthquake-induced landslides).
Site Response for Seismic Hazard They may also include the evaluation of basin
Assessment and topographic effects on ground motion (e.g.,
King and Tucker 1984; Bard and Bouchon 1985;
Gaetano Elia Geli et al. 1988; Bard 1994; Ashford et al. 1997;
School of Civil Engineering and Geosciences, Bouckovalas and Papadimitriou 2005; Semblat
Newcastle University, Newcastle Upon et al. 2005; Pagliaroli et al. 2011). In addition,
Tyne, UK the results of these studies can be incorporated
into microzonation and probabilistic seismic
hazard analyses (e.g., Tsai 2000; Bazzurro and
Introduction Cornell 2004; Papaspiliou et al. 2012).
Site response analyses have been traditionally
Lessons learned worldwide from historical (e.g., performed using a one-dimensional (1D)
Niigata, Japan, and Alaska in 1964) and recent frequency-domain numerical scheme based on
strong earthquakes (e.g., L’Aquila in 2009, Chile the equivalent viscoelastic approach. This
in 2010, Tohoku, Japan, and Christchurch in approach has been extensively adopted in the
2011, among others) have distinguished site last 30 years, and it is widely accepted in the
amplification and soil liquefaction as two of the engineering practice, although its limitations are
main causes of damage to man-made and natural well known. Being based on a total stress formu-
structures during seismic events. As illustrated in lation, it disregards the buildup of excess pore
Fig. 1, the ground shaking observed at surface water pressures in the soil deposit. Additionally,
during an earthquake depends on the seismic the adopted equivalent viscoelastic material
source characteristics and focal mechanism, the properties cannot properly represent the soil
Site Response for Seismic Hazard Assessment 3267
Seismograph Local site effects Subsidence-Liquefaction -

a Landslides
a a
t t t
a
t
Bedrock earthquake
Fault
Seismic source Deep wave propagation
Site Response for Seismic Hazard Assessment, Fig. 1 Wave propagation from seismic source to ground surface
and related geotechnical problems (Modified from Lanzo and Silvestri 1999)
behavior under cyclic loading for high seismic assessment of site response effects. Ground
intensities at bedrock. Finally, the 1D scheme response analyses are then discussed in the sec-
cannot take into account site effects related to tion “Ground Response Analysis,” describing lin-
surface and buried complex morphologies, i.e., ear and nonlinear approaches for the study of
topographic and valley effects. Time-domain one-dimensional wave propagation problems in
schemes are nowadays available to solve the free-field conditions. Soil–structure interaction
wave propagation problem in a more realistic and earthquake-induced ground failure problems,
way, accounting for the solid–fluid interaction such as soil liquefaction, landslides, and retaining
by means of a coupled effective stress formula- structure instability, are outside the scope of the
tion. In those schemes, the behavior of the soil entry and, therefore, are not discussed here.
can be described using either simple or sophisti- A review is given in Kramer (1996), Kramer
cated nonlinear soil constitutive models of differ- and Stewart (2004), and Semblat and Pecker
ent level of complexity. In addition, time-domain (2009). Finally, the section “Future Challenges”
analyses, usually performed with finite element provides an overview of future challenges in the S
codes, can also describe two- (2D) and three- field of geotechnical earthquake engineering.
dimensional (3D) complex geometries to model
topographic and basin effects. Nevertheless,
these nonlinear analyses are seldom adopted by Soil Dynamic Properties and
nonexpert users because the calibration of Measurement Techniques
advanced soil constitutive models can be chal-
lenging and the code usage protocols are often Soil Dynamic Properties
unclear or poorly documented in the literature. As schematically indicated by Fig. 1, the propa-
This entry begins with a short overview of gation of waves from bedrock to ground surface
available methods for site characterization and is commonly considered as a vertical process.
evaluation of soil dynamic properties (section According to Snell’s law (cf. Richart et al.
“Soil Dynamic Properties and Measurement 1970), in fact, waves traveling from higher-
Techniques”), factors that are essential for the velocity materials into lower-velocity materials
Site Response for

Seismic Hazard
Assessment,
Fig. 2 Mechanical
behavior of a soil element
subjected to an irregular
simple shear loading
history (Modified from
Lanzo and Silvestri 1999)
are refracted closer to the normal to the inter- characterized by state dependency, early irrevers-
faces. Therefore, earthquake waves propagating ibility, nonlinearity, buildup of excess pore pres-
upward through horizontal layers characterized sures, decrease of nominal stiffness, and related
by lower velocities and densities (such as in typ- hysteretic dissipation (e.g., Sangrey et al. 1969;
ical top soil deposits) are refracted closer to a Hardin and Drnevich 1972; Castro and Christian
vertical path. Moreover, saturated soils subjected 1976; Vucetic and Dobry 1991). Under a sym-
to earthquake loading behave essentially in metric cyclic loading condition, the hysteresis
undrained conditions, given the rapidity of the loop in the t g plane can be effectively
seismic action. The volumetric deformations described by two parameters: the shear modulus
induced by P-waves are, consequently, negligible and the damping ratio. As shown in Fig. 3, the
with respect to the distortional deformations tangent shear modulus Gtan represents the soil
associated with S-waves. For the above reasons, stiffness for a specific point of the loop, and it
many of the methods of ground response analysis changes continuously throughout the cycle.
presented in the section “Ground Response Anal- An overall indication of the average soil stiff-
ysis” simulate the seismic event as a loading ness is, instead, represented by the secant shear
process induced by SH-waves only, characterized modulus Gsec , defined as:
by a vertical traveling path associated with soil
particle motion in the horizontal plane. This phe- tc
Gsec ¼ (1)
nomenon requires the analysis of the mechanical gc
behavior of soils under simple cyclic shear load-
ing conditions. Considering a generic soil ele- where tc and gc are the maximum shear stress and
ment within the deposit in geostatic conditions shear strain amplitudes, respectively. The area
(when only the vertical, s0v0 , and the horizontal, included in the hysteresis loop is a measure of
s0h0, effective stresses are applied), the earthquake the energy dissipated by the soil during the cycle
action induces an additional simple shear stress and is described by the damping ratio D:
tðtÞ changing irregularly with time (Fig. 2).
The corresponding stress–strain (t g) curve WD Aloop
D¼ ¼ (2)
under this cyclic loading history is typically 4pWS 2pGsec g2c
overconsolidated reconstituted clays presented

by Vucetic and Dobry (1991) indicated that
(Fig. 4):
(a) The values of OCR and effective consolida-

tion stress have almost no effect on the posi-
tion and shape of G=G0 and D curves.
(b) The plasticity index PI is the key factor con-
trolling the dimensionless parameters G=G0
and D.
(c) The number of cycles N does affect both the
value of G=G0 (due to the degradation of the
shear modulus with N) and the D gc curve at
high cyclic strains (i.e., after the volumetric
threshold).
It should be noted that the modulus reduction

and damping curves reported in Fig. 4 for
PI = 0 are nearly identical to the average curves
Site Response for Seismic Hazard Assessment,
Fig. 3 Evaluation of shear moduli and damping ratio commonly used for sands (Seed and Idriss 1970).
from hysteresis loop (Modified from Kramer 1996) This suggests that the curves shown in Fig. 4 can
be used for both fine- and coarse-grained soils.
Modulus reduction and damping ratio curves are
where WD is the dissipated energy (equal to the influenced by the mean effective stress for
area of the hysteresis loop Aloop) and WS is cohesionless and low-plasticity soils, but this
the maximum strain energy represented by the influence decreases with increasing plasticity
shaded area in Fig. 3. index, being generally not significant for
Staring from a maximum value, i.e., G0 or PI 30 (Ishibashi 1992).
Gmax , the soil stiffness tends to decrease with
increasing shear strains. Its variation with cyclic Measurement Techniques
shear strain amplitude (gc ) is described graphi- The measurement of soil dynamic properties is a
cally by a normalized modulus reduction curve crucial task in the solution of any geotechnical
G=G0 gc . Conversely, the energy dissipation earthquake engineering problem. Particularly for
provided by the soil, increasing with loop ampli- the evaluation of local site effects, the character-
tude due to plasticity effects, is typically ization of the soil deposit above bedrock in terms S
depicted by a D gc curve. Therefore, the of variation of G0 and D with depth and
mechanical characterization of the soil stiffness G=G0 gc and D gc curves is essential. A vari-
and damping requires both the evaluation ety of field and laboratory techniques are avail-
of G and D at very low strains and the way able, each oriented toward the measurement of
in which the two properties change with low-strain properties and characterized by differ-
cyclic shear strain amplitude. Laboratory tests ent advantages and limitations. A complete
have shown that soil stiffness and damping review of the existing techniques is outside the
are also influenced by other factors, such as scope of this entry, and only the most significant
mean effective confining pressure, plasticity ones are discussed in the following.
index (PI), overconsolidation ratio (OCR), and Low-strain field tests induce seismic waves in
number of imposed cycles (N). In particular, the soil and measure the velocities at which these
extensive laboratory investigations on the cyclic waves propagate. The maximum shear modulus
response of normally consolidated and slightly can be computed using the measured shear wave
Site Response for a 1.0

Seismic Hazard
Assessment,
0.8
Fig. 4 Nonlinear behavior
as function of plasticity
index in terms of: (a) 0.6
PI = 200
Gmax
normalized shear modulus
G
100
and (b) damping ratio 0.4 50
30
(Vucetic and Dobry 1991) 15
OCR = 1-15 0
0.2
0.0
0.0001 0.001 0.01 0.1 1 10
Cyclic shear strain, γc(%)
b
25 PI = 0
15
20
Damping ratio (%)
30
OCR = 1-8
15 50
100
10
200
0
0.0001 0.001 0.01 0.1 1 10
Cyclic shear strain, γc(%)
velocity (VS) and the soil density (r) through the test, and the cross-hole technique. In the CSW
following equation based on the elasticity theory: test (Fig. 5a), a mechanical, servo-hydraulic, or
electromagnetic vibrator applies a single-
Gmax ¼ rV2S (3) frequency sinusoidal force to the ground surface.
Rayleigh waves traveling through the soil are
Shear wave velocities can be measured in situ by detected by a series of geophones (usually two)
several seismic tests, including cross-hole and displaced at a range of distances from the source.
down-hole, seismic cone penetrometer, suspen- By changing the input frequency, a profile of
sion logger, SASW (spectral analysis of surface phase velocity against wavelength is obtained,
waves), and MASW (multichannel analysis of and, consequently, a stiffness profile with depth
surface waves). A review of these test methods can be computed. Although less economical than
is given in Woods (1994) and Kramer (1996). an SASW test, the CSW method has been proved
Their accuracy can be sensitive to procedural to provide better data, as background noise can be
details, soil conditions, and interpretation tech- easily recognized and filtered. Likewise the CSW
niques. Fig. 5 shows the layouts and principles of test, the MASW approach uses a multiple of
three established geophysical tests: the continu- equally spaced receivers (usually 12 to 60) that
ous surface wave (CSW) method, the down-hole are deployed on the surface along a survey line.
Each receiver is connected to a common mobilized at large strains, can be indirectly used
multichannel recording instrument (i.e., a seis- to determine shear modulus profiles in situ by
mograph). This is the most significant difference using empirical correlations between penetration
between the CSW and the MASW techniques, as resistance (NSPT) or tip resistance (qc) values and
CSW is usually based on a two-receiver Gmax (a review is given in Kramer 1996). These
approach. Also the MASW method generally estimates are affected by high uncertainties and
uses an impulsive source, such as a sledgeham- should be used very cautiously, given the scatter
mer, to produce surface waves, whereas the CSW in the data on which they are based and the
technique makes use of a frequency-controlled variability in the results obtained by different
vibrator. correlations. Such correlations should be adopted
Down-hole and cross-hole tests are alternative only for preliminary estimates of Gmax in the
low-strain techniques which require one or more framework of simplified approaches.
borings. In a down-hole test, a vibration source is Finally, standard spectral ratio (SSR) and
placed on the ground surface adjacent to a bore- horizontal-to-vertical spectral ratio (HVSR)
hole. The arrival of seismic energy is detected at methods for the determination of the deposit fun-
depth either by geophones secured against the damental frequencies are becoming increasingly
borehole sides or by geophones within a seismic popular not only in the research field but also in
CPT (Fig. 5b). The test is repeated changing the professional practice (i.e., for microzonation
depth of the geophones (typically at 1 m inter- studies). Site amplification factors can, in fact,
vals) to plot the shear wave travel time as func- be inferred, at least in the linear strain range,
tion of depth. The average shear velocity can be using the SSR technique described by Field and
computed by knowing the distance between the Jacob (1993). HVSR amplifications obtained
source and the receiver. The cross-hole test experimentally can, instead, be used to validate
makes use of more than one boring (usually numerical model results (e.g., SESAME 2004;
two, less commonly three as this latter option is Lanzo et al. 2011).
more expensive): a source is placed in one boring In general, field tests have the advantage to
and a receiver is placed at the same depth in each describe the dynamic properties of the soil as it is
of the other boreholes (Fig. 5c). An impulsive in situ. Laboratory testing methods are, on the
disturbance is applied at the source and the travel contrary, usually performed on relatively small
time to each receiver is measured. A borehole specimens that can be disturbed by the sampling
verticality survey is required in order to calculate technique and may not be representative of the
the actual distance between the boreholes at each larger body of soil from which they are retrieved.
test depth (usually 1 m intervals). The wave prop- Nevertheless, cyclic and dynamic laboratory tests
agation velocity is, in fact, computed by knowing are complementary to field methods as they can
the distances between receivers. The use of two provide the description of soil nonlinearity over a S
sets of receivers (three boreholes) avoids the wide range of shear strains in terms of G=G0 gc
issue of trigger accuracy, but increases the cost and D gc curves. Typical low-strain element
of the test. While CSW and down-hole tests allow techniques are the resonant column (RC) and the
to determine the shear modulus for distortion in bender element (BE) tests. A schematic view of
vertical plane (Gv) as the source produces a ver- the resonant column apparatus is presented in
tically polarized horizontally traveling shear Fig. 6: a solid or hollow cylindrical specimen is
wave, the cross-hole technique allows to calcu- subjected to harmonic torsional or axial loading
late also the shear modulus in the horizontal plane by an electromagnetic loading system. The sys-
(Gh) by using horizontally polarized shear wave tem usually applies a harmonic load for which the
sources. frequency and amplitude can be controlled. The
The standard penetration test (SPT) and the fundamental frequency of the specimen can be
cone penetration test (CPT), although originally identified by gradually increasing the loading
developed for the measurement of soil properties frequency. Given the mass polar moment of
Site Response for

Seismic Hazard
Assessment, Fig. 5 Field
seismic tests: (a)
continuous surface wave
(CSW), (b) down-hole, and
(c) cross-hole (Clayton
2011)
inertia of the loading head and the specimen mass Damping can be determined using the half-power
and dimensions, the shear modulus of the soil can bandwidth method or from the logarithmic dec-
be calculated, assuming linear elasticity (i.e., rement by placing the sample in free vibration
using Eq. 3). Repeating the test with increasing (ASTM 2007). This technique has been modified
loading amplitude, the variation of secant shear to allow cyclic torsional shear testing to strain
modulus with shear strain can also be measured. levels above those typically achieved during a
Site Response for Seismic Hazard Assessment, Fig. 6 Schematic drawing of a typical resonant column apparatus:
(a) top view and (b) profile view (Kramer 1996)
conventional resonant column test. It is well the soil sample in which the device is embedded.
established that the large strain rates related to At the same time, a lateral disturbance of a bender
the high frequencies applied in an RC test can element produces a voltage, thus allowing the
affect the measured small-strain shear modulus detection of incoming waves. Therefore, BE can
and damping ratio. Dry cohesionless soils usually be used as both transmitters and receivers of
do not exhibit rate effects, while high strain rates S-waves within a soil sample (typically triaxial).
can produce an increase of shear stiffness at small By measuring the time required for the wave to
strains in fine-grained plastic soils leading to a G0 travel from the source to the receiver and knowing
overprediction with respect to cyclic simple shear the distance between each, the shear wave velocity
tests (e.g., Lo Presti et al. 1997; Cavallaro of the specimen can be measured.
et al. 2003). Moreover, the laboratory results Finally, laboratory tests able to measure the
indicate that some energy is always dissipated dynamic soil properties at high strain levels have
by the soil, even at very low strains, so that the been derived from conventional tests by adding
damping ratio is never equal to zero. This mech- cyclic loading capabilities to the testing appara-
anism cannot be justified by plasticity theory, as tus. Examples are represented by cyclic triaxial
no hysteretic dissipation of energy associated to tests with local strain measurements and cyclic
the development of plastic deformations takes simple shear tests. S
place at strain levels within the elastic domain. Table 1 summarizes the range and applicabil-
Therefore, the initial damping ratio D0 observed ity of the most common field and laboratory
in resonant column experiments can be probably cyclic and dynamic techniques, with particular
attributed to material viscous effects and/or the reference to the induced shear strain level, char-
inertia of the resonant column apparatus (e.g., acteristic frequency, and dynamic soil properties
Meng and Rix 2003; Lo Presti et al. 2007). which can be obtained from the tests.
Another laboratory technique that allows the
measurement of the shear wave velocity of a soil
sample is the bender element testing method. Ground Response Analysis
Bender elements consist of two piezo-ceramic
plates bonded together in such a way that applica- Analytical solutions have been developed to solve
tion of a voltage causes one plate to expand and the one-dimensional wave propagation problems,
other to contract. This generates seismic waves in assuming that the input motion is harmonic and
Site Response for Seismic Hazard Assessment, Table 1 Range and applicability of field and laboratory dynamic
tests (Modified from Vinale et al. 1996)
Shear strain Frequency Shear
Test type g (%) f (Hz) modulus Damping ratio
Field Standard SPT – – NSPT ! VS ! –
G0
CPT qc ! VS ! G0
Geophysics Down-hole <103 10 100 VS ! G 0 Possible
Cross-hole VS ! G 0
SASW VR ! VS !
G0
Lab Cyclic Triaxial >102 0.01 1 se!E! Hysteresis loop ! D
G
Simple shear >102 0.01 1 tg!G
Torsional 104 1 0.01 1 t g ! G, G0
shear
Dynamic Bender <103 >100 VS ! G 0 Possible
elements
Resonant 105 101 >10 fr ! G, G0 H-p or log.
column decrement ! D
Site Response for Seismic Hazard Assessment, ratio, and (d) maximum acceleration mobilized within the
Fig. 7 Profiles with increasing bedrock acceleration of soil deposit (Lanzo and Silvestri 1999)
(a) maximum shear strain, (b) shear modulus, (c) damping
the soil is a homogeneous elastic or viscoelastic nonlinear (i.e., G and D are functions of the shear
material (e.g., Roesset 1977). They make use of strain induced by the earthquake), the amplifica-
amplification functions to describe the dynamic tion function is no more constant (as in the analyt-
response of a soil deposit (considered as a single- ical solutions) but site specific. The effects of soil
degree-of-freedom system) in the frequency nonlinearity on wave propagation are qualitatively
domain. This approach is limited to the analysis illustrated in Fig. 7, where a homogeneous deposit
of linear systems, as it relies on the principle of overlying a horizontal bedrock is subjected to an
superposition. When the soil deposit is layered, input motion characterized by increasing values of
heterogeneous (i.e., G0 changes with depth), and the maximum acceleration (amax,r).
Site Response for

Seismic Hazard
Assessment, Fig. 8 Peak
acceleration on soft soil
sites as function of bedrock
maximum acceleration
(Idriss 1990)
Increasing the peak ground acceleration at be higher than on rock. At higher acceleration
bedrock: levels, on the contrary, nonlinearity can prevent
from the development of peak accelerations at
• The mobilized shear strain increases (Fig. 7a). surface as large as those observed on rock sites.
• The shear modulus reduces and the damping Local site conditions not only influence the
ratio increases correspondingly (Fig. 7b, c). peak acceleration amplitudes but can also
• For low input energy, the peak acceleration strongly affect the frequency content of surface
along the soil profile increases moving from motions. Fig. 9 shows the effects of local soil
the bedrock to the ground surface (profiles conditions on the shape of the normalized
1 and 2 in Fig. 7d). response spectra computed from ground motions
• For high input energy, the peak acceleration recorded on different sites: for periods above
along the profile can reduce because the 0.5 s, spectral accelerations for soil sites are
nonlinear behavior reduces the soil shear stiff- higher than those for rock sites. The figure
ness, thus preventing from the transmission of clearly indicates that deep and soft soil deposits
high frequencies, while the corresponding enhance the transmission of low frequencies
increase of damping ratio reduces the dis- (high periods). The results also show that the
placement and acceleration amplitude peaks use of a single response spectrum shape for all
(profile 3 in Fig. 7d). site conditions is not appropriate. This evidence
has been incorporated in a large number of seis- S
In terms of amplification factor (given as the mic codes worldwide which propose the use of
ratio between the max acceleration at surface to different spectral shapes for different subsoil
the max acceleration at bedrock), nonlinearity conditions.
effects can lead to a reduction of the amplification
at surface with increasing acceleration levels at Equivalent Linear Approach
bedrock. On the basis of data recorded during Nonlinear soil behavior can be approximated by
Mexico City (1985) and San Francisco Bay an equivalent linear characterization of soil
area (1989) earthquakes, and of additional dynamic properties. The method makes use of
ground response analyses, Idriss (1990) related the exact continuum solution of wave propaga-
peak accelerations on soft soil deposits to tion in horizontally layered viscoelastic materials
those on rock sites, as reported in Fig. 8. For subjected to vertically propagating transient
acceleration levels lower than about 0.4 g, the motions (e.g., Roesset 1977). It models the
peak acceleration at soft soil surface is likely to nonlinear variation of soil shear modulus and
Site Response for

Seismic Hazard
Assessment,
Fig. 9 Normalized
response spectra (5 %
damping) for different local
site conditions (Seed
et al. 1976)
ai
ai
f IFFT t
u1
1 =
G0
1......k, k+1 Hi integrate in t
G(γ)
hi,ρi,Gi,Di ui
iterations i
ui
1......k, k+1
D0 D(γ) un f t
n
X
0.0001 0.001 0.01 0.1 1 ur
γ (%) ρr Vr ar derive in z
ar γi
FFT γ max
t f t
γ = β γ max
NO
|γk+1-γk| < ε
YES
OUTPUT
Site Response for Seismic Hazard Assessment, Fig. 10 Equivalent linear algorithm (Modified from Lanzo and
Silvestri 1999)
damping with shear strain through a sequence of The iterative procedure, illustrated in Fig. 10,
linear analyses with iterative update of stiffness operates as follows:
and damping parameters. For a given soil layer,
G and D are assumed to be constant with time 1. Initial estimates of G and D are made for each
during the shaking. Therefore, an iterative proce- layer at low-strain values.
dure is needed to ensure that the properties used 2. These initial values are used in a linear
in the linear dynamic analyses are consistent with frequency-domain dynamic analysis where
the level of strain induced in each layer by the the acceleration time history at bedrock is
input motion. transformed through a fast Fourier transform
(FFT) algorithm and the transfer functions capable of representing changes in soil stiffness
between the bedrock and each layer (Hi in and hysteretic damping during the seismic action.
Fig. 10) are computed. For problems where strain levels remain low
3. The inverse fast Fourier transform (IFFT) (stiff soil profile and/or relatively weak input
algorithm is employed to obtain acceleration, motions), the equivalent linear method can pro-
displacement, and shear strain time histories at duce reasonable estimates of ground response.
each layer. For high seismic intensities at bedrock, nonlinear
4. The effective shear strain (g) in each layer is time-domain analyses should be preferred as
determined from the maximum shear strain they are likely to provide better results (section
(gmax) as follows: “Verification and Parametric Studies”).
g ¼ bgmax (4) Nonlinear Approach

In recent years, the use of fully coupled effective
where b ¼ ðM 1Þ=10 is usually related stress formulations, based on Biot’s theory for
to the expected earthquake magnitude M. solid–fluid interaction (Biot 1941; Zienkiewicz
5. From this effective shear strain, new equiva- et al. 1999), is becoming increasingly popular as
lent linear values of G and D are chosen for the the theoretical basis for the dynamic analysis of
next iteration. geotechnical structures (e.g., Dewoolkar
6. Steps 2–5 are repeated until a convergence et al. 2001; Elgamal et al. 2002a; Aydingun and
criterion is satisfied. Adalier 2003; Dakoulas and Gazetas 2005;
Liu and Song 2005; Madabhushi and Zeng
The most widely used computer software for 2007; Sica et al. 2008; Alyami et al. 2009;
one-dimensional ground response analysis, based Shahrour et al. 2010; Elia et al. 2011; Kontoe
on the equivalent linear approach described above, et al. 2011; Elia and Rouainia 2013, 2014). Spe-
is SHAKE (Schnabel et al. 1972) and its modified cifically for ground response analysis, nonlinear
version SHAKE91 (Idriss and Sun 1992). For approaches are able to solve the wave propagation
two-dimensional geometries, the equivalent linear problem by direct numerical integration in the
approach has been implemented in a number of time-domain, accounting for the development of
codes such as FLUSH (Lysmer et al. 1975) and plastic deformations and buildup of excess pore
QUAD4M (Hudson et al. 1994), among others. water pressures within the soil deposit induced by
Equivalent linear methods are extensively adopted the earthquake. In those schemes, the mechanical
in engineering practice for their simplicity, flexi- behavior of the soil can be described using either
bility, and low computational requirements. Nev- simple or sophisticated nonlinear constitutive
ertheless, their limitations are well known. The models of different level of complexity. In terms
analysis is performed adopting a total stress of spatial discretization, most numerical S
approach. This means that excess pore water pres- approaches are modelling the problem in 1D or
sures induced by the earthquake cannot be 2D plane strain conditions. Nonlinear three-
predicted and the displacements due to consolida- dimensional (3D) ground response analyses are
tion processes cannot be calculated. The model rarely found in the literature, being only performed
employed to describe the mechanical behavior of for critical structures, such as tunnels, bridges, and
soils is viscoelastic. Therefore, the method cannot earth dams (e.g., Elgamal 1992; Stamos and
predict permanent soil displacements or cumu- Beskos 1995; Elgamal et al. 2008; Ou 2009;
lated strains at the end of the analysis. Even though Hatzigeorgiou and Beskos 2010).
the iteration process allows to approximate A summary of the available computer soft-
nonlinear soil behavior, the approach is still a ware, employing equivalent linear and nonlinear
linear method of analysis. The strain-compatible approaches to solve the wave propagation prob-
soil properties are constant throughout the dura- lem in one- and two-dimensional conditions, is
tion of the earthquake. The method is thus not reported in Table 2.
Site Response for Geometry Software Type of analysis

Seismic Hazard
1D SHAKE (Schnabel et al. 1972) TS EL
Assessment,
Table 2 Ground response SHAKE91 (Idriss and Sun 1992)
analysis software PROSHAKE (EduPro Civil Systems 1998)
(Modified from AGI 2005) SHAKE2000 (www.shake2000.com)
EERA (Bardet et al. 2000)
TESS (Pyke 1992) TS NL
NERA (Bardet and Tobita 2001)
DEEPSOIL (Hashash 2009)
DESRA_2 (Lee and Finn 1978) ES
DESRAMOD (Vucetic 1986)
D-MOD_2 (Matasovic 1995)
D-MOD2000 (GeoMotions 2007)
SUMDES (Li et al. 1992)
CYCLIC 1D (www.soilquake.net)
2D QUAD4 (Idriss et al. 1973) TS EL
QUAD4M (Hudson et al. 1994)
FLUSH (Lysmer et al. 1975)
QUAKE/W (GeoSlope 2002)
SPECTRA (Borja and Wu 1994) TS NL
DYNAFLOW (Prevost 2002) ES NL
GEFDYN (Aubry and Modaressi 1996)
TARA-3 (Finn et al. 1986)
FLAC 2D (Itasca 2002)
PLAXIS 2D (www.plaxis.nl)
SWANDYNE II (Chan 1995)
OpenSees (McKenna and Fenves 2001)
TS total stress, ES effective stress, EL equivalent linear, NL nonlinear
Verification and Parametric Studies Seismic Test (LSST) site in Taiwan. The site was
The effectiveness of the different numerical established in 1985 to study the dynamic behavior
approaches, briefly described in the previous sec- of two scaled-down nuclear plant containment
tions, to simulate the complex wave propagation structures constructed by the Electric Power
process has been tested over the last decades Research Institute (EPRI) and the Taiwan Power
using real vertical array data and/or centrifuge Company.
test results. In these cases, in fact, the input Borja and coworkers studied the ground
motion is reasonably well defined, as it is directly response at Lotung employing both a standard
recorded on rock or it is imposed during the test. linear equivalent method (SHAKE) and an
In addition, validation studies have been advanced nonlinear approach, where a bounding
conducted in several international benchmark surface soil constitutive model (formulated in
projects, such as ESG-IASPEI/IAEE 1992 terms of total stresses) was implemented into the
(Turkey Flat and Ashigara Valley arrays), three-dimensional finite element (FE) code
ESG-IASPEI/IAEE 2006 (Grenoble basin), Tur- SPECTRA. Fig. 11a, b compares the recorded
key Flat 2008, and E2VP 2010 (EuroSeisTest, ground surface accelerations along the EW and
Greece). NS directions (i.e., FA1-5/EW and FA1-5/NS)
As an example, Borja et al. (1999) presented a with the corresponding predictions obtained
verification study using the recordings from accel- through SHAKE. The peak values of both EW
erometer arrays installed at the Lotung Large-Scale and NS accelerations recorded at ground surface
Site Response for

Seismic Hazard
Assessment,
ground surface
accelerations recorded at
Lotung (EW and NS
components) and
predictions by SHAKE and
SPECTRA (Borja
et al. 1999)
were slightly underpredicted. Moreover, a time directions. An excellent match also in terms of
shift in the acceleration peak between the zero crossings was obtained using the advanced
recorded and computed NS motions can be nonlinear time-domain approach. The advantage
observed. On the contrary, Fig. 11c, d shows that of the described approach lies in its capacity to
peak accelerations were predicted by the extend the analysis to three dimensions, thus
nonlinear model within 10–20 % error in both allowing the simultaneous application of the
Site Response for

Experimental
Seismic Hazard 0.2
Computed
Assessment,
Fig. 12 VELACS model
0
1 recorded and computed
acceleration time histories
(Elgamal et al. 2002b) −0.2
AH3 (surface)
0.2
0
Lateral Acceleration (g)
−0.2
AH4 (2.5m)
0.2
−0.2
AH5 (5.0m)
0.2
−0.2
AH1 (Input)
0 2 4 6 8 10 12 14 16
Time (sec)
three components of the earthquake (one vertical sand and subjected to a 2Hz harmonic base exci-
and two horizontal) at the base of the FE column. tation). Good agreement was achieved between
Another example of verification study, this the computed and recorded responses at different
time using centrifuge testing data, is provided depths within the soil deposit in terms of accel-
by Elgamal et al. (2002b). The authors adopted erations, although the numerically predicted set-
an advanced plasticity-based constitutive model tlement due to liquefaction was generally smaller
to simulate the cyclic mobility response mecha- than observations. In addition, the work by
nism typically observed in medium-dense sands Elgamal and coworkers demonstrates how,
subjected to cyclic loading. The model was under level ground conditions, shear strains can
implemented into an unnamed fully coupled be relatively small with minor cyclic mobility
finite element code to predict the dynamic effects, while mildly sloping ground may result
response observed during different centrifuge in large cyclic shear strain accumulation leading
tests (VELACS project, Arulanandan and Scott to unbounded flow failure. The importance of the
1993). In particular, Fig. 12 shows the compari- dominant excitation frequency on post-
son between the computed acceleration time his- liquefaction soil response is also highlighted.
tories and the ones recorded for VELACS model The two examples described above are repre-
1 (representing a level site composed of Nevada sentative of the state of the art in the field of
nonlinear ground response analysis, but other larger than the ratio fp/f1 between the predom-
verification studies are reported by Kramer and inant frequency of the input earthquake
Stewart (2004). Although a direct comparison motion (fp) and the fundamental frequency of
with recorded ground response during seismic the soil deposit (f1). This latter assumption was
actions is invaluable, the performance of based on the evidence that the higher modes of
advanced numerical approaches can also be use- a shear beam are odd multiples of the funda-
fully evaluated through parametric analyses. mental mode of the beam. Recently, Kwok
These numerical studies can, in fact, provide et al. (2007) used linear frequency-domain
helpful guidelines to nonexpert users of time- solutions to provide guidelines on the specifi-
domain nonlinear schemes by clarifying the cation of viscous damping adopted in four 1D
importance of the different ingredients required nonlinear codes (namely, D-MOD_2,
to perform sophisticated simulations. One of the DEEPSOIL, OpenSees, and SUMDES). The
most controversial factors affecting FE dynamic idea was to assume the results of 1D linear
simulations is the calibration of the viscous equivalent analyses as target solutions for the
damping, usually needed in nonlinear analyses nonlinear approaches, thus guiding the user in
when the adopted soil constitutive model is the calibration of the viscous damping param-
unable to provide enough hysteretic energy dis- eters. The main conclusions of this parametric
sipation through plasticity. Viscous damping is investigation can be summarized as follows:
typically introduced in FE codes by means of the
Rayleigh formulation, whose damping matrix is • The target damping ratio should be set to the
defined as follows: small-strain material damping.
• The two target frequencies should be set, as a
½C
¼ aR ½M
þ bR ½K
(5) first approximation, to the first mode site fre-
quency and five times the site frequency (i.e.,
where [M] and [K] are the mass and the stiff- n = 5).
ness matrix of the system, respectively. The • More generally, the target frequencies should
coefficients aR and bR are obtained consider- be established through an iterative process by
ing the following relationship with the which linear time-domain and frequency-
damping ratio D: domain solutions are matched.

aR 2D om on Similarly to Kwok et al. (2007), Amorosi
¼ (6)
bR om þ on 1 et al. (2010) performed a parametric investigation
using two different finite element codes (i.e.,
where om and on are the angular frequencies PLAXIS 2D and SWANDYNE II) and compared
related to the frequency interval fm fn over the results of 2D FE viscoelastic simulations with S
which the viscous damping is equal to or lower 1D linear equivalent solutions (i.e., EERA). In
than D. The amount of Rayleigh damping to order to provide a useful framework for finite
be introduced in an FE dynamic analysis is element users, some of the factors potentially
difficult to quantify a priori, but, at the same influencing the numerical results were critically
time, its magnitude can play a crucial role on discussed. In particular, the amount of viscous
the results of the simulations (Woodward and damping adopted in FE viscoelastic analyses,
Griffiths 1996). Different possible calibration the spatial and time discretization, and the nature
procedures have been proposed in the litera- of boundary conditions were examined. The
ture to identify the interval fm fn. A well- investigation showed that:
established one (e.g., Hudson et al. 1994) sug-
gests to select fm as the first natural frequency • The traditionally adopted procedures for the
of the deposit f1, while fn is assumed equal to calibration of the Rayleigh coefficients (e.g.,
n times fm, where n is the closest odd integer Hudson et al. 1994) can lead to large
overestimation of the peak ground accelera- and eigenvectors of the matrix [M]1[K]. The sec-
tion with respect to the EERA solutions. ond formulation introduces a reduction factor that
Therefore, a novel calibration procedure was modifies the extended Masing loading/unloading
proposed based on the linear equivalent ampli- stress–strain relationship (Masing 1926) to match
fication function and the frequency content of measured modulus reduction and damping curves
the input motion. It was suggested that the first simultaneously over a wide range of shear strains.
target frequency (fm) should be set to the first This modified hysteretic model allows to reduce
mode site frequency significantly excited by the typically observed overestimation of damping
the seismic motion, while the second (fn) predicted by advanced soil constitutive models for
should be selected equal to the frequency large strains (see next section). The proposed
where the amplification function gets lower models have been implemented in a total stress
than one. nonlinear code and, therefore, cannot predict accu-
• As regards the effect of boundary conditions mulation of excess pore water pressures during the
adopted in the FE simulations, tied nodes at shaking.
the lateral boundaries of the mesh should be
preferred, as they are properly representative
of a 1D condition and allow to obtain a perfect Future Challenges
match with EERA. When using Lysmer and
Kuhlemeyer (1969) boundaries, the 2D mesh While during recent years important advances
should be characterized by a width to height have been made in the field of geotechnical earth-
ratio between five and eight to avoid spurious quake engineering, some challenges still remain
wave reflections at the vertical boundaries. to be addressed.
• In terms of spatial discretization, the FE mesh In situ and laboratory characterization of soil
should always satisfy the condition that the dynamic properties represents a crucial aspect of
spacing between the nodes, Dlnode, must be any seismic design of geotechnical structures.
smaller than approximately one-tenth to Techniques for more accurate characterization of
one-eighth of the wavelength associated with in situ soil shear wave velocity and damping ratio
the maximum frequency component fmax of are required (e.g., Hall and Bodare 2000). Although
the input wave. expensive and time consuming, advanced labora-
• The time-step algorithm adopted in the FE tory cyclic and/or dynamic tests are essential for the
code can introduce some numerical damping correct understanding of soil behavior during seis-
in the integration of the governing equations mic excitations and for the proper calibration of
of motion, and, therefore, the time-stepping sophisticated constitutive assumptions introduced
coefficients need to be selected carefully in in nonlinear ground response analyses. At the same
order not to lose accuracy. time, these tests should be interpreted within the
framework of plasticity theory, as irreversible
Two alternative soil damping formulations for deformations in soil samples can develop also for
small and large strains have been recently small imposed strains.
implemented by Phillips and Hashash (2009) in With respect to site response analysis, the use
the nonlinear one-dimensional site response anal- of nonlinear time-domain approaches should be
ysis code DEEPSOIL. The first one introduces an encouraged as a viable alternative to the standard
approach to construct a frequency-independent linear equivalent method. It can, in fact, better
viscous damping matrix which reduces the predict soil deformation, degradation of stiffness,
overdamping at high frequencies and, therefore, and accumulation of excess pore water pressures
the filtering at those frequencies. A good match throughout the shaking. The issue of the accurate
with frequency-domain analysis results can be prediction of hysteretic damping with
obtained with this new approach, which, nonethe- advanced soil constitutive models still remains
less, requires the calculation of the eigenvalues controversial, as these models can significantly
underpredict the damping ratio for small strains ground motion records based only on distance,
and overestimate it in the large strain range. Some site soil conditions, and magnitude of
adjustments in the mathematical formulation of the characteristic event expected to dominate
advanced soil models have been proposed the seismic hazard. However, many other
recently (e.g., Phillips and Hashash 2009; factors, such as directivity of the rupture and
Seidalinov and Taiebat 2014) to address this presence of basin, affect the ground motion inten-
point, but more research work is needed. In gen- sity and frequency content, which ultimately gov-
eral, clear guidance and protocols should be pro- ern the nonlinear response and damage in
vided to nonexpert users of time-domain codes in geotechnical and structural systems (e.g., Frankel
order to improve the reliability of nonlinear et al. 2011). Despite this, current practice is not
ground response analysis results. adequately equipped to fully incorporate near-
Another open challenge in the area of seismic fault directivity, basin, and duration effects in
risk assessment is related to the selection of the design process. Methods to implicitly con-
appropriate input motions for nonlinear ground sider inelastic demands by amplifying the design
response simulations. The correct definition of spectra do not provide a reliable basis for
the design seismic actions, based on seismic haz- representing realistic ground input motions
ard and site response analyses, is, indeed, essen- (Kalkan and Kunnath 2006).
tial to fully implement the performance-based More research is still needed to address all
design approach proposed by code prescriptions these issues, requiring a better integration
(e.g., Eurocode 8). The options available to engi- between the work of geotechnical engineers, seis-
neers in terms of input acceleration time series mologists, geologists, and structural engineers.
are represented by artificial, synthetic, and real Finally, an ongoing transfer of research knowl-
accelerograms. Spectrum-compatible artificial edge into practice represents a significant requi-
records are characterized by excessive number site to support performance-based earthquake
of cycles of strong motion, and consequently engineering design and quantitative seismic risk
they possess unreasonably high-energy content. assessment.
It is now widely accepted that the use of these
artificial records is problematic and not suitable
particularly for nonlinear analyses (e.g., Bommer Summary
and Acevedo 2004). Synthetic accelerograms
generated from seismological source models are Experience from past strong earthquakes world-
highly sensitive to the definition of earthquake wide has demonstrated the relevance of local soil
source parameters, which involve a significant conditions on seismic ground response. The
degree of expert judgment. On the contrary, real changes in amplitude, frequency content, and
earthquake accelerograms represent a more via- duration of the seismic motion during its propaga- S
ble option for providing input to dynamic ana- tion in soil deposits, commonly referred to as site
lyses, being more realistic than artificial records effects, have a crucial impact on buildings and
and easier to obtain than synthetic accelerograms. infrastructure response during earthquakes.
In any case, current practice for selecting and Numerical methods allow geotechnical engineers
scaling ground motions for linear and nonlinear to quantify the effects of soil deposits on the wave
response-history analyses is based largely on propagation process from bedrock to ground sur-
engineering judgment. Very few systematic stud- face. These methods can be divided into:
ies provide impartial guidance to geotechnical (i) frequency-domain analyses (using the linear
engineers regarding appropriate methods to use equivalent method) and (ii) time-domain schemes
in site-specific applications. The majority of the (usually performed with finite element codes). The
research has been focused in the last years on benefits and limitations of the two approaches are
structural response more than on site response extensively discussed in this entry, together with a
(e.g., NIST 2011). Practitioners often select short description of the available methods for site
characterization and evaluation of soil dynamic Arulanandan K, Scott RF (eds) (1993) Proceedings of
properties. An overview of future challenges in VELACS symposium. A.A., Balkema, Rotterdam
Ashford SA, Sitar N, Lysmer J, Deng N (1997) Topo-
the field of geotechnical earthquake engineering graphic effects on the seismic response of steep slopes.
is also presented at the end of the entry. Bull Seismol Soc Am 87(3):701–709
Associazione Geotecnica Italiana (2005) Aspetti
Geotecnici della Progettazione in Zona Sismica.
Patron Editore, Bologna
List of symbols ASTM (2007) Standard test methods for modulus and
damping of soils by resonant-column method. ASTM
D4015-07. ASTM International, West Conshohocken
amax Earthquake maximum acceleration Aubry D, Modaressi A (1996) GEFDYN – Manuel
Aloop Area of the hysteresis loop Scientifique. Ecole Centrale Paris, Châtenay-Malabry
D Damping ratio Aydingun O, Adalier K (2003) Numerical analysis of
f Frequency seismically induced liquefaction in earth embankment
foundations. Part I. Benchmark model. Can Geotech
f1 Fundamental frequency of the soil deposit J 40(4):753–765
fp Predominant frequency of the input Bard PY (1994) Effects of surface geology on ground
earthquake motion motion: recent results and remaining issues. In: Pro-
fr Resonance frequency ceedings of the X European conference on earthquake
Gsec Secant shear modulus engineering, vol 1, Vienna
Gtan Tangent shear modulus Bard PY, Bouchon M (1985) The two-dimensional reso-
nance of sediment-filled valleys. Bull Seismol Soc Am
G0 or Initial shear modulus
75(2):519–541
Gmax
Bardet JP, Tobita T (2001) NERA – a computer program
M Earthquake magnitude for Nonlinear Earthquake site Response Analyses of
N Number of imposed cycles layered soil deposits. Department of Civil Engineer-
NSPT SPT number ing, University of Southern California, Los Angeles
OCR Overconsolidation ratio Bardet JP, Ichii K, Lin CH (2000) EERA – a computer
program for Equivalent-linear Earthquake site
PI Plasticity index
Response Analyses of layered soils deposits. Univer-
qc CPT test tip resistance sity of Southern California, Department of Civil Engi-
VR Rayleigh wave velocity neering, Los Angeles. User manual
VS Shear wave velocity Bazzurro P, Cornell CA (2004) Nonlinear soil-site effects
WD Energy dissipated in one hysteresis loop in probabilistic seismic hazard analysis. Bull Seismol
WS Maximum strain energy in one hysteresis Soc Am 94:2110–2123
loop Biot MA (1941) General theory of three-dimensional con-
solidation. J Appl Phys 12:155–164
aR Rayleigh damping coefficient
Bommer JJ, Acevedo AB (2004) The use of real earth-
b Ratio of effective shear strain to maximum quake accelerograms as input to dynamic analysis.
shear strain J Earthq Eng 8(4):1–50
bR Rayleigh damping coefficient Borja RI, Wu WH (1994) Vibration of foundations on
g Shear strain incompressible soils with no elastic region. J Geotech
gc Cyclic shear strain Eng ASCE 120(9):1570–1592
Borja RI, Chao HY, Montáns FJ, Lin CH (1999) Nonlinear
r Soil density
ground response at Lotung LSST site. J Geotech
o Angular frequency Geoenviron Eng ASCE 125(3):187–197
Bouckovalas GD, Papadimitriou AG (2005) Numerical
evaluation of slope topography effects on seismic
ground motion. Soil Dyn Earthq Eng 25:547–558
References Castro G, Christian JT (1976) Shear strength of soils and
cyclic loading. J Geotech Eng Div ASCE
Alyami M, Rouainia M, Wilkinson SM (2009) Numerical 102(GT9):887–894
analysis of deformation behaviour of quay walls under Cavallaro A, Lanzo G, Pagliaroli A, Maugeri M, Lo Presti
earthquake loading. Soil Dyn Earthq Eng DCF (2003) A comparative study on shear modulus
29(3):525–536 and damping ratio of cohesive soil from laboratory
Amorosi A, Boldini D, Elia G (2010) Parametric study on tests. In: Di Benedetto H et al (eds) Deformation char-
seismic ground response by finite element modelling. acteristics of geomaterials. Swets & Zeitlinger, Lisse,
Comput Geotech 37(4):515–528 pp 257–265
Chan AHC (1995) User manual for DIANA- landfills. GeoMotions LLC, Washington. User’s
SWANDYNE II. School of Engineering, University manual
of Birmingham, Birmingham GeoSlope Int. Ltd. (2002) QUAKE/W for finite element
Clayton CRI (2011) Stiffness at small strain: research and dynamic earthquake analysis. User’s Guide, Calgary
practice. Géotechnique 61(1):5–37 Hall L, Bodare A (2000) Analyses of the cross-hole
Dakoulas P, Gazetas G (2005) Seismic effective-stress method for determining shear wave velocities and
analysis of caisson quay walls: application to Kobe. damping ratios. Soil Dyn Earthq Eng 20:167–175
Soils Found 45(4):133–147 Hardin B, Drnevich VP (1972) Shear modulus and
Dewoolkar MM, Ko H-Y, Pak RYS (2001) Seismic damping in soils: measurements and parameter effects.
behaviour of cantilever retaining walls with liquefiable J Soil Mech Div ASCE 98:603–624
backfills. J Geotech Geoenviron Eng ASCE Hashash YMA (2009) DEEPSOIL V 3.7, Tutorial and
127(5):424–435 user manual. 2002–2009. University of Illinois at
EduPro Civil Systems Inc. (1998) ProShake – ground Urbana-Champaign, Urbana
response analysis program. Redmond, Washington. Hatzigeorgiou GD, Beskos DE (2010) Soil-structure inter-
User’s manual action effects on seismic inelastic analysis of 3-D
Elgamal A (1992) Three-dimensional seismic analysis of tunnels. Soil Dyn Earthq Eng 30:851–861
La Villita dam. J Geotech Eng ASCE Hudson M, Idriss IM, Beikae M (1994) QUAD4M: a
118(12):1932–1958 computer program to evaluate the seismic response
Elgamal A, Parra E, Yang Z, Adalier K (2002a) Numerical of soil structures using finite element procedures and
analysis of embankment foundation liquefaction coun- incorporating a compliant base. Center for Geotechni-
termeasures. J Earthq Eng 6(4):447–471 cal Modeling, University of California, Davis
Elgamal A, Yang Z, Parra E (2002b) Computational Idriss IM (1990) Influence of local site conditions on
modelling of cyclic mobility and post-liquefaction earthquake ground motions. In: Proceedings of IV U.-
site response. Soil Dyn Earthq Eng 22:259–271 S. national conference on earthquake engineering,
Elgamal A, Yan L, Yang Z, Conte J (2008) Three- vol 1, Palm Springs
dimensional seismic response of humboldt bay bridge- Idriss IM, Sun JI (1992) SHAKE91: a computer program
foundation-ground system. J Struct Eng ASCE for conducting equivalent linear seismic response ana-
134(7):1165–1176 lyses of horizontally layered soils deposits. Center for
Elia G, Rouainia M (2013) Seismic performance of earth Geotechnical Modeling, University of California,
embankment using simple and advanced numerical Davis
approaches. J Geotech Geoenviron Eng ASCE Idriss IM, Lysmer J, Hwang R, Seed HB (1973) QUAD-4:
139(7):1115–1129 a computer program for evaluating the seismic
Elia G, Rouainia M (2014) Performance evaluation of a response of soil structures by variable damping finite
shallow foundation built on structured clays under element procedures. Report No EERC 73-16, Earth-
seismic loading. Bull Earthq Eng 12(4):1537–1561 quake Engineering Research Center, University of
Elia G, Amorosi A, Chan AHC, Kavvadas M (2011) Fully California, Berkeley
coupled dynamic analysis of an earth dam. Ishibashi I (1992) Discussion to “effect of soil plasticity
Géotechnique 61(7):549–563 on cyclic response”, by M. Vucetic and R. Dobry.
Field EH, Jacob KH (1993) The theoretical response of J Geotech Eng ASCE 118(5):830–832
sedimentary layers to ambient seismic noise. Geophys Itasca Consulting Group Inc. (2002) FLAC – Fast
Res Lett 20(24):2925–2928 Lagrangian Analysis of Continua. Minneapolis.
Finn WDL, Yogendrakumar M, Yoshida N, Yoshida User’s manual
H (1986) TARA-3: a program for nonlinear static and Kalkan E, Kunnath SK (2006) Effects of fling step and
dynamic effective stress analysis. Soil Dynamics forward directivity on seismic response of buildings. S
Group, University of British Columbia, Vancouver Earthq Spectra 22(2):367–390
Frankel A, Stephenson W, Carver D, Odum J, Williams R, King JL, Tucker BE (1984) Observed variations of earth-
Rhea S (2011) Probabilistic seismic hazard maps for quake motion across a sediment-filled valley. Bull
seattle: 3D sedimentary basin effects, nonlinear site Seismol Soc Am 74(1):137–151
response, and uncertainties from random velocity var- Kontoe S, Zdravkovic L, Potts DM, Menkiti CO
iation. In: Proceedings of the 4th IASPEI/IAEE inter- (2011) On the relative merits of simple and advanced
national symposium: effects of surface geology on constitutive models in dynamic analysis of tunnels.
seismic motion, University of California, Santa Géotechnique 61:815–829
Barbara Kramer SL (1996) Geotechnical earthquake engineering.
Geli L, Bard PY, Jullien B (1988) The effect of topogra- Prentice Hall, Upper Saddle River, New Jersey
phy on earthquake ground motion: a review and new Kramer SL, Stewart JP (2004) Geotechnical aspects of
results. Bull Seismol Soc Am 78(1):42–63 seismic hazards. In: Bozorgnia Y, Bertero VV (eds)
GeoMotions (2007) D-MOD2000 – a computer program Earthquake engineering – from engineering seismol-
package for seismic response analysis of horizontally ogy to performance-based engineering. CRC Press
layered soil deposits, earthfill dams, and solid waste LLC, Boca Raton
Kwok AOL, Stewart JP, Hashash YMA, Matasovic N, Meng J, Rix GJ (2003) Reduction of equipment-generated
Pyke R, Wang Z, Yang Z (2007) Use of exact solutions damping in resonant column measurements.
of wave propagation problems to guide implementa- Géotechnique 53(5):503–512
tion of nonlinear seismic ground response analysis NIST (2011) Selecting and scaling earthquake ground
procedures. J Geotech Geoenviron Eng ASCE motions for performing response-history analyses.
133(11):1385–1398 NIST GCR 11-917-15, National Institute of Standards
Lanzo G, Silvestri F (1999) Risposta Sismica Locale and Technology, Gaithersburg
(Teoria ed esperienze). Hevelius Edizioni srl, Ou JH (2009) Three-dimensional numerical modelling of
Benevento interaction between soil and pore fluid. PhD thesis,
Lanzo G, Silvestri F, Costanzo A, d’Onofrio A, Martelli L, School of Engineering, University of Birmingham
Pagliaroli A, Sica S, Simonelli A (2011) Site response Pagliaroli A, Lanzo G, D’Elia B (2011) Numerical evalu-
studies and seismic microzoning in the Middle Aterno ation of topographic effects at the Nicastro ridge in
valley (L’Aquila, Central Italy). Bull Earthq Eng Southern Italy. J Earthq Eng 15(3):404–432
9(5):1417–1442 Papaspiliou M, Kontoe S, Bommer JJ (2012) An explora-
Lee MKW, Finn WDL (1978) DESRA-2, dynamic effec- tion of incorporating site response into PSHA-part II:
tive stress response analysis of soil deposits with sensitivity of hazard estimates to site response
energy transmitting boundary including assessment approaches. Soil Dyn Earthq Eng 42:316–330
of liquefaction potential, Soil mechanics series no 38. Phillips C, Hashash YMA (2009) Damping formulation
Department of Civil Engineering, University of British for nonlinear 1D site response analyses. Soil Dyn
Columbia, Vancouver Earthq Eng 29:1143–1158
Li XS, Wang ZL, Shen CK (1992) SUMDES: a nonlinear Prevost JH (2002) DYNAFLOW – a nonlinear transient
procedure for response analysis of horizontally layered finite element analysis program. Version 2002.
sites subjected to multi-directional earthquake loading. Release 01.A, Department of Civil Engineering &
Department of Civil Engineering, University of Cali- Operation Research, Princeton University, Princeton
fornia, Davis Pyke RM (1992) TESS: a computer program for nonlinear
Liu H, Song E (2005) Seismic response of large underground response analyses. TAGA Engineering Sys-
ground structures in liquefiable soils subjected to hor- tems and Software, Lafayette
izontal and vertical earthquake excitations. Comput Richart FE, Hall JR, Woods RD (1970) Vibrations of soils
Geotech 32(4):223–244 and foundations. Prentice-Hall, Englewood Cliffs
Lo Presti DCF, Jamiolkowski M, Pallara O, Cavallaro A, Roesset JM (1977) Soil amplification of earthquakes. In:
Pedroni S (1997) Shear modulus and damping of soils. Desai CS, Christian JT (eds) Numerical methods in
Géotechnique 47(3):603–617 geotechnical engineering. McGraw-Hill, New York,
Lo Presti DFC, Pallara O, Mensi E (2007) Characteriza- pp 639–682
tion of soil deposits for seismic response analysis. In: Sangrey DA, Henkel DJ, Esrig MI (1969) The effective
Ling HI et al (eds) Soil stress–strain behavior: mea- stress response of a saturated clay soil to repeated
surement, modeling and analysis. Springer, Dordrecht, loading. Can Geotech J 6(3):241–252
The Netherlands, pp 109–157 Schnabel PB, Lysmer J, Seed HB (1972) SHAKE: a com-
Lysmer J, Kuhlemeyer RL (1969) Finite dynamic model puter program for earthquake response analysis of
for infinite media. J Eng Mech Div ASCE horizontally layered sites. Report no EERC72-12,
95(EM4):859–877 Earthquake Engineering Research Center, University
Lysmer J, Udaka T, Tsai C-F, Seed HB (1975) FLUSH: a of California, Berkeley
computer program for approximate 3-D analysis of Seed HB, Idriss IM (1970) Soil moduli and damping
soil-structure interaction problems. Report EERC factors for dynamic response analyses. Report EERC
75-30, University of California, Berkeley 70-10, Earthquake Engineering Research Center, Uni-
Madabhushi SPG, Zeng X (2007) Simulating seismic versity California, Berkeley
response of cantilever retaining walls. J Geotech Seed HB, Ugas C, Lysmer J (1976) Site dependent spectra
Geoenviron Eng ASCE 133(5):539–549 for earthquake resistant design. Bull Seismol Soc Am
Masing G (1926) Eignespannungen und verfestigung 66(1):221–243
beim messing. In: Second international congress on Seidalinov G, Taiebat M (2014) Bounding surface
applied mechanics, Zurich, pp 332–335 SANICLAY plasticity model for cyclic clay behavior.
Matasovic N (1995) D-MOD_2 A computer program for Int J Numer Anal Methods Geomech 38(7):702–724
seismic response analyses of horizontally layered soil Semblat JF, Pecker A (2009) Waves and vibrations in
deposits, earthfill dams and solid waste landfills. soils: earthquakes, traffic, shocks, construction
GeoSyntec Consultants, Huntington Beach works. IUSS Press, Pavia
McKenna F, Fenves GL (2001) The OpenSees command Semblat JF, Kham M, Parara E, Bard PY, Pitilakis K,
language manual, version 1.2. Pacific Earthquake Makra K, Raptakis D (2005) Seismic wave amplifica-
Engineering. Research Center, University of Califor- tion: basin geometry vs soil layering. Soil Dyn Earthq
nia, Berkeley Eng 25:529–538
Site Response: 1-D Time Domain Analyses 3287
SESAME (2004) Guidelines for the implementation of the Introduction

H/V spectral ratio technique on ambient vibrations
measurements, processing and interpretation. SES-
AME European Research Project. WP12 – Deliverable One-dimensional (1-D) time domain analysis is
D23.12 one of several currently available methods to
Shahrour I, Khoshnoudian F, Sadek M, Mroueh H (2010) evaluate the influence of local site conditions on
Elastoplastic analysis of the seismic response of tun- input (i.e., bedrock) ground motions. However,
nels in soft soils. Tunn Undergr Space Technol 25:
478–482 this method is favored by Geotechnical Earth-
Sica S, Pagano L, Modaressi A (2008) Influence of past quake Engineers as it takes into account the
loading history on the seismic response of earth dams. unique geotechnical characteristic of a given
Comput Geotech 35(1):61–85 site, including soil nonlinearity and hysteretic
Stamos AA, Beskos DE (1995) Dynamic analysis of large
3-D underground structures by the BEM. Earthq Eng behavior and porewater pressure generation and
Struct Dyn 24(6):917–934 dissipation. It can also be applied for a wide range
Tsai C-CP (2000) Probabilistic seismic hazard analysis of shaking intensities, and it is relatively easy to
considering nonlinear site effect. Bull Seismol Soc implement in practice as it is coded in commer-
Am 90:66–72
Vinale F, Mancuso C, Silvestri F (1996) Dinamica dei cially available software.
terreni. Manuale di Ingegneria Civile Cremonese. Through back analysis of numerous case his-
Zanichelli/ESAC, Roma tories, the 1-D time domain analysis has been
Vucetic M (1986) Pore pressure buildup and liquefaction shown to work well when analyzed soil deposits
at level sand sites during earthquakes. Rensselaer
Polytechnic Institute, Troy are horizontally layered, there is a significant
Vucetic M, Dobry R (1991) Effects of the soil plasticity on impedance contrast within the profile, and when
cyclic response. J Geotech Eng Div ASCE 117(1): material (model) parameters are established
89–107 through a reasonable site investigation and char-
Woods RD (1994) Geophysical characterization of sites.
Balkema, Rotterdam acterization efforts.
Woodward PK, Griffiths DV (1996) Influence of viscous There are many circumstances under which
damping in the dynamic analysis of an earth dam using a 1-D time domain site response analysis is
simple constitutive models. Comput Geotech 19(3): performed. Such an analysis is commonly
245–263
Zienkiewicz OC, Chan AHC, Pastor M, Schrefler BA, performed in the following situations: (i) when
Shiomi T (1999) Computational geomechanics (with soil conditions cannot be reasonably categorized
special reference to earthquake engineering). Wiley, into one of the standard site conditions, (ii) when
Chichester empirical site factors for the site are not available
(e.g., the International Building Code Site Class
F), (iii) when special ground conditions govern
the design (e.g., soil liquefaction, seismic settle-
ment and/or lateral spreading, or slope stability),
Site Response: 1-D Time Domain (iv) for any case where the objective is to obtain S
Analyses ground motions considered to be more represen-
tative of the local geologic and seismic site con-
Neven Matasovic ditions than motions obtained from other
Geosyntec Consultants, Huntington Beach, CA, approaches, and/or (v) when calibration of
USA a more advanced, multidimensional models with
or without structural elements is required.
In general, site response analysis methods can
Synonyms be classified based upon the domain in which
calculations are performed (frequency domain
1-D nonlinear seismic site response analysis; 1-D or time domain), the sophistication of the consti-
nonlinear site response analysis; One- tutive model employed (linear, equivalent linear,
dimensional (1-D) time domain analysis and/or nonlinear), whether the effects of
3288 Site Response: 1-D Time Domain Analyses
porewater pressure generation are neglected or The dynamic equation of motion can be solved
not (total-stress and effective-stress analyses, by numerical integration. The numerical integra-
respectively), and the dimensionality of the tion calls for temporal discretization (i.e., system
space in which the analysis is performed of coupled equations is discretized temporally),
(one-dimensional, 1-D; quasi 2-D, 2-D, and 3-D). hence the term “solution in time domain.” The
Other considerations in classifying site response dynamic equation is solved by a time-stepping
analysis methods include modeling of cyclic scheme. Examples of time-stepping schemes
reduction and cyclic degradation in a total-stress include Wilson’s theta (y) algorithm (Clough
mode. This chapter focuses upon time domain and Penzien 1993) and several variations of the
nonlinear total- and effective-stress 1-D site Newmark’s b algorithms (Newmark 1959).
response analysis which has significant advan- In order to solve the dynamic equation of
tages over its frequency domain linear and motion, it is necessary to properly discretize the
equivalent-linear counterparts in areas of very domain of interest, which in this case is the 1-D
high seismicity (e.g., PGA 0.4 g) and/or when soil column. Two different approaches for
the effects of soft and/or potentially liquefiable discretization of the soil column are available:
soils in the profile cannot be ignored. (i) lumped mass discretization and (ii) finite ele-
ment discretization.
Figure 1 shows a lumped mass discretization
Theoretical Background (i.e., lumped mass model) that of a horizontally
layered soil deposit. Soil mass (m) is lumped at
Dynamic Response Model the layer interfaces (layers 1 through n). In each
In 1-D time domain analysis, seismic response layer, soil stiffness is represented by a nonlinear
of a horizontally layered soil deposit is computed spring. In addition to hysteretic damping inherent
by solving the dynamic equation of motion. to a nonlinear spring, a viscous damping coeffi-
The dynamic equation of motion is commonly cient (c) is also part of a model shown in Fig. 1, as
written as schematically represented by dashpots.
The finite element discretization requires solv-
½M
f€
ug þ ½C
fůg þ ½K
fug ing the dynamic equation of motion (Eq. 1) by
¼ ½M
u€g (1) means of an explicit time marching integration
algorithm. In this type of discretization, mass can
where [M], [C], and [K] are the mass matrix, be distributed over layer thickness.
viscous damping matrix, and nonlinear stiffness The thickness of the layers in both lumped
matrix, respectively; {u}, {ů}, and {€u} are the mass and finite element discretization have an
displacements, velocities, and accelerations of impact on calculated site response. The layer
the mass [M] relative to the base; and {€ug} is thickness determines the maximum and mini-
the acceleration of the base. mum frequency that can be propagated through
The stiffness matrix [K] is derived from the a soil column. If the layer is too thick, important
nonlinear (and effective-stress) soil constitutive components of the input ground motion may be
model selected to represent a cyclic response of filtered, and thus the calculated ground response
soil. In principle, all damping in the soil can be may be underestimated. If layer thickness is too
captured through the hysteretic loops of the con- small, representation of the actual site conditions
stitutive model used to model cyclic soil behav- might be inadequate (e.g., soil lens is modeled as
ior. However, most available soil constitutive a soil layer).
models cannot properly represent soil damping
at low strains. Therefore, it is necessary to add Nonlinear Constitutive Models with
damping through the use of velocity propor- Hysteretic Damping
tional viscous damping, as discussed in detail One-dimensional nonlinear total-stress site
below. response analysis is generally conducted using
Site Response: 1-D Time Domain Analyses, Fig. 1 Lumped mass discretization for 1-D dynamic response model
(Matasovic 1993)
relatively simplified constitutive models of soil (Pyke 1979; Vucetic 1990). Recently, Phillips
layers. These models evolved from the early and Hashash (2009) developed a set of rules that
stress–strain relationships of Ramberg and compensate for one of the most common defi-
Osgood (1943) and Kondner and Zelasko ciency of nonlinear models with hysteretic
(1963). The hyperbolic model, introduced by damping – overestimation of damping at large
Duncan and Chang (1970) for axial soil behavior, shear strains.
was based upon the basic formulation of the Bounding surface plasticity models may be
above-cited shear stress and strain behavior used to simulate nonlinear stress–strain behavior
models. It was accompanied by sets of generic in one dimension. Example of such models and
material properties and hence allowed for an ele- their application in site response analysis can be
gant and simple way to capture soil nonlinearity found in Borja et al. (2002).
at small axial strains. All three models provided
the basis for constitutive models that are pres- Viscous Damping Models
ently in use. These include models by Pyke The viscous damping models are an essential part S
(1979), Matasovic (1993), Matasovic and of nonlinear site response programs. This is
Vucetic (1993), and Darendeli (2001) which, in because: (i) viscous damping is a necessary input
addition to better simulation of nonlinear into a linear-elastic solution that is typically an
stress–strain behavior, allow for simulation of integral part of a time domain site response analy-
cyclic loading and reloading in accordance with sis program; (ii) viscous damping readily increases
certain rules. The stress–strain relationship in numerical stability of several numerical schemes
these models is generally established by: an ini- used to solve Eq. 1; and (iii) damping calculated by
tial loading curve; a series of rules that describe a hyperbolic model at very low strains (strains less
the backbone curve (see Fig. 2 for definition of than 0.1 %) is very small compared to the values
backbone curve); and unloading–reloading obtained by laboratory measurements. Viscous
behavior rules required to establish cyclic loops. damping compensates for that difference.
The most widely used rules are the Masing Given the above, most engineers introduce
rules (Masing 1926) and extended Masing rules a small amount of viscous damping into site
Site Response: 1-D Time Domain Analyses, Fig. 2 Backbone curve as stress–shear strain relationship for mono-
tonic loading
Site Response: 1-D Time

Domain Analyses,
Fig. 3 Schematic
illustration of viscous
damping change with
frequency
response analysis. The value of viscous damping and m and k are elements of the mass and stiff-
introduced is typically small (0.5–5 %). ness matrices, respectively.
The most commonly used formulation for Figure 3 illustrates how Rayleigh damping,
evaluation of the viscous damping coefficient expressed through c, changes with frequency.
(c) is the Rayleigh damping (model). The viscous The viscous damping ratio can be brought closer
damping coefficient can be evaluated as to a constant value of the target damping ratio
(xtar) by specifying c at only one frequency (e.g.,
c ¼ aR m þ bR k (2) at f2 in Fig. 3), which is termed the simplified
Rayleigh damping formulation, and/or at two
where aR and bR are the Rayleigh damping frequencies (at f1 and f2), which is termed
coefficients (Rayleigh and Lindsay 1945) the full Rayleigh damping formulation.
Kwok et al. (2007) recommended, for most prac- computed in a semiempirical way. At the begin-
tical applications, use of the full Rayleigh ning of shaking (i.e., at time t = 0), stress–strain
damping formulation in nonlinear (total-stress) relationships of the soil are identical to that of the
site response analysis whereby the first frequency total-stress models since PWP is zero. As shaking
is equal to the fundamental frequency of the soil progresses, PWP is generated and cyclic degra-
column, while the second frequency is equal to dation (of clay microstructure) starts. Subse-
five times the fundamental frequency. quently, the effects of porewater pressure
generation and, in some models, of cyclic degra-
dation are included by degradation of soil
Time Domain Analysis with Porewater strength and stiffness. Different models for deg-
Pressure Change radation of soil strength and stiffness are used in
some of the models. For example, Matasovic
General (1993) and Matasovic and Vucetic
The cyclic loading of saturated soils is accompa- (1995b) proposed degradation index functions
nied by porewater pressure change. This change that degrade strength and stiffness of sands at
includes simultaneous porewater pressure gener- different rates, while the concept of the degrada-
ation and dissipation. If the generated porewater tion index (Idriss et al. 1978) is generally used to
pressures are sufficiently large, the soil stiffness degrade strength and stiffness of soft clays.
and strength are significantly reduced, and ulti- A number of PWP generation models have
mately, in some soils, liquefaction can occur. been developed starting with Martin and Seed
In nonlinear site response analysis with (1978) and Martin et al. (1975). A more recent
porewater pressure generation (i.e., in effective- example of the semiempirical porewater pressure
stress analysis), the response of the soil to cyclic model for saturated sand is Dobry et al. (1985)
loading accounts for the generation of excess model. This model was developed based upon
porewater pressure (PWP) during cyclic shearing strain-controlled cyclic direct simple shear and
of the soil as well as dissipation of these excess cyclic triaxial testing. The model was later mod-
porewater pressures during and after the cyclic ified by Vucetic (1986) to allow for quasi-two-
loading. The representation of dissipation/redis- dimensional shaking and further by Matasovic
tribution of porewater pressure influences soil (1993) to more accurately model porewater
stiffness (modulus) and strength (shear stress) pressure-induced degradation of shear modulus
during shaking, thus resulting in a more realistic and shear stress. The porewater pressure genera-
simulation of site response. The PWP dissipation/ tion models described thus require the use of an
redistribution is discussed later. equivalent number of cycles to represent earth-
The influence of PWP changes during cyclic quake shaking. Polito et al. (2008) introduced an
loading is incorporated in soil constitutive energy-based model (GMP model) for the gener- S
models in two ways: (i) semiempirical PWP ation of porewater pressure based on a large num-
generation models used in combination with ber of laboratory tests, which does not require the
total-stress soil models and (ii) advanced development of an equivalent number of cycles.
effective-stress models whereby the model for- With exception of the modified Dobry
mulation is in terms of effective stress, and the et al. (1985) model, there is limited information
PWP change is computed as the change between available to guide the user in selecting the appro-
total stresses (or loads) and effective stresses via priate PWP model parameters.
the soil constitutive model. The effect of cyclic degradation on soil stiff-
ness and strength is illustrated in Fig. 4 for the
Semiempirical Porewater Pressure MKZ constitutive model (Matasovic 1993;
Generation Models Matasovic and Vucetic 1993). The initial hyster-
In this class of porewater pressure generation esis loop shown in the figure refers to the
models, PWP generation in a soil profile is first cycle of cyclic loading (i.e., at time t = 0).
τ
Gmo
Gso
1 Gmt 1
INITIAL HYSTERESIS LOOP τco 1
Gst
1
SUBSEQUENT DEGRADED τct
HYSTERESIS LOOP
−γct −γco ≈Gmt

1
γco γct γ
SUBSEQUENT DEGRADED ≈Gmo
BACKBONE CURVE
−τct 1
−τco
INITIAL BACKBONE CURVE
Site Response: 1-D Time Domain Analyses, Fig. 4 Stress–strain behavior modeling illustrating stiffness degrada-
tion due to porewater pressure buildup (Matasovic 1993)
The subsequent degraded hysteretic loop refers to constitutive law is developed in effective-stress
any subsequent cycle (i.e., at time t) for which space, and porewater pressures are computed as
enough porewater pressure has built up to the difference between effective stresses and total
degrade both initial shear modulus Gmo and initial stresses in the domain of interest. Examples of
shear stress tco at corresponding shear strain gco. plasticity-based constitutive models include
An example of a porewater pressure model for Roscoe and Schofield (1963), Pestana (1994),
clay is the Matasovic and Vucetic (1995a) model. and Elgamal et al. (2001). These advanced con-
This model was developed based upon the stitutive models are capable of simulating com-
results of cyclic simple shear testing. It should plex soil behavior under a variety of loading
be noted that porewater pressure in clay is of conditions. Key elements of these models
much lower intensity than in sand and that in include yield surfaces, flow rules, and hardening
overconsolidated clay, both positive and negative (or softening) laws. A review of advanced
(suction) porewater pressures may develop (e.g., constitutive models with application in site
Matasovic and Vucetic 1992). response analysis is provided in Potts and
Zdravković (1999).
Generic material parameters for advanced
Advanced Effective-Stress-Based Models constitutive models are often not available. Eval-
Another class of soil constitutive models used in uation of material parameters for these models
site response analysis is effective-stress models. requires significant expertise and detailed site-
In these models, the formulation of the specific soil properties.
Porewater Pressure Dissipation and TESS (Pyke 2000), with last three programs
Redistribution Models also having a provision for effective-stress anal-
The layers in a soil column have finite, saturated ysis. The study compared the programs in total-
hydraulic conductivities. During ground shak- stress mode and found that, when input is prop-
ing, even though loading is relatively rapid, it erly controlled, most of the software provided
is possible that porewater dissipation and redis- similar results.
tribution can occur due to differences in hydrau- Matasovic and Hashash (2012) performed
lic gradients and hydraulic conductivities a survey of the United States Department of
between layers. Transportation (DOT) practices. Part of the sur-
An early model for porewater pressure dissi- vey was an inquiry on what site response analysis
pation and redistribution was introduced by Mar- programs are used, both by DOT-s and their con-
tin and Seed (1978). Input parameters include sultants. The survey revealed that the most used
(effective-stress-dependent) constrained rebound program was D-MOD2000 (Matasovic and
modulus and (saturated) hydraulic conductivity. Ordonez 2007) followed by DEEPSOIL
A porewater pressure dissipation model for com- (Hashash et al. 2011). However, D-MOD2000
posite soil deposits (alternating sand and clay was the only program cited to have been used in
layers) was developed by Matasovic and Vucetic effective-stress mode. Coincidently, these two
(1995a). programs are the only time domain analysis pro-
Porewater pressure redistribution occurs and grams available in United States that have
is often modeled in accordance with the princi- a graphical user interface and the effective-stress
ples of Terzaghi’s theory of consolidation. analysis provision.
Time Domain Analysis Computer Calibration and Benchmarking Studies

Programs
A number of individuals and groups have
There are about 25 time domain analysis com- conducted back-analysis exercises to evaluate
puter programs that are available globally, either site response analysis procedures (and programs)
commercially or through direct contact with versus measured response to strong ground shak-
the author(s). However, only a handful of these ing. Most of these studies include use of record-
programs have a provision to perform an ings from downhole vertical arrays.
effective-stress analysis, including 2-D and 3-D Researchers have used a range of inverse anal-
programs which can be used for analysis of 1-D ysis techniques that include ad hoc system iden-
geometry. tification (e.g., Zeghal and Elgamal 1993;
Kwok et al. (2007) performed a comprehen- Assimaki and Steidl 2007) and evolutionary soil S
sive survey of available time domain analysis models (e.g., Tsai and Hashash 2007) to calibrate
computer programs in the United States and soil constitutive models in site response analysis
found that only five 1-D nonlinear programs actual (in situ) soil behavior.
were used at the time, including OpenSees A number of benchmarking exercises have
(Ragheb 1994) which is primarily used for anal- also been conducted to evaluate site response
ysis of 2-D and 3-D problems. Other programs analysis tools. A recent key exercise is the
identified by Kwok et al. (2007) include PEER benchmarking exercise for total-stress
DEEPSOIL (Hashash and Park 2001), site response analyses (Kwok et al. 2007).
D-MOD_2 (Matasovic 2006; subsequently Another interesting benchmarking exercise was
updated with graphical user interface as the evaluation of site response at Turkey Flat
D-MOD2000), SUMDES (Wang 1990), and (e.g., Kramer 2009). The interesting outcome of
this exercise is in highlighting the challenges References

encountered in computing the response at
a well-constrained relatively stiff soil site. Assimaki D, Steidl J (2007) Inverse analysis of weak and
strong motion downhole array data from the Mw 7.0
Sanriku-Minami earthquake. Soil Dyn Earthq Eng
27:73–92
Summary Borja RI, Duvernay BG, Lin CH (2002) Ground response
in Lotung: total stress analyses and parametric studies.
J Geotech GeoEnviron Eng 128(1):54–63
One-dimensional time domain analysis is
Clough RW, Penzien J (1993) Dynamics of structures,
a powerful tool for evaluation of site response of 2nd edn. McGraw Hill, London
horizontally layered soil deposits. Compared to Darendeli MB (2001) Development of a new family of
other types of 1-D analysis, it has a provision to normalized modulus reduction and material damping
curves. PhD thesis, University of Texas at Austin, Austin
simulate cyclic soil behavior in a more realistic
Dobry R, Pierce WG, Dyvik R, Thomas GE, Ladd RS
manner than its frequency domain counterparts, (1985) Pore pressure model for cyclic straining of
and it can accommodate models for seismically sand. Research Report, Civil Engineering Department,
induced porewater pressure buildup. Compared Rensselaer Polytechnic Institute, Troy, 56 p
Duncan JM, Chang C-Y (1970) Nonlinear analysis of
to its 2-D and 3-D counterparts, the advantage of
stress and strain in soils. J Soil Mech Found Div
this type of analysis lies in simple constitutive 96(SM5):1629–1653
models for which material parameters can be read- Elgamal A, Lai T, Yang Z, He L (2001) Dynamic soil
ily evaluated, or generic sets of material parame- properties, seismic downhole arrays and applications
in practice. In: Proceedings of the 4th international
ters are available to practicing engineers.
conference on recent advances in geotechnical earth-
Furthermore, compared to its 2-D and 3-D coun- quake engineering and soil dynamics, San Diego
terparts, 1-D analysis is far better calibrated Hashash YMA, Park D (2001) Nonlinear one-dimensional
and validated. Consequently, consistent with cur- seismic ground motion propagation in the Mississippi
embayment. Eng Geol (Amst) 62(1–3):185–206
rent trends, it is anticipated that the role of 1-D
Hashash YMA, Phillips C, Groholski D (2010) Recent
analysis will evolve into calibration of 2-D and advances in non-linear site response analysis. In: Pro-
3-D models. ceedings of the 5th international conference on recent
One-dimensional time domain analysis is not advances in geotechnical earthquake engineering and
soil dynamics, San Diego
perfect. Further calibration and validation of 1-D
Hashash YMA, Groholski DR, Phillips CA, Park D,
models are required both in effective-stress mode Musgrove M (2011) DEEPSOIL 4.0, User Manual
and at high levels of shaking (>0.4 g). There are and Tutorial. 98 p
also several issues related to the application of Idriss IM, Dobry R, Singh RD (1978) Nonlinear behavior
of soft clays during cyclic loading. J Geotech Eng Div
this type of analysis in engineering practice and
ASCE 104(12):1427–1447
research, as summarized by Hashash et al. Kondner RL, Zelasko JS (1963) A hyperbolic stress–strain
(2010). Future trends will certainly address formulation of sands. In: Proceedings of the 2nd Pan
these issues, yet also include improvement of American conference on soil mechanics and founda-
tion engineering, Sao Paulo, pp 289–324
graphical user interfaces that are essential not
Kramer SL (2009) Analysis of Turkey flat ground motion
only for effective use of the programs, but for prediction experiment – lessons learned and implica-
evaluation and validation of the results. tions for practice. In: SMIP09 seminar on utilization of
strong-motion data, San Francisco, pp 1–22
Kwok O-LA, Stewart JP, Hashash YMA, Matasovic N,
Pyke R, Wang Z, Yang Z (2007) Use of exact solutions
Cross-References of wave propagation problems to guide implementa-
tion of nonlinear ground response analysis procedures.
▶ Selection of Ground Motions for Response ASCE J Geotechn Geoenviron Eng 133(11):
1385–1398
History Analysis
Martin PP, Seed HB (1978) APOLLO, a computer pro-
▶ Site Response: Comparison Between Theory gram for the analysis of pore pressure generation and
and Observation dissipation in horizontal sand layers during cyclic
Site Response: Comparison Between Theory and Observation 3295
earthquake loading. Report No. EERC 78–21, Earth- Pyke RM (1979) Nonlinear soil models for irregular cyclic
quake Engineering Research Center, University of loadings. J Geotech Eng Div, ASCE 105(GT6):
California, Berkeley 715–726
Martin GR, Finn WDL, Seed HB (1975) Fundamentals of Pyke RM (2000) TESS: a computer program for nonlinear
liquefaction under cyclic loading. J Geotech Eng Div ground response analyses. TAGA Engineering Sys-
ASCE 101(GT5):423–438 tems & Software, Lafayette, http://www.tagasoft.com
Masing G (1926) Eigenspannungen und Verfestigung Ragheb AM (1994) Numerical analysis of seismically
beim Messing. In: Proceedings of the 2nd international induced deformations in saturated granular soil strata.
congress on applied mechanics, Z€ urich, pp 332-335 PhD dissertation, Department of Civil Engineering,
Matasovic N (1993) Seismic response of composite hori- Rensselaer Polytechnic Institute, Troy
zontally layered soil deposits. PhD dissertation, Civil Ramberg W, Osgood WR (1943) Description of
and Environmental Engineering Department, Univer- stress–strain curves by three parameters. Technical
sity of California, Los Angeles, 452 p Note 902, National Advisory Committee for Aeronau-
Matasovic N (2006) D-MOD_2 – a computer program for tics, Washington, DC
seismic response analysis of horizontally layered soil Rayleigh JWS, Lindsay RB (1945) The theory of sound.
deposits, earthfill dams, and solid waste landfills. Dover, New York
User’s Manual, GeoMotions, LLC, Lacey, 20 p (plus Roscoe KH, Schofield AN (1963) Mechanical behavior of
Appendices) an idealized ‘wet’ clay. In: Proceedings of the 2nd
Matasovic N, Hashash YMA (2012) Site response analysis European conference on soil mechanics, vol 1,
in transportation engineering practice – a TRB Survey. Wiesbaden, pp 47–54
In: Proceedings of the GeoCongress 2012, Oakland, Tsai C-C, Hashash YMA (2007) A novel framework inte-
CD-ROM paper, pp 1789–1798 grating downhole array data and site response analysis
Matasovic N, Ordonez GA (2007) D-MOD2000 – a com- to extract dynamic soil behavior. Soil Dyn Earthq Eng
puter program package for seismic response analysis 28(3):181–197
of horizontally layered soil deposits, earthfill dams, Vucetic M (1986) Pore pressure buildup and liquefaction
and solid waste landfills. User’s Manual, GeoMotions, of level sandy sites during earthquakes. PhD disserta-
LLC, Lacey, Washington, 182 p. http://www. tion, Rensselaer Polytechnic Institute, Troy, 616 p
geomotions.com Vucetic M (1990) Normalized behavior of clay under
Matasovic N, Vucetic M (1992) A pore pressure model for irregular cyclic loading. Can Geotech J 27:29–46
cyclic straining of clay. Soils Found J JSSMFE Wang ZL (1990) Bounding surface hypoplasticity model
32(3):156–173 for granular soils and its applications. PhD disserta-
Matasovic N, Vucetic M (1993) Cyclic characterization of tion, University of California at Davis
liquefiable sands. ASCE J Geotech Eng 119(11): Zeghal M, Elgamal AW (1993) Lotung sites: downhole
1805–1822 seismic data analysis. Electric Power Research Insti-
Matasovic N, Vucetic M (1995a) Generalized cyclic tute, Palo Alto
degradation-pore pressure generation model for
clays. ASCE J Geotech Eng 121(1):33–42
Matasovic N, Vucetic M (1995b) Seismic response of soil
deposits composed of fully saturated clay and sand
layers. In: Kenji Ishihara AA (ed) Proceedings of the
Site Response: Comparison Between
IS-Tokyo’95, the first international conference on Theory and Observation
earthquake geotechnical engineering, Balkema,
Rotterdam Takaji Kokusho
Newmark NM (1959) A method of computation for struc- S
tural dynamics. J Eng Mech Div 85:67–94
Department of Civil & Environmental
Pestana JM (1994) A unified constitutive model for clays Engineering, Chuo University, Tokyo, Japan
and sands. ScD thesis, MIT, Cambridge, MA
Phillips C, Hashash YMA (2009) Damping formulation
for nonlinear 1D site response analyses. Soil Dyn
Earthq Eng 29(7):1143–1158
Synonyms
Polito CP, Green RA, Lee JH (2008) Pore pressure gener-
ation models for sands and silty soils subjected to 1/4 wavelength formula; Two-layer system;
cyclic loading. J Geotech Geoenviron 134(10): Downhole array; Multilayer system; Multi-
1490–1500
Potts DM, Zdravković L (1999) Finite element analysis in
reflection theory; Nonlinear soil properties; Seis-
geotechnical engineering: theory. Thomas Telford, mic site amplification; Soil damping model; Sur-
London face array; S-wave velocity
3296 Site Response: Comparison Between Theory and Observation
Introduction
As As
Surface layer
Kanai et al. (1959) discovered from earthquake
r 1,Vs1,G ,D H
observations that a horizontal component in site 1 1
response can be explained by multi-reflection of A1 B1
SH-waves propagating vertically in soil layers.
Horizontal displacement of the vertically propa- Base layer A2 B2
gating SH-wave is expressed as r 2,Vs2,G2,D2
u ¼ Af ðz V s tÞ þ Bf ðz þ V s tÞ (1)
Site Response: Comparison Between Theory and
where z = vertical coordinate upward positive; Observation, Fig. 1 A simplified two-layer model to
t = time; Vs = S-wave velocity; f(), g() = an calculate site amplification
arbitrary functions; and A, B = amplitudes of
upward and downward waves, respectively. Site
response governed by this equation is largely and base layer, respectively. The wave equations
dependent on the composition of soil layers, in the surface and base layers are expressed as
S-wave velocities, and soil damping ratios of
those layers and their strain-dependent varia- u1 ¼ A1 eiðotk1 zÞ þ B1 eiðotþk1 zÞ
(4)
tions. In the following, site response observations u2 ¼ A2 eiðotk2 zÞ þ B2 eiðotþk2 zÞ
in array systems are compared with the SH-wave
multi-reflection theory to clarify its applicability where A1, B1 = amplitudes of upward and down-
in terms of dominant frequencies and spectrum ward waves in the surface layer and A2, B2 =
amplifications with special emphasis on soil those in the base layer.
damping mechanism and strain-dependent soil Utilizing the boundary conditions that at
nonlinearity. z = 0, u1 = u2 and G1 @u 1 @u2
@z ¼ G2 @z and at
z = H, G1@u1/@z = 0, and also introducing an
impedance ratio a as
Basic Mechanism on Site Amplification
k1 G1 r1 V s 1
a¼ ¼ , (5)
For a harmonic wave with angular frequency o k2 G2 r2 V s 2
and amplitudes A, B, Eq. 1 is replaced by
the transfer function (2As)/(2A2) is obtained as
iðotkzÞ iðotþkzÞ
u ¼ Ae þ Be (2)
2As 2
¼ (6)
Here, k is the wave number expressed as 2A2 ð1 þ aÞeik 1 H þ ð1 aÞeik1 H
k ¼ o=V s ¼ o=ðG=rÞ1=2 (3) Here, 2As is the wave amplitude to be observed at

the ground surface, and 2A2 is that at the base
Basic mechanisms of site amplification may be layer if it were outcropped and free from the
simplified by a two-layer model as illustrated in overburden. The site amplification to be used in
Fig. 1, where a surface layer with the thickness micro-zonation mapping is determined by Eq. 6.
H overlies a base layer of infinite thickness. Per- There are two types of site response monitor-
tinent properties are: r1, r2 = soil density; Vs1, ing: (a) surface array and (b) downhole array as
Vs2 = S-wave velocity; G1, G2 = shear modu- illustrated in Fig. 2. In the surface array (a),
lus; and D1, D2 = damping ratio of the surface ground motion is monitored at two different
a
Surface array
Needed for Seismic
Seismometer
Zonation
Surf,layer
Transfer function
2As/2A2
Incident wave in
surf. layer As
Incident wave in
base layer A2 Base layer
modification
b
Down-hole array (KiK-net system)
Seismometer
Surf. layer
Transfer function
Incident wave in Incident wave in 2As/(A2+B2)
base layer A2 surf.layer As
Base layer Downward wave in

base layer B2
Site Response: Comparison Between Theory and Observation, Fig. 2 Two types of earthquake observation array
systems to measure site amplification between ground surface and base layer
surface locations with different geologies, over- to derive the transfer function for micro-zonation
lying soft layer and outcropping stiff base layer. (2As)/(2A2), Eq. 7 cannot directly be used, but
If the soft layer is underlain by the same base some modification is necessary to extract the
layer and the upward wave in the base layer A2 is outcropping base motion 2A2 from observed
assumed the same at the two places, the site base motion (A2 + B2).
amplification in the soft soil site with respect to In a two-layer system, the resonant frequency
outcropping base layer is given by Eq. 6. can be computed by the next equation S
In the downhole array (b), surface and
downhole seismometers can evaluate the site
V s1
amplification exactly at the same location, and f ¼ o=2p ¼ ð2n 1Þ (8)
4H
its transfer function can be formulated as
2As 2 where n = the order of resonance. The most

¼ ik H (7)
A 2 þ B2 e 1 þ eik1 H important is the first-order resonant frequency
n = 1, and the equation f1 = Vs1/(4H) is named
Here, B2 is the amplitude of the downward wave as the 1/4 wavelength formula. In many cases,
in the base layer, which is influenced by the the site amplification in real site conditions
dynamic response of the surface layer. In order with multilayer systems can be simplified by a
Site Response: 10
Comparison Between *
8 earthquakes f
Theory and Observation, =
fk
Frequency in spectrum ratio f * (Hz)

Fig. 3 Frequency by ¼
8
wavelength formula fk
compared to peak *
f
2
frequency of observed 1.
6 =
spectrum ratio f* for main fk
shocks of eight strong EQs *
f
.8
fk =0
4
2 Peak : 1st 2nd 3rd

EW–direction :
NS–direction :
0
0 2 4 6 8 10
Frequency by 1/4 wave length formula fk (Hz)
two-layer model consisting of a soft surface layer 7 are compared in the same diagram. It is apparent
underlain by a stiff base layer. In this case, too, from the two equations that the properties of the
the 1/4 wavelength formula can be expressed in base layer are included in (2As)/(2A2) in terms of
an extended form as impedance ratio a but not in 2As/(A2 + B2). When
a = r1Vs1/r2Vs2 = 0 in Eq. 6, indicating r2Vs2 !
" Xj
#
1 (rigid base layer), Eqs. 6 and 7 are identical.
fj ¼ 1 4 ðH i =V si Þ (9) Hence, in the downhole array transfer function, the
i¼1
base layer is equivalent to a rigid base with the
prescribed motion A2 + B2 when no radiation
Here, j is the number of surface layers involved to
damping occurs, resulting in infinite amplification
create a specific resonant frequency. In Fig. 3,
in resonant frequencies if there is no soil damping
frequencies fj obtained from the above equation
in the surface layer, D1=0. In contrast, (2As)/(2A2)
are compared with peak frequencies f in the
produces a certain finite peak values
spectrum ratio of observed motions at a number
corresponding to a even for D1=0, because energy
of downhole array sites in Japan during recent
radiation into the base layer is possible. It is note-
strong earthquakes. The figure shows that the
worthy that in a multilayer system, the base layer
value f are mostly within 0.8–1.2 times fk, indi-
where the downhole seismometer is installed also
cating that the extended 1/4 wavelength formula
serves as the rigid base, whose properties have
Eq. 9 may be useful to roughly estimate the
nothing to do with the downhole array amplifica-
resonant frequency by simplifying actual multi-
tion (Schnabel et al. 1972).
layer soil systems by the two-layer system. Also
note in the diagram that Eq. 9 holds for not only
the first- but also second- and third-order peak
frequencies, which are generated by combina- Soil Damping Models and Site
tions of base layers of different depths and Amplification
corresponding overlying layers.
In Fig. 4, two types of transfer functions of the In order to incorporate soil damping in site ampli-
two-layer system in Fig. 1 calculated by Eqs. 6 and fications, the variables included in Eqs. 6 and 7, k,
Site Response: 2As /2A2: a = 0.1

Comparison Between 2As /2A2: a = 0.3
Theory and Observation,
2As /2A2: a = 0.5
Fig. 4 Comparison to two
types of transfer function D1 = 0%, D2 = 0% 2As /(A 2+ B 2)
2As/2A2 and 2As/(A2 + B2) 15
in a two-layer system
2As / 2A2 or 2As / (A 2 + B 2 )

Transfer functions
10
0
0 2 4 6 8 10
Normalized frequency f/f 1
Site Response: a b c
Comparison Between Kelvin (Voiget) Model Maxwell Model Non-Viscous Model
Fig. 5 Three mechanical
models for soil damping G
G x G iG⬘
Vs, and G, should be replaced by complex wave (time independent) dashpot with an S
number k, complex S-wave velocity Vs, and imaginary constant iG0 . For each model, shear
complex shear modulus G, respectively, so that stress t versus shear strain g relationship is
Eq. 3 is modified as expressed as
k ¼ o=V s ¼ o=ðG =rÞ1=2 (10) t ¼ G g (11)
Three mechanical models for soil damping can be where G is complex shear modulus considering
incorporated in site amplification: (a) Kelvin soil damping. The complex shear modulus G
(Voigt) model, (b) Maxwell model, and normalized by G is formulated for each
ffi damping
pffiffiffiffiffiffi
(c) nonviscous model as illustrated in Fig. 5, models a, b and c using i ¼ 1 as follows
consisting of a spring with shear modulus G and (Ishihara 1996):
a dashpot with viscosity x. For model (c),
the dashpot is replaced by a nonviscous G =G ¼ 1 þ iðx=GÞo (12a)
a 100 b 100
|G */G| Kelvin model tand Kelvin model

= 2D
10 10
Nonviscous model
1 1
Nonviscous
model
0.1 0.1
Maxwell model Maxwell model
0.01 0.01
0.01 0.1 1 10 100 0.01 0.1 1 10 100
Frequency (x/G)w Frequency (x/G)w
Site Response: Comparison Between Theory and Observation, Fig. 6 Variations of normalized shear modulus |
G*/G| and tan d along with normalized angular frequency (x/G)o
io From Eq. 2, the equation for a wave including soil

G =G ¼ (12b)
1=ðx=GÞ þ io damping propagating unidirectionally toward + z
is written as
G =G ¼ 1 þ iG0 =G (12c)

u ¼ Aeiðotk zÞ
(16)
In these equations, the normalized complex shear
modulus can be expressed as If k in this equation is substituted by Eq. 10, and
G in Eq. 10 is further substituted by Eqs. 12a, b,
G =G ¼ jG =Gjeid (13) and c, Eq. 16 can be transformed as follows:
where d is the phase lag angle between u ¼ Aebz eioðtz=V s Þ (17)

stress–strain relationships of the models, and tan
d is called a loss coefficient. In Fig. 6, the varia- Here, b is a positive real number called a wave
tions of normalized complex shear modulus |G/ attenuation coefficient by internal damping,
G| and tan d are shown versus dimensionless because it determines how the wave attenuates
angular frequency (x/G)o for models (a) and as it propagates by distance z. If delta is small,
(b), and constant values are shown for model beta is correlated to the damping ratio D as
(c). The damping ratio is correlated with the
loss coefficient as follows (Ishihara 1996): b ¼ oD=V s (18)
D ¼ ð tan dÞ=2 (14) Substituting D in Eq. 15a, b and c into Eq. 18

gives b for the corresponding damping models:
Hence, from Eqs. 12a, b, and c:

b ¼ o2 x= 2rV s 3 (19a)
D ¼ tan d=2 ¼ ox=ð2GÞ (15a)
D ¼ tan d=2 ¼ G=ð2oxÞ (15b) b ¼ rV s =ð2xÞ (19b)

D ¼ tan d=2 ¼ G0 =ð2GÞ (15c) b ¼ oG0 = 2rV s 3 (19c)
a 15 b 15
Down-hole array D1=5%, D2=0%
Transfer function for surface array
Surface array D 1=5%, D 2=0%
Transfer function for vertical array

a = r1Vs1/r2Vs2 = 0.1
10 Kelvin model 10
2As /(A b + B b )
Non-viscous model
2A s /2A b
Maxwell model
5 5
0 0
0 2 4 6 8 10 0 2 4 6 8 10
Normalized frequency f/f1 Normalized frequency f/f1
Site Response: Comparison Between Theory and 2Ab (a = 0.1) for surface array and (b) 2As/(Ab + Bb) for
Observation, Fig. 7 Transfer functions of a two-layer downhole array
model using three types of soil damping models: (a) 2As/
In Fig. 7, transfer functions of the two-layer records at six downhole array sites where PGA
model in Fig. 1 are shown for the three damping exceeds 0.2 g during strong earthquakes are
models assuming the damping ratios in the surface depicted with thin lines in EW and NS directions.
and base layers D1 = 5 % and D2 = 0 %, respec- For each site, the soil profile with density r and
tively. Figure 7a is for the surface array (2As)/(2Ab) S-wave velocity Vs is tabulated in Table 1
with the impedance ratio a = (r1Vs1)/(r2Vs2) = 0.1, together with installation levels of seismometers
and Fig. 7b is for the downhole array 2As/(Ab + Bb). indicated with arrows. Using the properties in the
The transfer functions are identical in the first peak table, transfer functions for each site, 2As/(Ab +
but obviously differ in higher-order peaks corres- Bb) for the downhole array and (2As)/(2Ab) for
ponding to the different damping models. In the the surface array, are calculated and superposed
Kelvin model, the amplification in higher-order in the diagram with thick curves, where Ab and
peaks tend to reduce more drastically than in the Bb = amplitudes of upward and downward
nonviscous model, and in the Maxwell model, it is waves in the base layer. In the calculation, the
unchanged at all peaks. These differences arise nonviscous damping model is used as normal
from the different formulations of b in Eqs. 19a, b, engineering practice, and a uniform damping
and c with respect to o. In cyclic loading tests on ratio D = 2.5 % is assumed tentatively for all S
soil elements, it is widely accepted that soil layers. If the observed spectrum ratios are com-
damping is almost frequency independent (Hardin pared with corresponding transfer functions for
1965; Hardin and Drnevich 1972). Hence, in most downhole arrays, 2As/(Ab + Bb), a fairly well cor-
engineering practice, the soil damping is assumed respondence in peak frequencies can be recog-
nonviscous as in Eq. 15c. In this case, the peak nized between the two in most sites despite some
amplification of the transfer function becomes minor effect of soil nonlinearity. This indicates
lower for higher-order peaks because the wave the applicability of one-dimensional soil models
attenuation coefficient b is proportional to o as in these sites to a certain extent.
shown in Eq. 19c. If 2As/(Ab + Bb) is compared with (2As)/(2Ab) at
In Figs. 8a–e, some typical Fourier spectrum each site, the difference in the peak values between
ratios calculated between surface and base the two transfer functions is obviously large.
a EQ.3 (TKCH08) c EQ.3 (NMRH02) e EQ.5 (FKOH03)

Thickness Depth Density Vs Thickness Depth Density Vs Thickness Depth Density Vs
Seismometer
Layer No. Layer No. Layer No.
H[m] [m] [t/m2] [m/s] H[m] [m] [t/m2] [m/s] H[m] [m] [t/m2] [m/s]
1 4 4 2.1 130 1 4 4 2 110 1 6 6 2.1 250
2 32 36 2.1 480 2 6 10 2.1 320 2 8 14 2.2 440
3 42 78 2.1 590 3 30 40 2.1 500 3 11 25 2.3 700
4 22 100 2.6 2800 4 8 48 1.6 430 4 75 100 2.6 2030
5 34 82 2.1 510
6 12 94 2.3 650
7 9 103 2.4 870
20 2A s /(Ab+Bb)
20 40
2A s /2Ab
Observed S. R (EW)
Transfer functions
16 32
Transfer functions
Transfer functions
Observed S. R (NS) 16
12 24
12
8 16
8
4 8
4
0 0
0 4 8 12 16 20 0
0 4 8 12 16 20 0 4 8 12 16 20
Frequency f (Hz) Frequency f (Hz) Frequency f (Hz)
b EQ.3 (TKCH11) d EQ.3 (TKCH04) f EQ.5 (SAGH03)

Thickness Depth Density Vs Thickness Depth Density Vs Thickness Depth Density Vs
Layer No.
H[m] [m] [t/m2] [m/s] Layer No. H[m] [m] [t/m2] [m/s] Layer No. H[m] [m] [t/m2] [m/s]
1 1 1 1.8 150
Seismometer
1 4 4 2.1 380 1 4 4 2 160

2 11 12 21 470 2 16 20 2.1 380 2 10 14 2.1 380
3 10 22 2.1 470 3 18 38 2.1 680 3 20 34 2.4 900
4 23 45 2.2 570 4 46 84 2.1 890 4 69 103 2.6 1980
5 6 51 2.3 730 5 88 172 2.1 1020
6 13 64 2.4 970
7 14 78 2.5 1500
8 22 100 2.6 2680
20 20 40
2A s /(Ab+Bb)
Transfer functions
16 32
Transfer functions
Transfer functions
16 2A s /2Ab
Observed S. R (EW)
12 12 Observed S. R (NS) 24
8 8 16
4 4 8
0 0 0
0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20
Frequency f (Hz) Frequency f (Hz) Frequency f (Hz)
Site Response: Comparison Between Theory and observed spectrum ratios at six downhole array sites dur-
Observation, Fig. 8 Transfer functions of a multilayer ing a strong earthquake
system 2As/(Ab + Bb) compared with 2As/2Ab and
However, the coincidence in peak frequencies is (f), though the Vs-value at the base layer is much
almost perfect in (a) and good in (b) but gets larger than the upper layers, the depth of seis-
poorer in (c) and (d) and very poor in (e) and mometer is too deep from the upper boundary of
(f). The reason may be gleaned by examining the clear Vs-contrast to properly detect the response
soil profiles. In (a) and (b), the Vs-value at the of the upper layers. This observation tells us
downhole seismometer is much larger than a significance of choosing appropriate seismom-
the upper layers, and the seismometer depth is eter depth in deploying a downhole array system
not so deep from the boundary of clear considering site specific soil profiles.
Vs-contrast. In (c) and (d), the Vs-value at the Another observation may be made on amplifi-
base layer is not so different from the upper cation values at individual peaks of observed
layers, and the seismometer depth is not so deep spectrum ratios compared with the downhole
from a boundary of major Vs-contrast. In (e) and array transfer functions 2As/(Ab + Bb).
Site Response: Comparison Between Theory and Observation, Table 1 Soil profiles and properties at two
vertical array sites corresponding to spectrum ratios in Figs. 9 and 10
For Fig. 9
IBUH03 Soil profiles and properties
Layer no. Thickness h (m) Depth H (m) Density (t/m3) Vs (m/s)
1 2 2 21 60
2 16 18 21 90
3 10 28 20 190
4 12 40 21 320
5 12 52 21 210
6 24 76 21 310
7 56 132 21 430
8 21 153 22 520
For Fig. 10
NMRH02 Soil profiles and properties
Layer no. Thickness h (m) Depth H (m) Density (t/m3) Vs (m/s)
1 4 4 2.0 110
2 6 10 2.1 320
3 30 40 2.1 500
4 8 48 1.6 430
5 34 82 2.1 510
6 12 94 2.3 650
7 9 103 2.4 870
As already mentioned, the damping ratio is ten- One major reason for this deviation may be
tatively set as 2.5 % in calculating 2As/(Ab + Bb). wave scattering in heterogeneous soils. The wave
At a glance, the peak amplification values of the scattering effect on frequency dependency of
transfer functions are almost monotonically damping in elastic waves propagating in heteroge-
decreasing with increasing peak frequencies, pre- neous earth crust is an important topic in seismol-
sumably reflecting the damping characteristics of ogy (e.g., Frankel and Clayton 1986; Wu 1982).
the non-viscous model used here, as previously The effect tends to be more conspicuous with
explained. In contrast, the peak values of decreasing strain in a stiff rock site. The damping
observed spectrum ratios seem to be quite site ratio D by wave scattering may be expressed as
specific and show no consistent increasing or
decreasing trend. This indicates that the soil D ¼ Dr ðf =f r Þm (20) S
damping is not perfectly nonviscous (frequency
independent) but to some extent frequency
dependent (the damping ratio D decreases with where fr = a reference frequency, Dr = refer-
increasing frequency f) in a similar manner as the ence damping ratio, and D = Dr at f = fr. The
Maxwell model. It is clear that the Kelvin model, power constant m is a positive number, and
for which the peak amplification tends to decrease m = 1 corresponds to the Maxwell model, as
more drastically with increasing f, has even less obviously seen in Eq. 15b. In general, the
applicability than the nonviscous model. Thus, damping ratio may be expressed as the sum of
the frequency-independent damping found in lab- the frequency-independent hysteretic part D0 and
oratory soil tests may not perfectly fit the actual the frequency-dependent part as
performance of site amplification based on earth-
quake observations. D ¼ D0 þ Dr ðf =f r Þm (21)
a EW-direction: IBUH03 b NS-direction: IBUH03

30 20
mainshock
Aftershock 1
25
Spectrum ratio 2As /(Ab+ Bb)

Aftershock 2
Aftershock 3 15
Aftershock 4
20
15 10
10
5
5
0 0
0 2 4 6 8 10 0 2 4 6 8 10
Frequency f (Hz) Frequency f (Hz)
Site Response: Comparison Between Theory and Observation, Fig. 9 Fourier spectrum ratios 2As/(Ab + Bb)
observed at a very soft soil site during main shock and aftershocks in directions EW (a) and NS (b)
However, the frequency-dependency of damping those for weaker aftershocks in directions EW

is normally ignored in geotechnical engineering (a) and NS (b) using earthquake records obtained
practice, because hysteretic nature of soil in the same downhole arrays. As indicated in
damping becomes more dominant during strong Table 1, the soil condition for the former site is
earthquakes as soil gets softer and strain- very soft (Vs < 100 m/s at top 18 m), while that
dependent nonlinearity becomes more dominant. of the latter site are relatively stiff (Vs > 320 m/s
deeper than 4 m). Thick curves for the main shock
are obviously different with thin curves for the
Soil Nonlinearity aftershocks with respect to peak values and peak
frequencies. The difference of peak values is
Seismically induced shear strain g is calculated greater in Fig. 9 than in Fig. 10 because of the
_
from particle velocity u: softer soil conditions, and the difference in the
same site tends to be greater in higher-order
_ s
g ¼ u=V (22) peaks than in the first peak because softer layers
near the ground surface tends to contribute more.
if upward propagating wave is chosen in Eq. 1, Basic effects of soil nonlinearity on site ampli-
for instance. This indicates that the shear strain fication can be examined using the two-layer
tends to be greater for strong earthquakes with system shown in Fig. 1. Three levels of induced
larger particle velocity amplitudes and in soft equivalent strain amplitude are assumed, and
soils with smaller Vs-values. Because corresponding shear modulus degradation G/G0
stress–strain behavior of soil is nonlinear and and equivalent damping ratio D (nonviscous) are
hence Vs and D are highly strain dependent, the determined based on empirical curves (Fig. 11)
site response during strong earthquakes may often used in engineering practice (Seed and
greatly differ from that during small earthquakes, Idriss 1970). Figures 12a and b depict the transfer
particularly in soft soil sites. The soil nonlinearity functions 2As/(Ab + Bb) for the downhole array
effect is normally evaluated using the equivalent and (2As)/(2Ab) for the surface array calculated
linear approximation method in engineering for the three levels of nonlinear soil properties.
practice (Schnabel et al. 1972). Obviously, soil nonlinear properties have
In each of Figs. 9 and 10, Fourier spectrum great effects on the peak frequencies and the
ratios for a strong main shock are compared with peak amplifications not only in 2As/(Ab + Bb) as
a EW-direction: NMRH02 b NS-direction: NMRH02

60 50
mainshock
50

Aftershock 1 40
Aftershock 2
Aftershock 3
40 Aftershock 4
30
30
20
20
10
10
0 0
0 2 4 6 8 10 0 2 4 6 8 10
Frequency f (Hz) Frequency f (Hz)
Site Response: Comparison Between Theory and Observation, Fig. 10 Fourier spectrum ratios 2As/(Ab + Bb)
observed at a stiff soil site during main shock and aftershocks in directions EW (a) and NS (b)
γ 5 ¥ 10−6 10−4 10−3 2As/(Ab + Bb) as indicated in Eq. 7. Under the

G/G0 1.0 0.65 0.25 paramount effect of radiation damping associated
D 0.02 0.05 0.15
with a, the difference in the amplification (2As)/
(2Ab) due to strain-dependent nonlinear proper-
ties becomes less conspicuous. Furthermore, the
1.0 25
impedance ratio, a = r1Vs1/r2Vs2, which
Shear modulus ratio G/G0
becomes smaller with degraded modulus or

Damping ratio D (%)
0.8 20
degraded S-wave velocity Vs1 in the surface
0.6 G/G0 15 layer, tends to give larger amplification compen-
D sating the effect of increased damping ratio in the
0.4 Seed-Idriss (1970) 10
Sand surface layer during strong earthquakes. Thus,
0.2 5 the difference in soil nonlinearity between the
main shock and aftershocks seems to have
0.0 0 smaller influence on the amplification in (2As)/
1E–6 1E–5 1E–4 1E–3 0.01 0.1 (2Ab) than in 2As/(Ab + Bb) as indicated from the
Normalized effective shear strain γ comparison of Fig. 12a, b. S
Site Response: Comparison Between Theory and
The Fourier spectrum ratio corresponding to
Observation, Fig. 11 Soil nonlinearity curves (G/G0- the transfer function (2As)/(2Ab) cannot be
g, D-g) used in calculating amplifications in a two-layer obtained directly from downhole array records,
system but the peak value can be calculated from
2As/(Ab + Bb). Figure 13 shows an example how
recognized in actual site amplifications in Figs. 9 this calculation is carried out. First, a transfer
and 10 but also in (2As)/(2Ab). However, the function 2As/(Ab + Bb) is calculated at
difference in the peak amplifications due to strain a downhole array site based on the multi-
level is less pronounced in (2As)/(2Ab) than in reflection theory. Among soil properties needed,
2As/(Ab + Bb) for the first peak in particular. It the S-wave velocities of individual layers are
is because the radiation damping effect presented given by in situ logging tests if strain-dependent
by the impedance ratio a affects (2As)/(2Ab) in soil nonlinearity is not so significant, and soil
Eq. 6, whereas no effect of a is involved in densities are judged from previous experiences.
Site Response: a 40
Comparison Between Down-hole array
Fig. 12 Transfer functions
30
Transfer function
of a two-layer system (a) D1 = 2.5%
2As /(Ab + Bb )
2As/(Ab + Bb) for
downhole array and (b) D1 = 5%
2As/2Ab for surface array in 20
D1 = 10%
different induced strain
levels
10
0
0 1 2 3 4
Normalized frequency f/f1
b 4
Surface array
3
Transfer function
2As /2Ab
0
0 1 2 3 4
Normalized frequency f/f1
The damping ratio D is tentatively assumed 2.5 % compared in this manner if possible, and the
in all layers and also postulated to be nonviscous values of D in the two directions, EW and NS,
or frequency independent. are averaged for individual peaks. Then, the
Then, the calculated transfer function transfer function (2As)/(2Ab) is computed using
2As/(Ab + Bb) is compared with the the modified damping ratio D based on the same
corresponding spectrum ratio obtained from multilayer system. In this way, the peak amplifi-
downhole array records. If a peak in the transfer cations for (2As)/(2Ab) between surface and base,
function can be found at about the same fre- to be used in seismic zonation study for surface
quency in the spectrum ratio of recorded motions, soil resting on the common base layer, can be
it is identified as the corresponding peak, and the obtained using the downhole array records.
damping ratio assumed as D = 2.5 % previously In Fig. 14, the peak amplification values in the
is modified to have the same peak value, by using first peak of the spectrum ratios are compared
the equation D = Q1/Q2 2.5(%), where Q1 is between small aftershocks (in the horizontal
the peak value of the calculated transfer function axis) where shaking is relatively smaller
and Q2 is that of spectrum ratio based on the (PGA 0.1 g) and main shock (in the vertical
actual records as indicated in Fig. 13. Not only axis) with PGA 0.1 ~ 2.4 g, based on vertical
the first peak but also the higher-order peaks are array records during strong earthquakes and
Site Response:
Observed spectrum ratio Transfer function
Comparison Between
20 EW NS 2As /(A b + B b)
Fig. 13 How to obtain
Q1 Q2
spectrum peak 2As/2Ab Transfer function
EW
from 2As/(Ab + Bb) of 16 2A s / 2A b
Amplification (surface/base)
vertical array observed
motions at the same site Q2 Q2
EW NS Q1
12
8
Q2
NS
0
0 2 4 6 8 10
Frequency (Hz)
Peak value of (2A s ) / (2A b)
Site Response: 60
Comparison Between k
oc
Theory and Observation, rsh
fte
Peak amplification : Main shock
Fig. 14 Comparison of 50 2A s /2A b A

spectrum peak k=
2A s /(A b+ B b) oc
amplifications between sh
40 in
aftershocks and main
Ma
shocks of downhole arrays
2As/(Ab + Bb) and surface
arrays in 2As/2Ab 30
20
10 1st peaks only,

Non-viscous damping
0 S
0 10 20 30 40 50 60
Peak amplification value: Aftershock
associated aftershocks recorded at many vertical possibly reduces due to soil nonlinearity. In con-
array sites in Japan (Kokusho and Sato 2008; trast, the solid circular symbols corresponding to
Kokusho 2013). The triangular symbols for (2As)/(2Ab) in surface arrays, though the amplifi-
downhole arrays 2As/(Ab + Bb) are dispersed in cation values are small, concentrate near the diag-
a wide range, and the majority is plotted around onal line, indicating that the effect of soil
or below the diagonal line (main shock = after- nonlinearity is less dominant in the first peak
shock), indicating that during strong shaking the site amplification, as demonstrated in Fig. 12 by
site amplification in terms of 2As/(Ab + Bb) using a simplified two-layer model.
Thus, the soil nonlinearity affects the site 6. Site amplification observations show that soil
amplification considerably with respect to peak nonlinearity affects the site amplification con-
frequency and peak amplification. However, for siderably with respect to peak frequency and
the amplification of the first peaks in the surface peak amplification. However, for the amplifi-
array, the nonlinear effect in obviously minor. cation of the first peak in the surface array, the
This may simplify the seismic zonation proce- nonlinear effect is obviously minor, which
dure considering strong shaking. may simplify the seismic zonation procedure
even for strong earthquakes.
Summary
Cross-References
1. Two types of site amplification monitoring
can be defined and implemented in earthquake
▶ Downhole Seismometers
observations, surface array and downhole
▶ Dynamic Soil Properties: In Situ
array. For micro-zonation mapping, the ampli-
Characterization Using Penetration Tests
fication for the surface array is needed.
▶ Site Response: 1-D Time Domain Analyses
2. For reproduction of site amplification
observed in the field, it is possible to simplify
a complicated multilayer system to a two-layer
system composed of a surface layer of References
a certain thickness underlain by a base layer
Frankel A, Clayton RW (1986) Finite difference simula-
of infinite thickness. Resonant frequencies of tions of seismic scattering: implications for the propa-
the multilayer system can roughly be evalu- gation of short-period seismic waves in the crust and
ated using the 1/4 wavelength formula. models of crustal heterogeneity. J Geophys Res
3. In the downhole array, the installation level of 91(B6):6465–6489
Hardin BO (1965) The nature of damping in sands. Proc
the downhole seismometer serves as a rigid Soil Mech Found Div, ASCE 91(SM1):63–97
boundary, and soil properties below have Hardin BO, Drnevich VP (1972) Shear modulus and
nothing to do with the amplification of the damping in soils: measurement and parameter effects.
ground surface with respect to the base J Soil Mech Found Div, ASCE 98(SM6):603–624
Ishihara K (1996) The representation of stress–strain
motion. relations in cyclic loading, Chapter 3. In: Ishihara K -
4. Among three soil damping models, Kelvin, (ed) Soil behaviour in earthquake geotechnics, Oxford
Maxwell, and nonviscous, the nonviscous engineering science series, 46. Clarendon, Oxford,
model is normally used for site amplification pp 16–39
Kanai K, Tanaka T, Yoshizawa S (1959) Comparative
evaluation in engineering practice. In this studies of earthquake motions on the ground and
case, the peak amplification of the transfer underground. Bulletin of the Earthquake Research
function becomes lower for higher-frequency Institute, Tokyo University, Vol. 37, 53–87.
peaks because the wave attenuation coeffi- Kokusho T (2013) Site amplification formula using aver-
age vs in equivalent surface layer based on vertical
cient b is proportional to the frequency. array strong motion records. ICEGE, Istanbul
5. There are evidences from earthquake observa- Kokusho T, Sato K (2008) Surface-to-base amplification
tions to suspect that some degree of evaluated from KiK-net vertical array strong motion
frequency-dependent damping similar to records. Soil Dyn Earthq Eng 28:707–716
Schnabel PB, Lysmer J, Seed HB (1972) SHAKE –
Maxwell damping due to a wave-scattering a computer program for earthquake response analysis
mechanism exists in the field. However, it is of horizontally layered sites, Report EERC 72-12.
normally ignored in geotechnical engineering, University of California, Berkeley
because hysteretic (nonviscous) nature of soil Seed HB, Idriss IM (1970) Soil Moduli and damping
factors for dynamic response analysis, Report EERC
damping becomes more dominant during 70-10. University of California, Berkeley
strong earthquakes as soil gets softer and Wu R-S (1982) Attenuation of short period seismic waves
more strain-dependent nonlinear. due to scattering. Geophys Res Lett 9(1):9–12
Social Media Benefits and Risks in Earthquake Events 3309
immediately after an earthquake (Young

Social Media Benefits and Risks in et al. 2013) for both first responders and survi-
Earthquake Events vors. In fact, social media, in particular Twitter,
may be the only immediate source of data from
Hamish McLean locations with limited sensors or other scientific
School of Humanities, Griffith University, instruments. Obviously, not all earthquakes are
Nathan, QLD, Australia reported on social media as many events are in
remote areas or undersea and in countries with
limited social media access or the magnitude is
Synonyms too small to be felt. A unique benefit of social
media is that it is user generated – disaster agen-
Citizen seismologists; Disaster communication cies, seismologists, and other parties do not have
analysis; Earthquake education; Enhanced situa- to motivate citizens to tweet – they will do it
tional awareness; Post-disaster recovery; Psycho- anyway, potentially by the thousands in a signif-
logical first aid; Real-time warnings on Twitter icant earthquake. The challenge is how to trans-
form the rapidly spreading flood of real-time
information, some of it inaccurate, into reliable,
Introduction useful, and valuable data. Part of the solution is to
train Twitter users to tweet messages that can be
The rapidly evolving social media platforms, more easily analyzed both manually and automat-
with an estimated 1.9 billion users worldwide, ically using a crisis-specific syntax (Starbird and
offer a myriad of communication benefits and Stamberger 2010). This approach, among others
risks in the context of a disaster. Social media discussed in this entry, will help guide disaster
generally refers to internet-based technologies response, galvanize ongoing humanitarian efforts,
that enable people to interact and share resources and add value to the expanding body of earthquake
and information using either text or multimedia sciences gathered since the development of the
applications (Lindsay 2011; Dabner 2011). modern seismograph in the late 1800s.
Advances in mobile devices allow access to any-
one who has the ability to connect online (Abbasi
et al. 2012). For example, the microblogging Sourcing Information in Disasters
platform Twitter allows followers to track what
an account holder is doing and thinking in real People facing a disaster seldom act on one source
time within the confines of 140 characters (Kaigo of information. Hunting for firsthand local infor-
2012). Tweets can be sent from a variety of plat- mation, they will “channel swap” across the tra-
forms ranging from cell phones to computers. ditional media, go online to news websites and S
Other examples of social media platforms today social media, and contact family and friends.
are Facebook, YouTube, Qzone, Pinterest, Without proof from a variety of trusted sources,
Instagram, and Flickr. This entry describes how people will believe there is no immediate threat
social media, in particular Twitter, can be used or that the situation does not apply to them. The
for a variety of applications before, during, and delay in taking action – known as milling – can
after an earthquake, both inside and outside the increase the risk of death or injury from a hazard
area of impact. These include detection, warn- before steps are taken for self-protection. The
ings, connecting to survivors, situational aware- milling effect is increased if warnings are vague
ness, notifying responders of where help is and conflicting across various channels or the
needed, and galvanizing humanitarian aid. credibility of the source is in doubt. Under these
Importantly, the increasing participation of “citi- circumstances, social media becomes a double-
zen seismologists” via social media is filling the edged sword. The speed of social media can
information gap with field observations reduce the decision-making lag time while, at
3310 Social Media Benefits and Risks in Earthquake Events
the same time, increasing confusion and uncer- disaster agencies is how social media data can be
tainty if the information is wrong or conflicting. It effectively incorporated into time-critical
is generally accepted that trust in social media decision-making processes. As Tapia et al.
information remains well below that of tradi- (2013, p. 770) note, “while data quality continues
tional media. The 2014 Edelman Trust Barometer to be a barrier, what is far more important to
found that 47 % of people trusted social media organizational use is the serving of this data at
sources, while 65 % trusted the traditional media. the appropriate time, in the appropriate form to
In practice this was reflected in the 2011 Japan the appropriate person and the appropriate level
earthquake. Although most people turned to of confidence.”
social media as their most reliable source of
information, one third described the “lack of
trust in formation as the greatest problem associ- Online Support on the Ground
ated with social media use in the disaster” (Perry
et al. 2012, p. 15). For disaster agencies, the Driven by developing mobile technology, social
challenge of earning trust and building credibility media is taking on an increasing role in
and relationships can be overcome by actively connecting people in disasters. Part of the reason
engaging online before the disaster. As noted by is that people reach out to both their “offline” and
the US-based Centers for Disease Control and “online” communities during a crisis or disaster
Prevention (CDC): “Organisations need to be (Dutta-Bergman 2006). In other words, they par-
regular users of social media before the crisis. allel their physical world with their virtual world
If not, social media users will go to other sources to garner “social support and gather information,
and groups with whom they already have rela- and vice versa.” In turn, on a much wider scale,
tionships for information” (2012, p. 268). Engag- this online convergence builds and strengthens
ing with social media before a disaster can also community resilience through “people power”
build preparedness and education for survival. (Duffy 2012). A bank of social capital is devel-
For example, more than 1.3 million people in oped by the exchange of information “during
New Zealand took part in the “ShakeOut” earth- difficult times” to build relationships between
quake drill in 2012, driven by a social media people (Kaigo 2012). Therefore, social media in
campaign. Many organizations took the drill real- disaster impact areas is frequently driven by the
istically. For example, a hospital in the country’s community to share knowledge and as a form of
North Island shut off its power and water for 24 h empowerment toward recovery. The CDC (2012)
to simulate the impact of an earthquake. Portable notes that the public uses social media on a
cooking, lighting, and other equipment greater scale in the hours after a disaster than
maintained hospital services (Civil Defence official agencies. Williams et al. (2012), who
Emergency Management 2012). Similar events have produced a practical guide to community-
are held in other countries to prepare communi- based social media in disasters based on lessons
ties for an earthquake. Although disaster agencies from a series of tornadoes and floods in the
are using social media to educate communities United States, found that post-disaster social
and disseminate information, they have been his- media was generated and driven by citizens
torically more reluctant to rely on social media rather than emergency agencies. In most cases,
data in their incident command systems. The community-managed social media was the pri-
reasons include fears about misinformation mary source of information for those impacted by
(Williams et al. 2012), the speed and spread of the disaster. For example, a University of Mis-
the information which makes validation difficult souri Extension Facebook site – Branson Tor-
(Gowing 2009), and a lack of understanding nado Info – attracted 14,000 followers within
about how they can make use of social media 12 h of a tornado in February 2012. One victim
(Duffy 2012; Palen 2008). For example, posted: “For the first few days after the storm, this
Tapia et al. (2013) argue that a major issue for Facebook page was our main source of
Social Media Benefits and Risks in Earthquake the level of response and resources that may be
Events, Table 1 The expanding role of social media in required. Earthquake alerts are now within the
earthquake events
scope of Twitter. Crooks et al. (2013) found that
2008 Sichuan The first alert of the severity of the velocity of Twitter can be used as a warning
(China) the earthquake was reported on
earthquake Twitter (Moore 2008; Cellan- system in large-scale events. Tweets and
Jones 2008) re-tweets spread from the epicenter faster than
2010 Haiti First major earthquake in the age the physical effects to distant locations. For
earthquake of social media. The purpose- example, tweets about the 2011 Virginia earth-
built Ushahidi crisis map became quake were read in New York 30 s before it was
an emergency reporting system to
locate people texting for help. felt there, “showing that information moves
Social media-generated faster through networks than the earthquakes
significant humanitarian aid themselves” (Perry et al. 2012, p. 6). Automatic
(Meier 2012; MacLeod 2010) earthquake warnings sourced from Twitter data
2011 Christchurch Social media became “the Church
are being developed using algorithms to analyze
(NZ) earthquake or meeting hall” for people to
support each other and share keywords, the number of words, and their context
information (Mathewson 2012) (Sakaki et al. 2010). Importantly, Tweets can
2011 Japan Social media platforms, include locations, which is fundamental in sens-
earthquake and particularly Facebook, Twitter, ing earthquake events (Sakaki et al. 2010).
tsunami and Japan’s own social media site
Mixi, provided information and
Another real-time online method of detecting
connected families when earthquakes in a general location is to trace the
telephone communication was IP addresses of visitors to earthquake information
damaged or became congested websites. For example, surges in visitor traffic to
(Wallop 2011)
the European-Mediterranean Seismological Cen-
2011 Turkey Facebook was used to coordinate
earthquake donations and aid requests. tre from people wanting information about what
People trapped in rubble tweeted they had felt can provide a snapshot within 2 min
for help (Turgut 2011) into the location and potential damage from an
2013 Ya’an Social media platforms become a earthquake (Bossu et al. 2011).
(China) place for mourning when people
earthquake turned their profile photographs
to gray in remembrance of the
victims (Hui 2013) Greater Situational Awareness
Tweets within the first minutes of an earthquake

information. Volunteers here answered our ques- are mostly generated from around the epicenter
tions about where to go to get help, what and provide potentially useful situational aware-
resources were available and what we needed to ness for both emergency responders, seismolo- S
do next” (Williams et al. 2012, p. 18). In the gists, and, importantly, survivors. In the 2011
context of earthquakes, social media has played Christchurch, New Zealand, earthquake, the first
an important and increasingly sophisticated role, tweet was within 30 s, the first photo was within
as the following Table 1 demonstrates: 4 min, and the first video was uploaded to
YouTube in 40 min. It is widely accepted that
information such as this supplements, rather than
Twitter Faster than Earthquakes replaces, data sourced from scientific instru-
ments. As Earle et al. (2011, p 709) note: “The
It is well documented that the real-time speed and qualitative descriptions contained in the tweets
user-input microblogging capability of Twitter are available at the same time as the seismically-
make it one of the most useful social media derived earthquake parameters and sometimes
tools for disaster management agencies to gain a provide a responding seismologist with a quick
rapid snapshot of the earthquake aftermath and indication of the severity of the earthquake
effects.” Although Twitter’s 140 characters may Social Media Benefits and Risks in Earthquake
be limiting, the narratives often produce a con- Events, Table 2 The benefits of various social media
platforms for communicating during a disaster
sensus on the intensity of the earthquake because
“citizens tend to report very similar experi- Twitter Instant messaging
(140 characters) One-to-many receivers (followers)
ences” (Young et al. 2013, p. 19). Hashtags are Monitoring first impressions of the
quickly formulated to spread information. For shaking in real time
example, within minutes of the 2011 Christ- Issuing warnings and alerts rapidly
church, New Zealand, earthquake, the hashtags Integrating mass or interpersonal
communication
#eqnz and #chch helped to share images and Initiating situational awareness from
videos of the damage (Edmond 2013). Within the field
2 min of a tremor in Victoria, Australia, more Channels to dispel rumors and
than 100 tweets were posted, giving an indica- correct information
Interactive mapping
tion of strength and reports of minor damage Rapid updating of traditional media
(Anderson 2012). In the 2011 Japan earthquake, Alternative method of seeking aid
Twitter was more effective in providing infor- for survivors
mation in devastated areas than traditional Linking to more detailed
information on other platforms
media and websites (Kaigo 2012). It has led the Ability to “snowball” information
Japanese authorities to consider making social by re-tweeting
media networks part of the country’s emergency YouTube (and Broadcasting live vision from a
call system (Dugan 2012). Scanning other social others) location
media platforms can also strengthen situational Providing a channel to group videos
Upload vision/audio for traditional
awareness. There are a myriad of social media and online media consumption
platforms that can convey video, photographs, Facilitating updates
audio, and written accounts of the earthquake Facebook Offers two-way communication
event. For example, the following chart outlines Connecting people inside and
outside the disaster area
key benefits of Twitter, YouTube, Facebook,
Providing more information than
and blogs (Table 2). Twitter
Immediate updates
Delivering vision/audio/images
Linking to other agencies and
Social Media Clutter, Misinformation, sources of information
and Rumors Accessible to anyone with an email
address
Social media is flooded with information in a Blogs Rapid updates
disaster event. For example, an estimated 2.3 Allows discussion
Space for more information
million tweets mentioned Haiti or the Red Cross Linking to other agencies and
in the 24 h following the 2010 Haiti earthquake. sources of information
In Japan, more than 2000 tweets were posted Providing opportunities to share
every second of the day following the 2011 earth- stories and experiences
quake and tsunami (Meier 2013). Significant
numbers of tweets are re-tweeted, adding to the
social media “clutter” and potential spread of and volunteers, such as micromappers.com, to
misinformation in the aftermath of a disaster. To quickly filter social media data during a disaster
make sense of the situation and to gather reliable using apps to tag tweets and photographs.
data, a number of social mapping and analysis Advances in technology are providing dividends.
projects are under development. They include Another project, the Artificial Intelligence for
machine-learning approaches for classifying and Disaster Response, reports that 40–80 % of
extracting “informative” Twitter messages to tweets containing disaster information can be
augment situational awareness (Imran et al. 2013) detected automatically, with an 80–90 %
accuracy rate on whether the tweet was from an sense and coming to terms with these events,
eyewitness. To obtain more structured firsthand even while they were still unfolding.”
accounts from social media users, crowdsourcing
approaches are utilized by dedicated not-for-
profit and government-based earthquake Psychological Support for Survivors
reporting and information sites. For example,
the US Geological Survey actively seeks contri- Social media plays a key role in the earthquake
butions through its “Did You Feel It” project by recovery phase. Unlike the limitations of the
“asking people where they were, what they one-way traditional media model, social media’s
observed and what they experienced during the two-way interaction helps rebuild communities
earthquake” (Young et al. 2013, p. 2). and bring together families, friends, and neigh-
Earthquake-Report.com, which describes itself bors. Importantly, social media provides “psy-
as the “best independent earthquake reporting chological first aid” where people “reported
site in the world,” utilizes multiple social media feeling a sense of connectedness and usefulness,
platforms to share firsthand accounts of earth- felt supported by others and felt encouraged by
quakes in real time along with merging data the help and support being given to people”
from scientific sources. Crowdsourcing questions (Taylor et al. 2012, p. 25). In the weeks and
from the impact area include location, scale of months following the 2011 Christchurch earth-
intensity, and a brief description of the experi- quake, Dabner (2011) found that online discus-
ence, including damage. sion provided support and information, with one
participant describing it as a lifeline “that helped
her (and therefore her children) cope with after-
Squashing Viral Rumors and False shocks by realizing normally (sic) would eventu-
Information ally return” (2011, p. 10). Due to the level of
destruction, the role of churches was resumed
People power – termed the Wikipedia effect – is a through social media, with one researcher
self-correcting social media. Many of the rumors observing: “Social media was really a way for
and misinformation in a disaster are identified people to feel like they weren’t being forgotten or
and corrected by social media users. Traditional like they were part of a larger community. As far
journalists, unable to compete with the speed of as someone sitting at home alone at 10 pm, they
social media, have assumed the mantle of “fact- were not able to go out for a cuppa. That’s where
checkers” to validate information. For example, social media really kicked in” (Chapman-Smith
fake information and images distributed on Twit- 2012).
ter during Hurricane Sandy in 2012 were quickly
ousted by other social media users and traditional S
media outlets. Although traditional media outlets Summary
have published inaccurate information sourced
from social media, it is often quickly corrected. Social media continues to rapidly evolve as a
Twitter and Facebook were used extensively by useful tool in earthquake communication. Devel-
the Queensland Police Service during the South oping technology will increase the accuracy of
East Queensland, Australia, floods in 2011 to information from “citizen seismologists” to
identify rumors and respond with factual infor- enhance situational awareness, improve warn-
mation. Those platforms became an important ings, coordinate aid and recovery, and galvanize
resource for traditional media. As Bruns humanitarian relief efforts. Impacted communi-
et al. (2012, p. 8) observed: “Additionally, ties will increasingly turn to social media as a
@QPSMedia also played a crucial role in way to communicate lifesaving information,
enabling affected locals and more distant gather support, and empower each other in the
onlookers to begin the difficult process of making recovery process.
Cross-References channel consumption. J Comput Mediat Commun

11(2):659–673
Earle PS, Bowden D, Guy M (2011) Twitter earthquake
▶ Community Recovery Following Earthquake detection: earthquake monitoring in a social world.
Disasters Ann Geophys 54(6):708–715
▶ Emergency Response Coordination Within Edmond S (2013) Social media pre, during and post-
Earthquake Disasters earthquake. http://blog.potentia.co.nz/social-media-
quake/
▶ Resilience to Earthquake Disasters Gowing N (2009) Skyful of lies and black swans. The new
tyranny of shifting information power in crises. The
Reuters Institute of Technology, Oxford
References Hui N (2013) Social media’s role in Ya’an earthquake
aftermath is revealing. http://www.chinafile.com/social-
Abbasi M, Kumar S, Filho A, Liu H (2012) Lesson learned medias-role-yaan-earthquake-aftermath-revealing
in using social media for disaster relief – ASU Crisis Imran M, Elbassuoni S, Castillo C, Diaz F, Meier P (2013)
Response Game. http://www.public.asu.edu/~huanliu/ Extracting information nuggets from disaster-related
papers/SBP12Game.pdf messages in social media. In: Proceedings of the 10th
Anderson M (2012) Earthquake social media response. international ISCRAM conference, Baden
http://www.cfaconnect.net.au/news/earthquake-social- Kaigo M (2012) Social media usage during disasters and
media-response.html social capital: Twitter and the Great East Japan Earth-
Bossu R, Gilles S, Mazet-Roux G, Roussel F, Kamb quake. Keio Commun Rev 34:19–35
L (2011) Flash sourcing, or rapid detection and char- Lindsay BR (2011) Social media and disasters: current uses,
acterization of earthquake effects through website traf- future options, and policy considerations. Congressional
fic analysis. Ann Geophys 54(6):716–727 Research Services Report 7-5700. http://fpc.state.gov/
Bruns A, Burgess J, Crawford K, Shaw F (2012) #qldfloods documents/organization/174191.pdf
and @QPSMedia: Crisis Communication on Twitter in MacLeod L (2010) New media vital in breaking Haiti
the 2011 South East Queensland Floods. Media Ecolo- earthquake story. http://www.bbc.co.uk/worldservi-
gies Project. ARC Centre of Excellence for Creative ce/worldagenda/2010/01/100122_worldagenda_haiti_
Industries & Innovation, Brisbane, Australia monitoring.shtml
Cellan-Jones R (2008) Twitter and the China earthquake, Mathewson N (2012) Social media excel after quakes.
BBC blog, 12 May 2008. http://www.bbc.co.uk/blogs/ http://www.stuff.co.nz/the-press/news/christchurch-
legacy/technology/2008/05/twitter_and_the_china_ earthquake-2011/7517088/Social-media-excel-after-
earthqua.html quakes
Centres for Disease Control (2012) Crisis and emergency Meier P (2012) How crisis mapping saved lives in Haiti.
communication. http://emergency.cdc.gov/cerc/pdf/ http://voices.nationalgeographic.com/2012/07/02/crisis-
CERC_2012edition.pdf mapping-haiti/
Chapman-Smith B (2012) Social media took ‘church’ Meier P (2013) Automatically extracting disaster-relevant
role in quakes. http://www.nzherald.co.nz/technol- information. http://irevolution.net/2013/04/01/auto-
ogy/news/article.cfm?buffer_share=a6e0e&c_id=5 extracting-disaster-info/
&objectid=10828401 Moore M (2008) China earthquake brings out citizen
Civil Defence Emergency Management (2012) New journalists. http://www.telegraph.co.uk/news/world
Zealand ShakeOut, more than 1.3 million news/asia/china/1950212/China-earthquake-brings-out-
involved. http://www.scoop.co.nz/stories/AK1209/ citizen-journalists.html
S00584/new-zealand-shakeout-more-than-13-million Palen L (2008) Online social media in crisis events.
-involved.htm EDUCAUSE Q 31(3):76–78
Crooks A, Croitoru A, Stefanidis A, Radzikowski Perry B, Shaw R, Takeuchi Y (2012) Utilisation of social
K (2013) #Earthquake: Twitter as a distributed sensor media in the East Japan earthquake and tsunami and its
system. Trans GIS 17(1):124–147 effectiveness. J Nat Disaster Sci 34(1):3–18
Dabner N (2011) Breaking ground in the use of social Sakaki T, Okazaki M, Matsuo Y (2010) Earthquake
media: a case study of University earthquake response shakes Twitter users: real-time event detection by
to inform education design with Facebook. www.educ. social sensors. In: Proceedings of the 19th interna-
utas.edu.au/users/afluck/. . ./wg332011_submission_12.doc tional WWW conference (WWW2010), Raleigh
Duffy N (2012) Using social media to build community Starbird K, Stamberger J (2010) Tweak the Tweet:
resilience. Aust J Emerg Manag 27(1):40–45 leveraging microblogging proliferation with a pre-
Dugan L (2012) Would you place a 911 call on Twitter? scriptive syntax to support citizen reporting. In: Pro-
http://www.mediabistro.com/alltwitter/911-call-on- ceedings of the 7th international ISCRAM conference,
twitter_b27752 Seattle
Dutta-Bergman MJ (2006) Community participation and Tapia A, Moore C, Johnson N (2013) Beyond the trustwor-
Internet use after September 11: complementarity in thy Tweet: a deeper understanding of microblogged
Soil-Structure Interaction 3315
data use by disaster response and humanitarian relief structure). Soil-structure interaction characteris-
organisations. In: Proceedings of the 10th international tics depend on several factors:
ISCRAM conference, Baden
Taylor M, Well G, Howell G, Raphael B (2012) The role
of social media as psychological first aid as a support – Intensity, wavelength, and angle of incidence
to community resilience building. A Facebook study of the seismic waves
for Cyclone Yasi Update. Aust J Emerg Manag – Soil stratigraphy
27(1):20–26
Turgut P (2011) Turkey’s earthquake: social media to the – Stiffness and hysteretic damping of the partic-
rescue. http://world.time.com/2011/10/24/turkeys- ular soil layers
earthquake-social-media-to-the-rescue/ – Geometry and rigidity of the foundation
Wallop H (2011) Japan earthquake: how Twitter and – Embedment depth of the structure
Facebook helped. http://www.telegraph.co.uk/tech-
nology/twitter/8379101/Japan-earthquake-how-Twitter- – Inertia characteristics, slenderness, and natu-
and-Facebook-helped.html ral vibration period (eigenperiod) of the
Williams R, Williams G, Burton D (2012) The use of superstructure
social media in disaster recovery. http://extension.mis- – Presence of nearby structures
souri.edu/greene/documents/PlansReports/using%20
social%20media%20in%20disasters.pdf
Young JC, Wald D, Earle P, Shanley L (2013) Various effects are associated to this
Transforming earthquake detection and science phenomenon:
through citizen seismology. Woodrow Wilson Interna-
tional Centre for Scholars, Washington, DC
– A building founded on compliant ground has
different vibrational characteristics, for exam-
ple, higher natural period compared to the
same building on rigid base (solid rock). The
softer the soil, the larger the difference.
Soil-Structure Interaction
– A part of the vibrational energy emanating
from the compliant structure foundation is
Christos Vrettos
transmitted into the surrounding soil through
Division of Soil Mechanics and Foundation
wave radiation in the unbounded soil medium
Engineering, Technical University of
and hysteretic energy dissipation. Such effects
Kaiserslautern, Kaiserslautern, Germany
do not occur in a rigidly supported structure.
– Due to the compliance of the foundation, the
motion at the foundation base contains
Synonyms
rocking and torsional components in addition
to the translational components.
Dynamic soil-structure interaction; Seismic soil-
structure interaction; SSI
The mechanisms governing soil-structure S
interaction can be divided into two distinct inter-
actions: inertial and kinematic interaction.
Introduction Kinematic interaction is the deviation of the
soil response from the free-field motion due to the
Seismic waves propagating through the soil resistance of the stiffer foundation to conform to
impinge upon structures founded on the soil sur- the distortions of the soil imposed by the travel-
face or embedded into it. Displacements are then ing seismic waves. It is commonly expressed in
produced both in the structure and in the soil. The terms of frequency-dependent transfer functions
mutual dependency of the displacements is called relating the disturbed motion at the interface
soil-structure interaction, abbreviated as SSI. foundation/soil to the free-field motion.
Consequently, the motion occurring at the base Inertial interaction arises as the structure
of the structure is different compared to the responds to the soil motion induced by kinematic
free-field motion (motion in the absence of the interaction at the foundation level. Inertial forces
3316 Soil-Structure Interaction
are developed in the structure being transmitted into account by the step-by-step numerical inte-
to the compliant soil. Frequency-dependent gration of the equations of motion within a time-
impedance functions are used to represent the domain algorithm. A drawback is the necessity to
stiffness of the foundation/soil system and the specify the input motion at the base of the model,
associated radiation damping. where it is not known a priori. Since the design
The relative impact of each contribution is a seismic motion is usually given at the free surface
function of the characteristics of the incoming or at outcropping rock, a deconvolution is neces-
waves, the foundation geometry and rigidity, sary to obtain the compatible bedrock motion.
and the soil conditions. Often the bedrock is located at large depths thus
The analysis is particularly challenging due to prohibiting the modeling of the entire soil layer,
the semi-infinite extent of the soil medium, the and some artificial boundary is defined at a
nonlinearity of the soil behavior, the inherent shallower depth. The deconvolution then
variability of the soil stratigraphy, and the depen- involves an iterative procedure. For convenience,
dency of the response on frequency. Several pro- the deconvolution is often carried out using algo-
cedures of different degrees of complexity have rithms that are based on 1D vertical shear wave
been proposed during the past five decades. propagation, thus requiring an adjustment of the
A historical overview is given by Kausel model parameters in order to achieve compatible
(2010). The book by Wolf (1985) provides a solutions between 1D and 2D analyses. Attention
rigorous and comprehensive treatment of the is further required in the selection of appropriate
topic including applications to seismic problems. boundary conditions at the side boundaries of the
Even with the computational facilities avail- discretized domain to avoid spurious reflections
able today, such analyses are associated with a that would contaminate the results. The compos-
major effort, both for modeling the soil-structure ite soil-structure model is finally subjected to the
system and carrying out the calculation. In par- previously determined base rock motion, and the
ticular during the early design stage, parametric evolution in time of displacement and stresses is
studies are necessary in order to assess the influ- computed.
ence of the various parameters and optimize
the system for purposes of cost estimation. Substructure Approach
This necessitates the application of simplified The underlying calculation method comprises
methods that capture the essential features of three steps. At first the seismic motion acting at
the system response. The next sections provide the foundation level is determined assuming a
a brief overview with emphasis on such simpli- rigid but massless foundation. This is referred to
fied methods. as foundation input motion (FIM), and for an
embedded structure, it will include both transla-
tional and rotational components. In the second
Soil-Foundation-Structure Analysis step, the complex-valued frequency-dependent
Models impedances for the foundation/soil system are
determined. The real part of the impedance func-
Two general approaches are commonly used for tion represents a linear spring and the imaginary
the analysis of soil-structure interaction part, a viscous dashpot accounting for the energy
problems. radiation into the soil medium. Finally, the struc-
ture supported by the frequency-dependent
Direct Approach springs and dashpots is subjected to the founda-
The soil and the structure are treated together in a tion input motion computed in the first
combined analysis by modeling them using finite analysis step.
elements or finite differences in two or three While impedance functions are sufficient for
dimensions. This offers the advantage that inelas- rigid foundations, distributed springs and dash-
tic behavior, particularly for the soil, can be taken pots placed around the foundation are used for
nonrigid embedded foundations when the distri- and wave radiation in the unbounded medium is
bution of sectional forces is sought. In this case, modeled by a pair of frequency-dependent dash-
due to the vertical variation of ground motion, the pots with coefficients Cx and Cy attached parallel
imposed differential ground displacements vary to the respective springs. This model may be
over the height of the basement walls. viewed as a single- or multistory building after
The validity of this approach – often called an appropriate reduction of the degrees of
superposition theorem – is shown by Kausel and freedom.
Roesset (1974). The main advantage of the Springs and dashpots for each degree of free-
method is that each step can be handled indepen- dom j can be condensed to complex-valued
dently and with different algorithms. Further, it impedances that are expressed in two equivalent
allows an insight into the contributions from each forms:
analysis step and is particularly suitable for para-

metric studies. K~ ¼ K j þ ioCj ¼ K j 1 þ i2bj (1)
The application of the principle of superposi-
tion requires linear behavior. Inelastic behavior is
where o is the circular frequency of the excita-
implemented using equivalent linearization by
tion, i is the imaginary unit, and bj is a damping
selecting soil modulus and radiation damping to
coefficient that is related to the viscous dashpot
correspond to the likely effective strain level the
coefficient of a simple oscillator by
soil will experience under the specific loading.
This is achieved by means of an iterative proce-
dure. Superposition is shown to be a reasonable Im K~ j oCj
bj ðoÞ ¼ ¼ (2)
approximation even when inertia forces produce 2Re K~ j 2K j
large strains in the vicinity of the foundation,
since shear strains due to kinematic interaction The use of bj has the advantage that at resonance
effects are usually significant in deeper soil of the compliant system bj corresponds to the
regions. percentage of critical damping.
The undamped natural vibration period of the
structure in its fixed-base condition is
Inertial Interaction
rffiffiffiffi
k
Shallow Foundations T ¼ 2p (3)
m
The illustration of the concepts is made on the
basis of a simple structure-soil system being com-
For the case of a compliant base, it can be shown
posed by a linear structure of height h, mass m,
that the respective natural period is (Veletsos and
lateral stiffness k, and damping ratio bstr that is S
Meek 1974)
connected to a rigid foundation of radius r resting
on the surface of a homogeneous elastic half- sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
space. The half-space is used to represent the k kh2
T~ ¼ T 1 þ þ (4)
unbounded soil medium and is characterized by Kx Ky
its shear modulus G, Poisson’s ratio n, and mass
density r. Mass and moment of inertia of the Hence, the period of the flexibly supported struc-
foundation are neglected for simplification. The ture is higher than that on rigid base. Since the
compliance of the soil is modeled by two spring stiffnesses are in general frequency depen-
frequency-dependent springs placed underneath dent, an iterative procedure is necessary to eval-
the rigid foundation: a horizontal translation uate the period T~ . A reasonable approximation
spring of stiffness Kx and a rotational spring of consists in using the spring values corresponding
stiffness Ky. Energy dissipation in the soil due to to the fixed-base natural period, and even simpler
friction within the material (hysteretic damping) is to use the static values of the springs.
Soil-Structure Interaction, Fig. 1 Effects of soil- and different values of the slenderness ratio h. Poisson’s
structure interaction: (a) elongation of natural period in ratio n = 0.45 (Adapted from Veletsos (1977). In: Hall WJ
dependency on the ratio of structure-to-soil stiffness s; (b) (ed) Structural and geotechnical mechanics, 1st edn, #
increase in effective damping in dependency on the natu- 1977. Reprinted by permission of Pearson, Inc., Upper
ral period ratio T~=T. Curves are for mass ratio m = 0.15 Saddle River, NJ)
The dimensionless parameters controlling the and 0.5 for stiff systems such as shear wall and
period lengthening are braced frame structures. The period lengthening
variation with the stiffness ratio is shown in
h Fig. 1a for typical values of the parameters
Stiffness ratio structure-to-soil s ¼ (5)
T vs involved.
For the overall effective damping ratio of the
h system, several approaches have been proposed
Slenderness ratio h ¼ (6)
r in the literature differing in the degree of approx-
imation involved. Usually products of damping
m
Mass ratio m ¼ (7) ratios are neglected as higher-order terms. The
rpr 2 h
most widespread among these solutions – that
also entered design codes – is that derived by
with
Veletsos and Meek (1974). Assuming structural
pffiffiffiffiffiffiffiffiffi damping of viscous nature, the overall effective
vs ¼ G=r (8) damping becomes
denoting the shear wave velocity in the soil.

The stiffness ratio will be larger for stiff struc- ~ ¼ b þ bstr
b (9)
0
T~=T
3
tural systems such as shear walls and smaller for
flexible systems such as moment frames. For soil
and weathered rock sites, this term is typically where b0 represents the contribution from the
smaller than 0.1 for flexible systems such as soil-structure interaction – being referred to as
moment frames and between approximately 0.1 foundation damping – that includes both material
and radiation damping (Veletsos 1977). The K j ¼ K 0j kj (11)

respective expression is written here in the more
general form
Exact closed-form solutions are available only
3 for perfectly rigid circular foundations and
T T~ bx by 2 relaxed boundary conditions at the soil-
b0 ¼ k þ h (10)
T~ T K~ x K~ y
foundation interface, i.e., normal stresses are
neglected for swaying and shear stresses for
From Eq. 9 it is evident that the effectiveness rocking. These solutions are
of the structural damping is reduced by soil- Horizontal translation
structure interaction as the period ratio T~=T
increases. This may lead to a decrease in overall 8
K 0x ¼ Gr (12)
damping unless this reduction is compensated 2n
by the increase in the foundation damping. In
practice, effective damping is taken higher than Rocking
the structural damping, the value 5 % used in the
development of design provisions being consid- 8
ered as a lower bound. Figure 1b shows the K 0y ¼ Gr 3 (13)
3ð 1 n Þ
significant increase of the foundation damping
with decreasing slenderness ratio h/r: rocking
motion that is characterized by small radiation These expressions may be used for square
damping dominates the response of slender foundations – and also for rectangular founda-
structures, whereas for squat structures the tions with aspect ratio less than 3 – by replacing
prevailing motion is horizontal translation that the radius by an equivalent value that yields the
radiates energy into the soil more efficiently. same footprint area for swaying and equal
Observations based on data from instrumented moments of inertia for rocking.
buildings confirmed the analytical findings. For A review of available solutions for foundation
the majority of structures, the stiffness ratio impedances is presented by Pais and Kausel
h/Tvs will be less than 0.5 and the mass ratio (1988) and Gazetas (1991) and the update by
will range between 0.1 and 0.2 with a typical Mylonakis et al. (2006). Approximate expres-
average of 0.15 (Stewart et al. 2003). The case sions and graphs are compiled for various config-
studies analyzed revealed that the governing urations and for all six modes of vibration. They
parameter for inertial interaction is the stiffness include static values for rectangular foundations,
ratio and that these effects can be neglected for stiffness modifiers, and expressions for the radi-
values less than 0.1. ation damping.
We restrict here the presentation of results to
S
Impedances for Shallow Foundations swaying and rocking motion for rectangular foun-
Frequency-dependent springs and dashpots for dations with footprint area 2a 2b with a b
shallow foundations have been determined in the with the x-axis running parallel to the longer
last decades by several authors for different foundation side. The subscripts yx and yy in the
geometries and soil stratigraphies. In most impedances indicate rotation around the x- and
cases radiation damping is expressed in terms y-axis, respectively. The weak coupling between
of the dashpot coefficient Cj, as given in Eq. 1. translational and rocking mode is neglected. The
The stiffness Kj at zero frequency is referred to frequency dependency is captured by the dimen-
as the static foundation stiffness and is denoted sionless parameter
by K0j . The effects of frequency on the spring
values for the particular vibrational mode j are ob
a0 ¼ (14)
then given by stiffness modifiers such that vs
and the foundation aspect ratio is denoted by 2 3

6 ð4=3Þv ‘ a20 7
a Rocking byx ¼6
4
7
5
‘¼ 1 (15) 0:4
b K 0yx =Gb 3
2:2 3 þ a20
‘

The approximate expressions obtained by Pais a0
and Kausel (1988) are displayed in the following. 2kyx
(26)
Surface Foundations 2 3
The static solutions are:
6 ð4=3Þv‘3 a20 7
Rocking byy ¼6
4
7
5
Gb
K 0yy =Gb3
1:8
þ a20
Swaying K 0x ¼ 6:8‘0:65 þ 2:4 (16) 1 þ 1:75ð‘ 1Þ
2n

a0
Gb 2kyy
Swaying K 0y ¼ 6:8‘0:65 þ 0:8‘ þ 1:6
2n (27)
(17)
where
Gb3
Rocking K 0yx ¼ ½3:2‘ þ 0:8 (18) pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1n
v¼ 2ð1 nÞ=ð1 2nÞ v 2:5 (28)
Gb
3
Rocking K 0yy ¼ 3:73‘2:4 þ 0:27 (19) is the ratio of compressional wave velocity to
1n
shear wave velocity in the soil.
The frequency-dependent stiffness modifiers are: It should be mentioned that the exact curves
for the stiffness modifiers and the damping fac-
Swaying kx ¼ 1 (20) tors have in general a smooth wavy form, and the
expressions given above consist approximations
Swaying ky ¼ 1 (21) to these curves.
Key features of the system behavior are:
2 3 Dynamic modifiers for translational stiffness
6 0:55a20 7 are almost unity independent of the foundation
Rocking kyx ¼ 1 6
4
7
5 (22) aspect ratio, whereas rocking modifiers are sig-
1:4
0:6 þ 3 þ a20 nificantly reduced with frequency in a very weak
‘
dependency on the foundation aspect ratio.
2 3
pffiffi 2 Radiation damping for the horizontal transla-
6 0:55 þ 0:01 ‘ a0 7 tional mode is only modestly influenced by the
Rocking kyy ¼ 1 6
4
7 (23)
5 direction of vibration or the foundation aspect
0:4
2:4 3 þ a0 2
ratio. For rocking on the other hand, the damping
‘
is strongly affected by the aspect ratio and the
The viscous damping coefficients accounting for direction of vibration, increasing with the foun-
radiation damping as determined from the dash- dation aspect ratio. At low frequencies damping
pot coefficients using Eq. 2 are: in rocking motion is smaller compared to that in
horizontal translation due to interference phe-

4‘ a0 nomena; it only overweighs translational
Swaying bx ¼ (24) damping at higher frequencies and for elongated
K 0x =Gb 2kx
foundations when excited in the direction of the
"
#
longer foundation side. Hence, translational
4‘ a0
Swaying by ¼ (25) foundation movement may often be predominant
K y =Gb 2ky
0
with respect to radiation damping.
Embedded Foundations associated with the stipulated design ground

The references cited above contain also informa- motion. In critical projects seismic site response
tion for embedded foundations. Embedment analyses are carried out with the soil properties
increases static foundation stiffness. According being determined from special dynamic labora-
to the review by Pais and Kausel (1988), dynamic tory tests on undisturbed samples. First-order
stiffness modifiers remain largely unaffected. estimates for the strain-compatible values are
The dynamic analyses for obtaining such imped- given in some code provisions. Typical values
ances usually assume a perfect contact between as recommended by Eurocode 8, Part 5 (CEN
the soil and the basement walls, a situation that 2004), are tabulated below in terms of their
seldom occurs in reality. This yields higher small-strain amplitude values G0 and vs0 in
damping values as observed from the actual dependency on the effective ground acceleration
response of buildings. A practical, conservative defined as the spectral acceleration at the plateau
approach consists in considering the embedment of the response spectrum divided by 2.5. Guide
effects only for the static stiffness and applying values for the hysteretic soil damping are also
the dynamic modifiers of surface foundations. given.
Alternatively, one may use the formulae given The small-strain values of the soil shear mod-
by Gazetas (1991) and by Mylonakis et al. (2006) ulus or the shear wave velocity may be deter-
that consider an effective height of the contact mined by a variety of methods, the choice
zone along the perimeter of the embedded depending on the variability of the soil condi-
foundation. tions, available knowledge on the material behav-
ior, and the importance of the structure. These
Soil Layering methods include: (i) empirical relationships in
Impedance functions for multilayered soils can terms of the SPT blow count or of the tip resis-
only be determined with specialized software that tance of the CPT tests, (ii) geophysical field
is not easily accessible to practicing engineers. methods based on wave propagation, and (iii)
Available algorithms are mostly based on finite dynamic laboratory tests. An overview of the
element procedures incorporating efficient con- testing procedures and available design equations
sistent boundaries for the proper energy radiation is summarized by Kramer (1996).
at the domain boundaries.
A particular case constitutes a soil layer of finite Effective spectral ground acceleration
[g]
depth on rock where a cutoff frequency exists,
0.10 0.20 0.30
below which there is no radiation damping. The G/G0 0.80 0.50 0.36
respective formulae given in the above references (0.10) (0.20) (0.20)
may be used for a two-layer system when the shear vs/vs0 0.90 0.70 0.60
wave velocity in the top layer is less than half of (0.07) (0.15) (0.15) S
that of the underlying stratum. Impedances for Damping 0.03 0.06 0.10
square foundations on uniform or nonuniform ratio
soil layer overlying a half-space are tabulated by
Wong and Luco (1985). Adaption in Design Codes and Implication
for the Design
Parameters for Soil Behavior The implications of inertial SSI for design are
The expressions given above assume linear elas- illustrated in Fig. 2 with reference to the acceler-
tic or viscoelastic soil behavior. However, for ation response spectrum used for evaluating seis-
moderate or strong seismic excitations, the mic base shear forces in buildings. Idealized
nonlinearity of the soil must be taken into envelope spectra in modern codes initially
account. Hence, the values of the shear modulus increase with period, attaining a plateau value,
entering the equations for the SSI effects must be and start decreasing monotonically after a certain
adjusted to reflect the strain level in the ground period that is in the order of 0.4 to 1.0 s.
Soil-Structure
Interaction, Fig. 2 Effect
of natural period elongation
and foundation damping on
a typical acceleration
design spectrum (Adapted
from Stewart et al. (2003)
by permission of the
Earthquake Engineering
Research Institute)
Soil-Structure
Interaction, Fig. 3 Ratio
of spectral acceleration to
peak ground acceleration
for 5 % structural damping
for some severe
earthquakes with long-
period components
compared to that of a
typical code for soft soil
(Adapted from Gazetas and
Mylonakis (1998) by
permission of the American
Society of Civil Engineers)
For buildings with periods larger than about 0.5 s, their peak at periods greater than 1.0 s. Spectra
consideration of period elongation and flexible from some prominent records are contrasted to a
base damping will lead to a reduction of the typical design spectrum for soil in Fig. 3. SSI
base shear demand. Hence, in most cases, SSI phenomena in these earthquakes had detrimental
effects are neglected in the frame of conservative effects as revealed by analyses linking site con-
design. ditions and building natural periods to observed
However, there are various seismic environ- damage. In the 1985 Mexico City earthquake, for
ments with recorded response spectra exhibiting example, due to SSI effects, the natural period of
10–12 story buildings founded on soft clay was from results by various authors are summarized
altered from about 1.0–1.5 s to nearly 2.0 s, thus by Gazetas (1991). The static stiffnesses are
coinciding with the peak of the response spec- expressed in terms of the pile diameter d and
trum at the particular site. The associated phe- Young’s moduli of the soil and the pile E and
nomena are elucidated among others by Gazetas Ep, respectively:
and Mylonakis (1998).
0:21
Hence, proper assessment of both the antici- Ep
pated seismic input and the prevailing soil condi- Swaying K 0x ¼ dE (29)
E
tions is an indispensable prerequisite in any SSI
analysis. In modern seismic codes the site char- 0:75
Ep
acterization for deep soil deposits is based almost Rocking K 0y ¼ 0:15d E 3
(30)
E
exclusively on the near-surface region of the soil
(often the top 30 m), disregarding the depth of the 0:50
Ep
underlying rock. The representative average Cross-swaying-rocking K 0xy ¼ 0:22d E
2
E
shear wave velocity to this depth in this deposit
(31)
is used as parameter for the classification;
cf. Dobry et al. (2000). The dynamic modifiers are approximately equal
to unity:
Pile Foundations
kx ¼ ky ¼ kxy 1 (32)
Single Pile
Consider a pile horizontally loaded at its head at The expressions for the radiation damping bj as
the ground surface. The deformed shape of the defined by Eq. 2 are given in dependency on the
pile extends down to a so-called active dimensionless frequency
(or effective) length below which it becomes
negligible. This length depends on the pile diam- o d=2
eter, the elastic modulus of the soil, the ratio of a0 ¼ (33)
vs
pile modulus to soil modulus, and the fixity con-
ditions. Expressions for static and dynamic load- as follows:
ing are given by Gazetas (1991). For static loads
this length is of the order of 10–20 pile diameters, 0:17
Ep
while for dynamic loading this length will be Swaying bx ¼ 0:35 a0 (34)
E
greater due to the wave propagation. With respect
to flexural response, the pile can be modeled 0:20
Ep
without significant error as an infinite-long Rocking by ¼ 0:11 a0 (35) S
beam when its length is greater than the active E
length. Two models are commonly used for the 0:18
analysis: elastic continuum theory or Winkler Ep
Cross-swaying-rocking bxy ¼ 0:27 a0
spring models (Pender 1993). E
Following the same principles as for shallow (36)
foundations, the horizontally loaded soil-pile sys-
tem may be represented by three impedances Pile Groups
corresponding to swaying, rocking, and cross- Building foundations are always constructed as
swaying-rocking. Consideration of the latter is groups of piles. In evaluating the dynamic stiff-
necessary since the reference level is located at ness of a pile group, the interactions between the
the pile head and the resultant of the reactions piles must be taken into consideration, just like in
acts at a specific depth thus inducing a bending the case of static loading. However, the cross-
moment at the pile head. Expressions synthesized interaction of individual piles is strongly
dependent on frequency, thus precluding descrip- Rayleigh waves, and torsion in the presence of
tion by simple explicit formulae. The rigorous SH- or Love waves. The torsion of symmetrical
solution methods available are based on the buildings observed in earthquakes is a conse-
thin-layer method (Kaynia and Kausel 1982; quence of obliquely incident seismic waves.
Waas and Hartmann 1984). Fortunately, a Further, the modification of the seismic motion
remarkable simple solution procedure was dis- depends on the frequency content of the seismic
covered by Dobry and Gazetas (1988) that is motion with high-frequency components being
straightforward to implement, thus facilitating filtered out by the slab when the respective appar-
the assessment of the associated SSI effects with ent wavelength is shorter than an effective length
a very good accuracy. The respective interaction of the foundation slab (the diameter for circular
coefficients between the piles are given in terms foundations).
of pile spacing, excitation frequency, and the Kinematic interaction effects are expressed in
wave velocity through the soil between the terms of transfer functions relating the amplitude
piles. The values for the overall stiffness and of the foundation input motion to that of the free-
damping of the pile group are then assembled field motion. The system considered consists of a
using the respective values of the single piles rectangular foundation with area 2a 2b, a b
and these interaction factors. that is subjected to harmonic SH waves of circu-
lar frequency o with particle motion in the direc-
tion of the x-axis impinging on the foundation at
Kinematic Interaction an angle av with the vertical and propagating
along the positive y-axis. The transfer functions
Shallow Foundations derived by Veletsos et al. (1997) include both
Kinematic interaction is induced by the presence coherent and incoherent seismic motions. They
of a stiff foundation that forces the foundation are given in dependency on the dimensionless
motions to deviate from the free-field motions. parameter
The associated phenomena are due to (i) base-
slab averaging of inclined or incoherent seismic sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
obe b
waves, and (ii) embedment of the foundation. a~0 ¼ k2 þ sin2 av (37)
vs be
Base-Slab Averaging pffiffiffiffiffi
Seismic waves impinging at directions other than where be ¼ ab is the half-side length of an
vertical arrive at different points along the foun- equivalent square foundation, vs is the shear
dation at different times giving rise to the wave velocity, and k is a ground motion incoher-
so-called wave passage effects. The apparent ence parameter. The curves shown in Fig. 4 rep-
propagation velocity of the waves is in the order resent the two limiting cases with k = 0 and
of 1.5–3.5 km/s and is controlled by the wave av = 0, respectively. The transfer functions for
propagation in the underlying rock. In addition torsional motions are referred to the foundation
to this, ground motion is in most cases inherently edge being the product of foundation half-width
incoherent resulting from inhomogeneities along b and rotational angular distortion.
the travel path from the source to the site. Lateral transfer functions are for both types of
Studies conducted hitherto mainly address the wave motion only very weak dependent on the
wave passage problem that is amenable to ana- aspect ratio a/b suggesting that the governing
lytical treatment. They show that the slab due to parameter is the foundation area. The induced
its stiffness and flexural rigidity averages the torsional component, however, is very sensitive
free-field displacement pattern by reducing the both to the aspect ratio and the type of wave
translational motions and at the same time intro- motion.
ducing rotational motions. The latter include Recent observations on buildings indicate that
rocking in the presence of inclined SV-, P-, or the apparent value of k (denoted by ka) is nearly
Soil-Structure Interaction, Fig. 4 Amplitude of trans- waves; (b) non-vertically incident, coherent waves. The
fer functions between free-field and foundation input solid lines are for the horizontal motion and the dashed
motion for rectangular foundations subjected to obliquely lines for the induced torsional component. Curves com-
incident shear waves: (a) vertically incident, incoherent puted from expressions in Veletsos et al. (1997)
proportional to the small-strain shear wave veloc- slenderness of the pile, and the frequency of
ity vs yielding roughly ka = 0.2 at a typical value excitation with high-frequency components
vs = 250 m/s, Kim and Stewart (2003). being filtered out especially by relatively short,
rigid piles.
Foundation Embedment Analyses of kinematic interaction that include
Embedment effects result from the scattering of variable stratigraphy are carried out mostly by
incoming waves. Rocking motions develop due using the Winkler spring model. The ends of the
to nonuniformly distributed tractions against the springs and dashpots that capture SSI effects are
side walls. Assessment is made by means of connected to the free field where the soil response
transfer functions relating the base-slab transla- is imposed. The latter is computed independently.
tional and rocking motions to the free-field The accuracy of this simplified approach depends
motions. An accurate numerical solution for on the selection of the springs and dashpots that
cylindrical foundations subjected to coherent are obtained adopting physically justified approx-
shear wave motions is provided by Day (1977). imations (Pender 1993).
The embedment depth e is normalized with Attention deserves the bending moment
respect to the radius of the foundation r, and the induced in the pile during the passage of seismic S
frequency dependency is captured by the dimen- waves. The maximum value occurs, as expected,
sionless parameter a0 ¼ or=vs . Figure 5 shows at soil layer interfaces, strongly increasing with
typical patterns for vertically propagating waves. the contrast in shear wave velocity between the
bottom and top layer (Nikolaou et al. 2001).
Piles The numerical study by Fan et al. (1991) using
Piles embedded in a soil stratum respond to inci- the continuum model by Kaynia and Kausel
dent vertical shear waves in dependence on their (1982) for pile groups excited by vertically prop-
flexural rigidity in relation to the stiffness of the agating shear waves provides graphs showing the
surrounding soil. The incoming wave field is effects of pile rigidity relative to that of the soil,
modified, the displacement at the pile head differs pile slenderness, pile spacing, number of piles,
from that of the free field, and pile-head rotation and pile-head fixity conditions. An idealized gen-
is induced. The displacement reduction depends eral shape of the frequency dependence of the
on the ratio of pile modulus to soil modulus, the kinematic response is defined in terms of a
Soil-Structure
Interaction,
Fig. 5 Amplitude of
transfer functions between
free-field and foundation
input motion for
cylindrical, embedded
foundation subjected to
vertically incident coherent
shear waves for different
normalized embedment
depths. (a) horizontal
translation; (b) rocking
component (Redrawn from
Day (1977))
displacement factor relating the pile-head dis- Concluding Remarks

placement to that of the free field. This factor is
approximately unity at low frequencies with the Despite its inherent complexity, the theory of
pile closely following the ground movement; in linear soil-structure interaction and the implica-
the medium frequency range, it decreases with tions in structural performance are now well
frequency and beyond a distinct frequency fluc- understood. Refinements, optimization, and vali-
tuates around a constant value of 0.2–0.4. The dation studies are subjects of ongoing research.
difficulty consists in defining the transition fre- The variability in the stratigraphy of soil deposits,
quencies for the particular system layout. the nonlinearity of the soil behavior, the fre-
It must be realized though that there is no quency dependency of the response, and the lim-
simple means for evaluating kinematic interac- ited availability of specialized software for the
tion for pile groups. In noncritical situations, analysis make the proper assessment of the SSI
however, this difficulty may be circumvented by effects still a difficult task, requiring physical
neglecting kinematic interaction. This is justified insight when applying such concepts. It should
by findings that kinematic effects for pile groups be self-evident that the effective implementation
are similar to those for individual piles, in partic- in an integrated structural design asks for a close
ular for horizontal translation and to a lesser collaboration between structural and geotechni-
extent for torsional and rocking vibration modes. cal engineers.
Summary system used in recent building seismic code provi-

sions. Earthq Spectra 16(1):41–67
Fan K, Gazetas G, Kaynia A, Kausel E (1991) Kinematic
The main effects of soil-structure interaction on seismic response of single piles and pile groups.
the seismic response of structures founded on J Geotech Eng, ASCE 117(12):1860–1879
compliant ground are presented. The modeling Gazetas G (1991) Foundation vibrations. In: Fang HY
concepts to capture the associated modification (ed) Foundation engineering handbook, 2nd edn. Van
Nostrand Reinhold, New York, pp 553–593
of the building natural period and the energy Gazetas G, Mylonakis G (1998) Seismic soil structure
dissipation due to radiation damping are interaction: new evidence and emerging issues. In:
highlighted by reference to relatively simple Geotechnical earthquake engineering and soil dynam-
structures. Both kinematic and inertial actions ics III, ASCE, Reston, pp 1119–1174
Kausel E (2010) Early history of soil-structure interaction.
are treated. Available expressions for the Soil Dyn Earthq Eng 30(9):822–832
dynamic impedance functions are summarized Kausel E, Roesset JM (1974) Soil-structure interaction
both for shallow foundations and piles. A brief problems for nuclear containment structures. In:
account is given of the implications in seismic ASCE power division specialty conference, electric
power and the civil engineer, Boulder, pp 469–498
design provisions for buildings. Kaynia AM, Kausel E (1982) Dynamic stiffness and seis-
mic response of pile groups. In: Research report
R82-03, Massachusetts Institute of Technology
Kim S, Stewart JP (2003) Kinematic soil-structure inter-
Cross-References action from strong motion recordings. J Geotech
Geoenviron Eng, ASCE 129(4):323–335
▶ Building Codes and Standards Kramer SL (1996) Geotechnical earthquake engineering.
Prentice Hall, Upper Saddle River
▶ Earthquake Response Spectra and Design Mylonakis G, Nikolaou A, Gazetas G (2006) Footings
Spectra under seismic loading: analysis and design issues
▶ Engineering Characterization of Earthquake with emphasis on bridge foundations. Soil Dyn Earthq
Ground Motions Eng 26:824–853
Nikolaou S, Mylonakis G, Gazetas G, Tazoh T (2001)
▶ Response Spectrum Analysis of Structures Kinematic pile bending during earthquakes: analysis
Subjected to Seismic Actions and field measurements. Géotechnique 51(5):425–440
▶ Seismic Analysis of Masonry Buildings: Pais A, Kausel E (1988) Approximate formulas for
Numerical Modeling dynamic stiffnesses of rigid foundations. Soil Dyn
Earthq Eng 7(4):213–227
▶ Selection of Ground Motions for Response Pender M (1993) Aseismic pile foundation design analy-
History Analysis sis. Bull NZ Nat Soc Earthq Eng 26(1):49–160
▶ Site Response: 1-D Time Domain Analyses Stewart JP, Kim S, Bielak J, Dobry R, Power MS (2003)
▶ Substructuring Methods for Finite Element Revisions to soil-structure interaction procedures in
NEHRP design provisions. Earthq Spectra 19(3):677–696
Analysis Veletsos AS (1977) Dynamics of structure-foundation sys-
tems. In: Hall WJ (ed) Structural and geotechnical
mechanics. Prentice-Hall, Englewood Cliffs, pp 333–361 S
References Veletsos AS, Meek JW (1974) Dynamic behavior of
building-foundation systems. Earthq Eng Struct Dyn
CEN-European Committee for Standardization 3:121–138
(2004) Eurocode 8: design of structures for earthquake Veletsos AS, Prasad AM, Wu WH (1997) Transfer func-
resistance part 5: foundations, retaining structures and tions for rigid rectangular foundations. Earthq Eng
geotechnical aspects Struct Dyn 26(1):5–17
Day SM (1977) Finite element analysis of seismic scatter- Waas G, Hartmann HG (1984) Seismic analysis of pile
ing problems. PhD thesis, University of California, foundations including pile-soil-pile interaction. In:
San Diego Proceedings, 8th World conference on earthquake
Dobry R, Gazetas G (1988) Simple method for dynamic engineering, San Francisco, vol V, pp 55–62
stiffness and damping of floating pile groups. Wolf JP (1985) Dynamic soil-structure interaction.
Géotechnique 38:557–574 Prentice-Hall, Englewood Cliffs
Dobry R, Borcherdt RD, Crouse CB, Idriss IM, Joyner Wong HL, Luco JE (1985) Table of impedance functions
WB, Martin GR, Power MS, Rinne EE, Seed RB for square foundations on layered media. Soil Dyn
(2000) New site coefficients and site classification Earthq Eng 4(2):64–81
3328 Source Characterization for Earthquake Early Warning
barrier-shaped, accelerometric network is

Source Characterization for deployed between the source region and the tar-
Earthquake Early Warning get site to be protected. The alert is issued when
two or more nodes of the array record a ground
Aldo Zollo, Gaetano Festa, Antonio Emolo and acceleration amplitude larger than a default
Simona Colombelli threshold value. For typical regional distances,
Department of Physics, University of Naples the peak acceleration at the barrier nodes is
Federico II, Federico II – AMRA S.c.ar.l, expected to be associated with the S-wave train,
Napoli, Italy so that the distance between the network and the
target is set to maximize the lead time (i.e., the
time available for warning before the arrival of
Synonyms strong ground shaking at the target sites), which
is, in this case, the travel time of S-waves from
Earthquake early warning; Earthquake ground the barrier to the target site.
motion; Earthquake source observation; Real- EEWS have experienced a very rapid
time location; Real-time magnitude improvement and a wide diffusion in many active
seismic regions of the world in the last three
decades (Fig. 1). They are operating in Japan,
Introduction Taiwan, Mexico, and California. Many other sys-
tems are under development and testing in other
Earthquake Early Warning Systems (EEWS) are regions of the world such as in Italy, Turkey,
real-time, seismic monitoring infrastructures that Romania, and China. Most of these existing
are able to provide a rapid notification of the EEWS essentially operate in the two different
potential damaging effects of an impending configurations described above, i.e., regional
earthquake. This objective is achieved through and on-site, depending on the source-to-site dis-
the fast telemetry and processing of data from tance and on the geometry of the considered
dense instrument arrays deployed in the source network with respect to the source area. The
region of the event of concern (regional EEWS) “front-detection” EEWS such as the barrier-
or surrounding/at the target infrastructure (front- type, Seismic Alert System (Espinosa-Aranda
detection or site-specific EEWS). et al. 2011) for Mexico City can be particularly
A regional EEWS is based on a dense sensor advantageous when the only potential seismic
network covering a portion or the entire area that sources are at some distance from the strategic
is threatened by earthquakes. The relevant source target to be protected.
parameters (event location and magnitude) are The regional EEWS approach is based on the
estimated from the early portion of recorded sig- detection of the initial P-wave signal at a number
nals (initial P-waves) and are used to predict, with of near-source stations, typically 4 to 6. Several
a quantified confidence, a ground-motion inten- methodologies have been proposed for the real-
sity measure at a distant site where a target struc- time estimation of the earthquake location and
ture of interest is located. Site-specific magnitude and are now implemented in the EW
(or on-site) EEWS consist of a single sensor or algorithms, such as ElarmS (Allen et al. 2009),
an array of sensors deployed in the proximity of Virtual Seismologist (Cua et al. 2009), and
the target structure that is to be alerted and whose PRESTo (Satriano et al. 2010) presently running
measurements of amplitude and predominant in California, Switzerland, and Southern Italy,
period on the initial P-wave motion are used to respectively. In the framework of EU REAKT
predict the ensuing peak ground motion (mainly (Strategies and Tools for Real-time Earthquake
related to the arrival of S- and surface waves) at Risk Reduction, FP7:ENV2011.1.3.1-1) and
the same site. Front-detection EEWS is essen- international collaboration projects, testing of
tially a variant of the on-site approach, where a PRESTo early warning platform is performed in
Source Characterization for Earthquake Early Warning 3329
Source Characterization for Earthquake Early Warn- operative systems, which are providing warnings to public
ing, Fig. 1 The map shows the distribution of Earth- users. In black, the systems which are currently under real-
quake Early Warning Systems around the world, with a time testing. Gray color is finally used for those countries
color indicating the status of the system. In purple, the where feasibility studies are currently being doing
Romania, Greece, Turkey, Spain, and South performed on the initial P-wave signal and the
Korea. The real-time magnitude estimation is final peak ground motion. Wu and Kanamori
generally inferred from the measurement of (2005) first showed that the maximum amplitude
peak displacement amplitude and/or the predom- of a high-pass filtered vertical displacement,
inant period measured in the first few seconds of measured on the initial 3 s of the P-wave
the recorded P-signal, typically 3–4 s. Although (named Pd), can be used to estimate the peak
the saturation of the P-wave parameters has been ground velocity (PGV) at the same site, through
observed for M > 6.5–7 earthquakes, several a power-law relationship. The main advantage is
methodologies making use of longer time win- that this relationship does not require an indepen-
dows of the P-wave and/or the S-wave to update dent estimate of the magnitude as for regional
magnitude estimates have been shown to be effi- EEWS. Although initially observed for near-
cient in minimizing the problem of magnitude source records (distances < 30 km), further ana-
underestimation (Colombelli et al. 2012b). The lyses on independent datasets have confirmed S
source location and magnitude estimations, that log PGV vs. log Pd scaling is still valid at
which are continuously updated by adding new relatively large distances (distances < 300 km)
station data, as the P-wave front propagates (Zollo et al. 2010; Colombelli et al. 2012a). Most
through the regional EW network, are then used of the currently operating on-site EEWS are
to predict the severity of ground shaking at sites threshold-based, alert methodologies: the alert is
far away from the source, by using regional- issued as the measured initial P-wave peak ampli-
specific, ground-motion prediction equations. tude overcomes a given threshold which is arbi-
The on-site early warning approaches are gen- trarily set according to the predicted S-wave peak
erally aimed at estimating the expected peak ground-motion amplitude. Since small magni-
ground shaking, associated with S- and surface tude earthquakes may have very large amplitudes
waves, directly from the recorded early P-wave driven by high-frequency spikes, such a basic
signal. This is achieved through the use of empir- threshold system can produce frequent false
ical regressions between measurements alarms. A more robust approach is to combine
the P-wave peak (which scales with distance and proposed, such as the strategy of expanding the
magnitude) and P-wave predominant period P-wave time window for the real-time signal
(which scales with the magnitude), into a single processing, the 2D mapping of the potential dam-
proxy to be used for on-site warning (Wu and age zone, and the use of continuous GPS mea-
Kanamori 2005). Following this idea, Zollo surements and methodologies to estimate fault
et al. (2010) and Colombelli et al. (2012a) have rupture extent in real time by classifying stations
proposed a threshold-based EW method based on into near source and far source. These innovative
the real-time measurement of the period (tc) and aspects of early warning will be discussed in the
peak displacement (Pd) parameters at stations present review, with a specific focus on methods
located at increasing distances from the earth- for rapid and reliable source characterization for
quake epicenter. The measured values of early early warning applications.
warning parameters are compared to threshold
values, which are set for a given minimum mag-
nitude and instrumental intensity. At each record- Methodology
ing site an alert level is assigned based on a
decisional table with four levels defined upon Point-Source Characterization for an
threshold values of the parameters Pd and tc. Earthquake Early Warning System
Given a real-time, evolutionary estimation of In EEWS the strong motion is generally synthe-
earthquake location from first P arrivals, the sized by a single parameter (in most cases the
method also furnishes an estimation of the extent peak ground velocity, PGV, the peak ground
of potential damage zone as inferred from con- acceleration, PGA, or the peak ground displace-
tinuously updated averages of the period param- ment, PGD, Fig. 2), which can be directly related
eter and from mapping of the alert levels to the damage that a building or an infrastructure
determined at the near-source accelerometer may undergo because of the earthquake. Two
stations. possible approaches can be explored for the pre-
P-wave-based, regional, and on-site EW diction/estimation of ground-motion parameters
methods can be integrated in a unique alert sys- at a given site.
tem (as actually done, e.g., in the new version of A first possibility is to relate the ground shak-
PRESTo, e.g., PRESTo Plus, Zollo et al. 2014), ing to simplified macroscopic description of the
which can be used in the very first seconds after a source, yielding ground-motion prediction equa-
moderate-to-large earthquake to determine the tions. In such a case, indicating with PGX the
earthquake location and magnitude and to map selected ground-motion parameter, the simplest
the most probable damaged zone, using data from attenuation relationship relates the logarithm of
receivers located at increasing distances from the PGX with the earthquake-to-site distance R and
source. the magnitude M:
Methodologies for regional earthquake early
warning assume a point-source model of the logPGX ¼ f ðM, RÞ (1)
earthquake source and isotropic wave amplitude
attenuation. These assumptions may be inade- By definition of magnitude, a linear function of
quate to describe the earthquake source of large M is in most cases adequate to describe the influ-
earthquakes and wave amplitude attenuation ence of the earthquake size on the ground motion.
effects, and they can introduce significant biases The distance effect is instead accounted for by
in the real-time estimation of earthquake location two terms describing the decay of the amplitude
and magnitude. This issue is critically related to owing to geometrical spreading and inelastic pro-
the EEWS performances in terms of expected cesses within the upper crust. Nevertheless, more
lead time and of uncertainties in predicting the complex ground-motion prediction equations
peak ground motion at the site of interest. Within may be used containing high-order terms, focal
this context, new developments have been depth dependence, and site effects. It is worth to
Source Characterization for Earthquake Early Warn- The gray small box shows a zoom on the first few seconds
ing, Fig. 2 Schematic illustration of early warning of P-wave on velocity (top) and displacement (bottom)
parameters. From top to bottom, an example of the vertical records. The initial peak displacement (Pd) is measured as
component of acceleration, velocity, displacement, and the absolute maximum of displacement waveform on the
integral of squared velocity (IV2) signals. The ground- early portion of P-wave (typically 2–4 s) while the period
motion parameters PGA, PGV, and PGD are measured as parameter tc is measured from the ratio between initial
the absolute maximum along the signal, using accelera- displacement and velocity waveforms in the same time
tion, velocity, and displacement records, respectively. window
note that while magnitude is an ensemble mea- is used by regional early warning systems, for
sure for earthquake size, the definition of the which the characterization of the source is S
distance requires a specific metric, which is sen- performed by a network installed in the source
sitive to the ratio between the source-to-site dis- vicinity.
tance and the earthquake size. At distances A complementary approach is based on empir-
significantly larger than the source size, a point- ical relationships between a ground-motion
source approximation for the earthquake can be parameter Py measured in the early portion of
generally assumed, and R refers to the epicentral the P-wave train and the final PGX. This is phys-
or the hypocentral distance. In the near-source ically grounded on the first-order approximation
range, instead, finite-fault effects may be relevant that log Py has the same magnitude and distance
and a different metric for the distance could be dependence of log PGX, being differences con-
required. Anyhow, whatever choice of attenua- centrated only on static and possibly frequency-
tion relationships and distance metrics are done, dependent effects. In such a case, estimation of
the ground-motion prediction requires the knowl- source parameters is hidden in the common
edge of earthquake location and size. This model dependence and the uncertainty may be
significantly reduced by avoiding the estimation between the observed arrival times otm and otn
of magnitude and distance. Additionally, this at the same stations, since they share the same
approach does not require a seismic network to earthquake origin time:
constrain the source parameters and can be effi-
ciently used for single stations. Such a model is ðtttm tttn Þi , j , k ¼ otm otn , with m 6¼ n
the one implemented in on-site early warning (2)
systems.
Finally both models can be combined If a homogeneous velocity model is assumed, the
together, with source parameters estimated by a previous equation defines a 3D hyperbolic sur-
regional network and ground motion locally ver- face whose symmetry axis passes through the two
ified at a specific site. stations. With NT triggered stations, we obtain
NT(NT 1)/2 surfaces and the hypocenter is
Real-Time Location then located at the region crossed by the maxi-
A main concern for any early warning system is mum number of EDT surfaces.
the reliable estimation of earthquake hypocenter This description is modified in the evolution-
in real time. Recently, Satriano et al. (2008) have ary approach introduced by Satriano et al. (2008)
developed an evolutionary approach aimed at in which, at each time step, EDTs are evaluated
constraining the earthquake location, which starts not only for pairs of triggered stations but also
when the first station is triggered by the seismic for those pairs where only one station has
event and is updated as time passes. This tech- already triggered. This means that when the
nique is based on the equal differential time first station, let us say Sn, is triggered by the
(EDT) formulation and provides a probabilistic earthquake at the time tn, it is immediately pos-
density function for the earthquake location in a sible to limit the hypocentral position (Fig. 3b)
three-dimensional space accounting, at each time from the EDT surfaces defined considering that
step, for information from both triggered and any operational but not-yet-triggered receiver Sl
not-yet-triggered stations. In particular, with a will identify the seismic arrival at a later time tl
single, initial, recorded arrival time, the hypocen- (with tl tn) such that
ter has to belong to the Voronoi cell containing
the triggered station, which is created using the ðtttl tttn Þi , j , k 0, with l 6¼ n (3)
travel times to the not-yet-triggered stations.
As more and more stations identify the seismic The equality in the previous equation corre-
event, the location is constrained in the volume sponds to the set of points in the volume of
defined by the intersection of the Voronoi cells interest for which the travel time at the first sta-
for the remaining not-yet-triggered stations and tion is equal to the travel time at any remaining,
the EDT surfaces evaluated for all pairs of not-yet-triggered, receivers (conditional EDT).
already-triggered stations. As a consequence, the inequality delimitates the
Let us consider a seismic network (Fig. 3a) region in the volume V, bounded by the condi-
with N operational recording stations S1, . . ., SN tional EDT, that must contain the earthquake
and a gridded volume V containing both the net- hypocenter. In the case of a homogeneous prop-
work and the earthquake source. Preliminarily, agation medium, the region bounded at the time tn
we computed the travel times from each grid by the conditional EDTs evaluated for all the
point (i, j, k) in the volume V to each station of stations corresponds to the Voronoi cell for the
the network. If the earthquake hypocenter is at the receiver Sn. As the time goes on, the source vol-
node (i*, j*, k*) of the searching grid, the classical ume is bounded by the system of equations
EDT formulation prescribes that the difference
between the theoretical travel times tttm and tttn
ðtttl tttn Þi , j , k tcur tn , with l 6¼ n
from the event source to two stations Sm and Sn of
the network is exactly equal to the difference (4)
Source Characterization for Earthquake Early Warn- has not yet triggered is gained and the EDT surface bends
ing, Fig. 3 Sketch of the evolutionary earthquake loca- around the first triggered station. As a consequence, the
tion approach. For simplicity of the presentation, it is source volume decreases. (d) As the second station trig-
shown only a plane view of the epicentral location. (a) gers, a true EDT surface is defined. The intersection
The Voronoi cells associated with each station between this surface and the conditional EDT surfaces
(operational and not) of the network are a priori known. limits the likely source volume whose dimension is con-
(b) As the first station triggers, it is possible to identify a tinuing to decrease. (e) Two more true EDT surfaces
volume that is bounded by the conditional EDT surface on become available when the third station triggers, thus
which the travel time for the first receiver is equal to the better constraining the earthquake location. (f) As more
travel times for each not-yet-triggered station. This vol- and more stations trigger, the earthquake location con-
ume is likely to contain the true earthquake location. (c) verges to the standard EDT solution which is based on
As the time passes, further information from station that true EDT only (After Satriano et al. (2008))
provided that, at the current time tcur, only the which varies in the range [0, 1] and represents
station Sn has triggered at the previous time tn. As the PDF for hypocentral location at the grid cell
tcur increases, the hypocentral volume becomes (i, j, k).
smaller since the conditional EDT surfaces fold As the time increases and more and more
around the station Sn (Fig. 3c). stations trigger, the number of not-yet-triggered
It is possible to prescribe a probability density stations becomes smaller and smaller, and the
function (pdf) for the hypocentral volume associ- earthquake location converges toward the hypo-
ated with each inequality in the previous equa- center which would be obtained through the stan-
tions and for each grid point. We assign a value dard EDT approach using data from all
pn.l(i, j, k) = 1 if the inequality is satisfied and a operational stations of the network (Figs. 3d–f).
value pn.l(i, j, k) = 0 if it is not. Summing up over Tests performed on both synthetic and real data
all the stations, we obtain, for each grid point, a have shown that when a dense seismic network
non-normalized PDF P(i, j, k) whose maximum (i.e., with a mean station spacing of about 10 km)
value is (N1) for those grid points for which all is deployed around the fault zone, a location
inequalities in the equation are satisfied. accuracy is achieved within 1–3 s after the first
When the second station (and, as the time arrival detection.
progresses, further receivers) triggers, the equa-
tion is evaluated for all the possible pairs of Real-Time Magnitude Estimation and
triggered and not-yet-triggered stations. Then, Earthquake Rupture
true EDT surfaces are defined for each pair Magnitude estimation for early warning appli-
(n, m) of triggered stations computing, for each cations is based on empirical relationships relat-
grid point, the quantity ing the earthquake size with parameters
measured in the early portion of the P- and
8h i2 9 S-wave trains. These parameters are generally
>
< ðtttn tttm Þi, j, k ðotn otm Þ > =
associated to the low-frequency content of the
qn, m ði,j,kÞ ¼ exp ,
>
: 2s2 >
; data, which is sensitive to the seismic moment,
and can be related to the maximum amplitude,
with n 6¼ m the dominant frequency, or the energy released
(5) by the event. Associated proxies are peak
values, predominant period and integrated
where the expression in square brackets corre- measurements, respectively. Several authors
sponds to the standard EDT and s represents showed that the initial portion of recorded
the uncertainty associated with the arrival time P-waves carries information about the event
picking and travel-time computation. The quan- magnitude, both through its frequency content
tity qn, m(i, j, k) varies between 0 and 1. We then and amplitude (Allen and Kanamori 2003;
sum the qn, m’s with the updated pn, l’s to obtain a Kanamori 2005; Zollo et al. 2006; Wu and
new P(i, j, k) which now has a maximum value Zhao 2006; Böse et al. 2007; Wu and Kanamori
equal to 2008). A review of the common used parame-
ters is detailed in the next section; here we want
N T ð N T 1Þ to note that peak and energy estimates are
Pmax ¼ ðN N T ÞN T (6) dependent on the source-to-site distance, while
2
predominant period is pretty insensitive to the
being NT the number of triggered stations. Once epicentral location (Allen and Kanamori 2003).
we have evaluated P, we can define the function In all cases, when dealing with dense regional
networks surrounding the fault that generated
the earthquake, location is generally available
Pði, j, kÞ N
Qði, j, kÞ ¼ (7) before or at the same time the first estimates of
Pmax magnitude are performed. Therefore, it is not a
disadvantage to use distance-dependent param- two-slope behavior is expected around the deter-
eters as compared to distance-independent ones. ministic threshold. Hence, the use of a single
All of the parameters are estimated from mea- relationship for a broad magnitude range should
sures in the early portion of the signal: with the be statistically checked to avoid under/over esti-
goal of issuing an early warning for earthquakes mations of the magnitude at the limits of the
with magnitude larger than 5.5, the time scale in investigated range. Additionally, if any scaling
which performing the measurements is few sec- occurs for events with magnitude larger than
onds (2–4 s generally). 6, this indicates a different initiation process
The definition of a time window t0 for the for earthquakes that reach different sizes. Spe-
measurements corresponds to image a specific cifically, since the scaling of parameters is
area on the fault plane, delimited by the observed up to magnitude 7.5, this implies that
corresponding isochrone. It is defined as the set when we look at the earthquake rupture on a
of points whose radiation arrives at a given sta- given space scale characteristic of a magnitude
tion at the same time t0. Hence, the portion of the 6.0 event, we can probabilistically forecast if this
fault enlightened by the first few seconds of sig- rupture will soon stop or it will grow up to a larger
nal depends both on the relative location of the space scale. Beyond that limit, standard regression
station as compared to the fault and on the phase, laws saturate and a different approach is required
either P or S. Specifically, for the same time to capture the earthquake size for very big-size
window, S-waves image a larger area on the events. As a final comment, since the parameters
fault, because their speed is closer to the rupture have a different sensitivity to the slip increase, the
speed as compared to the P-waves, and the uncertainty associated to the magnitude estimation
regions explored by the different stations do not may change. Generally, peak parameters are more
overlap as much as P-waves. We then expect that sensitive to slip changes, the slope of the scaling is
inclusion of early portion of S-waves may signif- larger, and the uncertainty in the magnitude esti-
icantly constrain the estimation of the earthquake mation is smaller as compared to predominant
magnitude, as compared to the only use of period estimations.
P-waves. This approach is possible for close sta- We now discuss a real-time approach for mag-
tions (with epicentral distances smaller than nitude estimation based on a probabilistic evolu-
30 km) for which the analysis of S-waves does tionary approach (Lancieri and Zollo 2008). This
not significantly affect the lead time for early approach generalizes standard averages coming
warning. Additionally, isochrone mapping from single-station estimates of magnitude and
enhances that few seconds of P/S-waves corre- can include also a priori probability density func-
spond to an earthquake size of magnitude 6–6.5. tions for magnitude characterization. As
This indicates that the early portion of the signal discussed before, we assume a linear scaling
captures almost the whole rupture process up to between the logarithm of the early warning S
this magnitude, while effective prediction is pos- parameter Py and the final magnitude
sible beyond that limit. Using a kinematic
description of the rupture, we can argue that logPytheo ¼ A þ BM þ CðR=R0 Þ (8)
early warning parameters up to magnitude 6–6.5
change because both the ruptured area and the where C is a function of the distance R and R0 a
total average slip increase. Beyond that threshold, reference distance. Coefficients are generally dif-
these parameters image almost the same portion ferent for P- and S-waves. For low-frequency
of fault and any change with magnitude is only estimators that depend on the distance, the pre-
ascribed to any increase of the average slip in that dominant effect comes from geometrical spread-
region. ing and we can assume C(R) = K log R. After
From this consideration we can retain that defining the duration of the window for P- and
scaling of early warning parameters may be dif- S-waves, the computation of Py is performed by
ferent for different magnitude ranges, and a isolating the corresponding time windows after
the P- and S-wave arrivals. P-wave arrival is distribution P(PyjM) is assumed uniform
automatically picked, while S-wave arrival is beyond the magnitude saturation threshold.
computed using a theoretical travel-time data-
base derived by a 1D or 3D velocity model suit- Parameters for Magnitude Estimation
able for the area of investigation. Since S-wave In the context of real-time applications, different
estimates are significantly different from P-wave amplitude and period parameters have been pro-
estimates, we need to be sure that the selected posed to get independent estimates of the earth-
P-wave window is not contaminated by the fol- quake size. We present here an overview of the
lowing S-wave, to avoid any bias in the magni- parameters used by the different systems for the
tude estimation. If the S-wave arrival is expected magnitude estimation.
in the P-wave window, that P-wave measurement Following an original idea of Nakamura
is discarded. For a single Py measurement at a (1988), Allen and Kanamori (2003) first proposed
given station and for a given phase, the magni- the use of a period parameter, measured on the
tude can be estimated using the Bayes theorem: first few seconds of P-wave signal, to infer the
size of the ongoing earthquake. The predominant
PðPyjMÞPa ðMÞ period is measured on the vertical component of
PðMjPyÞ ¼ ð (9) velocity record and is defined as
PðPyjMÞPa ðMÞ
M
pffiffiffiffiffiffiffiffiffiffiffiffi
T pi ¼ 2p Xi =Di (11)
In the above formula the probability Pa(M) rep-
resents the a priori probability on the magnitude,
where
the probability P(PyjM) comes from the regres-
sion relationship, and, assuming a Gaussian dis-
tribution for log(Py), this writes Xi ¼ aXi1 þ x2i (12)

1 1 2 2
PðPyjMÞ ¼ pffiffiffiffiffiffiffiffi exp logPy logPytheo dx
2ps 2s Di ¼ aDi1 þ (13)
dt i
(10)
(s is the standard deviation of the fit). Finally the where Tpi is the predominant period at the i th
denominator is a normalization factor. This sample, xi is the recorded ground velocity, Xi is
approach can be directly plugged into a real- the smoothed squared ground velocity, Di is the
time evolutionary magnitude estimation where smoothed squared velocity derivative, and a is a
the probability is updated as new information smoothing constant. They showed that an empir-
comes from additional stations or phases. With- ical log-linear correlation exists between the
out any information, we assume an a priori mag- event magnitude and the maximum of Tpi
nitude distribution coming from the long-term (named Tpi max) within 2–4 s after the P-wave
earthquake catalog inspection (generally a arrival. Using a dataset of Californian earth-
Gutenberg-Richter distribution). To build up an quakes, they derived two linear relations between
evolutionary scheme, assuming to have com- Tpi max and magnitude, for small earthquakes
puted the probability density function for the (in the magnitude range 3–5) and for larger earth-
magnitude after n1 measurements of Py, this quakes (magnitude range 5–7.3). In particular,
function is then used as the a priori distribution for the upper magnitude class they found a
for the n th measurement. It is worth to note that log-linear relationship of the form:
this scheme allows accounting for magnitude
saturation. In this case the probability M ¼ 7:0log T pi max þ 5:9 (14)
a b –1
1.0 log(τc) = 0.21 (± 0.01) M – 1.20 (± 0.07) P 4-s
sdv = 0.25
0.8
0.6 –2
0.4
log(PD10Km)
0.2
log τc (s)
0.0 –3
–0.2
–0.4
–4
–0.6
–0.8
–1.0 –5
4 5 6 7 8 4.0 4.5 5.0 5.5 6.0 6.5 7.0
M Magnitude (Mjma)
c P/S velocity integral d log(PGV) = 0.73 (± 0.01) log(Pd)+ 1.30 (± 0.02)

4
4s of P wave 100 sdv = 0.41
2s of S wave
Whole duration
3
2
10
log IV210Km
0
1
–1
–2
0.1
–3
3.5 4 4.5 5 5.5 6 6.5 7 7.5 0.001 0.01 0.1 1 1.0
Magnitude Pd (cm)
Source Characterization for Earthquake Early Warn- function of magnitude (After Lancieri and Zollo (2008));
ing, Fig. 4 Empirical scaling relationships between early (c) the correlation between Pd and PGV (After Zollo
warning parameters and magnitude. (a) The average et al. (2010)); and (d) the scaling of the integral of squared
period as a function of magnitude (After Zollo velocity (IV2) with magnitude (After Festa et al. (2008))
et al. (2010)); (b) the initial peak displacement as a
S
A similar period parameter has been proposed by where u and v are displacement and velocity,
Kanamori (2005), who defined the average period respectively; the integrals are computed over a
(tc) of the first seconds of P-wave signal as time window (0, t0) starting from the P-wave
arrival and t0 is generally set at 3 s.
vðffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u t0 Using a database of Taiwan, Japan, and
u Italy earthquakes (magnitude range 4 < M
u u2 ðtÞdt
u < 8.3), Zollo et al. (2010) determined the rela-
tc ¼ 2puð0t0 (15)
t
v2 ðtÞdt tionship between average period and magnitude
0 (Fig. 4a). By measuring tc in seconds, they found
logðtc Þ ¼ 0:21ð0:01ÞM 1:19ð0:08Þ (16) estimation. For larger magnitudes a saturation

effect of IV2 prevents from a correct evaluation
Both period parameters (Tpi max and tc) are empir- of the event size. The saturation effect disappears
ically related to the event magnitude but are when the velocity integral is evaluated along the
pretty independent of the distance. whole signal duration, showing a robust log-linear
An alternative estimate of the earthquake size correlation up to larger magnitudes (~7).
can be obtained using amplitude parameters. In addition to the event magnitude, another
Zollo et al. (2006) showed that the low-pass fil- relevant and complementary piece of information
tered, peak displacement amplitude of initial to be provided in real time is the estimate of the
P- and S-wave seismic signals correlates with expected ground shaking at target sites. This lat-
the earthquake magnitude (Fig. 4b). The P- and ter represents an important aspect of the practical
the S-peak amplitudes are measured in a short implementation of an EEWS and is crucial for the
time window (2–4 s) after the arrival times of decision-making processes and the prompt acti-
P- and S-waves, respectively. The initial peak vation of security actions and emergency proce-
amplitude can be measured on the single vertical dures. With this aim, an empirical correlation
component or on the modulus of the displacement between the initial peak displacement (Pd) and
vector, as proposed by Lancieri and Zollo 2008. the final observed peak ground velocity (PGV)
In both cases the functional dependence of the can be used. Analyzing a database of earthquake
peak amplitude follows the more general rela- records from Japan, Taiwan, and Italy (2009 Mw
tionship, including also the dependence on the 6.3 L’Aquila mainshock and aftershocks) and
distance (Eq. 8), with the reference distance considering a maximum recorded distance of
R0 = 10 km. Hereinafter, we will adopt the initial 60 km, Zollo et al. (2010) obtained the following
P-peak amplitude of the vertical component and regression relation (Fig. 4d):
will refer to it as Pd.
Finally, Festa et al. (2008) investigated the logðPGV Þ ¼ 0:73ð0:01ÞlogðPd Þ þ 1:30ð0:02Þ
scaling of the early-radiated energy, with the (18)
final size of the event. The radiated energy can
be inferred from the squared velocity integral where PGV is in cm/s and Pd is in centimeters.
(IV2), which is measured on the initial portion Amplitude and period parameters thus provide
of P- and S-wave signals and is defined as complementary information about the ongoing
follows: earthquake, being them related to the expected
ground shaking at recording sites and to the earth-
ð tc þDtc
quake magnitude, respectively. With this in
IV2c ¼ vc 2 ðtÞdt (17)
tc mind, Wu and Kanamori (2008) proposed an
original, on-site alert-level scheme, based on the
where the subscript c refers to the P- or S-phase, tc combination of measured Pd and tc parameters.
is the corresponding first arrival, and vc is the The idea is to rapidly distinguish the case of a
particle velocity measured on the seismograms. small/large and close/faraway event from the
Moreover Dtc is the length of the signal along observed values of Pd and tc at each recording
which the analysis is performed. They found an site. Following this idea, Zollo et al. (2010) pro-
evident log-linear scaling of IV2 with magnitude posed a threshold-based approach to EEWS
for both P- and S-wave data, up to M = 5.8 which is aimed at the setup of local alert levels
(Fig. 4c). Beyond this value, the early energy based on a decision table. The key element of the
increases less with respect to the final magnitude. method is the real-time, simultaneous measure-
Thus, they suggested that early-radiated energy ment of initial peak displacement (Pd) and period
can be used to discriminate whether the event has parameter (tc) in a 3 s window after the first
a magnitude larger or smaller than 5.8, and only P-arrival time and on the use of the initial peak
in the latter case it allows for real-time magnitude displacement as a proxy for the PGV.
Source Characterization for Earthquake Early Warn- values over the area are interpolated and the PDZ is
ing, Fig. 5 Alert levels and threshold values for observed obtained by delimiting the Pd = 0.2 isoline. (c) Example
early warning parameters (After Zollo et al. (2010)). (a) of potential damage zone resulting from the interpolation
Pd versus tc diagram showing the chosen threshold values of measured and predicted Pd values. Gray triangles rep-
and the regions delimiting the different alert levels. Level resent the stations triggered by the earthquake, while red
3 = damage expected nearby and far away from the sta- and blue triangles show the alert level recorded at each
tion; level 2 = damage expected only nearby the station; station, as soon as 3 s of signal after the P-picking are
level 1 = damage expected only far away from the sta- available. The color transition from light blue to red
tion; level 0 = no expected damage (After Zollo delimits the potential damage, PDZ (After Colombelli
et al. (2010)). (b) Conceptual scheme for the potential et al. (2012a))
damage zone definition. Measured and predicted Pd
S
From the analysis of strong-motion data of interpreted in terms of potential damaging effects
Japan, Taiwan, and Italy, Zollo et al. (2010) cal- nearby the recording station and far away from
ibrated the threshold values, for the definition of it. For example, following the scheme of Fig. 5a,
four alert levels (0, 1, 2, 3) (Fig. 5a). The thresh- the maximum alert level (level 3, i.e., tc 0.6 s
old values have been established according to the and Pd 0.2 cm) corresponds to an earthquake
Pd vs. PGV and to the tc vs. M empirical relation- with predicted magnitude M > 6 and with an
ships (Fig. 4a, d). The threshold values corre- expected instrumental intensity (at the site) IMM
spond to a minimum magnitude M = 6 and to 7. This means that the earthquake is likely to
an instrumental intensity IMM = 7, assuming that have a large size and to be located close to the
the peak ground velocity provides the instrumen- recording site and a high level of damage is
tal intensity through the relationship of Wald therefore expected either nearby or far away
et al. (1999). The alert-level scheme can be from the recording station. On the contrary, in
case of a recorded alert level equal to 0 (tc < 0.6 s window when compared to farther stations, an
and Pd < 0.2 cm), the event is likely to be small efficient procedure should manage in real time
and far from the site, thus no damage is expected both the increasing of the time window at single
either close or far away from the station. The stations and different time windows, when com-
application of this method to a series of large bining information coming from different stations.
Japanese events (M > 6) (Colombelli The analysis of strong-motion data of the
et al. 2012a) confirmed that the threshold-based recent Mw 9.0, 2011 Tohoku-Oki mega-thrust
approach is a robust strategy to rapidly predict the earthquake confirmed the necessity of consider-
expected damage at recording sites, showing a ing larger time windows to overcome the problem
very good matching between the real-time of parameter saturation. Colombelli et al. (2012b)
assigned local alert levels and the final, observed proposed an evolutionary approach to early warn-
peak ground velocity, carried by the later arrival ing, in which amplitude and period parameters
of S-waves. (Pd and tc, respectively) are measured in progres-
sively increasing P-wave time windows. The
real-time measurements are no more done within
Extended Source Approach a fixed portion of signal but starting from the
P-wave arrival and expanding the time window
Expansion of the P-Wave Time Window for up to the arrival of the S-waves at each recording
Large Magnitude Events site. With such an approach, the standard meth-
There are no concerns, in fact, on the effectiveness odologies and the empirical regression relation-
of EEWS for the real-time characterization of ships can be extended to very large earthquakes,
small and moderate events, for which the fracture provided that appropriate time windows are
process is concluded within a few seconds. Any- selected for the measurements. With this
how, regression relationships between peak dis- approach data from different stations require a
placement, predominant period, and other specific weight depending on the available
integrated parameters based on 2–4 s of P-wave P-wave time window (PTW). Specifically, a
saturate beyond a magnitude threshold that can be weight proportional to the square of PTW was
estimated into the range 7–7.5. This saturation is shown to provide a good compromise between
owed to the limited window range that images a the window length and the number of stations. As
too small portion of the fault plane. When an shown in Fig. 6a, that peak displacement
earthquake extends for hundreds of kilometers on increases with PTW, indicating a final level of
the fault, its dynamics is expected to be controlled 8.5. Final underestimation may be related in this
by large asperities that the rupture can meet during case to the regression relationship, validated for
its propagation also far away from the hypocenter. smaller magnitude events and shorter distances,
Another aspect to be carefully checked is the filter to the cutoff frequency, or to the specific dynam-
used to get a reliable low-frequency signal. Gen- ics of the event, which exhibited a frequency-
erally the high-pass cutoff frequency is 0.075 Hz dependent radiation (Fig. 6b). On the other
that is comparable with the corner frequency of an hand, predominant period does not monotoni-
M 7 event. This means that up to magnitude 7, the cally increase with larger time window, indicat-
cutoff filter still includes the ensemble radiation ing that its use in real time could be sensitive to
emitted by the fault while for larger magnitudes the specific event to be analyzed.
the selected frequency range extracts only the con-
structive interference coming from specific Real-Time Estimation of Fault Rupture Extent
regions of the fault. Investigation of large magni- Using Strong-Motion Data
tude events hence requires a progressive extension Even if we improve the magnitude estimation for
of the P-wave time window, with update of early very large earthquakes, distance metrics might be
warning parameters at individual stations. Since at inadequate, mostly for targets that are close to the
fixed time close stations experienced a large time causative fault but far from the epicenter. In such a
a 3.0 9.0
17 5
36 22 8.5
2.5 100 70
11
logPd + 1.14 logR
135 30 20 8.0
2.0
Magnitude
89 48 7.5
153
122
1.5 146 7.0
166
6.5
1.0 165
6.0
0.5 5.5
166
0.0 5.0
b
0.8 9.5
9.0
100 70 36 17
Magnitude
log[τc(s)]
0.6 135 22 5 8.5

48 11
166 153 89 30
146 122 20 8.0
0.4 7.5
166 165
7.0
0.2
0 10 20 30 40 50 60 70
P-wave Time Window [s]
Source Characterization for Earthquake Early Warn- Both parameters exhibit saturation when a 25–30 s PTW is
ing, Fig. 6 Real-time evolution of average values of peak used: the saturation level is shown by the gray-dashed
displacement (Pd) (a) and predominant period (tc) (b) as a lines. For each plot, the corresponding magnitude scale
function of the P-wave time window used. Error bars are is also represented; this has been derived based on the
computed as the standard deviation associated to each coefficients of Eq. 1 and on the tc vs. M relationship
value; the gray number close to each point represents the determined by Zollo et al. (2010) (After Colombelli
number of stations used for each considered time window. et al. (2012b))
case ground-motion prediction could be signifi- square root of the sum of the squares of the
cantly underestimated. To this end, Yamada peak velocity on the horizontal components,
et al. (2007) proposed a method to classify record- recorded at the ith station, respectively. Thus,
ing stations in near- and far-source distance. If a the quantity S
dense seismic network is available, this subdivi-
sion can be used to infer information about the 1
Pi ¼ (20)
fault geometry. Analyzing strong-motion data 1 þ eFi
from past events, the authors found that the com-
bination of vertical acceleration and horizontal provides the probability that the i th station is in
velocity produces the best performance for stations the near- or far-source region. Specifically, the
classification. Their discriminant is given by recording site is classified as a near-source station
if the probability is greater than 1/2.
Fi ¼ 6:046 logZai þ 7:885logHvi 27:091 For real-time application, peak values used for
(19) recording site classification are computed from
incoming data every 10 s for each station and then
where Zai and Hvi represent the peak acceleration used in the discriminant function. As example, in
on the vertical component (in cm/s2) and the Fig. 7 the procedure is applied to the case of
a b
25° 25°
km km
0 50 0 50
24.5° 24.5°
24° 24°
23.5° 23.5°
23° 23°
120° 120.5° 121° 121.5° 122° 120° 120.5° 121° 121.5° 122°
10 seconds 20 seconds
c d
25° 25°
km km
0 50 0 50
24.5° 24.5°
24° 24°
23.5° 23.5°
23° 23°
120° 120.5° 121° 121.5° 122° 120° 120.5° 121° 121.5° 122°
30 seconds 40 seconds
Source Characterization for Earthquake Early Warn- recording stations. White zones correspond to the source
ing, Fig. 7 Snapshots showing the recording stations area and rectangle s identify the surface projection of the
identified as near-source sites in the case of the causative faults. The darker is the marker, the higher is the
1999 M7.0 Chi-Chi (Taiwan) earthquake. The star probability that the corresponding site is located in near
marks the earthquake epicenter. Circles represent the source (After Yamada et al. (2007))
the 1999, M7.0, Chi-Chi (Taiwan) earthquake. Potential Damaged Zone

As time goes on, the near-fault stations identify As for the point-source analysis, on-site
quite well the fault extent. approaches can be used to predict finite source
In a recent paper, Yamada (2014) proposed an effects, such as directivity and azimuthal changes
improved version of this discriminant. This on the ground motion directly analyzing the early
approach identifies the fault rupture geometry motion at the single sites, without explicitly esti-
by classifying stations into near-source and far- mating any source parameters. The idea is to use
away source and provides reasonably good esti- the threshold-based approach to compute thresh-
mates of the extent of the near-source area. old levels at single stations and to interpolate the
levels to define the potential damage zone asso- characteristic period tc. The area covered by sta-
ciated with the earthquake (Kanamori 2005; tions is divided into cells, using a prefixed spatial
Zollo et al. 2010). grid, which is needed to fill the gaps where sta-
At a single site, the real-time measure of the tions are not available. At those stations where
P-wave parameters (Pd and tc) thus provide a the first 3 s of signal after the P-picking are
rough but rapid alert notification, although no available, an alert level is locally assigned,
information is given about the accurate earth- based on Pd and tc measurements. At the same
quake size and location. Considering that the time, the event location is obtained by using the
same measure can be performed at different available P-picks and a real-time location
nodes of a dense array of stations, deployed method. Furthermore, the expected Pd value can
nearby and far away from the earthquake epicen- be predicted at each node of the grid, through
ter, the mapping of the recorded maximum alert Eq. 21 and the tc averaged over the available
levels (3 or 2) provides a preliminary estimation measurements at the considered time step. Mea-
of the extent of the potential damage zone (PDZ), sured and predicted Pd values are thus interpo-
i.e., the area where the damages are expected to lated; the area within which the highest level of
be equal or greater to those predicted by the level damage is expected can be delimited by the iso-
VII according to the instrumental intensity scale. line corresponding to Pd = Pdth = 0.2 cm.
The same authors proposed that a rough but The predicted PGV values at stations and at
very rapid estimation of the PDZ extent can also grid nodes can be computed using the equation
be made from the updated averages of the period relating log(PGV) and log(Pd) (Eq. 1 in Zollo
parameter tc which is recorded at strong-motion et al. 2010) and finally converted into an instru-
sites in the near-source region. Fixing the param- mental intensity measure using the relationship of
eter Pd at its threshold value and using the pro- Wald et al. (1999). An interpolation is performed
gressively updated estimations of the period between all intensity values to produce a real-
parameter, the empirical equation time, continuously updated, intensity map
(Fig. 5b,c). This procedure is repeated every
logPd ¼ A þ B logtc þ ClogR (21) time step (typically 1 s); as the waves propagate
within the area and trigger other stations, the
(R is the hypocentral distance; A,B,C are deter- event location is refined by using P-pickings,
mined by the regression analysis) can be used to average tc value is updated, and more data are
determine the radius of the area within which the used for the interpolation procedure. The studied
strong ground motion is expected to produce an cases for Japan earthquakes displayed a very
instrumental intensity IMM > 7. The off-line good matching between the rapidly predicted
application of the method showed a very consis- earthquake potential damage zone inferred from
tent match between the rapidly predicted (within initial P-peak displacement amplitudes and the S
a few sec from the first recorded P-wave) and instrumental intensity map, the latter being
observed damage zone, the latter being mapped mapped after the event, using peak ground veloc-
from detailed macroseismic surveys a few ity and/or acceleration, or from field
days after the event. A more refined technique macroseismic surveys. The performance of the
for the real-time mapping of the PDZ has been method has been evaluated by defining success-
proposed by Colombelli et al. (2012a), who tested ful, missed, and false alarms in terms of observed
the method on data from ten M > 6 Japanese versus predicted instrumental intensities for all
earthquakes occurred in the period 2000–2009 the analyzed events and by counting their relative
and recorded at the K-Net/Kik-Net Japan percentage. A very high percentage (88 %) of
accelerometric arrays. The method is based on alert levels has been correctly assigned, and
the recursive use of the ground-motion prediction most of maximum alert levels correspond to the
equation for Pd (Eq. 21), with real-time, updated area within which the highest level of damage has
estimate of the earthquake location and been observed.
a b 9
Slip [m]
41° 0 MNFPS MFF
20 40 60
8 MJ MA MW
Magnitude
40° 7 9.0
6
8.5
39°
5
8.0
50 100 150 200
4
38° 0 50 100 150 200
c Time from O.T. [s]
Timeline (sec from O.T.)

37° 1st 1st GPS 2nd GPS
GPS plateau plateau
M 8.15 M 8.4 M 8.9
36° 1m 1st 3rd 4th

data J MA J MA J MA
model M 7.2 M 7.9 M 8.1
km 0 2nd 50 100 150 200

35° L
0 100 J MA
M 7.6 S-waves PGV PGV
in Sendai in Sendai in Tokyo
138° 140° 142° 144°
Source Characterization for Earthquake Early Warn- provided by GPS data using the near-field, point source
ing, Fig. 8 (a) Example of snapshot of the GPS-based approximation (dark blue solid line) and resulting from
strategy for the Mw 9.0, 2011 Tohoku-Oki earthquake. the slip inversion (small blue squares). For comparison,
The background color represents the predicted intensity the evolution of magnitude estimate provided by the JMA
distribution using the final magnitude value and the dis- early warning system is also shown as a dotted gray line,
tance from the finite fault. The purple color scale shows and the continuous gray line represents the real moment
the slip distribution on the seven-patch slip model. The magnitude value. (c) Warning timeline for the Tohoku-
total length estimate (L) is also plotted as vectors on the Oki earthquake showing when the GPS information is
fault plane. Black vectors represent the observed horizon- available with respect to the time at which the strongest
tal offset while white vectors show the static displacement shaking occurs in the Sendai and Tokyo regions and with
resulting from the inversion algorithm. (b) Magnitude respect to the JMA warnings. After Colombelli et al
estimate as a function of time from the origin time. The (2013)
magnitude is obtained by using the static displacement
Summary obtain displacement time series. Unfortunately,

for near-field records, this operation may intro-
An approach that may be more robust is the duce artificial effects and long-period drifts
inversion for the final slip on the fault plane, (Boore et al. 2002), which are usually removed
which allows consideration of the contributions by applying a high-pass filter. The application of
from the entire fault plane and may provide a the filter, while removing the artificial distortions,
more realistic estimate of both size and potential reduces the low-frequency content of the
damage of the ongoing event. recorded waveforms, resulting in the complete
One way to recover the slip distribution on the loss of the low-frequency energy radiated by the
fault plane is to use the permanent ground defor- source and of the static offset, which is the most
mation, which is directly related to the earth- relevant piece of information for a large
quake magnitude. The static component of earthquake.
ground motion could, in principle, be obtained GPS stations are able to provide a direct and
using dynamic range, accelerometric sensors, evolutionary measurement of the permanent
which are able to record unsaturated signals in a ground deformation, i.e., of the resulting
broad range of frequencies (0
100 Hz). co-seismic displacement after the dynamic vibra-
Accelerometric records are integrated twice to tion has finished. With the increasing diffusion of
high-rate 1Hz GPS stations, the seismological References

community has begun looking at GPS data as a
valid complement to the high-frequency informa- Allen RM, Kanamori H (2003) The potential for earth-
quake early warning in southern California. Science
tion provided by seismic data.
3:685–848
Once the permanent static offset is extracted Allen RM, Ziv A (2011) Application of real-time GPS to
from GPS displacement time series, a real-time earthquake early warning. Geophys Res Lett
static slip inversion scheme can be used to infer 38, L16310. doi:10.1029/2011GL047947
Allen RM, Brown H, Hellweg M, Khainovski O,
the slip distribution on the fault plane for both the
Lombard P, Neuhauser D (2009) Real-time earthquake
rapid determination of the event size and for the detection and hazard assessment by ElarmS across
near-real-time estimation of the rupture area California. Geophys Res Lett 36, L00B08.
(Fig. 8). Many authors have recently started doi:10.1029/2008GL036766
Boore DM, Stephens CD, Joyner WB (2002) Comments
applying GPS data to EEW (Allen and Ziv
on baseline correction of digital strong-motion data:
2011; Crowell et al. 2012; Wright et al. 2012; examples from the 1999 hector mine, California, earth-
Ohta et al. 2012; Colombelli et al. 2013); they quake. Bull Seism Soc Am 92:1543–1560.
showed that 1 Hz GPS data provide a rapid and doi:10.1785/0120000926
Böse M, Ionescu C, Wenzel F (2007) Earthquake early
remarkably robust magnitude estimate and can be
warning for Bucharest, Romania: novel and revised
used for the real-time estimation of the rupture scaling relations. Geophys Res Lett 34, L07302.
area, which, in turns, would allow for an doi:10.1029/2007GL029396
improved prediction of the earthquake damaging Colombelli S, Amoroso O, Zollo A, Kanamori H (2012a)
Test of a threshold-based earthquake early warning
potential.
using Japanese data. Bull Seismol Soc Am
Another possibility would be a fast kinematic 102:1266–1275. doi:10.1785/0120110149
inversion of the rupture process, searching for the Colombelli S, Zollo A, Festa G, Kanamori H (2012b)
fault geometry, the focal mechanism, and the slip Early magnitude and potential damage zone estimates
for the great Mw 9 Tohoku-Oki earthquake. Geophys
distribution on the fault plane. A first approach
Res Lett 39, L22306. doi:10.1029/2012GL053923
was proposed by Dreger et al. (2005), who com- Colombelli S, Allen RM, Zollo A (2013) Application of
bined preliminary information on location and real-time GPS to earthquake early warning in subduc-
magnitude with scaling laws for source size to tion and strike-slip environments. J Geophys Res
118:3448–3461. doi:10.1002/jgrb.50242
determine both the focal mechanism and a kine-
Crowell BW, Bock Y, Melgar D (2012) Real-time inver-
matic rupture model from inversion of strong- sion of GPS data for finite fault modeling and rapid
motion integrated displacement waveforms. The hazard assessment. Geophys Res Lett 39, L09305.
approach revealed useful for estimating the doi:10.1029/2012GL051318
Cua G, Fisher M, Heaton T, Wiemer S (2009) Real-time
ground shaking in the near-fault domain for the
performance of the virtual seismologist earthquake
2003, M 6.5, San Simeon, California, earthquake. early warning algorithm in southern California seism.
Nevertheless, computational times were esti- Res Lett 80:740–747. doi:10.1785/gssrl.80.5.740
mated however of the order of minutes, making Dreger DS, Gee L, Lombard P, Murray MH, Romanowicz
B (2005) Rapid finite-source analysis and near-fault
S
the results available in near real time but not in
strong ground motions: application to the 2003
real time. Mw 6.5 San Simeon and 2004 Mw 6.0 Parkfield Earth-
quakes. Seism Res Lett 76:40–48. doi:10.1785/gssrl.
76.1.40
Cross-References Espinosa-Aranda JM, Cuéllar A, Rodrı́guez FH,
Frontana B, Ibarrola G, Islas R, Garcı́a A (2011) The
seismic alert system of Mexico (SASMEX): progress
▶ Early Earthquake Warning (EEW) System:
and its current applications. Soil Dyn Earth Eng
Overview 31:2154–2162. doi:10.1016/j.soildyn.2010.09.011
▶ Earthquake Magnitude Estimation Festa G, Lancieri M, Zollo A (2008) Magnitude estima-
▶ Engineering Characterization of Earthquake tion from early radiated energy. Geophys Res Lett
35, L22307. doi:10.1029/2008GL035576
Ground Motions
Kanamori H (2005) Real-time seismology and earthquake
▶ Seismic Actions Due to Near-Fault Ground damage mitigation. Ann Rev Earth Planet Sci 33:
Motion 195–214
3346 Spatial Filtering for Structural Health Monitoring
Lancieri M, Zollo A (2008) A Bayesian approach to the using dense accelerometer networks. Geophys J Int
real time estimation of magnitude from the early P- 183:963–974
and S-wave displacement peaks. J Geophys Res Zollo A, Colombelli S, Elia L, Emolo A, Festa G,
113, B12302. doi:10.1029/2007JB005386 Iannaccone G, Martino C, Gasparini P (2014) An inte-
Nakamura Y (1988) On the urgent earthquake detection grated regional and on-site earthquake early warning
and alarm system (UrEDAS) In: Proceedings 9th system for southern Italy: concepts, methodologies
world conference on earthquake engineering, Tokyo- and performances. In: Wenzel F, Zschau J (eds)
Kyoto, Japan vol 7, pp 673–678 Early warning for geological disasters - scientific
Ohta Y et al (2012) Quasi real-time fault model estimation methods and current practices. Springer-Verlag Berlin
for near-field tsunami forecasting based on RTK-GPS Heidelberg, Dordrecht
analysis: application to the 2011 Tohoku-Oki earth-
quake (Mw 9.0). J Geophys Res 117:B02311.
doi:10.1029/2011JB008750
Satriano C, Lomax A, Zollo A (2008) Real-time evolu-
tionary earthquake location for seismic early warning.
Bull Seimc Soc Am 98(3):1482–1494
Spatial Filtering for Structural Health
Satriano C, Elia L, Martino C, Lancieri M, Zollo A, Monitoring
Iannaccone G (2010) PRESTo, the earthquake
early warning system for southern Italy: concepts, G. Tondreau1 and A. Deraemaeker2
capabilities and future perspectives. Soil Dyn Earthq 1
Postdoctoral Researcher, Building Architecture
Eng 31(2):137–153. doi:10.1016/j.soildyn.2010.06.
008 and Town Planning (BATir), Brussels, Belgium
2
Wald DJ, Quitoriano V, Heaton T, Kanamori H (1999) FNRS Research Associate, Building
Relationships between peak ground acceleration, peak Architecture and Town Planning (BATir),
ground velocity and modified mercalli intensity in
Brussels, Belgium
California. Earthq Spectra 15:557–564
Wright TJ, Houlié N, Hildyard M, Iwabuchi T (2012)
Real-time, reliable magnitudes for large earthquakes
from 1 Hz GPS precise point positioning: the 2011 Synonyms
Tohoku-Oki (Japan) earthquake. Geophys Res Lett
39, L12302
Wu YM, Kanamori H (2005) Rapid assessment of damage Control charts; Damage localization; Damage
potential of earthquake in Taiwan from beginning of detection; Sensor networks; Spatial filtering;
P-waves. Bull Seism Soc Am 95:1181–1185. Structural health monitoring (SHM)
doi:10.1785/0120040193
Wu YM, Kanamori H (2008) Development of an earth-
quake early warning system using real-time strong
motion signals. Sensors 8:1–9 Introduction
Wu YM, Zhao L (2006) Magnitude estimation using the
first three seconds P-wave amplitude in earthquake
Assessing the integrity of the structures in real
early warning. Geophys Res Lett 33, L16312.
doi:10.1029/2006GL026871 time is a very important topic for which many
Yamada M (2014) Estimation of fault rupture extent using methods have been developed in the last decades.
near-source records for earthquake early warning. In Today, structural health monitoring (SHM) is
Wenzel F, Zschau J (eds) Early warning for geological
gaining more and more attention: in the case of
disasters. Advanced technologies in earth sciences.
Springer, Berlin/Heidelberg, p 29ff. doi:10.1007/978- bridges, the maximum loads tend to increase
3-642-12233-0_2 (increase of the vehicle weights), while most of
Yamada M, Heaton T, Beck J (2007) Real-time estimation the structures are coming to the end of their
of fault rupture extent using neat-source versus
theoretical lifetime. In addition, exceptional
far-source classification. Bull Seism Soc Am
97:1890–1910. doi:10.1785/0120060243 events such as collisions or earthquakes can
Zollo A, Lancieri M, Nielsen S (2006) Earthquake mag- cause more severe and fast deteriorations. Opti-
nitude estimation from peak amplitudes of very early mal maintenance calls for an early detection of
seismic signals on strong motion. Geophys Res Lett
33, L23312. doi:10.1029/2006GL027795
small damage in structures, as it is well known
Zollo A, Amoroso O, Lancieri M, Wu YM, Kanamori H that limited and frequent repairs are much less
(2010) A threshold-based earthquake early warning costly than major repairs or total rebuilding after
Spatial Filtering for Structural Health Monitoring 3347
collapse. Current monitoring practice consists being the Stonecutters Bridge with more than
in scheduled maintenances including visual 1,500 sensors among which are 58 accelerometers
inspections, ultrasounds, eddy current, magnetic and 853 dynamic and static strain sensors
field, or radiography techniques (Hellier 2003). (Ni et al. 2012). In Europe, the Messina bridge
All these experimental methods require however project, designed to be the largest cable-stayed
that the vicinity of the flaw is known and that the bridge in the world would include a very large
proximity to be inspected is accessible. More- monitoring system with more than 3,000 sensors
over, these local inspections are labor intensive (De Neumann et al. 2011).
and therefore very expensive. A major problem is These technological advances have opened
that traditional monitoring is noncontinuous the way to real-time automated SHM of bridges.
which means that if a critical damage occurs A major problem related to the use of such very
between two inspections, it might lead to cata- large sensor networks is to find adequate tech-
strophic structural failure. One of the most rele- niques to post-process the data: intelligent
vant examples is the I-35W Mississippi River methods are needed in order to take advantage
Bridge case (Rofidal 2007): this bridge collapsed of the enormous amount of information provided
in August 2007 killing 13 people and injuring by these large networks. In fact, operational
145, despite annual inspection. modal analysis is not yet fit for automated
A general trend for new structures and brid- modal analysis using very large sensor networks.
ges is a lighter and more slender design, which With that perspective, rather than identifying
tends to increase the levels of vibrations under online the full set of modal properties of the struc-
ambient excitation. While these levels of vibra- ture, an alternative is to condense the measured data
tions need to be controlled as they could be while keeping the information about the potential
detrimental to the lifetime of the structure, they damage occurring in the structure. The technology
can also be used for the continuous monitoring presented in this entry is spatial filtering, which
of the structure without disruption or decrease consists in using a linear combiner to condense the
of functionality. The basic idea is that the occur- information from very large networks of sensors
rence of damage alters the structural parameters into one or just a few “virtual sensors.” Such virtual
which in turn affect the vibration characteristics. sensors can be designed in order to react strongly to
Vibration SHM of civil engineering structures damage while being insensitive to environmental
relies on ambient vibrations, as artificial excita- changes or even to react to a damage occurring in
tion of such large structures is often unpractical. a specific location along the bridge.
Based on this basic concept, many vibration- Rytter (Rytter 1993) has proposed a hierarchi-
based SHM techniques have been developed in cal decomposition of the SHM process in four
the last decades using mainly eigenfrequencies, levels, which has been widely accepted in the
damping ratios, or mode shapes (Doebling SHM literature: detection (level 1), localization S
et al. 1998). The reason of this popularity is the (level 2), quantification (level 3) of the damage,
ease of measuring modal parameters or frequency and prediction of the remaining service life of the
responses on real structures thanks to recent damaged structure (level 4). As the level of SHM
advances in sensors and sensing systems and in increases, the knowledge about the damage
the development of efficient operational modal increases and, usually, the complexity of the
analysis techniques (Reynders and De Roeck method increases as well. The method based on
2008; Reynders et al. 2012). Such advances are spatial filtering can deal with levels 1 and 2 by
so important that more and more very large brid- comparing data measured in the current unknown
ges are instrumented with larger and larger sensor state with data measured from the structure
networks. China has been the driving force in this assumed to be undamaged.
direction with the massive instrumentation of The general scheme of the method is shown in
bridges in the Hong Kong area, the largest one Fig. 1. It is divided in three parts: (i) the
Spatial and Modal Filtering

Time-domain Consider a structure excited with an ambient
measurements force f(t) equipped with a network of n sensors
whose time-domain output is denoted by yk(t) as
shown in Fig. 2. The dynamic time-domain
Feature response at each sensor can be decomposed into
extraction a sum of contributions of the N mode shapes
excited by the ambient force:
X
N
yk ðtÞ ¼ ai ðtÞfki (1)
i¼1
Spatial Filtering for Structural Health Monitoring,
Fig. 1 General scheme of vibration-based SHM where ai(t) is the modal amplitude of mode i and
fki is the projection of mode shape i on sensor k.
The application of spatial filtering with coeffi-
measurement of raw time-domain data, including
cients ak leads to a single sensor output g(t):
data reduction, (ii) the transformation of the data
into information using feature extraction, and (iii)
X
n N X
X n
the diagnostics of the structural health based on gðtÞ ¼ ak yk ðtÞ ¼ ak fki ai ðtÞ (2)
the monitored features. k¼1 i¼1 k¼1
The first section of this entry deals with mea-
surement of raw time-domain data and details the The general scheme of spatial filtering can be
spatial filtering technique for data reduction. The used for different purposes. A first idea is to
second section is devoted to feature extraction, condense the information into the modal coordi-
and the third section deals with the diagnostics, nates of the undamaged structure, which allows
both for damage detection and localization. The to reduce the data yk(t) from a very large network
last section presents an experimental illustration of n sensors to a limited set of ai(t) time
of the SHM technique on a 3.78 m long steel series from N modal sensors. This idea is moti-
I-beam tested in the laboratory. vated by the fact that the vibration of structures
typically involves only a few mode shapes
which are excited in a given frequency band of
Measurement of Raw Time-Domain interest.
Data To design a modal filter, the vector of the
linear combiner ak must be orthogonal to all the
The first building block of the SHM system is the modes of the structure in a frequency band of
sensor network. As stated earlier, a current trend interest, except mode l:
is to implement very large sensor networks
on large civil engineering structures. In such X
n
ak fki ¼ dli (3)
cases, it is often necessary to perform data reduc-
k¼1
tion in order to decrease the power consumption
and the bandwidth needed to transmit the data Equation 2 then reduces to
and to facilitate the data storage and post-
processing. For SHM applications, an optimal gð t Þ ¼ al ð t Þ (4)
reduction is one that significantly reduces
the amount of data while keeping most of the Equation 3 can be written in a matrix form:
information about the damage. The technique
presented in this entry for data reduction is spa-
½CT fag ¼ el , (5)
tial filtering.
Spatial Filtering for

Structural Health
Monitoring, y1(t)
α1
Fig. 2 Principle of spatial
y2(t) + g(t)
filtering on a network of Structure α2 +
n sensors + Sensor
. . .
f αn
yn(t)
where {a} = {a1 an}T and el = {0 0 1

0 0}T (all components set to 0 except the
lth component) and [C] is a matrix whose col- dB
umns correspond to the N mode shapes projected Undamaged
on the n sensors of the array. Equation 5 can be
solved only if there are at least as many sensors as
there are mode shapes (n N). In this case,
matrix [C] is rectangular and the system of equa-
tions is overdetermined (several solutions exist
which satisfy Eq. 5). The minimum norm solution
is usually adopted by computing the pseudo-
inverse (regularized with singular value decom-
position) of [C]T (Deraemaeker et al. 2008). ω
The modal filter can be tuned to any of the ωl
N mode shapes in the frequency band of interest.
In the frequency domain, the FRF of the modal Spatial Filtering for Structural Health Monitoring,
filter tuned on mode l is given by Fig. 3 Effect of a structural change on the modal filter
tuned on mode l (Deraemaeker et al. 2008)
bl
GðoÞ ¼ , (6)
o2l o2 þ 2jxl ol o a numerical model of the structure to be moni-
tored. This can be achieved thanks to operational
where bl depends on the excitation level and modal analysis, using, for instance, stochastic
position. It corresponds to the FRF of a single subspace-based methods (Reynders and De
degree of freedom system which presents a single Roeck 2008).
peak at frequency ol. S
Using modal filtering, the amount of data from Effect of Damage on Modal Filters
a large network of n sensors can be significantly Suppose now that damage initiates in the struc-
reduced to just a few modal filters. Such modal ture. This damage will alter the stiffness matrix,
filters are virtual sensors which measure the affecting the eigenfrequencies and mode shapes
amplitude of vibration of each mode separately. of the structure. The change of mode shapes will
Typically, for a bridge excited by the ambiance, be reflected in matrix [C]T so that Eq. 5 will now
only a few (up to ten) mode shapes are relevant, be violated. In other terms, the damage will alter
and the information can be drastically reduced. the mode shapes, and the coefficients of the linear
The computation of the linear combiner coeffi- combiner will not be tuned anymore. This will
cients ak is based on Eq. 5 which requires the result in the reappearance of the filtered peaks, as
knowledge of matrix [C]. This matrix should be illustrated in Fig. 3. This is the central idea of
built using experimentally identified mode vibration-based SHM based on modal filters, as
shapes in order to avoid the need for detailed in Deraemaeker et al. (2008).
Spatial Filtering for s(ω)

Structural Health
Monitoring,
Fig. 4 Example of IPeak IPeak = 0,59
values for an
increasing peak
IPeak = 0,73
IPeak = 0,99
ωa ωi ωb ω
From Data to Information The frequency band is typically defined by oa =

0.95oi and ob = 1.05oi, where oi is the angular
The raw time-domain output of modal filters is eigenfrequency. A peak indicator is then com-
not exploitable as such, as the changes in the puted in this frequency interval:
time-domain response due to damage will be
very small. The transformation in the frequency pffiffiffiffiffiffiffiffiffiffiffiffiffi
2 3RV F
domain is an important step which allows to I Peak ¼ , (8)
enhance these small changes by focusing on the ob oa
frequency bands away from the main peak where
the filtered peaks will reappear. In the case of where RV F is the root variance frequency
ambient vibrations, the input force is not known, defined by
and the power spectral density (PSD) Sgg(o) of
g(t) should be computed. This quantity is directly sÐffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ob
o a ðo FCÞ2 sðoÞdo
related to the amplitude of the FRF G(o) as RV F ¼ Ð ob , (9)
follows (Ewins 1984): oa sðoÞdo
Sgg ðoÞ ¼ jGðoÞj2 Sf f ðoÞ (7) and FC is the frequency center defined by
This equation shows that if peaks are filtered in Ð ob

a osðoÞdo
the FRF G(o), they will be filtered in Sgg(o) as FC ¼ oÐ ob (10)
well. When damage occurs, the reappearance of oa sðoÞdo
spurious peaks should be monitored based on the
power spectral density of the output of the modal Theoretically, IPeak is equal to 1 if s(o) is
filter Sgg(o). constant and decreases when the peak grows.
Figure 4 gives an example of IPeak values com-
Feature Extraction puted between oa and ob when a spurious
Because the spurious peaks are expected to peak grows around oi. The advantage of that
appear around the initial eigenfrequencies of feature is that it is very sensitive to the peak
the structure, the strategy consists in extracting growth but not to the level of the excitation
one feature in each frequency band around them. force, which is particularly needed for ambient
Let s(o) be the frequency dependent amplitude vibrations. More details on the peak indicator
in the frequency range (oa, ob) (Fig. 4). computation can be found in Deraemaeker and
For ambient vibrations, s(o) is the PSD Sgg(o). Worden (2010).
Spatial Filtering for Structural Health Monitoring, Fig. 5 A typical control chart
Diagnostics the process stays in control. Those limits are

governed by the statistical distribution f(x) of
The last building block of the SHM system is the the quality characteristic and the probability
diagnostics. It consists in assessing, based on the 1 g that any in-control sample will fall inside
monitored features (here the peak indicators) the control limits. There are control limits that
whether the structure is healthy or damaged, and can be computed to detect a shift of the mean
possibly in giving indication on the location of value of the process or a shift of the variance of
damage. the process.
Once a set of reliable control charts has been
Statistical Analysis of the Features established (phase I), the process is under moni-
When the excitations are random, the peak indi- toring (phase II). The process state is unknown
cators behave like random variables. They will (it might be in or out of control), and if a sample
therefore follow a statistical distribution which falls outside the control limits previously com-
can be inferred from several undamaged samples. puted, it is considered as an abnormal value, and
Many tools have been developed to detect a warning is triggered. Phase I fixes the probabil-
a change in that statistical distribution such as ity of type I (false alarms) and type II (missing
outlier analysis or hypothesis testing. In this con- alarms) errors. Because the control limit values
tribution, control charts (Montgomery 2009; are based on the number of samples in the
Ryan 2000) are presented. This tool of statistical in-control set of data, the statistical distribution S
quality control plots the features or quantities f(x), and the g value, the statistical analysis must
representative of their statistical distribution as be done very carefully.
a function of the samples. Different univariate or Typically, one can find two families of control
multivariate control charts exist but all these con- charts in the literature: the univariate control
trol charts are based on the same principle which charts and the multivariate control charts. The
is summarized in Fig. 5. first family will be used if there is only one
In phase I, a set of samples are collected and feature to be monitored, while the second one is
analyzed to infer statistical characteristics of the used when several features are monitored at the
process when it is assumed to be in control (i.e., same time. In the present application, this means
when the structure is undamaged). The aim of this that if one checks the appearance of only one
step is to compute the control limits (upper con- spurious peak around one given natural fre-
trol limit UCL and/or lower control limits LCL) quency, the univariate control chart will be
between which the feature should be included if applied on that feature, while a multivariate
control chart will be used if one checks the spu- several features simultaneously. Consider
rious peaks around several eigenfrequencies in p features following a p-normal distribution.
each modal filter. Finally, there are two catego- The Hotelling T2 control chart monitors the
ries of control charts in each family: the Shewhart Mahalanobis distance T2:
control chart and the time-weighted control
charts. The first category monitors each sample X1
T 2 ¼ ðx xÞT ðx xÞ, (14)
independently while the second category con-
siders the previous samples to monitor the current
sample, which allows to detect smaller shifts. where S is the p p estimated covariance matrix
It has been found that the best results are obtained of features, x is the current p 1 feature vector,
when the Shewart control charts are applied, and x is the p 1 vector of estimated mean
because the time weighted control charts values of x vectors (only the undamaged samples
increased too much the number of type I errors. are considered to obtain S and x ). Since the
For this reason, only the univariate and the Mahalanobis distance is always positive, only
Hotelling T2 control charts are presented. the upper control limit UCL based on an
Individual control chart Consider that only F distribution is considered:
one feature x following a normal distribution is
monitored (e.g., one IPeak value in each modal pð m þ 1Þ ð m 1Þ
UCL ¼ Fg, p, mp , (15)
filter). The individual control chart will monitor ðm2 mpÞ
that individual feature x. The control limits are
where p is the number of variables, m is the
MR number of samples in the set of data in phase I,
Upper Control Limit: UCL ¼ x þ 3
d2
(11) and g is such that there is a probability of 1 g
MR that any in-control sample will fall between
Lower Control Limit: LCL ¼ x 3 the control limits. Like the individual control
d2
chart, the Hotelling T2 control chart detects
If the number of samples in phase I is n, then a change of x.
1X n Damage Detection and Localization

x¼ xi , (12) When condensing all the sensors into a single
n i¼1
virtual modal sensor as shown in Fig. 2, the
1X n statistical analysis can only give an indication
MR ¼ jxi xi1 j, (13) on the deviation from the normal condition on
n i¼2
the structure as a whole, leading to damage detec-
tion. The methodology can be extended to dam-
and d2 = 1.128. In fact, MR
d2 is an estimate of the age localization: consider now that the n sensors
standard deviation s of x assumed to follow installed on the structure are grouped in several
a normal distribution in phase I. Equation 11 is smaller sensor networks, each consisting of
therefore based on a choice of g = 0.027. The m sensors. Modal filters can be built for each of
individual control chart is designed to detect these local sensor networks resulting in indepen-
a shift of x. dent local modal filters (Fig. 6).
Hotelling T2 control chart If two or more If the local network I contains sensors y1,I, . . . ,
features are monitored at the same time, monitor- ym,I, the output of its modal filter tuned to mode
ing these two quantities independently by apply- l is given by
ing two or more univariate control charts can be
very misleading, especially if those features are X
m
correlated. On the opposite, the Hotelling T2 con- g I ðt Þ ¼ ak, I yk, I ðtÞ, (16)
trol chart is designed for the monitoring of k¼1
Spatial Filtering for Structural Health Monitoring, Fig. 6 Principle of damage localization using local modal filters
where the ak,I coefficients are computed in order localization capabilities. This highlights the
to satisfy the following condition: importance of the choice of the type of sensor
which is part of the first building block of the
X
m SHM system. Applications of spatial filtering
ak, I fðk, IÞi ¼ dli , (17) techniques have been limited so far to acceler-
k¼1 ometers or dynamic strain sensors.
where f(k,I)i is the kth (k = 1, . . . ,m) component

of the ith mode shape projected on the Ith local Illustrative Example
sensor network. If a damage occurs under spatial
filter I and if the sensor responses are locally Description of the Case Study
sensitive to damage, the mode shape will only The experimental application consists in a 3.78 m
be altered in that spatial filter. As a result, only the long steel I-beam which is bolted on two big
spatial filter I will have spurious peaks, indicating concrete cubes. The structure is excited with
the location of the damage. a Modal 110 electrodynamic shaker from MB
The efficiency of the approach relies there- Dynamics, and a network of 20 13 mm 170 mm
fore on a very strong assumption: damage in 50 mm low-cost PVDF sensors have been fixed S
a local filter will cause a local change of the with double-coated tape, providing a continuous
mode shape which is limited to the very close measurement of the dynamic strains along the
vicinity of the damage location. The fulfillment beam between sensors 1 and 20. A National
of this requirement depends on the type of mea- Instrument PXIe-1082 data acquisition system is
sured quantity which is considered. Two differ- used to measure the sensor responses with
ent approaches coexist in the literature: in a sampling frequency of 6,400 Hz, as well as to
Mendrok and Uhl (2010), accelerometers are generate a band-limited white noise between
used, while in Tondreau and Deraemaeker 0 and 500 Hz (not measured) which drives the
(2013), dynamic strain sensors are used. These shaker. Figure 7 shows the experimental setup as
two approaches have been compared in well as the definition of the PVDF sensors
Tondreau and Deraemaeker (2011), showing (accelerometers installed for preliminary tests
that the method based on strain sensors has can also be seen, but are not used in the present
a higher sensitivity to damage and better study).
Test setup.
Artist front view: definition of the PVDF sensors.
Spatial Filtering for Structural Health Monitoring, Fig. 7 Experimental setup: 3.78 m steel I-beam equipped with
20 dynamic strain sensors (PVDF) for damage localization
The damage is introduced by fixing a very Spatial Filtering for Structural Health Monitoring,
Table 1 Damage scenarios
small steel stiffener (35 mm 65 mm 17 mm)
directly against one of the PVDF sensors (Fig. 7a) Location of Location of
damage damage (local
in order to induce a local change of stiffness at Case Samples (sensor) filter)
that position. It has been checked that such a local 1 1–350 None None
change of stiffness induces a local change of 2 351–400 1 [1:5]
strain similar to what happens with damage. The 3 401–450 2 [1:5]
network of 20 sensors is split in five local filters 4 451–500 3 [1:5]
of five sensors, with a small overlap: (i) [1:5], (ii) 5 501–550 4 [1:5] and [4:8]
[4:8], (iii) [8:12], (iv) [12:16], and (v) [16:20]. 6 551–600 5 [1:5] and [4:8]
The damage scenarios are described in Table 1.
For the undamaged case, the measurement is
performed 350 times in order to infer the statisti- at 64 Hz for the modal filter tuned to 230 Hz, and
cal properties of the peak indicators. For each appearance of peak at 230 Hz for the modal filter
damaged case, 50 measurements are performed. tuned to 64 Hz). Note that the peak indicator used
Each measurement is referred to as a statistical in this example is slightly different from the peak
sample. indicator presented in section “Spatial and Modal
The modal filters tuned on the two first bend- Filtering” but shares similar properties. As the
ing mode shapes of the beam at 64 and 230 Hz are feature vector is multivariate, the Hotelling T2
applied for each of the five local filters separately. control chart has been applied to automate the
The feature vector therefore consists in two peak damage localization in each local filter. The first
indicators in each local filter (appearance of peak 200 undamaged samples have been considered to
a b
Local filter (1:5). Local filter (4:8).
Spatial Filtering for Structural Health Monitoring, Fig. 8 Automated damage detection in local filter (1:5) and
local filter (4:8)
a b
Local filter (8:12). Local filter (12:16).
Spatial Filtering for Structural Health Monitoring, Fig. 9 Automated damage detection in local filter (8:12) and
local filter (12:16) S
estimate the covariance matrix, as well as to Summary

compute the control limit (g is fixed to 0.25 %).
Figures 8, 9, and 10 show the Hotelling T2 There is a strong incentive for the development of
control chart. There is only one missing alarm online automated SHM techniques for large civil
in local filter [1:5] for a damage at sensor four. infrastructures. The objective of such systems is
However, that missing alarm is compensated to be able to assess the structural integrity of
thanks to the overlapping of the local filters. safety critical civil infrastructure in real time.
Indeed, the damage at sensors four and five is This is particularly important to detect the onset
correctly located in local filter [4:8]. The results of damage due to aging or more severe damage
show that the method has successfully, and auto- due to accidental event such as an earthquake or
matically, localized all the damage cases. a collision.
▶ System and Damage Identification of Civil

Structures
▶ Vibration-Based Damage Identification: The
Z24 Bridge Benchmark
References
De Neumann S, Andersen JE, Enckell M, Vullo E (2011)

Messina bridge – structural health monitoring
system. In: Proceedings IABSE-IASS conference.
London
Deraemaeker A, Worden K (2010) New trends in vibra-
tion based structural health monitoring. CISM courses
and lectures, vol 520. Springer Wien, New York
Spatial Filtering for Structural Health Monitoring, Deraemaeker A, Reynders E, De Roeck G, Kullaa J (2008)
Fig. 10 Automated damage detection in local filter Vibration-based structural health monitoring using
(16:20) output-only measurements under changing environ-
ment. Mech Syst Signal Pr 22:34–56
Doebling SW, Farrar C, Prime MB (1998) A summary
With the deployment of very large sensor net- review of vibration-based damage identification
methods. Shock Vib Dig 30:91–105
works on structures, alternatives to modal identifi- Ewins DJ (1984) Modal testing: theory and practice.
cation techniques can be interesting when the Research Studies Press, Letchworth
focus lies in fast and efficient damage detection Hellier C (2003) Handbook of nondestructive evaluation.
McGraw-Hill, New York
and localization. This is the aim of the method
Mendrok K, Uhl T (2010) Experimental verification of the
presented in this chapter. The three important damage localization procedure based on modal filter-
ingredients are (i) the use of a linear combiner to ing. Struct Health Monit Int J 20:1615–1630
perform data reduction in the time domain through Montgomery DC (2009) Statistical quality control:
modal filtering, (ii) the transformation of the time- a modern introduction. Wiley, New York
Ni YQ, Wong KY (2012) Integrating bridge structural
domain output of modal filters to the frequency health monitoring and condition-based maintenance
domain and the subsequent feature extraction to management. In: Proceedings of the CSHM-4. Berlin
detect the appearance of spurious peaks, and (iii) Reynders E, De Roeck G (2008) Reference-based com-
the use of control charts to automate the damage bined deterministic-stochastic subspace identification
for experimental and operational modal analysis.
detection and localization process. Mech Syst Signal Pr 22:617–637
The fully integrated and automated methodol- Reynders E, Houbrechts J, De Roeck G (2012) Fully
ogy allows to process efficiently data from large automated (operational) modal analysis. Mech Syst
Signal Pr 29:228–250
sensor networks and to condense it into very
Rofidal K (2007) U.S. coast guard responds following the
limited information for diagnostics in the form collapse of the I-35W bridge in Minnesota. USCG
of control charts. The efficiency of this technol- Reservist 54:26–29
ogy has been illustrated on a laboratory experi- Ryan TP (2000) Statistical methods for quality improve-
ment of a 3.78 m long steel I-beam. ment. Wiley, New York
Rytter A (1993) Vibration based inspection of civil engi-
neering structures. PhD thesis, University of Aalborg
Tondreau G, Deraemaeker A (2011). Comparison of dam-
Cross-References age localization based on modal filters using strain
measurements and acceleration measurements. In
▶ Operational Modal Analysis in Civil IWSHM2011, Standford, Sept 2011
Tondreau G, Deraemaeker A (2013) Local modal
Engineering: An Overview
filters for automated data-based damage localization
▶ Stochastic Structural Identification from using ambient vibrations. Mech Syst Signal Pr 39:
Vibrational and Environmental Data 162–180
Spatial Variability of Ground Motion: Seismic Analysis 3357
Spatial Variability of Seismic Motion

Spatial Variability of Ground Motion: (SVGM): What It Is and How to
Seismic Analysis Describe It
Mounir Khaled Berrah1, Djawad Zendagui2 and Seismic Motion and Its Measurement
Mustapha Djafour2 The seismic motion is regarded as one of the most
1
Ecole Nationale Polytechnique, Algiers, Algeria important and unknown load which acts on
2
Risam Laboratory, Faculty of Technology, a structure. This huge motion results from mainly
University Abou Bakr Belkaı̈d, Tlemcen, two causes: (a) explosive volcanic eruptions
Algeria which are very common in areas of volcanic
activity and (b) tectonic activity associated with
plate margins and faults. The majority of earth-
Synonyms quakes worldwide are of this second type. Thus,
the main cause of earthquakes is either the sudden
Coherency; Dynamic; Pseudo-static; Seismic movement of various plate boundaries or when
analysis; Spatial variability; SVGM plates scrape against each other. Some earth-
quakes are also caused by old plate boundaries
or faults. The point from which starts this sudden
Introduction movement is the epicenter of the earthquake. This
point could be between 10 and 400 km below the
The seismic design of structures considers that all surface of the earth.
ground supports are subjected simultaneously to Following this sudden movement, many
identical seismic excitation. This assumption is waves radiate from the epicenter and propagate
not valid for extended structures since it has been through the earth. It has been recognized that
recognized that seismic motion exhibits no neg- both P and S waves, known as body waves,
ligible variability. This variability, called spatial emanate from the source and travel with
variability of ground motion (SVGM), must a velocity that exceeds 4 km/s in earth. When
hence be considered in the design of extended they arrive at the free field, they are followed by
structures. both Love and Rayleigh waves. These latter,
Many questions arise: what is SVGM and how known as surface waves, travel only at the
does one model it? Does SVGM increase or surface. The movements associated with body
decrease internal forces of a structure? Is there as well as surface waves are well known and
any general conclusion that can be drawn while are mainly decomposed into compressional
considering SVGM? and shear movement. At the free field, the seis-
This entry aims at answering these mic motion could be decomposed into two parts: S
questions. The first part of this entry addresses (a) motion caused by body waves and (b) motion
the description of SVGM through models in caused by surface waves.
terms of coherency function. Subsequently, the The seismic motion is recorded by seismome-
response of structures subjected to SVGM is ters which are deployed in seismic areas. The
derived. At this stage, the concept of pseudo- number of seismometers is growing from year
static and dynamic responses is introduced. to year denoting the need for a better understand-
To understand the effect of the SVGM on struc- ing of the seismic motion. While some of them
tures, two case studies are presented: a single are deployed in a single scheme, others are
degree of freedom (SDOF) system and a multi- deployed in grouped schemes. This latter scheme
degree of freedom (MDOF) system, subjected forms what is called the dense seismic array.
to SVGM. A typical dense array configuration is presented
3358 Spatial Variability of Ground Motion: Seismic Analysis
the separating distance between those points is

somewhat important. However, data from dense
seismic array show that even if the separating
distance between two points is not very important
(sometimes less than 50 m), the recorded motion
at both points exhibits some differences.
The effect of this difference, called spatial
variability of earthquake motion (SVEGM), on
extended structures is addressed in this entry.
Other sources of variability such as the result
from relative surface fault motion for recording
stations located on either side of a causative fault,
soil liquefaction, and landslides (Zerva and
Zervas 2002) are not considered herein. Finally,
although SVEGM has been extensively studied
for horizontal component and at free surface,
Spatial Variability of Ground Motion: Seismic Anal- some dense seismic arrays have been
ysis, Fig. 1 Typical dense array configuration instrumented to record motion with respect to
vertical component and/or at depth.
What causes SVEGM? It has been recognized
in Fig. 1 where in general a central station S1 is that SVEGM is caused by several effects (Der
surrounded by other stations such as station S2 Kiureghian 1996) (Fig. 1):
with a separating distance.
Many arrays have been and are being 1. Incoherence effects due to scattering in the
deployed around the world to assess the charac- heterogeneous ground and extended source
teristics of seismic ground motions. Among them effects.
SMART-1 array “Strong Motion ARray in Tai- 2. Traveling-wave effects, in which nonvertical
wan,” located in Lotung, in the north-east corner waves reach different points on the ground
of Taiwan deployed in 1980 remains one of the surface at different times, producing a time
most important. Other dense seismic arrays have shift between the motions at those points; this
been deployed (Zerva 2009). The deployment of effect is also known as “wave passage effect.”
these arrays enhanced greatly our knowledge on 3. Site effects due to the variation by the filtering
how the SVGM affects structure and what the effects of overlying soil columns; this effect is
main parameters used to describe the SVGM are. known as “site effect.”
What Is SVEGM? Although the attenuation of seismic motion

Let’s consider that the medium is described using increases the variability of seismic motion, it is
the Cartesian coordinate x = (x, y, z) and not usually included since its effect on structures
uðx, tÞ ¼ f uðx, tÞ vðx, tÞ wðx, tÞ gT is the seis- will not be visible below a separating distance of
mic motion measured along these axis. Hence, a dozen kilometers, which is greater than the
v(x, t) is the vertical motion, whereas u(x, t) and usual separating distance for structures.
w(x, t) are the horizontal motions. At any point of
the soil, both horizontal and vertical motions How to Measure and Describe SVEGM
could be recorded. For instance, let’s consider For the purpose of simplicity, two points A and
two points A and B at the free surface. Recorded B located at the free surface with z = 0 are con-
earthquake along a particular axis at two points is sidered. Thus, coordinates of both points are,
obviously different in terms of amplitude and respectively, xA = (xA, 0, 0) and xB = (xB, 0, 0).
frequency. This difference is well accepted if Even if the motion could be recorded along the
Spatial Variability of
Ground Motion: Seismic
Analysis,
Fig. 2 Propagation of
seismic waves
three axes, only the horizontal motion which is where

along the x axis, i.e., u(xA, t) and u(xB, t), is con- Im(SAB(o)) and Re(SAB(o)) stands for imagi-
sidered herein (Fig. 2). nary and real parts of SAB(o).
The SVEGM is described by models in terms jrAB(o)j is the lagged coherency and
of correlation or coherence functions. This latter, Re(SAB(o)) as unlagged coherency. It is worth not-
which is the most used, is the ratio of the crossing that the coherency function is complex, and
spectral density function of the motion at two lagged coherency is usually called coherence func-
points separated by a horizontal offset to the tion. This latter describes the similarity of the wave-
square root of the product between the auto- forms at two stations without taking into account
spectral density functions at the two points: the difference in the arrival times of the waves.
If the subscript A and B are skipped and
SAB ðoÞ replaced by the separating distance L, then the
rAB ðoÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (1)
SA ðoÞSB ðoÞ coherency function is rewritten as
where rðL, oÞ ¼ jrðL, oÞjexp½iyðL, oÞ (4)
L ¼ xB xA , This coherency function measures the coher-

ency along the x axis and at the free surface. It
SAB(o) is the cross-spectral density function of could be measured at depth by, for instance, con-
the motion along the x axis for the two points sidering that both points A and B are located at S
A and B. the same depth, i.e., y = h, which leads to the
SA(o) and SB(o) are the auto-spectral density following coherency function
functions of the motion along the x axis at,
respectively, points A and B. rðh, L, oÞ ¼ jrðh, L, oÞjexp½iyðh, L, oÞ (5)
The coherency function is often written as
Models in terms of coherency functions have
rAB ðoÞ ¼ jrAB ðoÞjexp½iyAB ðoÞ (2) been developed during last decades (Abrahamson
et al. 1991; Der Kiureghian 1996; Zendagui
et al. 1999). They could be divided into three
with
categories: empirical, semiempirical, and physi-
cally based models. An exhaustive list of
ImðSAB ðoÞÞ
1
yAB ðoÞ ¼ tan (3) these models could be found in Zerva (2009).
ReðSAB ðoÞÞ The common feature of these models is that the
coherency function decreases with separating To write the governing equations in a form
distance and frequency. similar to that of the case of a single excitation,
the displacements are separated into two parts:
Generation of Variable Seismic Excitation and
Response Spectra V us U
¼ þ (7)
The generation of variable seismic excitation has vs vs 0
been extensively studied. The method developed
In this equation, us is the vector of structural
by Shinozuka (1971) remains the most popular.
displacements due to static application of pre-
This method has been extended and modified by
scribed support displacements vs at each time
others to take into account nonstationarity of pro-
instant. The two parts are related as
cesses (Deodatis 1996). A spectral-representation-
based simulation algorithm has been developed by

k ks us 0
Deodatis (1996). The simulation uses the coher- ¼ (8)
kTs kss vs Rs ð t Þ
ency function and allows the generation of accel-
eration as well as displacement time histories. This Rs(t) are the support forces necessary to stati-
latter is necessary while performing non linear cally impose displacements vs that vary with
structural analysis. A recent method developed time; obviously, us varies with time and is there-
by Benmansour et al. (2012) solves the problem fore known as the vector of quasi-static displace-
of integrability of seismic motion to avoid the use ments. Observe that Rs(t) = 0 in the case of
of baseline correction. identical support ground motion. The displace-
Alternatively, Berrah and Kausel (1992) ments of the superstructure U are known as
developed an interesting method that modifies a dynamic displacements.
particular response spectrum to take into With the total structural displacements split
account SVGM. into quasi-static and dynamic displacements,
consider hence the first of the two partitioned
equations (Eq. 6):
Structure Subjected to SVGM: General
Derivation mV€ þ ms v€s þ cV_ þ cs v_s þ kV þ ks vs ¼ 0 (9)
Consider a structure with N DOFs. For the anal- Substituting Eq. 7 and transferring all terms
ysis of such a system, consider that the displace- involving vs and us to the right-hand side lead to
ment vector contains two parts: (1) V includes the
N DOFs of the superstructure, and (2) vs contains € þ cU_ þ kU ¼ Peff ðtÞ
mU (10)
the Ng components of support displacements. The
equation of dynamic equilibrium is written as where the vector of effective earthquake forces is

( )
( ) us þ ms v€s þ cu_s þ cs v_s þ kus þ ks vs Þ
Peff ðtÞ ¼ ðm€
m ms V€ c cs V_
þ T (11)
mTs mss v€s cs css v_s

(6)
k ks V 0 Equation 8 gives
þ ¼
kTs kss vs RðtÞ kus þ ks vs ¼ 0 (12)
The mass, damping, and stiffness matrices can This relation also enables us to express the
be determined from properties of the structure, quasi-static displacement us in terms of the spec-
while the support motions v€s ðtÞ, v_s ðtÞ, and vs(t) ified support displacements vs:
must be specified. It is desired to determine the us ¼ ½Lvs (13)
displacements V in the superstructure DOF and
the support forces R(t). where [L] = k1 ks is the influence matrix.
Equation 13 can be written in a different form as The solution of Eq. 19 can be written as
XN s XN s
us ¼ i¼1
Li vsi ðtÞ (14) qn ð t Þ ¼ Gni Dni ðtÞ (21)
i¼1
where Li is the ith column of the influence matrix

where Dni(t) is the displacement response of
[L].
the nth-mode SDOF system to support accel-
Substituting Eqs. 12 and 13 in Eq. 11, one gets
eration v€si ðtÞ:
The displacement response of the structure
vs ðtÞ ðc½L þ cs Þv_s
Peff ðtÞ ¼ ðm½L þ ms Þ€ (Eq. 7) contains two parts:
(15)
1. The dynamic displacements are obtained by
For many practical applications, further sim- combining Eqs. 18 and 21:
plification of Peff is possible in two stages. First,
the damping term is usually small relative to the XN s XN
inertia term and can be dropped. Second, for U ðt Þ ¼ i¼1 n¼1
Gni fn Dni ðtÞ (22)
structures with mass idealized as lumped at the
DOFs, the mass matrix is diagonal, implying that
2. The quasi-static displacements us are given by
ms is a null matrix and m is diagonal. With these
Eq. 14.
simplifications, Peff is expressed as
Combining the two parts gives the total dis-
Peff ðtÞ ¼ m½L€
vs ðtÞ (16) placements of the structure
By using Eq. 14, Peff can be expressed as XN s XNs XN

V ðtÞ ¼ i¼1
Li vsi ðtÞ þ i¼1 n¼1
Gni fn Dni ðtÞ
XNs (23)
Peff ðtÞ ¼ i¼1
mLi v€si ðtÞ (17)
The equivalent static forces in structural DOFs

The dynamic response U can be computed by
are given by
modal analysis, as the superposition of the modal
contributions
F ¼ kV þ ks vs (24)
XN
U ðt Þ ¼ fn qn ðtÞ (18)
n¼1 Substituting Eq. 7 for V and using Eq. 12 give
where fn are the natural modes and qn are scalar S

F ¼ kU ðtÞ (25)
multipliers called modal coordinates. Instead of
Eq. 18, the modal equation can be expressed for
These forces depend on the dynamic displace-
the nth mode as
ments only given by Eq. 22. Therefore,
XN s
q€n þ 2xn on q_n þ o2n qn ¼ Gni v€si ðtÞ XN s XN
i¼1
Fð t Þ ¼ i¼1 n¼1
Gni kfn Dni ðtÞ (26)
(19)
where The equivalent static forces along support

DOFs are also given by the last term on the left-
Lni hand side of Eq. 6:
Gni ¼ , Lni ¼ fTn mLi and Mn ¼ fTn mfn
Mn
(20) Fs ¼ kTs V þ kss vs (27)
Spatial Variability of u(t)

Analysis, Fig. 3 A SDOF
system subjected to SVGM ×G × G’
m
k1
k2
1 1’
2 2’
ug1(t)
ug2(t)
Substituting Eq. 7 and using Eq. 8 for the u þ ðk1 þ k2 Þu ¼ k1 ug1 ðtÞ þ k2 ug2 ðtÞ
m€ (29)
quasi-static support forces Rs(t) give
The natural frequency is defined by
Fs ðtÞ ¼ kTs U ðtÞ þ Rs ðtÞ (28)
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
k1 þ k2
Application 1: SDOF System Subjected o¼ (30)
m
to SVGM
The quasi-static displacement is thus
Derivation of the Relative Displacement
Consider a simple multi-support structure
k1 ug1 ðtÞ þ k2 ug2 ðtÞ
represented by a SDOF system with two supports us ðtÞ ¼ (31)
1 and 2 excited by, respectively, ug1(t) and ug2(t) k1 þ k2
(Fig. 3). The columns are assumed to be weightless
and inextensible in the vertical (axial) direction, At this stage, it is considered that the support
and the resistance to girder displacement provided motions are harmonic variable both in amplitude
by each column is represented by its spring con- and phase:
stant k1 and k2. For the purpose of simplicity, it is
assumed that the slab, which has a mass m, is ug1 ðtÞ ¼ A1 sin o1 t
(32)
infinitely rigid and that damping is neglected. ug2 ðtÞ ¼ A2 sin o2 t
The total displacement of the mass from the initial
position G to the final position G’ is u(t) The total dynamic and quasi-static displace-
The forces acting on the slab: ments are hence
• Inertial forces: f I ¼ m€
u. k1 A1
uðtÞ ¼ sin o1 t
• Elastic force on column (1): fe1 = k1(u ug1) k1 þ k2 1 o1 2
• Elastic force on column (2): fe2 = k2(u ug2). o
k2 A2
þ sin o2 t
k 1 þ k 2 1 o 2 2
Hence, the equation of motion is an expression (33)
of the equilibrium of these forces given by o
Spatial Variability of 20
Ground Motion: Seismic Uniform
Analysis, Fig. 4 Variation 15 Variable
of elastic forces at column
1 for case S-1 10
Elastic force (kN)

5 ← Funif max=3.03 kN
−5
−10
−15
← Fvar max=16.87 kN
−20
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (sec)
k1 k2 value than uniform load (continuous line)

us ð t Þ ¼ A1 sin o1 t þ A2 sin o2 t
k1 þ k2 k1 þ k2 (Figs. 4 and 5). Thus, for these cases, it seems
(34) that SVGM induces greater values of elastic
forces than uniform seismic motion.
Finally, the dynamic response could be What about the variation of the quasi-static
derived as follows: and dynamic components for both cases? Results
show great difference on the part of both compo-
U ðtÞ ¼ uðtÞ us ðtÞ (35) nents depending on frequency of excitations.
Hence, by analyzing the variations of the two
Case Study components, quasi-static and dynamic (Figs. 6
As the purpose of this entry is to see how SVGM and 7), it is found that for case S-1 the amplifica-
affects a structure, it will be considered hereafter tion caused by variable motion is mainly due to
two cases: a uniform seismic motion at supports the importance of the quasi-static component. In
1 and 2 equal to ug1 and a variable seismic this case, the frequency of the structure is larger
motion, i.e., ug1 at support 1 and ug2 at support than those of the imposed displacement. The
2. The natural frequency is equal to 24 rad/s. The SDOF system could be considered herein as
results are presented in terms of elastic forces of a rigid structure. However, for case S-2, the fre-
the column (1). For the case of spatially variable quency of the structure is close to the frequency
ground motion, two cases will be considered: of the excitation. In this case, the dynamic com- S
ponent is the main cause of the amplification of
ug1 ¼ 0:1 sin 10t the response.
Case S-1 ðm Þ
ug2 ¼ 0:1 sin 15t
(36)
Application 2: MDOF System Subjected
ug1 ¼ 0:1 sin 20t to SVGM
Case S-2 ðm Þ
ug2 ¼ 0:1 sin 25t
(37) Derivation of the Relative Displacements
In this part, responses of a MDOF system under
The frequencies of excitation have been cho- uniform and variable seismic input will be com-
sen below and above the natural frequency. It is puted. It is assumed herein that the MDOF is
interesting to see that for both cases (Eqs. 36 and composed of five masses connected by five
37), SVGM (dashed line) exhibits greater springs (Fig. 8). The motion, in terms of

← Fvar max=93.38 kN Uniform
80 Variable
Analysis, Fig. 5 Variation
of elastic forces at column 60
1 for case S-2
40
Elastic force (kN)

20
0
−20
−40 ← Funif max=
32.73 kN
−60
−80
−100
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (sec)
Ground Motion: Seismic ← Fq-s max=13.72 kN Quasi-static
Analysis, Fig. 6 Variation dynamic
of the pseudo-static and 10

dynamic components of the
Elastic force (kN)
elastic forces at column ← Fdyn max=5.95 kN

1 for case S-1 5
−5
−10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (sec)

← Fdyn max=93.09 kN
Ground Motion: Seismic Quasi-static
80 dynamic
Analysis, Fig. 7 Variation
of the pseudo-static and 60
dynamic components of the
40
Elastic force (kN)
elastic forces at column

1 for case S-2 20 ← Fq-s max=14.18 kN
0
−20
−40
−60
−80
−100
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (sec)
Initial
position
Deformed
position
Spatial Variability of Ground Motion: Seismic Analysis, Fig. 8 MDOF system under SVGM
displacements, is imposed at the outer masses into two parts fvs g ¼ h vs1 vs3 vs5 iT and
and the middle one, vs1, vs3 et vs5. Hence, only fV g ¼ h V 2 V 4 i T .
two masses do not have imposed displacements The equation of dynamic equilibrium for all
and their displacements are noted: V2 et V4. Thus, the DOFs is written as
the displacement vector could be decomposed
2 38 9
m1 > v€s1 >
> >
6 >€ >
7> >
6 m2 7< V 2 =
6 m3 7 v€s3
6 7> >
4 m4 > V€4 >
5> >
: >
> ;
m5 v€s5
2 38 9 8 9
k1 k1 > vs1 >
> > >
> R1 >
>
6 k1 k1 þ k2 þ k5 k2 k5 7>>
< V
>
>
=
>
>
< 0 >
>
=
6 7 2
þ6
6 k2 k2 þ k3 k3 7 vs3 ¼ R3
7> > > > (38)
4 k5 k3 k3 þ k4 þ k5 k4 5>> V4 >
> > 0 >
> > >
: > : >
; > ;
k4 k4 vs5 R5
By considering only DOFs 2 and 4, Thus,

( )
s
1
m2 0 V€ 2 k1 þ k2 þ k5 k5 V2 u2 k1 þ k2 þ k5 k5
þ fus g ¼ ¼
0 m4 V€ 4 k5 k3 þ k4 þ k5 V4 us 4 k5 k þ k þ k5
8 9 8 3 94 S

> vs1 >
vs1 >
>
k1 vs1 þ k2 vs3 k1 k2 0 < = k1 k2 0 < =
¼ ¼ vs3 vs3
k3 vs3 þ k4 vs5 0 k3 k4 > : > ; 0 k3 k4 > : > ;
vs5 vs5
(39) (41)
The pseudo-static component is obtained by If ki = k
neglecting the inertial forces 8 9

< vs1 =

s us 2 1 3 4 1
k1 þ k2 þ k5 k5 u2 fus g ¼ ¼ v (42)
0þ us 4 8 1 4 3 : s3 ;
k5 k þ k4 þ k5 us 4 vs5
83 9

> vs1 > (40)
k1 k2 0 < = The equation of movement could be rewritten
¼ vs3 in terms of the dynamic component
0 k3 k4 > : > ;
vs5 {U} = {V} {us}:
( )
8 9
m2 0 €2
U k1 þ k2 þ k5 k5 U2 < vs1 ¼ 0:1 sin 30t =
þ
0 m4 €
U4 k5 k3 þ k4 þ k5 U4 v ¼ 0:1 sin 35t ðmÞ (47)
: s3 ;

1 vs5 ¼ 0:1 sin 30t
m2 0 k 1 þ k2 þ k5 k5
¼
0 m4 k5 k3 þ k4 þ k5
8 9 Case M-3 variability in terms of temporal delay

> v€s1 >
k1 k2 0 < = and frequency close to those of the system:
v€s3
0 k3 k4 >
: > ; 8 9
v€s5 < vs1 ¼ 0:1 sin 30t =
(43) vs3 ¼ 0:1 sin 30ðt 0:03Þ ðmÞ (48)
: ;
vs5 ¼ 0:1 sin 30ðt 0:06Þ
If mi = m and ki = k, then

Case M-4 variability in terms of temporal delay
1 0 €2
3 1
U U2 and frequency greater than of the system:
m þk €
0 1 1 3
U4 U4 8 9
8 9 < vs1 ¼ 0:1 sin 50t =

v€
m 3 4 1 < s1 = vs3 ¼ 0:1 sin 50ðt 0:03Þ ðmÞ (49)
¼ v€s3 (44) : ;
8 1 4 3 : ; vs5 ¼ 0:1 sin 50ðt 0:06Þ
v€s5
It is worth noting that these cases model inco-
The natural modes are herence effect (Eqs. 46 and 47) and wave passage
effect (Eqs. 48 and 49). Results obtained in terms
1 1 of elastic forces at spring 1 for the four above
ff 1 g ¼ and ff2 g ¼ (45)
1 1 cases are presented in Figs. 9, 10, 11, and 12.
It is observed that SVGM caused by incoher-
Case Study ence effect leads to values of elastic forces greater
In this part, elastic forces at the spring k1 obtained than those obtained for uniform loading (Figs. 9
by considering uniform imposed displacements and 10). This could be readily observed while
will be compared to those obtained when impos- varying amplitude (Eq. 46, Fig. 9) or frequency
ing variable displacements vs1, vs3 et vs5. For the (Eq. 47, Fig. 10). In turn, SVGM caused by wave
sake of simplicity, it will be considered that the passage effect does not yield greater values of
uniform displacement is set equal to vs1 elastic forces than uniform loading for the case
For the purpose of study, let’s consider that where the frequency of excitation is close to the
natural frequency of the structure (Eq. 48, Fig. 11).
However, this conclusion is not valid for case M-4
m ¼ 1 t, k ¼ 600 kN=m )
where the excitation frequency is greater than the
o1 ¼ 34:64 rad=s et o2 ¼ 49 rad=s natural frequency of the system (Fig. 12).
For the case of spatially variable ground

motion, four cases will be considered: Summary
Case M-1 variability in terms of maximum This entry aims at describing the modeling
amplitude: of SVGM and its effects on structures. The
total force is divided into two components:
8 9 pseudo-static and dynamic forces. After deriving
< vs1 ¼ 0:10 sin 30t = a general formulation for the case of MDOF
v ¼ 0:15 sin 30t ðmÞ (46)
: s3 ; systems, two cases have been considered: SDOF
vs5 ¼ 0:30 sin 30t and MDOF systems.
It has been shown that SVGM could increase
Case M-2 variability in terms of phases: or decrease the elastic forces for both cases.
Ground Motion: Seismic ← Fvar max=16.79 kN Uniform
Analysis, Fig. 9 Variation 15 Variable
of elastic forces at spring
1 for case M-1 10 ← Funif max=8.99 kN
Elastic force (kN)

5
-5
-10
-15
-20
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Time (sec)
Analysis, 60 Variable
Fig. 10 Variation of
elastic forces at spring 1 for 40
Elastic force (kN)
case M-2
20
← Funif max=8.99 kN
0
−20
−40
−60
−80
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Time (sec)
← Funif max=8.99 kN
8 ← Fvar max=7.98 kN
Uniform
Variable
S
Analysis,
Fig. 11 Variation of 6
elastic forces at spring 1 for
4
Elastic force (kN)
case M-3
2
0
−2
−4
−6
−8
−10
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Time (sec)
3368 Spectral Finite Element Approach for Structural Dynamics
Variable
Analysis, 15
Fig. 12 Variation of
elastic forces at spring 1 for 10
Elastic force (kN)

case M-4 ← Funif max=5.77 kN
5
−5
−10
−15
−20
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Time (sec)
Cross-References Shinozuka M (1971) Simulation of multivariate and multi

dimensional random processes. J Acoust Soc Am
49(1):357–367
▶ Design of Cable-Supported Bridges: Control Zendagui D, Berrah M, Kausel E (1999) Stochastic
Strategies de-amplification of spatially varying ground motions.
▶ Earthquake Location Soil Dyn Earthq Eng 18(6):395–471
▶ Earthquake Magnitude Estimation Zerva A (2009) Spatial variation of seismic ground
motions. Modeling and engineering applications.
▶ Engineering Characterization of Earthquake CRC Press, Boca Raton
Ground Motions Zerva A, Zervas V (2002) Spatial variation of seismic
▶ Physics-Based Ground-Motion Simulation ground motions: an overview. Appl Mech Rev
▶ Random Process as Earthquake Motions 55(3):271–296
▶ Response-Spectrum-Compatible Ground
Motion Processes
History Analysis Spectral Finite Element Approach for
▶ Stochastic Ground Motion Simulation Structural Dynamics
Priscilla Brandão Silva, Danilo Beli and

References José Roberto de França Arruda
Department of Computational Mechanics,
Abrahamson N, Schneider J, Stepp J (1991) Empirical
spatial coherency functions for applications to soil-
Faculty of Mechanical Engineering, UNICAMP,
structure interaction analyses. Earthquake Spectra Campinas, SP, Brazil
7(1):1–27
Benmansour N, Djafour M, Bekkouche A, Zendagui D
(2012) Seismic response evaluation of bridges under
differential ground motion: a comparison with the new
Synonyms
Algerian provisions. Eur J Environ Civil Eng 16(7):1–19
Berrah MK, Kausel E (1992) Response spectrum analysis Forced response; Spectral element; Structural
of structures subjected to spatially varying motions. dynamics; Wave propagation
Earthq Eng Struct Dyn 21(6):461–470
Deodatis G (1996) Non-stationary stochastic vector pro-
cesses: seismic ground motion applications. Probab
Eng Mech 11(3):149–167 Electronic supplementary material: The online version
Der Kiureghian A (1996) Coherence model for spatially of this chapter (doi:10.1007/978-3-642-35344-4_284)
varying ground motions. Earthq Eng Struct Dyn contains supplementary material, which is available to
25(1):99–111 authorized users.
Spectral Finite Element Approach for Structural Dynamics 3369
Introduction reduction, making it more adequate for uncer-

tainty and optimization analyses. Furthermore, it
Earthquake disaster in the past has motivated can easily be associated with other numerical
studies on earthquake-resistant buildings to save methods such as FEM and BEM. The limitations
lives and resources (Newmark and Hall 1982). of SEM are general nonlinear analysis, or when
With this objective, physical models and numer- exact wave solutions are not available, as it
ical methods are used to predict the dynamic occurs in geometrically complex structures.
behavior of structures. The numerical methods When using SEM it is important to be careful
provide quantitative analyses of physical phe- with the time–frequency transformations to
nomena. They can be used to design structures avoid leakage and aliasing errors (Lee 2009).
with geometry and materials for an adequate This chapter presents spectral element for
dynamic behavior under seismic excitation. The straight and curved frame elements. First, the
most frequently used in structural dynamics are equations of motion are developed and physical
the finite element method (FEM) and the bound- and kinematic assumptions are presented. Then
ary element method (BEM). the spectrum relation is established and the spec-
Based on wave propagation, the spectral finite tral element is derived. Two simple examples – a
element or spectral element method (SEM) was two- and a three-dimensional frame structures,
introduced by Beskos in 1978, organized and excited by a real earthquake signal – are ana-
seemed by Doyle (1997) in the 1990s. It allows lyzed. The purpose is to present the methodology
calculating relatively complex structures with dif- and concepts of dynamic structural analysis via
ferent boundary conditions and discontinuities the spectral finite element method.
using simple theories. It combines important char-
acteristics of FEM, dynamic stiffness method Classical Straight Frame
(DSM), and spectral analysis (Lee 2009).
The dynamic stiffness matrices and shape Equations of Motion
functions used in SEM are exact within the Consider a straight, long, and slender beam, as
scope of the underlying physical theory, and shown in Fig. 1. A Cartesian reference frame is
the method allows a reduced number of degrees placed such that the x-axis is along neutral axis of
of freedom. The matrices are depended on fre- the beam. Thus, for a rectangular cross section, if
quency, but using spectral analysis, the dynamic h is the beam thickness, the position y will vary in
response can be easily composed by wave super- the range h=2 y h=2. Along the z-axis is
position. Harmonic, random, or damped tran- the beam width.
sient excitations can be decomposed using Here, it is considered a plane frame element
the discrete Fourier transform (DFT). The dis- under flexural moments and axial forces.
crete frequencies are used to calculate the spec- In straight beams, flexural and longitudinal behav- S
tral matrix and discrete responses. Then, the iors are decoupled, and they may be modeled
complete dynamic response is computed by the separately, the former via the Euler–Bernoulli
sum of frequency components (inverse DFT). beam theory and the latter via the elementary
As FEM, SEM uses the assembly of a global rod theory.
matrix using elementary matrices and spatial The Euler–Bernoulli theory assumes the verti-
discretization. However, differently from FEM, cal deflection is constant across any cross section
only discontinuities and locations where loads and that cross sections remain plane after defor-
are applied need to be meshed (Ahmida and mation, which means that the shear deformation
Arruda 2001). is neglected. On the other hand, the classical rod
It can be a useful tool because it joins DSM theory considers only axial displacements and
with spectral analysis to produce accurate solu- neglects lateral contractions due to Poisson’s
tions in both frequency and time domains at a low effect. Under these assumptions, the displace-
computational cost due to a dramatic system ment field for a straight beam is given by:
Spectral Finite Element

Approach for Structural
Dynamics,
Fig. 1 Straight frame
element model
uðx, yÞ ¼ uðxÞ yyz ðxÞ as Hamilton’s principle, may be used to deter-

vðx, yÞ ¼ vðxÞ mine the equations of motion for this classical
beam model, which, in the absence of damping, is
where ū is the displacement in the x-direction given by:
(axial displacement), v is the displacement in
the y-direction (vertical displacement), and yz is @2u @2v
EA 2 rA 2 ¼ qx ðx, tÞ
the rotation with respect to the z-axis. @x 2 @t 2
Then, from differentiation of the displacement @2 @ v @ v
EI 2 þ rA 2 ¼ qy ðx, tÞ
field, axial and transverse shear strains at a given @x 2 @x @t
cross section of the beam are expressed:
where E is the Young’s modulus, I is the area
@u @u @yz moment of inertia with respect to the z-axis, r is
exx ¼ ¼ y
@x @x @x the material density, A is the cross-section area,
1 @u @v 1 @v and qx(x,t) and qy(x,t) are, respectively, the lon-
exy ¼ þ ¼ yz þ :
2 @y @x 2 @x gitudinal and vertical distributed loads along the
beam length. The boundary conditions associated
The Euler–Bernoulli beam theory neglects the to this beam problem are:
shear deformation, although not the shear force.
This implies that the transverse shear strain must @u
vanish, which yields: u¼0 or Fx ¼ EA
@x
@ @2v
@v v¼0 or Vy ¼ EI 2
exy ¼ 0 ! yz ¼ : @x @x :
@x @2v
yz ¼ 0 or Mz ¼ EI 2
The expression for the cross-section rotation @x
can be used to write the axial strain, which is the
The term EI is usually called the bending
only nonzero strain in the Euler–Bernoulli beam
stiffness.
model, in terms of the vertical and the longitudi-
If some internal viscoelastic damping is pre-
nal displacements, as follows:
sent, the equations of motion for the straight
@u @2v frame are modified to:
exx ¼ y 2:
@x @x
@2u @u @2u
EA 2 A rA 2 ¼ qx ðx, tÞ
Moreover, neglecting all stresses except for @x 2 @t @t
the predominant axial stress, one can write: @2 @ v @v @2v
EI 2 þ A þ rA 2 ¼ qy ðx, tÞ:
@x 2 @x @t @t
@u @2v
sxx ¼ Eexx ¼ E yE 2 : Spectral Analysis
@x @x
Assuming a harmonic solution for the displace-
Once the stress and strains are determined, ments of the beam (or applying the Fourier trans-
Newton’s laws or variational formulations, such form), i.e., by writing:
Spectral Finite Element Approach for Structural Dynamics, Fig. 2 Spectrum relations of a damped frame
element: (a) real part of wavenumbers and (b) imaginary part of wavenumbers
X
uðx, tÞ ¼ u^n ðx, on Þeion t , k1 2 þ kx 2 ¼ 0 ! k1 ¼ kx
X
n k2 2 þ b2 ¼ 0 ! k2 ¼ b
vðx, tÞ ¼ v^n ðx, on Þeion t k3 2 b2 ¼ 0 ! k3 ¼ ib
n
where k1, k2, k3 are the wavenumbers. They are
in the homogenous case and considering constant function of the angular frequency o, and this
properties along the beam, the equations of relationship constitutes the spectrum or disper-
motion, in the frequency domain, are: sion relation, which is plotted in Fig. 2. Notice
that when some damping is present, longitudinal
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi and flexural modes are complex waves; thus, they
d2 u^ o2 rA ioA exhibit a decaying behavior while propagating. In
þ k x
2
^
u ¼ 0, k x ¼ the undamped case, there are two real
dx2 rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
EA
d4 v^ o 2
rA ioA wavenumbers (the longitudinal and first flexural
4
b4 v^ ¼ 0, b2 ¼ : waves), which are propagating waves, and an
dx EI
imaginary wavenumber (the second flexural
wave), which is a purely spatial decaying wave,
The second homogeneous equation, relative to the so-called evanescent wave. Phase and group
the transverse displacement, can be split into a speeds, c0 and cG, respectively, can be also
product of two terms which must vanish. Thus, it S
derived (see Fig. 3), and they are expressed as:
yields:
o do
c0 ¼ , cG ¼ :
k dk
d2 v^
þ b2 v^ ¼ 0
dx2 In the undamped case, unlike in the case of
d2 v^ longitudinal motion, the phase and group speeds
b2 v^ ¼ 0:
dx2 related to flexural waves depend on the cross-
section properties, given by the area and the sec-
Then, considering a second Fourier transforma- ond moment of inertia. The group speed is twice
tion, now in the spatial domain, u^ðx, oÞ ¼ the phase speed and both depend on frequency
X X
u~ðkm , oÞeikm x , v^ðx, oÞ ¼ v~ðkm , oÞeikm x ; and, hence, are dispersive. It should be noted that,
m m with the Euler–Bernoulli assumptions, the wave
gives: speeds tend to infinity as the frequency increases,
Spectral Finite Element Approach for Structural Dynamics, Fig. 3 Wave speeds of a frame element, real part in
solid line, and imaginary part in dotted line: (a) group speeds and (b) speed ratios between group and phase speeds
which is not reasonable. This theory is, therefore, Thus, it may be also written as a superposition
only suitable for low frequencies. of waves, as follows:
The displacement solution may be expressed X
in the spectral form as: yz ðx, tÞ ¼ iAþ
2e
ibn x
Aþ
3e
bn x
n
io t
X þ iA
2e
ibn x
þ A
3e
bn x
bn e n :
uðx,tÞ ¼ Aþ
1e
ikxn x
þ A
1e
ikxn x
e ion t
X
n
From the classical theories for beams and rods,
vðx,tÞ ¼ Aþ
2e
ibn x
þ Aþ
3e
bn x
þ A
2e
ibn x
the internal loads related to the frame may be
n
þ A
3e e ;
bn x ion t derived from the displacement field and are
given by:
i.e., as a superposition of waves propagating in @u

opposite directions. Fx ¼ EA
@x
It is worth recalling here that the cross-section @ @2v
Vy ¼ EI 2 :
rotation is related to the transverse displacement @x @x
by the following expression: @2v
Mz ¼ EI 2
@x
@v
yz ¼ :
@x Thus, they may be expressed in spectral form as:
X
Fx ðx, tÞ ¼ EA kxn iAþ1e
ikxn x
þ iA
1e
ikxn x ion t
e
Xn
Vy ðx,tÞ ¼ EI b3n iAþ
2e
ibn x
þ Aþ3e
bn x
þ iA 2e
ibn x
A
3e
bn x ion t
e
X
n
Mz ðx, tÞ ¼ EI b2n Aþ2e
ibn x
þ Aþ
3e
bn x
A 2e
ibn x
þ A3e e :
bn x ion t
n
Displacements and loads evaluated at an arbi-

trary position along the frame may be expressed
in matrix form, as follows:
2 3
Aþ1
3 2
2 3 6 þ7
u^ 1 0 0 1 0 0
þ 6 A2þ 7
L 0 66 A3 7
7
^ ¼ 4 v^ 5 ¼ 4 0
q 1 1 0 1 15 x 6 7
0 Lx 6 B1 7
y^z 0 ib b 0 ib b 4 B 5
2
B3
23
Aþ1
2 3 2 3 6 þ7
F^x iAkx 0 0 iAkx 0 0
þ 6 A2þ 7
^f ¼ 4 V^y 5 ¼ E4 0 3 5 Lx 0 66 A3 7
7
iIb3 Ib3 0 iIb3 Ib 6 7
^z 0 Lx 6 B1 7
M 0 Ib2 Ib2 0 Ib2 Ib2 4 B 5
2
B3
where going waves consists in analyzing the sense of

decreasing wave amplitudes or, if the amplitude
2 3
eikx x 0 0 remains constant, the sense of the time-averaged
þ 6 7 power transmission. The latter can be
Lx ¼ 4 0 eibx 0 5,
evaluated by regarding the sign of ℜ{iofTLqL}:
0 0 ebx
2 ikx ðxLÞ
3 if it is negative, the corresponding wave is a
e 0 0 positive-going wave; otherwise, it is a negative-
6 7
x
L ¼4 0 eibðxLÞ 0 5: going wave. Alternatively, the sign of the
0 0 ebðxLÞ group speed may be used, which may be
shown to follow the time-averaged power
Here, the reference for wave amplitudes transmission sign.
related to negative-going waves, i.e., those prop-
agating to the left or in the negative sense of the
x-axis, is placed at the end of an element of frame Spectral Element for a Classical Frame
( x ¼ L ). Thus, instead of A
1 , A2 , A3 , one has Consider an element of classical frame

B1 , B2 , B3 . This is done for the purpose of (Euler–Bernoulli beam combined with an
numerical conditioning in the case where there elementary rod) with length L. The displacements S
are highly evanescent waves. at both ends of the spectral element can be
From Mace et al. (2005) and Mead (1973), the expressed as a function of the wave amplitudes as
classification of waves into positive- or negative- follows:
2 3 2 32 3
u^ð0Þ 1 0 0 eikx L 0 0 Aþ
1
6 v^ð0Þ 7 6 0 1 1 0 eibL ebL 7 6 Aþ 7
6 7 6 76 2þ 7
6 y^z ð0Þ 7 6 0 ib b 0 ibeibL bebL 76 A 7
6 7 6 76 3 7
6 u^ðLÞ 7 ¼ 6 eikx L 0 0 1 0 0 76 B 7
6 7 6 76 1 7
4 v^ðLÞ 5 4 0 eibL ebL 0 1 1 54 B 5
2
y^z ðLÞ 0 ibeibL bebL 0 ib b B
3

" #

þ
^0
q Fþ F
q LL Aþ
Analogously, the external loads applied to the
¼ q
þ : frame may be written as:
^L
q Fþ
q LL

Fq B
2 3 2 32 þ 3
F^x ð 0Þ iAkx 0 0 iAkx eikx L 0 0 A1
6 V^y ð0Þ 7 60 iIb3 Ib3 0 iIb3 eibL Ib3 ebL 7 6 Aþ 7
6 7 6 76 2þ 7
6 M^ z ð 0Þ 7 6 Ib2 Ib2 Ib2 eibL Ib2 ebL 7 6 7
6 7 ¼ E6 0 0 76 A3 7
6 F^x ðLÞ 7 6 iAkx eikx L iAkx 0 0 7 6 7
6 7 6 0 0 7 6 B1 7
4 V^y ðLÞ 5 40 iIb3 eibL Ib3 ebL 0 iIb3 Ib3 54 B 5
2
^
Mz ðLÞ 0 Ib2 eibL Ib2 ebL 0 Ib2 Ib2 B
3

" #

þ
^0
F Fþ F
f LL Aþ
^ ¼ f
þ :
FL Fþ
f L L Ff B
2 3
With the general method of computing Aþ and Eð1 þ iÞ 4
iAkx 0 0

B from the displacement equation and replacing it K1 ¼ 0 2iIb3 ð1 þ iÞIb2 5:
2
in the expression for external loads, one can write: 0 ð1 þ iÞIb 2
2Ib
" #

^0 ^0
q Curved Beam
F
¼ KðoÞ ,
^L
F ^L
q
Equations of Motion
" þ
#" #
þ 1
Fþ f F f LL Fþq F
q LL
Consider a slender in-plane curved beam as shown
K¼ þ þ
: in Fig. 4. The circumferential reference frame is
Fþ
f LL Ff Fþ
q LL Fq
placed such that the s-axis is along the neutral axis
of the beam, here supposed to be equidistant of the
Throw-Off Spectral Element for a Straight Frame
upper and lower surfaces of the beam. The beam
A semi-infinite element of beam can be easily
thickness h varies in the range h=2 r þh=2
modeled via SEM by assuming the wave ampli-
along the radial r-axis and its width lies along the
tudes related to waves propagating to the left
z-axis. Based on the Euler–Bernoulli beam theory,
(negative-going waves) are zero, i.e., considering
which neglects shear deformation, the kinematics
there is no reflection. In this case, the displace-
of a curved beam can be written as:
ment and force vectors at the left side of the
element are given by:
@us ðsÞ vr ðsÞ
us ðr, sÞ ¼ us ðsÞ þ r
^ 0 ¼ Fþ
q qA
þ
@s R
vr ðr, sÞ ¼ vr ðsÞ
^ 0 ¼ Fþ Aþ :
F f
where ūs is the displacement in the s-direction

Thus, the spectral matrix for the throw-off
(tangential displacement), vr is the displacement
element of frame is:
in the y-direction (radial displacement), c is the
1 rotation with respect to the z-axis, and R is the
^ 0 ¼ K1 ðoÞ^ K1 ¼ Fþ þ
F q0 , f Fq radius of the centerline.

Dynamics, Fig. 4 Curved
beam element model

Then, from differentiation of the displacement @us vr
us ¼ 0 or T s ¼ EA
@s 2 R
field, considering a significant strain in

s-direction, a strain at a given cross section of 1 @ us @ 3 v r
the curved beam is expressed: vr ¼ 0 or V r ¼ EI þ 3
R @s @s
2
:
1 @us @ 2 vr
2 c¼0 or Mzs ¼ EI þ 2
@us vr @ vr 1 @us R @s @s
ess ¼ þr þ :
@s R @s2 R @s
If some internal viscoelastic damping is pre-
sent, the equations of motion are modified to:
Moreover, neglecting all stresses except for

the predominant circumferential stress, one can EI 1 @ 2 us @ 3 v r EA @vr @ 2 us
þ 3 þ þR 2
write: R R @s2 @s R @s @s

2 @us @ 2 us
@us vr @ vr 1 @us A rA 2 ¼ qs ðs, tÞ
sss ¼ Eess ¼ E þ rE þ : @t @t
@s R @s2 R @s 3
EI @ us @ 4 vr EA vr @us
þR 4 þ
R @s3 @s R R @s
Then, as a result of the application of New-
@vr @ vr
2
ton’s laws or variational formulations, the equa- þ A þ rA 2 ¼ qr ðs, tÞ:
tions of motion for the undamped curved beam @t @t
are derived:
Spectral Analysis
As before, in the straight beam case, two Fourier
EI 1 @ us @ vr
2 3
EA @vr @ us 2
3 þ þR 2 transforms can be successively applied to the
R R @s 2 @s R @s @s
displacement solution of a curved beam, which
@ 2 us allows writing: S
rA 2 ¼ qs ðs, tÞ
@t

EI @ 3 us @ 4 vr EA vr @us X XX
R þ us ðs,tÞ ¼ u^sn ðs,on Þeion t ¼ u~s ðkm ,oÞeikm s eion t
R @s3 @s4 R R @s n n m
X XX
@ vr
2
vr ðs, tÞ ¼ v^rn ðs, on Þeion t ¼ v~r ðkm , oÞeikm s eion t :
þ rA 2 ¼ qr ðs, tÞ
@t n n m
where qs(s,t) and qr(s,t) are, respectively, tangen- In the homogenous case and considering con-
tial and radial distributed loads along the curved stant properties along the beam, the displacement
beam length. The boundary conditions associated solution is substituted in the equations of motion,
to this problem are: which yields, in the frequency domain,
Spectral Finite Element Approach for Structural Dynamics, Fig. 5 Spectrum relations of a damped curved beam
element: (a) real part of wavenumbers and (b) imaginary part of wavenumbers
2 3
EIk2 iEIk3 iEAk

6 EAk2 þ iAo þ o2 rA þ 7
6 R2 R R 7½ u~s v~r ¼ 0 :
4 iEIk3 iEAk EA 5 0
þ EIk4 þ 2 iAo o2 rA
R R R
þiAo2
þiAo 2
A nontrivial solution for this system of equa- where b40 ¼ rAo EI and k20 ¼ rAo EA .
tions is obtained by setting the determinant to This equation provides three pairs of
vanish, which results in a sixth-order polynomial wavenumbers ks1, ks2, ks3. One pair is
equation in k, the characteristic equation, or spec- linked to extensional waves and the other two
trum relation: pairs to flexural waves (Fig. 5).

Then, the displacement solution may be
2 4 2 1 1 2 expressed in the spectral form as:
k 6
þ 2 k b0 þ k0 2 4 k
k20 4
R R R

1
þ k20 2 b40 ¼ 0
R
X þ þ þ

us ðs, tÞ ¼ As1 eiks1n s þ As2 eiks2n s þ As3 eiks3n s þ As1 eiks1n s þ As2 eiks2n s þ As3 eiks3n s eion t
Xn
vr ðs, tÞ ¼ Aþ s1 e
iks1n s
þ Aþ
s2 e
iks2n s
þ Aþ s3 e
iks3n s
þ A s1 e
iks1n s
þ A s2 e
iks2n s
þ A s3 e
iks3n s ion t
e ,
n
@vr ðs, tÞ X iks1n Aþ s1 e
iks1n s
iks2n Aþ s2 e
iks2n s
iks3n Aþ s3 e
iks3n s
cðs, tÞ ¼ ¼ iks1n s iks2n s iks3n s eion t ;
@s n
þik s1n A s1 e þ ik s2n A s2 e þ ik s3n A s3 e
As m
asm ¼
As m
EIk4smn þ EAR2 iAo rAo2
¼ :
i EIk3smn þ EAksmn =R
Expressions for the internal loads in curved

beams have been derived in terms of the displace-
ment field and are rewritten below:

@us vr
T s ¼ EA
@s R
1 @ 2 us @ 3 v r
Spectral Finite Element Approach for Structural V r ¼ EI þ 3
R @s @s
2
Dynamics, Fig. 6 Amplitude ratios for a damped curved
beam element 1 @us @ 2 vr
Mzs ¼ EI þ 2 :
R @s @s
i.e., as a superposition of waves propagating in Thus, substituting the spectral solutions for the
opposite directions. From the homogeneous sys- displacement field, it yields:
tem of equations for ũs, ṽr, amplitude ratios for
each wave mode are obtained (Fig. 6):
X ik þ iks1n s
iks2n as2 Aþ iks2n s
iks3n as3 Aþ iks3n s

s1n as1 As1 e s2 e s3 e
T s ðs, tÞ ¼ EA iks1n s iks2n s iks3n s eion t
þiks1n as1 As1 e þ iks2n as2 As2 e þ iks3n as3 As3 e
n
EA X Aþ s1 e
iks1n s
þ Aþ s2 e
iks2n s
þ Aþ s3 e
iks3n s
iks1n s iks2n s iks3n s eion t
R n þAs1 e þ As2 e þ As3 e

EI X k2s1n as1 Aþ e iks1n s
þ k 2
a A þ iks2n s
e þ k 2 þ iks3n s
s3n as3 As3 e
V r ðs, tÞ ¼ s1
iks1n s
s2n s2 s2
iks2n s iks3n s eion t
R n ks1n as1 As1 e
2
ks2n as2 As2 e
2
ks3n as3 As3 e
2
X ik3 Aþ eiks1n s þ ik3 Aþ eiks2n s þ ik3 Aþ eiks3n s :

EI s1n s1 s2n s2 s3n s3 eion t
ik3s1n A s1 e
iks1n s
ik3s2n A s2 e
iks2n s
ik3s3n A s3 e
iks3n s
n
EI X iks1n as1 Aþ s1 e
iks1n s
iks2n as2 Aþ s2 e
iks2n s
iks3n as3 Aþ s3 e
iks3n s
Mzs ðs, tÞ ¼ iks1n s iks2n s iks3n s eion t
R n þiks1n as1 As1 e þ iks2n as2 As2 e þ iks3n as3 As3 e
X k2 Aþ eiks1n s þ k2 Aþ eiks2n s þ k2 Aþ eiks3n s
EI s1n s1 s2n s2 s3n s3 eion t
n
k2s1n A s1 e
iks1n s
k2s2n A s2 e
iks2n s
k2s3n A s3 e
iks3n s
S
Displacements and loads evaluated at an arbi-

trary position along the curved beam may be
expressed in matrix form, as follows:
2 3
Aþ
s1
2 3 2 3 6 þ7
u^s as1 as2 as3 as1 as2 as3
þ 6 As2 7
L 0 6 Aþ 7
^CB
q ¼ 4 v^r 5 ¼ 4 1 1 1 1 1 1 5 s 6 s3 7
6 7
^ 0 Ls 6 Bs1 7
c iks1 iks2 iks3 iks1 iks2 iks3 4 B 5
s2
B
s3
0 2 3 1 3
2
iks1 iks2 iks3 iks1 iks2 iks3 Aþ
s1
2 3 B C1 4 k2s1 k2s2 k2s3 k2s1 k2s2 k2s3 5Gs C 6 þ7
T^s B C
þ 6 As2 7
B ik s1 ik s2 ik s3 iks3 3 C 0 6 Aþ 7
^
fCB 4 ^
¼ Vr ¼ B5 B 2 iks1 iks2 C Ls 6 s3 7
1 1 1 1 1 1 C 0 L 6 B 7
^ zs
M B C s 6 s1 7
@ þC2 4 ik3 ik3 ik3 ik3 ik3 ik3 5 A 4 B 5
s1 s2 s3 s1 s2 s3 s2
k2s1 k2s2 k2s3 k2s1 k2s2 k2s3 B
s3
where
2 3 2 3
eiks1 s 0 0 eiks1 ðss0 Þ 0 0
Lþ ¼ 4 0 eiks2 s 0 5, L ¼ 4 0 eiks2 ðss0 Þ 0 5;
s s
iks3 s iks3 ðss0 Þ
0 0 e 0 0 e
Gs ¼ diag½ as1 as2 as3 as1 as2 as3 ;
2 3 2 3
EA 0 0 EA=R 0 0
C1 ¼ 4 0 EI=R 0 5, C 2 ¼ 4 0 EI 0 5:
0 0 EI=R 0 0 EI
It should be noticed that the reference for the Spectral Element

wave amplitudes related to negative-going waves Consider an element of curved beam with cir-
is placed at the end of the beam element, i.e., at cumferential length s0. The displacements at
s ¼ s0 : both ends of the spectral element can be
expressed as a function of the wave amplitudes
as follows:
2 3 2 32 þ 3
u^s ð0Þ as1 as2 as3 as1 eiks1 s0 as2 eiks2 s0 as3 eiks3 s0 As1
6 v^r ð0Þ 7 6 eiks1 s0 eiks2 s0 iks3 s0 7 6 þ7
6 7 6 1 1 1 e 76 As2 7
6 ^ ð0Þ 7 6 7 6 7
6 c 7 6 iks1 iks2 iks3 iks1 eiks1 s0 iks2 eiks2 s0 iks3 eiks3 s0 7 6 Aþ 7
6 7 6 76 s3 7
6 u^s ðs0 Þ 7 ¼ 6 a eiks1 s0 as2 eiks2 s0 as3 eiks3 s0 76 B 7
6 7 6 s1 as1 as2 as3 76 s1 7
6 7 6 76 7
4 v^r ðs0 Þ 5 4 eiks1 s0 eiks2 s0 eiks3 s0 1 1 1 54 Bs2 5
^ ðs0 Þ
c iks1 eiks1 s0 iks2 eiks2 s0 iks3 eiks3 s0 iks1 iks2 iks3 B
s3

" #

þ
^0
q Fþ F
qs Ls0 As þ
¼ qs
þ :
^s0
q Fþ
qs Ls0 Fqs
Bs
Analogously, the external loads applied may

be written as:
2 2 3 3
iks1 iks2 iks3 iks1 eiks1 s0 iks2 eiks2 s0 iks3 eiks3 s0
6 6 k2s1 k2s2 k2s3 k2s1 eiks1 s0 k2s2 eiks2 s0 k2s3 eiks3 s0 7 7
6 6 7 7
6 6 7 7
6 6 iks1 iks2 iks3 iks1 eiks1 s0 iks2 eiks2 s0 iks3 e iks3 s0 7 7
2 3 6 C1 6 7 Gs 7 2 3
T^s ð0Þ 6 6 iks1 eiks1 s0 iks2 eiks2 s0 iks3 eiks3 s0 7 7 Aþ
6 6 iks1 iks2 iks3 7 7 s1
6 ^ 7 6 6 7 76 þ 7
6 V r ð0Þ 7 6 4 k2s1 eiks1 s0 k2s2 eiks2 s0 k2s3 eiks3 s0 k2s1 k2s2 k2s3 5 76 As2 7
6 7 6 76 7
6 M^ 7 6 iks1 s0 iks2 s0
iks3 eiks3 s0 iks1 iks2 iks3 76 Aþ 7
6 zs ð0Þ 7 6 iks1 e iks2 e 76 s3 7
6 7¼6 2 3 76 7
6 T^s ðs0 Þ 7 6 1 1 1 eiks1 s0 eiks2 s0 e iks3 s0 76 Bs1 7
6 7 6 76 7
6 ^ 7 6 6 3 iks3 s0 7 7
iks1 iks2 ik3s3 ik3s1 eiks1 s0 ik3s2 eiks2 s0 7 74 Bs2 5
3 3
4 V r ðs0 Þ 5 6 6 iks3 e
6 6 7 7
^
Mzs ðs0 Þ 6 6 2
ks1 2
ks2 k2s3 k2s1 eiks1 s0 k2s2 eiks2 s0 ks3 e2 iks3 s0 7 7
6 þC2 6 7 7 Bs3
6 6 eiks1 s0 iks2 s0 iks3 s0 7 7
6 6 e e 1 1 1 7 7
6 6 3 iks1 s0 7 7
4 4 iks1 e ik3s2 eiks2 s0 ik3s3 eiks3 s0 ik3s1 ik3s2 iks33 5 5
k2s1 eiks1 s0 k2s2 eiks2 s0 k2s3 eiks3 s0 k2s1 k2s2 k2s3

" #

þ
^0
F Fþ F
fs Ls0 As þ
^ ¼ fs
þ :
Fs0 Fþ L
fs s0 Ffs Bs
With the general method for computing As þ where c(y, z) is a function which represents the
and Bs from the displacement equation and warping of the cross section. In this chapter, only
replacing it in the expression for external loads, rods of circular cross section are considered.
one can write: In this case, there is no warping ðc ¼ 0Þ and the
" #
nonzero strain components are:
^0
F ^0
q
¼ KCB ðoÞ ,
^ s0 ^ s0
q
F 1 @u @v z @yx @yx
" þ
#" #
þ 1
exy ¼ þ ! exy ¼ , g ¼ z
Fþ fs Ffs Ls0 Fþ qs F
qs Ls0 2 @y @x 2 @x xy @x
KCB ðoÞ ¼ þ þ
: 1 @u @w y @yx @yx
Fþfs Ls0 Ffs Fþ
qs Ls0 F qs exz ¼ þ ! exz ¼ , g ¼y :
2 @z @y 2 @x xz @x
Rod Under Torsion (Shaft)
The associated stresses are given by:
Consider a rod element subjected to a distributed
torsion moment (torque) mx as shown in Fig. 7. @yx
Due to the applied load, at a position x, the cross sxy ¼ Ggxy ! sxy ¼ zG S
@x
section rotates about the x-axis by an angle yx(x). @yx
sxz ¼ Ggxz ! sxz ¼ yG
Here, it is assumed that the cross section remains @x
plane while twisting. For wavelengths which are
greater than ten times the cross-section dimen- where G is the shear modulus. In cylindrical
sions, the Saint-Venant theory is also valid (Petyt coordinates, the resultant stress is:
2010). In this case, the kinematics of a rod under
@yx
torsion may be approximated as: sxr ¼ Ggxr ! sxr ¼ rG :
@x
@yx ðxÞ
uðx, y, zÞ ¼ cðy, zÞ By applying Newton’s laws or variational for-
@x
vðx, y, zÞ ¼ zyx ðxÞ mulations, the equation of motion for this ele-
wðx, y, zÞ ¼ yyx ðxÞ ment is derived:

Dynamics, Fig. 7 Rod
element model under
torsion (shaft)

X
@ @yx @yx @ 2 yx yx ðx,tÞ ¼ Aþxx1 e
ikxxn x
þ A
xx1 e
ikxxn x ion t
e
GJ J rJ 2 ¼ mx X
@x @x @t @t n þ ik x
Mx ðx,tÞ ¼ ikxxn GJ Axx1 e xxn þ A xx1 e e :
ikxxn x ion t
n
where yx is the rotation with respect to the x-axis,
J is the torsion constant or the polar moment of Setting the reference for negative-going
area with respect to the x-axis for concentric waves at x ¼ L, the rotation and the internal load
circular rods (round shafts), is the damping are written in matrix form:
(viscous) coefficient per unit volume, and r is

ik x
þ
the mass density. e xx 0 Axx1
q^xx ¼ y^x ¼ ½ 1 1
The associated boundary conditions in this 0 e
ikxx ðxLÞ B xx1

case are the rotation yx and the internal torque ^ ^ eikxx x 0 Aþ
f xx ¼ Mx ¼ ikxx GJ ½1 1 xx1 :
Mx ¼ GJ @y 0 eikxx ðxLÞ B
@x .
x
xx1
Spectral Element
Spectral Analysis
The spectral element of a shaft element in torsion
As it was done in previous sections, by applying
of length L is straightforwardly obtained by writ-
two Fourier transforms (time and space), the dis-
ing the rotations with respect to the x-axis at both
placement field is written as spectral solutions, in
ends as a function of the wave amplitudes, as
the form:
follows:
X

yx ðs, tÞ ¼ y^xn ðx, on Þeion t y^x ð0Þ 1 eikxx L Aþ
¼ ikxx L xx1 ;
y^x ðLÞ B
n
XX e 1 xx1
¼ e
y x ðkxxm , oÞeikm s eion t :
n m
i.e.,

þ

Then, assuming geometric and material prop- q^xx0 Fþ F
qxx LL Axx þ
erties constant along the rod length and substitut- ¼ qxx
þ :
q^xxL Fþ
qxx LL

Fqxx Bxx
ing this expression in the homogeneous
differential equation of motion, the spectrum
The applied torque may be also evaluated at
relation is obtained:
both ends:

GJk2xx ioJ þ o2 rJ yx ¼ 0, M^ x ð 0Þ 1 eikxx L Aþ
^ x ðLÞ ¼ ikxx GJ eikxx L
xx1
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi M 1 B
o2 rJ io xx1
kxx ¼ :
GJ
or
Thus, the rotation yx and the internal moment

" #

þ
F^xx0 Fþ F
f xx LL Axx þ
Mx evaluated at an arbitrary point along the rod ¼ f xx
:
F^xxL Fþ
f xx LL
þ
Ff xx Bxx
may be expressed as:

Dynamics, Fig. 8 3D
frame member
Thus, by writing the wave amplitudes as func- It is important to emphasize at this point that the
tion of the rotation vector and substituting it into expression for the rotation w.r.t. the y-axis requires
the expression for external applied moments, the a negative sign as a result of the assumption that
spectral element matrix is obtained: translations are positive in the sense of the Carte-
sian coordinates and also the respect of the right-
" #
hand rule for the coordinate system. By superpos-
^xx0
F q^xx0
¼ Kxx ðoÞ , ing the spectral solutions for a beam element under
^xxL
F q^xxL
" #" #
axial, flexural, and torsion deformations, the dis-
þ þ 1
Fþ
f xx Ff xx LL Fþqxx F
qxx LL placement solution vector is given by:
Kxx ðoÞ ¼ þ þ
;
Fþf xx LL F Fþ
qxx LL F
f xx qxx
2 3
u^ðxÞ

6 v^ðxÞ 7
6 7

þ
cos ðkxx LÞ 1 6w ^ ðxÞ 7 þ Lþ 0 A
Kxx ðoÞ ¼ GJkxx sin ðkxx LÞ : ^3D ðxÞ ¼ 6 7 ¼ F F x
cos ðkxx LÞ 1 q 6 y^x ðxÞ 7 q q
0 Lx B
,
6 7
4 y^y ðxÞ 5
3D Frame Structure y^z ðxÞ
A 3D frame structure is a combination of
one-dimensional elements in the 3D space that where:
are capable to support axial, bending, and torsion
2 3
efforts (Fig. 8). Thus, using the expressions 1 0 0 0 0 0
derived in the previous sections, a spectral ele- 60 1 1 0 0 07
S
6 7
ment for a 3D frame element can be easily built. It 6 7
60 0 0 1 1 07
is composed of six degrees of freedom per node Fþ ¼ 6 7,
q 60 17
(three translations and three rotations), which 6 0 0 0 0 7
6 7
allow an axial deformation and torsion along the 40 0 0 ibz bz 05
main axis and bending in the two transverse 0 iby by 0 0 0
planes. Its displacement field may be expressed 2 3
1 0 0 0 0 0
as: 6 0
6 1 1 0 0 077
6 7
6 0 0 0 1 1 07
uðx, y, z, tÞ ¼ uðx, tÞ yx ðx, y, z, tÞ ¼ yx ðx, tÞ Fq ¼ 6

6 0
7;
@wðx, tÞ 6 0 0 0 0 177
vðx, y, z, tÞ ¼ vðx, tÞ yy ðx, y, z, tÞ ¼ yy ðx, tÞ ¼ 6 7
@x 4 0 0 0 ibz bz 05
@vðx, tÞ
wðx, y, z, tÞ ¼ wðx, tÞ yz ðx, y, z, tÞ ¼ yz ðx, tÞ ¼ 0 iby by 0 0 0
@x
2 3
eikx x 0 0 0 0 0
6 0 eiby x 7
6 0 0 0 0 7
6 7
6 0 0 eby x 0 0 0 7
Lx ¼ 6
þ
6 0 ibz x
7,
7
6 0 0 e 0 0 7
6 7
4 0 0 0 0 ebz x 0 5
ikxx x
0 0 0 0 0 e
2 ikx ðxLÞ 3
e 0 0 0 0 0
6 eiby ðxLÞ 7
6 0 0 0 0 0 7
6 7
6 0 0 e by ðxLÞ
0 0 0 7
L ¼ 6 7:
x 6 ibz ðxLÞ 7
6 0 0 0 e 0 0 7
6 7
4 0 0 0 0 e bz ðxLÞ
0 5
0 0 0 0 0 eikxx ðxLÞ
2 3
The corresponding internal loads acting on a F^x ðxÞ
3D frame member are: 6 V^y ðxÞ 7
6 7
6 7
^f 3D ðxÞ ¼ 6 V^z ðxÞ 7
6M ^ 7
@u @yx 6 x ðxÞ 7
Fx ¼ EA Mx ¼ GJ 4M ^ y ðxÞ 5
@x @x ^ z ðxÞ
@ @2v @2w M
Vy ¼ EI z 2 My ¼ EI y 2
þ

@x @x @x þ Lx 0 Aþ
¼ Ff Ff ;
@ @2w @2v 0 L B
Vz ¼ EI y 2 Mz ¼ EI z 2 x
@x @x @x
with:
which, in vector form, yields:
2 3
ikx EA 0 0 0 0 0
6 0 iEI z b3y EI z b3y 0 0 0 7
6 7
6 0 0 0 iEI y b3z EI Y b3z 0 7
Ff ¼ 6
þ
6 0
7;
6 0 0 0 0 ikxx GJ 7
7
4 0 0 0 EI y b2z EI y b2z 0 5
0 EI z b2y EI z b2y 0 0 0
2 3
ikx EA 0 0 0 0 0
6 0 iEI z b3y EI z b3y 0 0 0 7
6 7
6 0 0 0 iEI y b3z EI Y b3z 0 7
F ¼ 6 7:
f 6 0 0 0 0 0 ikxx GJ 7
6 7
4 0 0 0 EI y b2z EI y b2z 0 5
0 EI z b2y EI z b2y 0 0 0
A spectral element of frame with length L is required with the increase in frequency, and it is
straightforwardly obtained by writing displace- possible to build spectral models with very few
ment and external loads at both ends, as follows: degrees of freedom. On the other hand, as it
makes use of matrix formulations as FEM does.

" #

þ
^3D 0
q Fþ F
q LL Aþ The assemblage of spectral elements is quite sim-
¼ q
þ
^3D L
q Fþ
q LL Fq
B ple via the direct stiffness method (Doyle 1997).
It is also important to note that the global problem

is solved in the frequency domain. For every
^ 3D 0
F ^f 3D ðx ¼ 0Þ
^ ¼ ^ discrete frequency, an assemblage process iden-
F3D L f 3D ðx ¼ LÞ tical to that of the static case is performed and the
" #
þ
þ
Fþ F f LL A global response vector is obtained. The discrete
¼ f
þ :
Fþf L L F
f B Fourier transform is used to obtain the spectral
representation of the input data, and the inverse
DFT allows one to recover the response of the
Then, the spectral element matrix of a 3D
coupled system in the time domain.
frame member is readily obtained:
" #
Global Dynamic Stiffness Matrix Assembling
^ 3D 0
F ^3D 0
q
¼ K3D ðoÞ , Procedure
^ 3D L
F ^3D L
q Consider a frame element as shown in Fig. 9.
" þ
#" #
þ 1
Fþ f F f LL Fþq F
q LL
Two types of coordinate system are defined: a
K3D ðoÞ ¼ þ þ
: global one, defined by the orthogonal X, Y, and
Fþ
f LL F f Fþ
q LL Fq
Z directions, and a local one, defined by x, y, and
z axes with the x-axis parallel to the main axis
Spectral Element Modeling
(neutral axis) of the frame member.
In previous sections, it has been shown how to
The spectral element matrix, as formulated
formulate spectral element matrices for straight
previously for frame member, is written with
and curved frame members. However, in real
respect to the local reference frame, as follows:
engineering systems, those structures do not
appear isolated; they are usually connected to
^ xyz ðoÞ^
K ^ xyz ðoÞ:
qxyz ðoÞ ¼ F
other basic structures involving or not disconti-
nuities. In this section, it will be shown that the
spectral element method can be efficiently used Load and displacement vectors in the global
to compute the forced response of coupled sys- reference frame may be related to those defined in
tems in the frequency and in the time domains. the local coordinate system by application of
This is particularly useful in earthquake engineer- successive rotations from the local axes to the
ing, as it allows simulating the effect of seismic global axes: S
ground motion at a very low computational cost
compared to conventional computational ana- ^XYZ ¼ T^
q qxyz
lyses via FEM or BEM. This is possible as SEM ^ ^ xyz
FXYZ ¼ TF
allows building exact element matrices at each
frequency. Therefore, mesh refinement is not with:
2 3
Rx ðcÞRy ð’ÞRz ðyÞ 0 0
6 0 Rx ðcÞRy ð’ÞRz ðyÞ 0 7
T¼6
4
7;
5
⋮ ⋮ ⋱ ⋮
0 0 Rx ðcÞRy ð’ÞRz ðyÞ

Dynamics, Fig. 9 Frame
element relative to local
and global coordinate
systems
where Rx, Ry, and Rz are with:

rotation matrices, there-
fore unitary RT R ¼ I , defined as:
^ XYZ ðoÞ ¼ TK
K ^ xyz ðoÞTT :
2 3
1 0 0 The transformation of coordinate system is
6 7 performed for each structural member. This allows
Rx ¼ 4 0 cos c sin c 5,
one to enforce kinematic compatibility and equi-
0 sin c cos c
2 3 librium at all nodes connecting structural members.
cos ’ 0 sin ’ Then, the global dynamic system is constructed
6 7
Ry ¼ 4 0 1 0 5, following classic direct stiffness procedure:
sin ’ 0 cos ’
2 3 X ðiÞ
cos y sin y 0 ^ G ðoÞ ¼
K ^
LðiÞT K ðoÞLðiÞ ,
6 7
Rz ¼ 4 sin y cos y 0 5: i XYZ
0 0 1
where L(i) is the incidence matrix linking the
local vector of degrees of freedom of a structural
Applying the transformation matrix to the
member i to the global vector of degrees of
local system, the equilibrium system of equations
freedom.
for a frame member in the global reference frame
For each discrete frequency, the global prob-
is written:
lem to be solved is:
^ XYZ ðoÞ^
K ^ XYZ ðoÞ;
qXYZ ðoÞ ¼ F ^ Gq
K ^G;
^G ¼ F
Spectral Finite Element Approach for Structural Dynamics, Fig. 10 Seismic records using as input displacements
in numerical experiments
subjected to:
^F ¼ F
F ^0,
^P ¼ q
q ^0 ,
S
where the subscripts F and P are used, respec-
tively, to denote sets of degrees of freedom where
Neumann (prescribed loads) and Dirichlet
(prescribed displacements/rotations) boundary
conditions are applied.
The solution of the global system is possible
by rewriting it as:

^ FF K
K ^ FP q ^F ^
F
Spectral Finite Element Approach for Structural
^ ^ ¼ ^0 ;
Dynamics, Fig. 11 3D frame model KPF KPP q ^0 FP
Spectral Finite Element Approach for Structural Dynamics, Fig. 12 Response of the 3D frame structure to the
earthquake measured at (x = 10, y = 10) m
Numerical Examples
which allows one to get:
Earthquake Signal
^ 1 F
^F ¼ K
q ^0 K
^ FP q
^0
FF Seismic records from the Loma Prieta earthquake
^P ¼ K ^ PP K ^ 1 K
^ PF K ^ ^ 1 F
^ PF K
^0 þ K ^
F FF FP q FF 0 : which struck the San Francisco Bay area of
California in 1989 are used as input displace-

ments applied to the bottom nodes of the frame-
work. The input signal in time domain has
been obtained from an online database available
in Silva et al. (2000) and it is shown in Fig. 10.
3D Frame
Here, the dynamic response of a 3D framework
subjected to seismic ground motion is analyzed.
The model is presented in Fig. 11. Each
frame member is made of steel with mass
density r ¼ 7, 800 Kg:m3 , Young’s modulus
E ¼ 210 GPa, structural loss factor ¼ 0:04, Spectral Finite Element Approach for Structural
Dynamics, Video 1 Animation of a 3D frame under
and Poisson’s ratio n ¼ 0:3. All members have seismic ground motion
a circular cross section of radius 0.15 m.
The response of the structure at
ðx ¼ 10, y ¼ 10Þ m along the three orthogonal
directions is shown in Fig. 12, in the time and in
the frequency domains. It is worth pointing out
that the SEM has yielded a great reduction of the
problem size compared to classic FEM. The
response obtained from the SEM using 16 ele-
ments is in good agreement with the one obtained
by solving a classic FE problem composed of
1200 FE elements. An animation of the frame-
work dynamics during the passage of seismic
waves is simulated via the SEM and it is shown
in Video 1.

2D Frame Dynamics, Fig. 13 2D frame model
A 2D frame structure modeled using straight and
curved beam elements is shown in Fig. 13. The
frame has a square cross section 0.20 m 0.20 m ANSYS ®. As in the previous numerical prob- S
and steel as material, with Young’s modulus lem, fewer degrees of freedom were required to
E = 210 GPa, mass density r = 7,800 kg/m3, solve the problem via the spectral element
and structural damping = 0.02. method compared to the classic FEM, which
As in the previous section, the structure is results in CPU time savings. The 2D frame
excited at the base nodes by the Loma Prieta motion caused by the earthquake passage was
earthquake signal. Because of the in-plane con- computed via the SEM and it is available in
figuration, only the east–west and vertical dis- Video 2. The methodology presented in this
placement data are used. chapter can be used to derive other elements,
The output displacements at the node (x = 5, such as plates and shells (Doyle 1997), and to
y = 0) m are shown in Fig. 14. The results are study more complex structures under dynamic
validated with FEM results obtained using excitation.
Spectral Finite Element Approach for Structural Dynamics, Fig. 14 Response of the 2D frame structure to the
earthquake measured at (x = 5, y = 0) m
Summary
Elementary matrix of straight and curved beams,

rod under axial force, and torsion were developed
using spectral finite element approach. This
method can be used to analyze the structural
dynamic behavior. From the perspective of earth-
quake engineering, SEM allows to design struc-
tures subject to seismic excitation. The approach
has advantages such as low computational cost,
accuracy in frequency/time response, and simple
modeling of complex structures. The equations of
motion, spectral relation, and the spectral elemen-
tary matrix were derived. The simple elements
were combined into a 3D frame element and a
Dynamics, Video 2 Animation of a 2D frame under global matrix was assembled. Time and frequency
seismic ground motion responses of 3D and 2D frame structures under a
Steel Posttensioned Connections with Web Hourglass Pins: Toward Earthquake Resilient Steel Structures 3389
known seismic ground motion were computed

with SEM and results were validated using well- Steel Posttensioned Connections
known FEM commercial software. with Web Hourglass Pins: Toward
Earthquake Resilient Steel
Structures
Cross-References
George Vasdravellis1 and
▶ Nonlinear Dynamic Seismic Analysis Theodore L. Karavasilis2
▶ Nonlinear Finite Element Analysis 1
Institute for Infrastructure and Environment,
▶ Online Response Estimation in Structural Heriot-Watt University, Edinburgh, UK
Dynamics 2
School of Engineering, University of Warwick,
▶ Operational Modal Analysis in Civil Engineer- Coventry, UK
ing: An Overview
▶ Reliability Analysis of Nonlinear Vibrating
Systems-Spectral Approach Synonyms
▶ Robust Design Optimization for Earthquake
Loads Mitigation of residual drifts; Optimized seismic
▶ Steel Structures dampers; Seismic resilience; Steel self-centering
▶ Stochastic Analysis of Linear Systems frames
▶ Stochastic Finite Elements
▶ Structural Optimization Under Random
Dynamic Seismic Excitation Introduction
▶ Structures with Nonviscous Damping, Model-
ing, and Analysis Conventional steel moment-resisting frames
▶ Substructuring Methods for Finite Element (MRFs) are currently designed to form a global
Analysis plastic mechanism under the design basis earth-
▶ Time History Seismic Analysis quake (DBE) by developing plastic hinges at the
ends of the beams and the bases of the columns.
References This design methodology offers many advan-
tages, including collapse prevention and initial
Ahmida KM, Arruda JRF (2001) Spectral element-based economy; however, plastic hinges in structural
prediction of active power flow in Timoshenko beams. members involve difficulty to inspect and repair
Int J Solids Struct 38:1669–1679 damage and local buckling as well as residual
Doyle JF (1997) Wave propagation in structures, 2nd edn.
Springer, New York
drifts. The socioeconomic losses associated with
Lee U (2009) Spectral element method in structural damage and residual drifts are repair costs, S
dynamic. Wiley, Singapore increased downtime, and possibly demolition
Mace BR, Duhamel D, Brennan MJ, Hinke L (2005) Finite due to the complications associated with large
element prediction of wave motion in structural wave-
guides. J Acoust Soc Am 117(5):2835–2843
residual drifts (McCormick et al. 2008).
Mead DJ (1973) A general theory of harmonic wave A challenge of modern earthquake engineering
propagation in linear periodic systems with multiple is the development, standardization, and practical
coupling. J Sound Vib 27(2):235–260 implementation of resilient minimal-damage
Newmark NM, Hall WJ (1982) Earthquake spectra and
design. Earthquake Engineering Research Institute,
structures with the inherent potential to overcome
Berkeley the socioeconomic losses related to earthquake
Petyt M (2010) Introduction to finite element vibration anal- damage by minimizing or avoiding inelastic
ysis, 2nd edn. Cambridge University Press, Cambridge deformations and residual drifts. Examples of
Silva W, Bartling D, Spitzer C (2000) PEER strong motion
database. Regents of the University of California.
minimal-damage steel structures include frames
http://peer.berkeley.edu/smcat/search.html. Accessed equipped with rate-dependent passive dampers
9 July 2014 (Karavasilis et al. 2011; Whittle et al. 2012),
3390 Steel Posttensioned Connections with Web Hourglass Pins: Toward Earthquake Resilient Steel Structures
frames with steel energy dissipation devices hysteresis and fracture capacity of WHPs made of
(Karavasilis et al. 2012), and self-centering high performance steel, i.e., high-strength steel
moment-resisting frames (SC-MRFs) with and stainless steel (Vasdravellis et al. 2014).
posttensioned (PT) connections. It was found that the connection isolates inelastic
SC-MRFs exhibit softening force–drift behav- deformations in WHPs, avoids damage in other
ior while eliminating inelastic deformations and connection parts as well as in beams and col-
residual drifts under strong earthquakes as the umns, and eliminates residual drifts. The WHPs
result of gap openings developed in beam-to- do not interfere with the composite slab and are
column interfaces and self-centering capability very easy to replace without bolting or welding,
due to elastic pretensioning elements (e.g., high- and so, the connection enables nondisruptive
strength steel bars) which clamp beams to the repair and rapid return to building occupancy in
columns. PT connections use carefully designed the aftermath of a strong earthquake. This chapter
energy dissipation devices which are activated presents the development of the PT connection
when gaps open and can be classified into yield- through the experimental evaluation and the main
ing devices which dissipate energy through results of the numerical analyses.
inelastic deformations and devices which dissi-
pate energy through friction. Yielding devices
were proposed as angles bolted on the top and Steel Posttensioned Connection with
bottom flanges of the beam and on the column Web Hourglass Pins
flanges, dissipating energy through inelastic
bending (Ricles et al. 2001, 2002; Garlock Figure 1 shows a schematic representation of an
et al. 2005); buckling-restrained steel bars placed SC-MRF incorporating the proposed PT connec-
between the beam flanges and welded on the tion and the details of an exterior PT connection.
beam and column, dissipating energy through Two high-strength steel bars located at the
axial deformations (Christopoulos et al. 2002); mid-depth of the beam, one at each side of the
reduced flange plates welded around a square web, pass through holes drilled on the column
hollow section column and bolted on the beam flanges. The bars are posttensioned and anchored
flanges (Chou et al. 2006); and reduced-section or to the exterior column flange and hence clamp the
cross-shaped steel plates placed below the bottom beam to the column. Four cylindrical steel pins
flange of the beam (Chou and Lai 2009). Friction- (WHPs) are inserted in aligned holes drilled on
based devices were proposed as friction-bolted the web of the beam and on strong supporting
surfaces placed on the top and bottom flanges of plates. The supporting plates are welded to the
the beam (Rojas et al. 2004; Kim and column flanges and are of greater thickness to
Christopoulos 2008a), on the web of the beam provide fixed support boundary conditions to the
(Tsai et al. 2008), or on the bottom flange of the WHPs. Energy dissipation is provided by the
beam (Wolski et al. 2009). inelastic bending of the WHPs, which are sym-
A new steel PT beam–column connection metrically placed (close to the top and bottom
using web hourglass shape steel pins (WHPs) as beam flange) to provide increased lever arm and
EDs has been recently developed by the authors. hence increased internal moment resistance. As
The development phases included (a) large-scale shown in Fig. 2, the WHPs are designed to have
experimental evaluation (Vasdravellis an hourglass shape to provide enhanced energy
et al. 2013a) of the proposed connection, dissipation and fracture capacity (Kobori
(b) detailed finite element models (FEM) and et al. 1992). Both sides of the beam web are
associated parametric studies (Vasdravellis reinforced with steel plates to increase the contact
et al. 2013b), and (c) numerical simulations and surface of the WHPs with the web. In that way,
seismic analysis of SC-MRFs equipped with PT possible ovalization of the holes drilled on the
connections with WHPs (Dimopoulos web and the reinforcing plates under the
et al. 2013). A more recent study evaluated the WHP-bearing forces will be negligible, and
Steel Posttensioned
Connections with Web
Hourglass Pins: Toward
Earthquake Resilient
Steel Structures,
Fig. 1 (a) SC-MRF
incorporating the proposed
PT connection; (b) exterior
PT connection details
Internal Ext. part PT bar forces. In addition, the panel zone is

Ext. part
part strengthened with horizontal stiffeners and con-
r
tinuity plates along the web of the column.
De Di
Connection Behavior and Simplified

LWHP Analysis
symm.
Flexural Behavior
d VWHP The connection behavior is characterized by gap
opening and closing in the beam–column inter-
[M ] + face as a result of the re-centering force in the PT S
– bars. Figure 3 shows the gap-opening mechanism
in the connection where d1u and d1l are the dis-
[V ] + tances of the upper and lower WHPs from the
center of rotation (COR), respectively; d2 is the
Steel Posttensioned Connections with Web Hour- distance of the PT bars from the COR; FPT is
glass Pins: Toward Earthquake Resilient Steel Struc-
the total force in both PT bars; FWHP,u and
tures, Fig. 2 Geometry of half a WHP, assumed static
system, and internal force diagrams FWHP,l are the forces in the upper and lower
WHPs, respectively; and CF is the compressive
force on the beam–column bearing surface. It is
pinching behavior under cyclic deformations can assumed that the COR is located at the inner edge
be avoided. The connection includes beam of the beam flange-reinforcing plate. This
flange-reinforcing plates to avoid excessive assumption was verified during the connection
early yielding in the beam flanges under high experiments described later.
Steel Posttensioned
Steel Structures,
Fig. 3 Gap-opening
mechanism in the proposed
PT connection
Steel Posttensioned
Steel Structures,
Fig. 4 Force versus
displacement cyclic
behavior of one WHP
Figure 4 shows the assumed bilinear opens, and the connection behavior becomes
elastoplastic force–displacement (FWHP,i-d) nonlinear elastic. Following the procedure
behavior of one of the four (i = 1 to 4) WHPs. described in Garlock et al. (2007), the stiffness
Figure 5a plots the assumed cyclic K1 between points 1 and 2 is equal to the sum of
moment–rotation (M-y) behavior of the connec- the rotational stiffness contribution of the PT
tion. M is the sum of the moment contributions bars, KPT, and the rotational stiffness contribution
from the PT bars, MPT, and the WHPs, MWHP. of the WHPs, Ke:
Figure 5b shows the nonlinear elastic moment
contribution of both PT bars. In Fig. 5b, Md is K PT, a K b 2
K 1 ¼ K PT þ K e ¼ d 2 þ 2K fe d 21l þ d 21u
the decompression moment, i.e., equal to FPT,id2, K PT, a þ K b
where FPT,i is the total initial posttensioning force (1)
in both PT bars. Figure 5c shows the moment
contribution of the four WHPs. The MWHP-y where Kfe is the elastic stiffness of the force–dis-
curve changes slope two times since the upper placement relationship of one WHP shown in
WHPs yield at a rotation y2 and the lower WHPs Fig. 4, Kpt,a is the total axial elastic stiffness of
yield afterwards at rotation y3. Point 1 in Fig. 5a both PT bars, and Kb is the axial elastic stiffness
corresponds to the decompression moment, Md, of the beam. At point 2, the upper WHPs yield
of the connection. After decompression, gap and M continues to increase with slope K2:
a M When loading is reversed, the connection begins

K1 K2
4 to unload, from point 4 to point 5, ideally with the
K3
same elastic stiffness, K1. After point 5, the con-
2 3 nection unloads with stiffness equal to K2
6 5
1 between points 5 and 6 and with stiffness K3
7 between points 6 and 7 until the gap closes. The
04 05 06 03 02 behavior of the connection is symmetric, since
02 03 06 05 04 q
the WHPs are placed symmetrically to the beam
centerline. The moment in the connection, M, is
calculated as:
M ¼ MPT þ MWHP (4)

b MPT
KPT Following the procedure described by Garlock
Md et al. (2007), MPT is given by:
K PT, a K b 2
04 05 06 03 02
MPT ¼ Md þ d y (5)
02 0 3 0 6 0 5 04 q K PT, a þ K b 2
According to Fig. 3, MWHP can be obtained as:
MWHP ¼ FWHP, u d1u þ FWHP, l d 1l (6)

c M WHP
Ke Kp1 Plastic Design of WHPs
Kp2 Plastic analysis is used to calculate the force in
the WHPs as they deform due to gap opening and
q4 q3 q2 q6 q5
q5 q6 q 2 q3 q4 q closing. The assumed static system for half a
WHP is shown in Fig. 2. The yield strength,
VWHP, of half a WHP is controlled either by the
plastic moment of resistance, Mpl, or the plastic
Steel Posttensioned Connections with Web Hour- shear resistance, Vpl, (Eurocode 3 2003), i.e.:
tures, Fig. 5 Conceptual cyclic moment–rotation
D3e
behavior of the proposed PT connection Mpl ¼ f (7)
6 y
S
K 2 ¼ K PT þ K p1 pD2
0:9 i fy
K PT, a K b 2 Vpl ¼ p4ffiffiffi (8)
¼ d 2 þ 2 K fe d21l þ K fp d21u (2)
K PT, a þ K b 3
where Kfp is the post-elastic stiffness of the where fy is the yield strength of the WHP mate-
force–displacement relationship of one WHP rial, De is the equivalent external diameter (to be
shown in Fig. 4. At point 3, the lower WHPs defined later), and Di is the diameter at the
yield and M continues to increase with slope K3: mid-length of half a WHP, as indicated in
Fig. 2. The factor 0.9 in Eq. 8 accounts for the
relation between the average shear stress and the
K 3 ¼ K PT þ K p2
maximum shear stress in a circular section. The
K PT, a K b 2
¼ d þ 2K fp d21l þ d21u (3) internal WHP part is connected to the external
K PT, a þ K b 2 WHP part using a radius of 5 mm to avoid stress
Steel Posttensioned
Steel Structures,
Fig. 6 Prototype steel
building: (a) plan view and
(b) elevation and cross-
sections of the interior steel
MRF used in this study
concentration and early fracture. It is assumed where E is the modulus of elasticity and G the
that De, which controls the WHP bending resis- shear modulus of the material. The WHP post-
tance, is equal to the diameter at the start of the yield stiffness Kfp is assumed to be 2 % the initial
round-shaped part with radius r. Plastic analysis stiffness, i.e., Kfp = 0.02Kfe.
assumes that the plastic moment of resistance
should be reached at the ends before the plastic
shear resistance is reached at the mid-length of Design of Prototype SC PT Connection
half a WHP. By equilibrium, the aforementioned with WHPs
condition can be written as:
Prototype Building
2MPl Figure 6a shows the plan view of the prototype
VWHP ¼ < Vpl (9)
LWHP building used for this study. The study focuses on
the interior steel MRF shown in Fig. 6b. The yield
where LWHP is the clear length of the bending stress of structural steel is equal to 300 MPa. The
parts of half a WHP. From Eqs. 7 to 9, any design seismic action, referred to herein as DBE,
combination of fy, De, Di, and LWHP can be cho- has a return period equal to 475 years and is
sen to provide the desired VWHP. The yield force expressed by the elastic design spectrum of
of a WHP, Fy,WHP, is then calculated as: Eurocode 8 (2004) with peak ground acceleration
equal to 0.3 g and soil B. The interior MRF is first
Fy,WHP ¼ 2V WHP (10) designed as a conventional MRF according to
EC3 (2003) and EC8. The behavior (or strength
The virtual work method along with analytical reduction) factor q is equal to 6.5. Eurocode
integration was applied to derive the elastic stiff- 8 imposes a serviceability limit on the peak
ness Kfe of a WHP: story drift, ymax, under the frequent earthquake
(FOE) with a return period equal to 95 years. The
9pD3e Di EG
K fe ¼ 2 (11) FOE has intensity equal to 40 % the intensity of
40ED2e LWHP þ 48GL3WHP the DBE, and the limit on ymax is 0.5 % and
0.75 % for non-ductile and ductile nonstructural for drifts higher than the MCE drifts. Such a
components, respectively. A strength-based design approach is slightly different than the
design under the DBE was first performed. one proposed by Kim and Christopoulos
Beams and columns from this design had to be (2008b) where the connection is designed to
increased to satisfy the limit on ymax under the form a ductile plastic hinge for drifts approxi-
FOE. The final sections are shown in Fig. 6b and mately equal to the DBE drifts.
were found iteratively, i.e., by decreasing the The design of the PT connection is based on
value of q, designing the MRF under the DBE, drifts equal to 0.64 % under the FOE, 1.6 % under
and then checking drifts under the FOE. The con- the DBE, and 2.4 % under the MCE according to
ventional MRF has story drifts equal to 0.64 % the prototype building design. These drifts are
under the FOE, 1.6 % under the DBE, and 2.4 % significantly lower than those used in previous
under the maximum considered earthquake works which are based on the IBC 2 % drift limit
(MCE). The MCE has intensity equal to 150 % under the DBE (Rojas et al. 2004; Garlock
the intensity of the DBE. et al. 2007). To initiate the design procedure,
The sections of the conventional MRF are the moment corresponding to point 2 in Fig. 5a,
used for designing the proposed PT connection referred to herein as MIGO, is set equal to 66 % of
(see next section). In that way, a SC-MRF using the nominal plastic moment of resistance of the
the proposed PT connections would have the beam section. In addition, Md is set equal to 69 %
same initial stiffness and period of vibration of MIGO. These MIGO and Md values provide an
with the conventional MRF. energy dissipation factor, b (Garlock et al. 2007),
equal to 0.31. The posttensioning force and the
Performance-Based Design of PT Connection length of the reinforcing plates are then designed.
with WHPs PT bars are designed to avoid yielding for
The design focuses on the exterior connection of y 0.07 rad.
the 2nd floor of the prototype steel MRF shown in
Fig. 6b. This connection is designed as a PT
connection with WHPs using the methodology Hysteretic Characterization of WHPs
presented by Garlock et al. (2007) that adopts
two performance objectives and associated struc- Component tests on the WHPs were conducted to
tural limit states, namely, (1) immediate occu- assess their energy dissipation capacity and ductil-
pancy under the DBE by avoiding damage in ity under cyclic loading. Figure 7 shows the exper-
beams and columns while permitting gap opening imental setup for the tests. A pair of WHPs was
and (2) collapse prevention under the MCE by tested to reproduce their behavior in the proposed
avoiding PT bar yielding and beam local buck- PT connection. Two supporting plates were welded
ling while permitting minor yielding in beams on a fixed plate. A plate simulating the web of the S
and columns. beam was bolted to a thick plate attached to the
The design methodology proposed by Garlock actuator. The thicknesses of the plates were equal
et al. (2007) uses beam flange-reinforcing plates to those of the plates designed for the large-scale
to avoid early beam flange yielding and buckling. PT connection tests discussed later.
However, the connection can experience sudden Four coupon tests were conducted to identify
loss of strength and stiffness due to beam web the material properties of the WHPs. The results
local buckling when deformed beyond the MCE of the coupon tests are summarized in Table 1.
drift. In this work, the use of the structural details The WHPs were made of 1020-Grade carbon
(drilled holes on the beam flanges and beam web steel, with an average yield strength 557 MPa,
stiffeners) proposed by Kim and Christopoulos ultimate strength 598 MPa, and elongation at
(2009) in a connection designed according to the fracture equal to 20 %. This type of steel was
procedure described by Garlock is also examined chosen to achieve the required strength while
with the aim to further delay beam local buckling keeping the sizes of the WHPs relatively small.
Steel Posttensioned Connections with Web Hour-

tures, Table 1 Material properties of WHPs from
coupon tests
Yield Tensile %
Coupon stress stress E Elongation
no. (MPa) (MPa) (GPa) at fracture
1 551 595 207 17
2 550 594 209 20
3 564 599 210 23
4 563 604 208 20
Average 557 598 208 20
changes sign due to slight ovalization of the holes

of the supporting plates under the bearing forces
induced by the WHPs.
Large-Scale Experimental Program

glass Pins: Toward Earthquake Resilient Steel Struc- Test Setup
tures, Fig. 7 Component test setup for WHPs Experiments on PT connections equipped with
(dimensions in mm)
WHPs were conducted in the test setup shown in
Fig. 9. The test specimens were based on the
The hourglass shape of the WHPs was prototype second-floor exterior beam-to-column
designed to be consistent with their bending connection (see Fig. 6) at 0.6 scale. A strong
moment diagram shown in Fig. 2. Such a design 310UC158 column was used to minimize the col-
results in almost uniform distribution of inelastic umn deformations since they do not influence the
deformations along the length of the WHPs and, behavior of the main connection components
hence, in enhanced energy dissipation and frac- (beam, PT bars, and WHPs). Two additional steel
ture capacity. For the tested configuration, the members were welded to the column to form a
design resulted in WHPs with external diameter truss system which increases the horizontal stiff-
De = 20 mm, internal diameter Di = 14 mm, and ness of the test setup (310UC158 horizontal mem-
Lwhp,i = 40 mm, while the external part had a ber and 200UC52 diagonal member in Fig. 9). The
diameter equal to 30 mm. whole system was bolted on the strong floor. The
Figure 8a shows the deformed WHPs during displacement history was applied vertically by a
the component test. WHPs were capable of sus- hydraulic actuator positioned at a distance of
taining repeated large inelastic cycles without 1,800 mm from the inner face of the column.
fracture up to displacements associated with con-
nection rotation of 0.07 rad. Figure 8b plots the Test Specimens
force–displacement hysteresis of the WHPs. Two connection specimens (SC-WHP1 and
WHPs provide a yielding force equal to 150 kN SC-WHP2) were designed according to the pro-
which is in excellent agreement with the 147 kN cedure described earlier. The resulting beam and
WHP yielding force calculated by the plastic column sizes and details of the connection spec-
analysis procedure presented previously. The imens are summarized in Table 2. The specimens
ultimate strength of the WHPs at the initiation are identical except for the beam web stiffeners
of fracture is 160 kN. Slight pinching (short flat and beam flange drilled holes used in specimen
region) is observed at the points where the force SC-WHP2.
a b 200
100
Force (kN)
0
−100
−200
0 5 10 15
Displacement (mm)
Steel Posttensioned Connections with Web Hourglass Pins: Toward Earthquake Resilient Steel Structures,
Fig. 8 WHPs test results: (a) deformed WHPs and (b) WHPs hysteresis
Steel Posttensioned
Steel Structures,
Fig. 9 PT connection test
setup (dimensions in mm)
Steel Posttensioned Connections with Web Hour- thick plates as shown in Fig. 10b. The thickness
tures, Table 2 Specimen details
of the WHP-supporting plates is 36 mm, their
width is 80 mm, and a weld radius of 6 mm was
SC-WHP1 SC-WHP2
used for the welding of all the plates in the
Beam section 250UB37 250UB37
connection.
Column section 310UC158 310UC158
Figure 11 shows the SC-WHP2 specimen
FPT,i (kN)a 518 504
which is identical to the SC-WHP1 apart from S
Lrp (mm)b 700 700
Web stiffeners length n/a 540
four 540 mm long longitudinal stiffeners which
(mm) are welded on the web and four 27 mm holes
Drilled holes diameter n/a 27 drilled on the flanges immediately after the end
(mm) of the reinforcing plate. The longitudinal stiff-
a
FPT,i: Initial posttensioning force eners have thickness 10 mm, start 270 mm
b
Lrp: Length of reinforcing plates on the beam flanges before and are extended 270 mm beyond the
end of the reinforcing plates. The longitudinal
stiffeners aim to resist the web buckling in the
Figure 10a shows the SC-WHP1 specimen, beam which typically occurs after the end of the
while Fig. 10b shows a close-up view of its reinforcing plates due to the increased bending
connection region. Beam flanges are reinforced moment and compressive force developed in
with 700 mm long and 10 mm thick plates, while this section as the gap opening increases.
the web of the beam is reinforced with 10 mm The drilled holes aim to reduce the moment
Steel Posttensioned
Steel Structures,
Fig. 10 SC-WHP1
specimen: (a) overview and
(b) close-up view of the
connection region
capacity of the beam in this region in order to coupons cut from the flanges and webs of the
allow for the formation of a plastic hinge under beams. Table 3 summarizes the mean actual
large drifts. A dog-bone detailing could be used yield stress, tensile stress, and modulus of elas-
instead. ticity resulting from the material tests. The cou-
pon tests showed that the mean yield stresses are
Material Properties 361 and 388 MPa for the flanges and the web of
The beam and column sections were made of the beam, respectively, indicating considerable
steel with nominal yield strength equal to overstrength of the steel material. The material
300 MPa. The reinforcing plates, WHPs of the PT bars has nominal yield strength equal to
supporting plates, and stiffeners were made of 930 MPa, tensile stress 1,050 MPa, and elonga-
steel with nominal yield strength equal to tion capacity 6 %, according to the specifications
350 MPa. Tensile tests were carried out on of the supplier.

tures, Table 3 Material test results
Modulus
Yield Tensile of
Sample stress stress elasticity
Coupon no. (MPa) (MPa) (GPa)
Flange 1 354 535 208
2 370 536 207
3 359 530 208
Average 361 534 208
Web 1 385 547 205
2 392 548 199
3 387 545 176
Steel Posttensioned Connections with Web Hour- Average 388 547 193
tures, Fig. 11 Details of SC-WHP2 specimen
of the actuator tip to the inner face of the column.

Test Procedure Figure 12a shows that the connection maintains
The loading protocol for the two specimens con- self-centering capability for drifts up to 6 %. The
sists of cyclic vertical displacements of increas- initial stiffness of the connection is 5 kN/mm and
ing amplitude imposed in a quasi-static fashion is equal to the theoretical stiffness of a welded
with a speed approximately equal to 21 mm/min. (fully restrained) connection. During the last
The AISC (2005) loading protocol was used. This cycle of 9 % drift, the connection lost its self-
protocol consists of three initial sets of six cycles centering capability and experienced a residual
at 6.75, 9, and 13.5 mm displacements, four sub- drift equal to 3 %.
sequent cycles at 18 mm, and six sets of two To assess the reparability of the proposed
cycles at 27, 36, and 54 mm. These displacements connection after a strong earthquake, the
correspond to drifts equal to 0.00375, 0.005, SC-WHP1 specimen was initially imposed to
0.0075, 0.01, 0.015, 0.02, and 0.03. The speci- displacements up to the DBE level (i.e., 1.6 %
mens were further imposed to drifts equal to 0.06 drift). The test was then interrupted, and dam-
and, then, up to 0.09–0.10 to identify the connec- aged WHPs were substituted by new ones. The
tion failure mode. substitution was easily accomplished according
to the following steps: (a) a temporary support
was placed below the beam close to the
Experimental Results beam–column interface to accommodate the S
self-weight of the beam; (b) the PT force was
Specimen SC-WHP1 relaxed; (c) the damaged WHPs were taken out
Figure 12a shows the force–drift hysteresis of the with the aid of a wooden hammer; and (d) new
SC-WHP1 specimen, while Fig. 12b shows WHPs were placed in the connection. Since no
the associated moment–rotation hysteresis. In welding or bolting is needed for the WHPs
Fig. 12a, the drift is calculated as the ratio of the replacement, the proposed connection can be
vertical displacement at the tip of the actuator to repaired with minimal disturbance to building
the horizontal distance of the tip of the actuator to use or occupation in the aftermath of a major
the inner face of the column. The rotation is earthquake. After the substitution of the WHPs,
calculated as the gap-opening angle at the the test was restarted. Figure 12a plots
beam–column interface measured during the superimposed loops from the tests before and
tests. In Fig. 12b, the moment is calculated as after the replacement of WHPs. During the first
the product of the actuator force and the distance cycles, the connection loses part of its initial
a 100 b 200
Gap opening
50 100
Moment (kNm)
Force (kN)
Ultimate
Ultimate
response
0 response 0 Gap opening
Ki=5kN/mm
−100 1st Run
−50
1st run 2nd Run
2nd run Analytical
−100 −200
−10 −5 0 5 10 −0.08 −0.04 0 0.04 0.08
Drift (%) Rotation (rad)
c 200 d 200
100 100
MWHP (kNm)
MPT (kNm)
0 0
1st Run 1st Run

−100 −100 2nd Run
2nd Run
Analytical Analytical
−200 −200
−0.08 −0.04 0 0.04 0.08 −0.08 −0.04 0 0.04 0.08
Rotation (Rad) Rotation (Rad)
Fig. 12 Hysteretic response of SC-WHP1 specimen and comparison with analytical predictions
stiffness as a result of the unavoidable local Figure 12b shows that the connection achieves
yielding of the beam edges at the beam–column rotations larger than 0.035 rad with full self-
interface. Nevertheless, Fig. 12a shows that the centering capability. The experimental moments
connection gains its stiffness after drifts of 1 %. at decompression and gap opening are
In fact, the two loops are perfectly superimposed Md = 62 kNm and MIGO = 95 kNm, respectively.
to each other, demonstrating that the proposed These values are slightly smaller than the design
connection can be repaired after a strong earth- values of Md = 66 kNm and MIGO = 100 kNm.
quake without compromising its stiffness, However, the predicted analytical curve captures
strength, and energy dissipation capacity. well the cyclic behavior in terms of both stiffness
Figure 12b plots superimposed moment– and strength. The analytical prediction is also in
rotation loops from the tests before and after good agreement with both the MPT-y and the
replacement of WHPs along with the analytical MWHP-y loops, as shown in Fig.12c, d, respectively.
moment–rotation prediction using the analytical Displacement measurements at the
expressions of the simplified procedure described beam–column interface indicated a vertical trans-
earlier. The moment contribution of the PT bars, lation of the beam axis with respect to its initial
MPT, is calculated from the posttension force position of about 4 mm. This translation took
recorded by the load cells and is plotted in place gradually during the testing procedure and
Fig. 12c, while the contribution of the WHPs, is attributed to a gradual loss of friction between
MWHP, is found by subtracting MPT from the the beam and the column interface. Shear tabs
total moment of the connection and is plotted in could be welded on the column flange and bolted
Fig. 12d. on the beam web using slotted holes to
Steel Posttensioned Connections with Web Hour- Steel Posttensioned Connections with Web Hour-
glass Pins: Toward Earthquake Resilient Steel Struc- glass Pins: Toward Earthquake Resilient Steel Struc-
tures, Fig. 13 Gap opening at 6 % imposed drift of the tures, Fig. 14 Web buckling of the SC-WHP1 specimen
SC-WHP1 specimen
accommodate gravity loads and eliminate the 1200

vertical beam displacement without influencing Web local
the connection moment behavior. Another alter- buckling
PT force (kN)
native to accommodate gravity loads could be the 900

use of a seat angle positioned below the bottom
flange of the beam.
Figure 13 shows a close-up view of the 600
gap-opening angle corresponding to 6 % drift. 7kN
Apart from the intended damage in the WHPs
and local yielding in the beam edges at the 300
beam–column interface, no other evidence of −10 −5 0 5 10
damage was observed for drifts lower or equal Drift (%)
to 6 %. Yielding in the beam flanges at the
beam–column bearing surface was evident from glass Pins: Toward Earthquake Resilient Steel Struc-
the first cycles of the test. In fact, it was visually tures, Fig. 15 PT force versus drift for the SC-WHP1
verified that the extreme edges of the beam at the specimen
beam–column interface became rounded during
the test. As a result, the location of the COR was yielding at any point throughout the beam sec- S
not constant during the experimental process. tion when local buckling initiated.
At the initial cycles, the COR was located Figure 15 plots the variation of the force in the
approximately at the center of the reinforcing PT bars with respect to drift. The PT force
plate, while at larger drifts, it was stabilized at increases with gap opening and returns to its
the inner side of the reinforcing plates. initial value of 504 kN when gap closes. The
During the last cycle of 9 % drift, the speci- loss in posttensioning during the test was about
men failed due to local buckling of the web of 7 kN. The maximum force in the PT bars has
the beam immediately after the end of the values of 850 kN and 970 kN at 6 % and 9 %
reinforcing plates. Figure 14 shows a close-up drift, respectively. These values show that the PT
view of the buckled region. Web buckling took bars remained elastic during the test since their
place at 7 % drift, while local buckling of nominal yield limit is 1,227 kN. A drop in the PT
the beam flanges initiated at the same time. force due to beam local buckling is observed
Strain gauge measurements did not indicate during the last cycle.
a 100 b 200
Gap opening
50 100
Moment (kNm)
Force (kN)
Ultimate
response Ultimate
0 0 Gap opening
response
−50 K = 5kN/mm −100 Test

i
Analytical
−100 −200
−10 −5 0 5 10 −0.08 −0.04 0 0.04 0.08
Drift (%) Rotation (Rad)
c 200 d 200
100 100
MPT (kNm)
MWHP (kNm)
0 0
−100 Test −100 Test

Analytical Analytical
−200 −200
−0.08 −0.04 0 0.04 0.08 −0.08 −0.04 0 0.04 0.08
Rotation (Rad) Rotation (Rad)
Fig. 16 Hysteretic response of specimen SC-WHP2 and comparison with analytical predictions
Specimen SC-WHP2 agreement with the individual experimental

Figure 16a shows the cyclic force–drift behavior moment contributions MPT and MWHP.
of the SC-WHP2 specimen. The behavior is sta- The detailing employed for specimen
ble with full self-centering capacity up to 6 % SC-WHP2 resulted in a different failure mode.
drift similar to specimen SC-WHP1. A final No web or flange buckling was observed in the
experimental loop of 10 % drift caused the con- beam section after the end of the beam flange-
nection to fail, and a residual drift of 4 % was reinforcing plates. In addition, no plastic hinge
present at the end of the test. was developed in this region. The strain gauges
The moment versus relative rotation hystere- indicated that the yield strain was not reached in
sis is plotted in Fig. 16b. The experimental Md any point throughout the beam section at the end of
value was 60 kNm, while MIGO was 92 kNm. the beam flange-reinforcing plate. Figure 17 shows
These values are somewhat smaller than the that the SC-WHP2 specimen failed due to exces-
corresponding values of the SC-WHP1 specimen sive yielding in the beam flanges at the
due to the slightly smaller initial posttensioning beam–column interface under large bearing
force (504 instead of 518 kN). The simplified forces. Local yielding at the bearing interface of
design procedure predicts well the cyclic enve- specimen SC-WHP2 was more pronounced than in
lope with the predicted values for Md and MIGO specimen SC-WHP1 and resulted in a gradual
being 64 and 98 kNm, respectively. Figure 16c–d reduction in posttensioning force. Figure 18
shows that the analytical predictions are in good shows that the loss of force in the PT bars was
1200
15kN
PT force (kN)
900
600
Beam flanges 50kN
excessive
300 yielding
−10 −5 0 5 10
Drift (%)

Steel Posttensioned Connections with Web Hour- glass Pins: Toward Earthquake Resilient Steel Struc-
glass Pins: Toward Earthquake Resilient Steel Struc- tures, Fig. 18 PT force versus drift for the SC-WHP2
tures, Fig. 17 Local yielding in the beam flange at the specimen
bearing interface of the SC-WHP2 specimen
with the FEM model and its boundary conditions.

50 kN during the test, with the largest amount Since the actual geometry was symmetric, only
attributed to loading cycles associated with drifts one fourth of the specimens was modeled to
larger than the MCE drift. A reduction of 15 kN in decrease computational time.
the force of the PT bars was observed as a result of Figure 20a shows the deformed shape of the
the local failure at the final loading cycle. WHP and the equivalent plastic strain (PEEQ)
distribution at a displacement of 12 mm.
Figure 20a shows that the hourglass shape geom-
Finite Element Simulations and etry results in uniform distribution of plastic defor-
Parametric Study mation along the length of the internal part of the
WHP, while the external parts of the WHP and the
This section presents FEM models which can be supporting plates are essentially elastic. The
used to reliably assess the design and behavior of PEEQ distribution in the WHP’s supporting plates
the connection. A detailed nonlinear FEM model is plotted separately in Fig. 20b which shows
was developed in Abaqus (Dassault Systems minor plastic concentrations. These results agree
2010). The FEM model was calibrated against with the experimental test observations where a
experimental results and found capable to trace negligible ovalization of the holes of the
the nonlinear cyclic behavior of the connection supporting plates was evidenced. S
and capture all possible local failure modes. The Figure 21 compares the force–displacement
calibrated FEM model was used to conduct a hysteresis from FEM analysis with the experi-
series of simulations to study the effect of differ- mental hysteresis from the WHPs characteriza-
ent parameters on the connection behavior. tion tests. The FEM model can trace the nonlinear
cyclic behavior of the WHPs with good accuracy.
Models for the WHPs The model captures the pinching behavior at zero
The cyclic behavior of the WHPs was first simu- force which is observed in the experimental hys-
lated. The FEM model geometry reproduced the teretic curve as the result of the negligible
actual geometry of the test setup of the compo- ovalization of the holes of the supporting plates.
nent tests that were carried out to characterize the Figure 21 shows that the experimental hysteresis
behavior of the WHPs. Figure 19 shows the setup of the WHPs deteriorates after a displacement
used for the WHPs characterization tests along amplitude of 12 mm. The WHP failed due to
Fig. 19 The FEM model used for the WHPs
PEEQ PEEQ
(Avg: 75%) (Avg: 75%)
+7.158e−01 1 +1.773e−02
+6.561e−01 +1.625e−02
+5.965e−01 +1.477e−02
+5.368e−01 +1.330e−02
+4.772e−01 +1.182e−02
+4.175e−01 +1.034e−02
+3.579e−01 +8.865e−03
+2.982e−01 +7.387e−03
+5.910e−03
+2.386e−01 +4.432e−03
+1.789e−01 +2.955e−03
+1.193e−01 +1.477e−03
+5.965e−02 +0.000e+00
+0.000e+00
Fig. 20 Deformed shape of the WHP and equivalent plastic strain (PEEQ) contour plot
ductile fracture at the section close to the internal improve their bending behavior. The mesh was
support plate under a displacement amplitude of refined in regions where severe plastic deforma-
12 mm. tions or buckling phenomena were expected to
occur, i.e., close to the beam–column interface
Models for the Connection and at the end of the beam flange-reinforcing
A three-dimensional FEM model was developed plates. A coarser mesh was used for regions that
to simulate the behavior of the connections as were expected to remain essentially elastic, i.e.,
shown in Fig. 22. The geometry of the tests was the anchor block, the tip of the beam, and the top
reproduced in full detail. The column, the PT and bottom parts of the column. The WHPs and
bars, the WHPs, and the plates were modeled their supporting plates were modeled using the
using C3D8R solid elements. The beam was verified model for the WHPs described in the
modeled using solid elements with incompatible previous section.
modes (C3D8I). C3D8I are first-order elements The vertical load was applied as an imposed
that are enhanced by incompatible modes to displacement, U2, at a distance equal to 1,800 mm
from the connection face as shown in Fig. 22a. capable of producing the desired FPT,i in the
The analysis consisted of several steps. In the first beam. In the subsequent steps, the cyclically dis-
step, the contact interactions were established to placement history was applied. Displacement-
ensure that numerical problems due to contact controlled nonlinear analysis was performed.
formulation will not be encountered during the
next steps. The posttensioning force was applied Assessment of the Finite Element Model
during the second analysis step by imposing an Figure 23 plots the force–drift hysteresis from
axial displacement at the free ends of the PT bars FEM analyses along with the experimental hys-
teresis of the two connection specimens. The
FEM model can capture well the overall cyclic
200 behavior of the connections. The predicted values
for the Md and MIGO are in very good agreement
100 with the corresponding experimental ones, while
the initial and post-elastic stiffness are almost
Force (kN)
identical. The variation of FPT from experiments

0 is compared with the FEM prediction in Fig. 24.
The FEM simulations and experiments are in
reasonably good agreement. The numerical
−100
Tests model predicts a slightly smaller loss in FPT
FEM
than that measured one in the experiments for
−200 specimen SC-WHP1, while the loss in FPT is
0 5 10 15 more accurately predicted for specimen
Displacement (mm) SC-WHP2.
Following the loading protocol of the experi-
mental program, the FEM models were pushed to
tures, Fig. 21 Comparison of the experimental and the large drifts to investigate all possible global and
numerical force–displacement hysteresis of the WHPs local failure modes. Figure 23 shows that the two
Steel Posttensioned Connections with Web Hour- discretization, (b) model for specimen SC-WHP1, and
glass Pins: Toward Earthquake Resilient Steel Struc- (c) model for specimen SC-WHP2
tures, Fig. 22 FEM model details: (a) FEM
a 100 b 100
50 50
Force (kN)
Force (kN)
0 0
−50 −50
SC-WHP1 (Test) SC-WHP2 (Test)
SC-WHP2 (FEM)
SC-WHP1 (FEM) −100
−100
−10 −5 0 5 10 −10 −5 0 5 10
Drift (%) Drift (%)
Steel Posttensioned Connections with Web Hour- connection specimens tested in Kim and Christopoulos
glass Pins: Toward Earthquake Resilient Steel Struc- (2008b) and comparison with FEM results: (a)
tures, Fig. 23 Force–drift hysteresis of the two SC-WHP1 and (b) SC-WHP2
a 1000 b 1000
800 800
FPT (kN)
FPT (kN)
600 600
400 400
200 SC-WHP1 (Test) 200 SC-WHP2 (Test)

SC-WHP1 (FEM) SC-WHP2 (FEM)
0 0
−10 −5 0 5 10 −10 −5 0 5 10
Drift (%) Drift (%)
Steel Posttensioned Connections with Web Hour- specimens tested in Kim and Christopoulos (2008b) and
glass Pins: Toward Earthquake Resilient Steel Struc- comparison with FEM results: (a) SC-WHP1 and (b)
tures, Fig. 24 FPT-drift plots for the two connection SC-WHP2
large last cycles of 9 % drift for specimen three drift levels: DBE (i.e., Design Basis Earth-
SC-WHP1 and 10 % drift for specimen quake (Charney and Downs 2004)) = 1.6 %,
SC-WHP2 resulted in residual inelastic drifts MCE = 2.4 %, and Ultimate = 9–10 %. In this
and loss of self-centering capability. These large figure, darker areas indicate larger plastic strain
drifts resulted in different failure modes of the concentration. The FEM models captured the
two connection specimens, consistent with the failure modes observed in the experimental pro-
experimental observations. The FEM model for gram. Figure 25 shows that under the DBE drift,
specimen SC-WHP1 failed due to web buckling the connection damage is isolated in the WHPs.
immediately after the end of the reinforcing Under the MCE drift, damage is concentrated
plates. The FEM model for specimen SC-WHP2 in the WHPs and in the beam flanges in the
failed due to yielding in the region of the drilled beam–column interface. Damage in the
holes on the beam flange and in the bearing inter- beam–column interface is more evident in
face, while the web of the beam remained elastic the connection SC-WHP2. Under the ultimate
until the end of the analysis. This can be verified drift, damage in the connection is spread along
by the PEEQ value distribution in the deformed the beam web immediately after the end of the
configurations of the connections (Fig. 25) for reinforcing plates and in the beam flanges at the
Steel Posttensioned
Steel Structures,
Fig. 25 Contour plots of
the equivalent plastic strain
(PEEQ) in the connection
at different drift levels:
SC-WHP1 (left) and
SC-WHP2 (right)
S
bearing interface. Figure 25f shows that damage Seismic Analysis of SC-MRFs Using PT
in the beam flanges is more evident in the Connections with WHPs
SC-WHP2 model.
The comparisons between the FEM analyses Modeling of PT Connection with WHPs for
and experiments show that the proposed FEM Nonlinear Analysis in OpenSees
model is capable of reproducing the inelastic A model for the PT connection with WHPs and
response of the tested PT connections up to the associated beams and columns was developed
the ultimate deformation levels and to capture in OpenSees (Mazzoni et al. 2006) as shown in
accurately all possible failure modes. Therefore, Fig. 26. The beams and columns were modeled
it is a reliable tool for the simulation of the hys- using the nonlinear force-based beam–column
teretic behavior of PT steel connections (Figs. 23 fiber element which can strictly satisfy equilib-
and 24). rium and accurately capture the distribution of
Fig. 26 Model for an exterior PT connection with WHPs and associated columns and beam
inelasticity along the depth of the section and of the compression stiffness equal to 20 times the
along the length of the physical member. For axial stiffness of the beam Kb was assigned to
the beam, two fiber elements with cross-sections these contact springs. Larger values for this stiff-
having different flange thickness were used to ness were found to produce practically the same
account for the beam flange-reinforcing plates. results but with higher computational cost, i.e.,
Each fiber was associated with uniaxial bilinear more iterations to achieve convergence and equi-
elastoplastic stress–strain behavior (Steel01 in librium in nonlinear analysis. To capture the hys-
OpenSees) with a post-yield stiffness ratio equal teretic energy dissipation capacity of the
to 0.01. connection, two zero-length hysteretic springs
Rigid elastic beam–column elements were were placed at the exact locations of WHPs
used to model the beam–column interface where along the depth of the beam web. These springs
gap opening and closing take place. To accurately were associated with a smooth hysteretic
capture the gap-opening mechanism in the Giuffre–Menegotto–Pinto model with isotropic
beam–column interface, three zero-length con- hardening (Steel 02 material in OpenSees). To
tact spring elements were placed at equal spaces account for panel zone shear deformations and
along the beam flange thickness. These contact possible yielding, the panel zone was modeled
springs were associated with an elastic using the Scissors model which introduces four
compression – no tension force–displacement additional rigid elastic beam–column elements
behavior (ENT material in Openness). A value and two nodes in the center of the panel zone
MRF
0.5
FOE 0.64% DBE 1.60% MCE 2.40%
V/W
0.25
1st beam yielding

1st base column yielding
0
0 1 2 3 4 5
q r (%)
SC-MRF
0.5
FOE 0.64% DBE 1.60% MCE 2.40%
1st decompression point

V/W
0.25
1st WHP yielding
1st base column yielding
1st beam yielding
0
0 1 2 3 4 5
q r (%)
Fig. 27 Base shear coefficient – roof drift behavior from nonlinear monotonic (pushover) static analysis
connected with two zero-length rotational was increased to ensure that the posttensioning
springs. These springs are associated with bilin- force in the PT bars will be equal to FPT,i after
ear elastoplastic hysteretic rules (Steel01 mate- the beam shortening.
rial in OpenSees) with properties calculated to The accuracy of the developed model in
reflect the contribution of the column web OpenSees was assessed by comparing the
(including doubler plates) and the column flanges force–deformation and moment–rotation
in the force–shear deformation panel zone behav- responses of the connection with the experimen-
ior. This simple panel zone model has been found tal results. Detailed comparisons can be found in
to produce identical results to those of the more (Dimopoulos et al. 2013), where it is shown that
computationally expensive Krawinkler panel the numerical model is in acceptable agreement
zone model (Charney and Downs 2004). PT with the experimental response and can be con- S
bars were modeled using a truss element running sidered as reliable.
parallel to beam centerline axis and connected to
the exterior nodes of the panel zones of the exte- Nonlinear Monotonic and Cyclic Static
rior columns of the SC-MRF. The truss element Analysis
has a cross-section area APT equal to that of both Figure 27 shows the base shear coefficient
PT bars. To account for posttensioning, an initial (V/W) – roof drift (yr) behavior of the conven-
strain equal to FPT,i/(APT EPT) was first tional MRF and the SC-MRF from nonlinear
assigned to the truss element where EPT is the monotonic (pushover) static analysis. V is the
modulus of elasticity of the PT bar material. base shear and W is the seismic weight.
However, posttensioning results in shortening of An inverted triangular force distribution along
the beams which in turn decreases the with roof displacement control was used in these
posttensioning force. To account for this analyses. The MRF and the SC-MRF have com-
decrease, the initial strain in the truss element parable base shear strengths and comparable
Steel Posttensioned MRF SC-MRF

Connections with Web 0.4 0.4
Steel Structures,
Fig. 28 Base shear
V/W
V/W
coefficient – roof drift 0 0
behavior from nonlinear
cyclic (push–pull) static
analysis
−0.4 −0.4
−3 0 3 −3 0 3
q r (%) q r (%)
initial stiffness. The pushover curves are plotted (push–pull) static analysis. The first cycle of
along with points associated with structural limit the analysis is performed up to the DBE drift
states and vertical lines corresponding to roof while the next cycle up to the MCE drift. The
drifts expected under the FOE, DBE, and MC- SC-MRF shows full re-centering capability
E. The structural limit states for the conventional under the DBE, adequate energy dissipation,
MRF are beam yielding and base column yield- and a small residual drift under the MCE due to
ing and occur at yr equal to 0.82 % and 0.92 %, plastic deformations at the column bases. The
respectively. The conventional MRF avoids conventional MRF shows large energy dissipa-
damage under the FOE but experiences signifi- tion capacity due to plastic deformations at the
cant damage under the DBE. The structural limit beam ends and at the column bases and the
states for the SC-MRF are decompression in a possibility of experiencing large residual drifts
PT connection, WHP yielding, base column under the DBE and MCE.
yielding, and beam yielding. Figure 27 shows
that the beams of the SC-MRF are damage-free Nonlinear Dynamic Time History Analyses
for yr equal or lower than 3 %, i.e., drifts higher
than the MCE. Damage in the SC-MRF is expe- Ground Motions
rienced at the column bases that yield at yr equal A set of 20 earthquake ground motions recorded
to 0.97 %. No PT bar yielding is observed. The on ground type B were used in 2D nonlinear
first decompression occurs at yr equal to 0.4 % dynamic time history analyses to evaluate the
while WHPs yield at yr equal to 0.62 % which is performance of the SC-MRF and the perfor-
almost equal to the FOE drift. Decompression mance of the conventional MRF. None of the
does not involve damage, while yielding of the ground motions exhibit near-fault forward direc-
WHPs is acceptable under low drifts since tivity effects. The ground motions were scaled to
WHPs can be easily replaced without bolting or the DBE level using the scaling procedure of
welding. The conventional MRF experiences Somerville (1997). Table 4 provides the scale
softening at yr equal to 1.25 %, while the factors and information on the 20 earthquake
SC-MRF shows a more gradual softening behav- ground motions.
ior. In particular, the SC-MRF shows softening
due to decompression in the PT connections at Modeling for Nonlinear Dynamic Analysis
low drifts and further softening due to plastic Two-dimensional nonlinear analytical models
deformations at the column bases and yielding of the conventional MRF and the SC-MRF
of a large number of WHPs at yr equal to 1 %. were developed for nonlinear dynamic analyses
Figure 28 shows the V/W – yr behavior of the in OpenSees. Nonlinear beam–column fiber
MRF and the SC-MRF from nonlinear cyclic elements were used for the beams and columns
Table 4 Properties of the ground motions used for nonlinear dynamic analyses
Magnitude Distance Scale factor

Earthquake Station Component (Mw) (km) FOE DBE MCE
Imperial Valley Cerro Prieto H-CPE237 6.53 15.19 0.82 2.05 3.08
1979
Loma Prieta 1989 Hollister – S & P HSP000 6.93 27.67 0.29 0.72 1.08
Loma Prieta 1989 Woodside WDS000 6.93 33.87 1.40 3.49 5.24
Loma Prieta 1989 WAHO WAH090 6.93 17.47 0.48 1.20 1.80
Manjil 1990 Abbar ABBAR-T 7.37 12.56 0.28 0.70 1.05
Cape Mendocino Fortuna – Fortuna FOR000 7.01 15.97 0.99 2.47 3.71
1992 Blvd
Cape Mendocino Rio Del RIO360 7.01 14.33 0.50 1.25 1.88
1992 Overpass – FF
Landers 1992 Desert – Hot LD- 7.30 21.78 0.95 2.37 3.56
Springs DSP000
Northridge 1994 LA – W 15th St W15090 6.69 25.60 1.14 2.86 4.29
Northridge 1994 Moorpark – Fire Sta MRP180 6.69 16.92 0.78 1.94 2.91
Northridge 1994 N Hollywood – Cw CWC270 6.69 7.89 0.53 1.33 2.00
Northridge 1994 Santa Susana 5108-360 6.69 1.69 0.78 1.95 2.93
Ground
Northridge 1994 LA – Brentwood 0638-285 6.69 12.92 0.85 2.12 3.18
VA
Northridge 1994 LA – Wadsworth 5082-235 6.69 14.55 0.62 1.54 2.31
VA
Kobe 1995 Nishi-Akashi NIS090 6.90 7.08 0.48 1.19 1.79
Kobe 1995 Abeno ABN090 6.90 24.85 1.00 2.49 3.74
ChiChi 1999 TCU105 TCU105-E 7.62 17.18 0.96 2.39 3.59
ChiChi 1999 CHY029 CHY029-N 7.62 10.97 0.53 1.32 1.98
ChiChi 1999 CHY029 CHY041-N 7.62 19.83 0.56 1.40 2.10
Hector 1999 Hector HEC090 7.13 10.35 0.42 1.04 1.56
and the Scissors model (Charney and Downs Seismic Response Results
2004) for the panel zones of the conventional Figure 29 compares the roof drift time histories of
MRF and the SC-MRF. A diaphragm constraint the conventional MRF and the SC-MRF under
is imposed on each floor level of the conven- the 5082-235 ground motion scaled to the DBE S
tional MRF. Stiff truss elements were used to and MCE. Near the end of the time histories, the
connect the internal nodes of each bay of the SC-MRF oscillates around the origin, indicating
SC-MRF to allow beam shortening due to negligible residual drift, while the conventional
posttensioning, gap opening in PT connections, MRF experiences residual drifts. The peak roof
and to consider the use of a discontinuous slab displacements of the MRF and the SC-MRF are
proposed by Chou et al. (2009). The model of similar. Figure 30 shows the stress–strain hyster-
Fig. 26 was used to model the PT connections esis at points A and B (extreme column base
and the associated beams and columns of the flange fibers), and C and D (extreme beam flange
SC-MRF. Each dynamic analysis was extended fibers) of the first story of the conventional MRF
well beyond the actual earthquake time (the under the 5082-235 ground motion scaled to the
ground motions were padded with zeros) to DBE and MCE. Figure 31 presents similar infor-
allow for damped free vibration decay and accu- mation for the SC-MRF. The stress–strain hys-
rate residual drifts calculation. teresis immediately after the end of the beam
Steel Posttensioned DBE MCE

Connections with Web 3 3
Steel Structures, 1.5 1.5
Fig. 29 Comparison of the
q r (%)
q r (%)
roof drift time histories
0 0
under the 5082-235 ground
motion scaled to the DBE
and MCE −1.5 −1.5
SC-MRF
MRF
−3 −3
0 50 100 0 50 100
time (s.) time (s.)
A: DBE B: DBE C: DBE D: DBE

400 400 400 400
stress (Mpa)
stress (Mpa)
stress (Mpa)
stress (Mpa)
0 0 0 0
1st storey of MRF
−400 −400 −400 −400 C

−4 0 4 −4 0 4 −4 0 4 −4 0 4
strain (%) strain (%) strain (%) strain (%) D
A: MCE B: MCE C: MCE D: MCE

400 400 400 400 A B
stress (Mpa)
stress (Mpa)
stress (Mpa)
stress (Mpa)
0 0 0 0
−400 −400 −400 −400

−4 0 4 −4 0 4 −4 0 4 −4 0 4
strain (%) strain (%) strain (%) strain (%)
Steel Posttensioned Connections with Web Hour- and D of the conventional MRF under the 5082-235
glass Pins: Toward Earthquake Resilient Steel Struc- ground motion scaled at the DBE and MCE
tures, Fig. 30 Stress–strain hysteresis at points A, B, C,
A: DBE B: DBE C: DBE D: DBE E: DBE F: DBE

400 400 400 400 400 400
stress (MPa)
stress (MPa)
stress (MPa)
stress (MPa)
stress (MPa)
stress (MPa)
0 0 0 0 0 0
1st storey of SC-MRF
−400 −400 −400 −400 −400 −400 C E

−4 0 4 −4 0 4 −4 0 4 −4 0 4 −4 0 4 −4 0 4
strain (%) strain (%) strain (%) strain (%) strain (%) strain (%)
D F
A: MCE B: MCE C: MCE D: MCE E: MCE F: MCE
400 400 400 400 400 400
A B
stress (MPa)
stress (MPa)
stress (MPa)
stress (MPa)
stress (MPa)
stress (MPa)
0 0 0 0 0 0
−400 −400 −400 −400 −400 −400

−4 0 4 −4 0 4 −4 0 4 −4 0 4 −4 0 4 −4 0 4
strain (%) strain (%) strain (%) strain (%) strain (%) strain (%)
Steel Posttensioned Connections with Web Hour- D, E, and F of the SC-MRF under the 5082-235 ground
glass Pins: Toward Earthquake Resilient Steel Struc- motion scaled at the DBE and MCE
tures, Fig. 31 Stress–strain hysteresis at points A, B, C,
MRF - FOE MRF - DBE MRF - MCE

6 6 6
5 mean 5 5
4 median 4 4
Storey
Storey
Storey
mean & sd
3 3 3
2 2 2
1 1 1
0 0 0
0 2 4 0 2 4 0 2 4
qs-max (%) qs-max (%) qs-max (%)
SC-MRF - FOE SC-MRF - DBE SC-MRF - MCE

6 6 6
5 5 5
4 4 4
Storey
Storey
Storey
3 3 3
2 2 2
1 1 1
0 0 0
0 2 4 0 2 4 0 2 4
qs-max (%) qs-max (%) qs-max (%)
Steel Posttensioned Connections with Web Hour- conventional MRF and the SC-MRF under 20 earthquake
glass Pins: Toward Earthquake Resilient Steel Struc- ground motions scaled to the FOE, DBE, and MCE
tures, Fig. 32 Statistics of peak story drifts of the
flange-reinforcing plate (points E and F) of the uniform height-wise distribution for the conven-
SC-MRF is also shown. The SC-MRF eliminates tional MRF and large dispersion compared to
beam plastic deformations under both the DBE that of ys-max. The largest ys-res of the conven-
and MCE, while the conventional MRF experi- tional MRF occurs in the first story with mean
ences appreciable beam damage. Figures 28 and values equal to 0.1 % under the DBE and 0.3 %
29 show that the column bases of the SC-MRF under the MCE. The associated m + s ys-res
experience larger plastic deformations than those values are equal to 0.25 % under the DBE and
of the column bases of the conventional MRF. 0.6 % under the MCE. The latter ys-res values
Figure 32 shows m, m + s, and median ys-max indicate that repair of damage in the conven-
values under the earthquake ground motions of tional MRF would be costly and disruptive
Table 4 scaled to the FOE, DBE, and MCE. The after the DBE and not financially viable after
m, m + s and median height-wise ys-max distribu- the MCE (Mc Cormick et al. 2008). These S
tions show identical shapes. The MRF has the results highlight the need for Eurocode 8 to
largest ys-max in the fourth story with m values include residual deformations as an additional
equal to 0.75 % under the FOE, 1.65 % under the seismic performance parameter. The SC-MRF
DBE, and 2.2 % under the MCE, i.e., close to the practically eliminates residual story drifts apart
design values of 0.64 % under the FOE and 1.6 % from the first story that has m and m + s ys-res
under the DBE and smaller than the design value values equal to 0.1 % and 0.15 % under both the
of 2.4 % under the MCE. The SC-MRF has the DBE and MCE. The latter ys-res values are lower
largest ys-max in the fourth story with mean values than the global sway imperfections defined in
equal to 0.75 % under the FOE, 1.8 % under the EC3 (2003) and so it can be assumed that there
DBE, and 2.5 % under the MCE, i.e., slightly will be no need for these residual drifts to
larger than the DBE and MCE design ones. be straightened out. Figure 33 shows small m +
Figure 33 shows m, m + s, and median values of s ys-res values in the third story of the SC-MRF
the residual story drifts, ys-res. ys-res values show a due to modest yielding in the beam ends.
MRF - FOE MRF - DBE MRF - MCE

6 6 6
5 mean 5 5
4 median 4 4
Storey
Storey
Storey
mean & sd
3 3 3
2 2 2
1 1 1
0 0 0
0 0.5 1 0 0.5 1 0 0.5 1
qs-res (%) qs-res (%) qs-res (%)
SC-MRF - FOE SC-MRF - DBE SC-MRF - MCE

6 6 6
5 5 5
4 4 4
Storey
Storey
Storey
3 3 3
2 2 2
1 1 1
0 0 0
0 0.5 1 0 0.5 1 0 0.5 1
qs-res (%) qs-res (%) qs-res (%)
Steel Posttensioned Connections with Web Hour- conventional MRF and the SC-MRF under 20 earthquake
glass Pins: Toward Earthquake Resilient Steel Struc- ground motions scaled to the FOE, DBE, and MCE
tures, Fig. 33 Statistics of residual story drifts of the
Conclusions damaged WHPs. These tests showed that

WHPs can be easily replaced without
A new steel posttensioned connection was devel- welding or bolting, and hence, the pro-
oped for use in self-centering moment-resisting posed connection can be repaired with
frames. The connection performance was vali- minimal disturbance to building use or
dated through a large-scale experimental pro- occupation in the aftermath of a major
gram and extensive numerical analyses, earthquake.
including seismic analysis of steel frames • Web local buckling at the connection
equipped with the proposed connection. The region after the beam flange-reinforcing
main conclusions of this research are: plates can be avoided by using web stiff-
eners. However, this detailing results in
(a) Based on the experimental results: excessive local yielding at the
• The proposed connection has stable self- beam–column interface due to high bear-
centering behavior, enhanced energy dis- ing forces.
sipation capacity, and strength and stiff- • To accommodate the gravity loads of the
ness comparable to those of a welded frame and to facilitate the replacement
connection. procedure of the damaged WHPs, it is
• The proposed connection eliminates resid- necessary to include a vertical loading
ual deformations and avoids beam damage transfer system, e.g., a shear tab with slot-
for drifts lower or equal to 6 %. ted holes or a seat angle below the beam.
• A simplified analytical procedure using • It should be emphasized that the tests
plastic analysis and simple mechanics described herein were performed on a
was found to accurately predict the con- 300 mm deep beam section. For a given
nection behavior. rotation, the demand on the WHPs and PT
• Repeatable tests on a connection specimen bars is expected to increase for deeper
were conducted along with replacing sections. The response of the proposed
connection should be verified by using apart from the first story which sustains
deeper beam sections. small residual drifts due to plastic defor-
(b) Based on the finite element simulations: mations at the column bases. The mean
• The developed nonlinear FEM models can plus one standard deviation value of the
be reliably used to assess the design of the first story residual drift of the SC-MRF is
proposed connection as they are capable equal to 0.15 % under both the DBE and
to trace the hysteretic behavior and predict MCE which is considered small and does
the local failure modes both of the indi- not need to be straightened out.
vidual WHPs and the connection when • Seismic analyses show that the mean plus
subjected to either monotonic or cyclic one standard deviation value of the maxi-
loading. mum residual story drift of the conven-
• The FEM analyses confirm that for drifts tional MRF is 0.25 % under the DBE and
equal or lower than those expected under 0.6 % under the MCE. These values indi-
the design basis earthquake (DBE), dam- cate that repair of damage in the conven-
age in a carefully designed connection is tional MRF would be costly and disruptive
concentrated in the WHPs which are com- after the DBE and not financially viable
ponents that can be very easily replaced after the MCE.
without welding or bolting. • Seismic analyses show that the beams of
(c) Based on the seismic analyses of steel frames the SC-MRF do not exhibit any yielding
with the proposed connection: even under the MCE, while significant
• The proposed model for the PT connection inelastic deformations are developed in
with WHPs and the associated beams and the beams of the MRF under both the
columns has been calibrated against DBE and MCE. On the other hand, the
experimental results and found to accu- column bases of the SC-MRF experience
rately simulate the hysteretic behavior of larger inelastic deformations than those of
the PT connection. the conventional MRF.
• Nonlinear static monotonic (pushover)
analysis shows that the conventional
MRF and the SC-MRF have comparable Summary
base shear strength and initial stiffness.
The conventional MRF experiences sig- This contribution presents the experimental and
nificant damage in beams at the DBE analytical validation on a new self-centering steel
drift. On the other hand, the SC-MRF has posttensioned connection using web hourglass
damage-free beams for drifts even higher shape steel pins (WHPs) as energy dissipation
than the MCE drift. devices. The connection isolates inelastic defor- S
• Nonlinear static cyclic (push–pull) analy- mations in WHPs, avoids damage in other con-
sis shows that the SC-MRF has full re- nection parts as well as in beams and columns,
centering capability and adequate energy and eliminates residual drifts. WHPs do not inter-
dissipation capacity under the DBE. fere with the composite slab and can be very
• Seismic analyses show that the conven- easily replaced without bolting or welding, and
tional MRF and the SC-MRF have com- so, the connection enables nondisruptive repair
parable peak story drifts. In particular, the and rapid return to building occupancy in the
conventional MRF has slightly lower peak aftermath of a strong earthquake. The experimen-
story drifts than the SC-MRF. For both tal results are first presented followed by para-
frames through the mean peak story drifts metric studies using nonlinear finite element
are close to the design values. models. A simplified nonlinear model for the
• Seismic analyses show that the SC-MRF connection and the associated beams and col-
practically eliminates residual story drifts umns that consist of nonlinear beam–column
elements and hysteretic and contact zero-length Dimopoulos AI, Karavasilis TL, Vasdravellis G, Uy
spring elements appropriately placed in the B (2013) Seismic design, modelling and assessment
of self-centering steel frames using post-tensioned
beam–column interface was also developed in connections with web hourglass shape pins. Bull
OpenSees. The model was calibrated against Earthq Eng 11(5):1797–1816. doi:10.1007/s10518-
experimental results and found to accurately sim- 013-9437-4
ulate the connection behavior. A prototype build- EC3 Eurocode 3 (2003) Design of steel structures
EC8 Eurocode 8 (2004) Design of structures for earth-
ing was selected and designed as a conventional quake resistance
steel moment-resisting frame (MRF) according to Garlock M, Ricles JM, Sause R (2005) Experimental
Eurocode 8 or as a self-centering steel MRF studies of full-scale posttensioned steel connections.
(SC-MRF) using the connection with WHPs. J Struct Eng 131(3):438–448
Garlock M, Sause R, Ricles JM (2007) Behavior and
Seismic analyses results show that the conven- design of posttensioned steel frame systems. J Struct
tional MRF and the SC-MRF have comparable Eng 133(3):389–399
peak story drifts and highlight the inherent poten- Karavasilis TL, Ricles JM, Sause R, Chen C (2011) Exper-
tial of the SC-MRF to eliminate damage in beams imental evaluation of the seismic performance of steel
MRFs with compressed elastomer dampers using
and residual drifts. It is shown that repair of large-scale real-time hybrid simulation. Eng Struct
damage in the conventional MRF will be costly 33(6):1859–1869
and disruptive after the design basis earthquake Karavasilis TL, Kerawala S, Hale E (2012) Model for
and not financially viable after the maximum hysteretic behaviour of steel energy dissipation
devices and evaluation of a minimal-damage seismic
considered earthquake due to large residual drifts. design approach for steel frames. J Constr Steel Res
70:358–367
Kim HJ, Christopoulos C (2008a) Friction damped
Cross-References posttensioned self-centering steel moment-resisting
frames. J Struct Eng 134(11):1768–1779
Kim HJ, Christopoulos C (2008b) Seismic design proce-
▶ Seismic Analysis of Steel Buildings: Numeri- dure and seismic response of post-tensioned self-
cal Modeling centering steel frames. Earthq Eng Struct Dyn
▶ Seismic Collapse Assessment 38(3):355–376
▶ Seismic Strengthening Strategies for Existing Kim H, Christopoulos C (2009) Numerical models and
ductile ultimate deformation response of
(Code-Deficient) Ordinary Structures post-tensioned self-centering moment connections.
Earthq Eng Struct Dyn 38:1–21. doi:10.1002/eqe.
Kobori T, Miura Y, Fukuzawa E, Yamada T, Arita T,
References Takenaka Y, Miyagawa N, Tanaka N, Fukumoto
T (1992) Development and application of hysteresis
Charney FA, Downs WM (2004) Connections in steel steel dampers. In: Earthquake engineering, tenth world
structures V. ESSC/AISC workshop, Amsterdam, conference. Balkema, Rotterdam
3–4 June Mazzoni S, McKenna F, Scott M, Fenves G (2006) Open
Chou CC, Lai YJ (2009) Post-tensioned self-centering system for earthquake engineering simulation
moment connections with beam bottom flange energy (OpenSees). User Command Language Manual,
dissipators. J Constr Steel Res 65(10–11):1931–1941 Pacific Earthquake Engineering Research Center, Uni-
Chou CC, Chen JH, Chen YC, Tsai KC (2006) Evaluating versity of California, Berkeley
performance of post-tensioned steel connections with Mc Cormick J, Aburano H, Ikenaga M, Nakashima
strands and reduced flange plates. Earthq Eng Struct M (2008) Permissible residual deformation levels for
Dyn 35(9):1167–1185 building structures considering both safety and human
Chou C, Tsai K, Yang W (2009) Self-centering steel elements. 14th WCEE, Beijing
connections with steel bars and a discontinuous com- Ricles J, Sause R, Garlock M, Zhao C (2001)
posite slab. Earthq Eng Struct Dyn 38:403–422. Posttensioned seismic-resistant connections for steel
doi:10.1002/eqe. frames. J Struct Eng 127(2):113–121
Christopoulos C, Filiatrault A, Uang CM, Folz B (2002) Ricles J, Sause R, Peng SW, Lu LW (2002) Experimental
Posttensioned energy dissipating connections for evaluation of earthquake resistant posttensioned steel
moment-resisting steel frames. J Struct Eng connections. J Struct Eng 128(7):850–859
128(9):1111–1120 Rojas P, Ricles JM, Sause R (2004) Seismic performance
Dassault Systems (2010) Abaqus theory manual. Dassault of post-tensioned steel moment resisting frames with
Systèmes Simulia, Providence friction devices. J Struct Eng 131(4):529–540
Steel Structures 3417
Somerville P (1997) Development of ground motion time and aluminum. Cast iron, which was first
histories for phase 2 of the FEMA/SAC steel project, manufactured as early as fifth century BC, is
Report No. SAC/DB-97/04, Sacramento
Tsai KC, Chou CC, Lin CL, Chen PC, Jhang SJ typically considered to be too brittle and difficult
(2008) Seismic self-centering steel beam-to-column to join but received widespread application in
moment connections using bolted friction devices. construction through the late eighteenth century.
Earthq Eng Struct Dyn 37:627–645 The use of cast iron is uncommon in today’s
Vasdravellis G, Karavasilis TL, Uy B (2013a) Large-scale
experimental validation of steel post-tensioned con- construction industry, but some ductile cast
nections with web hourglass pins. J Struct Eng irons continue to be used in specialized applica-
139(6):1033–1042 tions such as water pipes. The use of cast iron was
Vasdravellis G, Karavasilis TL, Uy B (2013b) Finite ele- rapidly replaced by wrought iron, which was
ment models and cyclic behavior of self-centering
steel post-tensioned connections with web hourglass easier to work with, leading to the construction
pins. Eng Struct 52:1–16 of the first large metal structure, the Iron Bridge
Vasdravellis G, Karavasilis TL, Uy B (2014) Design rules, at Coalbrookdale by A. Darby in 1780 (www.
experimental evaluation and fracture models for high- greatbuildings.com/buildings/Iron_Bridge_at_
strength and stainless-steel hourglass shape energy
dissipation devices. J Struct Eng (ASCE). Coalbrookdale.html). Extensive use of iron
doi:10.1061/(ASCE)ST.1943-541X.00010 continued through the first half of the nine-
Whittle J, Williams MS, Karavasilis TL, Blakeborough teenth century, culminating in iconic structures
T (2012) A comparison of viscous damper placement such as the Crystal Palace in London (http://en.
methods for improving seismic building design.
J Earthq Eng 16:540–560 wikipedia.org/wiki/The_Crystal_Palace).
Wolski M, Ricles JM, Sause R (2009) Experimental study Wrought iron began to be replaced by steel in
of a self-centering beam-column connection with bot- civil structures when the latter became commer-
tom flange friction device. J Struct Eng 135(5): cially available in large quantities in the late
479–488
nineteenth century AD through the development
of the Siemens and Bessemer processes and the
development of rolling technologies (Bodsworth
2001; Ashton 1968; Halmos 2000). Initially steel
Steel Structures was rolled into flat plates and small sections
(angles and channels or L- and C-sections) that
Roberto Leon were assembled into larger structural sections
The Charles Edward Via, Jr. Department of Civil through riveting (Fig. 1), just as it had been
and Environmental Engineering, Virginia Tech, done for wrought iron. Riveted, or built-up sec-
Blacksburg, VA, USA tions, were used through the first half of the
twentieth century.
Steel rolling, which began to produce large
Synonyms sections at the turn of the twentieth century S
(Fig. 2), led to the development of optimal W-
Metal structures; Steel connections; Steel (or H-) sections for carrying flexure and axial
construction loads. Advanced fabrication techniques,
recycling, and improvements in energy consump-
tion during production have resulted in extremely
Introduction efficient and sustainable construction practices in
the steel industry in the twenty-first century.
The use of metals as the principal construction Much of today’s structural steel is manufactured
material for modern buildings and bridges dates in minimills that utilize scrap steel as their source
back to the beginning of the Industrial Revolution materials; some estimates are that up to 95 % of
(Bodsworth 2001; Ashton 1968). Three metallic steel is recycled in the USA.
materials have been commonly used in the con- The use of aluminum, which became common
struction industry: cast and wrought iron, steel, in the 1940s when the need to build light
3418 Steel Structures
Steel Structures,
Fig. 1 Riveted built-up
steel sections in a bridge
truss (thebridgehunter.
areavoices.com,
# RTLeon)
Steel Structures, Fig. 2 Rolling of deep sections (75 and 100 cm) in 1911. Note that flanges are not as wide as modern
sections (# J.C. Gerardy, ArcelorMittal)
Steel Structures,
Fig. 3 Modern rolling of
large steel wide-flange
section showing selective
cooling (darker area) to
improve metal performance
(# J.C. Gerardy,
ArcelorMittal)
Steel Structures, Fig. 4 Carefully executed weld (left) capacity as the weld does not fracture even if bent at
showing numerous passes and good workmanship; bend 180 (# J.C. Gerardy, ArcelorMittal)
test from this weld showing excellent deformation
airplanes made its strength-to-weight ratio very material to withstand the large cyclic loads
attractive, is also limited in today’s construction imposed by earthquakes.
industry due to cost and welding issues. Its use is The use of steel in multistory framed buildings S
limited mostly to facades, finishes, and began in the 1880s in Chicago, notably with the
nonstructural elements. Home Insurance Building (Condit 1968) and sim-
ilar skyscraper buildings. The use of steel frames
was still in its maturing stage when the excellent
Steel for Seismic Applications behavior of steel structures, when subjected to
large ground motions, became evident after the
A large number of improvements through better San Francisco earthquake and fire of 1906
metallurgy (heat treatments (Tylecote 1976) and (Kurzman 2001). These events highlighted the
the development of alloy steels with large defor- lateral resistance (strength) and deformation
mation capacity, Fig. 3) and joinery (welding and capacity (ductility) of steel structures (Ham-
high-strength bolts, Fig. 4) in the twentieth cen- burger and Meyer 2006), but also their suscepti-
tury have made steel the preferred construction bility to fire unless suitably protected (Fig. 5).
Steel Structures, Fig. 5 Damaged buildings from the The structure of the burned building in the center of the
1906 San Francisco earthquake and fire. The two tall steel photograph was undamaged (en.wikipedia.org)
frames were under construction and were not damaged.
Most structural members were made up of angles or plates. Initially, the encasement was
riveted plates and connections for larger struc- made from masonry rubble inside masonry
tures (Fig. 6) and riveted smaller rolled sections facades, but this was later replaced by encase-
for buildings (Fig. 7). Structures that contained ment with lightly reinforced concrete (Fig. 7).
steel columns and beams encased in masonry or Today, the synergistic use of steel in tension
concrete survived the earthquake and fire rela- and concrete in compression has led to the devel-
tively unscathed (Fig. 5). The superior perfor- opment of composite steel–concrete construction
mance of steel frames was enshrined in the particularly for high-rise construction (Viest et al.
report by ASCE (1906), which stated: The well- 1997).
designed steel frame offers the best solution of the Three important developments have occurred
question of an earthquake proof building, as all in the steel construction industry since the 1950s.
the stresses can be cared for. The first development is the change of joining
This and similar observations by many others methods, as rivets have been replaced by both
led to the widespread utilization of such systems high-strength bolts and welds. The use of indus-
in the earthquake-prone areas of the Western trialized welding, which was developed in the
USA in the first half of the twentieth century. early twentieth century for the shipping industry,
These structures consisted of rolled steel sections led to much simpler, stronger, and stiffer connec-
joined by connections made by riveted steel tions (Fig. 8). These connections were considered
Steel Structures,
Fig. 6 Typical built-up
riveted members and
connection for an elevated
railway structure
(Engineering News, 1914)
Steel Structures,
Fig. 7 Riveted building
beam-to-column
connection before
encasement (# RTLeon)
S
for many years to be the most reliable structural connection failures have been observed after
system in seismic areas. Failures in welded con- earthquakes, except for the case of under-
nections observed after the 1994 Northridge designed brace connections. Welded connections
(Fig. 9) and 1995 Kobe earthquakes (FEMA are still preferred by many designers because of
355E 2000a) have led to radical changes in the their perceived superior strength and stiffness
specifications for such joints. While both joining characteristics and ease of design when compared
methods can be used in seismic construction, the to bolted connections.
use of welding requires extensive planning, The second development is that most floors in
inspection, and QA/QC procedures to ensure steel structures have become composite ones,
desirable ductile performance (FEMA 355B with the use of shear studs to connect floor
2000b). In the aftermath of those failures, design beams and girders to concrete slab cast in metal
requirements for bolted connections have also deck (Viest et al. 1997). The use of the metal
been increased even though few bolted deck as both formwork and reinforcement, the
Steel Structures,
Fig. 8 Modern welded
connection to the beam
flanges; bolted connection
to the beam web is used for
erection (FEMA 355E,
2000)
Steel Structures,
Fig. 9 Laboratory
simulation of connection
weld brittle failure as
observed in the 1994
Northridge earthquake
(The photo shows the area
circled in red in Fig. 7
(# RTLeon))
ease of welding the shear studs that act as con- The third important change has been the
nectors between the steel beam and the concrete development of numerous new structural systems
floor, and the tremendous increases in strength and proprietary connections that have given
and stiffness have made composite floor con- designers a very wide choice of technologies for
struction a most economical system (Fig. 10). use in seismic design. The development of such
Floors of this type provide sufficient in-plane systems has been made possible by both exten-
stiffness, such that all parts of the lateral force- sive experimental testing and advanced simula-
resisting system can be assumed to work tion tools brought forth by the use of computers
together (rigid floor diaphragm assumption), so and improvements in analytical methods. Some
that all lateral load-resisting systems can act of these systems are discussed under Structural
concurrently. This characteristic is important as Systems later in this article. Striking evidence of
many frames in the USA utilize lateral load the advantages of these new systems can be seen
systems engaging only partially the perimeter in the excellent performance of steel eccentri-
frames. cally braced frames (EBF) in the recent
Steel Structures,
Fig. 10 Typical composite
CONCRETE
floor system (# RTLeon)
WELDED WIRE
FABRIC
SHEAR CONNECTORS
COMPOSITE
STEEL DECK
Steel Structures,
Fig. 11 Ductility of steel
as shown by the >25 %
elongation of the original
coupon (bottom) to the
fractured one (top)
(# RTLeon)
Christchurch earthquakes (Clifton et al. 2011) Steel is an ideal material for seismic design
as compared to that of modern reinforced special because both of its isotropic and homogeneous
moment-resisting reinforced concrete frame properties and its ability to undergo large local
(SMRF-CR) and shear wall structures (Leon et plastic deformations before failure. Figure 11
al. 2014). shows a typical mild steel test coupon elongating S
in excess of 15 % over a 2 in. length and more than
50 % locally. The stress–strain (or load-
Steel Properties deformation) behavior for this type of steel is
shown by the red line in Fig. 12. After an initial
For structural design purposes, steel is an iron- stiff elastic response, the steel reaches its yield
carbon alloy that contains many additional point (knee in the curve), the steel deforms plasti-
alloying elements such as Mn, Mo, Cr, Ni, and cally (i.e., undergoing large deformations with
V. A large variety of mechanical properties can little small increments of load, shown by the flat
be achieved with steels by either varying its portion of the curve), and picks up additional resis-
alloys (primarily carbon which constitutes tance as the deformation becomes large (last part
0.2–0.6 % of steel) and/or subjecting the material of the curve, termed the strain-hardening region).
to different heat treatments that change its grain The curve for a steel coupon subjected to large
microstructure. deformation reversals such as those experienced
Steel Structures,
Fig. 12 Cyclic
stress–strain behavior of a
mild structural steel
(# Alan Pense, Lehigh
University)
locally during a large seismic event is shown by Other important engineering properties
the rest of the curves in Fig. 12. If loaded cycli- needed for seismic design are notch toughness
cally, the steel will show fat, stable, and hardening and weldability. Notch toughness refers to the
hysteresis (stress–strain) loops, evidence of good ability of the material to resist fracture propaga-
energy dissipation capacity. Failure by ductile tion from an existing defect when subjected to
fracture will ideally occur after many cycles of dynamic loads. Fracture propagation is resisted
deformation. For seismic design, steel will per- through local plastic deformations and is primar-
form best when deformed in shear. ily a function of temperature and heat treatment.
Steels for structural use are classified as car- Weldability is primarily a matter of obtaining a
bon steels, high-strength low-alloy steels, and structural joint free of any undesirable defects by
alloy steels. For design purposes, these steels utilizing appropriate electrode materials and
can be assumed to have a density of 7.85 g/cm3, weld procedure specifications (WPS).
a modulus of elasticity of 210 GPa, and a
Poisson’s ratio of 0.3. Carbon steels are classified
based on the percentage of carbon. Mild carbon Structural Systems
steels (0.15–0.29 % C) with yield points in the
range of 220–250 MPa and tensile strengths of As noted earlier, one of the great developments in
400–500 MPa are the most common structural seismic construction has been the promotion of a
carbon steels. Typically, an increase in carbon large variety of steel structural systems. These
percent raises the yield point and increases hard- systems are classified as special, intermediate,
ness, but reduces ductility and makes welding and ordinary based on the amount of seismic
more difficult. These drawbacks can be mini- detailing present. A special system will require
mized by heat treatments. great care in design and detailing to obtain the
The stress–strain behavior of a mild carbon large deformation capacity needed to activate the
steel is characterized by a sharp yield point and plastic deformations and energy dissipation char-
large ultimate strains (percent elongation >20 %). acteristic of these systems. The benefit of a spe-
High-strength low-alloy steels exhibit higher yield cial system is lower design lateral forces, which
points (345–480 MPa) and somewhat larger tensile results in savings in both material and construc-
strengths (450–600 MPa) without an appreciable tion costs.
drop in ductility. Alloy steels are heat-treated At its most basic, lateral force-resisting sys-
steels (mostly quenched and tempered) with yield tems can be visualized as a continuum, with a
points of 550–760 MPa, but lower tensile to yield pure frame system (Fig. 13a) at one extreme and a
ratios and ductility than for other structural steels. wall system at the other end (Fig. 13e). In these
Steel Structures, Fig. 13 Deformation mechanisms for typical steel seismic force-resisting systems (# RTLeon)
figures, only the primary structural elements are be very ductile when properly designed, but pure
shown; secondary structural elements, such as frame structures tend to be rather flexible.
out-of-plane braces, floor slabs, chord, collectors, To eliminate or limit damage to nonstructural
and diaphragms, are not shown. Also not shown elements and contents under both small earth-
are nonstructural elements, such as partition quakes and large wind loads, special moment-
walls, which are assumed not to contribute sig- resisting frames need to meet some maximum
nificantly to lateral resistance and should be deformation criteria (typically 3 % drift under
designed so as not to interact with the primary earthquake loads and 0.25 % under wind loads)
structural members. If not properly isolated, and thus are said to be drift controlled. It is clear
nonstructural elements can have deleterious that rigid and strong connections that limit the
effects on frame behavior. shear deformations of the panel zones and pro-
The moment-resisting frame (MRF), a combi- mote beam yielding are desirable (note thick end S
nation of slender beams and columns, was the plate and large bolts in the connection in Fig. 14),
first type of steel structural system used and as long as the local buckling of the beam flanges
which showed excellent seismic behavior and web can be delayed. In addition, mechanisms
(Fig. 5). In a pure moment-resisting frame that can reduce the strength demand at the welds
(MRF, Fig. 13a), the lateral deformations are to at the joint without substantially decreasing stiff-
be accommodated primarily by bending of the ness, such as reduced beam sections (RBS,
beams and columns, shear deformation of the Fig. 15), are desirable. In an RBS, the steel in
panel zone, and the formation of plastic hinges, the beam flanges is cut to intentionally weaken
or areas of concentrated plasticity, in the beams. the beam and promote the formation of a plastic
The yielding of the beam steel and the formation hinge away from the welded connection.
of a plastic hinge are shown by the flaking of the At the other extreme, one can think of a wall
whitewash in the critical section of the beam in structure composed of a steel frame infilled with a
Fig. 14. All of these deformation mechanisms can steel plate (Fig. 13e) as the analogue to a
Steel Structures,
Fig. 14 End-plate
connection showing plastic
hinging followed shortly by
local buckling of the
flanges (# RTLeon)
Steel Structures,
Fig. 15 End-plate
connection showing
reduced beam section
(RBS) (# RTLeon)
reinforced concrete shear wall system. For econ- post-buckling strength of metal plates. Steel-plated
omy reasons, in steel structures it is desirable to shear walls are relatively modern structures, with a
use a rather thin plate. When displaced laterally, few implementations as far back as the late 1970s,
the plate will buckle in compression along one but have become more accepted only in the last
diagonal, allowing a tension field to develop in decade or so.
the other diagonal (Fig. 16). This creates a large In between these two limits, one can visualize
tensile strut mechanism that transfers the horizon- a large number of alternatives. Starting at the
tal forces similarly to a braced frame (Fig. 13c). bottom of Fig. 13, one can visualize a system
This force transfer is exactly the inverse of a con- where the struts in the walls are replaced by
crete shear wall structure, where a compression discrete, strong, and rigid braces (or diagonal
strut will form, as the concrete is strong in com- elements, Fig. 17). This system is known as a
pression but weak in tension. The great advantage buckling-restrained brace frames (BRBF,
of a steel shear wall over a concrete one is that Fig. 13d), because the braces in compression are
whereas the behavior of a concrete shear wall will so stout that they will not buckle but yield. The
degrade rapidly due to an X-crack pattern that buckling in these systems is limited to an interior
develops with cycling, the steel shear wall perfor- core of the brace, with the exterior portion being
mance will not degrade substantially due to the disconnected and acting only to restrict the
Steel Structures,
Fig. 16 Thin plate under
lateral loads showing
tension-field action from
top right to bottom left and
buckling in the opposite
direction (# M. Kurata)
Steel Structures,
Fig. 17 Use of large BRBs
in a modern structure
(Berkeley Animal Shelter)
(# ci. berkeley.ca.usa)
buckling of the brace. Many of these buckling- frames (CBF) originated when engineers began S
restrained braces are proprietary devices. BRBF to stiffen MRF for wind loads in the 1910s using
systems originated in Japan in the early 1990s and light diagonals, as in the Sather Tower in Berke-
are a very common type of structural system in ley, CA (Fig. 18). Modern CBF are much stiffer
the Western USA today. than SMRF, but generally less ductile as the
The next system is similar, except that the buckling of the braces can lead to a concentra-
braces are not so stiff and they will buckle tion of deformations in a few floors. It is not clear
slightly in compression. This system is known when CBF began to be used in seismic systems
as a special concentrically braced frame (SCBF, (i.e., going from CBF to SCBF), but seismic
Fig. 13c). In this system, the brace in tension is design recommendations for braced frames
assumed to carry most of the force (generally began to appear in the 1960s. The original CBF
about 70 %) as the compression brace will had slender X-shaped braces in which only
buckle and be able to provide only its post- the brace in tension provided resistance.
buckling resistance. Concentrically braced There are a large number of variations of this
Basis of Design
The seismic design of steel structures is predi-

cated on having little or no structural damage
under a service-level earthquake, moderate but
rapidly repairable damage under a moderate
earthquake, and not to collapse and cause loss
of lives in the event of the design-level earth-
quake. Until recently, design specifications typi-
cally addressed only the design-level earthquake
through a series of prescriptive design provisions
intended to address ultimate strength design
(ULD) criteria (limit states). However, the costs
and time related to both loss of contents and
functionality under smaller events and the desire
to provide some freedom from prescriptive build-
ing codes have begun to drive the design process
toward a performance-based design (PBD)
approach. A pure PBD approach will be
completely non-prescriptive, giving the structural
engineer complete freedom in determining both
the limit states and how to comply with them. In
the short term (next 5–15 years), it is likely that a
hybrid model for seismic design with elements of
both ULS and PBD will emerge.
Whether a ULD or a PBD approach is used,
the basis for design of steel structures remains the
need to provide ductility, redundancy, robust-
Steel Structures, Fig. 18 Sather Tower under construc- ness, and resiliency. Ductility comes from a syn-
tion (1914) ergistic use of the very ductile behavior of steel at
the material level to create elements, connec-
tions, and structural systems capable of sustain-
system (Fig. 19), each with its own design ing large inelastic deformations without
requirements and behavior characteristics. appreciable stiffness or strength degradation.
Finally, one can think of beginning to move From the design standpoint, most of our ductility
the brace in each bay horizontally, such that a is ensured through a series of prescriptive
hybrid structure between a pure moment frame requirements that have been developed through
and a braced frame is achieved. This is known as observed performance in past earthquakes
an eccentrically braced frame (Fig. 13b). In these (Fig. 19), advanced analytical studies (Fig. 20),
structures, the central portion of the beam, known and complex laboratory testing (Figs. 14, 15,
as the link, will deform and yield in shear if the and 16).
link is short and in a combination of flexure and Redundancy is the ability of the structure to
shear if the link is longer. Eccentrically braced redistribute forces as inelastic action occurs in
frames have performed extremely well in recent order to efficiently activate all major lateral load
earthquakes (Fig. 20), with numerous such struc- resistance systems. In its simplest form, redun-
tures providing superior performance during the dancy can be defined as providing multiple load
2011 Christchurch earthquake (Clifton et al. paths in a structure. Redundancy is primarily a
2011). function of the number and ductility of the lateral
Steel Structures, Fig. 19 Different braced frame configurations (# RTLeon)
Steel Structures,
Fig. 20 Damaged
eccentrically brace frame,
showing yielding of the
shear link element
(# RTLeon)
load-resisting systems employed, the system’s occupants). If the structure does not collapse,
rational 3D configuration, the ductility of connec- regardless of its damaged state, it will be consid-
tions, the strength of diaphragms and collector ered robust. Clearly, a client can establish more
elements, and the detailing of the structure. stringent criteria for robustness in her/his struc-
Importantly, redundancy implies activating not ture if deemed necessary.
only the lateral load-resisting system but also Resiliency refers to the ability of the structure
the gravity load system and other nonstructural to fulfill its intended functions after a large event. S
elements that can contribute to the strength and For example, a level-four trauma hospital would
stiffness of the system. be resilient if it can continue to function
Robustness refers to the ability of the struc- uninterruptedly after the design-level earthquake.
tural system to limit any failure to a relatively While it is possible today to provide that level of
small part of the structure so that the conse- performance for the structural system, it is rather
quences of the failure are not disproportionate more difficult to do it for all the other systems
to the initial failure. For example, a common (from electricity, water, oxygen, and the myriad
way of determining the robustness of a moment of IT systems) present in such hospitals. For a
frame is to remove an interior column in a lower large store, it may mean being able to get back in
floor (something that may happen during an business within a few days and at a minimum of
earthquake) and assess the stability of the system cost. Resiliency can be quantified in many ways,
under gravity loads (to make sure that the struc- including the ability to resist ground motions
tures does not “pancake” and crush its significantly larger than the design one, the
costs to repair the structure, the time needed to As part of that effort, known as the SAC Project
carry out such repairs, and the indirect costs asso- (FEMA 355B 2000b), extensive experimental
ciated with business interruption. studies were conducted on bolted and welded
connections. The probability of brittle failure in
connections can be minimized by (1) proper
Connections attention to selecting appropriate base materials
and welding consumables with high toughness;
In order to satisfy ductility, redundancy, robust- (2) careful consideration of surface preparation,
ness, and resiliency criteria, steel structures in preheating, and welding sequences; (3) inspec-
seismic areas need to be designed primarily to tion; and (4) nondestructive testing. This amounts
avoid any type of brittle failure modes, which can to the need for a very comprehensive quality
result in sudden losses of local or global strength, assurance/quality control (QA/QC) plan through-
stiffness, and stability. In general, it is possible to out all phases of the construction projects
design and fabricate steel members (beams, col- (materials, fabrication, and erection). In the
umns, braces, and walls) to avoid these type of USA, these requirements are now embodied in
failures through close attention to proper material the required weld procedure specifications
selection, using compact sections (i.e., keeping a (WPS).
small enough ratio of the width of an element to Connections with bolts are usually less sus-
its length.) and providing lateral bracing to pre- ceptible to material-type problems, but require
vent out-of-plane buckling. The potential weak more care in the fabrication as slight misalign-
links in steel structures are the connections, ments and exceeding tolerances can lead to
which must transfer very large forces through severe problems during erection. In addition, the
complex force paths. The design of connections force transfer mechanisms in connections with
for seismic loads is a complex design issue as it is bolts are less obvious and require more detailed
difficult to predict failure modes and their computations than those with welds. Finally, in
interactions. general, bolted connections will be less rigid than
From the standpoint of steel as a material, welded ones, requiring that this semirigid behav-
brittle failures can occur because of a combina- ior be included in the analyses, substantially com-
tion of poor selection of or unmatched materials, plicating the design process. Bolted connections,
large stress concentrations due to poor design or such as the end plates shown in Figs. 14 and 15,
execution, and large triaxial state of stresses. when properly detailed, can provide excellent
These conditions can often arise in connections. performance.
Problems with welded connections were From the structural standpoint, ductility
evidenced by many failures in the 1994 should arise in “plastic hinges” or zones of con-
Northridge earthquake (Fig. 9). In the welds to centrated plasticity primarily in beams. The
the beam flanges in these connections, large tri- behavior of a plastic hinge can be visualized as
axial stresses arise from a combination of that of a nonlinear rotational spring. The ability of
multidirectional forces, residual stresses from these hinges to rotate depends primarily on
welding, and the inevitable imperfections in the delaying the onset of any local or global buckling
welds. These conditions lead to brittle material of the section. Local buckling (Fig. 14) is mostly
behavior because shear forces, the source of most dependent on the slenderness (width/thickness)
deformations in steel, are small when compared ratio of the flanges and webs, and codes prescribe
to normal forces. Even if only minor overall strict limits to ensure that this type of buckling
damage to the structures was observed in struc- does not occur until large rotations are achieved.
tures with failed connections after the Northridge Global buckling of a beam or column, which in
earthquake, a major effort was launched in the this case implies large out-of-plane displace-
USA to better understand connection behavior. ments, is also controlled by slenderness criteria,
Bolt tension elongation

Yielding of T flange
Yielding of T web
a b
Shear tab
Bolt hole
elongation
K2
K3 Bolt slip
K1
K4
Steel connection Mechanistic model
Steel Structures, Fig. 21 (a) T-stub bolted connection and (b) corresponding component model (# RTLeon)
in the form of length/moment of inertia ratios. springs are arranged in series or in parallel and
Global buckling can be minimized by the addi- the overall moment-rotation (M-y) curve derived
tion of bracing along the member length, but this with the aid of simple computer programs that
leads to additional costs and complications in the conduct the analysis of the spring system. In this
construction process. Global buckling or collapse example, the K1 and K2 springs model the panel
of the structure can occur when large lateral zone deformation due to shear, while springs K3
deformations arise and the gravity forces result and K4 model the bending deformations of the
in large additional second-order effects. Global T-stubs. Springs K3 and K4 are made up of the
buckling is addressed in design by specifying contributions of several other springs that model
strict drift (lateral deformation) limits and different deformation components. With the aid
conducting advanced analyses that ensure overall of this approach, it is possible to achieve designs
stability. that meet and exceed current performance
As the results of the research began to filter requirements. Care in the design of these connec-
into practice, a number of new design technolo- tions can lead to very successful designs
gies became popular. For example, for the suc- (Fig. 22).
cessful design of a T-stub connection (Fig. 21a), Because the design of connections is a difficult S
it is necessary to first determine all the possible and time-consuming task, in the USA, the idea of
ductile and brittle failure modes and prioritize “prequalifying” connections has arisen (AISC
them from most brittle to most ductile. One 358 2010). In this case, an extensive experimental
approach to this task is the component approach, campaign is conducted to test a range of beam
in which each deformation mechanism in a joint and column sizes with a particular type of con-
is identified and individually quantified through a nection. If the results indicate reliable ductile
series of small laboratory component tests and cyclic behavior to rotations around 4 %, after
associated analytical studies. These tests are care- careful examination of all the results, the connec-
fully designed to measure one deformation com- tion will become “prequalified.” This implies that
ponent at a time. Each of these components is a series of simple, predefined design steps is all
then represented by a spring with either linear or that is necessary to design that type of
nonlinear characteristics (Fig. 21b). These connection.
Steel Structures,
Fig. 22 Successful design
of a T-stub connection
(# RTLeon)
Steel Structures,
Fig. 23 Innovative,
proprietary connection
(ConXTech, www.
conxtech.com)
An interesting and important consequence of Summary

the “prequalification” process is that a number of
proprietary connections have entered the market. Properly designed modern steel structures can
In these cases, the manufacturer of the connection provide superior performance when subjected to
will provide all the necessary connection design very large earthquake motions. The key to good
data and manufacture all proprietary pieces. For performance is in strict attention to detail in every
example, the ConXL connection (Fig. 23, phase of the construction process: (a) initial
ConXtech Inc.) consists of a concrete-filled tube selection of the structural form; (b) ductile design
HSS column, RBS beams, and a series of steel of members, particularly with respect to plastic
castings that make the field assemblage of the hinge formation; (c) careful attention to connec-
connection a fast and simple operation. tion strength and ductility; (d) frequent
Stochastic Analysis of Linear Systems 3433
interaction between the designer and fabricator to

ensure proper fabrication procedures and toler- Stochastic Analysis of Linear
ances; and (e) proper erection, along with exten- Systems
sive QA/QC for any field welding and bolting
needed. Giuseppe Muscolino
Dipartimento di Ingegnria Civile, Informatica,
Edile, Ambientale e Matematica Applicata,
Università degli Studi di Messina, Messina, Italy
Cross-References
▶ Seismic Strengthening Strategies for Existing

Synonyms
(Code-Deficient) Ordinary Structures
Evolutionary frequency response function; Evo-
References lutionary power spectral density function; Gauss-
ian zero-mean random models of seismic
AISC 358 (2010) Prequalified connections for special and accelerations; Non-geometric spectral moments;
intermediate moment frames for seismic applications.
American Institute of Steel Construction, Chicago
Stochastic analysis
ASCE (1906) The effects of the San Francisco earth-
quake of April 18th, 1906, on Engineering Construc-
tions: report of Committee on Fire and Earthquake Introduction
Damage To Buildings. Trans ASCE 57(2):208–263
Ashton TS (1968) Iron and steel in the industrial revolu-
tion. Augustus M. Kelley, New York The stochastic analysis of structural vibrations
Bodsworth C (ed) (2001) British iron and steel AD1800‐ deals with the description and characterization
2000 and beyond, Book 472. IOM Communications, of structural loads and responses that are
London
Clifton C et al (2011) Steel structures damage from the
modeled as stochastic processes. The probabi-
Christchurch earthquake of February 22, 2011. Bull N listic characterization of the input process could
Z Soc Earthq Eng 44(4):297–318 be extremely complex in time domain where the
Condit CW (1968) American building. The University of probability density functions depend on the
Chicago Press, Chicago
FEMA 355E (2000a) State of the art report on past per-
autocorrelation functions which experimentally
formance of steel moment-frame buildings. FEMA, have to be specified over given set points. Since
Washington, DC this approach is difficult to be used in applica-
FEMA 355B (2000b) State of the art report on welding tions, stochastic vibration analysis of structural
and inspection. FEMA, Washington, DC
Halmos GT (2000) Roll forming handbook. CRC Press,
linear systems subjected to Gaussian input pro-
Boca Raton cesses is quite often performed in the frequency
Hamburger R, Meyer JD (2006) The performance of steel- domain by means of the spectral analysis. This S
frame buildings with infill masonry walls in the 1906 analysis is a very powerful tool for the analytical
San Francisco earthquake. Earthq Spectra, EERI 22
(2):S43–S68
and experimental treatment of a large class of
Kurzman D (2001) Disaster! The Great San Francisco physical as well as structural problems subjected
earthquake and fire of 1906. Harper Perennial, New to random excitations. The main reasons are
York (a) the spectrum has an immediate physical
Leon et al (2014) Performance of RC beam-column joints
in the Christchurch 2010–2011 earthquakes. In: SP‐
interpretation as a power-frequency distribution;
296: symposium honoring James O. Jirsa’s contribu- (b) the spectrum provides information on the
tions in structural concrete: a time to reflect, CD. ACI, stochastic structure of the process; and (c) the
Farmington Hills spectrum may be estimated by fairly simple
Tylecote RF (1976) A history of metallurgy. Metals Soci-
ety, London
numerical techniques which do not require any
Viest et al (1997) Composite construction design for specific assumption of the structure of the
buildings. McGraw-Hill/ASCE, New York, 416 pp process.
3434 Stochastic Analysis of Linear Systems
In the framework of earthquake engineering, crossing rates, distribution of peaks, and extreme
the stationary non-white input models were values. The latter quantities can be evaluated, for
suggested first. These models, which account for stationary input process, as a function of the well-
site properties and for the dominant frequency in known geometric spectral moments (GSMs) intro-
ground motion, fail to reproduce the time-varying duced by Vanmarcke (1972). For stationary sto-
intensity typical of real earthquakes ground- chastic response processes, the GSMs are defined
motion accelerograms. In order to overcome this as the geometric moments of the one-sided power
drawback, the so-called quasi-stationary spectral density (PSD) of the response process.
(or uniformly modulated) random processes have Application of spectral methods to nonstationary
been introduced (see, e.g., Shinozuka and Sato random processes is more difficult than for
1967; Jennings et al. 1969; Hsu and Bernard the stationary ones; indeed for nonstationary
1978). These processes are constructed as the processes the geometric approach fails (Di Paola
product of a stationary zero-mean Gaussian ran- 1985; Muscolino 1991). To perform the structural
dom process by a deterministic function of time; reliability in the latter cases, the so-called
for this reason they are also called separable nongeometric spectral moments (NGSMs) have
nonstationary stochastic processes. been introduced (Michaelov et al. 1999a, b).
Furthermore, a time-varying frequency con- In this study the before outlined topics will be
tent is observed in actual accelerogram records. addressed in order to evaluate the spectral char-
This nonstationary frequency is prevalently due acteristics of the structural response that are use-
to different arrival times of the primary, second- ful to perform the reliability assessment of linear
ary, and surface waves that propagate at different systems subjected to stationary or nonstationary
velocities through the Earth’s crust. To take into mono-/multi-correlated excitations.
account both the simultaneous amplitude and fre-
quency non-stationarity, Spanos and Solomos
(1983) proposed a non-separable model introduc- Spectral Representation of Stochastic
ing a particular evolutionary power spectral den- Processes
sity (EPSD) function; Fan and Ahmadi
(1990) proposed a generalization of the Kanai- To carry out the spectral analysis, it is necessary
Tajimi filter model with time-dependent coeffi- to determine the spectral properties of the
cients; Conte and Peng (1997) defined the involved functions. These properties may be
ground-motion accelerations as the sum of determined through the Fourier-Stieltjes trans-
a finite number of pairwise independent uniform (Priestley 1999).
formly modulated zero-mean Gaussian stochastic Let us consider now a zero-mean stationary
process, the so-called sigma-oscillatory process. stochastic process, F(t). This process is character-
Once the problem is formulated from ized by the feature that its statistical moments do
a mathematical point of view, the further step not change over time and generally arise from any
deals with the evaluation of the structural response “stable” system which has achieved a “steady
to perform the prediction of the safety of structural state.” Moreover, the probabilistic structure of
systems. In this framework, the maximum absolute a stationary process is invariant under a shift of
peak of stationary or nonstationary stochastic the time origin. This is a consequence of the fact
responses may be useful in design information of that a sample of the process will almost certainly
several engineering situations (see, e.g., Lin 1976; not “decay” to zero at infinity. Then the stationary
Lutes and Sarkani 2004; Muscolino and Palmeri processes possess infinite energy. Since
2005; Li and Chen 2009). Approximate proce- a stationary process possesses infinite energy, its
dures to calculate the statistics of the maximum kth sample, F(k)(t), cannot be represented by the
absolute peak of the response have been proposed. Fourier transform. In fact in this case, the Dirichlet
These procedures lead to the probabilistic assess- condition is not satisfied. It follows that the spec-
ment of structural failure as a function of barrier tral representation of a sample of a stationary
stochastic process can be performed only by the In postulating the stationarity of the stochastic
Fourier-Stieltjes integral (Priestley 1999): process, very strong assumptions regarding the
structure of the process are made. Once these
ð
þ1
assumptions are dropped, the process can become
ðkÞ
F ðt Þ ¼ expði o tÞdN ðkÞ ðoÞ (1) nonstationary in many different ways. In the
1 framework of the spectral analysis of
pffiffiffiffiffiffiffi nonstationary processes, Priestley (see, e.g.,
where i ¼ 1 is the imaginary unit and N (k)(o)
Priestley 1999) introduced the evolutionary
is the kth sample of the complex stochastic pro-
power spectral density (EPDS) function. The
cess N(o), satisfying the condition
EPSD function has essentially the same type of
EhdN ðo1 ÞdN ðo2 Þi ¼ dðo1 o2 ÞSNN ðo1 Þdo1 do2
physical interpretation of the PSD function of
stationary processes. The main difference is that
(2) whereas the PSD function describes the power-
where d(•) is the Dirac delta, the symbol Eh•i frequency distribution for the whole stationary
means stochastic average, and the asterisk indi- process, the EPSD function is time dependent
cates the complex conjugate quantity. Notice that and describes the local power-frequency distribu-
in Eq. 2 sometimes the Kronecker delta is intro- tion at each instant time. The theory of EPSD
duced instead of the Dirac delta; this is to avoid the function is the only one which preserves this
inconsistence of this relationship for o1 = o2 physical interpretation for the nonstationary pro-
(Spanos and Solomos 1983). The relationship (2) cesses. Moreover, since the spectrum may be
shows that the stochastic process N(o) is a process estimated by fairly simple numerical techniques,
with orthogonal increments, in the sense that its which do not require any specific assumption of
increments dN(o1) and dN(o2) at any two distinct the structure of the process, this model, based on
points o1 and o2 are uncorrelated random vari- the EPSD function, is nowadays the most adopted
ables. Furthermore, in Eq. 2 SNN (o), which is a real model for the analysis of structures subjected to
and symmetric function, SNN (o) = SNN (o), is nonstationary processes as the seismic motion
the power spectral density (PSD) function of the due to earthquakes.
process N(o). According to the theory of stationary In the Priestley spectral representation of
stochastic process, the autocorrelation function, nonstationary processes, a sample of the
RFF (t), of the zero-mean stationary stochastic pro- nonstationary stochastic process is defined by
cess F(t) is a real function given as (Lin 1976; Lutes the Fourier-Stieltjes integral as follows:
and Sarkani 2004; Li and Chen 2009)
ð
þ1
RFF ðtÞ ¼ EhFðt þ tÞFðtÞi ðkÞ
F ðtÞ ¼ expði o tÞ aðo; tÞd N ðkÞ ðoÞ (5)
ð þ1
þ1 ð S
1
¼ exp½iðo1 ðt þ tÞ o2 tÞ (3)
1 1
where a(o, t) is a slowly varying complex
Ehd N ðo1 Þ d N ðo2 Þi deterministic time-frequency modulating func-
tion which has to satisfy the condition a(o, t)
which in virtue of Eq. 2 leads to a (o, t) and N(o) is an orthogonal process
ð
1 satisfying the condition (2). In Eq. 2 SNN(o)
RFF ðtÞ ¼ EhFðt þ tÞFðtÞi ¼ expðiotÞSFF ðoÞdo is the PSD function of the so-called “embed-
ded” stationary counterpart process, N(o)
1
(Michaelov et al. 1999a). It follows that the
(4)
autocorrelation function of the zero-mean
In this equation SFF (o) is the PSD function of Gaussian nonstationary random process F(t)
the stationary process F(t). can be obtained as
RFF ðt1 , t2 Þ ¼ EhFðt1 ÞFðt2 Þi SFF ðo; tÞ ¼ jaðo; tÞj2 SNN ðoÞ (9)
ð þ1
þ1 ð
¼ exp½iðo1 t1 o2 t2 Þaðo1 , tÞ is the so-called EPSD function of the
1 1 nonstationary process F(t). In the previous equa-
a ðo2 , t2 ÞEhdN ðo1 Þ dN ðo2 Þido1 do2 tions the symbol j • j denotes the modulus of the
(6) function in brackets. The processes characterized
by the EPSD function SFF(o, t) are called fully
It is a real function which, in virtue of Eq. 2, nonstationary or non-separable random process,
leads to since both time and frequency content change,
and they cannot be decoupled. If the modulating
ð
1 function is a time-dependent function,
RFF ðt1 ,t2 Þ ¼ exp½ioðt1 t2 Þaðo, t1 Þa ðo,t2 Þ a(o, t) a(t), the nonstationary process is called
1
quasi-stationary or uniformly modulated or sep-
1ð arable random process. In this case the time
SNN ðoÞdo ¼ exp½ioðt1 t2 ÞSFF ðo, t1 ,t2 Þdo content change is independent by the frequency
content change; indeed, the EPSD function
1
assumes the following expression:
(7)
where SFF ðo; tÞ ¼ a2 ðtÞ SNN ðoÞ (10)
SFF ðo, t1 , t2 Þ ¼ aðo, t1 Þ a ðo, t2 Þ SNN ðoÞ (8) In the stochastic analysis the one-sided PSD is
generally used; the latter can be suitably defined
According to the Priestley evolutionary pro- in the Priestley representation by the following
cess model (Priestley 1999), the function equation:

aðo, t1 Þ a ðo, t2 Þ GNN ðoÞ 2SFF ðo, t1 , t2 Þ, o 0;
GFF ðo, t1 , t2 Þ ¼ (11)
0, o < 0

where GNN (o) (GNN (o) = 2SNN (o), o 0; Re RFF ðt1 , t2 Þ RFF ðt1 , t2 Þ: It can be easily
GNN (o) = 0, o < 0) is the one-sided PSD func- proved that the complex function (12) is the auto-
tion of the stationary counterpart of the input correlation function of a complex process FðtÞ
process F(t). In this case the autocorrelation func- defined as (Di Paola and Petrucci 1990)
tion of the process F(t) is given by the following
pffiffiffi ð
relationship: þ1
FðtÞ ¼ 2 expði o tÞ aðo, tÞ d N ðoÞ (13)

ð
1
0
RFF ðt1 , t2 Þ ¼ exp½ioðt1 t2 Þaðo, t1 Þ
(12)
1 The real part of FðtÞ is proportional to the
process F(t), while the imaginary part of FðtÞ is
a ðo, t2 Þ GNN ðoÞdo
a nonstationary process having stationary coun-
Note that since the one-sided PSD function terpart proportional to Hilbert transform of
GNN (o) is not symmetric, the corresponding the real part of the stationary counterpart of the
autocorrelation function, RFF ðt1 , t2 Þ, is a complex process itself (Di Paola 1985; Di Paola and
function (Di Paola 1985), whose real part coin- Petrucci 1990. Muscolino 1991). The complex
cides with the function defined in Eq. 6: process, FðtÞ , which generates the complex
autocorrelation function (12), has been called where X(b, t) is the so-called hazard function
pre-envelope process by Di Paola (1985). and LjXj(b, 0) is the reliability at time t = 0,
which for the nonstationary case can be assumed
as unity.
Reliability of Linear Structural Systems For narrow-band zero-mean stationary Gauss-
Subjected to Stochastic Excitations ian process, the hazard function has been derived
by Vanmarcke (1972) as
The structural systems are conceived and
designed to survive to natural actions. If the exci- 2 rffiffiffiffiffiffiffiffiffiffiffiffi!3
p
tations are modeled as random processes, the sffiffiffiffiffiffiffiffiffi61 exp b dX1:2 7
dynamic responses are random processes too, 1 l 2, X 6 6
2 l 0, X 7
7
X ðbÞ ¼ 2
and the structural safety needs to be evaluated in p l 0, X 6 4 b 7
5
a probabilistic sense. Among the models of fail- exp 1
2 l 0, X
ure, the simplest one, which is also the most
widely used in practical analyses, is based on (15)
the assumption that a structure fails as soon as
the response at a critical location exits with
a prescribed safe domain for the first time. The
probability of failure, in this case, coincides with sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
the first passage probability, i.e., the probability l21, X
dX ¼ 1 (16)
that the absolute value of the random response l 0, X l 2, X
process X(t) of a selected structural response (e.g.,
strain or stress at a critical point) will exceed
a specified safety bound, b, within a specified In this equation dX is the so-called bandwidth
time interval (Lin 1976). In random vibration parameter of the process X(t) (Vanmarcke 1972,
theory, the problem of probabilistically 1975). This parameter measures the variation of
predicting this event is termed first passage prob- the narrowness of the stochastic process X(t).
lem. Unfortunately, this is one of the most com- Usually, a stochastic process with 0 dX 0.35
plicated problems in computational stochastic is called a narrow-band stochastic process.
mechanics. The solution of this problem has not Finally, in the previous equations l i,X
been derived in exact form, even in the simplest (i = 0, 1, 2) are the so-called geometric spectral
case of the stationary response of a single-degree- moments (GSMs), introduced by Vanmarcke
of-freedom (SDoF) linear oscillator under zero- (1972) as the geometric moments of the
mean Gaussian white noise (Lin 1976; Lutes and one-sided PSD of the response process:
Sarkani 2004; Muscolino and Palmeri 2005). S
Hence, a large number of approximated tech- ð
1
niques have been proposed in literature, which l i, X lGi, X ¼ oi GXX ðoÞ do (17)
differ in generality, complexity, and accuracy. 0
In the framework of approximate methods, the
time-dependent reliability of the structure, based
In this equation the apex G emphasizes the
on the first passage failure criterion, can be
geometric evaluation of the GSMs. Notice that
expressed, for a symmetric barrier, as (Lutes and
l0,X coincides with the variance of the zero-mean
Sarkani 2004)
2 t 3 process X(t), l0,X s2X = EhX2(t)i; l2,X coin-
ð cides with the variance of the zero-mean process

LjXj ðb, tÞ ¼ LjXj ðb, 0Þexp4 X ðb, rÞ dr5 X_ðtÞ, l 2, X s2__ ¼ E X_2 ðtÞ ; while l1,X does not
X
0 coincide with the cross-covariance of the pro-
(14) cesses X(t) and X_ðtÞ.
For narrow-band zero-mean nonstationary of this approach is to establish a time-domain

Gaussian process, the hazard function has been interpretation of the SM. In order to do this, the
derived by Corotis et al. (1972) as pre-envelope covariances as the covariances of
structural systems subjected to a complex-valued
2 rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!3
p random process (the so-called pre-envelope pro-
sffiffiffiffiffiffiffiffiffiffiffiffiffiffi61 exp bdX1þd ðtÞ 7
1 l 2, X ðtÞ6 2l 0, X ðtÞ 7 cess) have been introduced (Di Paola and
X ðb;tÞ ¼ 6 7
p l 0, X ðtÞ6 4 b2 7
5 Petrucci 1990). The real part of this process is
exp 1
2l 0, X ðtÞ proportional to the original nonstationary pro-
cess, while the imaginary part is an auxiliary
(18)
random process related to the real part in such
where d (d = 0 or d = 0.2) is an empirical a way that the complex process exhibits power in
parameter and dX (t) is the bandwidth parameter the positive frequency range only. Since the use
which in the nonstationary case is time depen- of complex pre-envelope process is not very intu-
dent. Because of the non-stationarity of itive, Michaelov et al. (1999a, b) evaluated the
random process, this parameter involves complex pre-envelope covariances as a function of the
functions, and it is defined by Michaelov EPSD of the response and recalled them
et al. (1999a, b) as as nongeometric spectral moments (NGSMs).
It has to be emphasized that the NGSMs contain
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 more information than the “conventional” covari-
Re l 1, X ðtÞ ances. Indeed, the NGSMs have been proved to
dX ðtÞ ¼ 1 (19)
l 0, X ðtÞl 2, X ðtÞ be more appropriate for describing nonstationary
process and can be effectively employed in struc-
The evaluation of the time-dependent quanti- tural reliability applications (Di Paola 1985; Di
ties li, X (t) is conceptually more complicated Paola and Petrucci 1990; Muscolino 1991;
than for stationary processes; indeed for these Michaelov et al. 1999a, b).
processes the geometric approach fails for i = 1 It has to be emphasized that in the framework
and i = 2 (Di Paola 1985; Di Paola and Petrucci of nonstationary analysis of structures, other
1990; Muscolino 1991; Michaelov et al. 1999a, time-dependent parameters, very useful in
b), that is describing the time-variant spectral properties of
the stochastic process, are (i) the mean frequency,
l i, X ðtÞ 6¼ lGi, X ðtÞ n+X (t), which evaluate the variation in time of the
ð
1 mean up-crossing rate of the time axis, and
¼ oi GXX ðo; tÞ do, i ¼ 1, 2 (20) (ii) the central frequency, oC, X (t), which scruti-
0
nizes the variation of the frequency content of the
stochastic process with respect to time. The two
The physical inconsistency on the evaluation functions introduced before can be evaluated as
of the geometric GSMs, in the nonstationary case, a function of NGSMs and have been defined,
as the moments of the one-sided EPSD function respectively, as (Michaelov et al. 1999a, b)
was pointed out by Corotis et al. (1972). In fact
they discovered that for the case of the transient sffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 l 2, X ðtÞ Re l 1, X ðtÞ
response of an oscillator subjected to stationary nþ
X ð tÞ ¼ ; o C , X ð tÞ ¼
2p l 0, X ðtÞ l 0, X ðtÞ
Gaussian white noise processes, the second GSM
does not exist because it is unbounded. At same (21)
time in the stationary case, this GSM, which is the
limit of the transient as the time approaches infin- In stationary case the mean frequency, n+X, and
ity, is finite. The first that considered the problem the central frequency, oC, X, are not time depen-
of spectral characteristics from a nongeometric dent. Moreover, in the latter case, because of the
point of view was Di Paola (1985). The basic idea Im{l1, X} = 0, it follows Re{l1, X} l1,X.
Response of Single-Degree-of-Freedom Denoting by yp(t) the particular solution vec-

(SDoF) Oscillators tor, the solution of Eq. 24 can be evaluated as
(Borino and Muscolino 1986)
Fundamental of Deterministic Analysis
The theory of deterministic linear systems plays yðtÞ ¼ yp ðtÞþY0 ðt t0 Þ yðt0 Þ yp ðt0 Þ (26)
a fundamental role in the dynamic analysis of
structures subjected to stochastic excitations. where Y0(t) is the so-called transition matrix:
For this reason in this section the fundamental

of deterministic analysis of SDoF subjected to o20 g0 ðtÞ h0 ðtÞ
Y0 ðtÞ ¼ expðD0 tÞ ¼
deterministic excitation is synthetically o20 h0 ðtÞ h_0 ðtÞ
reviewed. Particular care has been devoted to (27)
the state-space approach. This approach is the
best suited for the development of formulations with
in the framework of random vibrations. In fact, its

1 x 0 o0
adaptability to numerical method of solution of g 0 ð tÞ ¼ expð x 0 o 0 t Þ cos ð o 0 t Þ þ sin ð o 0 t Þ ;
o20 o0
differential equations and its extension to multi- 1
h0 ðtÞ ¼ g_0 ðtÞ ¼ expðx0 o0 tÞsin ðo0 tÞ;
degree-of-freedom (MDoF) systems are very o0

x o0
straightforward. h_0 ðtÞ ¼ expðx0 o0 tÞ cos ðo0 tÞ 0 sin ðo0 tÞ
The equation of motion of a linear oscillator o0
with mass, m; viscous damping, c; stiffness, k; (28)
and subjected to the excitation f (t) and at rest at pffiffiffiffiffiffiffiffiffiffiffiffiffi
and o0 ¼ o0 1 x0 is the damped natural cir-
initial time of motion, t = t0, can be written as
cular frequency. Notice that the contribution of
the last term in the right member of Eq. 26
m u€ðtÞ þ c u_ðtÞ þ k uðtÞ ¼ f ðtÞ;
(22) decreases in the time because the transition
uðt0 Þ ¼ 0, u_ðt0 Þ ¼ 0 matrix satisfies the following condition:
or in canonical form as follows: lim Q0 ðtÞ ¼ 0 (29)

t!1
u€ðtÞ þ 2 x0 o0 u_ðtÞ þ o20 uðtÞ ¼ FðtÞ ; Alternatively, by applying the so-called

(23) parameter variation method, the solution of
uðt0 Þ ¼ 0, u_ðt0 Þ ¼ 0
Eq. 24 can be written in integral form as follows:
where u(t) ispthe displacement
ffiffiffiffiffiffiffiffi
ffi response of the
ðt
mass, o0 ¼ k=m is
pffiffiffiffiffiffiffi the natural circular fre-
yðtÞ ¼ Q0 ðt t0 Þyðt0 Þþ Q0 ðt tÞ v FðtÞdt (30)
quency, x0 ¼ c=2 m k is the damping ratio, and
t0 S
F(t) = f(t)/m; a dot over a variable denotes dif-
ferentiation with respect to time t. In state vari- For quiescent systems, because the relation-
ables the equation of motion of the oscillator, in ship y(t0) = 0 is satisfied, Eqs. 26 and 30 become
canonical form, can be written as a set of two
first-order differential equations: yðtÞ ¼ yp ðtÞ Q0 ðt t0 Þyp ðt0 Þ
ðt
y_ ðtÞ ¼ D0 yðtÞ þ v0 FðtÞ; yð t 0 Þ ¼ 0 (24)
¼ Q0 ðt tÞ v FðtÞdt (31)
t0
where

Notice that in this case the first element of
uðtÞ 0 1 0
yðtÞ ¼ , D0 ¼ , v ¼ vector y(t), written in integral form, coincides
u_ðtÞ o20 2x0 o0 0
1
with the well-known Duhamel integral. Starting
(25) by the integral form solution in state variables,
it is possible to derive a very powerful uncondi- one-sided PSD, the input process is a complex
tionally stable numerical procedure for the eval- one. It follows that the response processes, u(t),
uation of the structural response (see, e.g., Borino is a complex function too. After some algebra,
and Muscolino 1986). for the quiescent oscillator (Eq. 23), the
NGSMs can be evaluated, in time domain, as
Stochastic Response (Di Paola 1985; Di Paola and Petrucci 1990,
Mathematically strictly speaking, as Muscolino1991)
a consequence of the introduction of the
ðt ðt

l0, uu ðtÞ EhuðtÞ u ðtÞi ¼ hðt t1 Þhðt t2 ÞRFF ðt1 , t2 Þ dt1 dt2 ;
00
ðt ðt
l1, uu ðtÞ ¼ i hðt t1 Þh_ðt t2 ÞRFF ðt1 , t2 Þ dt1 dt2 ; (32)
00
ðt ðt

l2, uu ðtÞ Ehu_ðtÞ u_ ðtÞi ¼ h_ðt t1 Þh_ðt t2 ÞRFF ðt1 , t2 Þ dt1 dt2
00
where RFF ðt1 , t2 Þ is the complex autocorrelation Notice that the function Z0(o, t) is the
function defined in Eq. 12. Moreover, the pres- so-called evolutionary frequency response func-
ence of the imaginary unit in the second of Eq. 32 tion of the oscillator (Li and Chen 2009).
inverts the roles of the real and imaginary parts of Remarkably, since the integrals (Eq. 34) are con-
l1, uu(t) with respect to the variances l0, uu(t) volution integrals of Duhamel’s type, they can be
and l2, uu‘(t); furthermore, while l0, uu(t) interpreted as the response, in terms of state vari-
and l2, uu(t) are real functions, l1, uu(t) is ables, of the quiescent oscillator, at time t = 0,
a complex one. Substituting Eq. 12 into Eq. 32, subjected to the deterministic complex function
the following relationships are obtained: f(o, t) = exp(iot) a(o, t). By introducing the
state variables, the evolutionary frequency
ð
1
response vector function can be defined as
l0, uu ðtÞ ¼ Z 0 ðo; tÞZ 0 ðo; tÞGNN ðoÞdo;

0 Z ðo; tÞ
ð
1 Y0 ðo; tÞ ¼ _0 (35)
Z0 ðo; tÞ
l1, uu ðtÞ ¼ i Z 0 ðo; tÞZ_0 ðo; tÞGNN ðoÞdo; (33)
0 It follows that relationships (33) can be rewrit-
ð
1
ten in compact form as follows:
l2, uu ðtÞ ¼ Z_0 ðo; tÞZ_0 ðo; tÞGNN ðoÞdo

0 l 0, uu ðtÞ il1, uu ðtÞ
Suu ðtÞ ¼
il 1, uu ðtÞ l2, uu ðtÞ
where
ð
1
ðt
GNN ðoÞ Y 0 ðo; tÞYT0 ðo; tÞdo (36)
Z0 ðo, tÞ ¼ hðt tÞexpðiotÞaðo, tÞ dt;
0
0
(34)
ðt This matrix coincides with the so-called pre-
Z_0 ðo, tÞ ¼ h_ðt tÞexpðiotÞaðo, tÞ dt envelope covariance matrix introduced by Di
0 Paola and Petrucci (1990) which is a complex
matrix. As a conclusion, for input processes char- ð

1
acterized by one-sided EPSD function, the cross- lim Z 0 ðo,tÞ ¼ h0 ðt tÞexpðiotÞdt ¼ expðiotÞ H 0 ðoÞ;
t!1
covariance matrix (Eq. 36) of an oscillator is 0
ð
1
a complex matrix whose elements are the lim Z_0 ðo,tÞ ¼ h_0 ðt tÞexpðiotÞdt ¼ ioexpðiotÞ H 0 ðoÞ
t!1
NGSMs. 0
Generalizing Eq. 32, it can be easily proved (41)
that the complex cross-correlation function
matrix of the zero-mean response process, where h0(t) and h_0 ðtÞ are the functions defined in
which collects the cross-correlation of the Eq. 28 and H0(o) is the frequency response func-
pre-envelope response processes, can be evalu- tion of the oscillator defined as
ated as
ð
1
ð
1 H0 ðoÞ ¼ h0 ðtÞexpðiotÞdt
Ruu ðt1 , t2 Þ ¼ GNN ðoÞ Y 0 ðo, t1 ÞYT0 ðo, t2 Þdo 0
0 1
¼ (42)
(37) o20 o2 þ i 2x0 o0 o
Finally, according to the Priestley evolution- By substituting Eq. 41 into Eq. 35 and the
ary process model (Priestley 1999), this complex results into Eq. 36, the following relationship is
function matrix Ruu(t1, t2) can be also rewritten obtained:
as
ð
1
Suu ¼ GNN ðoÞ H 0 ðoÞH0T ðoÞ do

ð
1
Ruu ðt1 , t2 Þ ¼ exp½ioðt1 t2 ÞGuu ðo, t1 , t2 Þdo
0

l0, uu il1, uu
0 (43)
il 1, uu l2, uu
(38)
where
where

H 0 ðo Þ
Guu ðo , t1 , t2 Þ ¼ GNN ðoÞ Y 0 ðo, t1 ÞYT0 ðo, t2 Þ H0 ðoÞ5 (44)
ioH 0 ðoÞ
(39)
The elements of this matrix Suu are the GSMs,
Consequently the NGSMs of the oscillator introduced by Vanmarcke (1972), that is
response are the elements of the pre-envelope S
covariance matrix, given as li, uu lGi, uu
ð
1
ð
1
¼ oi jH 0 ðoÞj2 GNN ðoÞdo; i ¼ 0,1, 2
Suu ðtÞ Ruu ðt, tÞ Guu ðo, tÞ do (40)
0
0
(45)
Let us assume now the modulating function Response of Multi-degree-of-Freedom

aðo, tÞ ¼ UðtÞ. Starting from the nonstationary (MDoF) Systems
formulation, it is possible to deduce the formula-
tion in the case of stationary input. This result is Fundamental of Deterministic Analysis
obtained by performing the limit as t ! 1 into Let us consider the equation of motion of a linear
Eq. 34: quiescent n-degree-of-freedom (n-DoF) classically
damped structural system whose dynamic behavior quantities 2xjoj, xj being the modal damping
is governed by the equation of motion: ratio. It follows that the jth differential Eq. 49
can be written as
M u€ðtÞ þ Cu_ ðtÞ þ KuðtÞ ¼ f ðtÞ (46)
q€j ðtÞ þ 2xj oj q_j ðtÞ þ o2j qj ðtÞ ¼ fTj f ðtÞ (51)
where M, C, and K are the (n n) mass,
damping, and stiffness matrices of the structure; In the state space, Eq. 51 can be written in the
u(t) is the (n 1) vector of displacements, hav- first-order form as
ing for ith element ui(t); and f(t) denotes the
external load vector. Under the assumption of
y_ j ðtÞ ¼ Dj yj ðtÞ þ Vj f ðtÞ (52)
classically damped system, the equation of
motion can be decoupled by applying the modal
analysis. To this aim let us introduce the modal where
coordinate transformation: " # " #
qj ð t Þ 0 1
X
m yj ðtÞ5 ; Dj ¼ ;
q_j ðtÞ o2j 2xj oj
uðtÞ ¼ F qðtÞ ¼ f j qj ð t Þ ) ui ð t Þ " #
j¼1 0
X Vj ¼
m
fTj
¼ fi j qj ðtÞ (47)
j¼1 (53)
In this equation, F ¼ ½ f 1 f 2 . . . f m Stochastic Response for Mono-correlated

is the modal matrix, of order n m, collecting Stochastic Input Processes
the m eigenvectors fj, normalized with respect to Let us assume now that the forcing term is
the mass matrix M, solutions of the following a mono-correlated zero-mean Gaussian random
eigenproblem: process vector given by the relationship:
K1 MF ¼ FV2 ; FT MF 5 Im (48) f ðtÞ ¼ b FðtÞ (54)
In this equation V is a diagonal matrix listing where b is the (n 1) vector of spatial distribu-
the undamped natural circular frequency oj, Im is tion of loads and FðtÞ is a zero-mean Gaussian
the identity matrix of order m, and the apex nonstationary random process. It follows that the
T means transpose operator. Once the modal jth differential Eq. 51 can be written as
matrix F is evaluated, by applying the coordinate
transformations (47) to Eq. 46, the following set
q€j ðtÞ þ 2xj oj q_j ðtÞ þ o2j qj ðtÞ ¼ pj FðtÞ, j ¼ 1, 2, . . ., m
of decoupled second-order differential equations
is obtained: (55)
where
q€ðtÞ þ J q_ ðtÞ þ V2 qðtÞ ¼ FT f ðtÞ (49)
in which J is a generalized damping matrix pj ¼ fTj b (56)

given by
After very simple algebra it can be shown that
J ¼ FT CF (50) the jth NGSMs, l j, ui ui ðtÞ (i = 0, 1, 2), of the ith
nodal displacement response, ui(t), are given as
For classically damped structures the modal a function of modal NGSMs, pj lj,k‘(t)
damping matrix J is a diagonal matrix listing the (‘, k = 1 . . . , m), by the following relationships:
m X
X m between the response process and the response
l 0, ui ui ðtÞ ¼ pk p‘ fik fi‘ l0, k‘ ðtÞ; velocity process of the same linear system
k¼1 ‘¼1
m X
X m subjected to a nonstationary input whose station-
l1, u i u i ðtÞ ¼ pk p‘ fik fi‘ l1, k‘ ðtÞ; (57) ary counterpart is proportional to its Hilbert trans-
k¼1 ‘¼1 form (Di Paola 1985; Langley 1986; Di Paola and
Xm X m
l 2, u i u i ðtÞ ¼ pk p‘ fik fi‘ l2, k‘ ðtÞ Petrucci 1990; Muscolino 1991). As shown in
k¼1 ‘¼1 Eq. 57 the nodal NGSMs can be evaluated as
a function of lr,k‘(t), r = 0, 1, 2, which are the
It has to be emphasized that the zeroth NGSM, so-called time-dependent modal NGSMs
l 0, ui ui ðtÞ, and the second-order NGSM, l 2, ui ui ðtÞ, “purged” by pj factors. After some algebra,
are real functions that coincide with the variances these quantities, which are complex ones, can
of the response in terms of displacement and be evaluated, in time domain, for quiescent struc-
velocity, respectively, while the first-order tural systems as (Di Paola 1985; Di Paola and
NGSM, l1, ui ui ðtÞ , is a complex quantity whose Petrucci 1990; Muscolino 1991)
real part can be evaluated as the cross-covariance
ðt ðt
l0, k‘ ðtÞ ¼ hk ðt t1 Þh‘ ðt t2 ÞRFF ðt1 , t2 Þdt1 dt2 ;
00
ðt ðt
l1, k‘ ðtÞ ¼ i hk ðt t1 Þh_‘ ðt t2 ÞRFF ðt1 , t2 Þdt1 dt2 ; k ¼ 1, 2, . . . , m; ‘ ¼ 1, 2, . . . , m;
00
ðt ðt
l2, k‘ ðtÞ ¼ h_k ðt t1 Þh_‘ ðt t2 ÞRFF ðt1 , t2 Þdt1 dt2
00
(58)
where RFF ðt1 , t2 Þ is the complex autocorrelation process itself; it follows that zj(t) = qj(t)/pj is
function defined in Eq. 12. Notice that the pres- a complex process too. Notice that, with the posi-
ence of the imaginary unit in the second term of tion pj = 1, Eq. 59 coincides with Eq. 55. Then
the second of Eq. 58 inverts the roles of the real by substituting the complex function RFF ðt1 , t2 Þ,
and imaginary parts of l1,k‘(t) with respect to the defined in Eq. 12, into Eq. 58, after very simple
cross-covariance l0,k‘(t) and l2,k‘(t); further- algebra, it is possible to evaluate the “purged” S
more for k = ‘ while l0,k‘(t) and l2,k‘(t) become NGSMs as
real quantities, l1,k‘(t) remains a complex one.
It can be easily proved that the “purged”
NGSMs, given in Eq. 58, can be evaluated as ð
1
a function of the statistics of the response of l0, k‘ ðtÞ ¼ Z k ðo, tÞZ ‘ ðo, tÞGNN ðoÞdo;
a dummy oscillator whose motion is governed by 0
ð
1
z€j ðtÞ þ 2xj oj z_j ðtÞ þ o2j zj ðtÞ ¼ Fð t Þ (59) l1, k‘ ðtÞ ¼ i Z k ðo, tÞZ_‘ ðo, tÞGNN ðoÞdo;
0
ð
1
where FðtÞ is a complex process whose imaginary
part is a process having stationary counterpart l2, k‘ ðtÞ ¼ Z_k ðo, tÞZ_‘ ðo, tÞGNN ðoÞdo
proportional to Hilbert transform of the real part 0
of the stationary counterpart of the complex (60)

where, for j = k, ‘, the following positions have pre-envelope response processes, according to
been made: the Priestley evolutionary process model
(Priestley 1999), can be evaluated as
ðt
Z j ðo, tÞ ¼ hj ðt tÞexpðiotÞaðo, tÞdt; ð
1
Rk‘ ðt1 , t2 Þ ¼ GNN ðoÞ Y k ðo, t1 ÞYT‘ ðo, t2 Þdo

0
(61)
ðt 0
Z_j ðo, tÞ ¼ h_j ðt tÞexpðiotÞaðo, tÞdt ð

1
0
exp½ioðt1 t2 ÞGk‘ ðo, t1 , t2 Þdo
0
with (65)
1 where
hj ð t Þ ¼ exp xj oj t sin oj t (62)
oj
qffiffiffiffiffiffiffiffiffiffiffiffiffi Gk‘ ðo, t1 , t2 Þ ¼ GNN ðoÞ Y k ðo, t1 ÞYT‘ ðo, t2 Þ
and oj ¼ oj 1 x2j (j = k, ‘) is the damped (66)
circular frequency of the jth dummy oscillator.
Moreover, introducing the state variable, the is the cross-modal EPSD function matrix between
modal evolutionary frequency response function the k, ‘th dummy oscillators. By means of the
vector is defined as modal transformation (Eq. 47), the nodal autocor-
relation and the EPSD function matrices of the

Z j ðo, tÞ displacement response, ui(t), can be evaluated,
Yj ðo, tÞ ¼ _ ; j ¼ k, ‘ (63)
Z j ðo, tÞ after very simple algebra, respectively, as follows:
Then, relationships (33) can be rewritten in

m X
X m
Ru i u i ðt1 , t2 Þ ¼ pk p‘ fik fi‘ Rk‘ ðt1 , t2 Þ,
compact form as follows: k¼1 ‘¼1
Xm X m

Gu i u i ðo, t1 , t2 Þ ¼ pk p‘ fik fi‘ Gk‘ ðo, t1 , t2 Þ
l 0, k‘ ðtÞ il1, k‘ ðtÞ k¼1 ‘¼1
Sk‘ ðtÞ ¼
il 1, k‘ ðtÞ l2, k‘ ðtÞ (67)
ð
1
GNN ðoÞ Y k ðo, tÞYT‘ ðo, tÞdo (64) Consequently the nodal displacement NGSMs
0
are the elements of the nodal pre-envelope
covariance matrix, given as
which coincides with the cross-modal pre- ð

1
envelope covariance matrix which is a complex Su i u i ðtÞ Ru i u i ðt, tÞ Gu i u i ðo, tÞ do
matrix (Di Paola and Petrucci 1990). As
0
a conclusion for input processes characterized m X
X m
by one-sided PSD function, the cross-covariance ¼ pk p‘ fik fi‘ Sk‘ ðtÞ (68)
matrix (Eq. 64) between the k, ‘th dummy oscil- k¼1 ‘¼1
lators is a complex matrix whose elements are
the “purged” NGSMs. It can be easily proved that Let us assume now the modulating function
the complex cross-correlation function matrix aðo, tÞ ¼ UðtÞ . Starting from the nonstationary
of the zero-mean modal “purged” response pro- formulation, it is possible to deduce the formula-
cess, which collects the cross-correlation of the tion in the case of stationary input. This result
is obtained by performing the limit as t ! 1 Stochastic Response for Multi-correlated

into Eq. 61: Stochastic Input Processes
In earthquake engineering it is common to
ð
1 assume that the entire base of a structure is
lim Z j ðo, tÞ ¼ hj ðt tÞexpðiotÞdt subjected to a uniform ground motion. This
t!1
0 hypothesis inherently implies that the ground
¼ expði o tÞ H j ðoÞ; motion is a result of spatially uniform motion.
(69) Thus it is certainly true when the base dimensions
ð
1
of the structure are small relative to the seismic
lim Z j ðo, tÞ ¼ h_j ðt tÞexpðiotÞdt
_
t!1 vibration wavelengths. From an analytical point
0
of view, this hypothesis leads to mono-correlated
¼ i o expði o tÞ H j ðoÞ zero-mean Gaussian models of the ground-
motion acceleration. It follows that this model
where hj(t) is the functions defined in Eq. 62 and of earthquake excitations is undoubtedly advan-
Hj(o) is the frequency response function of the jth tageous, because it substantially facilitates the
modal oscillator: stochastic analysis of dynamic problem. Indeed,
in this case the stochastic analysis requires the
ð
1
knowledge of only one PSD of the input process.
Hj ðoÞ ¼ hj ðtÞexpðiotÞdt
The assumption of uniform ground motion is
0 inappropriate for structures, such as long span
1 bridges, which require accounting for spatial var-
¼ (70)
o2j o2 þ i 2xj oj o iability of the support motions. The main sources
of ground-motion spatial variability are the loss
By substituting Eq. 69 into Eq. 63 and the of coherency of seismic waves with distance, the
results into Eq. 64, the following relationship is difference in the arrival times of waves at sepa-
obtained: rate supports, and the difference in soil conditions
underneath the supports. In these cases the multi-
ð
1
correlated model of the seismic excitation is more
Sk‘ ¼ GNN ðoÞ H k ðoÞH‘T ðoÞ do appropriate.
0 In order to take into account of the spatial

l 0, k‘ il1, k‘ variability of earthquake-induced ground
(71)
il 1, k‘ l2, k‘ motions, let us consider an n-DoF structural sys-
tem subjected to an N support motion. It follows
where that the stochastic forcing vector process has to
be modeled as a multi-correlated zero-mean S

H j ðoÞ Gaussian random vector process:
H j ðo Þ ¼ j ¼ k, ‘ (72)
ioH j ðoÞ
f ðtÞ ¼ B FðtÞ (74)
The elements of the matrix Sk‘ are the cross
GSMs, introduced by Vanmarcke (1972), that is, where B is the (n N) matrix of spatial distribu-
tion of loads and FðtÞ is a zero-mean Gaussian
l i, k‘ lGi, k‘ nonstationary multi-correlated random vector
process of order N. Following the Priestley spec-
ð
1
tral representation of nonstationary processes
¼ oi H k ðoÞH ‘ ðoÞGNN ðoÞdo; i ¼ 0,1, 2 (Priestley 1999), this vector can be defined by
0 the Fourier-Stieltjes integral as follows
(73) (Di Paola and Petrucci 1990):
pffiffiffi ð
þ1 This relationship shows that the stationary
FðtÞ ¼ 2 expði o tÞ Aðo, tÞdNðoÞ (75) counterpart of the multi-correlated stochastic
1
process is a vector process, N(o) (of order N),
with orthogonal increments. Furthermore,
where A(o, t) is a slowly varying complex deter- GNN(o) is an (N N) Hermitian matrix function
ministic time-frequency modulating diagonal which describes the one-sided PSD function
function matrix, of order N N, which has to matrix of the so-called “embedded” stationary
satisfy the condition A(o, t) A (o, t), while counterpart vector process, N(o). After some
N(o) is an orthogonal vector process satisfying algebra it can be proved that the autocorrelation
the following conditions: function matrix of the zero-mean Gaussian
nonstationary random vector process FðtÞ can
E d Nðo1 Þ dN T ðo2 Þ be obtained as
¼ dðo1 o2 Þ GNN ðo1 Þd o1 do2 (76)
D ð
E 1
T
RFF ðt1 , t2 Þ ¼ E Fðt1 ÞF ðt2 Þ ¼ exp½ioðt1 t2 ÞAðo, t1 Þ GNN ðoÞ A ðo, t2 Þdo
0
ð
1 (77)
¼ exp½ioðt1 t2 ÞGFF ðo, t1 , t2 Þdo
1
where dissipation in the ground medium, and (iv) the

site-response effect associated to spatially vary-
GFF ðo, t1 , t2 Þ ¼ Aðo, t1 Þ GNN ðoÞ A ðo, t2 Þ ing of the local soil profiles. These effects are
(78) taken into account by the so-called coherency
function, gNN,s r(o), defined as
According to the Priestley evolutionary pro-
cess model (Priestley 1999), the EPSD function GNN, s r ðoÞ
gNN, s r ðoÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (80)
matrix of the nonstationary multi-correlated pro- GNN, r r ðoÞ GNN, s s ðoÞ
cess FðtÞ is given as
where GNN,s r(o) is the sth and rth element of the
matrix GNN(o) that is the one-sided PSD at the
GFF ðo; tÞ ¼ Aðo; tÞ GNN ðoÞ A ðo; tÞ (79)
sth and the rth supports. The coherency function
is usually written as
In the framework of stochastic seismic analy-
sis, the main distinct phenomena that give rise to gNN, s r ðoÞ ¼ rs r ðoÞexpði o ds r =cÞ (81)
the spatial variability of earthquake-induced
ground motions are (i) the incoherence effect in which exp(i o ds r/c) is a measure of the
associated to loss of coherency of seismic waves wave-passage delay due to the apparent velocity
due to the differential superpositioning of waves of waves, c represents the velocity of the seismic
arriving from an extended source, (ii) the wave- waves through the ground (c decreases as the
passage effect due to difference in the arrival distance between the support points increases or
times of waves at separate supports, (iii) the the soil is softer), dsr is the distance between the
attenuation effect due to gradual decay of wave rth and sth support points, and rsr(o) is the real
amplitudes with distance due to energy frequency-dependent spatial correlation function
between the two support points. Several models response in terms of displacement and velocity,
of the coherency function have been proposed in respectively, while the first-order NGSM,
literature (see, e.g., Harichandran and l1, ui ui ðtÞ, is a complex quantity (Di Paola 1985;
Vanmarcke 1986; Zerva 1991). Moreover, the Langley 1986; Di Paola and Petrucci 1990;
one-sided cross-PSD of ground accelerations in Muscolino 1991).
a particular direction between surface points As shown in Eq. 85 the nodal NGSMs can be
r and s, GNN,sr(o), may be written as the product evaluated as a function of lj,k‘rs(t), j = 0, 1, 2,
of the coherence function by the target one-sided which are the so-called time-dependent modal
PSD of ground acceleration G0(o) as follows: NGSMs “purged” by pjv factors. After some alge-
bra, these quantities, which are complex ones,
gNN, s r ðoÞ G0 ðoÞ, for r 6¼ s can be evaluated, in frequency domain, for qui-
GNN, s r ðoÞ ¼
G0 ðoÞ, for r ¼ s escent structural systems, at time t = 0, as
(82) (Di Paola 1985; Di Paola and Petrucci 1990,
Muscolino1991):
Let us evaluate now the NGSMs of the struc-
tural response. In the case of multi-correlated ð
1
excitations, the forcing vector f(t) assumes the l 0, k ‘r s ðtÞ ¼ Z k r ðo,tÞZ‘ s ðo,tÞGNN, s r ðoÞdo;
expression (Eq. 74); it follows that the differen- 0
tial equation (49), governing the motion in modal ð
1
subspace, can be written as l 1, k ‘r s ðtÞ ¼ i Z k r ðo, tÞZ_‘s ðo, tÞGNN, s r ðoÞdo;

0
q€ðtÞ þ J q_ ðtÞ þ V2 qðtÞ ¼ P F ðtÞ (83) ð
1

l 2, k ‘r s ðtÞ ¼ Z_k r ðo,tÞZ_‘ s ðo,tÞGNN, s r ðoÞdo
where 0
(86)
P ¼ FT B (84)
where for j = k, ‘, the following positions have
After very tedious algebra it can be shown that been made:
the jth NGSMs, l j, ui ui ðtÞ (j = 0, 1, 2), of the jth
nodal displacement response, ui(t), are given as ðt
a function of modal NGSMs, pj v lj,k‘rs(t) Z j v ðo, tÞ ¼ hj ðt tÞexpðiotÞavv ðo, tÞ dt;
(‘, k = 1, . . . , m ; r, s = 1, , N), by the fol-
0
lowing relationships:
ðt
Z_j v ðo, tÞ ¼ h_j ðt tÞexpðiotÞavv ðo, tÞ dt;
X
m X
m X
N X
N
l 0, ui ui ðtÞ ¼ pk r p‘s f i k f i ‘ l 0, k ‘r s ðtÞ; 0
S
k¼1 ‘¼1 r¼1 s¼1 j ¼ k, ‘; v ¼ r, s
X
m X
m X
N X
N
l 1, ui ui ðtÞ ¼ pk r p‘s f i k f i ‘ l 1, k ‘r s ðtÞ; (87)
k¼1 ‘¼1 r¼1 s¼1
Xm X m X N X N
with hj(t) the function defined in Eq. 62 and
l 2, ui ui ðtÞ ¼ pk r p‘s f i k f i ‘ l 2, k ‘r s ðtÞ avv(o, t) the vth element of the diagonal matrix
k¼1 ‘¼1 r¼1 s¼1
A(o, t). Moreover, introducing the state variable,
(85)
the modal evolutionary frequency response func-
tion vector is defined as
where pjv is the jth and vth element of the matrix
P( j = k, ‘; v = r, s). Once again it has to be
emphasized that the zeroth NGSM, l 0, ui ui ðtÞ, Zj v ðo, tÞ

Yj v ðo, tÞ ¼ _ vðo, tÞ ; j ¼ k, ‘; v ¼ r, s
and the second-order NGSM, l 2, ui ui ðtÞ, are real Zj
functions that coincide with the variances of the (88)
a 1.2
b
1 GFF (w,t)
tmax=15 sec 1.2

0.8
0.8
a(t )
0.6 tmax=5 sec 0.4 0

0
0.4
10
10
0.2 tmax=1sec w
t
20 20
0
0 10 20 30
t 30
Stochastic Analysis of Linear Systems, Fig. 1 Hsu which it takes the maximum value; (b) EPSD function
and Bernard (1978) model of the uniformly modulated GFF(o, t) = |a(t)|2 GNN(o) with GNN(o) = 1 cm2/sec3
nonstationary excitation: (a) modulating and a = 1/5
function for different value of the time instant, tmax, in
Then relationships (33) can be rewritten in ð

1
compact form as follows: Rk ‘r s ðt1 ,t2 Þ ¼ Y k r ðo, t1 ÞGNN, sr ðoÞYT‘s ðo, t2 Þdo

0
l 0, k ‘ r s ðtÞ il 1, k ‘ r s ðtÞ ð
1
S k ‘ r s ðtÞ ¼
il 1, k ‘ r s ðtÞ l2, k ‘ r s ðtÞ exp½ioðt1 t2 ÞGk ‘ r s ðo,t1 , t2 Þdo
ð
1 0
Y k r ðo, tÞ GNN, s r ðoÞ YT‘ s ðo, tÞdo (90)

0
where
(89)
where Sk ‘ r s(t) coincides with the cross-modal Gk ‘ r s ðo, t1 , t2 Þ ¼ Y k r ðo, t1 ÞGNN, sr ðoÞYT‘ s ðo, t2 Þ
pre-envelope covariance matrix (Di Paola and (91)
Petrucci 1990). The one-sided PSD GNN, s r(o)
can be evaluated as a function of the coherence is the cross-modal EPSD function matrix
function introduced in Eq. 81. Notice that between the k, ‘th dummy oscillators at the rth
according to the Priestley evolutionary process and sth support points. By means of the modal
model (Priestley 1999), the complex cross- transformation (47), the nodal autocorrelation
correlation function matrix of the zero-mean and the EPSD function matrices of the displace-
modal “purged” response process can be evalu- ment response, ui(t), can be evaluated, after very
ated as simple algebra, respectively, as follows:
m X
X m X
N X
N
Ru i u i ðt1 , t2 Þ ¼ pk r p‘ s f i k f i ‘ Rk ‘ r s ðt1 , t2 Þ,
k¼1 ‘¼1 r¼1 s¼1
(92)
Xm X m X N X N
Gu i u i ðo, t1 , t2 Þ ¼ pk r p‘ s f i k f i ‘ Gk ‘ r s ðo, t1 , t2 Þ
k¼1 ‘¼1 r¼1 s¼1
0.4
0.08
0.06 0.3
Re {λ1(t)} [cm2/sec]
tmax=15 sec
tmax=15 sec
λ0(t ) [cm2]
0.04 0.2
tmax=5 sec
tmax=5 sec
0.02 0.1
tmax=1 sec tmax=1 sec
0 0
0 10 20 30 0 10 20 30
t t
2.5
1
2 0.8
λ2(t ) [cm2/sec2]
tmax=15 sec
1.5 0.6
δuu(t)
1 0.4 tmax=15 sec

0.5 0.2
tmax=1 sec
tmax=1 sec
0 0
0 10 20 30 0 10 20 30
t t
Stochastic Analysis of Linear Systems, Fig. 2 NGSM (solid line) and adopting the Hsu and Bernard (1978)
functions, lj,uu(t) (j = 0, 1, 2) and bandwidth parame- model, for different value of the time instant, tmax, in
ter, duu(t), of the transient response of an oscillator which the modulating a(t) function take the maximum
(o0 = 2p rad/sec, x0 = 0.05) in the stationary case value (dashed lines)
S
Consequently the nodal displacement NGSMs Calculating Nonstationary Stochastic
are the elements of the nodal pre-envelope Responses
covariance matrix, given as
ð
1 In this section, in order to evidence the
Su i u i ðtÞ Ru i u i ðt, tÞ Gu i u i ðo, tÞ do main differences between the stochastic
responses of structural systems subjected to both
0
X
m X
m X
N X
N uniformly and fully nonstationary models of
¼ pk r p‘s f i k f i ‘ Sk ‘ r s ðtÞ seismic excitations, the spectral characteristics
k¼1 ‘¼1 r¼1 s¼1 of the response of two SDoF oscillators are
(93) evaluated.
0.001
0.025
0.0008 0.02
Re {λ1(t)} [cm2/sec]
tmax=15 sec
tmax=15 sec
λ0(t ) [cm2]
0.0006 0.015
0.0004 0.01
0.0002 0.005
0 0
0 10 20 30 0 10 20 30
t t
0.7
1
0.6
0.8
0.5
λ2(t ) [cm2/sec2]
tmax=15 sec
0.4 0.6
δuu(t)
0.3
0.4 tmax=15 sec
0.2 tmax=5 sec tmax=5 sec
0.2
0.1
tmax=1 sec
tmax=1 sec
0 0
0 10 20 30 0 10 20 30
t t
Stochastic Analysis of Linear Systems, (solid line) and adopting the Hsu and Bernard (1978)
Fig. 3 NGSMs, lj,uu(t) (j = 0, 1, 2) and bandwidth model, for different value of the time instant, tmax, in
parameter, duu(t), of the transient response of an oscillator which the modulating a(t) function take the maximum
(o0 = 8p rad/sec, x0 = 0.05) in the stationary case value (dashed lines)
Comparison Between Steady-State and a(t) is the normalized to one modulating function
Nonstationary Responses of SDoF Systems proposed by Hsu and Bernard (1978):
Let us consider an oscillator, whose differential
equation governing the motion is written in aðtÞ ¼ a expð1Þ t expða tÞ UðtÞ (95)
canonical form in Eq. 23, forced by the uniformly
modulate Gaussian zero-mean nonstationary pro- which takes its maximum value at time, tmax =
cess F(t), defined as 1/a, and UðtÞ is the unit step function defined as

0, t t0 ;
FðtÞ ¼ aðtÞ N ðtÞ (94) Uðt t0 Þ ¼ (96)
1, t > t0
where N(t) is a Gaussian white noise process, In Fig. 1 the modulating function (95) is
with one-sided PSD GNN (o) = 1 cm2/sec3, and depicted for different value of a = 1/tmax.
a 1.2 b
1 GFF (w,t)
0.8 1.5
ω=π
ω=2π 1
a(t)
0.6
0.5 0
0.4 ω=5π 0
0 10
0.2 ω=8π
10 w
0 t
20 20
0 10 20 30
t
30
Stochastic Analysis of Linear Systems, Fig. 4 Spanos modulating function a(o, t) at different values of abscissa
and Solomos (1983) model of the fully nonstationary o; (b) EPSD function GFF(o, t) = |a(o, t)|2 GNN(o) with
excitation model: (a) sections of the normalized to one GNN(o) = 1 cm2/sec3
In the same figure the EPSD function GFF (o, t) are obtained for the second oscillator for which
= ja(t)j2 GNN (o) of the quasi-stationary tSS 2.39 sec. Moreover, the analysis of the
(separable) Hsu and Bernard (1978) model bandwidth parameters shows that if tmax < tSS
is also shown assuming a = 1/5 and GNN (o) = the parameter duu(t) is narrower than the value
1 cm2/sec3. obtained in the relative stationary case.
In Fig. 2 the NGSM, lj,uu(t) (j = 0, 1, 2),
and bandwidth parameter, duu(t), functions of
the response of an oscillator with natural circu- Comparison Between Uniformly and Fully
lar frequency o0 = 2p rad/sec and damping Nonstationary Responses of SDoF Systems
ratio x0 = 0.05 for the Hsu and Bernard (1978) Let us consider now the Spanos and Solomos
model of the uniformly modulated (1983) model of the fully nonstationary Gaussian
nonstationary excitation are depicted. These zero-mean process. For this model the normal-
NGSMs are compared to the transient NGSMs ized to one evolutionary modulating function can
of the stationary case (aðtÞ ¼ UðtÞ), for different be written as
value of the time instant, tmax. In Fig. 3 the
comparisons are performed for an oscillator
aðo, tÞ ¼ eðoÞ t expðaðoÞ tÞUðtÞ (97)
with natural circular frequency o0 = 8p rad/ S
sec and damping ratio x0 = 0.05. These figures
shows that when the response approaches to its Selecting the following parameters
pffiffiffi
steady-state condition, defined by the steady- eðoÞ ¼ o 2=5 p amax and a(o) = 0.15/2 +
state time tSS 3/(x0 o0), the maximum values e2(o)/4, the normalizing to one coefficient is
of the NGSM functions coincide with the amax = 1.34. The unitary maximum is reached
corresponding values of the NGSM for station- at o = 1.937 p rad/sec and at t = 6.667 sec.
ary input processes. If this condition is not sat- In Fig. 4, the section of the modulating function
isfied, the stationary approximation of input at different values of abscissa o is depicted
process leads to overestimated results. In fact, together with the one-sided EPSD function
for the first oscillator for which the steady-state GFF(o, t) = ja(o, t)j2 GNN(o) with GNN(o) =
time is tSS 9.55 sec, the NGSMs with tmax < 1 cm2/sec3. This figure evidences the frequency
tSS possess maximum values lesser than the dependence of this model of nonstationary input
corresponding stationary case. Similar results process.
0.12 0.7
0.1 0.6
0.5
Re {λ1(t )} [cm2/sec]
0.08
λ0(t ) [cm2]
0.4
0.06
0.3
0.04
0.2
0.02 0.1
0 0
0 10 20 30 0 10 20 30
t t
0.012 4
3.5
0.008
3
Im {λ1(t )} [cm2/sec]
0.004
λ2(t ) [cm2/sec2]
2.5
0 2
1.5
−0.004
1
−0.008
0.5
−0.012 0
0 10 20 30 0 10 20 30
t t
Stochastic Analysis of Linear Systems, Fig. 5 NGSMs of the response of an oscillator with o0 = 2p rad/sec, x0 =
0.05 for the Hsu and Bernard (1978) model (solid line) and for the Spanos and Solomos (1983) model (dashed line)
In Figs. 5 and 6 the NGSMs, lj,uu(t) (j = characteristics of the structural systems. More-
0, 1, 2), of the two oscillators with natural circu- over, in some cases, the quasi-stationary
lar frequency o0 = 2p rad/sec and o0 = 8p modellation of the modulating function can lead
rad/sec and damping ratio x0 = 0.05, obtained to very unconservative results (see Fig. 5) of the
by means of the two modulating functions before stochastic structural response.
described, are depicted and compared. Analyzing In Figs. 7 and 8 the mean frequencies, n+uu(t),
these figures it can be evidenced that the temporal and the normalized time-varying central frequen-
variation of the frequency content of the EPSD cies, oC,uu(t)/o0, defined in Eq. 21, of the
function, often neglected for mathematic conve- response of the two oscillators before analyzed
nience, has substantial effects on the structural for the Hsu and Bernard (1978) and Spanos and
response. In fact the maximum values of the Solomos (1983) modulating functions of the
NGSM functions depend on the form of modu- nonstationary zero-mean Gaussian input process,
lating function as well as the dynamic are depicted.
0.001 0.025
0.0008 0.02
Re {λ1(t )} [cm2/sec]
λ0(t ) [cm2]
0.0006 0.015
0.0004 0.01
0.0002 0.005
0 0
0 10 20 30 0 10 20 30
t
t
0.7
0.0002
0.6
0.5
Im {λ1(t )} [cm2/sec]
0.0001
λ2(t ) [cm2/sec2]
0.4
0
0.3
−0.0001 0.2
0.1
−0.0002
0
0 10 20 30 0 10 20 30
t t
Stochastic Analysis of Linear Systems, Fig. 6 NGSMs of the response of the oscillator with o0 = 8p rad/sec,
x0 = 0.05 for the Hsu and Bernard (1978) model (solid line) and for the Spanos and Solomos (1983) model (dashed line)
Figure 7 shows that for the Hsu and Bernard Summary S

(1978) modulating function, the mean frequen-
cies, n+uu(t), get to an asymptotic value very close The dynamic behavior of structural systems
to the natural frequency of the oscillators. This subjected to uncertain dynamic excitations can
behavior is not verified for the oscillator with be performed through the stochastic analysis,
higher natural circular frequency subjected to which requires the probabilistic characterization
the Spanos and Solomos (1983) model of the of both input and output processes. The charac-
nonstationary input process, where the mean terization of output processes can be extremely
frequency decreases evidencing the frequency complex, when nonstationary and/or non-
dependence of the structural response. Similar Gaussian input processes are involved. However,
results are obtained for the normalized time- in several cases the approximate description of
varying central frequencies, oC,uu(t)/o0, the dynamic structural response based on its spec-
depicted in Fig. 8. tral characteristics may be sufficient.
Stochastic Analysis of a 5 b 6
Linear Systems,
Fig. 7 Mean frequencies 5
of the response of the 4
oscillator for the Hsu and 4
Bernard (1978) model 3
ν+uu(t)
ν+uu(t)
(solid line) and for the 3
Spanos and Solomos (1983) 2
model (dashed line); (a) 2
oscillator with o0 = 2p
rad/sec and x0 = 0.05; 1 1
(b) oscillator with
o0 = 8p rad/sec 0 0
and x0 = 0.05 0 10 20 30 0 10 20 30
t t
a 2.8 b 1.2
2.4 1
2 0.8
ωc,uu(t)/ω0
ωc,uu(t)/ω0
1.6 0.6
1.2 0.4
0.8 0.2
0.4 0
0 10 20 30 0 10 20 30
t t
Stochastic Analysis of Linear Systems, Fig. 8 Time- (1983) model (dashed line); (a) oscillator with o0 = 2p
varying central frequency, normalized by the natural cir- rad/sec and x0 = 0.05; (b) oscillator with o0 = 8p rad/
cular frequency, of the oscillator for the Hsu and Bernard sec and x0 = 0.05
(1978) model (solid line) and for the Spanos and Solomos
In this study a unitary approach to evaluate or nonstationary mono-/multi-correlated zero-

the spectral characteristics of the structural mean Gaussian seismic excitations.
response, to perform the reliability assessment,
of classically damped linear systems subjected
Cross-References
to stationary or nonstationary mono-/multi-
correlated zero-mean Gaussian excitations, is
▶ Probability Density Evolution Method in
described.
Stochastic Dynamics
The main steps of the described approach are
▶ Stochastic Analysis of Nonlinear Systems
(i) the use of modal analysis to decouple the
▶ Stochastic Ground Motion Simulation
equation of motion; (ii) the determination, in
state variable, of the evolutionary frequency
response vector functions and of the evolutionary
power spectral density function matrix of the References
structural response; and (iii) the evaluation of
Borino G, Muscolino G (1986) Mode-superposition
the nongeometric spectral moments as well as
methods in dynamic analysis of classically and
the spectral characteristics of the stochastic non-classically damped linear systems. Earthq Eng
response of linear systems subjected to stationary Struct Dyn 14:705–717
Stochastic Analysis of Nonlinear Systems 3455
Conte JP, Peng B-F (1997) Fully nonstationary analytical

earthquake ground-motion model. J Eng Mech Stochastic Analysis of Nonlinear
(ASCE) 123:15–24
Corotis RB, Vanmarcke EH, Cornell CA (1972) First pas- Systems
sage of nonstationary random processes. J Eng Mech
(ASCE) 98:401–414 Annibale Luigi Materazzi and Marco Breccolotti
Di Paola M (1985) Transient spectral moments of linear Department of Civil and Environmental
systems. SM Arch 10:225–243
Di Paola M, Petrucci G (1990) Spectral moments and Engineering, University of Perugia, Perugia, Italy
pre-envelope covariances of nonseparable processes.
J Appl Mech (ASME) 57:218–224
Fan FG, Ahmadi G (1990) Nonstationary Kanai-Tajimi Synonyms
models for El Cento 1940 and Mexico City 1985
earthquakes. Probab Eng Mech 5:171–181
Harichandran RS, Vanmarcke EH (1986) Stochastic var- Earthquake engineering; Hysteresis;
iation of earthquake ground motion in space and time. Nonlinearity; Random vibrations
J Eng Mech (ASCE) 112:154–174
Hsu T-I, Bernard MC (1978) A random process for earth-
quake simulation. Earthq Eng Struct Dyn 6:347–362
Jennings PC, Housner GW, Tsai C (1969) Simulated Introduction
earthquake motions for design purpose. In: Proceed-
ings of the 4th world conference earth engineering,
The present chapter is devoted to the probabilis-
Santiago, A-1, pp 145–160
Langley RS (1986) Structural response to non-stationary tic analysis of the random response of nonlinear
non-white stochastic ground motion. Earthq Eng structural systems exposed to random excitation
Struct Dyn 14:909–924 with special attention to earthquake action. The
Li J, Chen JB (2009) Stochastic dynamics of structures.
random system response may be due to random
Wiley, Singapore
Lin YK (1976) Probabilistic theory of structural dynam- excitation, to random system properties, to
ics. Krieger, Huntington random boundary conditions. The nonlinear
Lutes LD, Sarkani S (2004) Random vibrations – analysis of character of the response is mainly due to
structural and mechanical vibrations. Elsevier, Boston
the nonlinear materials properties and to the
Michaelov G, Sarkani S, Lutes LD (1999a) Spectral char-
acteristics of nonstationary random processes – a effect of large displacements (the so-called
critical review. Struct Saf 21:223–244 P-D effect).
Michaelov G, Sarkani S, Lutes LD (1999b) Spectral char- The first pioneering studies on this topic took
acteristics of nonstationary random processes –
place in the 1960s and 1970s (VanMarcke
response of a simple oscillator. Struct Saf 21:245–267
Muscolino G (1991) Nonstationary pre-envelope covari- et al. 1970; Iwan 1973; Atalik and Utku 1976;
ances of nonclassically damped systems. J Sound Vib Spanos 1976), when equivalent linearization
149:107–123 techniques were used, taking advantage of the
Muscolino G, Palmeri A (2005) Maximum response sta-
availability of first digital computers.
tistics of MDOF linear structures excited by
non-stationary random processes. Comput Method The attention is limited herein to systems hav- S
Appl Mech Eng 194:1711–1737 ing deterministic properties, including determin-
Priestley MB (1999) Spectral analysis and time series. istic boundary conditions, considering those
Academic, London
types of nonlinearity that can really occur during
Shinozuka M, Sato Y (1967) Simulation of nonstationary
random process. J Eng Mech (ASCE) 93:11–40 earthquakes.
Spanos P, Solomos GP (1983) Markov approximation to After a brief review of the types of nonlinear
transient vibration. J Eng Mech (ASCE) 109:1134–1150 behavior that can be expected, the available
Vanmarcke EH (1972) Properties of spectral moments
methods for the analysis of the response are
with applications to random vibrations. J Eng Mech
(ASCE) 98:425–446 discussed.
Vanmarcke EH (1975) On the distribution of the first- They can be classified into analytical and sim-
passage time for normal stationary random processes. ulation methods. Among the first ones are con-
J Appl Mech (ASME) 42:215–220
sidered in the following: the Fokker-Planck-
Zerva A (1991) Effect of spatial variability and propagation
of seismic ground motions on the response of multiply Kolmogorov (FPK) equation, the equivalent lin-
supported structures. Probab Eng Mech 6:212–221 earization, the perturbation method, and the
3456 Stochastic Analysis of Nonlinear Systems
stochastic averaging. The simulation methods The Masing model is also suitable to represent
based on Monte Carlo techniques are then the friction dissipation which takes place in
presented. masonry structures under earthquakes.
An interesting and powerful nonlinear model
is represented by the Bouc-Wen equation (Bouc
Types of Nonlinearity in Earthquake 1967; Wen 1976).
Engineering The restoring force F is given by
During severe earthquakes most structures F ¼ ak0 x þ ð1 aÞQZ (1)

undergo great amplitude time-varying displace-
ments which can induce inelastic behavior in the where k0 is the initial system stiffness, a is the
structural members. ratio between the post- and the pre-yielding stiff-
In the case of reinforced concrete structures, ness, x is the displacement, Q is the yield strength,
several phenomena take place, like cracking and and Z is a nondimensional parameter which takes
crushing, together with yielding and strain hard- into account the hysteresis, satisfying the follow-
ening of steel. Also the bond between steel and ing nonlinear first-order differential equation:
concrete may be imperfect, as great shear forces
must be exchanged and slippage can occur. dZ
q ¼ Ax_ bjx_jZ jZ jn1 tx_jZjn (2)
Moreover, the reversal of the sign of displace- dt
ments, due to the structural vibration, leads to the
phenomenon of the hysteresis and to the subse- where b, t, A, and n are the dimensionless param-
quent dissipation of energy. eters which control the shape of the hysteresis
This apparently undesirable behavior is delib- loop, q is the yield displacement, and x_ is the
erately taken into account by the modern strate- velocity.
gies for aseismic design of reinforced concrete Typical force-deformation hysteresis loops
structures, based on the concept of the capacity generated using Eqs. 1 and 2 are shown in
design, as it allows the reduction of the inertia Fig. 2. In the examples the dynamical system is
forces acting upon the structures by means of acted upon by a sinusoidal motion having ampli-
energy dissipation. tude 0.1 m and frequency 1.0 Hz. The other
The modeling of the interaction of concrete parameters of the model are q = 0.025 m,
and steel under severe earthquakes was studied b = t = 0.5, k0 = 1, and A = 1. Thus, by changing
in depth in the past (Park et al. 1972, Takeda the different parameters of the Wen’s model,
et al. 1970, Popov et al. 1972). Among the other many hysteretic behaviors can be obtained such
aspects, the hysteretic behavior after steel yield- as the Masing type, the elastic-perfectly plastic,
ing, the stiffness deterioration due to cracking and the elastic plastic with hardening.
of concrete, the strength deterioration and soft-
ening consequent to cumulative severe large
deformations, and the pinching behavior due to Analytical Techniques
shear, along with bond deterioration, were
investigated. The solution of the stochastic differential equa-
Simple hysteretic models that can be used in tions that govern the motion of nonlinear dynam-
the practice are presented in Fig. 1. Model ical systems has been first achieved by means of
(a) represents an elastic-perfectly plastic behav- analytical techniques. Exact and approximate
ior, while model (b) is a variant allowing for solutions can be obtained by these techniques
strain hardening. Model (c) (Masing 1926) takes depending on the complexity of the problem.
into account the stiffness deterioration, using The stochastic response of nonlinear single
smooth curves, and model (d) is the degrading degree of freedom system with elastic behavior
stiffness model due to Takeda. and Gaussian external excitation can be generally
Stochastic Analysis of
Nonlinear Systems,
Fig. 1 (a) Elastic-
perfectly plastic model; (b)
bilinear strain hardening
model; (c) Masing-type
model; (d) degrading
stiffness model due to
Takeda
Stochastic Analysis of Nonlinear Systems, Fig. 2 Different hysteresis loops from the Bouc-Wen equation
solved exactly through the FPK equation. been searched for nonlinear multi-degree of free-
Nonlinear elastic multi-degree of freedom can dom systems and for inelastic systems. In the next
be solved exactly via FPK equation only for paragraphs a brief description of the FPK equa-
very restrictive conditions on the stochastic exter- tion and of some approximate methods is
nal excitation. Thus, approximate solutions have provided.
Fokker-Planck-Kolmogorov Equation subjected to a (scalar) white-noise forcing func-

tion x(t) with second moment rate b.
Introduction By placing y ¼ m dxdt the system may be rewrit-
The theory of stochastic processes began in the ten in the state space as a system of first-order
nineteenth century when physicists were trying to differential equations:
show that heat in a medium is essentially a random
motion of the constituent molecules. At the end of dx y
that century, some researches began to adopt more ¼
dt m (4)
direct mathematical models of random distur- dy @GðxÞ c
¼ y þ xðtÞ
bances instead of considering random motion as dt dx m
due to collisions between objects having a random
distribution of initial positions and velocities. In This is a special case of the system:
this context several physicists, among which Fok-
ker (1914) and Planck (1915), developed partial dx @H
differential equations, which were versions of ¼
dt @y
(5)
what was subsequently called the Fokker-Planck dy @H @H
equation, to study the theory of Brownian motion. ¼ c þ xð t Þ
dt @x @y
The theory of the Fokker-Planck equation
was made considerably more general by Kol-
where the symbol H represents the total energy
mogorov (1931). He assumed the process to be
(potential plus kinetic):
continuous with respect to time and Markovian,
i.e., a process for which the future probability
density conditional on the present and past is 1 2
H ¼ GðxÞ þ y (6)
actually independent of the past. On introducing 2m
further assumptions, Kolmogorov was able
to show that the probability density of the pro- The steady-state probability density p(x,y) for the
cess obeys a partial differential equation of the system of Eq. 5 can be determined by solving the
Fokker-Planck type. He also gave another par- appropriate stationary FPK equation:
tial differential equation which the future prob-
ability density, conditional on the present state,

@ @H @ @H @ @H
obeys with respect to the present state. The p þ p þ c p
@x @y @y @x @y @y
latter equation is called Kolmogorov’s first
equation, and the Fokker-Planck-type equation 1 @2p
þ b 2¼0
is called Kolmogorov’s second equation. Some- 2 @y
times the Fokker-Planck equation is called the (7)
Fokker-Planck-Kolmogorov (FPK) equation.
If p is any function of the energy H, the terms in
Solution of the Fokker-Planck-Kolmogorov the square brackets vanish, and therefore if it can
Equation for an SDOF Elastic Nonlinear Second- be found a particular function p(H) which makes
Order System the remaining two terms vanish, p(H) is a solution
Let us consider the nonlinear nonconservative of Eq. 7.
stochastic system By substituting p(x, y) = p(H(x, y)) and assum-
ing null the terms in the square bracket, Eq. 7
d2 x dx @GðxÞ becomes
m 2
þc þ ¼ xðtÞ (3)
dt dt dx

@ @H 1 @2
that describes the displacement x of a body with c pð H Þ þ b 2 pð H Þ ¼ 0 (8)
mass m, damping c, and potential energy G(x) @y @y 2 @y
Nonlinear Systems,
Fig. 3 Parameters
describing a continuous
piecewise-linear restoring
force
The integration with respect to y yields used, for instance, to determine the response of
cracked reinforced concrete beams with cracks
@H 1 @ that alternatively open and close (Breccolotti
c p ð H Þ þ b p ð H Þ ¼ Lð x Þ (9)
@y 2 @y et al. 2008).
In this case the motion of the vibrating system
where L(x) is an arbitrary function. can be described by the equation
As boundary conditions for the stationary FPK
equation can be assumed, that p is zero at infinity.
x€ þ b1 x_ þ Fi ðxÞ ¼ f ðtÞ (13)
Thus, the left side of Eq. 9 vanishes when y ! 1
and hence L(x) = 0. Equation 9 simplifies to
where Fi(x) = o2i (x x0i) for xi x xi + 1,
@H 1 @p @H i = 1, 2, . . . (see Fig. 3), oi and x0i are positive
c pðH Þ þ b ¼0 (10)
@y 2 @H @y constants, and f(t) is a Gaussian stochastic pro-
cess having null mean and uniform power spec-
Since @H@y cannot be identically null (otherwise
trum S0 over the entire frequency domain.
H would be independent of y and hence so will The following boundary conditions have to be
be p), then Eq. 10 yields satisfied:
@p 2c S
þ p¼0 (11) _ xÞ ¼ wiþ1 ðxiþ1 ,x,€
wi ðxiþ1 ,x,€ _ xÞ (14)
@H b
whose general solution representing the steady- X ð ðð

state probability density is _ xÞ dx dx_ d€
wi ðx,x,€ x¼1 (15)
i
x x_ x€

2c
pðx, yÞ ¼ Cexp Hðx, yÞ (12)
b
o2i ðxiþ1 x0i Þ ¼ o2iþ1 ðxiþ1 x0iþ1 Þ (16)
being C a normalizing constant.
For the case of an SDOF system with unit that represent the continuity of the probability
mass and continuous linear piecewise restoring density function of the response, the normalizing
force, a closed form solution of the FPK equa- condition, and the continuity of the piecewise-
tion can, thus, be determined. This model can be linear restoring force characteristics, respectively.
It can be demonstrated that the solution of the wðx1 , x2 , . . . , x_1 , x_2 , . . .Þ

FPK equation corresponding to this system, " !#
1 1X n 2
representing the stationary probability density ¼ Cexp mi x_ þ U (20)
function (PDF) of the response, can be written k 2 i¼1
i
in the form
This form of the probability density function is

o2
2 b1 2
2x0i o2 known as the Maxwell-Boltzmann distribution,
S0 b1 x þS0 x_ S0 b1 x
i i
wi ðx,x_Þ ¼ Ci e being the expression inside the parentheses the
o2 2x0i o2 b
total mechanical energy of the system.
i b x2 þ ib
1x S1 x_2
¼ Ci e S0 1 S0
e 0
¼ wi, x ðxÞ wx_ðx_Þ (17) Equivalent Linearization

The very restrictive conditions that make avail-
It can be noted, as demonstrated by Lin and Cai able exact solutions for the stochastic dynamical
(1995), that for an SDOF system having systems motivated the development of approxi-
nonlinear stiffness, linear damping, and exposed mate solution techniques. Methods such as the
to a Gaussian white-noise excitation, the station- equivalent linearization have been developed to
ary displacement (x) and the velocity ðx_Þ are generate first-order approximate solutions. These
independent random variables. techniques can also be applied in some cases to
single-degree-of-freedom nonlinear oscillator
Solution of the Fokker-Planck-Kolmogorov with hysteretic behavior.
Equation for an nDOF Elastic Nonlinear The adaptation of the classical equivalent lin-
Second-Order System earization technique of the deterministic theory
Several solutions of the FPK equation for to systems subjected to random excitations was
nonlinear elastic systems with n degrees of free- independently developed by Booton (1953) and
dom (nDOF) subjected to random excitation have Caughey (1953). Later on other researchers have
been developed in the past (Piszczec and Nizioł extended the method to encompass approximate
1986). Let us consider the following set of solutions of the stationary random response of
n equations for an nDOF system, characterized multi-degree-of-freedom nonlinear oscillators.
by elastic nonlinear properties: In this case the mathematical equation that
describes the response of the system is
1 @U
x€i þ bi x_i þ ¼ f i ðt Þ (18) x þ Cx_ þ Kx þ f x, x_ ¼ G
M€ (21)
mi @xi
where fi(t) are independent and uncorrelated where M, C, and K are the constant mass,
white noises with null mean and spectral density damping, and stiffness
square matrices with
Si and bi are positive constants. dimension n, f x, x_ is an n-vector function of
The term U(x1, x2, . . .) represents the potential the dependent variable x and its derivative, and
energy.
This system of differential equations can be G T ¼ ð g1 , . . . , gn Þ (22)
solved assuming the validity of the following
with n stationary Gaussian processes gi.
conditions:
The principle of the method is the replacement
of the nonlinear dynamical system described by
Si mi Eq. 21 by an auxiliary linear system for which the
¼k i ¼ 1, 2, . . . (19)
2bi exact analytic solution is known. The replace-
ment is made so as to be optimum with respect
being k a positive constant. In this case the solu- to some measure of the difference between the
tion of the FPK equation is original and the auxiliary system.
Nonlinear Systems,
Fig. 4 Bilinear hysteretic
restoring force
This optimum auxiliary linear system, which 5. Solution of the equivalent linear system to
will be called equivalent, is defined as obtain equations for the specification of the
identification parameters of x.
x þ ðC þ Ce Þx_ þ ðK þ K e Þx ¼ G
M€ (23)
The application of the method can be conve-
where Ce and Ke are t-independent matrices. niently shown by analyzing the hysteretic sto-
These matrices must be such that the difference chastic dynamical single DOF system described
d between systems of Eqs. 21 and 23, defined by by the following differential equation (Iwan and
Lutes 1968):
d ¼ f x, x_ Ce x_ K e x; (24)
nðtÞ
x€ þ 2b0 o0 x_ þ o20 ’ðxÞ ¼ (25)
is minimized for every x which belongs to the m
class of solutions of the system (23).
Since the excitation G of the linear system is where m denotes the mass, o0 is the undamped
Gaussian, it is well known that the response x will natural circular frequency for small-amplitude
be Gaussian as well. Therefore, the matrices Ce and vibrations, b0 is the fraction of critical damping
Ke must be such that the difference d is minimized for small-amplitude vibrations, and ’(x) is the S
for every stationary Gaussian random vector x. bilinear hysteretic restoring force, shown in
The equivalent linearization technique is thus Fig. 4, having a unit slope for small amplitudes
composed of the following steps: and a second slope a. The excitation n(t) is a
stationary random function with a uniform
1. Identification of a class of approximate solu- power spectral density S0 and a Gaussian proba-
tion functions and the parameters defining bility distribution.
each member of the class. The equivalent linear system
2. Selection of the norm of the difference
vector d. nðtÞ
3. Selection of the averaging operator G. x€ þ 2beq oeq x_ þ o2eq x ¼ (26)
m
4. Determination of the matrices Ce and Ke of
the equivalent linear system in terms of the may be identified by choosing the parameters oeq
identification parameters of x. and beq so as to minimize the mean-squared
difference between the nonlinear inelastic equa- Perturbation Method

tion of motion (25) and the linear equation (26). The random response of slightly nonlinear vibrat-
For the case of bilinear hysteretic oscillator ing systems can also be obtained by applying the
with “small” nonlinearity, a solution for the classical perturbation method (Crandall 1963).
parameters oeq and beq has been found by The method is based on the assumption that the
Caughey (1960) under the following assumptions: nonlinearity is small enough to allow the solution
of the stochastic differential equation of motion
1. The response of the nonlinear system is to be expressed as a power series. If the following
contained within a narrow frequency band. SDOF system is considered
2. The probability density of the amplitude of
this narrowband response follows the Ray- x€ þ 2Bo0 x_ þ o20 x þ eðx,x_Þ f ðtÞ ¼ 0 (31)
leigh distribution.
where is a nonlinear function and e is a suffi-
These assumptions lead to the following ciently small parameter, the solution can be
expressions for oeq and beq: expressed as a power series in e:
2

oeq 8ð 1 a Þ xðtÞ ¼ x0 ðtÞ þ ex1 ðtÞ þ e2 x2 ðtÞ þ . . . (32)
¼1
o0 p
ð1 (27) Substituting Eq. 32 into Eq. 31 and grouping the
1 z2
23 þ ðz 1Þ2 e l dz
1
different terms having the same power of e, a set
lz
1
of linear equation in x0, x1, x2, . . . can be obtained.
and If the external excitation f(t) is weakly stationary,
the steady-state response to Eq. 31 can be
2 1 constructed from the following solutions of the
o0 o0 ð 1 aÞ
beq ¼ b0 þ pffiffiffiffiffiffi erfc l2 set of linear equations:
oeq oeq pl
(28) ð1
x0 ðtÞ ¼ f ðt tÞhðtÞdt
where ð1
1
x1 ðtÞ ¼ ½x0 ðtÞ, x_0 ðtÞhðtÞdt (33)
2s2x ð1
1
l¼ 2 x2 ðtÞ ¼ ½x1 ðtÞ, x_1 ðtÞhðtÞdt
Y
... 1
In general Eq. 27 must be evaluated numerically.

However, when l 1, the asymptotic expansion being h the impulse response function
shown in Eq. 29 can be used: corresponding to e = 0.
2

oeq 8ð1 aÞ Stochastic Averaging
¼a þ The method of stochastic averaging, firstly intro-
o0 p
duced by Landau and Stratonovich (1962), has
0:6043l4 0:2451l4 0:1295l4
3 5 7
proved to be a very useful tool for deriving

(29) approximate solutions to problems involving the
dynamical response of lightly damped systems to
After finding oeq and beq, the RMS levels of broadband random excitation. It is based on the
response of the equivalent linear system can be principle that the rate of change with respect to
obtained from time of the oscillator’s total energy is equal to the
power input due to random excitation, minus the
pS0
s2x_ ¼ o2eq s2x ¼ 2
(30) power dissipation due to the damping
4m beq oeq mechanism.
In lightly damped structures the slowly vary- It has been shown that the stochastic averaging
ing energy can be treated as a constant over an method is applicable to an oscillator with bilinear
appropriate period of oscillation, and oscillatory restoring force-displacement characteristic when
terms can be approximated by their time averages the energy dissipation due to hysteresis is rela-
over one period of oscillation. Furthermore, tively low. In this case, after a preliminary aver-
under broadband random excitation, the relaxa- aging operation, the equation of motion could be
tion time of the oscillator response is much cast into the standard form of Eq. 34, enabling the
greater than the correlation time of the excitation. stationary response distribution to be determined
Thus, it is possible to model the power input due together with statistics such as the standard devi-
to the excitation as a nonzero mean component ation of the displacement and the average
plus an additional, fluctuating component with yielding rate.
the character of white noise. Nevertheless, the standard stochastic averag-
Thanks to these properties, the method can ing technique cannot be used for examining the
also be applied to strongly nonlinear stiffness effect of strongly nonlinear restoring forces
oscillators and to certain nonlinearities of the since, to O(e2), this effect vanishes. In these
hysteretic kind. cases it is possible to combine the equivalent
The essence of the standard stochastic averag- linearization method with stochastic averaging
ing method is embodied in a limit theorem due to and treat the equivalent frequency as amplitude
Stratonovich (1963) and Khasminskii (1966): dependent.
An alternative approach to the analysis of
X_ ¼ e2 f ðX, tÞ þ egðX, t, Y ðtÞÞ (34) nonlinear oscillators based on a consideration of
the energy envelope has been developed by Rob-
where X(t) is an n-vector stochastic process, usu- erts and Spanos (1986).
ally representing the response, and Y(t) is an Consider an oscillator with the following
m-vector stochastic excitation process. If the ele- equation of motion:
ments of Y(t) are broadband processes, with zero
means, and the vectors f and g satisfy certain x€ þ e2 hðx,x_Þ þ GðxÞ ¼ ezðtÞ (36)
requirements (which are almost invariably met
in practice), then it can be shown that X(t) may
with e assumed small enough to ensure that x(t) is
be uniformly approximated over a time interval
O(e0). The symbol G(x) denotes an arbitrary
of order O(e1) by an n-dimensional Markov
nonlinear stiffness term, while the energy enve-
process, which satisfies the Ito equation:
lope V(t) may be defined as the sum of the kinetic
and the potential energy:
dX ¼ e2 mðXÞdt þ esðXÞdW (35)
x_2
S
The symbol W(t) denotes an n-vector of indepen- V ðtÞ ¼ þ U ðx Þ (37)
dent Wiener (or Brownian) processes, with unit 2
variance, and m and s are, respectively, the “drift
vector” and the “diffusion matrix.” and
If e is small, then the elements of X(t) must be ðx
slowly varying, with respect to time. The equa- U ðxÞ ¼ GðxÞdx (38)
tions of motion which occur in random vibration 0
problems can be written in state space form,

involving displacement and velocity response For the special case where the damping is linear
variables. However, these variables are usually e2 hðx,x_Þ ¼ 2Bo0 x_ and the nonlinear stiffness has
rapidly fluctuating with respect to time. Evidently the power-law form
a transformation of variables is required to cast
the equations of motion into the form of Eq. 34. GðxÞ ¼ kjxju sgnðxÞ; (39)
it is possible to obtain analytical solution for the white-noise processes, and combined normal
transition density function p(V, t|V1, t1) of V(t). and Poisson white noise (Pirrotta and Santoro
By introducing the nondimensional energy 2011).
variable The PI method is based on the fact that the
state space vector, Yt say, obtained as a solution
V ðtÞ of a stochastic differential equation is a Markov
wð t Þ ¼ ; (40)
gksuþ1 vector process. This makes it possible to use
a time stepping procedure to obtain the joint
the following transition density p(w, tjw1, t1) = probability density function p(y, t) of Yt as a
p(w, tjw1) can be found: function of time t by exploiting the fundamental
equation:
r2
pðw, tjw1 Þ ¼
1 w ð1
ð 1 qÞ w1 q pðy, tÞ ¼ pðy, tjy0 , t0 Þpðy0 , t0 Þdy0 (44)
pffiffiffiffiffiffiffiffiffiffi 1
w þ qw1 2 ww1 q
exp Ip
1q 1q
where p(y, tjy0 , t0 ) denotes the conditional proba-
(41) bility density function of Yt given that Y t0 ¼ y0 :
For small time increments Dt = t t0 , p(y, tjy0 , t0 )
being will be referred to as the short-time transition
probability density function. It can be demon-
1 strated that for a numerical solution of a stochas-
r¼ 1 (42)
a tic differential equation, the short-time transition
probability density function can always be given
q ¼ e2aBo0 t (43) as an analytical, closed-form expression. Hence,
if an initial probability density function,
and with Ip(.) as the modified r – order Bessel p0(y) = p(y, t = 0), is given, then Eq. 44 can be
function of the first kind. invoked repeatedly to produce the time evolution
of p(y, t). If the stochastic differential equation
Path Integral Solution has an invariant measure there exists a stationary
The FPK equation that describes the evolution of probability density function ps(y), then, eventu-
the response’s probability density (PD) of a ally, assuming that p0(y) 6¼ ps(y), p(y, t) will
nonlinear system excited by an external white approach this stationary probability density func-
noise can be solved numerically by path integral tion. The number of times Eq. 44 has to be
(PI) solution procedures. In essence the PI repeatedly used to reach the stationary situation
method is a stepwise calculation of the joint depends, of course, on the dynamic system and on
probability density function of a set of state the specified initial probability density function
space variables describing a white-noise-excited p0(y).
nonlinear dynamic system. Among the first
efforts to develop the PI method into numerical
tools are those of Wehner and Wolfer (1983), Sun Monte Carlo Simulation Methods
and Hsu (1990), and Naess and Johnsen (1993).
The PI method has been proved to provide Generality
extremely accurate results for the tail behavior The probabilistic structure and the statistical
of the joint probability density function of the moments of the response of any type of nonlinear
state space vector and thus for the estimation of mechanical systems can be evaluated using sim-
extreme responses of nonlinear dynamical sys- ulation techniques, like the Monte Carlo method.
tems excited by forces, external or parametric, This method operates in the time domain by
that can be approximated as white noise, filtered repeating a great number of times deterministic
analyses, each one consisting in a step-by-step Univariate One-Dimensional Stochastic

nonlinear analysis of the structure subjected to Processes
an earthquake record. The basic method to generate a realization of a
In order to evaluate properly the statistics of univariate, one-dimensional (1V-1D) stationary
the response, each earthquake record must belong stochastic process f0(t) with zero mean and
to the same “family,” in the sense that all the one-sided PSD function Gf 0 ðoÞ was proposed
records must be compatible with the same by Shinozuka (Shinozuka 1972).
power spectral density (PSD) function. The process may be simulated by the series
The major advantage of Monte Carlo simula-
tion is that it can deal with almost any type of pffiffiffiX
N 1
f ðtÞ ¼ 2 An cos ðon t þ Fn Þ (45)
nonlinearity, maybe using the same commer- n¼0
cially available packages that are used for the
deterministic analyses. Its only disadvantage is where
that it is time-consuming, as many analyses are qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
requested in order to have reliable estimates of An ¼ Gf 0 ðon ÞDo, n ¼ 0, 1, 2, . . . N 1
the statistical properties of the response.
(46)
The number of needed runs may be many
thousands, even if 300–400 analyses can give on ¼ nDo; (47)
acceptable results in practical applications. ou
Probably the most important and delicate part Do ¼ (48)
N
of Monte Carlo simulation is the generation of
realizations of the stochastic process or field that ou is the upper cutoff frequency and the phase
represents the earthquake. angle Fn is the realization of a random variable
In principle the earthquakes may be modeled uniformly distributed over the interval [0, 2p].
as nonstationary non-Gaussian random fields, Shinozuka and Deodatis (1991) showed that it
even if the null-mean Gaussian approximation is must be
considered acceptable by most authors.
If the analyzed system is a building, it is A0 ¼ 0 and Sf 0 ðo0 ¼ 0Þ ¼ 0 (49)
reasonable, due to the reduced area of ground
that it occupies, to model the process as to ensure that the realization of the process is
one-dimensional. If the hypothesis that the earth- ergodic.
quake induces ground acceleration in one direc-
tion only may be accepted, the process is Univariate Multidimensional Stochastic
univariate. Processes
In the case of structures that have a notable For the sake of simplicity, let us consider the S
extension over the ground, like long span bridges, special case of a univariate two-dimensional sto-
the earthquake input is not the same in different chastic process (1V-2D), which may represent
part of the system and the process must be the random field of the one-direction ground
considered as multidimensional. If only acceleration interesting a great-extension struc-
one-direction component of the ground accelera- ture frozen in time.
tion is considered, it is also univariate. Otherwise The process f0(x1, x2), having cross-PSD
it is multivariate. Gf 0 ðk1n1 , k2n2 Þ, can be simulated by the series
pffiffiffiNX X
1 1 N 2 1 h i
f ðx1 , x2 Þ ¼ 2 An1 n2 cos k1n1 x1 þ k2n2 x2 þ Fðn11Þn2 þ A~n1 n2 cos k1n1 x1 k2n2 x2 þ Fðn21Þn2
n1 ¼0 n2 ¼0
(50)
where
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Gf 0 ðk1 , k2 Þ, A~n1 n2 ¼ Gf 0 ðk1n1 , k2n2 ÞDk1 Dk2
(51)
k1n1 ¼ n1 Dk1 , k2n2 ¼ n2 Dk2 ; (52)
k1u k2u
Dk1 ¼ , Dk2 ¼ (53)
N1 N2
n1 ¼ 0, 1, 2, . . . N 1 1, n2 ¼ 0, 1, 2, . . . N 2 1
(54) Stochastic Analysis of Nonlinear Systems,
Fig. 5 Kanai-Tajimi power spectrum
and
where G0 is the constant value of the spectrum of
A0n2 ¼ An1 0 for n1 ¼ 0, 1, 2, . . . N 1 1 and the white-noise process and og and xg are the
n2 ¼ 0, 1, 2, . . . N 2 1 predominant ground frequency and the ground
damping.
(55)
The values of og = 4p and xg = 0.60 have
been suggested as representative of earthquakes
A~0n2 ¼ A~n1 0 for n1 ¼ 0, 1, 2, . . . N 1 1 and on firm ground.
n2 ¼ 0, 1, 2, . . . N 2 1 The typical aspect of the Kanai-Tajimi PSD is
(56) shown in Fig. 5.
Anyway in most cases it is better to derive the
k1u and k2u are the upper cutoff wave numbers PSD from the response spectrum.
and the phase angles Fðn11Þn2 and Fðn21Þn2 are the For this purpose let us consider the following
realization of two statistically independent ran- SDOF deterministic dynamical system:
dom variables uniformly distributed over the x€ðtÞ þ 2xox_ðtÞ þ oxðtÞ ¼ f 0 ðtÞ (58)
interval [0, 2p].
The extension to the case of 1D-3V process acted upon by a 1V-1D stationary stochastic pro-
f0(x1, x2, t), including also the time variability, is cess f0(t) which represents the ground accelera-
straightforward. tion. If the process is Gaussian, its probabilistic
structure is fully represented by its PSD Gf 0 .
Earthquake Power Spectra The random dynamic analysis of the system in
A well-known PSD of the ground motion avail- the frequency domain allows finding the PSD of
able in literature is the so-called Kanai-Tajimi the response x(t), which is the system
spectrum. It is based on the hypothesis that the displacement:
ground acceleration during earthquakes may be
considered as a filtered band-limited white-noise Gx ðoÞ ¼ Gf 0 ðoÞjH ðoÞj2 (59)
process expressed by the following function:
The variance of the response is represented by the
h 2 i area under the PSD function:
1 þ 4x2g o=og
GðoÞ ¼ h 2 i2 2 G0 ð
1
1 o=og þ 4x2g o=og
s2x ¼ Gf 0 ðoÞjH ðoÞj2 do (60)
(57) 0
where jH(o)j2 is the squared frequency-domain A measure of the spread of the PSD function is
transfer function between loading and system
response: qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
d¼ 1 l21 =l0 l2 (66)
1
jH ðoÞj2 ¼ 2 (61) Given the above, an approximate expression for
o2n o2 þ 4x2 o2n o2
the peak factor rs;p is given by
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
n h
with on the natural frequency of the system. pffiffiffiffiffiffiffiffiffiffiffiffiffiffi io
The variance of the pseudo-acceleration r s;p ¼ 2log 2n 1 exp dx€ðsÞ plog2n
response is
(67)
ð
1
where
s2x€ ¼ o4n s2x ¼ o4n Gf 0 ðoÞjH ðoÞj2 do

0
n ¼ Ox€ðsÞs0 =2p
(68)
Gf 0 ðon Þon
p
1 (62) logp
4x

s0 ¼ s exp 2 sx€2 ðsÞ=s2x€ðs=2Þ 1 (69)
The ordinates x€s;p of the acceleration response
spectrum are the maxima of the response x€ðtÞ of
an oscillator whose frequency o varies from 0 to Ox€ðsÞ and dx€ðsÞ are the spectral moments of the
some stipulated maximum value. The parameter response.
“s” is the duration of the strong motion and “p” is Then the PSD function is
the non-exceedance probability of the maxima.
The solution of the equation 1
G f 0 ðo n Þ
p
x€s;p ¼ r s;p sx€ (63) on 1
4xs
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (70)
u ðn
uo 2 S ð o Þ 2 o
implies the evaluation of the peak factor rs;p by u n v n s, p
t Gf 0 ðoÞdo
solving the corresponding “first-passage” r 2s, p
0
problem.
First of all, it must be reminded that the
moment of order i of the generic one-sided PSD where xs is the time-dependent damping for the
S(o) is duration s S
and Sv is the velocity response spectrum.
ð
1 As the PSD Gf 0 ðoÞ appears on both sides of
li ¼ oi SðoÞdo (64) Eq. 65, an iterative computational procedure
must be used.
0
while a measure where the spectral mass is con-

Summary
centrated along the frequency axis is
pffiffiffiffiffiffiffiffiffiffiffiffi The present chapter is devoted to the probabilistic

O¼ l2 =l0 (65) analysis of the random response of nonlinear
structural systems exposed to the random excita-
parameter that resembles the root mean square of tion due to earthquakes. The system response is
a random variable. nonlinear due to the nonlinear character of the
material properties and to the effects of large Fokker A (1914) The median energy of rotating electrical
structural displacements. dipoles in radiation fields. Annalen Der Physik
43:810–820
The attention is focused on systems having Iwan WD (1973) A generalization of the concept of equiv-
deterministic properties, including deterministic alent linearization. Int J Nonlinear Mech 8:279–287
boundary conditions, taking into account those Iwan WD, Lutes LD (1968) Response of the bilinear
types of nonlinearity that can really occur during hysteretic system to stationary random excitation.
J Acoust Soc Am 48(3):545–552
earthquakes. Khasminskii R (1966) A limit theorem for the solution of
After a brief review of the types of nonlinear differential equations with random right-hand sides.
behaviors that can be expected, some available Theory Probab Appl 11:390–405
methods for the analysis of the response are Kolmogorov A (1931) Ober analytische Methoden
in der Wahrscheinlichkeitsrechnung. Math Ann
discussed. 104:415–458
They can be classified into analytical and sim- Landau P, Stratonovich R (1962) Theory of stochastic
ulation methods. Among the first ones are transitions of various systems between different states.
presented: the Fokker-Planck-Kolmogorov Moscow University, Vestinik
Lin Y, Cai G (1995) Probabilistic structural dynamics
(FPK) equation, the equivalent linearization, the advanced theory and applications. McGraw-Hill,
perturbation method, and the stochastic averag- New York
ing. The simulation methods based on Monte Masing G (1926) Eigenspannungen und Verfestigung
Carlo techniques are then presented. beim Messing (in German). In Proceedings of the
2nd international congress of applied mechanics,
Zurich, pp 332–335
Naess A, Johnsen J (1993) Response statistics of
nonlinear, compliant offshore structures by the path
Cross-References integral solution method. Probab Eng Mech 8(2):
91–106
▶ Plastic Hinge and Plastic Zone Seismic Park R, Kent D, Sampson R (1972) Reinforced concrete
Analysis of Frames members with cyclic loading. ASCE J Struct Div
98(7):1341–1360
▶ Reinforced Concrete Structures in Earthquake-
Pirrotta A, Santoro R (2011) Probabilistic response of
Resistant Construction nonlinear systems under combined normal and Poisson
▶ Stochastic Analysis of Linear Systems white noise via path integral method. Prob Eng Mech
26(1):26–32
Piszczec K, Nizioł J (1986) Random vibration of mechan-
ical system. Ellis Horwood, Chichester
References Planck M (1915) Sgr. preuss. Akad. Wiss. p 512
Popov E, Bertero V, Krawinkler H (1972) Cyclic behavior
Atalik T, Utku S (1976) Stochastic linearization of multi- of three r.c. flexural members with high shear. In:
degree of freedom nonlinear systems. Earthquake Eng EERC report 72-5. Earthquake Engineering Research
Struct Dyn 4(4):411–420 Center, University of California, Berkeley
Booton R (1953) The analysis of nonlinear control sys- Roberts J, Spanos P (1986) Stochastic averaging: an
tems with random inputs. Polytechnic Inst, Brooklyn approximate method of solving random vibration
Bouc R (1967) Forced vibrations of mechanical systems problems. Int J Nonlinear Mech 21(2):111–134
with hysteresis. Proceedings of the Fourth Conference Shinozuka M (1972) Monte Carlo solution of structural
on Nonlinear Oscillation, Prague dynamics. Comput Struct 2(5/6):855–874
Breccolotti M, Materazzi A, Venanzi I (2008) Identifica- Shinozuka M, Deodatis G (1991) Simulation of stochastic
tion of the nonlinear behavior of a cracked RC beam processes by spectral representation. Appl Mech Rev
through the statistical analysis of the dynamic 44(4):191–204
response. Struct Control Health Monitor 15:416–435 Spanos P-TD (1976) Linearization techniques for
Caughey T (1953) Response of nonlinear systems to ran- nonlinear dynamical systems, EERL 76-04. California
dom excitation. Lecture Notes California Institute of Institute of Technology, Pasadena, Calif
Technology, Pasadena, Calif Stratonovich R (1963) Topics in the theory of random
Caughey T (1960) Random excitation of a system with noise. Gordon & Breach, New York
bilinear hysteresis. J Appl Mech 27:649–652 Sun J, Hsu C (1990) The generalized cell mapping method
Crandall S (1963) Perturbation techniques for random in nonlinear random vibration based upon short-
vibration of nonlinear systems. J Acoust Soc Am time Gaussian approximation. J Appl Mech 57:
35(11):1700–1705 1018–1025
Stochastic Finite Elements 3469
Takeda T, Sozen M, Nielson N (1970) Reinforced con- (Shinozuka 1971) of wide-sense homogeneous
crete response to simulated earthquakes. Proc ASCE random fields.
J Struct Div 96(ST12):2257–2573
VanMarcke E, Yanev P, De Estrada M (1970) Response of The expression SFEM came in use in the early
simple hysteretic systems to random excitation, s.l., 1980s (Contreras 1980; Baecher and Ingra 1981).
Research report R70-66, Department of Civil Engi- Der Kiureghian and Ke (1988) defined SFEM as
neering, MIT “a finite element method which accounts for
Wehner M, Wolfer W (1983) Numerical evaluation of
path-integral solutions to Fokker–Planck equations. uncertainties in the geometry or material proper-
Phys Rev A 27(5):2663–2670 ties of a structure, as well as the applied loads”
Wen Y (1976) Method for random vibration of hysteretic where “the uncertainties are usually spatially dis-
systems. ASCE J Eng Mech 120:2299–2325 tributed over the region of the structure and
should be modelled as random or stochastic
fields.” The distinguishing feature of an SFEM
is that it involves the discretization of the random
Stochastic Finite Elements field and the computation of solution statistics.
Discretization methods for random fields
Carsten Proppe replace the random field by a finite set of random
Institut f€
ur Technische Mechanik, Karlsruhe variables. They can be broadly classified into two
Institute of Technology, Karlsruhe, Germany groups: direct discretization schemes and series
expansion techniques. Direct discretization
schemes either assign the values of the random
Introduction field at a given set of nodes to the finite set of
random variables (point discretization schemes)
The finite element method (FEM) has become the or compute the values from local averages of the
dominant computational method in structural random field over a spatial domain (local averag-
engineering. In general, the input parameters in ing schemes).
the standard FEM assume deterministic values. Possible sets of nodes for point discretization
In earthquake engineering, at least the excitation schemes are the nodes of the finite element mesh
is often random. However, considerable uncer- (Hisada and Nakagiri 1981; Liu et al. 1986), the
tainties might be involved not only in the excita- centroids of the finite elements (Der Kiureghian
tion of a structure but also in its material and and Ke 1988), or the Gauss points used by the
geometric properties. A rational treatment of integration rules of the finite elements (Brenner
these uncertainties needs a mathematical concept and Bucher 1995). If the set of nodes for the point
similar to that underlying the standard FEM. discretization scheme of the random field corre-
Thus, FEM as a numerical method for solving sponds to the set of nodes of the finite elements,
boundary value problems has to be extended to the shape functions for the representation of the S
stochastic boundary value problems. The exten- random field may differ from those for the dis-
sion of the FEM to stochastic boundary value placements (Liu et al. 1986). Point discretization
problems is called stochastic finite element schemes yield a positive definite correlation
method (SFEM). matrix that is easily computed. The distribution
The first developments of the SFEM can be function of the random variables is the same as
traced back at least to Cornell (1970), who stud- for the random field. Nearly all schemes can be
ied soil settlement problems, and to Shinozuka applied to Gaussian as well as non-Gaussian ran-
(Astill et al. 1972), who combined FEM with dom fields. The point discretization schemes that
Monte Carlo simulation for reliability analysis do not use different shape functions have the
of structures with random excitation, random disadvantage that the FE mesh depends on the
material properties, or random geometric proper- correlation structure of the random field and that
ties. He introduced random fields and discretized the shape and size of all these elements should be
them based on spectral representation theory the same. In general, the discretization produces a
3470 Stochastic Finite Elements
huge number of random variables and leads to However, according to Stefanou and
inefficient numerical procedures. Papadrakakis (2007), homogeneity and ergodic-
In contrast, local averaging schemes may yield ity of sample functions generated by the KL
accurate results even on rather coarse FE meshes. expansion are questionable. The spectral repre-
They can be based directly on averages of the sentation method approximates a homogeneous
random parameter field (Vanmarcke and Gaussian random field by a superposition of
Grigoriu 1984) or on integrals computed over harmonics with fixed frequencies and random
the domain of a finite element, e.g., for the deter- amplitudes or random frequencies and ampli-
mination of the element stiffness matrix tudes. Due to the central limit theorem, the gen-
(Deodatis 1991). However, direct averaging erated samples are Gaussian, if a sufficient
yields random variables whose distribution func- number of harmonics (at least 128, according
tions are difficult to obtain (except for Gaussian to Stefanou (2009)) are superposed.
random fields) and the approximation of In some cases, non-Gaussian random fields
non-rectangular elements may lead to a can be represented by nonlinear memoryless
nonpositive definite covariance matrix (Matthies transformations of Gaussian random fields
et al. 1997), and averaging based on element (Yamazaki and Shinozuka 1988). The
integration introduces again a dependence of the memoryless transformation introduces a compat-
FE mesh on the correlation structure. ibility condition between the marginal distribu-
Series expansion techniques represent the tion function and the autocorrelation function of
random field by a series involving countably the non-Gaussian random field. If this condition
infinite random variables and a complete set of is not satisfied, only an approximation of the
deterministic functions. The discretization is non-Gaussian random field by a nonlinearly
obtained by truncation of the series expansion. transformed Gaussian random field is possible.
Lawrence (Lawrence 1987) considers a series Iterative procedures have been proposed
expansion with an a priori set of orthogonal (Deodatis and Micaletti 2001; Yamazaki and
deterministic base functions that are multiplied Shinozuka 1988) that calibrate the power spectral
by random variables. In the Karhunen-Loève density of the Gaussian random field in order to
(KL) expansion, the set of orthogonal base func- approximately match the marginal distribution
tions is obtained as eigenfunctions of a homoge- function and the autocorrelation function of the
neous Fredholm integral equation of second kind non-Gaussian random field. The underlying
that involves the covariance kernel of the ran- Gaussian random field can then be discretized
dom field. The KL expansion is optimal in the by any of the methods described above.
sense that it reduces the mean-square error A series expansion of a non-Gaussian random
resulting of its truncation. In principle, it can field in terms of independent random variables is
be applied to homogeneous as well as inhomo- obtained by projection on a set of orthogonal
geneous random fields and to Gaussian as well as polynomials. This method is known as polyno-
non-Gaussian random fields. However, in prac- mial chaos expansion and is discussed exten-
tical cases, the KL expansion is applied to sively in Ghanem and Spanos (1991), where
Gaussian random fields almost exclusively Hermite polynomials and Gaussian random vari-
because the random variables in the expansion ables are considered. In Xiu and Karniadakis
are then independent and standard normally dis- (2002), other families of orthogonal polynomials
tributed. In general, the eigenfunctions have and corresponding distributions of the random
to be computed numerically and the accuracy variables are discussed. The polynomial chaos
of the numerical solution influences the accu- expansion is convergent in mean-square sense.
racy of the random field representation. For Field and Grigoriu (2004) critically discuss poly-
strongly correlated random fields with smooth nomial chaos expansions and show that the accu-
covariance kernel, few terms of the series expan- racy of the PC approximation is not always
sion are sufficient to represent the random field. improved as additional terms are retained and
that the polynomial chaos expansion might be determination of the unknown coefficients in the
computationally demanding due to the large approximation scheme (Acharjee and Zabaras
number of expansion coefficients that have to be 2007; Babuška et al. 2007; Baroth et al. 2007;
computed. Huang et al. 2007). This leads to nonintrusive
Once the random field involved in the stochas- algorithms that allow to combine the solution
tic boundary value problem has been discretized, procedure with repetitive runs of a finite element
a solution method has to be adopted in order to (FE) solver for deterministic problems.
solve the boundary value problem numerically. A nonintrusive algorithm based on least squares
The choice of the solution method depends on the regression has been presented recently in
required statistical information of the solution. Berveiller et al. (2006).
If only the first two statistical moments of the In the following, discretization methods for
solution are of interest (second moment analysis), the random parameter field are illustrated, and a
the perturbation method can be applied. However, mathematical theory for the approximate solution
if a full probabilistic analysis is necessary, Galerkin of stochastic elliptic boundary value problems
schemes can be utilized or one has to resort to involving a discretized random parameter field
Monte Carlo simulations eventually in combination that is represented as a superposition of indepen-
with a von Neumann series expansion. dent random variables is outlined. In the random
The perturbation method starts with a Taylor domain, global and local polynomial chaos
series expansion of the solution, the external expansions are employed. The relation between
loading, and the stochastic stiffness matrix in local approximations of the solution and Monte
terms of the random variables introduced by Carlo simulation is considered, and reliability
the discretization of the random parameter assessment is briefly discussed. Finally, an exam-
field. The unknown coefficients in the expansion ple serves to illustrate the different solution
of the solution are obtained by equating terms of procedures.
equal order in the expansion. From this, approx-
imations of the first two statistical moments
can be obtained. The perturbation method is Discretization of Random Fields
computationally more efficient than direct
Monte Carlo simulation. However, higher- Let D be a convex bounded open set in ℝn and
order approximations will increase the compu- (O, F , P) be a complete probability space, where
tational effort dramatically, and therefore accu- Ω is the set of outcomes, F the s-field of events,
rate results are obtained for small coefficients of and P : F ! ½0 : 1 a probability measure.
variation only. A function a : D Ω ! ℝm is a random field, if
In the spectral SFEM, the random parameter a(x, o) is a random variable for any x D. In the
fields are discretized by a KL or a polynomial following, we consider scalar valued random S
chaos expansion, the solution is expanded with fields (m = 1).
Hermite polynomials, and a Galerkin approach The finite dimensional distribution of order
is applied to solve for the unknown expansion q of a at x1, x2, . . ., xq D is the probability of
coefficients. The theoretical foundation has been the set \qi¼1 faðxi , oÞ ai g. The random field is
laid in Deb et al. (2001) and Babuška homogeneous (in strict sense), if the finite dimen-
et al. (2005), where local and global polynomial sional distributions are invariant under a space
chaos expansions for linear elliptic boundary shift and thus depend only on the space lag.
value problems with stochastic coefficients Suppose that the random field a is square
were investigated and where a priori error esti- integrable on D Ω and denote by
ð
mates have been proved for a fixed number of
E½X ¼ xd P the expectation of the random
terms of the KL expansion. O
Instead of a Galerkin projection, several variable X(o). Then, the mean, correlation, and
authors employed collocation schemes for the covariance are given by E[a(x, o)], R(x, y) = E
[a(x, o)a(y, o)], and C(x, y) = E[(a(x, o) E where fi are random variables uniformly distrib-
[a(x, o)])(a(y, o) E[a(y, o)]), respectively. uted on [0, 2p]. A representation involving a
A random field is homogeneous (in weak sense), finite number of random variables is thus
if the mean is constant and the correlation
depends only on the space lag y x. X
M pffiffiffiffiffiffiffiffiffi
2DS cos ðvi x þ fi Þ: (5)
Bochner’s theorem allows to introduce the i¼1
spectral distribution S(n) of a weakly homoge-
neous random field with continuous correlation As a consequence of the central limit theorem, it
function by converges for M ! 1 to a Gaussian random field
with the same mean value and autocorrelation
ð structure as the target Gaussian random field.
Rð x Þ ¼ expðix vÞd SðvÞ: (1)
ℝn
Karhunen-Loève Expansion
Due to the properties of the covariance function,
If S is absolutely continuous, the Radon- the operator T : L2(D) ! L2(D),
Nikodym derivative s(n) is called spectral density
function: d S(n) = s(n) dn. ð
Tu ¼ Cðx, yÞuðxÞd x; (6)
Spectral Representation D
If a is a weakly homogeneous mean-square con-

tinuous random field, it can be represented by is compact and self-adjoint and thus admits a
spectrum of decreasing nonnegative eigenvalues
ð f li g1
i¼1 : The corresponding eigenfunctions
aðx, oÞ ¼ expðix vÞd W ðvÞ; (2) ffi ðxÞg1 i¼1 are orthonormal in L2(D). The random
ℝn variables given by
ð
where the random field W(n) has mean zero and 1
satisfies E[d W(n) d W(m) ] = d(n m) d S(n). The xi ðoÞ ¼ pffiffiffiffi ðaðx, oÞ E½aðxÞÞfi ðxÞd x (7)
li D
asterisk denotes complex conjugation and d is the
Dirac d-distribution. are uncorrelated (but in general not independent),
This distribution can be used to approximately have zero mean and unit variance, and allow to
represent a homogeneous Gaussian random field represent the random field by the KL expansion
by a superposition of harmonics. For a real-valued
random field, one obtains the representation X
1 pffiffiffiffi
aðx, oÞ ¼ E½a ðxÞ þ li xi ðoÞfi ðxÞ (8)
ð i¼1
aðx,oÞ ¼ ð cos ðv xÞd U ðvÞ þ sin ðv xÞd V ðvÞÞ;
ℝn that converges in L2(D Ω) (Loève 1977).
(3) Conditions for stronger convergence properties
are given in Babuška et al. (2005). The KL expan-
where E[d U(n)2] = E[d V(n)2] = d S(n). Starting sion is usually truncated by retaining only the first
from a partition of the wave number domain, the M terms. In order to keep the computational effort
increments d U(n) and d V(n) are approximated by small, a fast decay of the spectrum of Eq. 6 is
important. It is shown in Todor and Schwab
pffiffiffiffiffiffiffiffiffi
DUðvi Þ ¼ 2DS cos fi ; (2006) that fast eigenvalue decay corresponds to
pffiffiffiffiffiffiffiffiffi smoothness of the covariance function. Moreover,
DV ðvi Þ ¼ 2DS sin fi ; (4)
for a decreasing correlation length, the number
DS ¼ Sðviþ1 Þ Sðvi Þ M of retained terms increases if the accuracy of
the approximation is kept constant. The KL expan- several authors (Grigoriu 2006; Stefanou and
sion reduces to the spectral representation method Papadrakakis 2007; Sudret and Der Kiureghian
for homogeneous random fields defined over an 2000) observed problems regarding the homoge-
infinite domain (Huang et al. 2001). neity of samples generated from the truncated
For a prescribed, uniformly bounded random expansion.
field a(x, o), the random variables xi(o) in Eq. 8
would be dependent non-Gaussian random vari-
ables whose joint distribution function is very Kriging
difficult to identify. If, on the other hand, inde- Application of Kriging to the discretization of
pendent but bounded distributions are prescribed random fields has been introduced by Li
for xi(o), the random field a(x, o) is not neces- and Der Kiureghian (1993). The random field
sarily bounded for N ! 1. Thus, one is left with a(x, o) is approximated by a linear function of
Gaussian distributions for xi(o) and a(x, o), with M nodal values ai(o) = a(xi, o)
transformations of Gaussian random fields or
with some situations, where nonnegative distri- X
M
butions for xi(o) lead to meaningful (e.g., Erlang) ~a ðx, oÞ ¼ f0 ðxÞ þ ai ðoÞfi ðxÞ: (9)
distributions for a(x, o). i¼1
The KL expansion is optimal in the sense that

the error measured in L2(D Ω) resulting from a The functions fi (x), i = 0,. . ., M are determined
truncation after M terms is smaller than for any by minimizing in each point x the variance of the
other linear combination of M functions. It is a error aðx, oÞ ~a ðx, oÞ under the condition that
representation for homogeneous as well as ~a ðx, oÞ is an unbiased estimator of a(x, o). This
nonhomogeneous random fields. However, yields
X
M
E½aðx, oÞ ¼ f0 ðxÞ þ E½ai ðoÞfi ðxÞ,
i¼1
(10)
X
M
fi ðxÞCov ai ðoÞaj ðxÞ ¼ Cov aðx, oÞaj ðoÞ , j ¼ 1, 2, . . . , M;
i¼1
with the covariance operator KL expansion provides the lowest mean error
over the domain.
Cov½X, Y ¼ E½ðX E½XÞ ðY E½Y Þ: (11) S

Polynomial Chaos Expansion
In contrast to the KL expansion, the polynomial
Li and Der Kiureghian (1993) introduced a spec- chaos expansion does not need an a priori knowl-
tral decomposition of the nodal covariance edge of the covariance structure of the random
matrix. They showed that the maximum error field. A functional representation of the random
of the KL expansion is not always smaller than field in terms of a vector of basic random vari-
the error of Kriging for a given number of ables x,
retained terms. The point-wise variance error
estimator of the KL expansion for a given order aðx, oÞ ¼ f ðx, xÞ; (12)
of truncation is smaller than the error of Kriging
in the interior of the discretization domain but is necessary. Such a representation is given for
larger at the boundaries. Note however that the the solution field of a stochastic boundary value
problem, if the random parameter field has been more detail in Ghanem and Spanos (1991). Xiu
discretized. and Karniadakis (2002) extended the approach to
A nonlinear expansion of the functional f(x, x) general families of orthogonal polynomials, the
is obtained by projecting it on a set of polyno- Wiener-Askey chaos.
mials {Gp} the basic random variables x. The Field and Grigoriu (2004) pointed out some
space spanned by {Gp} is called the pth homoge- limitations of polynomial chaos approximations.
neous chaos. The polynomial Gp, called the poly- They demonstrated that the convergence rate of
nomial chaos of order p, is a polynomial of order polynomial chaos approximations may be slow,
p that is orthogonal to all polynomials with order that the accuracy of the polynomial chaos
less than p. approximation is not always improved by adding
Assuming symmetry of the polynomials, terms, that higher-order moments may be inaccu-
which is always possible (Ghanem and Spanos rate, and that the polynomial chaos approxima-
1991), the random field can be approximated by tions of homogeneous non-Gaussian processes
may not be homogeneous.
X
N
~
a ðx, oÞ ¼ f0 ðxÞ þ fi1 ðxÞG1 xi1 Transformation Techniques for Non-Gaussian
i1 ¼1
Random Fields
N X
X i1
þ fi1 i2 ðxÞG2 xi1 , xi2 Transformation techniques for non-Gaussian ran-
i1 ¼1 i2 ¼1 dom fields seek to represent the non-Gaussian
N X
X i1 X
i2 random field as a nonlinear transformation of a
þ fi1 i2 i3 ðxÞG3 xi1 ,xi2 ,xi3 þ . .. Gaussian random field:
i1 ¼1 i2 ¼1 i3 ¼1
(13)
aðx, oÞ ¼ F1 ðFðgðx, oÞÞÞ; (15)
or more briefly by
where F is the standard Gaussian cumulative
X
M
~
a ðx, oÞ ¼ Fi ðxÞCi ðxÞ: (14) distribution function, F is the non-Gaussian mar-
i¼1 ginal cumulative distribution function of a(x, o),
and g(x, o) is the underlying Gaussian random
The expansion is convergent in the mean- field.
square sense. For Gaussian random variables, The transformation imposes a correlation
the polynomial chaos expansion is described in structure to a(x, o), namely,
ð1 ð1

Rð x Þ ¼ F1 ðFðuÞÞF1 ðFðvÞÞf u, v, Rg ðxÞ d u d v; (16)
1 1
where f (u, v, Rg (x)) denotes the joint Gaussian random field. It allows to calculate analytically
probability density function. If the correlation many important quantities, such as crossing rates
structure of a(x, o) does not match the prescribed and extreme value distributions.
values, one has to resort to nonlinear transforma- Phoon et al. (2002, 2005) used the KL
tions that match the target marginal distribution expansion for the simulation together with an
and/or the correlation structure approximately iterative mapping scheme to fit the target mar-
(Grigoriu 1998). ginal distribution function of non-Gaussian ran-
The nonlinear transformation technique can dom fields. The method allows to simulate
be combined with any series expansion schemes homogeneous as well as nonhomogeneous ran-
described above for the underlying Gaussian dom fields.
Stochastic Linear Elliptic Boundary reads: find u H 10 ðDÞ L2P ðOÞ, such that for all
Value Problems v H 10 ðDÞ L2P ðOÞ,
ð ð ð ð
Consider the following model problem with aðx,oÞ∇u ∇vdxdPðoÞ ¼ f ðxÞvðx,oÞdxdPðoÞ:
stochastic operator and deterministic input on O D O D
D O : find D O ! ℝ , such that P-almost (21)

surely
The assumptions on the random field a(x, o)
∇ aðx, oÞ∇uðx, oÞ ¼ f ðxÞ on D, guarantee the continuity and coercivity of the
(17) bilinear form in Eq. 21, and thus, the existence
uðx, oÞ ¼ 0 on @D:
and uniqueness of a solution to Eq. 21 follows
from the Lax-Milgram lemma.
It is assumed that the deterministic input function
f(x) is square integrable and that the random field
a : D Ω ! ℝ is bounded and coercive, i.e., there
Numerical Solution of the Stochastic
exists positive constants amin, amax, such that
Boundary Value Problem
Pðo O : amin < aðx, oÞ < amax 8x DÞ ¼ 1 In a first step, the random parameter field is
(18) discretized and replaced by a finite sum of ran-
dom variables. Assume that a suitable approxi-
and that the random field has a continuous and mation is given by a linear combination of
square-integrable covariance function. continuous and independent random variables
We are interested in the probability that a xi(o) with zero mean and unit variance,
functional F(u) of the solution u(x, o) exceeds a
threshold F0, i.e., we want to evaluate the integral X
M
aM ðx, oÞ ¼ E½a ðxÞ þ xi ðoÞfi ðxÞ; (22)
ð i¼1
PF ¼ wðF0 ,1Þ ðFðuðx, oÞÞÞd PðoÞ; (19)
O where Gi = xi(Ω) are bounded intervals in ℝ.
Under these assumptions, the stochastic
where wI(), the indicator function, assumes the variational problem involving the random field
value 1 in the interval I and vanishes elsewhere. aM(x, o) has the following deterministic equiva-
The variational formulation of the stochastic lent: find u H 10 ðDÞ L2P ðGÞ, such that
boundary value problem necessitates the intro-
duction of the Sobolev space H10(D) of functions ð ð
having generalized derivatives in L2(D) and aM ðx, yÞ∇x uðx, yÞ ∇x vðx, yÞd xpðyÞd y
vanishing on the boundary @D with norm
G D
ð ð S
ð 1=2 ¼ f ðxÞvðx, yÞdxpðyÞdy;
kukH1 ðDÞ ¼ j∇uj d x
2
, the space L2P(O) of G D
0
D (23)
square-integrable random variables, and the ten-
sor product space H 10 ðDÞ L2P ðOÞ of H10(D)- for all v H 10 ðDÞ L2P ðGÞ, where p L1(G) is
valued random fields with finite second order the joint probability density functions of the ran-
moments, equipped with the inner product dom variables xi(o), i = 1, 2, . . . , M; see
ð ð Babuška et al. (2005). Here, G ¼ ∏M i¼1 Gi ℝ .
M
ðu,vÞH1 ðDÞL2 ðOÞ ¼ ∇u ðx,oÞ ∇v ðx,oÞdxdPðoÞ: This variational formulation is now discretized
0
O D on finite dimensional approximation spaces. For
(20) H10(D), a family of standard finite element
The variational formulation of the stochastic lin- approximation spaces Xh H10 ðDÞ of continuous
ear elliptic boundary value problem Eq. 17 then piecewise linear functions in a regular
triangulation Th of D with mesh parameter h is interpreted as finite element stiffness matrices

considered. Following Deb et al. (2001), discon- with a spatial variation of Young’s modulus.
tinuous finite elements are applied on the domain If the basis fck gPk¼1 consists of discontinuous
G, which is partitioned into elements finite element base functions, the system of equa-
g g
g ¼ ∏M i¼1 ai , bi , with (ai, bi) Gi. Elements of
tions Eq. 25 decouples and can be treated sepa-
Y qk L2p ðGÞ are functions that are polynomials of rately for each element g. Moreover, due to
degree at most q (i.e., yq1 q2 qM q the presence of terms in one single variable
1 y2 . . . yM yk if q1 + q2
+ + qM q) when restricted to each element g ys, s = 1, 2,. . ., M on the left-hand side of
G. The parameter k = (k1, k2,. . . kM), with Eq. 25, it is possible to construct the basis
fck gPk¼1 in a way that the problem decouples
ki ¼ maxgG bgi bgi , represents the mesh para-
meter. If the partition consists of a single element into problems that have the same size as the
only and the degree q is varied, global approxi- deterministic FE problem (Babuška et al. 2005).
To see this, consider the polynomial basis
mations are obtained. g Pg
Denote with Ni(x), i = 1. . . N a basis of Xh and ci i¼1 in one element g only. This basis can
with ck(y), k = 1. . . P, a basis of Yqk . The solution be constructed by multiplying polynomials in one
u(x, y) is approximated on Xh Y qk by single variable ys, s = 1, 2, . . ., M. Let P(y) be the
monomial basis (1, y, y2, . . ., yq)T. We want to
X
N X
P
find a transformation matrix S(s) g , such that the
uik N i ðxÞck ðyÞ: (24)
i¼1 k¼1
transformed basis cðgsÞ ðyÞ ¼ SðgsÞ PðyÞ satisfies
ð bgs
In order to determine the unknown coefficients ðsÞ ðsÞ
uik, this representation is inserted into Eq. 23 ci ðyÞcj ðyÞps ðyÞdy ¼ dij and
ags
together with test functions v(x, y) = Nj(x)cl(y), ð bgs (27)
j = 1, 2,. . ., N, l = 1, 2, . . ., P, yielding ðsÞ ðsÞ ðsÞ
ci ðyÞcj ðyÞys ps ðyÞdy ¼ li dij :
ags
P ð
X X
M
ðKð0Þ þ KðsÞ ys cj ðyÞcl ðyÞpðyÞdyuj Defining the matrices A(s) (s)
g and Bg by
j¼1 G s¼1
ð
ð bgs
¼ f cl ðyÞpðyÞdy; ðsÞ
G
Agij ¼ Pi ðyÞPj ðyÞyps ðyÞdy
ag
(25) ð bsgs (28)
ðsÞ
Bgij ¼ Pi ðyÞPj ðyÞps ðyÞdy;
where ags
ð
ð0Þ
K ij ¼ E½aðxÞ∇N i ðxÞ ∇N j ðxÞdx, it is easily seen that l(s)
i are the eigenvalues and
ðD S(s)
g is the matrix of the eigenvectors of the gen-
ðsÞ
K ij ¼ fs ðxÞ∇N i ðxÞ ∇N j ðxÞdx, eralized eigenvalue problem
ðD
fi ¼ f ðxÞN i ðxÞdx, i, j ¼ 1, 2, . . . , N; AðgsÞ SðgsÞT ¼ BðgsÞ SðgsÞT L, with SðgsÞ BðgsÞ SðgsÞT ¼ I:
D
(26) (29)
and uj is the N 1 dimensional matrix obtained Now, by introducing multiindices j = (j1, j2, . . .,
ðsÞ
from uij for fixed value of j. In structural mechan- jM) for the basis function cgi ¼ ∏M
s¼1 cis ðys Þ on
ics, the matrices K(s), s = 0, 1,. . ., M can be each element g, we find from Eq. 25
! ð
X
M
Kðg0Þ þ KðgsÞ uj ¼ fg cgj ðyÞpðyÞdy, 1 ji q þ 1, i ¼ 1, 2, . . . , M (30)
s¼1 g
due to the orthogonality properties Eq. 27 of If the vectors u(yj) are computed by means of
this basis. Eq. 31, it is not obvious how to interpolate the
As can be seen from this equation, the solution in probability space for other values of y.
parallelization of the algorithm for computing A simple interpretation in terms of a local
the local chaos approximations is easily possible approximation would be a partition of G into
due to the fact that: N subdomains, where each subdomain contains
the nearest neighbors of the sampling point yj. In
1. The approximations on each element g are each subdomain, u(y) is then approximated by a
independent. constant value, namely, the value of the sampling
2. The linear systems for the expansion coeffi- point associated with the subdomain. If the num-
cients decouple due to the choice of ber of sampling points tends to infinity, the error
biorthogonal polynomials instead of Hermite of the approximation vanishes.
polynomials. This approach can be related to Latin hypercube
sampling, if the number of subdomains g is very
As a consequence, any degree of high and the approximations are not computed for
parallelization (from coarse grained to fine every subdomain of the partition. For computing
grained) is possible, depending on the number n samples according to the Latin hypercube sam-
of processors at disposal. Note, however, that the pling method, G is partitioned into nN subdomains
introduction of biorthogonal polynomials is pos- of equal size. Then, the n N matrix P containing
sible only if the random variables xi are indepen- random permutations of 1,. . ., n and an n N
dent and that approximations with biorthogonal matrix G R of independent and uniformly over
polynomials require in general an upper limit [0, 1] distributed random numbers are constructed,
q for the polynomial degree in each single vari- and the sampling plan S = (P R)/n is
able ys, s = 1, . . ., M, i.e., yq1
1 y2 . . . yM is an
q2 qM
established. Mapping the elements
of the sampling
approximation function, if qs q, s = 1, 2, . . ., plan to G via xij ¼ P1 G S ij , i = 1, . . ., n, j = 1, . . .,
1
M. For the same level q, this truncation leads N, where PG ðÞ is the inverse of the standard
to much more expansion coefficients than Gaussian distribution, one computes the solution
requiring q1 + q2 + qM q, especially if only in those subdomains, in which the sample
M is large. xi = (xi1, . . ., xiN), i = 1, . . ., n falls. In this way,
a hybrid method, i.e., a combination of sampling
techniques with approximation techniques, is
Relationship with Monte Carlo obtained that leads to a considerable reduction of
Simulation the sampling variance.
There is a close correspondence between Monte S

Carlo simulation and local polynomial approxi- Evaluation of Response Quantities
mations. The equations for Monte Carlo simula-
tion can be obtained if instead of the Galerkin Once the algebraic problem is solved and the
approximation, a collocation method is applied approximation coefficients have been deter-
with respect to G, leading to mined, an expression of the displacement field
that depends on the input random variables has
! been obtained. This expression can be considered
X
M
ð0Þ
K þ KðsÞ yjs u yj ¼ f, j ¼ 1, 2, . . . , N; as a response surface. This response surface has
s¼1 local character (Proppe 2008) and depends on the
(31) size and location of the elements g, if a partition
of G is adopted.
with N sampling points yj, j = 1, 2, . . ., N, where Approximations for the moments of u(x, o)
finite element solutions are obtained. can be computed by evaluating this expression
with respect to the input random variables. Com-

putation of approximations for the distribution of
u(x, o) or the failure probability Eq. 19 is a more
l =1
complex task, and resort to Monte Carlo simula- q =1
tion via the obtained expression for u(x, o) seems y
to be the easiest way to accomplish it.
x
For solving reliability problems, it is very
helpful to consider the approximation for
the displacements as a local response surface,
which leads to a functional relationship between Stochastic Finite Elements, Fig. 1 Thin square plate
the input random variables and u(x, o). It is under uniform in-plane tension
then possible to compute the most probable
point of failure (MPP), i.e., the point x with Spanos 1991 and Fig. 1). The product of Young’s
F(u(x, x)) = F0 with lowest Euclidean norm, modulus and the thickness of the plate is assumed
and to refine the grid at its vicinity. In this way, to be an isotropic Gaussian random field with
it is possible to solve the reliability problem with covariance function
a high degree of accuracy in an adaptive manner.
The MPP may also be useful for the evaluation jx1 x2 j jy1 y2 j
Cðx1 , y1 ; x2 , y2 Þ ¼ s exp ;
of the integral in Eq. 19 by means of variance- lc lc
reduced Monte Carlo simulation (importance (33)
sampling). To this end, a sampling density p~ðyÞ
is introduced by shifting the original probability standard deviation s = 0.2, correlation length
density function p(y) of the random variables xi, lc = 1, and unit mean value. The Poisson’s ratio
i = 1, . . ., M, to the previously obtained MPP, and is 0.3. The plate has unit length and the external
Eq. 19 is approximated by excitation is deterministic and of unit magnitude.
The longitudinal displacement of one of the free
XN j pð y j Þ j corners is considered in the following.
PF w ð F 0 ,1 Þ F u x, y p~ y ;
j¼1 p~ðyj Þ
(32) Approximation of Higher-Order Moments:
Skewness and Kurtosis
where the sampling points yj, j = 1, 2, . . ., M are The random field has been represented by two
generated according to p~ðyÞ and u(x, yj) is com- random variables (M = 2). Figures 2 and 3 dis-
puted from the approximation on the element that play the approximations obtained with polyno-
contains yj. mials up to third order for the skewness and the
The accuracy of the MPP is influenced by the kurtosis of the displacement of the free corner
FE mesh, the truncation level M of the random with increasing number of intervals of equal
parameter field, the partition of G, and the choice probability. Reference values have been obtained
of the ansatz functions both in spatial and random by Monte Carlo simulation with 30 106 samples.
domain. These parameters can be gradually For one interval, the result corresponds to a
adapted such that the MPP is computed with a global approximation. It can be seen that global
prescribed accuracy. approximations are rather inexact for the skew-
ness and kurtosis and that it is sufficient to split
the intervals into two parts in order to improve the
Example results significantly.
In order to illustrate the stochastic finite element Hybrid Method

techniques, consider a clamped thin square plate Figures 4 and 5 display relative (with respect to
under uniform in-plane tension (cf. Ghanem and the estimated values) 95 % confidence intervals
Stochastic Finite 1.4

Elements,
Fig. 2 Approximation of 1.2
the skewness
1
skewness
0.8
0.6
0.4
MCS
0.2 Pol. order 1
Pol. order 2
Pol. order 3
0
0 2 4 6 8
intervals
Stochastic Finite 8
Elements,
Fig. 3 Approximation of 7
the kurtosis
6
5
kurtosis
2
MCS
1 Pol. order 1
Pol. order 2
Pol. order 3
0
0 2 4 6 8
intervals
Stochastic Finite 25 S
rel. size of 95%-confidence intervals [%]
Elements, LHS
Pol.Order1-LHS
Fig. 4 Approximation of
the skewness by Latin 20
hypercube sampling
techniques
15
10
0
0 250 500 750 1000
intervals
Stochastic Finite 25
rel. size of 95%-confidence intervals [%]

LHS
Elements,
Fig. 5 Approximation of Pol.Order1-LHS
the kurtosis by Latin 20
hypercube sampling
techniques
15
10
0
0 250 500 750 1000
intervals
local approx.
Elements,
global approx.
Fig. 6 Accuracy of 30
estimates for the MPP
25
rel. error [%]
20
15
10
0
1 2 3
pol. order, global approx.
for the estimation of the skewness and kurtosis computation of the displacements. These compu-
obtained from Latin hypercube sampling and tations can be bypassed, if the approximations
Latin hypercube sampling combined with linear obtained from the numerical solution of the sto-
approximation in the sampling intervals. A three- chastic boundary value problem are employed as
term (M = 3) representation of the random field response surfaces. Figure 6 compares the relative
by truncated Gaussian random variables has been error (i.e., the relative Euclidean distance)
employed. For a low to moderate number of between the “true” MPP computed with FE calls
intervals, the confidence intervals obtained by and approximations obtained with local and
the hybrid method are narrower than those of global response surfaces of the displacement.
the standard Latin hypercube sampling. The random field has been represented by four
random variables (M = 4). The functional F(u) is
Reliability Assessment the value of the displacement field at the free
For most reliability approximation techniques, it corner and the threshold value F0 has been set
is necessary to compute the MPP. In order to to 2.0. In order to obtain local approximations, an
obtain this point, a constrained optimization approximation of the MPP has been computed
problem has to be solved. This implies expensive with a linear global approximation, and an inter-
function calls (e.g., FE solutions) for the val of length 1s has been inserted at that point for
Stochastic Finite 0.002

Elements, direct computation
Fig. 7 Prediction of the 0.0018 global approx., order 1
global approx., order 2
failure probability
0.0016 global approx., order 3
local approx., 1 direction
0.0014 local approx., 2 directions
failure probability
0.0012
0.001
0.0008
0.0006
0.0004
0.0002
0
2 2.1 2.2 2.3 2.4 2.5 2.6 2.7
F0
the coordinate with largest partial derivative of This allows to compare the efficiency of local and
the longitudinal displacement for the free corner global approximations. From Fig. 8, it is evident
at the MPP. Linear polynomials have been that local approximations have considerable
applied for local approximations. From Fig. 6, advantages over global approximations. This
one can see that this method yields very accurate behavior is even more pronounced for smaller
approximations for the MPP, while for global values of F0.
approximations, polynomials of third degree are Finally, for F0 = 2.7 (corresponding to a prob-
necessary in order to achieve a comparable ability of failure of 8.2 105) and a tolerance of
accuracy. 2 % for the Euclidean norm of the MPP, the
From this fact, a corresponding approximation procedure described in section “Relationship
quality of the failure probability can be deduced, with Monte Carlo Simulation” yields M = 4 and
if variance-reduced sampling techniques a partition of G into 48 elements leading to an
which rely on the MPP together with the overall effort of 768 deterministic FE runs for the
abovementioned response surface techniques are reliability estimation problem. This is consider-
employed. This can be clearly seen from Fig. 7, ably lower than 104 runs for importance sampling
where the probability of failure obtained with and 105 runs for a direct Monte Carlo simulation.
various approximation techniques has been plot- S
ted over the threshold value F0. Importance sam-
pling at the predicted MPP with 30 batches of Concluding Remarks
10,000 samples has been employed for each fail-
ure probability estimate. For global approxima- For the SFE solution of stochastic boundary value
tions, again polynomials of third degree are problems, a mathematical theory is available that
necessary in order to obtain accurate predictions. is in many aspects comparable to its deterministic
On the other hand, local approximations with counterpart, the finite element method. However,
linear polynomials lead already to quite accurate for random fields with short correlation lengths
results, if only the principle direction is (requiring a high number M of random variables),
partitioned. the solution methods become inefficient, due to
In Fig. 8, the relative error of the failure prob- the series expansion of the solution. This is also
ability for F0 = 2.3 is displayed over the true for most other SFE approximations, be they
corresponding number of deterministic FE runs. global or local, and recourse to efficient sampling

Elements, local approx.
Fig. 8 Efficiency of global global approx.
and local approximation
techniques for the
prediction of the failure
probability
rel. error [%]
101
0 50 100 150 200 250 300
FE runs
techniques combined with efficient model reduc- Baroth J, Bresslotette P, Chauvière C, Fogli M (2007) An
tion might be the only way to circumvent this efficient SFE method using Lagrange polynomials:
application to nonlinear mechanical problems with
problem. uncertain parameters. Comput Methods Appl Mech
A fundamental question that still has to be Eng 196:4419–4429
addressed in detail concerns the error of the solu- Berveiller M, Sudret B, Lemaire M (2006) Stochastic
tion due to the discretization of the random finite element: a non intrusive approach by regression.
Rev Eur Mècanique Numèrique 15:81–92
parameter field. Beyond this aspect of verifica- Brenner CE, Bucher CG (1995) A contribution to the
tion, the validation of the random field model SFE-based reliability assessment of nonlinear struc-
itself, either from experimental data or from tures under dynamic loading. Probab Eng Mech
information pertaining to the microscale, remains 10:265–273
Contreras H (1980) The stochastic finite-element method.
an important issue (cf. the critique of the SFE Comput Struct 12:341–348
method raised, e.g., in Ostoja-Starzewski 2011). Cornell CA (1970) First order uncertainty analysis in soils
deformation and stability. In: Proceedings of first inter-
national conference of statistics and probability in soil
References and structural engineering, Hong Kong
Deb MK, Babuška IM, Oden JT (2001) Solution of sto-
Acharjee S, Zabaras N (2007) A non-intrusive stochastic chastic partial differential equations using Galerkin
Galerkin approach for modeling uncertainty propaga- finite element techniques. Comp Methods Appl Mech
tion in deformation processes. Comput Struct Eng 190:6359–6372
85(5–6):244–254 Deodatis G (1991) Weighted integral method I: stochastic
Astill CJ, Noseir SB, Shinozuka M (1972) Impact loading stiffness matrix. J Eng Mech 117:1851–1864
on structures with random properties. J Struct Mech Deodatis G, Micaletti RC (2001) Simulation of highly
1:63–77 skewed non-Gaussian stochastic processes. Trans
Babuška IM, Tempone R, Zouraris GE (2005) Solving ASCE, J Eng Mech 127:1284–1295
elliptic boundary value problems with uncertain coef- Der Kiureghian A, Ke JB (1988) The stochastic finite
ficients by the finite element method: the stochastic element method in structural reliability. Probab Eng
formulation. Comput Methods Appl Mech Eng Mech 3:83–91
194(1):1251–1294 Field RV Jr, Grigoriu M (2004) On the accuracy of the
Babuška I, Nobile F, Tampone R (2007) A stochastic polynomial chaos approximation. Probab Eng Mech
collocation method for elliptic partial differential 19(1–2):65–80
equations with random input data. SIAM J Numer Ghanem RG, Spanos PD (1991) Stochastic finite ele-
Anal 45:1005–1034 ments: a spectral approach. Springer, New York
Baecher GB, Ingra TS (1981) Stochastic FEM in settle- Grigoriu M (1998) Simulation of stationary non-Gaussian
ment predictions. J Geotech Eng Div 107:449–463 translation processes. J Eng Mech 124:121–126
Stochastic Ground Motion Simulation 3483
Grigoriu M (2006) Evaluation of Karhunen-Loève, spec- Vanmarcke E, Grigoriu M (1984) Stochastic finite ele-
tral and sampling representations for stochastic pro- ment analysis of simple beams. J Eng Mech
cesses. J Eng Mech 132:179–189 109:1203–1214
Hisada T, Nakagiri S (1981) Stochastic finite element Xiu D, Karniadakis G (2002) The Wiener-Askey polyno-
method developed for structural safety and reliability. mial chaos for stochastic differential equations. SIAM
In: Proceedings of third international conference on J Sci Comput 24:619–644
structural safety and reliability, Trondheim, Yamazaki F, Shinozuka M (1988) Digital generation of
pp 395–408 non-Gaussian stochastic fields. Trans ASCE, J Eng
Huang SP, Quek ST, Phoon KK (2001) Convergence Mech 114:1183–1197
study of the truncated Karhunen-Loève expansion for
simulation of stochastic processes. Int J Numer
Methods Eng 52:1029–1043
Huang S, Mahadevan S, Rebba R (2007) Collocation-
based stochastic finite element analysis for random
field problems. Probab Eng Mech 22:194–205 Stochastic Ground Motion
Lawrence M (1987) Basis random variables in finite ele-
ment analysis. Int J Numer Methods Eng
Simulation
24:1849–1863
Li CC, Der Kiureghian A (1993) Optimal discretization of Sanaz Rezaeian1 and Xiaodan Sun2
random fields. J Eng Mech 119:1136–1154 1
U.S. Geological Survey, Golden, CO, USA
Liu WK, Belytschko T, Mani A (1986) Probabilistic finite 2
elements for nonlinear structural dynamics. Comput
Southwest Jiaotong University, Chengdu, China
Methods Appl Mech Eng 56:61–81
Loève M (1977) Probability theory. Springer, Berlin
Matthies HG, Brenner CG, Bucher CG, Guedes Soares Synonyms
C (1997) Uncertainties in probabilistic numerical anal-
ysis of structures and solids – stochastic finite ele-
ments. Struct Saf 19:283–336 Earthquake ground motion simulation;
Ostoja-Starzewski M (2011) Stochastic finite elements: Nonstationary stochastic process; Stochastic
where is the physics? Theor Appl Mech 38:379–396 models; Strong ground motion simulation; Syn-
Phoon KK, Huang HW, Quek ST (2002) Simulation of
second-order processes using Karhunen-Loève expan-
thetic accelerograms
sion. Comput Struct 80:1049–1160
Phoon KK, Huang SP, Quek ST (2005) Simulation of
strongly non-Gaussian processes using Karhunen- Introduction
Loève expansion. Probab Eng Mech 20:188–198
Proppe C (2008) Estimation of failure probabilities by
local approximation of the limit state function. Struct Strong earthquake ground motion records are
Saf 30(4):277–290 fundamental in engineering applications. Ground
Shinozuka M (1971) Simulation of multivariate and motion time series are used in ▶ response-history
multidimensional random processes. J Acoust Soc
Am 49:357–367
dynamic analysis of structural or geotechnical
Stefanou G (2009) The stochastic finite element method: systems. In such analysis, the validity of
past, present and future. Comput Methods Appl Mech predicted responses depends on the validity of S
Eng 198:1031–1051 the input excitations. Ground motion records are
Stefanou G, Papadrakakis M (2007) Assessment of spec-
tral representation and Karhunen-Loève expansion
also used to develop ground motion prediction
methods for the simulation of Gaussian stochastic equations (GMPEs) for intensity measures such
fields. Comput Methods Appl Mech Eng as spectral accelerations that are used in
196:2465–2477 ▶ response-spectrum dynamic analysis. Despite
Sudret B, Der Kiureghian A (2000) Stochastic finite ele-
ment methods and reliability – state of the art. Tech-
the thousands of available strong ground motion
nical report, UCB/SEMM-2000/08, Department of records, there remains a shortage of records for
Civil & Environmental Engineering, University of large-magnitude earthquakes at short distances or
California, Berkeley in specific regions, as well as records that sample
Todor RA, Schwab C (2006) Convergence rates for sparse
chaos approximations of elliptic problems with sto-
specific combinations of source, path, and site
chastic coefficients. Research Report No. 2006–05, characteristics. The limited number of recordings
Seminar f€ur Angwandte Mathematik, ETH Z€ urich has become problematic in the emerging field of
3484 Stochastic Ground Motion Simulation
▶ performance-based earthquake engineering discretization of the medium and require a thor-

(PBEE), which considers the entire spectrum of ough knowledge of the environment (i.e., earth-
structural response, from linear to grossly quake source, crustal structure, material
nonlinear and even collapse, and thereby requires properties, basin and local site effects); this infor-
ground motions with various levels of intensity mation is not available for many regions or is not
for various earthquake design scenarios (e.g., a accessible to engineers for well-studied regions.
design scenario can be defined by the earthquake On the other hand, stochastic simulation models
magnitude, distance, and site conditions). are simple and powerful tools that incorporate
To obtain the desired ground motions for the what is known about ground motion, source,
purposes of PBEE, it is common engineering path, and site characteristics into simple func-
practice to scale or modify acceleration time tional forms, and as a result are widely used to
series that were recorded during previous earth- simulate earthquake ground motions. There is a
quakes to represent certain ground motion char- specific entry dedicated to ▶ physics-based
acteristics for the design of structural or ground motion simulation in this encyclopedia
geotechnical systems. However, scaling and by Taborda and Roten that provides more details
modification methods can significantly alter on deterministic ground motion simulation
other ground motion characteristics and result in models. Here, we focus on stochastic ground
unrealistic earthquake ground motion time series. motion simulation.
Synthetic ground motions can be used instead to A word of caution may be in order here about
replace or supplement recorded motions when the terminology “physics-based.” The use of this
scarcity of previously recorded motions becomes term for deterministic models is mostly out of
a problem, provided they accurately capture char- convenience and should not imply that stochastic
acteristics of real earthquake ground motions and models are not based on physics. In fact, some
their natural variability. experts disagree with this terminology because
Synthetic ground motions can be based on many stochastic models are valid representations
deterministic or stochastic simulations. of the physics of earthquakes, and although they
A determinist model is one in which variables do not necessarily solve the mathematical prob-
are uniquely determined, and the model performs lem of wave propagation, they integrate the phys-
the same way for a given set of initial conditions. ical characteristics of the earthquake source, path
Conversely, in a stochastic model, randomness is effects, and site effects implicitly through their
present and variables are not described by unique formulations and parameters.
values, but rather by probability distributions. In Both deterministic and stochastic simulations
earthquake engineering, deterministic ground can be used for response-history dynamic analy-
motion simulation is commonly referred to as sis, but only stochastic simulations can be utilized
“physics-based” ground motion simulation. for stochastic dynamic (i.e., random vibration)
These simulation models synthesize the earth- analysis, because the latter analysis method
quake source by defining a source model (e.g., requires a random process model of the earth-
kinematic or dynamic rupture models) and quake ground motion. Synthetic ground motions
describe the seismic wave travel path by defining are particularly useful for nonlinear dynamic
a material model (e.g., seismic velocity model); analysis due to the scarcity of recorded motions
then, they utilize numerical methods (e.g., finite for large-magnitude earthquakes that are capable
element or finite difference methods) to estimate of causing nonlinear responses. Two approaches
the solution to the wave propagation equation. are available for nonlinear dynamic analysis of
In other words, these models explicitly incorpo- structures subjected to earthquakes: (1) nonlinear
rate the physics of the propagation of seismic response-history analysis by the use of a selected
waves. They produce realistic synthetic ground set of ground motion time series and (2) nonlinear
motions at low frequencies (typically <1Hz). But stochastic dynamic analysis by the use of a sto-
they are computationally intensive due to the fine chastic representation of the ground motion.
Well-developed methods and tools are available content with time). Whereas most simulation
for nonlinear response-history analysis; a set of methods account for temporal nonstationarity,
recorded or synthetic ground motions or a collec- which controls most common intensity measures
tion of both can be used for this type of analysis. and duration of motion, many ignore the spectral
Nonlinear stochastic methods are not as devel- nonstationarity, which can be very important in
oped, but research in that direction is continuing nonlinear dynamic analysis.
(e.g., Au and Beck 2003; Der Kiureghian and Many synthetic ground motion models exist,
Fujimura 2009). Several stochastic ground but not all represent the natural variability of real
motion simulation models exist that can satisfy earthquake ground motions. Some models under-
the requirements of these nonlinear stochastic estimate the natural variability of ground motions
methods. These methods typically require the because they underestimate the variability in
input excitation to be represented in a discrete their model parameters. On the other hand, if
form and in terms of a finite number of random correlations between the model parameters are
variables. For example, in Der Kiureghian and not taken into account or are underestimated,
Fujimura (2009), it is necessary to represent the the variability of the resulting ground motions
input excitation in the following form can be overestimated compared to real earth-
Xn quake ground motions. In probabilistic analysis,
mðtÞ þ si ðtÞui , where m(t) is the deterministic it is important to capture the natural variability of
i¼1
mean of the random process, n is the ground motions in the simulations to accurately
discretization of the time series and provides a quantify the variation of response due to the var-
measure of the resolution, and ui and si(t) are, iation in input excitation.
respectively, representatives of standard normal In summary, stochastic ground motion simula-
random variables and deterministic basis functions have many applications. They can be used in
tions of time. linear and nonlinear response-history dynamic
Regardless of the method of simulation, the analysis as well as in stochastic dynamic analysis,
resulting synthetic time series should properly or they can be used to develop GMPEs for various
represent the main characteristics of real earth- intensity measures. Stochastic models are generally
quake ground motions. For most engineering simpler, more widely used, and more accurate at
applications, the intensity, duration, and fre- high frequencies than deterministic (physics-based)
quency content of the input ground motion are models. A good stochastic ground motion model
the characteristics that control the response of must represent the temporal and spectral
structural and geotechnical systems. In addition nonstationary characteristics of real earthquake
to the duration of motion, which results from ground motions and must properly represent their
variation of the ground motion intensity over natural variability. More details on stochastic
time, variation of the frequency content of the ground motion models, their classification, formu- S
motion is also an important characteristic that lation, parameters, and practicality are given below.
influences the response, particularly in nonlinear Discussions on adjustment of models for special
analysis. Consequently, two important character- situations including multicomponent, near-fault,
istics of real earthquake ground motions that sep- and multi-station simulations are also provided.
arate them from a simple stochastic process (e.g.,
a ▶ white-noise signal) are temporal and spectral
nonstationarities. Temporal nonstationarity Classification of Ground Motion
refers to the nonstationarity of ground motion in Simulation Models
the time domain (i.e., variation of intensity with
time). Spectral nonstationarity refers to the There are two main categories of models
nonstationarity of ground motion in the fre- for generating synthetic ground motions,
quency domain (i.e., variation of frequency source-based models and site-based models.
Source-based models explicitly describe the fault site effects and then combines it with a random
rupture process at the source, the propagation of phase spectrum. This method assumes ground
the resulting seismic waves through the ground motion to be a band-limited white Gaussian
medium, and the effects of local site conditions noise with finite duration. The white-noise is
on the seismic waves to generate a time series at a then adjusted to have the following FAS:
specific site. Site-based models describe the
ground motion time series as it is observed at a Y ðM0 , f , RÞ ¼ SðM0 , f Þ PðR, f Þ Gðf Þ I ðf Þ
specific site by fitting a stochastic process to
previously recorded motions with known earth- where S(M0, f ) is the source spectrum with
quake and site characteristics; in this way, they f denoting the frequency and M0 representing
implicitly account for the source, path, and the seismic moment; R is the distance from the
site effects. Due to their need for empirical earthquake source to the site and P(R, f) accounts
recordings of ground motions, site-based for the wave propagation effects, including the
models are sometimes referred to as empirical geometrical attenuation with distance, which is
models, but note that this terminology is not usually described by a three-segmented function,
strictly correct because many source-based and the energy dissipation (combining intrinsic
models also use data to empirically identify and scattering attenuation); G( f ) accounts for
their parameters. the effects of local site conditions, including
near-surface amplification and reduction of high
Source-Based Models frequencies that result from the path-independent
Source-based models can be deterministic or loss of energy; and finally, I( f ) is the indicator of
stochastic. An early review of source-based the instrument or the type of motion (i.e., accel-
models is presented by Zerva (1988). Determinis- eration, velocity, or displacement). More details
tic (physics-based) simulations were briefly are provided in Boore (2003).
described in the Introduction and have their own This method was first developed to model
entry in this encyclopedia. They can produce real- far-field ground motions where the earthquake
istic accelerograms at low frequencies (typically source can be considered as a point (Boore
<1 Hz) but often need to be adjusted for high 1983), resulting in “point-source stochastic
frequencies by combining with a stochastic com- models.” For simulation of ground motions closer
ponent, resulting in hybrid models. For many to the earthquake source, the method was
years, source-based simulations have been utilized improved to consider the rupture progress on a
to develop GMPEs for locations where earth- finite fault (Beresnev and Atkinson 1998),
quakes are rare and the collection of recorded resulting in “finite-fault stochastic models.”
ground motions is sparse, for example, in the sta- In finite-fault modeling, the fault is discretized
ble continental regions of the Central and Eastern into many subfaults, and each subfault is treated
United States (e.g., Somerville et al. 2001); as a point-source. The ground motion from each
research is continuing to improve these simula- subfault is modeled using the point-source sto-
tions and the resulting GMPEs. chastic model with the amplitude spectrum
One family of widely used synthetic ground described in the above equation. The total ground
motions are the source-based stochastic simula- motion at a site is the superposition of the contri-
tions based on the work of David Boore and a butions of all subfaults with a proper time lag
number of other researchers in the past several considering the difference from the triggering
decades (e.g., Boore 2003; Beresnev and time of the subfault and from the travel time
Atkinson 1998; Motazedian and Atkinson 2005; between the subfault and the site. Using finite-
Boore 2009). This simulation method roots in the fault modeling, the synthetic motion is able to
work of McGuire and Hanks (1980), which iden- capture some near-fault characteristics of ground
tifies the Fourier amplitude spectrum (FAS) of a motion, such as rupture directivity and hanging-
ground motion considering the source, path, and wall effect.
Most improvements in source-based stochas- small- or intermediate-magnitude events. One of

tic modeling focus on enhancing the source spec- the latest improvements to the source-based sto-
trum, S(M0, f). The originally used source chastic simulation is the work of Boore (2009)
spectrum was according to the o-square model that allows selection of different functions for
with single corner frequency (Aki 1967) calculating the scaling factor Hij.
Source-based stochastic models are simple and
CM0 practical compared to deterministic (physics-
Sð M 0 , f Þ ¼ 2
1 þ ff based) models. Availability of source codes
c
for these models has increased their use in recent
years. For example, EXSIM_dmb is a simulation
where C is a constant related to the radiation
software that is developed by Boore and
pattern and fc is the corner frequency inversely
can be downloaded from his website (http://
controlled by the dimension of the source. By
www.daveboore.com/software_online.html), and
applying the above equation in finite-fault model-
EXSIM_beta is a simulation software that is devel-
ing, however, the sum of acceleration amplitude
oped by Motazedian and can be found on his
spectrum at high-frequency range (f > fc)
website (http://http-server.carleton.ca/~dariush/);
strongly depends on the discretization scheme
both EXSIM_dmb and EXSIM_beta are based on
of the fault. The proper subfault size is required
stochastic finite-fault modeling with dynamic cor-
to be within 5–15 km in order to obtain a reliable
ner frequency. One of the disadvantages of source-
synthesis (Beresnev and Atkinson 1998). To
based stochastic models is that they assume
solve this problem, Motazedian and Atkinson
stationarity of the frequency content with time,
(2005) made a conceptual improvement by intro-
which is usually not the case in real earthquake
ducing a “dynamic corner frequency” in the
ground motions as will be discussed in more detail
source spectrum
in the next section. Also, care should be taken in
CM0ij H ij using these models for probabilistic analyses
Sij ðf Þ ¼ 2 because they usually underestimate the variability
1 þ ff in ground motions by fixing their parameters
cij
instead of assigning a probability distribution

where fcij is the dynamic corner frequency for the to them.
ij-th subfault and Hij is the scaling factor to com- In general, source-based models tend to
pensate the high-frequency spectral loss from heavily employ seismological principles to
using a dynamic corner frequency. describe the source mechanism and wave travel
The dynamic corner frequency of the ij-th path and depend on physical parameters that vary
subfault depends on the cumulative number of significantly from region to region. This limits
subfaults ruptured by the time the subfault is their use in regions where seismological data S
triggered. Therefore, each subfault will have a are lacking. This is more critical for deterministic
different corner frequency in the source spec- (physics-based) models than it is for stochastic
trum. Generally, a subfault that is triggered in models that use simpler parameters and models
the early part of the rupture will have a higher for the source, path, and site effects. In the current
corner frequency than a subfault triggered in the practice, most engineers prefer using methods of
late part. This describes the difference in the scaling and spectrum matching of recorded
frequency contents of the generated motions motions instead of incorporating source-based
from different subfaults. The most significant models. This is partly due to the lack of under-
advantage of using a dynamic corner frequency standing the seismological principles underlying
is that the high-frequency spectral amplitude is these models and the fact that these models
no longer dependent on the size of the subfault; require a thorough knowledge of the source,
thus a subfault can be as small as 1 km, which wave path, and site characteristics, which
allows the finite-fault modeling to be used for typically are not available to a design engineer.
In this regard, source-based stochastic models of the motion. Furthermore, the model should
have an advantage compared to deterministic refrain from complicated analysis, involving
(physics-based) models; but the user is still extensive processing of recorded motions for
required to make certain assumptions depending parameter identification.
on the region in order to define the models for the Site-based stochastic ground motion simula-
source S(M0, f ), path P(R, f ), and site G(f ). More tion models can be classified into four categories:
recently developed site-based stochastic models
can be more useful in engineering applications Filtered white-noise Processes obtained by
because their relatively simple formulations facil- processes passing a white-noise
itate time-efficient simulations (similar to source- signal through a filter.
based stochastic models) and their parameters are These models are
expressed directly in terms of earthquake and site appealing due to the
characteristics such as magnitude and distance that efficient digital
describe a design scenario and are readily avail- simulation of sample
able to a design engineer. functions for a filtered
white-noise process
Site-Based Models (i.e., generating
Site-based models are stochastic and represent realizations). Many
the earthquake ground motion time series by a models of this type use
stochastic process. A large number of site-based subsequent modulation
stochastic ground motion models have been in time to achieve
developed. Formal reviews are presented by sev- temporal
eral authors including Liu (1970), Shinozuka and nonstationarity, but
Deodatis (1988), and Conte and Peng (1997). The these processes have
papers by Rezaeian and Der Kiureghian (2008, essentially time-
2010, and 2012) provide reviews and are repre- invariant frequency
sentative of the more recent work. As previously content. This type of
mentioned, a good synthetic ground motion simulation can be
model must represent both the temporal and the improved by varying the
spectral nonstationary characteristics of earth- filter parameters in time
quake ground motions. Whereas temporal to achieve
nonstationarity can be easily modeled in site- nonstationarity of the
based stochastic simulations by time-modulating frequency content (e.g.,
a stationary ▶ stochastic process, spectral Rezaeian and Der
nonstationarity is not as easy to model. Neverthe- Kiureghian 2008).
less, this spectral nonstationarity is of particular Filtered Poisson Processes obtained by
importance in nonlinear dynamic analysis processes passing a train of
because of the moving resonance effect of Poisson pulses through a
nonlinear structures (i.e., when the natural period linear filter. Through
of a structure varies during the ground shaking modulation in time,
period due to inelastic behavior and gets closer to these processes can
the varying frequency content of the ground possess both temporal
motion) and should not be ignored. In general, and spectral
for a stochastic model to be of practical use in nonstationarities (e.g.,
earthquake engineering, it should be parsimoni- Lin 1986). However,
ous, i.e., it must have as few parameters as pos- matching with recorded
sible. Preferably, the model parameters should ground motions is
provide physical insight into the characteristics difficult.
Auto-regressive- By allowing the model aðtÞ ¼ qðt, aÞ f ðt, lÞ

moving-average parameters to vary with
(ARMA) models time, these models (e.g., where a(t) is the acceleration time series with
Conte et al. 1992) can t denoting time; q(t, a) is a deterministic nonneg-
have both temporal and ative time-modulating function with a
spectral representing a vector of parameters; and f(t, l) is
nonstationarities. a filtered white-noise process with l representing a
However, it is difficult vector of parameters. Several models have been
to relate the model proposed for the q(t, a) modulating function, e.g.,
parameters to any
physical aspects of the • The piecewise function by Housner and
ground motion. Jennings (1964)
Time-varying Instead of working in the • The double exponential function proposed by
spectral time domain, as is done Shinozuka and Sato (1967)
representation in the previous three • The gamma function proposed by Saragoni
categories, these models and Hart (1974)
work in the frequency
domain and use various These models or their modified versions have
forms of spectral been used by many other modelers. In general, a
representation (e.g., time-modulating function gradually increases
Conte and Peng 1997). from zero to achieve a nearly constant intensity
The focus in these that represents the “strong shaking” phase of an
models is in developing a earthquake ground motion and then gradually
time-varying spectral decays back to zero (see Fig. 1). The parameters
representation by a control the intensity and shape of the function.
matching to a target Although it is not a common practice, two or
recorded ground motion. more modulating functions can be combined to
These models require simulate ground motions that have more than a
extensive processing of single strong shaking phase, a feature possibly
the target motion. caused by multiple sub-events.
Virtually all these models The conventional “filtered white-noise” pro-
assume the ground cess f(t, l) is the stationary response of a linear
motion to be a zero-mean time-invariant filter subjected to a white-noise
Gaussian process. process. White-noise w(t) is a stationary random
process in time that has a zero mean and a con- S
stant spectral density for all frequencies. The
Simulating a Single Ground Motion response of a linear filter to a white-noise process
Time Series Using a Site-Based may be calculated by using the Duhamel convo-
Stochastic Model lution integral, and hence the general formulation
of a filtered white-noise process can be written in
Model Formulation the following form:
Regardless of the specific formulation, every site-
ðt
based model is essentially a stochastic process
with a certain number of parameters. As an exam- f ðt, lÞ ¼ hðt t, lÞwðtÞdt
1
ple, the general formulation for a site-based sto-
chastic model based on the filtered white-noise where h(t, l) denotes the impulse response func-
process is given below (Rezaeian and Der tion or IRF of a linear filter and the parameters l
Kiureghian 2008) can be the natural frequency and damping of the
Stochastic Ground Motion Simulation, Fig. 1 Development of a fully nonstationary stochastic process according to
Rezaeian and Der Kiureghian (2008)
filter or other parameters that shape the filter l(t) are identified by fitting certain characteristics
response. Different filters are used and various of the stochastic process to those of the target
tricks are incorporated by modelers to make the accelerogram. The approaches to parameteriza-
resulting process “fully nonstationary” and rep- tion and model fitting are entirely different from
resentative of real earthquake ground motions. model to model. Currently, there are no standards
One approach taken by Rezaeian and Der on the exact characteristics that modelers should
Kiureghian (2008) is illustrated in Fig. 1, where match to identify their parameters for a target
the filtered white-noise process is normalized by motion. However, all these different characteris-
its standard deviation sf(t) to separate the tempo- tics are designed to capture the intensity, dura-
ral from the spectral characteristics; and the filter tion, and frequency content of the ground motion
parameters are functions of time, l(t), to achieve at certain points in time or in frequency. In some
spectral nonstationarity. Like all other models models, this is done over the entire duration of the
that use a time-modulating function, temporal recorded ground motion and its entire frequency
nonstationarity is achieved by q(t, a). In this spectrum; in this case, the parameters are identi-
figure, the nonstationary characteristics of ground fied using the following generalized formulation:
motions are demonstrated using the trace of an
ð tn h i
example stochastic process.
^ ¼ argmin
p fobj ðrecordedÞ fobj ðsimulationÞ dt
p 0
Model Parameters
Given a target accelerogram (i.e., a recorded where p is a vector of model parameters and tn
ground motion), the model parameters a and denotes the entire duration of the recorded ground
motion. Linear or nonlinear optimization baseline correction methods or high-pass filtering

methods are used to minimize the difference methods used for recorded accelerograms to sim-
between an objective function for the recorded ulated motions in order to overcome this disad-
motion, fobj(recorded), and the same function for vantage (e.g., Papadimitriou 1990; Liao and
the simulation, fobj(simulation), resulting in the Zerva 2006). For example, Rezaeian and Der
vector of identified parameters p ^ for that partic- Kiureghian (2008) use a high-pass filter
ular recorded motion. The misfit of the objective representing a critically damped oscillator to
functions can be characterized by their differ- post-process the stochastically simulated ground
ence, as is shown in the above equation, or by motion to obtain zero velocity and displacement
their absolute difference, or by their square root residuals and appropriate response spectral
difference. Many objective functions are used in values for periods as long as 5–10 s. An example
the literature by various modelers. Objective of a simulated ground motion before and after
functions such as the expected cumulative energy post-processing is shown in Fig. 2. Even though
of the process or the variance of the process can the difference between the acceleration processes
be used to identify the parameters of the time- is insignificant, the integration over time results
modulating function; objective functions such as in unacceptably high nonzero velocity and dis-
the zero-level up-crossing rate can be used to placement residuals as well as high spectral
identify the frequency parameters (e.g., Yeh and intensities at long periods before post-processing.
Wen 1990; Rezaeian and Der Kiureghian 2008). After the post-processing, the residuals are zero
Once a set of model parameters has been iden- and the spectral content is lower at long periods.
^, the model formulation is used to simulate
tified, p
realizations of the ground motion. These realiza-
tions are all different due to the stochasticity of the Simulating a Suite of Ground Motions
model (e.g., the white-noise process in the above for a Specified Design Scenario
example), but they all have the same model param-
eters and expected characteristics similar to those Here, a design scenario refers to a set of earth-
of the target motion. In this way, the target quake and site characteristics such as earthquake
accelerogram may be regarded as a single realiza- magnitude, source-to-site distance, and soil con-
tion of the ground motion process for a specified set ditions of the site. In the broader context of
of model parameters, while the simulated motions PBEE, an ensemble of ground motions that rep-
may be regarded as other random samples of the resents all possible ground shakings for a speci-
process for the same set of model parameters. fied design scenario is of interest, not just ground
motion simulations with characteristics similar to
Post-Processing for Long Spectral Periods those of a previously observed motion. A good
(Low Frequency) stochastic ground motion model must properly S
In general, site-based stochastic ground motion represent the natural variability of real earth-
models fail to match the response spectrum asso- quake ground motions. It also must be practical
ciated with the target accelerogram in the long for use by a practicing engineer.
spectral period range and tend to overestimate the
structural response (typically beyond 2–4 s Variability
depending on the model). This is because a sto- The variability among an ensemble of simulated
chastic model developed for an acceleration pro- ground motions for a specified design scenario
cess cannot guarantee zero velocity and comes from two sources: (1) the randomness of
displacement residuals (final values at the end ground motions for a specified set of model param-
of the record) upon integration of a sample real- eters (the stochasticity mentioned in the discussion
ization. This shortcoming of site-based models of the model parameters) and (2) the randomness
has been recognized by some modelers who of the model parameters for a specified design
“post-process” the simulations by extending scenario. The former is due to the stochastic nature
0.4
Acceleration,g
0.2
0 101
–0.2 Simulation before post-processing
5% Damped Pseudo-Acceleration
Simulation after post-processing
–0.4
100
0.15
Response Spectrum
Velocity, g.s
0.1
0.05 10–1
0
–0.05
10–2
Displacement, g.s2
0.5 10–3
0
–0.5 10–4
0 2 4 6 8 10 12 14 16 18 20 10–1 100 101
Time, s Period, s
Stochastic Ground Motion Simulation, Fig. 2 An example stochastic simulation before and after post-processing
using a high-pass filter with a cutoff frequency of 0.2 Hz
of the model and it is achieved when fitting to and physical meaning of a parameter. If correlations
simulating a target accelerogram. The latter is between the parameters are ignored, the variabil-
referred to as “parametric uncertainty.” Both ity in the resulting ensemble of ground motions
should be modeled properly to reproduce in the may be overestimated because the parameter
synthetic motions the variability present in real realizations are not well constrained.
earthquake ground motions.
The underestimation of parametric uncertainty Practicality
results in the underestimation of the natural vari- There is a need for generating synthetic motions
ability of ground motions. This has been a major that correspond to a specified design scenario, but
problem in the current practice of seismic hazard most site-based stochastic models limit their scope
analysis and the generation of synthetic ground to generating synthetic motions similar to a target
motions. A few exceptions are some recent site- recorded ground motion. In such models, no
based models (e.g., Rezaeian and Der Kiureghian attempt is made to select a different but appropriate
2010), where model parameters are fitted to a set of model parameters for the specified design
database of ground motions and are randomized scenario. Moreover, for a model to be practical and
at the time of simulation to achieve variability to be used in the engineering community, it should
similar to the sample database. Most source- be expressed in terms of variables that are readily
based models account for the variability in ground available to a design engineer.
motions by manually varying the source parame- In practice, GMPEs are widely used because
ters; whether the realizations follow the “true” they predict measures of ground motion intensity
underlying probability distribution for each in terms of common earthquake and site charac-
parameter is the question in these models. teristics that define a design scenario to a practic-
To properly account for parametric uncer- ing engineer. Typical intensity measures
tainty, a good model needs to randomly generate modeled by GMPEs are peak ground motion
parameters according to their underlying proba- values (i.e., peak ground acceleration, velocity,
bility distributions and also to account for the and displacement) and elastic response spectra at
correlations between the parameters. Probability specified oscillator periods and damping ratios.
distributions for parameters and their correlations A few GMPEs for inelastic response spectra have
can be estimated empirically or based on the also been developed. These GMPEs are useful for
linear response-spectrum analysis or crude characteristics. The error can then be randomized
nonlinear analysis, but not for response-history to generate the realizations of the model parame-
analysis, as they do not predict ground motion ter. To generate the realizations of multiple param-
time series. Such simplified analysis methods eters, correlations between the parameters should
have proven to be adequate for simple code- be taken into account in the simulation.
based design purposes, but not for PBEE.
Stochastic ground motion models can comple-
ment GMPEs by generating ground motion time Extensions of Site-Based Stochastic
series in terms of similar earthquake and site Models
characteristics used by GMPEs. This is done by
developing predictive relations for the model Multiple-Component Simulation
parameters in terms of commonly used earth- Earthquake ground motions are
quake and site characteristics. multidimensional. Neglecting the rotational com-
To adjust a stochastic model for practical use, ponents, three translational components are usu-
the model parameters are identified for a database ally expressed in two orthogonal horizontal
of recorded ground motions. Then, through directions and one vertical direction. For proper
regression analyses and empirical modeling dynamic analysis of complex structural and geo-
methods, predictive relations for each parameter technical systems, for example, asymmetric
in terms of earthquake and site characteristics are structures that are vulnerable to torsion or struc-
developed to relieve the design engineer from the tural components that are sensitive to vertical
task of predicting the parameter values for a ground shaking, it is important to simulate
design scenario (e.g., Sabetta and Pugliese multicomponent ground motion time series. The
1996; Stafford et al. 2009; Rezaeian and Der vast majority of stochastic models are restricted
Kiureghian 2010). A generic form of predictive to simulating a single horizontal ground motion
relations for a model parameter p is given below component, but the research to simulate multiple
components is continuing and under develop-
F1 ½FP ðpÞ ¼ mðEarthquake, Site, bÞ þ e ment (e.g., bidirectional ground motion simula-
tion by Rezaeian and Der Kiureghian (2012).
where F1 denotes the inverse of the standard When modeling and simulating multiple
normal cumulative distribution function, FP ground motion components, differences as well
denotes the marginal cumulative distribution func- as similarities and dependencies between the com-
tion estimated for parameter p, m is the conditional ponents must be taken into account. Because the
mean of F1[FP(p)] given the earthquake and site ground motion components are generated from the
characteristics of interest, and b is the vector of same earthquake source and from seismic waves
regression coefficients. The regression error e is a that travel through the same ground medium, high S
zero-mean normally distributed random variable correlations between model parameters of the
with standard deviation s representing the uncer- components are expected. Correlations and coher-
tainty of the parameter. The earthquake and site encies between the components can be estimated
characteristics are typically the earthquake magni- empirically by analyzing a large number of
tude, source-to-site distance, faulting mechanism, recorded ground motion components.
and various proxies to describe the site soil condi- For stochastic ground motions, in addition to
tions such as VS30 (i.e., the time-averaged shear cross-correlations between the model parameters,
wave velocity in the top 30 m of soil). These dependencies between the stochastic components
regression variables change depending on the (e.g., the seed white-noise processes) should also
region and the parameter under consideration. be taken into account. This is a difficult task, but
Given the predictive relations as described Penzien and Watabe (1975) showed that there is a
above, one can calculate the mean of each model unique set of orthogonal axes, referred to as “prin-
parameter given a set of earthquake and site cipal axes,” along which the three translational
components of ground motion may be considered presence of a two-sided, long-period, large-

uncorrelated. These are the major, intermediate, amplitude velocity pulse that typically shows as a
and minor principal axes along which the ground long-period pulse in the response spectrum and
motion components have intensities in decreasing may impose extreme demands on structures.
order and the stochastic processes (e.g., the seed Because near-fault recordings that contain direc-
white-noise processes) are statistically indepen- tivity pulses are relatively scarce and because scal-
dent. Although the stochastic processes can be ing and modification methods cannot create a pulse
assumed statistically independent, there can still in a record that does not initially contain a pulse,
be high correlations between the model parameters simulated near-fault ground motions are needed.
of the two ground motion components (see But only a few stochastic models currently exist
Rezaeian and Der Kiureghian 2012). Based on that simulate the forward directivity effect (e.g.,
examination of few real accelerograms, Penzien Mavroeidis and Papageorgiou 2003; Dabaghi
and Watabe (1975) suggested that the major prin- et al. 2011). This is usually done by modeling a
cipal axis usually points in the general direction of long-period velocity pulse that is a function of the
the epicenter, the intermediate principal axis is earthquake design scenario, including the relative
horizontal and perpendicular to the major principal location of the site to the fault rupture. The simu-
axis, and the minor principal axis is almost verti- lated velocity pulse is added to the derivative of the
cal. The concept of principal axes has been the simulated acceleration time series resulting in a
basis for many subsequent studies on simulated pulse-like near-fault ground motion.
multicomponent stochastic modeling of ground Not all near-fault ground motions contain a
motions, but more studies are required to confirm pulse. For example, backward directivity occurs
the direction of the principal components. If sim- when the fault rupture propagates away from the
ulation is done along the horizontal principal axes, site. The resulting ground motion tends to be of
an orthogonal pair of acceleration time series a1(t) low intensity and long duration. A valid near-fault
and a2(t) can easily be rotated to the horizontal ground motion model should take into account
direction of interest using a rotation matrix both the pulse-like and non-pulse-like cases.

The fling step is caused by the permanent
a1, y ðtÞ cos ðyÞ sin ðyÞ a1 ð t Þ displacement of the fault and is usually charac-
¼
a2, y ðtÞ sin ðyÞ cos ðyÞ a2 ð t Þ terized by a one-sided velocity pulse in the fault-
parallel direction. Modeling of the fling step is
where a1,y(t) and a2,y(t) are the counterclockwise not as well developed in stochastic ground
rotations by angle y. motion modeling, but its effect on structural
response is not as extreme as the forward direc-
Near-Fault Simulation tivity effect. The fling step is difficult to model
Ground motions close to fault ruptures (within because information is usually lost in the typical
about 30 km), i.e., ▶ near-fault ground motions, processing of recorded ground motions.
can have distinct characteristics such as the direc-
tivity effect in the fault-normal direction and the Spatial Simulation (Multiple Stations)
fling step in the fault-parallel direction. These Most stochastic ground motion models are devel-
characteristics can have a strong influence on oped for single-station simulations. To simulate
the structural response and are usually modeled ground motions from a single earthquake event at
separately and then added to a site-based stochas- multiple stations that are located close to each
tic ground motion simulation. other, the spatial coherency between records
A site near a fault may experience forward should be taken into account. This coherency usu-
directivity, which occurs when the fault rupture ally depends on the frequency content of the
propagates toward the site with a velocity almost ground motions and separation distance between
equal to the shear wave velocity. The resulting the stations and can be estimated from the power
ground motion time series is characterized by the spectral density functions of recorded motions.
Several models exist in the literature that can be recorded accelerogram by identifying the model
used to adjust a ground motion time series for parameters through optimization and fitting into
spatial variation (e.g., Liao and Zerva 2006; certain target characteristics of real earthquake
Konakli and Der Kiureghian 2012, Konakli ground motions. The parameters are then used
et al. 2014). These models can be applied to a in the formulation of the stochastic process.
stochastic simulation of ground motion at a single Post-processing is usually required for site-
station to generate consistent simulations at nearby based stochastic models to ensure zero residuals
stations. and proper modeling of long-period spectral
values. Simulating a suite of ground motions for
a specified design scenario is discussed, where
Validation of Ground Motion Simulation the natural variability of ground motions must be
Models modeled properly by accounting for parametric
uncertainty and correlations between the model
Although many stochastic ground motion simula- parameters. Practicality of a model is also
tion models exist, no standard validation tech- discussed, where the model parameters are
niques are available. There are almost as many expressed in terms of earthquake and site charac-
validation approaches undertaken as there are sim- teristics that define a design scenario and are
ulation models. Comparisons against empirical readily available to a design engineer. Finally,
data and trusted models in engineering practice adjusting site-based stochastic ground motion
aid in identifying the limitations of the proposed simulations for special cases including
simulation models and may encourage implemen- multicomponent, near-fault, and multi-station
tation of simulations in practice. The next step in simulations is discussed. Currently, many
the field of ground motion simulation that has been research efforts are directed toward developing
gaining attention in recent years is development of ground motion simulation validation procedures
standard techniques for ground motion simulation for various engineering applications.
validation for use in various engineering applica-
tions (see, e.g., the efforts of the Southern Califor- Acknowledgment We would like to thank Mayssa
nia Earthquake Center ground motion simulation Dabaghi, Morgan Moschetti, Ricardo Taborda, and the
anonymous reviewer of this entry for their review and
validation technical activity group http://collabo-
valuable feedback. Any use of trade, product, or firm
rate.scec.org/gmsv/Main_Page). names is for descriptive purposes only and does not
imply endorsement by the US government.
Summary
Cross-References
Stochastic ground motion simulations are used to S
replace or supplement recorded earthquake ▶ Physics-Based Ground-Motion Simulation
ground motions when recorded motions are
sparse. Stochastic ground motion models are cat- References
egorized into source-based and site-based
models. Definitions and examples for each cate- Aki K (1967) Scaling law of seismic spectrum. J Geophys
gory are given. Source-based simulations explic- Res 72:1217–1231
Au SK, Beck JL (2003) Subset simulation and its applica-
itly model the source spectrum, path, and site tions to seismic risk based on dynamic analysis. J Eng
effects. Site-based stochastic models are classi- Mech ASCE 129:901–917
fied into four groups depending on their formula- Beresnev IA, Atkinson G (1998) Stochastic finite-fault
tions. Regardless of the formulation, the model modeling of ground motions from the 1994 northridge,
California, earthquake (I): validation on rock sites.
must exhibit the characteristics of real earthquake
Bull Seismol Soc Am 88(6):1392–1401
ground motions including temporal and spectral Boore DM (2003) Simulation of ground motion using the
nonstationarities. Site-based models simulate a stochastic method. Pure Appl Geophys 160:635–676
3496 Stochastic Structural Identification from Vibrational and Environmental Data
Boore DM (1983) Stochastic simulation of high-fre- Rezaeian S, Der Kiureghian A (2008) A stochastic ground
quency ground motions based on seismological motion model with separable temporal and spectral
models of the radiated spectra. Bull Seismol Soc Am. nonstationarities. Earthq Eng Struct 37:1565–1584
73:1865–1894 Rezaeian S, Der Kiureghian A (2010) Simulation of syn-
Boore DM (2009) Comparing stochastic point-source and thetic ground motions for specified earthquake and site
finite-source ground-motion simulations: SMSIM and characteristics. Earthq Eng Struct 39:1155–1180
EXSIM. Bull Seismol Soc Am 99:3202–3216 Rezaeian S, Der Kiureghian A (2012) Simulation of
Conte JP, Peng BF (1997) Fully nonstationary analytical orthogonal horizontal ground motion components for
earthquake ground-motion model. J Eng Mech ASCE specified earthquake and site characteristics. Earthq
12:15–24 Eng Struct 41:335–353
Conte JP, Pister KS, Mahin SA (1992) Nonstationary Sabetta F, Pugliese A (1996) Estimation of response spec-
ARMA modeling of seismic motions. Soil Dyn Earthq tra and simulation of nonstationary earthquake ground
Eng 11:411–426 motions. Bull Seismol Soc Am 86:337–352
Dabaghi M, Rezaeian S, Der Kiureghian A (2011) Sto- Saragoni GR, Hart GC (1974) Simulation of artificial
chastic simulation of near-fault ground motions for earthquakes. Earthq Eng Struct 2:249–267
specified earthquake and site characteristics. In: Pro- Shinozuka M, Deodatis G (1988) Stochastic process
ceedings of the 11th international conference on appli- models for earthquake ground motion. Probabilist
cations of statistics and probability in civil engineering Eng Mech 3:114–123
(ICASP11), Zurich, pp 759–760 Shinozuka M, Sato Y (1967) Simulation of nonstationary
Der Kiureghian A, Fujimura K (2009) Nonlinear stochas- random process. J Eng Mech ASCE 93:11–40
tic dynamic analysis for performance-based earth- Somerville P, Collins N, Abrahamson N, Graves R, Saikia
quake engineering. Earthq Eng Struct 38:719–738 C (2001) Ground motion attenuation relations for the
Housner GW, Jennings PC (1964) Generation of artificial Central and Eastern United States – Final report,
earthquakes. J Eng Mech Div ASCE 90:113–150 30 June 2001: Report to U.S. Geological Survey for
Konakli K, Der Kiureghian A (2012) Simulation of spa- award 99HQGR0098, 38 p
tially varying ground motions including incoherence, Stafford PJ, Sgobba S, Marano GC (2009) An energy-
wave-passage and differential site-response effects. based envelope function for the stochastic simulation
Earthq Eng Struct 41:495–513 of earthquake accelerograms. Soil Dyn Earthq Eng
Konakli K, Der Kiureghian A, Dreger D (2014) Coher- 29:1123–1133
ency analysis of accelerograms recorded by the Yeh CH, Wen YK (1990) Modeling of nonstationary
UPSAR array during the 2004 Parkfield earthquake. ground motion and analysis of inelastic structural
Earthq Eng Struct 43:641–659 response. Struct Safety 8:281–298
Liao S, Zerva A (2006) Physically compliant, condition- Zerva A (1988) Seismic source mechanisms and ground
ally simulated spatially variable seismic ground motion models, review paper. Probabilist Eng Mech
motions for performance-based design. Earthq Eng 3:64–74
Struct 35:891–919
Lin YK (1986) On random pulse train and its evolutionary
spectral representation. Probabilist Eng Mech
1:219–223
Liu SC (1970) Synthesis of stochastic representations of Stochastic Structural Identification
ground motions. Bell Sys Tech J 49:521–541
Mavroeidis GP, Papageorgiou AS (2003) A mathematical from Vibrational and Environmental
representation of near-fault ground motions. Bull Data
Seismol Soc Am 93(3):1099–1131
McGuire RK, Hanks TC (1980) RMS accelerations and Minas D. Spiridonakos and Eleni N. Chatzi
spectral amplitude of strong ground motion during the
San Fernando, California, earthquake. Bull Seismol Department of Civil, Environmental and
Soc Am 70(5):1907–1919 Geomatic Engineering, ETH Zurich, Institute of
Motazedian D, Atkinson GM (2005) Stochastic finite-fault Structural Engineering, Zurich, CH, Switzerland
modeling based on a dynamic corner frequency. Bull
Seismol Soc Am 95(3):995–1010
Papadimitriou K (1990) Stochastic characterization of
strong ground motion and application to structural Synonyms
response. Earthquake Engineering Research Labora-
tory, California Institute of Technology, Pasadena, ARMA model; Environmental conditions;
Report No. EERL 90–03
Penzien J, Watabe M (1975) Characteristics of 3- Output-only identification; Polynomial chaos
dimensional earthquake ground motions. Earthq Eng expansion; Structural identification; Vibration
Struct 3:365–373 response
Stochastic Structural Identification from Vibrational and Environmental Data 3497
Introduction Structures Operating Under Varying

Conditions and Their Responses
In recent years, Structural Health Monitoring
(SHM) of engineering structures has become an A linear viscously damped structure may be
important area of research. This is a broad term described by the following lumped parameter,
which by large also deals with the online detec- time-varying ordinary differential equation:
tion and identification of structural damage and
therefore holds a critical significance for the case
MðtÞx€ðtÞ þ CðtÞx_ ðtÞ
of large-scale civil structures where a failure inci- (1)
dence is connected not only to high cost losses þ KðtÞxðtÞ ¼ uðtÞ, for t t0 , tf
but potentially to human loss in the worst-case
scenario. where t designates analog time with t0 denoting
Vibration-based methods are today the fastest the start and tf the end of the structure’s lifetime
growing research area in the SHM field (Fassois or inspection interval, x(t) the structural displace-
and Kopsaftopoulos 2013). However, while such ments response vector, u(t) the applied excitation
methods are already used for industrial mechan- forces vector, and M(t), C(t), K(t) the mass,
ical structures, these are still far from being damping, and stiffness matrices of the structure,
successfully implemented in large-scale civil respectively.
structures such as high-rise buildings and brid- The variability of structural response may be
ges. The main reason for this discrepancy is the usually attributed to a small number of parame-
fact that these structures do not operate in the ters related with either inherent properties of the
controlled environment of an industry or structure or exogenous random variables, e.g.,
a laboratory facility but rather they are affected environmental and loading conditions, geometry,
by a number of uncontrollable environmental mass distribution, and so on. Gathering these
conditions, such as temperature, temperature L parameters in a single vector j(t) = [x1(t),
spatial gradients due to solar irradiation, tempo- x2(t), . . ., xL(t)]T the previous model may be
ral temperature gradient due to thermal capacity, rewritten as
humidity, and others (Sohn 2007). The short-
comings of vibration-based SHM methods have
been recognized already in one of the first review Mðj ðtÞÞx€ðtÞ þ Cðj ðtÞÞx_ ðtÞ þ Kðj ðtÞÞxðtÞ ¼ uðtÞ
papers on the subject (Doebling et al. 1998) (2)
exposing the advantages and capabilities of sev-
eral methodologies in this area.
Focusing on environmental conditions and
For the aforementioned reason, one is faced
assuming that they are characterized by slow
with the problem of identifying a dynamic model S
(compared to the structural dynamics) variation,
that is able to represent a given structure for an
the previous model may be rewritten as
extended spectrum of operational conditions – if
not its entire life cycle. Toward this end, the
combined vibrational response and environmen- xðtÞ þ Cðj Þx_ ðtÞ
Mðj Þ€
tal data, which are normally available in current
þ Kðj ÞxðtÞ ¼ uðtÞ, for t ½ta , tb t0 , tf
SHM systems, have to be utilized along with
(3)
appropriate models that can account for the var-
iability of structural dynamics due to environ-
mental conditions. The approach detailed herein The latter may be considered as a local time-
treats this issue through a functional representa- invariant model which represents the structure
tion that introduces this dependence into the when monitored during a small time interval.
model’s parameters. Within this interval the input parameter vector,
j and as a consequence the mechanical proper- sequence e[t]. The latter is typically considered to
ties of the structure, may be considered to be be a normally independently distributed (NID)
“fixed” to constant values. Then, the input sequence, that is, a white noise process, with
parameter vector j may be considered as zero mean and constant variance s2e . Therefore,
a realization of the random vector X with joint the analytical model of Eq. 5 is approximated by
probability density function (pdf) fX(j). In this a model of the following form:
way the initial linear time-varying (LTV) model
of Eq. 1 may be written as a linear time-invariant
(LTI) model: xðtÞ þ a1 ðXÞx½t 1 þ þ ana ðXÞx½t na
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
AR part
MðXÞx€ðtÞ þ CðXÞx_ ðtÞ ¼ e½t þ c1 ðXÞe½t 1 þ þ cnc ðXÞe½t nc ,
(4) |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
þ KðXÞxðtÞ ¼ uðtÞ, with XðoÞ O
MA part
e½t NID 0, s2e , for t ¼ 1, .. .,N
which is however characterized by parameters (6)
that are depending on the input variable vector
X. In the equation above, O designates the event
which has the form of a conventional ARMA
space of the random vector X, and o an elemen-
model (Ljung 1999), with the important differ-
tary event of this space.
ence that it is characterized by stochastic param-
The above continuous-time model obviously
eters which depend on the input variable
has a discrete-time counterpart, which for the
vector X. These parameters may actually be
simpler single-input single-output case may be
represented by a deterministic mapping describ-
approximated by a scalar difference equation of
ing their relation with the input random variables.
the following form (Andersen 1997):
Specifically, assuming that the ARMA model
parameters ai(X) and ci(X) have finite variance,
xðtÞ þ a1 ðXÞx½t 1 þ þ ana ðXÞx½t na they admit the following representation (Soize
¼ b0 ðXÞu½t þ b1 ðXÞu½t 1 þ and Ghanem 2004):
þ bnb ðXÞu½t nb , for t ¼ 1, . . . , N
X
1
(5) ai ðXÞ ¼ ai, j ’dðjÞ ðXÞ ði ¼ 1, . . . , na Þ,
j¼1
in which t designates normalized discrete time X

1
ci ðXÞ ¼ ci, j ’dðjÞ ðXÞ ði ¼ 1, . . . , nc Þ
(absolute time normalized by the sampling period j¼1
Ts), N the signals’ length in samples, x[t], u[t] (7)
the discretized versions of the observed
vibration response and excitation force, respec-
with ai, j and ci, j designating unknown determin-
tively, ai(X), bi(X) the model parameters
istic coefficients of projection onto the probabil-
depending on X, and finally na, nb the equation
ity space of X and ’d(j) are multivariate basis
orders.
functions that are orthonormal with respect to
Structural identification is an inverse problem
the joint pdf of X, that is,
which concerns the estimation of an appropriate
mathematical representation of the structural
n
dynamics based on vibrational measurements. E ’a ðXÞ’b ðXÞ ¼ 01,, for a¼b
otherwise (8)
In the case of ambient loading, the excitation
may not be measured with accuracy, and thus,
the input sequence u[t] in the theoretical model of and d(j) is the multi-indices of the multivariate
Eq. 5 has to be replaced by an unknown excitation basis. Each probability density function may be
associated with a well-known family of orthogo- of deterministic coefficients of projection ai, j

nal polynomials. For instance, a normal distribu- and ci, j while a specific PC-ARMA model struc-
tion is associated with Hermite polynomials ture is fully defined by the model orders na, nc,
while a uniform distribution with Legendre. and the basis total degree P.
A list of the most common probability density At this point it should be noted that similar
functions along with the corresponding orthogo- models have been used for the structural identifi-
nal polynomials and the relations for their con- cation and damage detection issues in a number
struction may be found in Xiu and Karniadakis of recent studies (Functionally Pooled ARX
(2002). For the case of independent random vari- models; see Kopsaftopoulos and Fassois 2013
ables Xj the expansion of Eq. 7 is called polyno- and the references therein) and automatic control
mial chaos (PC) expansion (Blatman and Sudret within the Linear Parameter Varying (LPV)
2010), and the resulting models may be referred models framework (Tóth 2010, Zhao et al. 2012).
to as PC-ARMA models.
Of course, for purposes of practicality, the
infinite series of PC expansion of Eq. 7 must be PC-ARMA Model Identification
truncated by selecting an appropriate functional
subspace consisting of a finite number of terms. The complete PC-ARMA structural identifica-
A usual approach lies in the selection of tion problem may be stated as follows: Given
a functional subspace consisting of polynomial a set of dynamic vibration response signals
basis functions with total maximum degree P, obtained from the structure under study for vari-
that is, jd(j)j = L‘ = 1d(j, ‘) P for all j. In ous measured environmental conditions, select
this case the dimensionality of the functional the PC-ARMA model structure and estimate the
subspace is equal to corresponding model parameters that best fit the
available records. The issues of PC-ARMA
ðL þ PÞ! model parameter estimation and model structure
p¼ (9) selection are discussed in the following sections.
L!P!
Model Parameter Estimation
where L is the number of random variables and
As already mentioned, the estimation of a
P the maximum basis total degree. For instance,
PC-ARMA model refers to the determination of
for a random input vector of dimension equal to
the coefficients of projection assembled in the
three (L = 3) and maximum degree equal to two
parameter vector
(P = 2), the multi-indices vectors are:
T
l u ¼ a1, 1 a1, 2 ana , p c1, 1 c1, 2 cnc , p
1 2 3
S
d(1) 0 0 0
or a subset of u selected by the procedure
d(2) 1 0 0
described in the next section. Its estimation is
d(3) 0 1 0
based on the availability of structural vibration
d(4) 0 0 1
response data and measurements of the environ-
d(5) 2 0 0
d(6) 0 2 0
mental conditions of interest. In order to identify
d(7) 0 0 2 a stochastic model that will be able to describe the
d(8) 1 1 0 structure’s dynamics for the wide range of its
d(9) 1 0 1 operational spectrum, data from a sufficient num-
d(10) 0 1 1 ber of operating conditions should be available.
Considering K such datasets have been acquired
In this way, the resulting PC-ARMA model is during an equal number of experiments conducted
fully parametrized in terms of a finite number under different input random vector realizations
T vibration response signals xNk = {xk[1], xk[2], . . .,

j k ¼ j 1k j 2k . . . j Lk , for k ¼ 1, 2 . . . , K: Based
xk[N]}T follow the PC-ARMA model of Eqs. 6
on the validity of the assumptions described in the
and 7:
previous section, it is considered that the recorded
X
p X
p
xk ½t þ a1, j ’dðjÞ ðj k Þxk ½t 1 þ þ ana , j ’dðjÞ ðj k Þxk ½t na
j¼1 j¼1
X
p X
p (9a)
¼ ek ½t þ c1, j ’dðjÞ ðj k Þek ½t 1 þ þ cnc , j ’dðjÞ ðj k Þek ½t nc ,
j¼1 j¼1
for t ¼ 1, . . . , N, and k ¼ 1, . . . , K

ek NID 0, s2ek (9b) coincide with the model’s residual sequence
ek[t] (Ljung 1999), and thus, u has to be estimated
through the following minimization problem:
s2ei dt, t for i¼j
E ei ½tej ½t t ¼ 0 for i6¼j (9c)
( )
XK X N
ðxk ½t x^k ½tjt 1Þ2
with di, j designating the Kronecker delta func- u ¼ arg min
y
k¼1 t¼1
g2k
tion. In the relationships above, the ( )
uncorrelatedness of the residual series between XK X N
e^k ½t2
¼ arg min
different vibration response records has been y
k¼1 t¼1
g2k
added to the model assumptions.
(10)
The estimation of the PC-ARMA model
parameter vector u may be based on the minimi-
zation of the Prediction Error (PE) criterion. The with x^k ½tjt 1 denoting the model’s one-step-
PE criterion consists of the sum of squares of the ahead prediction, gk user defined weights, and
model’s one-step-ahead prediction errors for the argmin minimizing argument with respect to the
complete set of datasets. As it may be easily indicated quantity. Toward this end, Eq. 9 may be
shown, the one-step-ahead prediction errors rewritten in a regression type form as
na X
X p nc X
X p
xk ½t ¼ ai, j ’dðjÞ ðj k Þxk ½t i þ ci, j ’dðjÞ ðj k Þ½t i þ ek ½t
2i¼1 j¼1 3 2 i¼1 j¼13
’dð1Þ ðjk Þxk ½t 1 a1, 1
6 ’dð2Þ ðjk Þxk ½t 1 7 6 a1, 2 7
6 7 6 7
6 ⋮ 7 6 ⋮ 7
6 7 6 7
6 ’dðp1Þ ðjk Þxk ½t na 7 6 ana, p1 7
6 7 6 7
6 ’dðpÞ ðjk Þxk ½t na 7 6 ana, p 7
6 76 7 (11)
6 ’dð1Þ ðjk Þek ½t 1 7 6 c1, 1 7 þ e^k ½t
6 7 6 7
6 ’dð2Þ ðjk Þek ½t 1 7 6 c1, 2 7
6 7 6 7
6 ⋮ 7 6 ⋮ 7
6 7 6 7
4 ’dðp1Þ ðjk Þek ½t nc 5 4 cnc, p1 5
’dðpÞ ðjk Þek ½t nn cnc, p
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflffl{zfflfflfflfflfflffl}
’½jk, t
T u
or by stacking all time instants in a single vector sets and expanding their parameters onto the PC
basis (Zhao et al. 2012).
2 3
a1, 1 However, in the simpler PC-AR case, the
6 a1, 2 7 regression model of Eq. 13 is linear, and thus, u
6 7
6 ⋮ 7 may be estimated by the linear weighted least
2 3 2 3 6 7 2 3
x k ½ 1 w½j k , 1T 6 ana, p1 7 ek ½1
6 7 squares (WLS) estimator:
6 xk ½2 7 6 w½j , 2T 7 6 ana, p1 7 6 ek ½2 7
6 7 6 7 6 7 6 7
4 ⋮ 5 ¼ 4 ⋮ 5 6 c1, 1 7 þ 4 ⋮ 5
k
1
6 7 u^ 5 Fðj ÞT G Fðj Þ Fðj ÞT G Y (15)
xk ½N w ½j k , N T 6 c1, 2 7 ek ½N
ffl} 6
|fflfflfflfflffl{zfflfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl 7
6 ⋮ 7 |fflfflfflfflffl{zfflfflfflfflffl}
xNk Fðj k Þ 6 7 eNk
4 cnc, p1 5 The simplest and obvious choice for the weights
cnc, p gk is to set them equal to unity, with this selection
|fflfflfflfflfflffl{zfflfflfflfflfflffl}
u leading to the ordinary least squares (OLS) esti-
(12) mator. However, it has been shown for the similar
FP-ARX models (Kopsaftopoulos and Fassois
Finally, by pooling all the available datasets, the 2006) that the weights that lead to consistent
following regression model is obtained: and asymptotically efficient estimates are those
equal to the standard deviation of the residual
2 3 2 3 2 N3 sequence, that is,
xN1 F ðj 1 Þ e1
6 xN 7 6 Fðj 2 Þ 7 6 eN 7
6 27¼6 7 6 27 g k ¼ s ek , for k ¼ 1, 2, . . . , K
4 ⋮ 5 4 ⋮ 5 u þ 4⋮5 (13) (16)
xNK Fðj K Þ eNK Due to the fact that the true residual standard
|fflfflffl{zfflfflffl} |fflfflfflfflfflfflffl{zfflfflfflfflfflffl
ffl} |fflfflffl{zfflfflffl}
X Fðj Þ E deviation is typically unknown, it has to be
replaced by an appropriate estimate, typically
where F(j) is the regression matrix. Thus, the obtained via the OLS estimation which is used
minimization problem of Eq. 7 may be rewritten as a first stage. The WLS estimator may then be
as follows: realized in an iterative manner till convergence of
the PE criterion is achieved.

u 5 arg min ET G E (14)
u Model Structure Selection
N Model structure selection is the optimization pro-
where G = IN g with g = [g1 gk]T denoting cedure during which models corresponding to
the vector of weights, IN the N N identity
N various candidate “structures” are estimated and
matrix, and the Kronecker product. weighted according to an appropriately selected
Due to the nonlinear dependence of the a priori, fitness criterion. Model “fitness” may be consid- S
unknown residual sequence E on the parameter ered in terms of various criteria, such as the
vector u, the last equation leads to a nonlinear- residual sum of squares (RSS)
weighted least squares problem, which has to be
tackled by nonlinear optimization methods. How- X
K X
N
ever, nonlinear least squares techniques are sensi- RSS ¼ ek ½t2 (17a)
tive to the initial parameter values and if no k¼1 t¼1
accurate estimates are available, the minimization

or the equivalent mean squared error (MSE)
procedure is very likely to converge to a local
minimum. In order to avoid potential inaccurate RSS
convergence problems associated with arbitrary MSE ¼ (17b)
NK
initial estimates, initial values for the coefficients
of projection may be obtained by identifying con- However, the limitation of using such criteria is
ventional ARMA models for each of the K data that they may monotonically decrease with
increasing model structure and as a result may

lead to overfitting. For this reason, caution has to
be exercised and a compromise has to be made
between model accuracy and model structure
complexity.
The model structure selection procedure for
a PC-ARMA model concerns the determination
of the set of 3 integers (na, nc, P) for obtaining the
best fitting model, and a combinatorial approach,
that is, estimating PC-ARMA models for various
integer sets, may be used. However, a simpler
approach that may lead to sparse PC basis by
selecting only a subset of the initial PC functional
subspace may be described as follows
(Poulimenos and Fassois 2006):
Phase I. AR/MA orders selection. In order to

“decouple” the selection of the model orders
(na, nc) from that of functional subspaces, their
interaction has to be minimized. For this rea-
son a high-degree P and the complete set of PC
basis functions may be initially adopted.
When employing these, AR/MA model order
selection may be achieved through trial-and-
error techniques based on the values of the
fitness function.
Phase II. PC basis subspace selection. The aim
of this phase is the optimization of the PC
functional subspace, in the sense of reducing
model complexity without significantly reduc-
ing model accuracy. This may be accomplished
by dropping, one at a time, the PC basis func-
tions whose removal does not cause significant
loss of the criterion employed.
A flowchart of the complete PC-ARMA iden-

tification procedure is given in Fig. 1.
Validation
Once a PC-ARMA model has been obtained, it
must be validated with respect to the assumptions Stochastic Structural Identification from Vibrational
behind the identification method. A standard val- and Environmental Data, Fig. 1 Flowchart of the
PC-ARMA model identification procedure
idation procedure in the simple ARMA case
involves the testing of Gaussianity and, in partic-
ular, the whiteness of the identified model’s additional assumption of cross-uncorrelatedness
one-step-ahead prediction error sequence. In the between the datasets has to be validated.
PC-ARMA case this assumption has to be vali- The validation procedure of the estimated
dated for each of the recorded datasets, while the PC-ARMA model may be based on the typical
Stochastic Structural Identification from Vibrational and Environmental Data, Fig. 2 The complete stochastic
structural identification framework based on PC-ARMA models
cross-validation principle which presumes the The complete stochastic structural identification
separation of the processed signal into an estima- framework based on PC-ARMA models is illus-
tion set and a smaller validation set of data (Ljung trated in the abstract schematic representation of
1999). Fig. 2. It is noted that the estimated PC-ARMA
model parameters may be used for the calculation
Summary of the Method of the structure’s modal characteristics, such as
Step 1. Collect K datasets of vibration response the modal frequencies and damping ratios, as
and environmental variables measurements a function of the input variable vector X for the
from the structure under study for a range of entire event space O.
operating conditions.
Step 2. Fit a probability density function to the
environmental variables based on the avail- Numerical Case Study
able data and select the corresponding family S
of PC basis functions. The modeling of a 3-DOF system (Fig. 3) is
Step 3. Select a high total degree for the PC basis presently considered for demonstrating the work-
P; estimate PC-ARMA models of various ings of the PC-ARMA identification framework.
orders (e.g., considering common AR/MA The mechanical properties of the system are:
order, from na = nc = nmin up to
na = nc = nmax); and select the final orders m1 ¼ 102 , m2 ¼ 101:5 , m3 ¼ 10 ðkgÞ
based on the “fitness” criterion (e.g., MSE).
Step 4. Estimate PC-ARMA (na, nc) models by k1 ¼ 106 , k2 ¼ 105:5 , k3 ¼ 105 ðN=mÞ
rejecting each time a number of PC basis
function from the complete PC subspace. c1 ¼ 100, c2 ¼ 10, c2 ¼ 10, N=ms1
Reject the functions whose removal does not
imply significant reduction of the “fitness” The system is considered to operate under vary-
criterion value. ing conditions of temperature and humidity.
Stochastic Structural Identification from Vibrational and Environmental Data, Fig. 3 Three degree of freedom
system and its properties
The system’s stiffness coefficients are considered 15 and variance equal to 25 for the temperature,
to be linearly depended on the temperature that is, X2 N (15, 25). The 50 samples of the
through the following relationship: input variables are shown in Fig. 4.
In a real-world structural identification appli-
k0i ¼ ki ½1 ðT T ref ÞkT cation, where no information is available regard-
ing the true pdfs of the input random vector,
with kT = 5 103 denoting the thermal coeffi- someone could use maximum likelihood estima-
cient and Tref = 20 C the reference temperature, tion fitting of the environmental condition data
while humidity is modeled as mass increment due values to a parametric distribution. The results
to trapped humidity. For the sake of simplicity, in of such a fitting of the data onto pdfs are shown
this test case study, it is assumed that the temper- in Fig. 5. Based on this fitting and after
ature is uniform with no spatial gradients or tem- transforming the pdf of the mass load into
poral variations within each experiment, while a normal distribution by using the natural loga-
the mass load is evenly distributed over the rithm, the Hermite polynomials may be selected
system. for the construction of the multivariate PC basis
In order to identify an inverse model for this functions.
system under various operating conditions, the For each set of these “environmental” condi-
system’s response (acceleration of mass m3) is tions, a white noise realization of the excitation
simulated for various operating condition scenar- force u(t) N(0, 104) is used for the excitation
ios. Specifically, 50 samples of the added mass of the system. System’s response is obtained
and temperature variables are drawn, by means of by using the Runge-Kutta algorithm for 1,800
the Latin hypercube sampling (LHS) method, points with sampling frequency equal to
from their theoretical pdfs which are considered 64 Hz. After discarding the first 1,000 samples
to correspond to a log-normal distribution with in order to avoid transient effects, the rest of the
mean value equal to 1 and variance equal to 0.25 data are separated into an estimation set
for the mass load, that is, X1 ln N (1, 0.25), and consisting of 500 samples and a validation set
a normal distribution with mean value equal to of 300 samples.
Stochastic Structural Identification from Vibrational and Environmental Data, Fig. 4 Input variable values for
the 50 simulation experiments conducted
Stochastic Structural Identification from Vibrational corresponding probability density and cumulative distri-
and Environmental Data, Fig. 5 The empirical density bution functions estimated by the maximum likelihood
and distribution functions calculated based on the 50 real- method
izations of the input variables, compared to the
The estimation dataset (50 sets 500 sam- MSE criterion and the procedure described in
ples) is used for the identification of the system section 2.2. The maximum total degree of the
through simple PC-AR models in what follows. PC multivariate basis is initially selected
PC-AR model structure selection is based on the equal to 5, which using Eq. 9 with P = 5, and
Stochastic Structural Identification from Vibrational . . ., 20 and P = 5, and (b) MSE criterion values obtained
and Environmental Data, Fig. 6 PC-AR model struc- from a PC-AR(6) model by dropping, one at a time, the PC
ture selection results: (a) MSE criterion values for basis functions of the initial subspace
WLS-based estimated PC-AR models with na = 2, 3,
L = 2 gives a total number of 21 candidate basis Stochastic Structural Identification from Vibrational
functions. These functions are firstly used for the and Environmental Data, Table 1 Multivariate indi-
ces of the finally selected PC basis functions of the PC-AR
AR order selection by estimating PC-AR models (6) model
of various orders (na = 2, 3, . . ., 20) by means of
l
the WLS estimator. The results illustrated in
1 2
Fig. 6a show an AR order equal to six as appro-
d(1) 0 0
priate for capturing the dynamics of the simu-
d(2) 1 0
lated system. At the second stage of the structure d(3) 0 1
selection procedure, a sparse PC basis is sought. d(4) 2 0
As shown in Fig. 6b, this comes down to a basis
of only 4 basis functions which are given in
Table 1. It may be observed that the basis The estimated PC-AR(6) model is success-
indicates a linear relationship of the PC-AR fully validated by calculating the one-step-
model parameters with respect to temperature ahead prediction errors for the validation dataset
and quadratic with respect to humidity (mass (50 sets 300 samples). Indicative results for
load). the autocorrelation function of the resulting
Stochastic Structural
Identification from
Vibrational and
Environmental Data,
Fig. 7 PC-AR(6) model
validation results. (a) The
autocorrelation function for
the prediction error
sequence of the validation
set (simulation experiment
2; time lags 100; 95 %
confidence limits) and (b)
the covariance matrix of the
prediction error matrix E
prediction error sequence corresponding to the Summary

second simulation experiment are shown in
Fig. 7a, while also the assumption of Mechanical properties, and thus the dynamic
uncorrelatedness between the various prediction behavior of an engineering structure, are signifi-
error sequences may be qualitatively validated by cantly affected by environmental and operational
visually inspecting the residual covariance matrix conditions. A complete identification procedure
which is shown in Fig. 7b. has to lead to a consistent representation of the
Finally, the estimated coefficients of projec- structural dynamics for a great percentage, if not S
tion may be used for calculating the PC-AR(6)- the whole, range of the structure’s operational
based estimates of the system natural frequen- spectrum. This may be achieved either by
cies and thus for constructing a surface that a multi-model approach, that is, estimating local
shows the dependence of the natural frequencies models for every new set of environmental condi-
of the 3-DOF system on the two input random tions data or using functional models that incorpo-
variables. The results are plotted against the rate into their parameters the structure’s
theoretical curves in Fig. 8. It may be observed dependency on a set of random input variables
that there is a good agreement between the the- which correspond to the measured environmental
oretical and the estimated natural frequencies conditions. In this way, after the estimation
with the highest errors, which are below 1 Hz, (training) of the functional model on a given set
occurring at extreme values of the input of measurements, its parameters may be extrapo-
variables which were not included in the estima- lated for any new set of environmental conditions.
tion set. Presently, the functional models used are of the
True
PC-AR(6) based
estimates
Error
Stochastic Structural Identification from Vibrational curves. The natural frequencies are plotted versus the
and Environmental Data, Fig. 8 PC-AR(6)-based nat- input variables, that is added mass and temperature
ural frequency estimates compared to the theoretical
ARMA form with parameters that are expanded on References

polynomial chaos basis. The latter has the impor-
tant feature of being orthogonal to the probability Andersen P (1997) Identification of civil engineering
structures using vector ARMA models. Aalborg Uni-
density function of the input variable vector, pro-
versity, Department of Mechanical Engineering
viding in this way the best convergence rate with Blatman G, Sudret B (2010) An adaptive algorithm to
increasing functional subspace. build up sparse polynomial chaos expansions for sto-
chastic finite element analysis. Probabilist Eng Mech
25:183–197
Doebling S, Farrar C, Prime M (1998) A summary review
Cross-References of vibration-based damage identification methods.
Shock Vib Dig 30:91–105
Fassois SD, Kopsaftopoulos FP (2013) Statistical time
▶ Operational Modal Analysis in Civil
series methods for vibration based structural health
Engineering: An Overview monitoring. In: Ostachowicz W, Guemes A (eds)
▶ Response Variability and Reliability of New trends in structural health monitoring. Springer,
Structures Wien/London, pp 209–264
Kopsaftopoulos FP, Fassois SD (2006) Identification of
▶ Stochastic Analysis of Linear Systems
stochastic systems under multiple operating condi-
▶ System and Damage Identification of Civil tions: the vector dependent FP–ARX parameteriza-
Structures tion. In: Proceedings of the 14th Mediterranean
▶ Uncertainty Quantification in Structural Health conference on control and automation. Ancona
Kopsaftopoulos FP, Fassois SD (2013) A functional
Monitoring
model based statistical time series method for vibra-
▶ Vibration-Based Damage Identification: The tion based damage detection, localization, and magni-
Z24 Bridge Benchmark tude estimation. Mech Syst Signal Process 39:143–161
Strengthened Structural Members and Structures: Analytical Assessment 3509
Ljung R (1999) System identification: theory for the user, quality of materials, and various other weak-
2nd edn. Prentice Hall, Upper Saddle River nesses such as increased loading due to change
Poulimenos AG, Fassois SD (2006) Parametric time-
domain methods for non-stationary random vibration of use and corrosion of reinforcement.
modelling and analysis: a critical survey and compar- The majority of multistory R/C buildings in
ison. Mech Syst Signal Process 20:763–816 southern Europe were built in the first half of the
Sohn H (2007) Effects of environmental and operational twentieth century. Structures were designed for
variability on structural health monitoring. Philos
Trans R Soc A 365:539–560 gravity loads only by implementing the allowable
Soize C, Ghanem R (2004) Physical systems with random stress design philosophy which did not allow any
uncertainties: chaos representations with arbitrary control of the mode of failure and the
probability measure. SIAM J Sci Comput 26:395–410 corresponding deformation capacity of the indi-
Tóth R (2010) Identification and modeling of linear
parameter-varying systems. Lecture notes in control vidual members. Taking Greece as an example,
and information sciences, vol 403. Springer, Berlin the first seismic code was introduced in 1959, and
Xiu D, Karniadakis GE (2002) The Wiener–Askey poly- R/C walls were introduced in construction in the
nomial chaos for stochastic differential equations. 1960s (often without extending up to the founda-
SIAM J Sci Comput 24:619–644
Zhao Y, Huang B, Su H, Chu J (2012) Prediction error tion of the building). The modern seismic
method for identification of LPV models. J Process codes were introduced more than 20 years later,
Control 22:180–193 in the mid-1980s. The multistory R/C buildings
of the 1950s represent the cutting edge of the
construction technology for gravity load-
designed frame buildings. Information regarding
Strengthened Structural Members the material, detailing, and geometrical charac-
and Structures: Analytical teristics of representative low- to medium-rise
Assessment (up to 8 stories) R/C buildings real structures
found in the urban areas built between the 1920s
Georgia E. Thermou and 1960s is presented in Table 1 (Thermou and
Department of Civil Engineering, Aristotle Palaioxorinou 2013). The lack of a continuous
University of Thessaloniki, Thessaloniki, Greece vertical load path along the height of the build-
ings is a common feature of the buildings of that
era. There is no typical floor since the dimensions
Synonyms of the columns and beams change from story to
story. Often in-plan column layout does not fol-
Assessment; Buildings; Interface; Jacketing; low a grid pattern, hence leading to indirect sup-
Repair; Retrofitting; Shear transfer; Strengthen- ports. Representative typical floor plan layouts
ing; Structural interventions; Structures are presented in Fig. 2.
The result of these systematic deficiencies of S
existing buildings is a decreased level of seismic
Introduction protection, increased seismic vulnerability, and
hence extensive damage expected in future seis-
The catastrophic earthquakes that struck several mic excitations. This is a rather alarming issue
countries over the last 20 years brought to light considering the socioeconomic impact of
the susceptibility of the existing building stock severely damaged buildings or collapses in future
(Fig. 1). Old-type reinforced concrete (R/C) strong ground motions. The recommended solu-
structures are characterized by insufficient rein- tion for this category of buildings, which com-
forcement detailing (lack of stirrups for ensuring prises the vast majority of the existing stock, is
a certain ductility level, indirect supports, insuf- retrofitting (strengthening) with a view to
ficient anchorages of bars), nonuniform distribu- upgrading their seismic capacity and meeting
tion of stiffness and/or mass along the height of the current standards for seismically designed
the building, insufficient foundation system, poor structures. Adopting global intervention (GI)
3510 Strengthened Structural Members and Structures: Analytical Assessment
Strengthened Structural Members and Structures: R/C residential buildings with pilotis in Greece (Source of
Analytical Assessment, Fig. 1 Damage patterns in figures: author’s personal files)
old-type R/C members and structures; typical multistory
Strengthened Structural Members and Structures: Analytical Assessment, Table 1 Information of the typical
construction practice followed in the 1920s–1960s in Greece (Thermou and Palaioxorinou 2013)
Construction practice characteristics
Concrete grade B120
B160a (fck ffi 8–10 MPa)
Steel grade (longitudinal and transverse Smooth StI (fyk = 220 MPa)b
reinforcement)
Typical column cross section dimensions 250
600 mm
Typical diameters of column longitudinal Ø14
Ø20
reinforcement bars
Typical column transverse reinforcement Ø6/250
300 mmc
Typical column longitudinal 7‰
9‰ < 10‰
reinforcement ratio
Typical beam cross section dimensions 150 300
300 600 mm
Typical diameters of beam longitudinal Ø10
Ø18
reinforcement bars
Typical beam transverse reinforcement Ø6/200
250 mmc
Typical thickness of masonry infill walls 100 mm for internal, 200 mm for external
Typical wall cross section dimensions Length: 1.5 ~ 3.6 m, thickness: 150 ~ 250 mm, boundary elements:
length 200 mm, 4Ø12, web: #Ø8/250
Anchorage/lap splice construction Longitudinal reinforcement with hooks with arbitrary lengths
practice Stirrups anchored with 90 hooks
Unconfined lap splices
a
As per Greek Royal Decree of 18/2/1954. Later concrete grade B225 (fck ffi 14 MPa) was introduced in construction
b
As per DIN 1045 (1936). Later longitudinal reinforcement steel grade was changed to StIII (fyk = 400
420 MPa)
c
Diameter Ø8 was rarely applied
methods (e.g., R/C jacketing, R/C infill walls) Fig. 1). GI methods cannot be used separately
leads to the modification of strength and stiffness from local intervention (LI) methods since they
(Fig. 3). They are used in systems with a high are used to accommodate premature failure
flexibility to sway, in torsionally unbalanced sys- modes (Thermou et al. 2012a). LI methods as,
tems, or where strengthening of the existing for example, the intervention methods with com-
building is required (e.g., increase of base shear posite materials are considered not to alter the
strength in buildings with an open first story, stiffness or flexural strength of the retrofitted
Strengthened Structural Members and Structures: Analytical Assessment, Fig. 2 Representative ground floor
plan layouts of the R/C multistory buildings in Greece of the (a) 1920s, (b) 1950s, and (c) 1960s
Strengthened Structural Members and Structures: Analytical Assessment, Fig. 3 Retrofitting strategy: stiff-
ness, strength, and ductility increase (Thermou and Elnashai 2006)
members, but only affect their post-yielding due to the discontinuity in the normal strain along
deformation capacity through confinement and the interfaces is rather determinant for the
by suppressing premature modes of failure (e.g., response at member level. One of the objectives
brittle shear or lap splice failure). is to highlight the role of the interface character-
The decision on the extent of the interventions istics and present the parameters that influence
(i.e., which floors, which elements of the floor, the interconnection between the existing and new
and which design parameters are going to be member. The shear transfer mechanisms mobi-
modified) and its impact on the modification of lized due to relative slippage of the contact inter-
selected global response indices may be assessed faces are presented in the following, and their
through analytical modeling of the strengthened interaction is discussed. Information is also pro-
structure. This implies that there should sufficient vided for the influence of cyclic loading on the
information relative to the effectiveness of the shear resistance mechanisms. The fundamentals
intervention methods adopted at member level. for the development of analytical models which
The addition of a new concrete layer (e.g., flex- take into account slip at the interfaces are
ural strengthening of beams) or a new R/C mem- discussed. The interrelation between the effects
ber (e.g., R/C jacketing, R/C infill walls) in that the changes at local level may have on global
various retrofit techniques entails issues related level is identified as a key issue in the assessment
to the connection between existing and newly of strengthened structures. Moreover, the alterna-
cast concrete. The response of the composite tive simplified design approach suggested by
member, and subsequently of the whole structure, codes, where the monolithicity factors are uti-
depends largely on the response characteristics of lized as to indirectly account for the adverse
the interface, since slip takes place and shear effect of slippage on response, is presented. The
transfer mechanisms are mobilized. Estimating last section of this entry focuses on the alternative
response indices such as stiffness and strength analytical design approaches that could be
in a reliable manner requires the development of followed for one of the most popular global inter-
analytical models that will consider the phenom- vention techniques, the R/C jacketing technique.
ena that take place along interfaces due to slip and
their interaction.
This contribution aims to present the frame- Implications of the Addition of New
work for the analytical assessment of strength- Concrete Layers or New Elements
ened structures when the intervention methods
adopted result in composite members with mul- The response of a composite beam, as the one
tiple monolithic phases. The slippage introduced shown in Fig. 4, depends on the degree of
Strengthened Structural
Members and
Structures: Analytical
Assessment,
Fig. 4 Strain distribution
profiles for (a) no
connection; (b) perfect and
(c) partial connection
between beam A and beam
B. (d) Shear flow along the
interface
interconnection between beam A and beam B. cross sections where an outer shell is placed
In one extreme case, the case of zero friction around the core of the cross section, as in the
(no connection), beam B could slide relative to case of R/C jacketing of an existing column, or
beam A without mobilizing any kind of shear when an infill wall is added to an existing bay by
resistance (Fig. 4a). In an ideal situation, the incorporating the existing columns.
composite members (beam A and beam B) The preparation of the interface and the inter-
would behave as monolithic (perfect connection, connection measures taken between existing and
Fig. 4b). In real conditions, partial connection new member play a key role in the interfacial
between the two bodies is expected, which force transfer and relative slip occurring along
implies that relative slip, s, between the two bod- the contact surface. The bond between the
ies is allowed, thus mobilizing the shear transfer monolithic phases of the composite member
mechanisms at the interface (Fig. 4c). The dis- can be enhanced by roughening of the substrate
continuity in the strain profile corresponds to the surface, addition of dowels placed through holes
slip, s, at the horizontal interfaces (Fig. 4c). Cur- driven on the substrate surface, and when appli-
vature, which is the slope of the strain profile on cable (e.g., R/C jacketing technique) welding of
the cross section, is the same between the various the new longitudinal bars to the existing ones
slipping parts. This is a common assumption in through deformed reinforcing bars bent into a
the analysis of composite (layered) beams (e.g., U-shape. Moreover, experimental evidence has
glulam timber beams, composite steel beams, shown that differential shrinkage and stiffness of
etc.). The shear flow, T, at the interface is esti- new-to-old concrete interfaces, as well as the S
mated by compressive strength of the added concrete to
an existing concrete substrate, influence the
VS response of composite system where concrete-
T¼ (1)
I to-concrete interfaces exist (e.g., Choi
et al. 1999). For example, in the experimental
where S is the first moment of area, I is the study conducted by Júlio et al. (2006), it was
moment of inertia of the composite cross section, found that the increase of the concrete compres-
and V is the shear force on the member. sive strength of the new layer compared to that of
Flexural enhancement of beams with the addi- the existing one leads to an enhancement of the
tion of a new reinforced concrete layer in the compressive strength of the interface, i.e., taking
tension zone follows the principles presented for into account the compressive strength of the
the composite beam of Fig. 4. The same approach weakest concrete seems to be conservative.
can also be followed in the case of composite Recently, Santos and Júlio (2014) proposed
Members and
Assessment, Fig. 5 Slip
at a concrete interface
crossed by reinforcement
design expressions where the effects of differen- Mechanisms that resist sliding (slip) are
tial shrinkage and differential stiffness at the (i) aggregate interlock between contact surfaces,
concrete-to-concrete interface are considered. including any initial adhesion of the jacket con-
crete on the substrate; (ii) friction owing to the
clamping action of reinforcement normal to the
interface; and (iii) dowel action of any properly
Shear Transfer Along Concrete-to-
anchored reinforcement crossing the sliding
Concrete Interfaces
plane. Dowel action develops by three alternative
mechanisms, namely, by direct shear, by kinking,
Shear transfer along interfaces has been a subject
and by flexure of the bars crossing the contact
of continuous research since the 1970s. A variety
plane. The relationship that describes the contri-
of analytical models are available for modeling
bution of the individual shear transfer mecha-
the main shear transfer mechanisms (friction and
nisms is
dowel resistance) by considering that they act
separately or jointly. There are models where all
ttot ¼ tagr þ tf þ tD ¼ tagr þ msN þ tD ,
forces are transferred through reinforcement
(e.g., Birkeland and Birkeland 1966; Walraven ttot ¼ tagr þ mðsc þ rss Þ þ tD
1981), whereas others that, apart from reinforce- ¼ tagr þ mðnf c þ rss Þ þ tD
ment contribution, include a cohesion term (e.g., (2)
Mattock and Hawkins 1972; Vecchio and Collins
1986; Tassios 1986). In the modified compres- where tagr represents the shear resistance of the
sion field theory (MCFT) (Vecchio and Collins aggregate interlock mechanism, m is the interface
1986), the role of the cohesion term is played by shear friction coefficient, sN is the normal
the aggregate interlock mechanism, whereas in clamping stress acting on the interface, and tD is
the model of Tassios (1986), by friction owing to the shear stress resisted by dowel action, FD, in
the clamping action of reinforcement normal to cracked reinforced concrete. The clamping stress
the interface. In the rest of the models, the cohe- represents any normal pressure, p, externally
sion term corresponds to the friction resistance applied on the interface, but also the clamping
developed along the interface. action of reinforcement crossing the contact
Shear transfer is affected by the roughness of plane, ss is the axial stress of the bars
the sliding planes, by the characteristics of the crossing the interface, r is the corresponding
reinforcement, by the compliance of concrete, reinforcement area ratio, n = Ν/(scfc) = sc/fc is
and by the state of stress in the interface zone. the normalized axial load at the interface of
area Ac, and fc is the concrete compressive shear transfer mechanisms by estimating the
strength (Fig. 5). The first two terms in Eq. 2 shear resistance of the interface as
collectively represent the contribution of con-
crete as they depend on the frictional resistance ttot ¼ bF tf þ bD tD (4)
of the interface planes.
As it is mentioned in Model Code (2010) and where the bF and bD are the interaction factors of
the Greek Code for Interventions (KANEPE the friction and dowel resistance, which are equal
2013), in real structures the various mechanisms to bF = 0.4 and bD = 0.7 for slip values s
interact, thereby affecting each other as a func- 1.00 mm. These coefficients can be further
tion of the shear slip. The aggregate interlock decreased to 0.5 when either the expected slip
including any initial adhesive decreases at low value cannot be predicted or when the external
slip values, whereas the maximum contribution compressive force at the interface is negligible.
of frictional and dowel resistance occurs at dif- The friction resistance, tf (Eq. 2), is calculated
ferent slips. Therefore, Model Code (2010) as by considering the coefficient of friction, m,
well as the Greek Code for Interventions which depends on the classification of the rough-
(KANEPE 2013) introduce interaction factors ness of the interface. The majority of the experi-
which consider that the dowels crossing the inter- mental studies provide a qualitative description
face are subject to bending and axial forced of the roughness of the interface after application
simultaneously and that the maximum values of of the selected type of treatment (e.g.,
the different shear resistance mechanisms occur sandblasting, use of pneumatic chipping device).
at different slips. The interaction factors depend It is obvious that the characterization by visual
on various parameters such as the magnitude of inspection only may lead to inaccuracies. There-
the expected slip at the interface, the diameter of fore, the roughness of the interface should be
the reinforcing bars crossing the interface, the quantified by specific indicators. Although sev-
concrete compressive strength, cyclic loading, eral methods have been developed that can suc-
etc. Equation 2 is thus modified according to cessfully quantify the roughness of concrete
Model Code (2010) as follows: interfaces (Santos and Júlio 2013), design codes
in their majority still provide qualitative descrip-
tions of the interface (e.g., smooth, rough, or very
ttot ¼ tagr þ mðsc þ k rss Þ rough). An exception is Model Code (2010)
qffiffiffiffiffiffiffiffiffiffi (3)
þ a r fy fc b n fc which suggests two methods: the sand area
method and the one where the average roughness
is estimated (i.e., measuring the average devia-
where the coefficients k and a refer to the inter- tion of the profile from a mean line).
action factors for interface roughness and flexural From the above, it is seen that describing in S
resistance, respectively, whereas b is a coefficient detail the mechanisms mobilized along interfaces
related to the compressive struts. For the case due to slip and their interaction is a complex
where dowels are driven into the interface, mechanical issue which becomes even more
k = 0.5 for roughened interfaces and k = 0 for complicated considering the response of the
smooth interfaces, whereas in the latter case interface under cyclic loading conditions where
a = 1.5 denoting that the main resistance degradation should also be accounted for. The
mechanism is provided by the dowel action. research conducted on the derivation of constitu-
Model Code (2010, Chap. 7) provides detailed tive models that describe the combined shear
information regarding the design of the interface force resistance mobilized along interfaces due
in shear and how the interaction coefficients are to sliding both under cyclic imposed deforma-
modified for different surface roughness. tions is limited. Palieraki et al. (2012) based on
The Greek Code for Interventions (KANEPE the previous studies of Tassios and Vintzileou
2013) considers also the interaction between the (1987) and Vintzileou and Tassios (1986, 1987)
Strengthened Structural 1/2 fc 1/2 1/3

a t f,n = t f,1 1 – 0.05(n-1) S
Members and sN Su
Assessment, τ f,1=τ f (S)
Fig. 6 Response to τ f,n
symmetric cyclic loading: τ f,n+1
(a) friction and (b) dowel
0.4 τ f,n
resistance
0.5MPa
−s −0.5s n 0.5s s
0.5MPa
−0.4 τ f,n
cycle n+1
cycle n −0.75τ f,n
FD
b 1
FD,n = FD,1 1 n–1
7
FD1=FD (S)
FD,n
FD,n+1
0.25FD,n
−s −0.7s −0.4s s
0.4s 0.7s s
−0.25FD,n
cycle n+1
cycle n
−0.70FD,n
and after carrying out a new experimental pro- the first cycle, n is the number of cycles, s is the
gram proposed the degradation rules for frictional slip value, su is the ultimate slip value, sN is the
and shear resistance, which were also adopted by normal clamping stress acting on the interface,
the Greek Code for Interventions (KANEPE and fc is the concrete compressive strength.
2013). The frictional and dowel resistance is
reduced at each cycle, n, according to (see Fig. 6)
Analytical Assessment of Strengthened
tf; n ¼ tf;1 tdeg Members and Structures
( " 1=2 1=3 #)
1=2 fc s
¼ tf;1 1 0:05ðn 1Þ Analytical Model: Slip at the Interface
sN su
The success of repair/strengthening, as quanti-
(5) fied by the improvement of strength and stiffness

of the retrofitted members, depends on its
1 pffiffiffiffiffiffiffiffiffiffiffi entirety on the degree of collaboration between
FD, n ¼ FD, 1 Ddeg ¼ FD, 1 1 n 1 (6)
7 the various monolithic phases of the composite
member. The response of the retrofitted member
where tf,1 and FD,1 are the peak frictional and is modified proportionally to the slip that takes
dowel resistance value, respectively, attained in place along the interface. The calculation of slip
and its effects on the mechanical characteristics may be evaluated rapidly through standard
of the retrofitted members presuppose the use of Rayleigh-type or Stodola-type iteration
analytical models where the interface character- (Clough and Penzien 1993), using secant-to-
istics need to be explicitly modeled. This com- yield stiffness values for the individual members
prises a quite challenging subject of continuous (Thermou et al. 2012c). The estimated funda-
research. mental translational response shape and period
In case of R/C jacketing of columns, may guide definition of retrofit objectives. Addi-
Thermou et al. (2007, 2012b) have developed tional lateral strengthening of the building
an analytical model for predicting the response (owing to the implicit relationship between stiff-
under monotonic and reversed cyclic loading of ness and strength) by means of controlled stiff-
structural members with old-type detailing, ness addition along the building height may be
strengthened with R/C jacketing. The analytical required. Proportioning the stiffness of the indi-
model introduces one additional degree of free- vidual floors is determined so as to even out large
dom between the existing member and its outer discrepancies in relative drift between succes-
R/C shell, thus allowing slip to take place at the sive floors detected in the fundamental response
interface between the existing member and the shape pattern.
jacket. Shear resistance mechanisms, such as
aggregate interlock, friction, and dowel action, Simplified Design Approach: The Use of
are mobilized to resist slip. The magnitude of Monolithicity Factors
shear flow sustained along the contact surface is A pragmatic design approach commonly
calculated by considering the states of stress of adopted by codes of practice considers the
the composite member at a cracked cross section monolithic approach for the analysis of compos-
and at a point between successive cracks, in ite members making use of properly defined
order to introduce in the flexural behavior the monolithicity factors for obtaining the mechan-
effect of the moment gradient (shear force mag- ical properties of the strengthened member. The
nitude). A detailed description of the analytical redesign procedure is thus simplified substan-
model for R/C jacketed members is presented in tially and is used extensively in practice for
the last section of this article. The same various intervention methods for different
approach has also been implemented after the types of structural members (slabs, beams, col-
necessary modifications for predicting the umns, walls, or footings). The values given so
response of thin reinforced self-compacting far to these modification factors are empirical or
three-jacketed beams (Chalioris et al. 2014). semiempirical. The use of reliable analytical
Moreover, Tsioulou and Dritsos (2011) devel- models (e.g., Thermou et al. 2007, 2012b) can
oped an analytical model for predicting the flex- be utilized for the derivation of monolithicity
ural capacity of beams strengthened by the factors after extensive study of their sensitivity S
addition of a concrete layer in the tension or to the characteristics of the intervention method
compression zone. The algorithm developed (e.g., Kappos et al. 2012).
aimed at the evaluation of slip and shear stress The use of monolithicity factors simplifies
distribution along the interface. substantially design calculations and is applica-
The analytical models, as the ones previously ble to various types of structural members (slabs,
presented, may provide information relative to beams, columns, walls, foundation elements) and
the effectiveness of the selected intervention intervention methods (e.g., R/C jacketing, R/C
method applied at member level. Stiffness and infill walls). These reduction factors are used to
strength of the retrofitted (composite) members obtain the strength and deformation indices of the
are estimated and assessed. The interrelation jacketed members and are applied to the respec-
between the effects that the changes at local tive properties of monolithic members with iden-
level may have on global level is a key issue in tical geometry. The monolithicity factors are
the assessment of strengthened structures. These defined as follows:
Response index of the composite member

K¼ (7)
Response index of the monolithic member with identical geometry
The various monolithicity factors usually used in action between old and new concrete,
design are: (ii) application of full axial load to the jacketed
member, and (iii) application of the concrete
– The monolithicity factors that refer to the properties of the jacket over the full section of
deformation capacity indices such as the the element. It is noted that the monolithicity
chord rotations at yield and ultimate, which factors for the design approach as proposed by
are defined as the Greek Code for Interventions (KANEPE
2013) are subject to certain limitations, i.e., that
y y, J the target strength increase of the jacketed mem-
Chord rotation at yield : K yy ¼ ;
y y, M ber should not exceed twice that of the original.
(8)
yu , J On the other hand, considering the code minima
Chord rotation at ultimate : K yy ¼ regarding the percentage of longitudinal rein-
yu, M
forcement of the jacket, but also of the entire
– The strength related monolithicity factors composite cross section (equal to 1 %), and the
pertinent detailing rules (minimum thickness of
V J, max the jacket is 70 mm), it is seen that the strength of
Shear strength : K v ¼ ;
V M, max the strengthened member far exceeds twice its
(9)
M y, J original strength. In the case of EC8-Part 3
Moment at yield : K My ¼
M y, M (2005), there is no restriction related to the
increase of resistance of the R/C member due to
– The stiffness monolithicity factor jacketing.
In case of R/C column jacketing, the sensitiv-
K y, J ity of the monolithicity factors to the construction
Stiffness at yield : K k ¼ (10)
K y, M materials of the existing cross section (core) and
the jacket, as well as to the percentage of longi-
where the subscripts J and M correspond to the tudinal reinforcement of the jacket for increasing
composite (jacketed) cross section and to axial load, has been highlighted by Thermou
the identical monolithic cross section, yy, yu are et al. (2007). The analytical model for R/C
the chord rotations at yield and ultimate, Vmax is jacketed columns, which is presented in the last
the maximum strength of the cross section, My section of this article, was utilized by Thermou
is the yield moment, and Ky is the secant flexural et al. (2007, 2014) for performing an extensive
stiffness at yield, defined as the ratio of the yield parametric study for the derivation of
moment (My) to the yield curvature (’y). monolithicity factors. For the range of parameters
Eurocode 8, Part 3 (§A.4.2, 2005), and the considered in their study, the lower and upper
Greek Code for Interventions (KANEPE 2013) limits for the monolithicity factors were found
suggest the monolithicity factors presented in to be Kyy = 1.17 ~ 4.85, Kyu = 0.45 ~ 4.13,
Table 2 to be used in the design of R/C members KMy = 0.32 ~ 0.99, Kr = 0.35 ~ 1.02 and
(slabs, columns, beams, and foundation system) Kk = 0.19 ~ 0.95. Lampropoulos et al. (2012)
when strengthened with additional reinforced utilized a computational (finite element) model
concrete layers. In the case of Eurocode 8, Part3 of experimental specimens, and for the specific
(§A.4.2, 2005), the values of the monolithicity range of parameters examined, the monolithicity
factors for columns of Table 2 should be used factors received the following values: Kyy = 1.05
under the assumptions of: (i) full composite ~ 3.00, Kyu = 0.95 ~ 2.85, Kr = 0.70 ~ 0.90 and
Strengthened Structural Members and Structures: Analytical Assessment, Table 2 Suggested values of the
monolithicity factors by EC8-Part 3 (2005) and KANEPE (2013)
Element type Ki KV KMy Kk Kyy Kyu
Slab KANEPE (2013) 0.95 0.85 1.15 0.85
Column EC8-Part 3 (2005) 0.90 1.00 – 1.05a 1.00
1.20b
KANEPE (2013) 0.90c – 0.80c 1.25c 0.75c
Beam KANEPE (2013) 0.85 – 0.80 1.25 0.75
Foundation KANEPE (2013) 0.90 0.70 1.30 0.80
a
Roughening
b
Rest of measures or no measures
c
Roughening and connection measures
Kk = 0.25 ~ 0.75. The comparison of the code considered to be a global intervention method if
suggested monolithicity factors to the experimen- longitudinal reinforcement placed in the jacket
tal values for monolithicity factors revealed that passes through holes drilled in the slab and new
there is large dispersion in the case of concrete is placed in the beam-column joint
monolithicity factors Kyy , Kyu, and Kk (Thermou (Fig. 7). The main advantage of the R/C jacketing
et al. 2014). This observation has to be further technique is the uniformly distributed lateral load
assessed by considering the limited range of capacity throughout the structure, thereby
parameters of the experimental database avoiding concentrations of lateral load resistance,
(Thermou et al. 2011) as well as the fact that which occur when only a few shear walls are
deformation and stiffness values are difficult to added. A disadvantage of the method is the pres-
measure experimentally and “ultimate” condi- ence of beams which may require most of the new
tions are not defined in a uniform way in all longitudinal bars in the jacket to be bundled into
tests (Thermou et al. 2014). The suggested values the corners of the jacket. Because of the presence
for the monolithicity factors to be used in of the existing column, it is difficult to provide
EC8-Part 3 (2005) and KANEPE (2013) apply crossties for the new longitudinal bars, which are
for the specific properties of the construction not at the corners of the jacket.
materials and level of applied axial load. These The analytical model presented in section
limits are defined by the experimental data as: “Analytical Model for R/C Jacketed Columns”
(i) percentage of the longitudinal reinforcement explicitly accounts for the slip at the interface
of the existing cross section 0.81 ~ 2.01 %; between the existing member (core) and the
(ii) percentage of the longitudinal reinforcement jacket. The transfer of normal and shear stresses
of the jacket 0.75–1.64 %; (iii) concrete compres- in the interface between the external layers of the
sive strength of the existing cross section jacket and the existing member determines the S
23–56 MPa; (iv) concrete compressive strength composite action developed by the strengthened
of the jacket 18–69 MPa; and (v) dimensionless members (Thermou et al. 2007). An alternative to
axial load (with the assumption that it is applied this approach is presented in section “Simplified
to the jacketed cross section) 0–0.23. Analytical Expressions for R/C Jacketed Col-
umns” where simplified analytical expressions
in combination with the monolithicity factors
Analytical Assessment of R/C Jacketed may be used for designing the retrofit scheme.
Columns
Analytical Model for R/C Jacketed Columns
Reinforced concrete (R/C) jacketing is one of the The proposed analytical model for predicting the
most commonly applied methods for the rehabil- flexural response of existing R/C members
itation of concrete members. Jacketing is strengthened with concrete jacketing under
Strengthened Structural Members and Structures: Analytical Assessment, Fig. 7 Column R/C jacketing
(Source of figures: author’s personal files)
Strengthened Structural Members and Structures: Analytical Assessment, Fig. 8 (a) Strain profile of the
jacketed cross section; (b) shear transfer mechanisms at the interface between the jacket and the core
monotonic and cyclic loading conditions intro- According to the analytical model of Thermou
duces a degree of freedom allowing the relative et al. (2007) for R/C jacketed members, shear
slip at the interface between the existing member transfer at the interface between the existing
and the jacket (Thermou et al. 2007). Slip along member and the jacket takes place between half
the member’s length is attributed to the differ- crack intervals along the length of the jacketed
ence in normal strains at the contact interfaces member, as commonly considered in bond anal-
(Fig. 8a). For flexural analysis, the cross section ysis. At the initial stages of loading, cracks form
is divided into three layers which bend with the only at the external layers (jacket) increasing in
same curvature, ’ (Fig. 8a). The two external number with increasing load, up to crack stabili-
layers represent the contribution of the jacket, zation. This occurs when the jacket steel stress at
whereas the internal one represents both the the crack, ss,cr, exceeds the limit (fib 2010)
core (existing cross section) and the web of
the jacket shell. Slip at the interface mobilizes 1 þ rs, eff
ss, cr > f ctm, J (11)
the shear transfer mechanisms such as aggregate rs, eff
interlock, friction due to clamping action, and
dowel action provided by the stirrup legs of the where fctm,J is the tensile strength of concrete,
jacket and by the dowels placed at the interface (=Es/Ec) is the modular ratio, and rs,eff is the
between the core and the jacket in case that such a effective reinforcement ratio defined as the total
connection measure is taken (Fig. 8b). steel area divided by the area of mobilized
a b A B
fct,c ΣF1/ jacket ΣF1
N.A.
lc core lc τd.1
core
ΣF2/ core ΣF2
τ
jacket tJ jacket tJ τd.2
ΣF3/ jacket ΣF3
f ms,J f crs,J
c c/2
c
Strengthened Structural Members and Structures: Analytical Assessment, Fig. 9 (a) Free body equilibrium in
the tension zone of the core of the composite section; (b) section equilibrium between adjacent cracks; (c) crack spacing
concrete in tension, usually taken as a circular applied axial load, Next (considered to be applied
domain with a radius of 2.5Db around the bar to the jacketed section), is used to calculate the
(Model Code 2010). Using the same consider- vertical shear stress demand in the member, td,i
ations in the combined section it may be shown (Fig. 9b). With the assumption that the shear
that a number of the external cracks penetrate the flow, q, reversal takes place at length equal to
second layer (core) of the jacketed member c/2 (where c is the crack spacing), the average
(Fig. 8a). The distance between those cracks, stress demand td,i is equal to
taken as c, is a key element of the proposed
methodology (Fig. 9a). SFi
td , i ¼ (13)
After crack stabilization and assuming that the 0:5 c bJ
neutral axis depth is about constant in adjacent
cross sections, from the free body equilibrium in where SFi is the layer force resultant, bJ is the
the tension zone of the core of the composite width of the jacketed cross section, and c is the
section (Fig. 9a), the crack spacing is defined as crack spacing length.
follows (Thermou et al. 2007):
Shear Transfer Mechanisms Under Monotonic
0:64 bJ lc f ct, c and Cyclic Loading
c¼ (12)
nc Db, c f b, c þ nJ Db, J f b, J The monotonic and cyclic behavior of interfaces is
described by the constitutive model developed by
where bJ is the width of the jacketed cross sec- Vassilopoulou and Tassios (2003) where the
tion, lc is the height of the tension zone in the core models of friction and dowel resistance of Tassios
of the composite cross section, fct,c is the tensile and Vintzileou (1987), Vintzileou and Tassios S
strength of concrete core, nc, nJ are the number of (1986, 1987) are adopted. The interface model
bars in the tension steel layer of the core and the accounts for the combined shear force resistance
jacket, respectively, Db,c, Db,J are the bar diame- mobilized along interfaces due to sliding both
ter of the core and jacket longitudinal reinforce- under monotonic and cyclic imposed deformations.
ment, respectively, and fb,c and fb,J are the This model was further modified by Palieraki
average bond stress of the core and the jacket et al. (2012) and adopted by the current Greek
reinforcement layer, respectively. Code for Interventions (2013) (see Fig. 6). In case
Shear stress demand at the interfaces, td,i, is of cyclic loading, additional modifications and
determined by examining the cross section along extensions were applied (Thermou et al. 2012b).
the height and along a member length equal to the
distance between successive cracks (Fig. 9b). Solution Algorithm
Τhe layer force resultant SFi (sum of concrete The objective of the calculation algorithm at each
and steel forces at each layer), for the externally loading step was twofold: simultaneous
Members and
Assessment,
moment–curvature
response histories with the
corresponding
experimental ones (SS1 and
SS3 specimens from
Rodriguez and Park (1994),
R and E from Vandoros and
Dritsos (2008))
establishment of equilibrium between the shear moment–curvature histories derived by the ana-
stress capacity and demand at the interfaces for lytical model with those for the specimens stud-
relative slip, s, and force equilibrium at the cross ied by Bousias et al. (2007). The decision to
section. An iterative procedure was followed, and select this experimental study was based not
equilibrium is established until convergence is only on its scope, but also on the fact that it
achieved. Due to the complexity of the proposed provides detailed test results in terms of
solution algorithm, a program was necessary to moment–curvature curves. The comparison
be developed where fiber analysis was consid- between the experimental and the analytical
ered. The analytical model was set up to predict moment–curvature histories for some of the spec-
moment–curvature response curves. More details imens of Bousias et al. (2007) is presented in
can be found in Thermou et al. (2012b). Fig. 11. It is observed that the analytical curves
reasonably match the strength and stiffness level
Comparison with Experimental Data of the experimental curves at each loading step.
The validity of the proposed analytical model for When slip is taken into account, pinching is more
predicting the flexural response of R/C jacketed pronounced in the analytical model, indicating
members was examined by comparing: (i) for less energy dissipation at each loading cycle,
monotonic loading, the analytical lateral load which is an indication of conservatism.
versus lateral displacement curves along with
the envelope experimental curves of the recorded Simplified Analytical Expressions for R/C
lateral load versus lateral displacement hysteretic Jacketed Columns
loops from various experimental studies The composite cross section is considered as
(Thermou et al. 2007) (the monotonic monolithic, assuming that there is full connection
moment–curvature response curves were derived in the interface between old and new concrete.
according to the analytical model and then The monolithicity factors may be utilized to mod-
converted to force–displacement response ify the response indices in order to indirectly
curves). In Fig. 10, the experimental lateral load account for the adverse effect of slip on the inter-
versus drift curves of a representative number of face (see section “Simplified Design Approach:
test specimens are compared to the corresponding The Use of Monolithicity Factors”). A simplified
analytical curves; (ii) For cyclic loading, the model of a typical jacketed cross section of a
300 300
QRCR QRCD
Moment (kNm)
Moment (kNm)
200 200
100 100
−0.20 −0.15 0.05 0.10 0.15 0.20 −0.20 −0.15 0.05 0.10 0.15 0.20
Curvature (rad/m) Curvature (rad/m)
Experimental
Experimental
Cyclic α=0.5, μ=0.4,
Cyclic α=0.5, μ=0.8 dowels
−300 −300
300 300
QRCRD QRCW
Moment (kNm)
Moment (kNm)
200 200
100 100
−0.20 −0.15 0.05 0.10 0.15 0.20 −0.20 −0.15 −0.10 0.05 0.10 0.15 0.20
Curvature (rad/m) Curvature (rad/m)
Experimental Experimental
Cyclic α=0.5, μ=0.8, Cyclic α=0.5, μ=0.4,

dowels dowels
−300 −300
Strengthened Structural Members and Structures: corresponding experimental ones for specimens QR/CR,
Analytical Assessment, Fig. 11 Comparison of QR/CD, QR/CRD, and QR/CW from Bousias et al. (2007)
moment–curvature response histories with the
column member is depicted in Fig. 12 (Thermou is rcbcdc(0.5hc dc)2/(0.5hJ dJ)2). Further-

et al. 2012c). Dimensions of the initial section are more, any other existing web longitudinal rein-
bc and hc, whereas those of the section after forcement is neglected, as it is considered to have
jacketing are bJ, hJ. Compression zone depth a small influence on post-jacketed flexural S
c is expressed as a fraction of the depth of the strength. Thus, the equivalent longitudinal tensile
jacketed section, c = xyJ dJ. For calculations of reinforcement ratio re, of the jacketed cross sec-
yield moment and flexural stiffness, all reinforce- tion (total reinforcement area divided by the total
ment is considered to act at the location of the area of the jacketed member) is given by
added (jacket) reinforcement. Note that
according to Steiner’s theorem, existing tension ð0:5hc d c Þ2 bc hc
re ¼ rJ þ rc 2
(14)
longitudinal reinforcement (equal to compression ð0:5hJ d J Þ bJ hJ
reinforcement, Fig. 12), given by the area ratio
rc, contributes to flexural stiffness of the jacketed where rc(=Ac/bchc) and rJ(=AJ/bJhJ) are the ten-
section through the term rcbcdc(0.5hc dc)2; sion longitudinal reinforcement ratios of the
thus, in order to maintain the same contribution existing cross section (equal to compression rein-
in this calculation, the equivalent amount that is forcement) and the jacket, respectively, hc and hJ
transferred to the location of jacket reinforcement are the heights of the existing and jacketed cross
Strengthened Structural Members and Structures: reinforcement is neglected as it is considered to have a

Analytical Assessment, Fig. 12 Simplified model for small influence on post-jacketed flexural strength)
R/C jacketed members (existing web longitudinal
sections, respectively, bc and bJ are the widths of ratio of steel and concrete, and xy,jJ is the normal-
the existing and jacketed cross sections, respec- ized depth of compression zone of the R/C
tively, and dc and dJ are the depths of the existing jacketed member.
and jacketed cross sections, respectively. Depending on the magnitude of the axial load
The jth member translational stiffness (secant ratio, nj = Nj/bJ hJ fc/, yielding may occur when
to yield) is defined as either the tension steel reinforcement reaches
yielding or the compressive concrete strain
12Ec I Jy, j 12MJy, j exceeds the limit of linear response in the com-
K Jy, j ¼ ¼ (15) pressive stress–strain envelope, estimated in the
h3st ’Jy, j h3st
range of ec = 1.8 fc//Ec. From the basic cross-
sectional equilibrium, the normalized depth of
The moment at yield, My,jJ, at the center of
compression zone, xy,jJ, associated with these
gravity of the simplified jacketed cross section
two alternative definitions of phenomenological
is estimated equal to
yielding is obtained from:

Mjy, j ¼ ’Jy, j bj h3J Ec 0:40r xy ð1 0:25N c Þ
(a) Upon yielding of tension steel
i
þ 1:15nc þ 0:10 þ 0:25x2y 1 0:66xy

vj f 0c
(16) xJy, j ¼ ð2c 1Þre, j þ
Ec esy
(
2
Hence, the jth member translational stiffness is vj f 0c
equal to þ ð2c 1Þre, j þ
Ec esy

0:5
bJ , j h3J , j Ec n h vj f 0c
K Jy, j ¼ 4:8re, j 1:15 c xJy, j þ 2 ð1:10c 0:10Þre, j þ
3
hst Ec esy
i 2 (18a)
ð1 þ 0:25c Þ þ 0:1 þ 3 xJy, j 1 0:66xJy, j
(17) where fc/ is the concrete compressive strength

and esy is the steel strain at yielding.
where re is the equivalent longitudinal tensile (b) At the onset of concrete strain nonlinearity,
reinforcement ratio, c(=Es/Ec) is the modular ec = 1.8 fc//Ec
h i
Mattock HA, Hawkins MN (1972) Shear transfer in
xJy, j ¼ ð2c 1Þre, j þ 0:55vj
reinforced concrete- recent research. PCI J
h i2 0:5
17(2):55–75
þ ð2c 1Þre, j þ 0:55vj þ 2ð1:10c Þre, j Palieraki V, Vintzileou E, Zeris C (2012). Behaviour of
interfaces in repaired/strengthened R/C elements
(18b) subjected to cyclic actions: experiments and model-
ling. In: Third international symposium on life-cycle
From the above, the secant to yield stiffness and sustainability of civil infrastructure systems
and the moment at yield may be calculated (IALCCE’12), Vienna
Rodriguez M, Park R (1994) Seismic load tests on
directly by utilizing Eqs. 15 and 16, respectively, reinforced concrete columns strengthened by
after being multiplied by the adequate jacketing. ACI Struct J 91(2):150–159
monolithicity factors according to Table 2. Royal Decree (1954) Design regulation of reinforced con-
crete building works. Royal Decree (18.2/26.07.1954),
Ministry of Public Works, Greece (in Greek)
Santos PMD, Júlio ENBS (2013) A state-of-the-art review
References on roughness quantification methods for concrete sur-
faces. Constr Build Mat J 38:912–923
Birkeland PW, Birkeland HW (1966) Connections in pre- Santos PMD, Júlio ENBS (2014) Interface shear transfer
cast concrete construction. J ACI 63(3):345–368 on composite concrete members. ACI Struct
Bousias S, Biskinis D, Fardis MN, Spathis A-L J 111(1):113–121
(2007) Strength, stiffness, and cyclic deformation Tassios TS (1986) Fundamental mechanisms of force-
capacity of the concrete jacketed members. ACI Struct transfer across reinforced concrete critical interfaces.
J 104(5):521–531 In: CEB workshop, West Germany,pp 381–397
Chalioris KΕ, Thermou GE, Pantazopoulou SJ Tassios T, Vintzileou E (1987) Concrete-to-concrete fric-
(2014) Behaviour of rehabilitated R/C beams with tion. ASCE J Struct Eng 113(4):832–849
self-compacting concrete jacketing – analytical Thermou GE, Elnashai AS (2006) Seismic retrofit
model and test results. Constr Build Mat schemes for R/C structures and local–global conse-
J 55(31):257–273 quences. J Progr Struct Eng MatWiley InterSci
Choi D-U, Fowler DW, Jirsa JO (1999) Interface shear 8(1):1–151
strength of concrete at early ages. ACI Struct Thermou GE, Palaioxorinou T (2013) Rapid evaluation of
J 96(3):343–348 non-ductile multistory RC buildings. In: Proceedings
Clough RW, Penzien J (1993) Dynamics of Structures, of ECCOMAS thematic conference “computational
2nd edn. McGraw-Hill International Editions, New methods in structural dynamics and earthquake engi-
York, p 738 neering (COMPDYN 2013)”, Kos, paper no 1459
Deutsches Institut f€ur Normung [German Standards Insti- Thermou GE, Pantazopoulou SJ, Elnashai AS (2007) Flex-
tute] (DIN) (1936) Beton und Stahlbetonbau: ural behavior of brittle R/C members rehabilitated with
Bemessung und Ausf€ uhrung. DIN 1045, Berlin [Con- concrete jacketing. J Struct Eng 133(10):1373–1384
crete and reinforced concrete: design and construction] Thermou, GE, Papanikolaou VK, Kappos AJ (2011) Ana-
Eurocode 8 (2005) Design of structures for earthquake lytical model for predicting the response of old-type
resistance, Part 3: assessment and retrofitting of build- columns rehabilitated with concrete jacketing under
ings. EN1998-3, European Committee for Standardi- reversed cyclic loading. In: Proceedings of
zation (CEN), Brussels ECCOMAS thematic conference “computational
fib (2013) Model Code for Concrete Structures 2010. methods in structural dynamics and earthquake engi- S
Ernst & Sohn, Hoboken, New Jersey neering (COMPDYN 2011)”, Corfu, paper no 137
Greek Code for Interventions (KANEPE) (2013) Earth- Thermou GE, Pantazopoulou SJ, Elnashai AS (2012a)
quake Planning and Protection Organization, Athens Global interventions for seismic upgrading of substan-
Júlio ENBS, Branco FAB, Silva VD, Lourenco JF dard R/C buildings. J Struct Eng 138(3):387–401
(2006) Influence of added concrete compressive Thermou GΕ, Papanikolaou VK, Kappos AJ
strength on adhesion to an existing concrete substrate. (2012b) Cyclic response of R/C jacketed columns
Build Environ 41:1934–1939 including modelling of the interface behavior. In: Pro-
Kappos AJ, Thermou GE, Papanikolaou VK (2012) Investi- ceedings of 15th world conference on earthquake engi-
gation of the behaviour of old type detailing R/C col- neering, Lisbon, 24–28 Sep 2012, paper no 2855
umns strengthened with R/C jackets, Report. Earthquake Thermou GE, Elnashai AS, Pantazopoulou SJ (2012c)
Planning and Protection Organization 179 (in Greek) Retrofit Yield Spectra–a practical device in seismic
Lampropoulos AP, Tsioulou OT, Dritsos SE (2012) Mono- rehabilitation. Earthq Struct 3(2):141–168
lithic coefficient values for design when seismically Thermou GΕ, Papanikolaou VK, Kappos AJ (2014)
strengthening RC columns with jackets. J Earthq Eng Monolithicity factors for the design of R/Ccolumns
16:1023–1042 strengthened with R/C jackets. In: Proceedings of
3526 Strengthening Techniques: Bridges
2nd European conference on earthquake engineering connections under the span’s supporting truss
and seismology, Istanbul, paper no 766 which led to its unseating (Buckle et al. 1990).
Tsioulou OT, Dritsos SE (2011) A theoretical model to
predict interface slip due to bending. Mat Struct The bridge failure of the Hanshin elevated
44:825–843 expressway during the 1995 Great Hanshin
Vandoros KG, Dritsos SE (2008) Concrete jacket con- Kobe earthquake (Fig. 1b) is attributed to column
struction detail effectiveness when strengthening R/C failure caused mainly by poor anchorage of the
columns. Constr Build Mater 22:264–276
Vassilopoulou I, Tassios P (2003) Shear transfer capacity circular hoop ends which were lap spliced in the
along a R/C crack under cyclic sliding. In: Proceedings cover concrete (Park 1996) and inadequate
fib symposium, Technical Chamber of Greece, Athens, amount and details of the longitudinal and trans-
paper no 271 verse reinforcement (Anderson et al. 1996). This
Vecchio FJ, Collins MP (1986) The modified compression
field theory for reinforced concrete elements subjected entry presents a summary of observed damage in
to shear. J ACI 83(2):219–231 past earthquakes and seismic strengthening tech-
Vintzileou E, Tassios TS (1986) Mathematical models for niques of existing bridges. The stages of evolu-
dowel action under monotonic and cyclic conditions. tion of bridge design in the USA from the 1970s
Mag Concr Res 38(134):13–22
Vintzileou E, Tassios TS (1987) Behavior of dowels under including vulnerabilities and typical design
cyclic deformations. ACI Struct J 84(1):18–30 details are outlined.
Walraven JC (1981) Fundamental analysis of aggregate
interlock. Struct Div ASCE 107(ST11):2245–2270
Seismic Bridge Design
The seismic design of bridges is separated into

Strengthening Techniques: Bridges three periods, based on the State of California
provisions (Caltrans 2006).
Chris P. Pantelides
Department of Civil and Environmental Pre-1971 Design
Engineering, University of Utah, Salt Lake City, In this period, it was recognized that earthquakes
UT, USA generate forces proportional to the structure’s
dead weight. Until 1965, the maximum lateral
seismic design force was 6 % of the structural
Synonyms dead weight. Vulnerabilities of bridges built dur-
ing this period include: (a) column shear failure,
Bridge; Damage; Design; Earthquakes; Rehabil- (b) column longitudinal reinforcement pullout,
itation; Repair; Retrofitting; Seismic; and (c) unseating of expansion hinges. Typical
Strengthening design details of the period are 13 mm diameter
column ties at 300 mm spacing (regardless of
column or bar size), very short seat widths at
Introduction expansion joints (150–200 mm), inadequate
lap splices of column bars from the footing
The goal of seismic design is to prevent life- (20 bar diameters), inadequate development of
threatening damage and to allow moderate dam- column bars into the footing (20 bar diameters
age that can be repaired after an earthquake. From without standard hooks), and lap-spliced column
a life safety perspective, the most serious bridge ties in the cover concrete (absence of seismic
damage occurs when the deck collapses, because hooks).
of inadequate seat details, column failure, or cap
beam-column joint failure. The collapse of the 1971–1994 Design
link span of the San Francisco-Oakland Bay After the 1971 San Fernando earthquake, the
Bridge during the 1989 Loma Prieta earthquake importance of ductility and detailing was recog-
(Fig. 1a) was caused by failure of two bolted nized, and the concept of capacity design was
Strengthening Techniques: Bridges 3527
Strengthening Techniques: Bridges, Fig. 1 Bridge (CALTRANS)); (b) Hanshin elevated expressway
failure: (a) San Francisco-Oakland Bay Bridge (# Cali- (https://www.fhwa.dot.gov/publications/publicroads/96fall/
fornia Department of Transportation (2003) imgs/p96au18.jpg (FHWA))
introduced in the design code. Vulnerabilities of Damage to Bridges in Past Earthquakes

bridges built during this period include:
(a) column shear failure of plastic hinge regions, After an earthquake, lessons are learned which
(b) shear failure of flared columns, and influence seismic strengthening of existing brid-
(c) unseating of expansion joint hinges. In ges as well as the development of new design
retrofitted bridges, vulnerabilities were found at codes. This section describes damage and its
expansion joint hinge restrainers, particularly location as observed in bridges from past earth-
for skewed bridges. Typical details for this quakes (Caltrans 2006). Deck damage includes
period include: closer spacing and improved complete collapse of a span, as shown in Figs. 1a
column shear detailing (spacing 100–150 mm, and 2a, b, or movement of the deck, as shown in
but no confinement/anti-buckling requirement Fig. 2c. Cap beam, cap beam-column joint, and
for plastic hinges), top reinforcement in footings girder shear damage are shown in Fig. 3. Damage
and pile caps (but no shear reinforcement), col- to columns can be flexural as shown in Fig. 4 or
umn longitudinal splices were prohibited at shear damage as shown in Fig. 5; shear damage is
maximum moment locations, short seat widths brittle and frequently leads to collapse. Soil set-
at expansion joints (300 mm), no cap beam- tlements can occur around columns, and gaps
column joint reinforcement, and poor column between columns and soil may become large as
flare details. shown in Fig. 6. Abutment failures are also com-
mon, as shown in Fig. 7. S
Post-1994 Design
After the 1994 Northridge earthquake, a capacity
design philosophy was adopted that would ensure Bridge Seismic Strengthening
ductile flexural failure of the columns while all
other bridge components remained elastic. Typi- There is historic evidence which supports the
cal design details include: tight confinement rein- assertion that seismic strengthening of bridges is
forcement in plastic hinge regions (100 mm effective. Damage in past earthquakes, such as
spacing), long seat widths at expansion joints the 1994 Northridge earthquake, has been
(600 mm), improved flare column details (gap reported from cable restrainers breaking through
between top of flare and superstructure), no lap the hinge diaphragm; however, no bridge
splices in plastic hinge zones, shear reinforce- retrofitted after the 1989 Loma Prieta earthquake
ment in footings, and cap/column and footing/ experienced any serious damage (Yashinsky
column joint reinforcement. 1998). Similarly, most of the observed structural
Strengthening Techniques: Bridges, Fig. 2 Deck failure: (a), (b) span collapse (# California Department of
Transportation (2006)); (c) span displacement
Strengthening Techniques: Bridges, Fig. 3 Cap beam, Transportation (2006); beam-column joint: # California
cap beam-column joint, and girder shear damage (Cap Department of Transportation (2003))
beam, girder shear damage: # California Department of
Strengthening Techniques: Bridges, Fig. 4 Column flexural damage (# California Department of Transportation
(2006))
Strengthening Techniques: Bridges, Fig. 5 Column shear damage (# California Department of Transportation
(2006))
Strengthening
Techniques: Bridges,
Fig. 6 Soil settlement and
gap around columns (#
California Department of
Transportation (2006))
Strengthening
Techniques: Bridges,
Fig. 7 Abutment failures
(# California Department
of Transportation (2006))
damage in the 2011 Great East Japan earthquake seat extenders, column strengthening, infill wall
has reportedly occurred in older bridges that had between columns, footing and abutment strength-
not yet been retrofitted after the 1995 Great ening, and cap beam-column joint strengthening.
Hanshin Kobe earthquake or only partially so In addition, retrofit measures such as seismic
(Marsh et al. 2011). isolation reduce the demand imposed on a bridge
Several measures are used to strengthen brid- system, and in this sense they improve its seismic
ges: restrainers, steel girder plate connections, performance significantly.
a b
Restrainer Cables
CL Hinge d
c
Pipe
New New
Bolster Bolster
Strengthening Techniques: Bridges, Fig. 8 (a) Cable restrainer, (b) steel bar restrainer, (c) pipe seat extender, (d)
bumper blocks
Restrainers/Expansion Joint Seat Width because of the need to limit restrainer forces to
Extenders/Steel Plates/Bumper Blocks protect other structural elements, while increas-
Cable restrainers (Fig. 8a) and longitudinal steel ing restrainer forces to accomplish their intended
bar restrainers (Fig. 8b) are used to prevent purpose. Alternatives that should be considered
girders from falling off their supports when loss of support is a possibility include seat
(Yashinsky 1998). The typical cable restrainer is extensions, bearing strengthening, and replace-
a galvanized 19 mm diameter steel cable with ment with conventional or isolation bearings.
galvanized cold-swaged fittings and 25 mm
diameter threaded studs at each end. Loss of Column Jacketing: RC/Steel/Fiber Reinforced
support of the superstructure can be achieved Polymer (FRP)
using beam seat extenders, pipe seat extenders Strengthening of concrete columns can be
shown in Fig. 8c, steel plates used to connect achieved by reinforced concrete (RC) jackets.
two steel girders at expansion joints, and bumper This is an intrusive and time-consuming proce-
blocks as shown in Fig. 8d (Pantelides dure, but in some cases it is necessary. However,
et al. 2004). the effect of strengthening a column using RC
Failure of restrainer cables has been observed jackets should be examined carefully as this
in the 1989 Loma Prieta, 1994 Northridge, and method will increase column flexural strength
1995 Kobe earthquakes (FHWA 2006). and stiffness more than a steel jacket, FRP com-
Restrainers ruptured, anchorage plates pulled posite jacket, or wire wrap, with potentially unde-
through concrete diaphragms, fittings pulled sirable effects on bridge performance.
away from cables, and anchorage nuts worked Steel jacketing of circular columns by circular
loose from the ends of cable units. Restrainers steel jackets (Fig. 9a), and of rectangular columns
alone may not be the best solution in many cases by elliptical steel jackets, was found to be
Strengthening Techniques: Bridges, Fig. 9 Column 06/21/fauntleroy-expressway-wearing-new-jackets/#sthash.

jackets: (a) steel (Photo credit: Washington State Depart- nlxozUnm.dpbs), (b) FRP square, (c) FRP circular
ment of Transportation. http://sdotblog.seattle.gov/2012/
effective in enhancing shear strength and flexural Street Bridge on I-80 in Salt Lake City, Utah
ductility (Priestley et al. 1994). Encasing circular (Pantelides et al. 2004).
columns with a bonded steel jacket was found to The advantage of using FRP composites
inhibit bond failures in lap splices of longitudinal stems from their superior resistance to corrosion,
reinforcement in plastic hinge regions (Chai high stiffness-to-weight ratio, high strength-to-
et al. 1991). Rectangular steel jackets are effec- weight ratio, and ability to control the material’s
tive in enhancing the performance of shear criti- behavior by selecting the proper fiber orienta-
cal columns. These jackets can improve column tion. FRP composites must be protected against
ductility by eliminating the brittle shear mode of ultraviolet radiation; even though carbon fibers
failure, but it is important to recognize that the themselves are not affected by moisture, the
failure mode may then shift to a flexural one for quality of the epoxy matrix is critical in the S
which the rectangular jacket can provide only performance of the composite. Carbon fiber sys-
limited assistance. tems have shown loss of tensile strength at high
Seible et al. (1997) used continuous FRP temperatures due to moisture absorption.
jackets for flexural hinge confinement of circular Another consideration when using FRP systems
and rectangular columns, shear strengthening, is their high initial cost; however, they have
and lap splice clamping. Full-scale tests a superior life-cycle cost as compared to steel
(Fig. 9b) have shown that significant increases because of their resistance to corrosion. In shear
can be gained in the ductility of bridge column and flexural critical applications, bond of FRP
bents strengthened with FRP composites composites to concrete is a critical parameter
(Pantelides et al. 1999, 2007). FRP composites which must be considered in the design (ACI
have been used widely for seismic strengthening 440 2002); mechanical anchorage of FRP lami-
of existing bridges (Priestley et al. 1996; Ogata nates to concrete has been used successfully to
and Osada 2000), as shown in Fig. 9c for the State postpone debonding.
Another effective strengthening technique is Dowels New Infill Wall

to wrap prestressing wire under tension around
a column (Hawkins et al. 1999). Rectangular steel
columns have been strengthened with steel angles
and circular steel columns with circular steel
pipes (Nishikawa et al. 1998).
Infill Wall Strengthening

Infill shear walls are reinforced concrete walls
cast in place between the columns of
a multicolumn bent, as shown in Fig. 10. The New Footing
reinforced concrete columns can have a circular
or rectangular shape. They have been used suc- Strengthening Techniques: Bridges, Fig. 10 Pier
cessfully to increase transverse shear capacity. strengthening using infill wall
These walls prevent the formation of plastic
hinges in the columns during transverse loading
and help overcome deficiencies in the flexural or case, it is desirable to add hold-down anchors or
shear strength of the cap beam. To be effective, supplemental tension piles. An alternative is to
infill shear walls should be designed to act com- drill holes into the piles from the top of the pile
positely with existing members. This is done by cap and epoxy high strength steel bars into the
providing drilled and bonded dowels in the col- pile cap and piles (Pantelides et al. 2001); these
umns and the bottom of the cap beam, so that vertical bars are shown in Fig. 11 (Pantelides
shear is transferred at the interfaces through et al. 2007).
a shear friction mechanism; this may require the
existing concrete surfaces to be roughened. Abutment Strengthening
A footing should be provided under the infill Strengthening of seat-type abutments in the lon-
wall which is tied into the existing column foot- gitudinal and transverse directions can reduce
ings. There needs to be sufficient reinforcement forces on the piers. Anchor slabs are used to
to transfer all seismic forces, and the potential for increase the ability of abutments to carry both
differential settlement between the existing and longitudinal and transverse loads, and resist dis-
new footings should be considered. placement in either direction, as shown in Fig. 12.
Transverse abutment shear keys and anchors can
Foundation Strengthening reduce forces on the piers (FHWA 2006).
Footings that support columns may be structur-
ally unable to resist the forces transmitted from Cap Beam Joints: RC Bolsters/Posttensioning/
those columns. This usually occurs when there is Steel Jackets/FRP Jackets
a lack of reinforcement in the top of the footing. Cap beam-column joints are subjected to large
A concrete overlay can be designed to provide stresses and are vulnerable to damage in earth-
increased negative moment capacity in the foot- quakes transverse to the bridge. Existing cap
ing. Existing positive moment reinforcement is beams and joints were not usually designed to
utilized along with the additional footing depth behave in a ductile manner, so any strengthening
provided by the overlay to increase positive measure must ensure that these elements are
moment capacity. For isolated pile caps a grade either capable of accommodating the ductility
or link beam is utilized to improve shear capacity, demands placed on them or are capable of elasti-
as shown in Fig. 11. In addition, longitudinal cally resisting the forces that will result from
reinforcement in reinforced concrete piles may plastic hinging in the columns.
be terminated too close to the top of the pile, RC bolsters are used to improve flexural
resulting in insufficient tensile capacity; in this capacity, with or without prestressing. Another
b GRADE BEAM
COLUMN
PILE CAP DYWIDAG

BAR
STEEL BAR
SECTION A-A
PILE DYWIDAG
BAR
A A
Strengthening Techniques: Bridges, Fig. 11 Footing strengthening using grade beam and pile cap to pile steel bars
Dowels Anchor Slab Seismic Isolation: Elastomeric Bearings/

Friction Pendulum Bearings
Seismic isolation bearings can shift the natural
frequency of the bridge away from the region of
dominant earthquake energy; in addition, they
Pipe
increase damping and reduce and redistribute
Restrainer
CIDH the earthquake-induced lateral forces to levels
Pile
within the elastic capacities of the substructure
and foundation (Mayes et al. 1994). An elasto-
meric bearing (Fig. 14a) consists of alternating
layers of rubber and steel plates and often has
Strengthening Techniques: Bridges, Fig. 12 Abutment a lead core to dissipate seismic energy. A friction
strengthening using anchor slab pendulum bearing (Fig. 14b) consists of a stain-
less steel concave dish and articulated slider sur-
faced with a composite liner. During an
strengthening method is to use external earthquake, the slider moves back and forth on S
posttensioning as shown in Fig. 13a. Steel jackets the concave dish; the spherical surfaces of the
bonded to the sides of the joint and anchored to slider and the dish define a motion similar to
the bent cap are also effective (Thewalt and that of a pendulum. A friction pendulum bearing
Stojadinovic 1995). Finally, FRP jackets in the isolates a bridge from an earthquake through
form of an ankle wrap have been tested (Gergely pendulum motion and absorbs energy through
et al. 2000) and used to strengthen column to cap friction.
beam-column joints as shown in Fig. 13b
(Pantelides et al. 2004). The nominal principal
tensile stress developed in the beam-column joint Summary
is used to design the carbon FRP composite
layers. When strengthening of these joints is The entry presents a number of methods for seis-
required to improve the longitudinal response, mic strengthening of existing bridges. A brief
transversely prestressed bolsters may be used. review of seismic bridge design evolution and
Strengthening Techniques: Bridges, Fig. 13 Cap beam-column joint strengthening: (a) posttensioning (Photo Credit:
VSL International Ltd. http://vsl-sg.com/sps-images/4-004.jpg), (b) FRP jacket
Strengthening Techniques: Bridges, Fig. 14 Bridge publicroads/99marapr/images/seismic3.gif), (b) friction

bearings: (a) elastomeric (Photo Credit Federal Highway pendulum (Photo credit: Federal Highway
Administration. https://www.fhwa.dot.gov/publications/ Administration)
an overview of damage to structural elements of progress made in designing new and strengthen-
bridges have been presented. Strengthening tech- ing existing bridges, it is likely that future earth-
niques covered include restrainers, expansion quakes will expose additional vulnerabilities of
joint seat width extenders, column jackets, foot- existing and retrofitted bridges.
ings, abutments, and cap beam-column joints.
Retrofit measures such as seismic isolation
reduce the demand imposed on a bridge system Cross-References
thus improving the seismic performance. Hinges
and abutments are the two areas where retrofitted ▶ Bridge Foundations
bridges continue to experience damage. The 1994 ▶ Resilience to Earthquake Disasters
Northridge and 2011 Great East Japan earth- ▶ Retrofitting and Strengthening of Structures:
quakes have shown that retrofitted bridges did Basic Principles of Structural Interventions
not experience any serious damage. This clearly ▶ Seismic Resilience
demonstrates that seismic strengthening of brid- ▶ Seismic Strengthening Strategies for Existing
ges is valuable and should be continued. A unique (Code-Deficient) Ordinary Structures
feature of earthquakes is that they often expose ▶ Strengthened Structural Members and
the weaknesses of a structural system. Despite the Structures: Analytical Assessment
Strengthening Techniques: Code-Deficient R/C Buildings 3535
References Pantelides CP, Duffin JB, Reaveley LD (2007) Seismic

strengthening of reinforced-concrete multicolumn
American Concrete Institute, ACI Committee 440 (2002) bridge piers. Earthq Spectra 23(3):635–664
Guide for the design and construction of externally Park R (1996) An analysis of the failure of the columns of
bonded FRP systems for strengthening concrete struc- a 600 m length of the Hanshin elevated expressway
tures. ACI 440.2R-02, Farmington Hills during the Great Hanshin Earthquake of 17 January
Anderson DL, Mitchell D, Tinawi RG (1996) Performance 1995. Bull N Z Natl Soc Earthq Eng 29(2):73–82
of concrete bridges during the Hyogo-Ken Nanbu Priestley MJN, Seible F, Xiao Y, Verma R (1994) Steel
(Kobe) earthquake on January 1995. Canadian J Civil jacket retrofitting of reinforced concrete bridge col-
Eng 23:714–726 umns for enhanced shear strength-part 2: test results
Buckle IG, Goodson M, Cassano R, Douglas B, Hegemier G, and comparison with theory. ACI Struct
Imbsen R, Jones D, Liu D, Mayes R, North P, Priestley N, J 91(5):537–551
Rojahn C, Seible F, Selna L, Viest I (1990) Bridge Priestley MJN, Seible F, Calvi GM (1996) Seismic design
structures. Earthq Spectra 6(S1):151–187 and retrofit of bridges. Wiley, New York, 704 pp
California Department of Transportation (2003) The race Seible F, Priestley MJN, Hegemier GH, Innamorato
to seismic safety: protecting California’s transporta- D (1997) Seismic retrofit of RC columns with contin-
tion system. Department of Transportation Advisory uous carbon fiber jackets. J Compos Constr 1(2):52–62
Board Report, Sacramento, 186 pp Thewalt CR, Stojadinovic B (1995) Behavior of bridge
California Department of Transportation (2006) Visual outrigger knee joint systems. Earthq Spectra
catalog of reinforced concrete bridge damage. Struc- 11(3):477–509
ture Maintenance and Investigations Division, Sacra- Yashinsky M (1998) Performance of bridge seismic retro-
mento, 350 pp fits during Northridge Earthquake. J Br Eng 3(1):1–14
Chai YH, Priestley MJN, Seible F (1991) Seismic retrofit
of circular bridge columns for enhanced flexural per-
formance. ACI Struct J 88(5):572–584
Federal Highway Administration (2006) Seismic
retrofitting manual for highway structures: part
1-bridges. U.S. Department of Transportation, Strengthening Techniques:
McLean, 656 pp Code-Deficient R/C Buildings
Gergely I, Pantelides CP, Reaveley LD (2000) Shear
strengthening of R/C T-joints using CFRP composites.
J Compos Constr 4(2):56–64
Andreas P. Lampropoulos
Hawkins NM, Kaspar II, Karshenas M (1999) Seismic School of Environment and Technology,
retrofit of Poplar Street interchange in East St. Louis. University of Brighton, Brighton, UK
Structures Congress. ASCE, Reston
Marsh ML, Buckle IG, Yen WP, Dasti S, Frost D, Monzon
E (2011) Bridge performance during the Great East
Japan Earthquake of March 11, 2011. http://www. Synonyms
pwri.go.jp/eng/ujnr/tc/g/pdf/27/27-1-3_Marsh.pdf
Mayes RL, Choudhury D, Crooks RS, Jones DM, Knight Epoxy grouting; Interfaces; Monolithicity coeffi-
RP (1994) Seismic isolation retrofit of existing brid-
cients; RC jacketing; Repair; Shrinkage;
ges. In: Proceedings of the materials engineering con-
ference. ASCE, Reston, pp 1179–1186 Strengthening
Nishikawa K, Yamamoto S, Natori T, Terao K,
S
Yasunami H, Terada M (1998) Retrofitting for seismic
upgrading of steel bridge columns. Eng Struct 20(4–6):
540–551
Introduction
Ogata T, Osada K (2000) Seismic retrofitting of express-
way bridges in Japan. Cem Concr Compos 22:17–27 Nowadays, many buildings need to be upgraded,
Pantelides CP, Gergely I, Reaveley LD, Volnyy VA especially in earthquake-prone areas, because of
(1999) Retrofit of R/C bridge pier with CFRP advanced
increased strength demands introduced by mod-
composites. J Struct Eng 125(10):1094–1099
Pantelides CP, Gergely J, Reaveley LD (2001) In-situ ern design codes or damage due to strong earth-
verification of rehabilitation and repair of reinforced quakes. As a result, many techniques have been
concrete bridge bents under simulated seismic loads. developed to improve the performance of the
Earthq Spectra 17(3):507–530
existing buildings. There is a wide range of appli-
Pantelides CP, Alameddine F, Sardo T, Imbsen R (2004)
Seismic retrofit of State Street bridge on Interstate 80. cations of fiber-reinforced polymers (FRPs) in
J Br Eng 9(4):333–342 the form of rebars, plates, and sheets for the
3536 Strengthening Techniques: Code-Deficient R/C Buildings
strengthening of the existing buildings. These strengthening deficient buildings depends on the
materials can be used for flexural and/or shear level of damage and the requirements. In case of
strengthening of existing structures, and a com- lightly damaged concrete elements, where there
prehensive description of this technique can be are no requirements for further upgrading of the
found in Essay 382,109 “Retrofitting and structure, the technique of epoxy grouting is
Strengthening of Contemporary Structures: widely used. Epoxy grouting can be used to
Materials Used.” improve the performance and the durability of
In this entry, special emphasis is given on the cracked reinforced concrete elements.
technique of reinforced concrete (RC) jacketing. In case of heavily damaged structural ele-
RC jacketing is a widespread technique used for ments or structures that need to be upgraded to
the strengthening of existing columns, beams, conform with modern codes, seismic strengthen-
and walls. The use of epoxy resins for the crack ing is required. There are a number of available
repair of damaged structural elements is also techniques for the earthquake strengthening
presented. of existing buildings, such as jacketing of
In earthquake-prone areas, the majority of the load-bearing elements, addition of infill walls or
old buildings could be characterized as “code bracing systems, and installation of energy
deficient,” and these buildings need to be dissipation systems. For the jacketing of struc-
upgraded. tural elements, steel, fiber-reinforced polymers
“Code-deficient” buildings are the buildings (FRPs), and reinforced concrete can be used.
that do not satisfy the requirements proposed by The decision for the selection of the most suitable
the new codes (Eurocode 8 2005; Code of Struc- technique is not always easy and depends on a
tural Interventions 2012; fib 2010). There are dif- number of parameters such as cost, required time,
ferent factors responsible for structural and disruption of the occupants. Detailed descrip-
deficiencies in existing buildings. The structural tions of a number of different repair and strength-
system of many old buildings was designed with ening techniques have been published by Tassios
configuration problems. Lack of regularity in (2009) and Dritsos (2005).
geometry, in strength or stiffness, in plan, or in In this entry, the technique of repair of dam-
elevation are common characteristics of several aged buildings with epoxy grouting and the
existing buildings. Also, in the past, approxima- strengthening with RC jackets are described in
tions and simplifications were adopted in carrying detail, and practical recommendations are
out structural analysis. Computers were not avail- presented. Special focus is given on the modeling
able, 3D analysis was not feasible, even 2D anal- and design of the jacketed elements. Critical
ysis was rarely used, and (continuous) beams and parameters for the effectiveness of this technique
columns were considered as independent ele- are the bond between the old and new concrete
ments. Some critical requirements concerning the elements and shrinkage of the new concrete.
behavior of structures under earthquakes, such as To take into account these two parameters, finite
ductility, capacity design, and detailing provisions element method can be used. However, for the
(minimum amount of stirrups, lower limits for practical design of strengthened elements the use
compressive strength, upper limit for tensile rein- of monolithicity coefficients is proposed, which
forcement), were also ignored. Also, in the past, is a simplified procedure. Monolithicity coeffi-
the seismic actions used for the design of structures cients are used to correlate the strength, stiffness,
were much lower than the actions currently used and deformation at yield and failure of strength-
for the design of new structures (Dritsos 2012). ened columns to those of respective monolithic
As a consequence, the majority of existing columns, the behavior of which can be deter-
buildings are deficient and in need of strengthen- mined by following conventional concrete design
ing. Also, some of the old buildings are already procedures. Monolithicity coefficients for jackets
damaged after previous strong earthquakes. The with different characteristics are presented at the
type of intervention that will be used for end of this entry.
Strengthening
Techniques:
Code-Deficient R/C
Buildings, Fig. 1 (a)
Preparation of cracks
before the use of epoxy
resin and (b) repaired
columns. http://www.
episkeves.civil.upatras.gr
Repairing Existing RC Structures Using improve substrate conditions. All the existing coat-
Epoxy Resins and Bonded Mortar ings along the crack, at intervals of not less than the
thickness of the concrete member or 10 cm (ETEP
Epoxy grouting is one of the most commonly 2012), should be removed.
used techniques for the repair of cracked struc-
tural elements. For the successful application of Port Installation
this technique, the surface of the damaged or Entry ports (tube-like devices that provide for the
cracked section needs to be properly prepared successful transfer of the epoxy resin under pressure
(Fig. 1a). Depending on the type of damage, into the crack) together with “cap” crack sealing
different epoxy resin types and installation pro- should be installed on the face of the crack to ensure
cedures can be used (Fig. 1b). containment of the epoxy as it will be injected under
In this section, the technique of epoxy resin pressure into the crack (ACI 2003; Fig. 2). In case of
injection for crack sealing and the use of epoxy- crack penetration throughout a section, the use of
bonded concrete replacement for the repair of cap seals on both sides of the crack elements is S
damaged concrete are described. suggested (ACI 2003).
The distance of the ports depends on the open-
Epoxy Resin Injection for Crack Sealing ing at the face of the crack. The recommended
The resin characteristics and the application port distance values proposed by ETEP (2012)
methods used for the epoxy resin injection are presented in Table 1.
depend on the environmental and concrete tem- Before the cap seal installation, the location of
perature, the crack width, the thickness of the the widest portion of the crack should be marked.
concrete member, and the extent of cracking. The material used for the seal cap should be
carefully prepared with accurate batching of
Substrate Preparation components and with a consistent application of
The first critical step for the successful application the material over the crack. The use of 25 mm
of this method is the preparation of the substrate. wide and 5 mm thick cap seals is proposed by
Any existing contaminants should be removed to ACI (2003).
Strengthening
Techniques:
Code-Deficient R/C
Buildings,
Fig. 2 Installation of ports
for crack sealing with
epoxy resins
Strengthening Techniques: Code-Deficient R/C Strengthening Techniques: Code-Deficient R/C

Buildings, Table 1 Recommended port spacing for dif- Buildings, Table 2 Recommended resin viscosity and
ferent crack widths (ETEP 2012) injection pressure values for different crack widths (ETEP
2012)
Crack width (mm) Port spacing (mm)
0.3–0.5 100 Dynamic Required
0.5–1 100–135 Crack width viscosity of the injection
(mm) resin (cps) pressure (MPa)
1–2 135–170
0.3–0.5 1,000 –
2–3 170–200
500 0.8
3–10 200
250 0.4
130 0.2
Epoxy Resin Injection 0.5–1 1,000 0.8
Epoxy resins cure to form solids with high 500 0.4
strength and relatively high modulus of elasticity. 250 0.2
Epoxy resin used for crack injection should be a 1–2 1,000 0.4
100 % solid resin that meets the requirements of 500 0.2
specification ASTM C-881 (2010) for type I or 250 0.1
2–3 1,000 0.3
IV, grade 1 (low viscosity), class B or C
500 0.1
(depending on the ambient temperature). If the
250 0.005
purpose of injection is to restore concrete to its
original load-bearing capacity, a type IV epoxy
should be used. If full restoration of load-bearing
capacity is not envisaged, a type I epoxy is suffi- The injection pressure and the dynamic vis-
cient. No solvents or nonreactive diluents are cosity of the resin depend on the crack width. In
permitted in the resin (USBR 1997). Class Table 2, the recommended values proposed by
B epoxy is used between 4 oC and 16 oC, while ETEP (2012) are presented.
class C is used above 16 oC (ASTM C-881 2010). Upon completion of the resin injection, the
The injection should start from the bottom in ports and the cap seals will be removed by heat
the case of vertical cracks, while for horizontal chipping or grinding (ACI 2003).
cracks the starting point should be the widest
crack section. The injection should continue Quality Control
until refusal or bleeding of the adjacent port The first check should be done by hand 48 h
(ACI 2003; ETEP 2012). after the injection, to check if the coagulation
(polymerization) of the epoxy resin has been (USBR 1997). The surface preparation of the
completed. old concrete and the materials used for the repair
Then, an additional check is required to ensure are critical parameters for the effectiveness of the
that the crack is filled with the resin. Cores should technique.
be taken with a diameter of 25–50 mm and depth
equal to the thickness of the cracked element and Old Concrete Preparation
at least equal to 15 cm (ETEP 2012). At least one The first step for the old concrete preparation is to
core is required for 30 m crack length, and the saw cut the perimeter of the damaged area to a
core will be visually inspected to check if the depth of 25–40 mm in order to provide a retaining
crack is filled to the appropriate level with resin. boundary for the compaction and consolidation
The cores can also be tested for compressive and of the repair material. Then, the deteriorated con-
tensile loading following ASTM C42 (1999) and crete needs to be removed to create a sound
(ACI 2003). The removed core should be patched substrate for the bonding of the new repair mate-
with a non-shrink or expansive cementitious or rial (USBR 1997). There are several techniques
epoxy grout (ETEP 2012; ACI 2003). for the damaged concrete removal such as the use
The use of an endoscope is also recommended of high-pressure hydroblasting, jackhammer,
for quality control. In this method, holes with a shotblasting, and dry or wet sandblasting
diameter 2 mm larger than the diameter of the (USBR 1997).
tube of the endoscope should be drilled, to a depth
equal to the thickness of the cracked element and Materials Used for Epoxy-Bonded Repair
at least 15 cm. At least two points (in two differ- The epoxy resin used for epoxy-bonded mortar
ent crack segments) for every 30 m of crack repairs (less than 40 mm depth) should be two
injected (total length of the injected crack seg- component, 100 % solid type, meeting the
ments) should be inspected. There are also non- requirements of specification ASTM C-881
destructive methods suggested for the inspection (2010) for type III, grade 2 (medium viscosity),
of the repaired elements. Use of the ultrasonic class B or C (depending on the ambient
pulse velocity method is also an option (ETEP temperature).
2012; ACI 2003). At least three measurements The material used in this technique is conven-
should be taken (in three different crack seg- tional concrete and epoxy resin bonding agent. The
ments) for every 30 m of crack injected (total concrete should be the same as the old concrete but
length of the injected crack segments) (ETEP with consistency (slump value) within 40 mm. The
2012). resin should be of two components, 100 % solid
type, meeting the requirements of specification
Epoxy-Bonded Mortar or Concrete for ASTM C-881 (2010) for type II or type V, class
Concrete Repair B or C (depending on the ambient temperature). S
Epoxy-bonded mortar or concrete is used for Type II can be used for use in non-load-bearing
repairs involving shallow replacement of con- applications to bond fresh to hardened concrete,
crete. Epoxy-bonded mortar is used for depth of while for load-bearing applications, type V should
replacement less than 40 mm, while epoxy- be used (ASTM C-881 2010).
bonded concrete is used in depths between
40 and 140 mm (USBR 1997). Shallow concrete
replacements are subject to poor curing condi- Strengthening Using Additional Layers/
tions as a result of moisture loss to evaporation Jackets
and to capillary absorption by the existing con-
crete element with subsequent poor bond to the This technique involves casting of an additional
existing structure. The use of epoxy-bonded mor- reinforced concrete element in connection to the
tar or concrete can be used to ensure adequate existing load-bearing elements and is extensively
bond between the old and the new material used to upgrade existing buildings (Fig. 3).
Strengthening
Techniques:
Code-Deficient R/C
Buildings,
Fig. 3 Jacketing of
existing column and wall.
http://www.episkeves.civil.
upatras.gr
Two crucial issues for the effectiveness of this Four mechanisms contribute to the shear resis-
technique are the interface between the old and tance at the interface: concrete-to-concrete adhe-
the new concrete and shrinkage of the new con- sion, concrete-to-concrete friction (Fig. 4a), the
crete layer or jacket (Lampropoulos and Dritsos connecting action from either steel bars placed
2011). Concrete shrinkage induces an initial slip across the interface between the old and the new
at the interface which in some cases can be high concrete (Fig. 4b), and bent-down bars welded
enough and even lead to debonding (Beushausen between the bars of the old and the new concrete
and Alexander 2006, 2007). Also, tensile stresses (Fig. 4c; Dritsos 2007). These four mechanisms
are developed at the new concrete, which may can be subdivided into the two groups of
lead to a reduced concrete compressive strength unreinforced and reinforced interfaces,
due to biaxial stress state (Lampropoulos and depending on whether or not additional steel is
Dritsos 2010, 2011; Lampropoulos et al. 2012a). placed across the interface or welded between the
Practical recommendations for the strengthening bars of the old and the new concrete (Dritsos
of the additional elements using additional layers 2007).
or jackets are presented in section “Interface The roughening of the interfaces is crucial for
Preparation”, while in section “Analysis and the performance of the strengthened specimens
Design of Strengthened Elements”, analysis and especially in case of unreinforced interfaces. The
design methods for strengthened/composite spec- roughening of the surface of the existing elements
imens are presented. can be achieved using jet with water and sand
mixture or light air equipment or electric needle
Practical Recommendations for Jacketing (Code of Structural Interventions 2012). In Fig. 5,
Procedure the use of air chipping hammer for roughening
the surface of a beam is presented.
Interface Preparation There are two main techniques currently used
In order to guarantee a sufficient connection to characterize the interface roughness. The one of
between contact surfaces, the shear force acting the two is a qualitative approach, wherein the
at the interface should be lower than, or equal to, surface of the existing elements is compared to
the shear resistance. standard surface samples with different roughness
Strengthening
Techniques:
Code-Deficient R/C
Buildings,
Fig. 4 Friction, steel bars
across the interface, and
bent-down bars used in
concrete-to-concrete
interfaces
Strengthening Techniques: Code-Deficient R/C Buildings, Fig. 5 (a) Roughening procedure and (b) roughened
surface
grades. The second one, which is a quantitative perform, both in place and in laboratory, and as
approach, is the sand patch test. This method con- all interfaces are horizontal top surfaces, it is easy
sists in uniformly spreading a standard volume of to be applied (ASTM E965 2003; Santos and Júlio
sand (V) and then measuring the diameter (D) of 2010, 2013).
the covered area from which the roughness is Based on the roughness value, the surface can
calculated (=4 V/pD2). This method presents the be classified as very smooth when the roughness
advantage of being inexpensive and easy to is not measurable, smooth when the roughness is
Strengthening Anchorage with

Techniques: double bolts
Code-Deficient R/C
Buildings,
Fig. 6 Construction detail
of a typical jacketed
column (Tassios 2009) At least one bent down
bar in the joint
Epoxy resin
injection
New longitudinal bars

New stirrups
Bent down bars
Concrete jacket
Existing longitudinal reinforcement
lower than 1.5 mm, rough when the roughness is layer, while use of non-shrinkage concrete is also
higher than 1.5 mm, and very rough when the recommended (Júlio et al. 2003).
roughness is higher than 3 mm (Santos and Júlio The anchorage of the added longitudinal rein-
2013; fib 2010). forcement to the existing element is essential to
Apart from the two techniques presented prevent debonding. In case of jacketed columns,
above, there are other “relatively new” tech- continuity between floors is recommended (Júlio
niques used to determine the roughness of the et al. 2003). However, it is not always feasible to
interface (i.e., outflow meter, mechanical stylus, continue a jacket to the other floors. In such cases,
circular texture meter, digital surface roughness or in order to minimize occupant disturbance, the
meter, microscopy, ultrasonic, slit-island following construction details are required
method, roughness gradient method, photogram- (Fig. 6) (Tassios 2009):
metric method, shadow profilometry, air leakage
method, processing of the digital image (PDI) (i) Filling of the gap between the jacket and the
method, two-dimensional laser roughness ana- slab with epoxy resin injection to ensure
lyzer (2D-LRA) method, and three-dimensional effective compressive load transfer. The
laser scanning method). Most of these methods epoxy resin should be injected at least one
are based on optical techniques, and their main month after the casting of the jacket.
benefits are the improved accuracy and the fact (ii) The longitudinal bars of the jacket should be
that they are not destructive methods and can be anchored to the slab with double bolts to
applied both in the laboratory and in situ (Santos ensure tensile load transfer.
and Júlio 2010). (iii) Welding of at least one bent-down bar with
each reinforcing bar of the initial column.
Casting of New Concrete and Reinforcement
Details In case of “open” jackets, where the jacket
For the casting of new concrete, small aggregate cannot be cast on all four sides of the column,
size of about 2 mm is normally used because of special provisions are required to prevent
the lack of space in the jacket or the additional debonding. In three-sided jackets, epoxy grout
Strengthening
Techniques:
Code-Deficient R/C
Buildings, Fig. 7 Typical
shear stress-slip model for
roughened interface and for
interface with dowels
should be used to integrate the stirrups of the et al. 2012a). Detailed investigations have
jacket into the initial column. In case of strength- recently been published where monolithicity
ening with additional layers only on one or two coefficient values have been calculated for differ-
sides, the connection of the new with the old ent values of the normalized axial load and jacket
reinforcement using bent-down bars is essential thicknesses (Lampropoulos et al. 2012b;
(Dritsos 2005). Thermou et al. 2014). The effect of jacket’s
restraint shrinkage has also been studied
Analysis and Design of Strengthened (Lampropoulos et al. 2012b), and it has been
Elements shown that in some of the examined cases, the
To take into account the effect of the interface “code”-proposed values are not conservative.
and concrete shrinkage, finite element analyses In the following section (section “Concrete-to-
can be conducted (Lampropoulos and Dritsos Concrete Interface”), different models for the
2011). However, in practice, it is not feasible to shear stress calculation at the interface are
perform sophisticated finite element analyses and presented. The proposed method for modeling
take into account all these parameters, so the use concrete shrinkage effects is presented in section
of monolithicity coefficients is proposed by the “Concrete Shrinkage Effect”.
codes (Eurocode 8 2005; Code of Structural
Interventions 2012). Monolithicity coefficients Concrete-to-Concrete Interface
correlate the characteristics of the strengthened There are a number of available models for the
members to those of the respective monolithic calculation of the interface strength between old
ones. This procedure is quite old; however, most and new concrete. In most of these models, the
of the proposed monolithicity coefficients are shear stress at the interface between the old and
empirical, and single values are proposed without new concrete is related to the slip at the interface
accounting for the different values of the normal- (Tassios 1986; Tassios and Vintzileou 1987; S
ized axial load, the thickness of the jacket, and the CEB-FIP 1993; Code of Structural Interventions
shrinkage of the new concrete parameters that 2012). In these studies, concrete-to-concrete
considerably affect the response of the strength- interfaces with different roughening grades were
ened specimens (Lampropoulos et al. 2012b). In examined. A typical shear stress-slip model for
the literature, there are studies (Dritsos 2007; roughened interfaces with and without steel
Thermou et al. 2007; Thermou et al. 2014) dowels is presented in Fig. 7.
where monolithicity coefficients values are pro- For the shear resistance due to cohesion, the
posed, and the effect of jacket concrete shrinkage following model is proposed by CEB-FIP (1993)
has been ignored. It must be stated that and Code of Structural Interventions (2012). The
the restrained shrinkage of the jacket concrete distribution of the shear stress with the slip at the
significantly influences the response of strength- interface is considered linear until the limit value
ened columns and cannot be overlooked for the shear resistance (tfud) which is the shear
(Lampropoulos and Dritsos 2010; Lampropoulos stress value for slip equal to 0.01-0.02 mm.
After that point, shear resistance is considered to

be constant and equal to the limit value (tfud). The
tfud value depends on the interface type Eq. 1.
8
< 0:25 f ct , smooth interface
tfud ¼ 0:75 f ct , rough interface (1)
:
f ct , shotcreting
where:
tfud is the shear resistance at the interface, Strengthening Techniques: Code-Deficient R/C
fct is the concrete tensile strength. Buildings, Fig. 8 Shear stress-slip model for concrete-
to-concrete interfaces due to cohesion (CEB-FIP 1993;
The shear stress-slip distribution is presented Code of Structural Interventions 2012)
in Fig. 8.
In case of smooth interfaces, there are also
similar models for the shear stress distribution A similar model is presented in the Code of
due to friction (Tassios and Vintzileou 1987; Structural Interventions (2012), where the shear
CEB-FIP 1993; Code of Structural Interventions stress distribution with the slip is defined using
2012; Fig. 9). two different equations, before and after half of
According to this model, the shear stress dis- the maximum slip value Eq. 4 (Fig. 11):
tribution is linearly increasing up to the point of
s tf
its limit value, while after that point the stress 0:5 !
sfu tfud
remains constant. The maximum shear stress rffiffiffiffiffi
point is calculated using Eq. 2. s s tf
¼ 1:14 3 , > 0:5 !
sfu sfu tfud
pffiffiffiffiffi 1=3
tfud ¼ 0:4 sc , sfu ¼ 0:15 sc (2) ¼ 0:81 þ 0:19
s
where tfud ¼ 0:4 f 2c sc
sfu
where (4)
sfu is the slip at maximum shear strength (tfud)
and sc is the design compressive strength at the The recommended maximum shear stress at
interface. the interface (tfud) proposed by Model Code
Various models have also been proposed for 2010 (fib 2010) is in the range of 1.5–2.5 MPa
roughened interfaces. The first model for rough- for strengthened interfaces with sandblasting,
ened interfaces is presented in Fig. 10 (Tassios while for a well-roughened interface with high-
1986; CEB-FIP 1993). pressure hydroblasting, the maximum shear
According to this model, the shear strength of stress is in the range of 2.5–3.5 MPa. These
roughened interfaces is linearly increasing up to a values are valid for concrete with strength
slip equal to 0.1 mm. At this point, the shear stress lower than C50/60.
is equal to half of its maximum value. For slip Values for the coefficients of friction and
values higher than 0.1 mm and lower than cohesion at the interface of strengthened ele-
2.0 mm, the shear resistance can be derived ments have also been proposed. The suggested
using Eq. 3. values according to CEB (1983), together with
the coefficient of friction values proposed by
4 3
tf tf Eurocode 2 (1996) and Model Code 2010 (fib
0:5 ¼ 0:3 s 0:03
tfud tfud 2010) are presented in Table 3.
1=3 From the values of Table 3, it can be observed
where tfud ¼ 0:4 f 2c sc
that the design values for the coefficient of fric-
(3) tion proposed by Eurocode 2 (1996) are lower for
Strengthening
Techniques:
Code-Deficient R/C
Buildings, Fig. 9 Shear
stress-slip models due to
friction for smooth
interfaces (Tassios and
Vintzileou 1987; CEB-FIP
1993; Code of Structural
Interventions 2012)
Strengthening τf 4 3
τf τf
Techniques: – 0.5 • = 0.3 • s – 0.03
Code-Deficient R/C τ fud τ fud
stress-slip distribution due
to friction for roughened τfud
interfaces (Tassios 1986;
CEB-FIP 1993)
τfud/ 2 1/3
τ fud = 0.4 • ( f c2 • s c )
0.0
0.1 mm Sfu = 2.0 mm S [mm]
Strengthening
Techniques:
Code-Deficient R/C
stress-slip distribution due
to friction in roughened
interfaces (Code of
Structural Interventions
2012)
well-roughened interfaces compared to the values of the normal to the interface stresses is
values proposed by CEB Bulletin No. 162 (CEB proposed (Eq. 5).
1983) and Model Code 2010 (fib 2010). The
coefficient of friction values proposed by Model 2=3
sc
Code 2010 (fib 2010) are close to the values m ¼ 0:44 (5)
proposed by CEB (1983). fc
In the study of Tassios and Vintzileou (1987)
and in the Code of Structural Interventions Another investigation has been conducted for
(2012), an analytical relationship for the calcu- dowel-reinforced interfaces, and models for the
lation of coefficients of friction for different distribution of the shear stress with the friction
Strengthening Techniques: Code-Deficient R/C Buildings, Table 3 Coefficient of friction and cohesion values for
strengthened elements with various roughening levels (CEB 1983; Eurocode 2 1996; fib 2010)
Model code 2010 (fib
CEB Bulletin No. 162 (1983) Eurocode 2 (1996) 2010)
Coefficient of Cohesion Coefficient of
friction (MPa) friction Coefficient of friction
Well-roughened 1.5 1.9 0.9 1–1.4
interface
Roughened interface 0.9 1.7 0.7 0.7–1
Smooth interface 0.7 1.0 0.6 0.5–0.7
Very smooth interface – – 0.5 –
Strengthening
Techniques: V / Vu
Code-Deficient R/C V u = 1.3 * d b2 * f c * f sy
stress-slip relationship for
dowel action (Vintzileou 1.00
and Tassios 1986)
0.80
0.60 V 4 s
= *4
Vu 3 db
0.40
0.20 V = (200 f c ) s
* f *d s [mm]
Vu sy b
0.0
0.14 1.0 2.0 3.0 4.0
have been proposed (Vintzileou and Tassios the shear stress with the slip is defined by two
1986; CEB-FIP 1993; Code of Structural Inter- relationships, depending on the slip values
ventions 2012). In the model proposed by (Eq. 6):
Vintzileou and Tassios (1986), the variation of
sffiffiffiffiffiffi! rffiffiffiffiffi
V fc s V 4 4 s
¼ 200 for s 0:14 mm, and ¼ for s > 0:14 mm
Vu f sy db Vu 3 db (6)
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where Vu ¼ 1:3 d2b f c f sy
where: (s 0.14 mm) is similar to Eq. 6 with the only

V is the shear stress (N), difference that a coefficient equal to 250 is pro-
Vu is the shear strength (N), posed instead of 200. After the initial elastic
fsy is the yield stress of steel (MPa), and branch, there is a second linear one up to the
db is the diameter of the dowels (mm). maximum shear resistance.
A different model has been presented in
This model is presented in Fig. 12. CEB-FIP (1993). According to this model, the
A similar bilinear model has been proposed by shear resistance due to dowel action linear
CEB (1983). The equation for the initial branch increases up to half of the maximum shear stress.
At this point, the slip equals to 0.05 of the max- shear stress-slip distribution is presented in
imum slip and after this point Eq. 7 is used. The Fig. 13.

4
3 !
s V V s
¼ 1:7 0:5 þ 0:05 for 0:05 1 where su ¼ 2:0 mm
su Vu Vu su

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (7)
1:30 2 pffiffiffi
Vu ¼ db 1 þ ð1:3 eÞ2 1:3 e f cd f yd 1 z2 < As f yd = 3
gRd
where reduction after each loading cycle. Characteristic

diagrams of the shear stress distribution with the
gRd is considered equal to 1.30, slip under cyclic loading are presented in Fig. 15
As is the dowel cross section, and
(Vintzileou and Tassios 1986, 1987; Tassios and
e is the load eccentricity.
Vintzileou 1987).
The first model for the calculation of the
In the model proposed by the Code of Struc-
reduced shear stress due to cyclic loading is the
tural Interventions (2012), the maximum shear
one proposed by Tassios and Vintzileou (1987)
strength value is calculated using the same equa-
(Eq. 10)
tion proposed by CEB-FIP (1993). The variation
of the shear stress with the slip is considered
linear for slip equal or lower than 0.005 dB. For tf , n ¼ tf , 1 1 ½0:002 ðn 1Þ ðs=sfu Þ=ðscn =f c Þ1=3
any value between 0.005 and 0.05 dB, the shear (10)
stress value is calculated using Eq. 8. This model
is presented in Fig. 14.
where

4
3 !
V V tf,n is the shear stress at the interface in the nth
s ¼ 0:1 du þ 1:80 du 0:5 loading cycle,
Vu Vu
tf,1 is the shear stress at the interface in the first
du ¼ 0:05 db loading cycle,
(8) N is the number of cycles, and
scn is the normal to the interface stress in the nth
The model for the dowel action proposed by loading cycle.
Model Code 2010 (fib 2010) is given by Eq. 9
In the Code of Structural Interventions (2012),
two models have been presented for the shear S
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi s 0:5
Vu ¼ k As f cd f yd , stress reduction in smooth (Eq. 11) and rough-
su (9)
ened interfaces (Eq. 12).
su ¼ 0:1 0:2 db Smooth interfaces:
where k ~ 1.6 for circular reinforcement cross pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

section and concrete strength equal to or lower tf , n ¼ tf , 1 1 0:15 ðn 1Þ (11)
than C50/60.
There are also models for the reduction of Roughened interfaces:
shear stress due to cyclic loading in roughened
interfaces and for dowel action (Vintzileou and 1=2 1=3
Dtf , n fc s
Tassios 1986, 1987; Tassios and Vintzileou 1987; ¼ 0:05 ðn 1Þ1=2
Code of Structural Interventions 2012). tf , 1 scn sfu
According to these models, there is a shear stress (12)
Concrete Shrinkage Effect

Concrete shrinkage can considerably affect the
performance of strengthened elements with addi-
tional layers or jackets. Tensile stresses are
induced to the new concrete of the layer or jacket
while there is also an induced slip at the interface
between the old and the new concrete. Denarie
and Silfwerbrand (2004) presented a model
where the tensile stresses of the new concrete
due to restrained concrete shrinkage are propor-
tional to coefficient “R.” Coefficient “R” is char-
acteristic of the connection between the old and
the new concrete and the degree of restraint
Strengthening Techniques: Code-Deficient R/C
(Eq. 14).
Buildings, Fig. 13 Shear stress-slip relationship for
dowel action (CEB-FIP 1993) s ¼ Ε efsh R (14)
where
E: modulus of elasticity of concrete,

efsh: free shrinkage strain, and
R: degree of restraint.
Values for the coefficient R have been pro-

posed by Abbasnia et al. (2005) based on exper-
imental investigations. The risk of debonding and
cracking of the new layer of strengthened beams
was investigated by Beushausen and Alexander
(2006, 2007) where the tensile stress relaxation
due to concrete creep was also taken into
consideration.
Buildings, Fig. 14 Shear stress-slip relationship for
The biaxial stress state in strengthened ele-
dowel action (Code of Structural Interventions 2012) ments due to the development of restrained
shrinkage tensile stresses was numerically inves-
The following model has been proposed by tigated, and the importance of the restrained con-
Vintzileou and Tassios (1986) and Vintzileou crete shrinkage in case of jacketed columns was
and Tassios (1987) for dowel action under cyclic highlighted (Lampropoulos and Dritsos 2010,
loading (Eq. 13) 2011). An extensive experimental study was
also conducted to validate the reliability of the

1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi numerical results (Lampropoulos et al. 2012a).
Dn ¼ D1 1 ðn 1Þ (13) In this study, specimens with and without steel
7
plates (with expanded polystyrene) were cast,
where to simulate the concrete of the jacket under
restrained and free shrinkage, respectively.
Dn is the maximum shear stress due to dowel action in Specimens were stored until testing in a room
the nth loading cycle, with a constant temperature and relative humid-
D1 is the maximum shear stress due to dowel action in ity, and shrinkage strain at the end of the speci-
the 1st loading cycle.
mens was measured using micrometer dial
Strengthening Techniques: Code-Deficient R/C Buildings, Fig. 15 Characteristic shear stress-slip diagrams for
friction and dowel shear stress due to cyclic loading
Strengthening
Techniques:
Code-Deficient R/C
Buildings,
Fig. 16 Specimens with
and without steel plates and
strain measuring setup
gauges. The specimens together with the setup In the tested square prism specimens, the
for the shrinkage strain measurements are reduction of the concrete strength was found to
presented in Fig. 16. range from 20 % to 30 %. This reduction was
Each pair of specimens (with steel plates or more significant for larger values of tensile
expanded polystyrene cast from the same batch of stresses caused by the restrained concrete shrink- S
concrete) was tested in compression at a different age (Lampropoulos et al. 2012a).
age, to obtain results for different values of the
shrinkage strain. The specimens with restrained Modeling of Strengthened Elements and
and free concrete shrinkage were tested under Monolithicity Coefficients for the Design of
compression (Fig. 17a). The ratio of the tensile Strengthened Elements
stress to the tensile strength together with the
respective reduction of the concrete strength Finite Element Modeling
was calculated. It was found that there is a sig- For modeling the strengthened elements, the
nificant reduction of concrete strength due to finite element method has been used. Different
restrained concrete shrinkage and the subsequent assumptions have been examined, and the accu-
biaxial stress state, and this is in agreement with racy of the numerical models was validated using
the model proposed by Kupfer et al. (1969) experimental data (Lampropoulos and Dritsos
(Fig. 17b). 2011).
a b 0.9
0.8
Reduction of concrete
compressive strength
0.7
0.6
0.5
0.4
0.3
0.2 Experimental results
0.1 Kupfer et al. (1969)
0.0
0.30 0.35 0.40 0.45
Tensile stresses
(normalized to the tensile strength)
Strengthening Techniques: Code-Deficient R/C Buildings, Fig. 17 (a) Compressive test setup and (b) Biaxial
stress state due to restrained concrete shrinkage (Lampropoulos et al. 2012a)
Interface Between the Old and the New columns, with the only difference that there is
Concrete The interface between the old and full connection (no slip) between the original
the new concrete can be simulated using spring column and the jacket, and concrete shrinkage is
elements or using special contact elements. To neglected.
determine the characteristics of the interface, the The load and the deflection at the points of
models presented in section “Concrete-to- yield, maximum load, and failure of the speci-
Concrete Interface” can be used. Strength deteri- mens were estimated, and Eq. 15 was used to
oration due to cyclic loading should also be taken calculate monolithicity coefficient values.
into account (Lampropoulos and Dritsos 2011).
h i
Fy
½Fmax STR dY
Concrete Shrinkage Concrete shrinkage effect KF ¼ , K k ¼ h i STR ,
can be simulated by applying volumetric strain to ½Fmax MON Fy
dY MON (15)
the elements of the jacket. The effect of concrete
dy ½du STR
creep should also be included to take into account K d, yy ¼ STR , K d, yu ¼
the tensile stress relaxation. In a previous study dy MΟ ½du MΟ
(Lampropoulos and Dritsos 2010), in order to
include creep effects, application of a volumetric where KF, Kk, and K d, yy, u are monolithicity coef-
strain equal to half of the free concrete shrinkage ficients for the strength, stiffness, and deflection
strain was proposed. or rotation angle at yield and failure, respectively,
Fy and Fmax are the yield and maximum load, and
Monolithicity Coefficients dy, du are the deflections at yield and failure.
The accuracy of the numerical models was vali- Subscripts STR and MΟΝ indicate strengthened
dated using a number of different experimental and monolithic specimens, respectively.
data. The same assumptions were used in a para- The proposed design values are based on the
metric study where columns with different char- assumption of concrete with free shrinkage strain
acteristics were examined (Lampropoulos of 800 microstrains. A roughened interface was
et al. 2012b). Analyses were performed for the assumed without any other connection technique
strengthened and the respective monolithic spec- (bent-down bars or dowels) which is a conserva-
imens (Fig. 18), and monolithicity coefficients tive assumption. Different values have been pro-
were calculated using Eq. 15. Monolithic speci- posed depending on the normalized axial load (n)
mens are specimens with exactly the same char- and the ratio of the concrete cross-sectional area
acteristics as the examined strengthened of the jacket (Acj) to the cross-sectional area of
Strengthening [Fmax]MON
Techniques: [Fmax]STR
[Fy]MON
Code-Deficient R/C 200
Buildings, [d y]MON
Fig. 18 Typical load 150 Experimental results [d u]MON
deflection relationship for
Numerical strengthened
the strengthened and the [Fy]STR
100 Numerical monolithic
[d u]STR
respective monolithic [d y]STR
specimens (Lampropoulos
50
and Dritsos 2011, 2012b) Load (kN)
0
−50
−100
−150
−200
−100 −75 −50 −25 0 25 50 75 100
Deflection (mm)
the original column (Aco) (Acj/Aco). The normal- Strengthening Techniques: Code-Deficient R/C
ized axial load (n) can be calculated using Eq. 16. Buildings, Table 4 Monolithicity coefficient values for
a four-sided jacket with different normalized axial load
values and an Acj/Aco ratio of 1.5 (Lampropoulos and
N Dritsos 2012b)
v¼ (16)
Aco f co þ Acj f cj Monolithicity coefficients
n KF Kk K d,yy K d,yu
where N is the axial load, Aco is the cross-
Proposed 0.1, 0.85 0.55 1.50 1.00
sectional area of the original column, fco is the design values 0.2
concrete strength of the original column, Acj is the 0.4 0.70 0.50 1.30 1.10
cross-sectional area of the concrete jacket, and fcj Eurocode 8, 0.90 0.95 1.20 1.00
is the strength of the jacket concrete. Part 3 (2005)
Four- and three-sided jackets have been exam- Code of 0.90 0.80 1.25 0.80
ined. In case of three-sided jacket, the stirrups of Structural
Interventions
the jacket were merged into the initial column in (2012) S
order to simulate anchorage, which is essential
and in practice is ensured by using epoxy grout
(Lampropoulos et al. 2012b). The proposed values for the monolithicity
The proposed values for a four-sided jacket coefficients for a three-sided jacket with an Acj/
with Acj/Aco ratio of 1.5 and normalized axial Aco ratio of 1.0, and normalized axial load values
load values of 0.1, 0.2, and 0.4 are presented in of 0.1, 0.2, and 0.4 are presented in Table 6
Table 4 (Lampropoulos et al. 2012b together with (Lampropoulos et al. 2012b).
the values proposed by existing codes (Eurocode Table 7 presents proposed monolithicity coef-
8 2005; Code of Structural Interventions 2012). ficients values for a three-sided jacket and a nor-
Table 5 presents monolithicity coefficient malized axial load value of 0.2 with different Acj/
values for a normalized axial load of 0.2 and Aco ratio values (Lampropoulos et al. 2012b).
different Acj/Aco ratio values for a four-sided Monolithicity coefficients for any case with a
jacket (Lampropoulos et al. 2012b). different Acj/Aco ratio or normalized axial load
Strengthening Techniques: Code-Deficient R/C Strengthening Techniques: Code-Deficient R/C

Buildings, Table 5 Monolithicity coefficient values for Buildings, Table 7 Monolithicity coefficient values for
a four-sided jacket with different Acj/Aco ratio values and a a three-sided jacket with different Acj/Aco ratio values and
normalized axial load of 0.2 (Lampropoulos and Dritsos a normalized axial load of 0.2 (Lampropoulos and Dritsos
2012b) 2012b)
Monolithicity coefficients Monolithicity coefficients
Acj/Aco KF Kk K d,yy K d,yu Acj/Aco KF Kk K d, yy K d, yu
Proposed 0.5 0.90 0.70 1.30 0.95 Proposed 0.5 0.85 0.55 1.70 1.15
design values 1.5 0.85 0.55 1.50 1.00 design values 1.5 0.80 0.55 1.40 1.10
4.0 0.75 0.75 1.05 2.85 4.0 0.75 0.25 3.00 1.00
Eurocode 0.90 0.95 1.20 1.00 Eurocode 8, 0.90 0.95 1.20 1.00
8, Part 3 Part 3
(2005) (Eurocode
Code of 0.90 0.80 1.25 0.80 8 2005)
Structural Code of 0.90 0.80 1.25 0.80
Interventions Structural
(2012) Interventions
(2012)

Buildings, Table 6 Monolithicity coefficient values for values proposed by codes are not conservative.
a three-sided jacket with different of normalized axial load The proposed values of the present work can also
values and an Acj/Aco ratio of 1.0 (Lampropoulos and be used when considering other techniques to
Dritsos 2012b) connect between the original column and the
Monolithicity coefficients new jacket (such as dowels crossing the interface
n KF Kk K d, yy K d, yu or bent-down bars welded between the two sets of
Proposed 0.1, 0.80 0.55 1.40 1.10 longitudinal bars), as these values are conserva-
design values 0.2 tive (only roughening at the interface was consid-
0.4 0.70 0.35 1.90 1.70 ered) and are suitable for design purposes
Eurocode 0.90 0.95 1.20 1.00 (Lampropoulos et al. 2012b).
8, Part 3
(2005)
Code of 0.90 0.80 1.25 0.80
Structural Summary
Interventions
(2012) In this entry, the technique of concrete repair with
epoxy grouting and the strengthening of existing
structures with additional reinforced concrete
value from those presented in Tables 4, 5, 6, and 7 layers and jackets were described in detail. Prac-
can be estimated by linear interpolation tical recommendations were presented for both
(Lampropoulos et al. 2012b). techniques. For the strengthening of the existing
The monolithicity coefficient values for the structures using additional layers and jackets,
design of strengthened RC columns presented recommendations for the analysis and
herein depend on the type of the jacket (four or monolithicity coefficients for the design of the
three sided), the normalized axial load value, and strengthened elements were also presented.
the thickness of the jacket. The proposed values The following general recommendations can
are more realistic than using one value to cover be drawn:
all situations, as is the case with present code
recommendations (Eurocode 8 2005; Code of • Epoxy resins can be used for the crack sealing
Structural Interventions 2012). It was also found and for the replacement of damaged concrete.
that in many cases, monolithicity coefficient Due consideration should be given to
successfully apply this method. In this chapter, ▶ Retrofitting and Strengthening of Structures:
a step-by-step procedure was presented, with Basic Principles of Structural Interventions
detailed instructions about the epoxy resin ▶ Seismic Strengthening Strategies for Existing
injection technique, including details about (Code-Deficient) Ordinary Structures
the substrate preparation, the nozzle placement, ▶ Seismic Strengthening Strategies for Heritage
and the quality control. The use of epoxy resins Structures
for shallow concrete replacements was also ▶ Strengthening Techniques: Bridges
described, and information about the suitable ▶ Strengthening Techniques: Masonry and
material for these applications was presented. Heritage structures
• The technique of strengthening using addi-
tional layers and jackets is one of the most References
commonly used techniques. Two crucial
parameters for the effectiveness of this tech- Abbasnia R, Godossi P, Ahmadi J (2005) Prediction of
nique are the connection between the old and restrained shrinkage based on restraint factors in
the new elements and shrinkage of the new patching repair mortar. Cem Concr Res 35:1909–1913
ACI (2003) Structural Crack repair by epoxy injection,
concrete. field guide to concrete repair application procedures.
• Models for the shear stress at the interface for ACI Committee E706, RAP Bulletin 1
reinforced and unreinforced interfaces and ASTM C42/C42M–99 (1999) Standard test method for
models for the strength deterioration of the obtaining and testing drilled cores and sawed beams
of concrete. West Conshohocken, Pennsylvania
interfaces due to cyclic loading were ASTM C881/C881M (2010) Standard specification for
presented. Epoxy-resin-base bonding systems for concrete.
• For the modeling of the strengthened speci- ASTM International, West Conshohocken
mens, it is essential to simulate the interface ASTM E965 (2003) Standard test method for measuring
pavement macrotexture depth using a volumetric tech-
between the old and the new concrete and the nique. ASTM International, West Conshohocken
shrinkage effect of the new concrete. The Beushausen H, Alexander MG (2006) Failure mechanisms
interface can be simulated using contact ele- and tensile relaxation of bonded concrete overlays
ments, while concrete shrinkage effect can be subjected to differential shrinkage. Cem Concr Res
36:1908–1914
simulated by applying a volumetric strain to Beushausen H, Alexander MG (2007) Localised strain and
the elements of the jacket. stress in bonded concrete overlays subject to differen-
• For the design of strengthened elements, tial shrinkage. Mater Struct 40:189–199
monolithicity coefficient values can be used. CEB Bulletin No. 162 (1983) Assessment of concrete
structures and design procedures for upgrading
Various values have been proposed for the (Redesign), CEB, Paris
design of strengthened RC columns, based on CEB-FIP (1993) Model code 1990. Thomas Telford,
the type of the jacket (four or three sided), the London
normalized axial load value, and the thickness Code of Structural Interventions (2012) Team for devel-
opment of code of interventions on reinforced concrete
S
of the jacket. buildings harmonization team of code of interventions
to Eurocodes. Earthquake Planning and Protection of
Greece (E.P.P.O.), Athens
Cross-References Denarie E, Silfwerbrand J (2004) Structural behaviour of
bonded concrete overlays. In: Proceedings interna-
▶ Assessment and Strengthening of Partitions in tional RILEM workshop on bonded concrete overlays,
Stockholm, pp 37–45
Buildings
Dritsos SE (2005) Repair and strengthening of reinforced
▶ Retrofitting and Strengthening Masonries of concrete structures. Dritsos, Patras (in Greek)
Heritage Structures: Materials Used Dritsos SE (2007) Seismic strengthening of columns by
▶ Retrofitting and Strengthening Measures: adding new concrete. Bull N Z Soc Earthq Eng
40(2):49–68
Liability and Quality Assurance
Dritsos SE (2012) Seismic assessment and retrofit of
▶ Retrofitting and Strengthening of existing buildings under the eurocodes framework.
Contemporary Structures: Materials Used Assessment and Retrofitting Framework, Boku
3554 Strengthening Techniques: Code-Deficient Steel Buildings
ETEP (2012) Recommended technical specifications for Vintzeleou E, Tassios T (1986) Mathematical model for
retrofitting. Technical Chamber of Greece, Athens dowel action under monotonic and cyclic conditions.
(in Greek) Magazine of Concrete Research 38(134):13–22
Eurocode 2 (1996) Design of concrete structures Vintzileou E, Tassios T (1987) Behavior of dowels under
Part 1.3: general rules – precast concrete elements and cyclic deformations. Am Concr Inst Struct J
structures. British Standards Institution; ENV 1992-1-3, 84(1):18–30. http://www.episkeves.civil.upatras.gr
London (Repair and Strengthening of Structures, University of
Eurocode 8 (2005) European standard EN. Design of Patras webpage)
structures for earthquake resistance, Part 3, assessment
and retrofitting of buildings. CEN Technical Commit-
tee CEN/TC250, Brussels
fib Bulletin No 55 (2010) Model code 2010. Thomas,
Telford Strengthening Techniques:
Júlio ES, Branco F, Silva VD (2003) Structural rehabili- Code-Deficient Steel Buildings
tation of columns with reinforced concrete jacketing.
Prog Struct Eng Mater J 5:29–37
Konstantinos Daniel Tsavdaridis
Kupfer H, Hilsdorf H, Rusch H (1969) Behaviour of
concrete under biaxial stresses. Am Concr Inst School of Civil Engineering, University of Leeds,
J 66(8):656–666 Leeds, UK
Lampropoulos AP, Dritsos SE (2010) Concrete shrinkage
effect on columns strengthened with concrete jackets.
Struct Eng Int 20(3):234–239
Lampropoulos AP, Dritsos SE (2011) Modelling of RC Synonyms
columns strengthened with RC jackets. J Earthq Eng
Struct Dyn 40(15):1689–1705 Braced frames; Composite materials; Connec-
Lampropoulos AP, Tsioulou OT, Dritsos SE (2012a)
tions; Cyclic behavior; Deformation capacity;
Biaxial stress due to shrinkage in concrete jackets
of strengthened columns. ACI Mater J 109(3): Energy dissipation; Fuses; Repair; Seismic
331–340 retrofitting; Steel buildings; Strengthening
Lampropoulos AP, Tsioulou OT, Dritsos SE (2012b) techniques
Monolithic coefficient values for design when seismi-
cally strengthening RC columns with jackets. J Earthq
Eng 16(7):1023–1042
Santos P, Júlio E (2010) Comparison of methods for Introduction
texture assessment of concrete surfaces. ACI Mater
J 107(5):433–440
The design of steel buildings is often governed by
Santos P, Júlio E (2013) A state-of-the-art review on
roughness quantification methods for concrete sur- lateral wind loads and not seismic loads. Also,
faces. Construct Build Mater 38:912–923 statistics indicate that the number of fatalities dur-
Tassios T (1986) Fundamental mechanisms of force- ing earthquakes due to failure of all types of steel
transfer across reinforced concrete critical interfaces.
buildings is significantly less compared to other
CEB workshops of Commissions II and IV, Karlsruhe
Tassios T (2009) Design theory for repair and strengthen- types of buildings. Consequently, much effort has
ing. Symmetria Publications, Athens (in Greek) been invested to seismically retrofit buildings hav-
Tassios T, Vintzileou E (1987) Concrete to concrete fric- ing unreinforced masonry walls and reinforced
tion. Jf Struct Eng ASCE 113(4):832–849
concrete frames. However, recently steel buildings
Thermou GE, Pantazopoulou SJ, Elnashai AS (2007) Flex-
ural behavior of brittle RC members rehabilitated with have received significant attention, while this
concrete jacketing. ASCE J Struct Eng 133(10): interest mainly stems from the realization, follow-
1373–1384 ing the 1994 Northridge earthquake, that the
Thermou GE, Papanikolaou VK, Kappos AJ
welded beam-to-column connections in moment-
(2014) Monolithicity factors for the design of R/C
columns strengthened with R/C jackets. In: The 2nd resisting frames were likely to fail in a brittle
European conference on earthquake engineering and manner, prior the development of significant
seismology, Istanbul inelastic response, therefore negating the design
USBR (1997) United States Department of the Interior
intent and possibility causing safety hazards.
Bureau of Reclamation Technical Service Centre,
Guide to concrete repair. http://www.usbr.gov/pmts/ Recent research has expanded the variety and
materials_lab/concrete/ versatility of the tools available in the structural
Strengthening Techniques: Code-Deficient Steel Buildings 3555
engineer’s toolbox to meet the seismic performance structural engineering design and construction
objectives. This entry provides an overview of how phase with some earthquake-proof measures
this research is expanding the available options for taken into consideration. It is the side-to-side
the seismic strengthening of steel buildings, by load which causes the worst damage. Poorly
reporting on some selected research projects. designed buildings often collapse on the first
Structural strengthening and proving seismic shake. The side-to-side load can be even worse
resistance for steel building, but also masonry if the shocks come in waves, as taller buildings
and reinforced concrete, may be done by first can vibrate like a huge tuning fork, while each
considering the direction of the weak links in the new sway is bigger than the last one, until failure.
structures. For instance, for a heavy building with Usually, significant weight is added in time to
large dead load, this would be the major factor that such code-deficient steel buildings (i.e., walls,
contributes to the increase of lateral seismic load. partitions to make more and smaller rooms, etc.)
Therefore, it is reasonable to first consider reduc- or even due to extreme reinforcing techniques.
ing the overall existing dead load and then provide The more weight there is, and the higher this
the necessary strengthening technique for the lat- weight is located in the building, the stronger
eral load-resisting system of the structure. the building and its foundations must be to with-
The use of structural steel in buildings’ stand the earthquake actions. Many buildings
retrofitting can be often considered economical have not been strengthened when such extra
and efficient because: weight was added. These buildings are then
more vulnerable to even a weak aftershock, per-
haps from a different direction or at a different
• Steel buildings are particularly effective under
frequency, and can cause collapse. Moreover, in a
performance-based design.
lot of multistory steel buildings, the ground floor
• Steel members exhibit ductile behavior
has increased headroom with taller slender col-
beyond elastic limit, hence dissipate consider-
umns as well as with more large openings and
able amount of energy before damages occur.
fewer walls. So, these columns, which carry the
• Steel members have higher strength-to-weight
largest loads from both the self-weight and the
and stiffness-to-weight ratios; hence, the
cumulative sideways actions from the seismic
buildings attract less base shear under an
event, are vulnerable and they are often the first
earthquake.
to fail. It only takes one to fail for the worst
• A better quality control is practiced in the
disaster; therefore, it is deemed necessary to cau-
production of the material as well as the fab-
tiously strengthen steel buildings with the most
rication and erection of them, while ensuring
appropriate method.
results close to the theoretical predictions.
The potential deficiencies are different for dif-
• Steel can be generally used to retrofit all types
ferent types of steel buildings (i.e., steel moment S
of structures without increasing the dead
frames, steel braced frames, steel frames with
weight dramatically, making the works less
concrete shear walls, steel frames with infill
intrusive and time consuming.
masonry shear walls). The indicators such as the
global strength and stiffness, the configuration,
Code-Deficient Buildings the load path, the component detailing, the dia-
phragm, and the foundation design demonstrate
All buildings can carry their own weight. They the performance under seismic actions and the
can usually carry a bit of snow and a few other margins for improvement in specific ways;
floor loads vertically, so even badly built build- hence, they should be studied carefully before
ings and structures can resist some up-and-down any decision is taken.
loads. However, buildings and structures are not Retrofitting of existing code-deficient steel
necessarily resistant to lateral loads, unless this buildings accounts for a major portion of the
has been taken into account carefully during the total cost of hazard mitigation. Therefore, it is
important to identify correctly the structures that The vibration occurred from the shock waves is
need and can accept strengthening or weakening, tend to be damped out. This phenomenon is called
while the overall cost should be also monitored. If “plastic hinging” and is easily demonstrated in
appropriate, seismic retrofitting should be steel beams. In extreme earthquake sway, the
performed through several methods such as beams should always be able to form hinges some-
increasing the load, deformation, and energy dis- where, while the columns should behave elasti-
sipation capacity of the structure (FEMA cally. In this way, the frame can deflect and the
356 2000). plastic hinges can absorb energy, while the reso-
nant frequency of the structure is altered without
major loss of strength and inevitable collapse. All
Code-Efficient Buildings Resistant to floors should be connected to the framing in a
Earthquake robust and resilient way and should be as light as
possible. They should possibly span around each
To be earthquake proof, the buildings and their column and be fixed to every supporting beam
foundations need to be built to be resistant to using enough shear connectors (e.g., studs). An
sideways loads. The lighter the building is, the effective way of reducing the vulnerability of
less the loads are. In steel, especially in high- large buildings is to isolate them from the floors
rise buildings, the sideways resistance mainly using bearings or dampers; however, this is an
comes from the diagonal bracing which must be expensive process, and it is not applied to low to
placed equally in both directions. Where possi- medium rise buildings which have not been clas-
ble, the diagonal bracing should be strong sified as important, due to the content they carry
enough to accept tension as well as compression and the occupancy usage.
loads; the bolted or welded connections should Nothing can be though guaranteed to behave
resist more tension than the ultimate tension as such, even in code-based designs; hence, most
value of the brace or much more than the design of the steel buildings and especially those under-
load. If the sideways load is to be resisted with designed with older seismic codes can be consid-
moment-resisting framing, then great care has to ered as code-deficient buildings in certain
be taken to ensure that the joints are stronger circumstances.
than the beams and that the beams will fail
before the columns. Also in such a case, special
care should be provided to the foundation-to- Design Concept for EC8
first-floor level, avoiding soft-story effects,
while the columns should be much stronger Eurocode 8 (EC8) follows three general design
than at higher levels. The foundations could be concepts based on the ductility requirements and
enhanced by having a grillage of steel beams at capacity design considerations of steel buildings:
the foundation level able to resist the high col- the concept of the low-dissipative structural
umn moments and keep the foundations in behavior of DCL structures, the concept of
place. The main beams should be fixed to the dissipative structural behavior of DCM and
outer columns with full capacity joints, which DCH structures satisfying the ductility and
almost mean hunched connections, and care capacity design requirement, and the concept of
should be taken to consider the shear within dissipative structural behavior with steel
the column at these connections. dissipative-controlled zones. In the latter case,
When the steel beams are able to yield and bend when composite action may be considered from
at their highest stressed points, without losing Eurocode 4 (EC4) in the presence of the steel and
resistance, while the connections and the columns concrete (slab) interaction, specific measures
remain full strength, then the resonant frequency have been stipulated to prevent the contribution
of the whole frame changes, while the energy is of concrete under seismic conditions, hence
absorbed and evenly dissipated across the framing. apply general rules for steel frames.
Introduction to Strengthening • The susceptibility to local buckling of out-

Techniques standing flanges as well as the lack of connec-
tion ductility
Preliminary Investigation
It is becoming preferable, both environmentally Often, the nonstructural costs will likely
and economically, to upgrade building structures exceed the structural costs; therefore, the true
rather than to demolish them and rebuild them. costs of a retrofit project are primarily dependent
Engineers assessing structures for increased or on the number of locations of work than the
special loadings are finding that new methods of amount of work done in each location, and thus,
analysis using, for instance, computer models are this influences the structural design and analysis
revealing shortcomings under service and ultimate decisions.
conditions. Under such circumstances, a method The general approach to strengthen existing
has to be found to bring the structure up to the structures includes the following aspects:
required standard. There is a range of techniques
which can be used on structures, but, one must take • Risk assessment and structural vulnerability
into consideration that disruption to normal stage assessment
must be minimal while work is in progress. • Preliminary analysis
Evaluation and subsequent strengthening of • Consideration of alternatives (structural and
existing structures require a realistic and prag- nonstructural)
matic design approach. However, some of the • Detailed design and impact of connections
solutions proposed by researchers do not lie
within this category and they will be eliminated The common goal is to:
in the current study. Also, effective communica-
tion between the owner, structural engineer, • Protect specific structural elements.
architect, risk analyst, insurance provider, and • Provide redundancy to structural systems.
other stakeholders is paramount to a successful • Strengthen a specific part of the structure.
finished solution. In general, in the case where
additional load-carrying capacity is required of Assessing Existing Conditions and
an existing building, engineers have the option of Strengthening Methods
either reinforcing the existing framing or adding
new framing to replace or supplement the Introduction
existing. A site visit should also be performed to inspect
Where a decision is made to strengthen some the building, especially for structures more than
parts of an existing facility or a specific structural 30 years old. Some key things to look for when
system or element, the design approach is assessing the existing condition of a steel build- S
influenced by a series of factors such as: ing are: damage to framing, noticeable corrosion,
signs that modifications to the structure that may
• Information about relevant existing conditions have been performed without engineering
which is often limited review, unusual deflections in floor framing,
• Part of the structure to be strengthened is com- cracks in supported slabs, signs of foundation
monly hidden or obstructed by existing archi- settlement, signs for new rooftop equipment,
tectural or building services systems that are heavy hung piping loads, folding partitions, rig-
difficult or costly to remove ging, or other suspended loads that may have
• Structural renovation work is typically been added without proper structural engineering
constrained by the need for continuity of review (Schwinger 2014). A valuable resource
building operations available to structural engineers working with
• Level of ductility of the existing construction existing building structures is the AISC Steel
may limit its strength Design Guide 15 – AISC Rehabilitation and
Retrofit Guide (Brockenbrough 2002). Other oxy-fuel – while avoiding cuts at areas
publications for further reference are ASCE subjected to high shear):
41-06 2006, FEMA 274 1997, and FEMA • Place reinforcement (e.g., stiffeners)
547 2006. before cutting holes.
The strengthening methods can be categorized (g) Replacement of members (may be
as follows: economical).
(h) Strengthening columns.
1. Passive against active methods. (i) Convert gravity frame to moment-resisting
2. Strengthening techniques: frame.
(a) Reinforcing beams by welding (enlarge
section with plates). Determining Load Capacity of Existing Buildings
(b) Reinforcing (or weakening) connections: Knowing the yielding strength of the steel used in
• Framing. the framing is essential for computing the load
• Seated angles. capacity; therefore, testing should be performed
• Partial-depth end plate. to ascertain and verify the actual yield strength.
• Replace with high-strength fasteners. One technique is to test the steel to determine its
• Add welds at the perimeter of the con- actual yield strength, in hope of finding it to be a
nection and/or properly clean existing higher value than the one that was used in the
welds. original design. Another technique applied in
• Converting single- to double-shear existing structures is to analyze the framing
connections by adding angles or plates. using the load and resistance factor design
• Add web stiffener plate. (LRFD) method (AISC 1999); in LRFD usable
• Add steel cover plates. strength is approximately 1.5 times greater than
• Enhance column splices. the older allowable stress design (ASD) service
• Enhance braced frame connection. level strength.
(c) Shortening span (provided that there are
no fitting issues): Increasing Capacity of Connections
• Add beams. The technique by which existing shear and
• Add columns and girders. moment connections can be strengthened is lim-
• Add diagonal braces. ited only by the imagination of the engineer.
• Add walls with openings. Various techniques are going to be presented
• Add steel braced frame. thereafter, based on well-established but also
• Add concrete, masonry, or steel plate recent research outcomes obtained from numer-
walls. ous computational analyses and experimental
• Enhance strength and ductility of campaigns. It is worth to be aforementioned that
braced frames. the capacities of existing connections must be
(d) Introducing composite action: determined when existing framing is modified
• Steel (partially or fully) encased with or additional capacity is sought.
concrete
• Shear connectors Increasing Flexural Strength Floor Framing
(e) Post-tensioning (or external pre-stressing) Members
of beams and connections (considering There are two options for reinforcing existing
eccentricities of brackets on member flooring systems to support additional loads:
capacity – some need protection from cor- (a) add new framing to supplement the existing
rosion, fire, and vandalism). framing and (b) reinforce the existing beams,
(f) Openings in existing beams (using thermal girders, and connections. The easiest solution is
cutting – plasma arc cutting is faster than usually that of reinforcing the existing structural
Strengthening
Techniques:
Code-Deficient Steel
Buildings,
Fig. 1 Examples of
strengthened beams
Strengthening
Techniques:
Code-Deficient Steel
Buildings,
Fig. 2 Examples of
strengthened columns
elements, provided that the floor slab has suffi- Connecting New Frame to Existing Frame
cient capacity to carry the loads. The most effi- Similarly to the connections, there are numerous
cient way is to weld rectangular high-strength ways that new framing can be connected to
steels (HSS) to the flanges as shown in Fig. 1. existing framing. Welding the new steel mem-
bers to the existing members is a straightforward
Increasing Axial Load Capacity of Columns approach which requires less precision as
The buckling limit state and its variable slender- compared to the bolting process, while drilling
ness should be determined in order to evaluate the new holes through existing steel and bolting
axial load capacity of columns. Column strength- in the field. Various details for connecting
ening serves both to reduce slenderness by new framing to an existing one can be found by
increasing the radius of gyration of the section Schwinger (2014). S
and to reduce stress. Column buckling is a
mid-height phenomenon; therefore, increasing
column stiffness between the supports, not at Detailed Description of Retrofitting and
the supports, is required to increase column Strengthening Techniques
capacity. Both methods shown in Fig. 2 are effec-
tive; however, the one on the left better increases Introduction
the weak axis stiffness of an H-shaped section. The performance of steel frames can be synop-
sized in three very different behaviors:
Dealing with Weldability Issues
Weldability is verified by mechanical and chemi- 1. Formation of plastic hinges
cal testing. The former measures ductility and the 2. Local and global instabilities
latter determines the “carbon equivalent” value. 3. Fracture and structural discontinuity
Strengthening Techniques: Code-Deficient Steel Buildings, Fig. 3 (a) Reduced beam section (RBS) connection
(Crawford 2002), (b) welded haunch connection (Uang et al 2000), (c) bolted brackets
These three behaviors and/or combinations of contribution factors to the brittle fractures. Brittle
them are likely to occur and govern the capacity fractures originated in the beam flange groove
of a connection or member with result on the welds and often propagated to rupture beam
structural continuity and integrity of the system. flanges or columns. A cooperative effort by
Overall, it is known from seismic studies that NISC, AISC, the University of California at San
frame capacity is related to two different aspects Diego, the University of Texas at Austin, and
of frame behavior: Lehigh University examined three techniques
for the retrofit of existing code-deficient steel
1. The member response, as controlled by plas- moment connections trying to force plastic hing-
tic rotational strength and deformation char- ing of the beam away from the column face,
acteristics (including local and global namely, (a) the reduced beam section (RBS) con-
buckling) cept to weaken a portion of the beam near the
2. The connection response, as controlled by column so that plastic hinging would occur at the
bolt fracture, premature brittle weld failure, designated location, (b) the addition of a welded
and panel zone failure haunch to strengthen the steel beam near the
welded connection, and (c) the use of bolted
Determining the capacity of columns is diffi- brackets to reinforce the connection (Fig. 3).
cult as in many situations code-deficient build- More RBS patterns have been developed by
ings are not designed for large lateral loading. Plumier in 1997 and appeared in different config-
urations, as it is shown in Fig. 4.
Steel Connections: Fuses Further analytical research on the same con-
nection complement this work, as design model
Beam-to-Column Connections: Developing and guidelines have been recommended. A target
Ductile Behavior (Fuse Concept) plastic rotation capacity of 0.02 rad was selected.
In parallel with the FEMA/SAC steel research In 2004, Engelhardt recommended (Bruneau
program, the National Institute of Standards and 2004) the following as potential positive
Technology (NIST) and the AISC initiated a solutions:
research project to upgrade existing special
moment frames (SMFs) and investigate the effec- • The use of a bottom flange RBS combined
tiveness of two rehabilitation schemes. Modifica- with the replacement of top and bottom beam
tions to pre-Northridge moment connections to flange groove welds with high toughness weld
achieve improved seismic performance focused metal provided plastic rotations on the order of
on reducing or eliminating some of the 0.02 to 0.025 radian. The presence of a
Strengthening Techniques: Code-Deficient Steel Buildings, Fig. 4 Various RBS patterns
composite slab had little effect on the perfor- It is worth to note that conventional beam
mance of this retrofit technique. theory cannot provide a reliable prediction for
• The addition of a welded bottom haunch, with neither of the above structural systems. Uang
the existing low toughness beam flange groove et al. (2000) and Yu et al. (2000) proposed a
welds left in-place, resulted in significantly simplified model that considers the interaction
improved connection performance, which of forces and deformation compatibility between
was dramatically influenced by the absence the beam and the haunches.
or presence of a composite concrete floor Exhaustive research works have been
slab. In the former case, the specimens devel- conducted on RBS connections varying the geo-
oped plastic rotations of 0.015 to 0.025 radian, metric characteristics of both the beam and the
whereas in the latter case developed plastic connection assembly itself. More recently, RBS
hinges in excess of 0.03 radian. moment-resisting connections have been also
• The use of bolted brackets at the top and investigated by researchers in Europe using Euro-
bottom flanges provided plastic rotations in pean HEA-profile sections (Pachoumis
excess of 0.03 radian. et al. 2008), since so far they have been only
investigated by the US design construction prac- S
Design recommendations have been provided tices. The result is the readjustment of the geo-
for fully restrained, radius cut RBS connections. metric characteristics of the RBS in order to
A step-by-step procedure is presented, with com- apply to the European profiles. Limitation in
mentary for various design considerations. A sim- using RBS is the shear connection between the
ilar procedure is included in FEMA 351 (2000) top steel flange and the metal decking of the steel-
and AISC 358 (2005), which also provides design concrete composite (SCC) slab due to the signif-
guidelines for other prequalified post-Northridge icant width reduction.
connections such as the bolted flange plate (BFP) More recently, the same concept has been
moment connections, the bolted unstiffened applied to steel frames as a strengthening-
(BUEEP) and stiffened extended end plate weakening technique, while introducing a circular
(BSEEP) moment connections, and the so-called opening (Fig. 5) in the beam’s web instead, at a
ConXtech ConXL and Kaiser bolted bracket certain distance from the beam-to-column connec-
(KBB) moment connections. tion, as an effective method to improve the
show that the ultimate displacement of the mod-

ified buildings increases a lot due to the web
openings, and thus, the building ductility is
improved greatly. Moreover, brittle weld frac-
tures can be avoided and the maximum plastic
zone moved to the weakened areas. RWS con-
nections can easily be applied to the beams of
new as well as existing code-deficient buildings
as the web cutting does is away from the com-
posite slab that sits on top of the compression
steel flange. It is worth to mention that different
geometric characteristics and limitations of
beams and columns should be used for different
RWS connection types. There is a need to bring
the attention and propose more experiment phys-
Strengthening Techniques: Code-Deficient Steel
Buildings, Fig. 5 Failure mode of connection with cir- ical testings to validate and establish RWS con-
cular web opening (Qingshan et al. 2009) nections in the current European and American
practices.
Similarly, welded haunches and bolted
aseismic behavior of MRFs (Qingshan brackets are used to move the plastic hinges
et al. 2009). The accurate position and size of the away from the column, but also strengthen the
circular opening has been investigated through existing connection and seek to maintain the
numerical modeling as well as experimental original flexural capacity of the beam.
works, while the plastic hinge positions is effec-
tively controlled. Similar studies have demon- Ductile Behavior: Fuses in Bracing Members
strated the effect of various nonstandard web The concept of adding ductile fuses in bracing
opening shapes (Fig. 6), in enhancing the ductility members of steel concentrically braced frames
but also the strength of the connections (CBFs) resisting seismic loads is well linked to
(Tsavdaridis et al. 2014). Step-by-step procedures the RBS technique. Current code provisions
have been proposed to determine the most suitable require that steel CBFs are designed to exhibit
geometries to achieve adequate connection ductile energy dissipation. Limits on brace over-
strength, ductility, stiffness, and rotational capac- all slenderness ratio must be satisfied to achieve
ities. Such techniques prove suitable in cases ductile inelastic behavior. It is apparent that
where large plastic rotations are required (i.e., implementation of this design approach may
larger than 0.03 rad). Tsavdaridis et al. (2014) result in significant increases in design loads for
have further proposed the use of previously pat- brace connections, beams, and columns (Egloff
ented novel elliptically based web opening shapes, et al. 2012).
which can also be used for perforated beams (e.g., In order to reduce seismic design loads, ductile
cellular and castellated beams) adding numerous fuses in bracing members have been recently
advantages from the manufacturing process to proposed, as they control their axial resistances.
their life-span, while they can develop rotational Such behavior can be achieved by locally reduc-
capacities up to 0.05 rad with insignificant strength ing the brace cross-sectional area or by introduc-
degradation (Fig. 7). ing ductile components that yield in both tension
The so-called reduced web section (RWS) and compression. In the former case, the reduced
connections have been studied, yet not exten- section might need to be confined to prevent
sively, assessing local connection models mainly local buckling, or it can be resized (reshaped -
computationally with cyclic (quasi-static), and restructured) to yield in tension while remaining
pseudo-dynamic (PSD) analyses. The results elastic in compression, which means that the
S
Strengthening Techniques: Code-Deficient Steel Buildings, Fig. 6 Types of perforated beam webs (Hedayat and
Celikag 2009)
effect of fatigue loading will be minimized. In the open bracing members has been proposed by
latter case, the overall buckling is eliminated, and Vincent in 2008. A part of the flange to web
hence, the strength degradation is limited due to intersection is removed to limit the impact on
symmetrical hysteretic behavior. This type of the brace flexural stiffness and buckling resis-
fuse technique can be applied in open and closed tance. In 2012, Egloff et al. introduced a new
profile sections of the bracing members, while it local buckling restraining system (LBRS) which
has been noticed that the former ones perform includes two cold-formed channels that support
better exhibiting higher ductility. A special fuse the web and the flanges. Moreover, external cover
(Fig. 8) for controlling the tension resistance of plates can be bolted to the channels in order to
Strengthening Techniques: Code-Deficient Steel Buildings, Fig. 7 Geometric parameters of various novel
perforated beam-to-column moment connections and von Mises plastic stresses (Tsavdaridis et al. 2014)
Strengthening Techniques: Code-Deficient Steel Buildings, Fig. 8 Ductile fuse for H-shaped bracing members
(Vincent 2008)
prevent local buckling of the brace flanges. Splice improvements and the design procedure for the
plates can also be used to provide lateral support fuse and the fuse LBRS are available in the
to the flanges in the fuse (Fig. 9). This LBRS can literature.
slip longitudinally with respect to the brace so Cast Connex, a high-strength steel connector
that it does not attract any axial forces. Further for round hollow structural section brace
Strengthening Techniques: Code-Deficient Steel Buildings, Fig. 9 Proposed fuse local buckling restraining
mechanism (Egloff et al. 2012)
Strengthening Techniques: Code-Deficient Steel Buildings, Fig. 10 Structural details of pin-fuse connection
(Coedova and Hamburger 2011)
members in CBF, has also developed a yielding column joint. The bolts are designed to slip S
fuse connector for CBF, called the Scorpion within slotted holes allowing the pin joint to
Yielding Brace System (SCBF), that relies on rotate, dissipating the energy through frictional
the flexural yielding of fingerlike plates that are resistance. This joint acts as the fuse for the
specially designed to dissipate energy locally and system, while the rest of the steel frame can be
can be used in both architecturally exposed and designed to remain elastic. Following an earth-
nonexposed braced bays. quake, and avoiding damages, the frame can be
adjusted to its initial position, reducing the
Pin-Fuse Joints potential for permanent residual displacements
A new type of connection, which begun its while both the connection and the frame main-
prequalification process in 2011, is the pin-fuse tain their structural integrity and reduce the need
connection (Fig. 10) which incorporates a for costly structural repairs. The simplicity of
curved plated end connection using slip-critical the pin-fuse connection offers the ability to
bolts and a steel pin adjacent to the beam-to- accommodate braces and dampers.
Strengthening Techniques: Code-Deficient Steel Buildings, Fig. 11 Proposed connections with replaceable link
(Shen 2009)
Replaceable Links which can be removed and replaced with smaller

Using RBS- and RWS-type connections has been flexural capacities are used at the locations of the
proved very efficient for certain applications; expected inelastic actions. Consequently, the other
however, there are some drawbacks. As the yield- structural elements in the frame will remain elastic
ing fuse is a part of the beam, strength design and during an earthquake. This efficient method of
drift design of the structure are interlinked. For repair for MRFs allows for quick inspection and
instance, due to increased drift requirements, the replacement of damaged links while it minimizes
capacity of the yielding fuse may be also the disruption time. Further, the welding of critical
increased, which then leads to higher demands elements of beam-to-column connections can be
on the other parts of the structure including col- done in the shop while improving construction
umns, floor slabs, connections, and foundations quality and significantly reducing the initial erec-
and often resulting in overdesigned buildings tion time (Shen 2009).
with increased overall cost. Further, significant In particular, two types of replaceable links
damage can result in the beam from repeated have been proposed by Shen (2009):
inelastic deformation and localized buckling (i) H-section with end plates and (ii) back-to-
(hence crack propagation) during a design-level back channels eccentrically bolted to the beam
earthquake. As this cumulative inelastic action of web. The former one is prepared in the shop using
the building cannot be precisely anticipated, it is complete joint penetration welds. The end plate
not trivial to assess the extent of damage on site (flush or extended) is then bolted to the column
and the residual capability of the structure to flange using prestressed high-strength bolts. The
adequately provide the required level of safety latter type of these double channel-built sections
for any subsequent loading. In such a case, repair intended to act as truss girders in special truss
of the beam is not a straightforward procedure moment frames, and they have been connected
and it can be disruptive and costly. using welded reinforcing gusset plates. In cer-
The replaceable link concept (Fig. 11) effec- tain circumstances, lateral bracing is deemed
tively eliminates the aforementioned concerns necessary in the region adjacent to the plastic
instead of reducing the beam section size; hinge to achieve large (i.e., 0.06 rad) plastic
dismountable dissipative elements (Fig. 12) rotation of the hinge. However, sometimes
Strengthening Techniques: Code-Deficient Steel Buildings, Fig. 12 End plate model boundary conditions (Shen
2009)
large overstrength has been observed in these Steel Connections: Stiffeners

buildup channel sections.
The end plate links can exhibit 0.04 rad con- Introduction
tributing to 90 % of the total story drift and Forcing the plastic deformation to the beam end
demonstrate higher energy-dissipating capacity away from the connection is a common practice
than the double channel links, while some in seismic moment-resisting frames, but in con- S
strength degradation occurs due to ductile local trast to the fuse concept, this can be also achieved
flange and web buckling. On the other hand, by increasing the relative stiffness of the column
higher story drift can be reached by the double and the connection with respect to the beam.
channel links before experiencing strength deg- Eventually, this is an alternative in effectively
radation at 0.06 rad. The degradation has been controlling the position and intensity of the plas-
also caused due to ductile tearing of the flanges tic hinge in the connection zone when such mod-
and the webs. Overall, double channel links type ifications are allowed.
has been considered to be preferable as it pro-
vides a more gradual transmission of forces at the SidePlateTM Connections
connections via friction. For more stability, fur- A well-promising retrofitting method for
ther modifications can take place enhancing the upgrading an existing traditional moment-resisting
connection of the channel webs and the connec- connection is shown in Fig. 13. This concept uses
tion of the beam segments. the so-called SidePlateTM retrofit system where the

Buildings, Fig. 13 SidePlateTM retrofit connection
(Crawford 2002)
physical separation between the face of the column

flange and the end of the beam, for mitigating the Strengthening Techniques: Code-Deficient Steel
stress concentration, is achieved using parallel Buildings, Fig. 14 Strengthening retrofit concept:
full-depth side plates which act as discrete conti- concrete-filled hollow section of column (Crawford 2002)
nuity elements to sandwich the connecting
beam(s) and column (Crawford 2002). Similar
design concept can be used for steel and Stiffeners at Connections
concrete-filled hollow section columns (Fig. 14). In addition to the prequalified connections for
Whole steel frame is eventually stiffened, and SMFs and intermediate moment frames (IMF)
the zone panel deformation is eliminated using presented in AISC (1999) and the SidePlateTM
this type of connections due to the increased system, research has been focused on the effect of
stiffness of the side plates that ultimately provide stiffeners on the strain patterns of the welded
the three panel zones. This connection system connection zone.
uses all fillet-welded fabrication which predomi- For example, the effect of both internal and
nately carries all shear actions as well as external stiffeners on the behavior of I-beam to
moments through the combination of vertical hollow-column section connections (Fig. 15) has
shear plates and fillet welds. The side plates been initially thoroughly investigated by Chen
should be designed with sufficient strength and and Lin (1990). It has been observed that the
stiffness to force all significant plastic behavior of connections with triangular stiffeners have the
the connection system into the beam. lowest rigidity, in contrast to those with side
The same system can be used to upgrade con- stiffeners which present significantly higher
struction of deficient buildings. The difference is moment rotation capacities, stiffness, and ductil-
that an initial hole is required in each side of the ity. Moreover, the performance of the retrofitted
plates to permit welding access, while the holes connections with side stiffeners has been investi-
are closed with the same cutoff plate following gated (Ghobadi et al. 2009) and design guidelines
the completion of the welding process. All new proposed. The benefits of using side stiffeners
welds are again fillet welds loaded in shear along have been also introduced on concrete-filled tab-
their length, whereas if there are any existing ular (CFT) columns connected to I-beams, while
Complete Joint Penetration (CJP) welds, they stable hysteresis and adequate ductility are pro-
are removed by air arcing to eliminate the reli- vided. Overall, it has been concluded that con-
ance on through thickness properties and triaxial nections with both column stiffeners and
stress concentrations. More information can be top-flange stiffeners have the highest value of
found from FEMA 351 (2000). energy dissipation, while the beam top-flange
to girders or cables with connections to beams is

very important especially for high-rise buildings.
Self-centering braces have been designed and built
using prestressed aramid fiber strands in conjunc-
tion with friction pads or memory alloys. Energy
dissipation is implemented using yielding seat
angles, friction dampers, or energy-dissipating
bars confined in tubes. Researchers have investi-
gated self-centering column bases that use post-
tensioned (PT) bars or spring-loaded wedges.
Tendons can span over multiple floors, while elas-
tomeric spring dampers and fuse bars can be used
to provide energy dissipation.
Furthermore, self-centering structural systems
(Fig. 17) have been proposed, for the seismic
retrofit of special moment-resisting frames, by
Christopoulos et al. (2002). This is a post-
tensioned energy-dissipating (PTED) steel frame
design, where high-strength bars or tendons pro-
Strengthening Techniques: Code-Deficient Steel vide the post-tension at each floor. Confining steel
Buildings, Fig. 15 Typical I-beam to box-column con-
sleeves have been often used to prevent the
nection (Kiamanesh et al. 2010)
energy-dissipating bars from buckling during
cyclic inelastic loading. It has been concluded
stiffener is the most effective one, especially that these economical innovative systems:
when it is incorporated with the column stiffeners
of a hollow section. • Incorporate the nonlinear characteristics of
yielding structures and, thereby, limit the
Steel Frames: Modifications induced seismic forces and provide additional
damping characteristics.
Frame Modification at Beam’s Mid-Span (Fuse • Encompass self-centering properties allowing
Concept) the structural system to return to its original
A retrofitting method which can be used for new position after an earthquake.
construction as well as a strengthening technique • Reduce or eliminate cumulative damage to the
for existing moment-resisting frames has been main structural elements.
developed by Leelataviwat et al. (1998). This tech- S
nique replaces certain beams and is introducing a Later, Garlock et al. (2004) has proposed a
ductile fuse element in shear at their mid-span similar structural system with high-strength steel
instead of modifying the beam-to-column connec- PT strands, after bolted replaceable top-and-seat
tions (Fig. 16). A braced rectangular opening is angles have been installed (Fig. 18). Here, the
created in the web of each girder at the mid-span, vertical shear is supported by both the angles and
to move the plastic deformation away from the the friction between the beam and the column, and
critical connection regions, while ensuring the it is expected to continue to perform if failure of
development of a ductile mechanism. one or more strands occurs. It is proved that this
system can achieve greater strength and ductility.
Cabling: Self-Centering Systems Recently, researchers have designed and exper-
The use of the cables is another promising tech- imentally evaluated a new self-centering PT con-
nique, which can be applied to both slabs and nection using yielding web hourglass shape pins
connections. Placement of cables with connections (WHPs) as seismic energy dissipaters. WHPs do
Strengthening Techniques: Code-Deficient Steel Buildings, Fig. 16 Frame modified with mid-span truss opening
(Leelataviwat et al. 1998)

Buildings, Fig. 17 PTED system (Christopoulos
et al. 2002)
not interfere with the composite slab and can be

very easily replaced without the need for welding Strengthening Techniques: Code-Deficient Steel
and bolting and, therefore, can significantly Buildings, Fig. 18 Post-tensioned moment connection
decrease downtime in the aftermath of a strong with top and bottom seat angles (Garlock et al. 2004)
earthquake. Repeated experiments are described
in detail and proved the reparability of the PT enlarging the existing members. Bracings,
connection with WHPs (Vasdravellis et al. 2013). masonry (Fig. 19), and reinforced concrete (pre-
and post-cast infill) shear walls are the most pop-
Structural System: Adding Structural ular and efficient strengthening techniques as
Elements (Walls, Blocks, Bracings) they provide lower overall cost and they are
easy to use. Braces are more effective due to
Introduction their much higher ductility, but the shear walls
Conventional retrofitting methods include the are indeed the most commonly applied method,
addition of new structural elements to the system, as they also reduce the demand on the other
Strengthening Techniques: Code-Deficient Steel Buildings, Fig. 19 Ordinary hybrid wall (Adopted by http://
masonryedge.com/site/mim-archives-thestorypole-vol-38-no-4/143-filling-in-the-frame-)
structural members resisting large lateral loads, wall, polyurethane spray to prevent punching
hence increasing their safety in a simple manner. shear failure. Moreover, openings can be accom-
The actual capacity of bearing walls has been modated in such walls, especially with multistory
often underestimated or even ignored. However, buildings, while further enhancements are
it can be a major contributor, and it can provide needed.
the required capacity, without the need of more
complex strengthening techniques (Sarhosis et al. Steel Plate Shear Walls
2014). In addition to concrete and reinforced concrete
walls with SCC beams, research has been initially
Bearing Walls conducted by Thorburn et al. (1983) later on steel
Walls must go equally in both directions, and plate shear wall (SPSW) design (Fig. 20a) and
they must be strong enough to add stiffness to retrofitting methods. SPSWs can be used as the S
the steel framing system while they are tied in to primary lateral force resisting system in steel
any framing in order to take load in their weakest buildings allowing the occurrence of shear buck-
direction. Also, they must not fall apart and must ling. Following buckling, diagonal tension field is
remain in place after the worst shock waves, so as developed to transfer the lateral load in the panel,
to retain strength for the aftershocks. while the forces in beams and columns are
In particular, three approaches have been iden- reduced. Furthermore, the use of low-yield-
tified (Crawford 2002) for enhancing the resil- strength steel panels and RBS connections as
ience of a building’s bearing walls under the well as light-gauge cold-formed steel plates has
progressive collapse and seismic scenarios. been examined as potential applications by
These are the following: (i) backup wall, build a Berman and Bruneau (2004). The former one
second wall or gravity-carrying frame to support demonstrates an earlier onset of energy dissipa-
the existing wall; (ii) strong wall, employ fabric tion by the panel, while perforated panel speci-
retrofit to control the breach area; and (iii) ductile mens (Fig. 20b) can be used to control the
Strengthening Techniques: Code-Deficient Steel Buildings, Fig. 20 (a) SPSW specimen with cutout corners
(right), (b) buckled panel and RBS yielding of SRW specimen (Vian and Bruneau 2004)
stiffness and overstrength issues using hot-rolled adding zipper columns which connect the inter-
plates. This option is also useful in a retrofit section point of the braces above the first floor.
situation, providing access for utilities to pene- The zipper columns tie all brace-to-beam inter-
trate the predesigned system. section points together and force compression
braces in a braced bay to buckle simultaneously,
Braced Frames hence better distribute the dissipated energy over
Braced frames can be constructed of single diag- the height of the building (Bruneau 2004).
onal, x and k braces, chevron and split braces, and A suspension system has been proposed later,
lattice or knee braces, and they can be used in ensuring that the top story braces are designed
interior cores – so connections could be easily to remain elastic, whereas all other compression
made with wall panels, as well as in the exterior. braces are designed to buckle, while the
Composite braced frames are also becoming pop- suspended zipper struts are designed to yield in
ular where concrete bracings are supporting steel tension. Therefore, adequate ductility is pro-
frames. vided, with superior seismic performance.
A simplified design procedure for suspended Engaging the fuse, the replaceable links, the
zipper frames, initially proposed by Khatib self-centering, and the braced frame retrofitting
et al. (1988), has been introduced by Leon and concept, Eatherton et al. (2008) proposed a con-
Yang (2003) and consists of inverted V-braces trolled rocking system which virtually eliminates
Strengthening Techniques: Code-Deficient Steel Buildings, Fig. 21 A non-buckling brace
residual drifts and concentrates the majority of effective technique to add strength, and it can be
structural damage in replaceable fuse elements. applied for a particular group of members, such
The system is consisted of three components: as on the ground floor. Thus, concrete encase-
(a) a stiff steel braced frame which remains elas- ment of columns (Figs. 14), and floor beams
tic, but it is not tied to the foundation and hence (Fig. 22) as proposed (Tsavdaridis et al. 2013),
allowed to rock, (b) a vertical post-tensioning that constitute forms of strengthening techniques for
strands the top of the frame down to the founda- new and existing steel buildings.
tion and brings the frame back to the center, and
(c) the replaceable fuses that absorb the energy as
the frame rocks. Materials
Non-Buckling Braces Innovative ways have been explored for the

Conventional braces tend to buckle under the strengthening and rehabilitation of deficient
compression cycle of the seismic load, hence dis- steel buildings, due to the demand to access the
sipating little energy under compression. This specified load capacity and the deterioration as a
causes pinching of the hysteresis loop and failure result of corrosion. In particular, externally
of the braces within a few cycles, due to the bonded fiber-reinforced polymer (FRP) compos-
formation of plastic hinge close to mid-length of ites can be applied to various structural members
the member. The use of non-buckling braces (also such as columns, beams, slabs, and walls in order
known as buckling restrained braces or un-bonded to improve their structural performance in terms
braces) bypasses this problem. In this type of of stiffness, load-carrying and deformation
bracing system, the requirements of adequate capacity, and ductility, while simultaneously pro-
strength to resist compression, as well as rigidity viding environmental durability. Generally, S
to avoid buckling, have been addressed separately FRPs have been widely used mostly in applica-
to a core and a sleeve (Fig. 21). tions that allow complete wrapping of the mem-
In the last decade, buckling restrained braced ber, while attention deemed necessary to avoid
(BRB) frames have received much attention in brittle shear and de-bonding failures, especially
the United States, as they demonstrate stable hys- prone when used on steel. In such cases, the
teretic behavior and excellent low-cycle fatigue actual member can entirely waste the strengthen-
life characteristics. However, buckling and ing application, or the composite material might
cracking of gusset plates are expected in certain harm the member itself by decreasing its ductility
cases, similarly to all types of braced frames. (Buyukozturk et al. 2003). It has been proposed
that for buildings with large seismic deficiencies,
Strengthening Members a combination of conventional and FRP strength-
Strengthening members by adding plates or ening techniques may prove to be an effective
encasing/filling them with concrete provides an retrofitting solution.
Tie-bar Any Depth

Pre-cast Steel
units decking
75mm min 50mm min

bearing for bearing for
PC units steel decking
Strut
action Bearing
pressure on
flange
Local loading
from slab
Cross section of a Flexural action of

USFB at the web flanges of USFB
opening
Strengthening Techniques: Code-Deficient Steel Buildings, Fig. 22 Schematic representation of the USFB
system and the internal actions (Tsavdaridis et al. 2013)
The preformed high-strength carbon fiber stresses necessary to appreciate the full strength
plates currently being used for concrete struc- of these materials (Lenwari et al. 2006).
tures are typically 4 mm thick. To strengthen
steel beams, they would need to be at least Energy Dissipation and Active/Passive
20 mm thick, in order to achieve a significant Structural Control Systems
increase in bending moment for the steel or SCC A quite different retrofitting method, which can
beam. Consequently, new high-modulus carbon be quite cost-efficient, is the installation of com-
fiber-reinforced polymer (CFRP) materials are plementary energy dissipation devices in struc-
likely to provide solutions for steel structures’ tures as a means of passive, semi-active, or
deterioration issue (Schnerch et al. 2005). Time active structural control systems. These are not
should be allowed for the surface preparation, described thoroughly here, as it is beyond the
application of the adhesive, and curing time scope of the present entry. The main objective of
(usually between 4 and 8 h). Specific surface structural control is to minimize structural vibra-
preparation and detailing are critical to ensure tions improving safety and serviceability limits
adequate bond interaction between steel and mainly under wind and earthquake actions.
FRP materials, both in the short and long term, Up to date, the majority of passive energy dissi-
and capable of sustaining the high interfacial pation devices have been found very effective in
Strengthening Techniques: Code-Deficient Steel Buildings, Fig. 23 Supplemental energy dissipation devices
(Adopted from MIT.edu website)
controlling the seismic response of steel frames. which are not within the domain of traditional
Further advanced techniques seem very promis- civil engineering. In particular, the control the-
ing such as the introduction of an “inerter” ory is elaborated with computer science, data
(Smith 2002; Marian and Giaralis 2014), and processing, sensing technology and materials
its combined use with the already well-operated science using the knowledge and principles of S
tuned mass dampers, with scope to reduce earthquake (and wind) engineering, structural
the size of the mass required to control and dynamics, as well as stochastic processes. It is
dissipate the energy of high-rise buildings. essential to mention though, that the effective-
Figure 23 shows the basic principles of various ness of such dissipation devices is predomi-
control systems commonly used on building nantly dependent of the deformation capacity
structures. of the structure. Consequently, the application
There is a vast research conducted on energy of such devices to code-deficient buildings with
dissipation devices during the past 20 years, inadequate seismic detailing or post-earthquake
while their use becomes more direct and appar- damages should be carefully considered, and
ent with the upsurge of technology. However, a perhaps it should be combined with an appropri-
diverse background of researchers is required, ate strengthening technique with deformation
integrating a number of disciplines, some of enhancement measures as proposed above.
Ongoing research on special strengthening ▶ Retrofitting and Strengthening of

techniques involving the use of simple and robust Contemporary Structures: Materials Used
active control systems, while emphasizing on the ▶ Steel Structures
performance of various controllers (Demetriou ▶ Strengthened Structural Members and
et al. 2014) as well as new conceptual methods Structures: Analytical Assessment
introducing the ability of structures to adapt ▶ Tuned Mass Dampers for Passive Control of
under dynamic loads (Slotboom et al. 2014), are Structures Under Earthquake Excitations
posing great expectations.
References
Summary
AISC 1999 LRFD (1999) Specification for structural steel
It is worth to mention that there are factors which buildings. American Institute of Steel Construction,
inhibit retrofit design when it comes to strength- Chicago
AISC 358 (2005) Specification for structural steel build-
ening existing buildings. Many times important ings. American Institute of Steel Construction,
issues are misaddressed due to the complexity of Chicago
strengthening design concepts, lack of technol- ASCE 41–06 (2006) Seismic rehabilitation of existing
ogy understanding or even ignoring it, while buildings. American Society of Civil
Engineers – ASCE, Reston
uncertainties about the design of the building Brockenbrough RL (2002) AISC Rehabilitation Bernam
involved. Therefore, existing buildings should on and retrofit guide – a reference for historic shapes
be best approached with a risk-based retrofit and specifications, Steel design guide series 15. Amer-
scheme in order to concentrate the works where ican Institute for Steel Construction, Chicago
Bruneau M (2004) Seismic retrofit of steel structures. In:
they are actually needed most. In this way, more Sociedad Mexicana de Ingeniera Sismica, VIII SNIS,
safe, effective, and cost-efficient steel buildings Tlaxcala
will stay operated in the future. This inherently Buyukozturk O, Gunes O, Karaca E (2003) Progress on
requires more skilled designers and engineers, understanding de-bonding problems in reinforced con-
crete and steel members strengthened using FRP com-
better design tools, and more truly innovative posites. Construct Build Mater 18(1):9–19
“smart” retrofitting techniques to be developed Chen SJ, Lin HY (1990) Experimental study of steel
for the seismic strengthening of code-deficient I-beam to box-column moment connection. In: 4th
steel buildings. international conference on steel structures and space
frames, Singapore, pp 41–47
Christopoulos C, Filiatrault A, Uang CM (2002) Self-
centering post-tensioned energy dissipating (PTED)
steel frames for seismic regions. In: Structural systems
Cross-References research project report no. SSRP-2002/06, Department
of Structural Engineering, University of California,
▶ Assessment of Existing Structures Using San Diego, p 292
Inelastic Static Analysis Coedova PP, Hamburger RO (2011) Steel connections:
proprietary or public domain. Modern Steel Construct
▶ Buckling-Restrained Braces and Their 51(10):24–31
Implementation in Structural Design of Steel Crawford JE (2002) Retrofit methods to mitigate progres-
Buildings sive collapse, 2002. Online: available in May 2014
▶ Buildings and Bridges Equipped with Passive Demetriou D, Nikitas N, Tsavdaridis KD (2014) Perfor-
mance of proportional-integral-derivative controllers
Dampers Under Seismic Actions: Modeling
on structures with variable damping tuned mass
and Analysis dampers. In: 6th World conference on structural con-
▶ Passive Control Techniques for Retrofitting of trol and monitoring, Barcelona
Existing Structures Eatherton MR, Hajjar JF, Deierlein GG, Krawinkler H,
Billington S, Ma X (2008) Controlled rocking of steel-
▶ Performance-Based Design Procedure for
framed buildings with replaceable energy-dissipating
Structures with Magneto-Rheological fuses. In: The 14th World conference on earthquake
Dampers engineering, Oct 12–17, Beijing
Egloff O, Tremblay R, Vincent R, Dussault S (2012) subjected to cyclic loading: experimental analysis
Finite element analysis of ductile fuses for W-shape and FEM simulation. Eng Struct 31:216–223
steel bracing members. In: 15th World conference of Plumier A (1997) The dog-bone: back to the future.
earthquake engineering, Lisboa Eng J 34(2):61–67
FEMA 274 (1997) NEHRP commentary on the guidelines Qingshan Y, Bo LA, Na Y (2009) Aseismic behaviors of
for the seismic rehabilitation of buildings. Report steel moment resisting frames with opening in beam
no. FEMA 274. Federal Emergency Management web. J Construct Steel Res 65:1323–1336
Agency, Washington, DC Sarhosis V, Tsavdaridis KD, Giannopoulos I (2014) Dis-
FEMA 351 (2000) Recommended seismic evaluation and crete element modelling of masonry infilled steel
upgrade criteria for existing welded steel moment frames with multiple window openings subjected to
frame buildings. Report No. FEMA 351. Federal lateral load variations. The Open Construction and
Emergency Management Agency, Washington, DC Building Technology Journal 8:93–103
FEMA 356 (2000) Prestandard and commentary for the Schnerch DA (2005) Strengthening of steel structures with
seismic rehabilitation of buildings. In: Report high module carbon fiber reinforced polymer (CFRP)
No. FEMA 356. Federal Emergency Management materials. PhD dissertation, North Carolina State Uni-
Agency, Washington, DC versity, p 265
FEMA 547 (2006) Techniques for the seismic rehabilita- Schwinger C (2014) Quality assurance for structural
tion of existing buildings. Report No. FEMA 547. engineering firms. Online: available in May 2014
Federal Emergency Management Agency, Washing- http://www.aisc.org/store/p-1795-quality-assurance-
ton, DC for-structural-engineering-firms.aspx
Garlock M, Ricles JM, Sauce R (2004) Experimental Shen Y (2009) Seismic performance of steel moment-
studies on full-scale post-tensioned steel moment con- resisting frames with nonlinear replaceable links. Mas-
nections. In: 13th World conference of earthquake ter Thesis, Department of Civil Engineering, Univer-
engineering, Vancouver sity of Toronto
Ghobadi MS, Ghassemieh M, Mazroi A, Abolmaali Slotboom M, Robbemont A, Habraken A, Teuffel
A (2009) Seismic performance of ductile welded con- P (2014) Safety and redundancy of adaptive buildings
nections using T-stiffener. J Construct Steel Res structures. In: Civil engineering for sustainability and
65:766–775 resilience international conference, CESARE’14,
Hedayat A, Celikag M (2009) Post-Northridge connection Amman
with modified beam end configuration to enhance Smith MC (2002) Synthesis of mechanical networks: the
strength and ductility. J Construct Steel Res 65 inerter. IEEE Trans Autom Contr 47(10):1648–1662
(7):1413–1430 Thorburn LJ, Kulak GL, Montgomery CJ (1983) Analysis
Khabit IF, Mahin SA, Pister KS (1988) Seismic behavior of steel plate shear walls. In: Structural engineering
of concentrically braced steel frames. Report no. UCB/ report no. 107. Department of Civil Engineering, Uni-
EERC-88/01. Earthquake Engineering Research Cen- versity of Alberta, Edmonton
ter, University of California, Berkeley Tsavdaridis KD, D’Mello C, Huo BY (2013) Experimental
Kiamanesh R, Abolmaali A, Ghassemieh M (2010) and computational study of vertical shear behaviour of
The effect of stiffeners on the strain patterns of the partially encased perforated steel beams. J Eng Struct
welded connection zone. J Construct Steel Res 66(1): 56:805–822
19–27 Tsavdaridis KD, Faghih F, Nikitas N (2014) Assessment
Leelataviwat S, Goel SC, Stojadinovic B (1998) Drift and of perforated steel beam-to-column connections
yield mechanism based seismic design and upgrading subjected to cyclic loading. J Earthquake Eng
of steel moment frames. Research report no. UMCEE 18(8):1302–1325
98–29. Department of civil and environmental engi- Uang CM, Yu QS, Noel SA, Gross JL (2000) Cyclic test- S
neering, The University of Michigan, Ann Arbor ing of steel moment connections rehabilitated with
Lenwari A, Thepchatri T, Albrecht P (2006) Debonding RBS or welded haunch. J Struct Eng 126(1):57–68
strength of steel beams strengthened with CFRP plates. Vasdravellis G, Karavasilis TL, Uy B (2013) Large-scale
J Comp Constr 10(1):69–78 experimental validation of steel post-tensioned con-
Leon RT, Yang CS (2003) Special inverted-V-braced nections with web hourglass pins. J Struct Eng ASCE
frames with suspended zipper struts. Georgia Institute 139(6):1033–1042
of Technology Vian D, Bruneau M (2004) Testing of special LYS steel
Marian L, Giaralis A (2014) Optimal design of a plate shear walls. In: 13th World conference on earth-
novel tuned mass-damper-inerter (TMDI) passive quake engineering, Vancouver, Paper no. 978
vibration control configuration for stochastically Vincent RB (2008) Minimizing the strength of bracing
support-excited structural systems. Prob Eng Mech. connections. In: 6th international workshop connec-
http://www.sciencedirect.com/science/article/pii/S02668 tions in steel structures VI. Chicago, pp 127–141
92014000216 Yu QS, Uang CM, Gross JL (2000) Seismic rehabilitation
Pachoumis DT, Galousis EG, Kalfas CN, Christitsa AD design of steel moment connection with welded
(2008) Reduced beam section moment connections haunch. J Struct Eng ASCE 126(1):57–68
3578 Strengthening Techniques: Masonry and Heritage Structures
(exhaustive surveys, in situ and laboratory tests,

Strengthening Techniques: Masonry monitoring, sophisticated analyses, etc.) leading
and Heritage Structures to the selection of adequate intervention tech-
niques by multidisciplinary national or interna-
Elizabeth Vintzileou tional teams. Thus, they are left out of the scope
Department of Structural Engineering, Faculty of of this text. Furthermore, structures other than
Civil Engineering, National Technical University buildings (e.g., towers, bridges, etc.) present
of Athens, Athens, Greece peculiarities that need to be treated separately.
Therefore, those structures as well are not cov-
ered by this text. However, the repair and
Synonyms strengthening techniques, concisely presented
and treated herein, are applicable also to the cat-
Arch; Box action; Deep rejointing; Diaphragm; egories of structures that are not explicitly
Dome; Foundation; Grouting; Masonry; Mortar; covered.
Repair and strengthening; Stitching; Ties; Vault
Prerequisites for the Selection of

Introduction Adequate Intervention Techniques:
Intervention Strategies
Masonry structures constitute a large part of the
existing structures throughout the world. Interventions to individual structural members or
A significant part of the existing masonry struc- to parts of an existing masonry structure should
tures belong to the built cultural heritage. Fur- be part of a global intervention strategy. The
thermore, many countries with rich cultural selection of a strategy and, hence, of appropriate
heritage are situated in earthquake-prone areas intervention techniques (for the repair or
(in Europe, in Asia, in South America). The strengthening of a structure) is based on a series
built cultural heritage includes unique monu- of investigations and studies that should precede
ments (temples, churches, castles, palaces, brid- the redesign of the structure, as briefly described
ges, etc.) and (inhabited and leaving) historic herein:
centers. Therefore, there is a continuous need
for the preservation of this wealth, whose cul- (a) Documentation of the history of the struc-
tural, historical, social, and economic values can- ture: It comprises the date of construction,
not be overestimated. information on construction phases, events
On the other hand, the international scientific that may have affected the structural system
community is working quite intensively in the (earthquakes, floods, fires, wars, etc.), previ-
field of preservation of the existing stock of struc- ous damages and interventions, etc. This
tures, and, thus, the knowledge acquired in the phase is of paramount significance especially
last decades allows for interventions to be more for cultural heritage structures.
rationally selected and better applied. (b) Documentation of the structural system: It
It is obvious that the vast variety of existing comprises the documentation of both the
masonry structures, in terms of typology, of structural system and its current state. The
materials used for their construction, of their cul- documentation of the structural system
tural significance, etc., does not allow this text to includes the survey of the bearing system
be of general applicability and to cover all possi- (bearing elements, their interconnections,
ble cases. Thus, to make this text to be of some path of forces from the region of their appli-
use, one has to limit its scope to a part of the cation to the Earth), as well as type and prop-
existing masonry structures. Unique monuments erties of materials, construction type of
are usually the subject of specific studies masonry, etc. The documentation of the
Strengthening Techniques: Masonry and Heritage Structures 3579
current state of the structure includes the stiffness or the interconnection of limited
survey of its pathology (survey of damage number of elements, without major alter-
and decay), as well as qualitative interpreta- ation of the bearing system.
tion of damage and decay. This step may (ii) Alleviation of irregularities: This strategy
require in situ and laboratory investigations is applicable, when the irregular distribution
(destructive and/or nondestructive, of masses, stiffnesses, and/or strengths leads
depending on the significance of the struc- to respective excessive requirements in
ture, its state, its future use, the characteristic some regions of the structure. In such
to be detected, etc.). a case, the alleviation of irregularities
(c) Diagnosis and assessment: The input from (by adding shear walls, by modifying the
steps (a) and (b) serves the needs of diagnosis arrangement of existing bearing walls, etc.)
(i.e., the identification of the causes of the may constitute the best strategy. It should be
observed pathological image) and assessment noted, however, that this strategy may be of
(i.e., the margins of safety of the structure in limited applicability in the case of cultural
its current state). At this step, the structure is heritage structures.
adequately modeled and subjected to normal (iii) Global enhancement of bearing capacity:
(self weight, live loads, etc.) and to accidental This strategy is adequate when (a) a large
actions (e.g., seismic action) with the purpose number of bearing elements are severely
to reproduce the main damages of the struc- damaged or (b) large inelastic deformations
ture. This is a step of major significance, as it are expected to be imposed to the structure.
allows for identification of the causes of dam- In such a case, a global enhancement of the
age and of the regions of the structure that are bearing capacity of the structure is advisable
vulnerable, for checking the adequacy of (strengthening of existing elements and/or
modeling, and for assessing the margins of addition of new bearing elements). Needless
safety of the existing structure. to say, the resulting enhancement of the
global stiffness of the structure has to be
The output of steps (a) to (c) allows the engi- taken into account.
neer to select the strategy of interventions, to (iv) Global enhancement of stiffness: This
choose among the available intervention tech- strategy may be selected when the calcula-
niques those techniques that serve better the pur- tions show that the expected deformations
pose of repair or strengthening of the structure, are larger than those the (brittle) masonry
and to proceed with dimensioning of the elements can sustain. If this is the case,
interventions. global enhancement of stiffness may be
Depending on the results of the assessment of achieved by strengthening of the bearing
the existing structure and taking into account the elements and/or by improving the box S
actions that are expected to act on the structure in action of the structure, e.g., by enhancing
the future (e.g., change of use, required seismic the diaphragm action of floors. Thus, the
upgrading of the structure, etc.), one of the fol- deformations to be imposed to the structure
lowing strategies of interventions may be are reduced to a level the masonry elements
applied: can sustain.
(v) Reduction of masses: This is a strategy that
(i) Local interventions to bearing elements: is desirable for structures to be subjected to
This strategy is applicable when the bearing seismic actions. In case of masonry struc-
system is in principle adequate, whereas tures, it may consist in using lighter mate-
local inadequacies are detected in a limited rials for roof and floor pavements, as well as
number of bearing elements. The interven- by reducing live loads (selection of appro-
tions consist in enhancing the bearing priate use of the structure). It should be
capacity or the deformation capacity or the noted, however, that this strategy does not
lead to spectacular results, as – in the case of depends on those connections. Subsequently,

masonry structures – their self weight con- (b) techniques for the enhancement of the dia-
stitutes the most significant part of the ver- phragm action of floors and roofs as well as for
tical loads. their connection with the vertical elements are
(vi) New bearing system: When the existing presented. The purpose of those techniques is, in
bearing system is inadequate for seismic most cases, not to enhance the bearing capacity of
actions, a new bearing system may be floors and roofs, but to ensure almost equal dis-
constructed inside the existing structure. placements of all vertical elements and to prevent
The new bearing system (e.g., made of diaphragms from moving independently of the
reinforced concrete (RC) or steel) is vertical elements during an earthquake, thus,
connected to the existing structure, thus causing severe damage to the walls. Finally,
contributing to the resistance to seismic (c) some indicative solutions frequently applied
actions. This is a strategy applied in some to cope with problems such as sliding soil or
cases to cultural heritage structures, where insufficient dimensions of the foundation are
extensive interventions to the existing bear- briefly presented.
ing system are invasive and, thus, undesir-
able. It should be noted that, the Repair and Strengthening Techniques for
dimensioning of the new bearing system Masonry Walls
will be governed by the compatibility of
deformations with those of the existing Deep Rejointing
one, rather than by the required additional Deep rejointing consists in the removal of the
bearing capacity. mortar from the joints to a depth at least equal
to the thickness of the joints and filling with
a mortar of improved properties. The technique
Repair and Strengthening Techniques should preferably be applied on both faces of
masonry. In order to reach a significant enhance-
The information included in this chapter regards ment of the mechanical properties of masonry,
the most commonly applied techniques for the and taking into account that the contribution of
repair and strengthening of existing masonry the mortar to the strength of masonry is rather
structures. It should be noted that some of the limited, a large part of the existing mortar has to
techniques presented here, although of frequent be replaced. Therefore, this technique may be
use, are not sufficiently documented through efficient only when applied to masonry of limited
experimental work or through monitoring of thickness (say 300 mm max.). This limits the
structures after intervention. The research on applicability of deep rejointing to brick or block
intervention techniques, as well as on the produc- masonry, excluding stone masonry walls
tion, testing, and use of new materials, is ongoing. (of minimum thickness around 400 mm). It has
The material included in this chapter is orga- to be noted that rejointing may be needed for
nized as follows: (a) Repair and strengthening reasons other than strengthening (e.g., to avoid
techniques for masonry as well as for the connec- water permeability of masonry or to prepare
tion between masonry elements are first masonry for the application of grouts). In those
presented. The aim of those techniques is to rein- cases, usually, there is no need for deep
state the pre-damage bearing capacity of individ- rejointing.
ual elements or to enhance their bearing capacity, The application of the new mortar has to be
so that they may sustain seismic actions in the made very carefully, so that the voids created due
future. Finally, the repair of damage in the con- to the removal of the existing mortar are fully
nection between masonry elements as well as its filled. Furthermore, the new mortar should not be
strengthening is examined, since the so-called excessively strong compared to the existing one.
box action of masonry structures strongly Otherwise, the compressive stresses on the
Strengthening a b
Techniques: Masonry
and Heritage Structures, layer of
Fig. 1 (a) Qualitative bricks
interpretation of spalling of
masonry units due to
new old mortar new
defective application of mortar mortar
deep rejointing, (b) A brick
wallette tested in void
compression after layer of
a defective application of bricks σ1 σ2>σ1
deep rejointing (Vintzileou
2001)
exterior (limited in thickness) part of masonry checked whether the frescoes are fixed on
may be excessive (Fig. 1a) and lead to spalling masonry. If not, their adequate fixing should
of the masonry units (Fig. 1b). As a result, the precede the application of grout. Further-
compressive strength of masonry may be more, during the application of grout, spe-
reduced, instead of being increased. cial care should be taken for immediate
In conclusion, (a) deep rejointing may be effi- cleaning of the face of masonry, in case of
cient in case of brick masonries, 300 mm thick at leaking of the grout.
maximum, and (b) provided that the technique is (b) Holes are drilled to masonry at distances
applied with due care and using adequate mate- along the crack. The distance of consecutive
rials, (c) in case of stone masonries, it is not holes should not exceed the thickness of
advisable to apply deep rejointing both because masonry. Transparent plastic tubes are
it is not expected to contribute to the improve- installed into the holes. Part of the length of
ment of the mechanical properties of masonry the tubes is protruding.
and because it is a quite expensive technique (c) The crack is sealed along its entire length,
(mainly in workmanship). Even if applied to using an adequate mortar.
stone masonries, deep rejointing is considered (d) The materials for the grout (see also section
as a repair technique, reinstating the initial “Generalities”) are mixed and the grout is
mechanical properties of masonry. introduced to the crack from the lower tube
until it leaks from the immediately higher up S
Grouting of Cracks placed tube and so on, until the crack is filled
This repair technique is applicable to cracks hav- over its entire length. The grout is applied
ing a width that does not exceed 10 mm with a pressure not exceeding 1 bar.
approximately: Depending on the state of the masonry, in
order to avoid excessive hydrostatic pressure
(a) Preparation of masonry: When masonry is that might cause further damage to masonry,
plastered, plaster is removed in the vicinity the crack length to be grouted should be lim-
of the crack. In cases where plaster cannot be ited within 1 day. For damaged, low strength
removed (e.g., when masonry is covered masonries, this length should be limited to
with paintings or frescoes or mosaics or 1 m per day.
other decorative elements), special care has (e) After hardening of the grout, the protruding
to be taken for the protection of the decora- parts of the tubes are cut and masonry is
tion of masonry. For example, it should be plastered, if this is the case.
Strengthening Techniques: Masonry and Heritage Structures, Fig. 2 Stitching of wide cracks
Strengthening
Techniques: Masonry
and Heritage Structures,
Fig. 3 Repair of wide
cracks by means of
reinforced zones
(indicative reinforcement)
Stitching of Cracks This technique cannot be applied to masonry

In case of almost vertical wide cracks (width walls that are cracked due to shear (diagonal or
exceeding 10 mm), grouting alone cannot lead bi-diagonal cracks), as the removal of cracked
to reinstatement of the initial strength of masonry units along the cracks is not feasible.
masonry. In such a case, stitching of the cracks, In those cases, the wall should be either
prior to grouting, is needed. demolished (and reconstructed) or it should be
Stitching consists in the removal of cracked carefully grouted, provided that its construction
masonry units along the crack and replacement type and its state allow grouting (see section
with new sound masonry units laid in mortar of “Grouting”).
adequate composition (Fig. 2). If necessary, some An alternative, applicable exclusively to build-
of the bed joints along the crack may be locally ings without any heritage value, would be that of
reinforced. To this purpose, steel plates are vertical and horizontal reinforced zones (Fig. 3).
inserted in the mortar joints (Fig. 2). The plates Grooves are created in the wall, both horizontally
are smooth and, hence, of limited bond with the and vertically on both sides of masonry. Steel
mortar. It is therefore advisable to artificially reinforcement is placed and a mortar of adequate
roughen them. composition or concrete is poured.
Stitching of Vertical Cracks Between It should be noted that (a) the separation
Perpendicular Walls between the walls is not reversible and (b) the
The same technique is applicable to large cracks in steel plates inserted into mortar joints are passive
the connection of walls. In that case, stones or (i.e., they are mobilized in case further relative
bricks are removed and replaced by intact ones displacement of the walls takes place). Therefore,
positioned diagonally (Fig. 4), to connect the two the cracks – after stitching – have to be grouted.
walls.
Alternatively, steel plates can be introduced to Grouting
mortar joints (Fig. 5), provided that the thickness
thereof is large enough for accommodating the Generalities Grouting of masonry (termed also
plates. as “homogenization” of masonry) consists in fill-
ing all voids and cracks within the mass of
masonry using a hydraulic mixture. Organic
binders (mainly epoxy resin) are not in use for
masonry; the exothermal reaction of their hard-
ening, combined with the (quite often) large
voids within the mass of masonry may lead to
disintegration of the epoxy resin-based grout
before the application of any actions to the build-
ing. Furthermore, due to the dust present in the
mass of masonry, the bond between the grout and
the in situ materials may be defective. Finally,
due to durability considerations, organic binders
are excluded. It should also be noted that hydrau-
lic grouts, even of low to medium mechanical
properties can ensure significant enhancement
of the mechanical properties of masonry (e.g.,
Binda et al. 1994; Valluzzi 2004; Vintzileou and
Miltiadou-Fezans 2008), thanks to their high
bond with the in situ materials.
Strengthening Techniques: Masonry and Heritage
A special mention needs to be made here of
Structures, Fig. 4 Stitching of vertical cracks between
perpendicular walls. Stones 1 and 2 are substituted by cement-based grouts (with a cement content
stone 3 exceeding 50 % per weight). Although they
Strengthening S
Techniques: Masonry
Fig. 5 Steel plates
accommodated in
horizontal mortar joints for
stitching cracks
Strengthening Techniques: Masonry and Heritage Structures, Fig. 6 Grid of holes in a masonry wall to be
grouted
constitute the first application of grouts to mechanical properties of the grouted masonry
masonry, the experimental work carried out in (i.e., strength and deformability characteristics),
several countries (Vintzileou and Tassios 1995; depending on the actual state of masonry, the
Toumbakari 2002) has proven that cement-based actions to be imposed, the overall scheme of
grouts are not more efficient than ternary grout interventions, etc.
(cement content of approximately 30 % per Therefore, the design of a grout requires suf-
weight, lime (and additives)) or hydraulic lime- ficient knowledge on the construction type of
based grouts. Furthermore, several cases of dura- masonry, on its state, on the in situ materials,
bility problems due to the use of cement are and on the aim of the intervention, all these com-
reported (e.g., efflorescence). Thus, in the recent bined with knowledge on material science.
years, the use of cement-based grouts is very
limited, if not fully abandoned. Preparation of Masonry, Mixing of Grout, and
Application The preparation of masonry for
The Design of the Grout Hydraulic grouts grouting is similar to that described in section
injected into masonry have to comply with “Grouting of Cracks.” Holes are drilled to
a set of requirements, namely, rheological masonry at distances, forming a grid. In the gen-
(injectability, i.e., penetrability into fine cracks eral case of application of grouts from both faces
and voids, according to the design, and sufficient of masonry elements, holes are drilled on both
fluidity for the grout to be diffused into masonry, faces of masonry. As shown in Fig. 6, the holes on
as well as stability), physical (low hydration heat, the interior face of masonry are drilled halfway
limited shrinkage, adequate hardening time, and between consecutive holes drilled on the exterior
hygroscopic properties), chemical (e.g., resis- face of masonry. Thus, the resulting grid of holes
tance to expansion and chemical stability of the is dense and it ensures better filling of the internal
products of chemical reactions), and mechanical voids and cracks. The holes should be drilled
requirements that are related to the desirable deep enough (Fig. 7) to reach the core of
tube (Fig. 11), equipped with a device that pro-

vides a mechanical stirring of the grout during
application (to avoid segregation).
When the preparation of masonry is com-
pleted (Fig. 12) and the grout mix is ready, the
grout is applied with low pressure from the bot-
tom to the top of masonry. The consumed quan-
tity of grout is recorded. This is a very significant
piece of information; it is taken into account
when the enhanced mechanical properties of
masonry are estimated.
Masonry is cured (kept wet) during the entire
procedure that may last for several days,
depending on the size and state of the injected
element.
Reinforced Plaster
This is a technique applicable to masonry of
limited thickness (not exceeding 0.40 m). It is
applied after repair of masonry (by grouting or
by stitching of cracks, by rejointing masonry,
etc.). Reinforced plaster should preferably be
Strengthening Techniques: Masonry and Heritage applied on both sides of masonry. Actually, it is
Structures, Fig. 7 Drilling holes in masonry
beneficial especially against out-of-plane bend-
ing, and, therefore, either the interior plaster or
masonry, thus ensuring that the grout will fill all the exterior one is mobilized (in tension)
the internal voids and cracks. Transparent plastic depending on the direction of the seismic action.
tubes are installed into the holes (Fig. 8) with an The plaster is removed and masonry is cleaned
inclination toward the bottom of the element from any loose material. The (metallic or nonme-
(to facilitate the flow of the grout). The tallic) grid of reinforcement is placed on the faces
(horizontal and vertical) distance between con- of masonry and fixed on it (using anchors).
secutive tubes may vary between 0.5 m and In order to achieve better bond of the reinforced
1.0 m. However, it should in no case exceed the plaster, it is advisable to remove the mortar from
thickness of the elements to be injected. the joints as deep as possible and fill them during
The plastic tubes are numbered and reported the application of the plaster. In case of metallic S
on a drawing (Fig. 8). This is of major impor- reinforcement (steel grid), the plaster is usually
tance, because it allows for full control of the a cement-based one. In case of nonmetallic rein-
application procedure (tubes from which forcement (e.g., fiber-reinforced wires or bars),
the grout is introduced and those from which the the plaster can be a hydraulic lime + pozzolan
grout overflows are marked during the injection). mortar.
All cracks are sealed (Fig. 9). Recently, fiber-reinforced mortars are also
The materials to be used for the production of applied to masonry.
the grout should be available at the vicinity of the It should be noted that the use of reinforced
region to be injected, together with the mixer plasters may not be advisable in the case of cul-
(Fig. 10). The use of high turbulence mixer is tural heritage assets, where masonry is quite
recommended, to provide adequate mix of the thick, and the application of the reinforced plaster
materials. After mixing of the constituent mate- may be possible only on the interior face of
rials, the grout is transferred to a transparent masonry (therefore, in its least efficient position).
Strengthening Techniques: Masonry and Heritage Structures, Fig. 8 Plastic tubes installed into drilled holes.
Tubes are numbered and reported to drawings
Local Demolition and Reconstruction large cracks in other than the collapsed regions
In case of local collapse of walls or in case of of the building, etc.).
partial collapse of the exterior leaf of a double- or
three-leaf masonry or in case of significant Strengthening of Walls Using FRP Rods,
out-of-plane displacement of a wall (Fig. 13), Laminates, or Sheets
the only solution is the reconstruction of the part The use of FRPs in strengthening (bearing and
that has collapsed or undergone significant infill) masonry is quite extensive in the recent
deformations. years. FRPs with various types of fibers and in
Needless to say, reconstruction is usually various forms (rods, laminates, or sheets) are
combined with repair or strengthening techniques used for strengthening masonry walls (Fig. 14)
described in the previous paragraphs (e.g., either against out-of-plane (bending) or in-plane
grouting of three-leaf masonry, stitching of (shear) actions.
The use of FRP materials is supported by ElGawady 2005). The available experimental
extensive experimental research (collected, e.g., data prove that the technique is quite efficient in
in Morbin 2001; Rashadul 2008), whereas enhancing the bearing capacity of masonry, as
models for the dimensioning of the intervention well as its ductility, provided that a good bond
are also proposed (e.g., Faella et al. 2004; with the substrate (masonry) is ensured.
The technology of FRP applications is not
presented in detail in this text for several reasons,
namely, (a) most of the tests available in the
Literature refer to monotonic loading of masonry
specimens; in a limited number of cases, also
repeated loading (without change of sign) or
cyclic loading (e.g., Grillo 2003, on concrete
masonry) was applied to the specimens. On the
other hand, (b) in laboratory tests, masonry walls
are subjected either to out-of-plane bending or to
in-plane shear. However, during an earthquake,
masonry walls are subjected to cyclic actions, as
well as to simultaneous in-plane and out-of-plane
actions. This is a very adverse condition which is
Strengthening Techniques: Masonry and Heritage not sufficiently covered by the international Lit-
Structures, Fig. 9 Sealing of cracks erature. The problem that may arise is
Strengthening Techniques: Masonry and Heritage Structures, Fig. 10 Materials for preparation of the grout mix.
High turbulence mixer
a premature de-bonding of the FRPs from the needs to be very carefully examined is
substrate with subsequent loss of the strengthen- (d) whether, after the occurrence of damage in
ing of the walls. Furthermore, (c) fire protection FRP strengthened masonry, its repair or strength-
of the intervention is an issue. Another issue that ening is still possible. Actually, as proven by
tests, the failure of FRP strengthened
walls – even under monotonic actions – may be
associated with extensive cracking and disinte-
gration of masonry. Especially in the case of
historic masonry, for which the possibility to
re-intervene after a seismic event is
a prerequisite, this aspect may be governing.
Last but not least, (e) there are limited data
regarding strengthening of historic masonry
using FRP materials. It is questionable whether
poor-quality (double-leaf or three-leaf) stone
masonry, having very low elastic properties, can
be strengthened using high-strength and
low-deformability materials. In such a case,
prior strengthening of masonry using “tradi-
tional” techniques would be required, whereas
the issue of compatibility of the in situ and
added materials should be carefully examined.
For example, the application of FRP sheets, cov-
ering large portions of masonry elements, may
create problems of building physics (e.g., entrap-
ment of humidity in the mass of masonry walls),
thus leading to decay of the covered masonry.
Another example would be that of inserting FRP
Structures, Fig. 11 Device for mechanical mixing of rods in (horizontal and vertical) mortar joints.
the grout before application Even if the practical difficulties in finding
Strengthening
Techniques: Masonry
Fig. 12 Masonry walls
ready for the application of
grouts
Strengthening Techniques: Masonry and Heritage a three-leaf masonry wall, (c) significant out-of-plane
Structures, Fig. 13 (a) collapse of the corner of displacement of a masonry wall
a building, (b) partial collapse of the exterior leaf of
Strengthening
Techniques: Masonry
Fig. 14 Alternative
configurations of FRPs for
masonry strengthening
(schematic)
FRP rods FRP laminates FRP sheets
inserted in S
mortar joints
continuous bed joints in rubble stone masonries, bending, there is a danger of stress concentration
the fact that grooves should be created to insert in the reinforced portions of the masonry section.
vertical FRP rods, and the difficulties in anchor- The stress concentration may lead to spalling of
ing the rods at the top and at the base of a wall, as the masonry units and to subsequent adverse
well as into transverse walls, are neglected, the effects on the bearing capacity of the masonry
problem remains of the behavior of rather soft element (comp. with Fig. 1). Moreover, (f) in case
original mortar joints that are partially filled with of historic buildings, the application of FRPs may
a (by orders of magnitude) stronger material. In be not be acceptable, as it alters the appearance of
real buildings, where the wall will be subjected to (unplastered) masonry. Thus, although the appli-
simultaneous cyclic shear and cyclic out-of-plane cation of FRPs seems to be a technically
Strengthening Techniques: Masonry and Heritage Structures, Fig. 15 Typical damages due to out-of-plane
bending of walls
promising intervention, it is believed that this increase of their stiffness and to subsequent
technique should be applied with caution for reduction of their natural period. In this way, the
strengthening masonry buildings that are to be building “moves” to the left region of the
subjected to seismic actions. It is believed that response spectrum and, hence, to lower seismic
the (under investigation) use of inorganic bond- actions.
ing material (mortar) instead of an organic one
(resins) may be a sensible alternative. Enhancement of the Diaphragm Action of Floors
and Roofs
Enhancement of the Box Action of Masonry In existing masonry buildings, typically, timber
Buildings floors and roofs are simply resting on vertical
elements (walls or piers). Furthermore, due to
Introductory Remarks the flexibility of floors and roofs in their plane,
It is well known that masonry is a brittle material; masonry walls are quite vulnerable to out-of-
it reaches its maximum resistance at small defor- plane bending (Fig. 15). Under such actions, ver-
mation value, whereas an abrupt decrease of its tical cracks close to the connections between
resistance takes place for imposed deformations longitudinal and transverse walls, as well as
exceeding the deformation corresponding to the out-of-plane collapse of portions of walls, consti-
maximum resistance. The vulnerability of tute typical damages of masonry buildings.
existing masonry buildings to large deformations The enhancement of the diaphragm action of
imposed by an earthquake is due to various typ- floors and roofs leads to almost equal deformations
ical construction features, namely, the defective of all vertical elements, thus limiting the out-of-
connection of walls among them, the flexibility of plane vulnerability. Thanks to the overall increase
floors and roofs, the large openings (especially of the stiffness of the building, the imposed defor-
when located close to the corners of the building), mations are also significantly reduced.
the large distance between transverse walls, There are several techniques for the enhance-
etc. All measures taken with the purpose of alle- ment of the diaphragm action of floors and roofs.
viating those sources of vulnerability contribute The Engineer has to select those most appropriate
to an improved seismic behavior of masonry. for each specific case and document (through
Last but not least, all the techniques that improve analysis) the efficiency of the selected
the box action of masonry buildings lead to an techniques.
indicative connection between

longitudinal and transverse beam
existing pavement
existing floor beams new transverse beams

transverse
Strengthening Techniques: Masonry and Heritage wall
Structures, Fig. 16 Enhancement of the diaphragm
action of a floor, arrangement of transverse beams wall supporting
connected to the existing ones (schematic)
the floor
Some of the possible alternatives are briefly

presented herein:
Structures, Fig. 17 Arrangement of steel ties for the
(a) Providing transverse beams to the existing enhancement of the in-plane stiffness of a floor
floor: Typically, in existing structures, floors (schematic)
consist of beams spanning in one direction
(normally, the smaller of the two). The
in-plane flexibility of the floor can be reduced (c) Providing a second pavement: In the frequent
by arranging (timber or steel) beams perpen- case of a timber pavement of floors and roof,
dicular to the existing ones (Fig. 16), thus significant enhancement of the in-plane stiff-
forming a grid. It should be noted that the ness can be reached through a second pave-
cross-sectional dimensions of the transverse ment. This second pavement can be located
beams, their spacing, and the dimensioning of either on top of the existing one or at the
their connections to the existing members are bottom of the beams (to avoid changes in
to be determined for each specific case. Sim- the height of the story). The second pavement
ilarly, in case of a trussed roof, beams per- should be either perpendicular to the existing
pendicular to the existing trusses can be one or, preferably, arranged at an angle of 45
arranged. (Fig. 18). In this case too, the Designer should
(b) Providing a system of bi-diagonal steel ties: check the efficiency of the solution through
Significant enhancement of the in-plane stiff- modeling and analysis of the structure.
ness of a floor (or roof) can be achieved by (d) Alternatively, instead of providing a second
arranging steel ties along the two diagonals of pavement, the in-plane stiffness of the floor
the floor (Fig. 17). A system of multiple ties may be enhanced by arranging bi-diagonal S
may be needed in the case of rectangular FRP laminates (Fig. 19). This alternative
floors, in which one in-plan dimension is solution should be applied with caution: The
significantly larger than the other. As in the efficiency of the FRP plates depends on their
previous case, the ties (adequately connected bond with the existing pavement. The simul-
to the existing floor beams or to the existing taneous flexural (due to vertical loads) and
trusses of the roof) are to be adequately in-plane (due to seismic actions) deforma-
dimensioned. It should be noted that, typi- tions of the floor may lead to detachment of
cally, ties should be dimensioned to develop FRP laminates from the substrate. In such
a tensile stress not exceeding 50 % of their a case, the efficiency of the intervention
yield strength, in order to avoid excessive may be significantly reduced.
strains of the steel elements and, hence, (e) Replacement of the existing floors and roof
avoid significant in-plane deformations of by reinforced concrete slabs: This alternative,
the floor. quite popular in previous decades, is
Strengthening Techniques: Masonry and Heritage Structures, Fig. 18 Existing timber pavement and additional
timber pavement at an angle of 45
be ensured. To this purpose, one of the following

two techniques may be applied: (i) Grooves are
made in masonry walls at distances (Fig. 20),
steel reinforcement is positioned, and the groves
are filled with concrete, during the construction
of the slab. The spacing, the dimensions, and the
reinforcement of the grooves are determined so
that they are able to transfer the shear (due to the
design seismic action) from the slab to the walls.
The respective resistance of masonry should also
be checked, (ii) steel beams are positioned along
the masonry walls (Fig. 20), and they are fixed
Strengthening Techniques: Masonry and Heritage onto masonry. The RC slab is supported by the
Structures, Fig. 19 FRP laminates glued on the existing
steel beams. In this case too, the connection of the
pavement (schematic)
steel beams to masonry walls should be ade-
quately dimensioned.
nowadays of rather limited application. Actu- Independently of the alternative selected for
ally, it is rather invasive in case of cultural the enhancement of the in-plane stiffness of floors
heritage buildings and, thus, not accepted by and roofs, there are two prerequisites for the
the competent authorities. Furthermore, efficiency of the solution, namely, (i) adequate
recent research (Zaopo 2011; Mazzon connection of the diaphragm to masonry walls
et al. 2009; Magenes et al. 2012; Mouzakis should be ensured and (ii) the supporting
et al. 2012) on floors stiffened using several masonry walls should be adequately repaired or
alternative techniques, as well as shaking strengthened. Actually, a very stiff diaphragm not
table tests on model buildings, has proven connected to masonry walls may be displaced as
that solutions (a) to (d) provide sufficient a rigid body during the earthquake, causing dam-
in-plane stiffness to the floor, and, hence, age to the masonry walls. If the stiff diaphragm is
the replacement of the existing floors and connected to masonry walls made of
roofs with RC slabs is not inevitable. unstrengthened poor-quality masonry, cata-
strophic damage may occur (Fig. 21).
When this alternative is selected, adequate Figure 22 shows the connection between floor
connection of the RC slab with the walls should and walls provided to a building model tested on
Strengthening
Masonry wall
Techniques: Masonry
Fig. 20 Alternative RC slab
connections of RC slabs to Section
masonry walls (schematic)
Plan
Masonry wall
RC slab
Steel beam
Masonry wall
Anchors
RC slab
Strengthening Techniques: Masonry and Heritage Structures, Fig. 21 Catastrophic effects of RC slabs
supported by poor-quality masonry (Courtesy of Prof. C. Modena)
S
the earthquake simulator (Mouzakis et al. 2012). and roof, the weight of pavements, as well as the
The diaphragm action of the floor was enhanced live loads of the building) are concentrated at
as shown in Fig. 18, and the walls (made of three- intervals. In case of poor-quality masonry, this
leaf masonry) were grouted. The connection was may lead to local damage. Furthermore, during
proven to be efficient, and no damage was caused an earthquake, the floors are moving rather inde-
to masonry walls even for an input acceleration pendently from the vertical masonry elements,
exceeding 1g. thus, causing damage to the walls at their top
(Fig. 23). Finally, the out-of-plane deformation
Improving the Connection Between Floors/Roofs of the walls is not prevented, due to loose con-
and Vertical Elements nection with the horizontal elements. For all
Frequently, in existing masonry buildings, roof those reasons, it is desirable to provide better
and floors are simply resting on masonry walls. connection between horizontal and vertical
Thus, the vertical loads (the self weight of floors elements.
Strengthening
Techniques: Masonry
Fig. 22 Detail of the
connection between
diaphragm and masonry
walls. The building model
after strengthening
Strengthening
Techniques: Masonry
Fig. 23 Damaged
masonry walls due to the
movement of simply
supported roof
The improvement of the connection between connection of the (timber or steel) beams of
horizontal and vertical elements can be achieved the floor with the collector beam. Although
as follows: the beam shown in Fig. 24 was positioned
during construction, it is feasible to provide
(1) Arrangement of a timber or a steel beam in a collector beam during interventions as well,
the perimeter of floors and roof (Fig. 24) and by creating a recess in masonry walls. Floor
Strengthening
Techniques: Masonry
Fig. 24 The timber beams
of the floor are connected to
a perimeter collector beam
beams or roof trusses have to be connected to previous paragraph (1), where all the application
the collector beam, which should be is dry and (c) durability problems may arise due
connected to the walls (Fig. 22). to the continuous contact of timber elements and
(2) When the roof is replaced or temporarily concrete.
removed during interventions, a reinforced
concrete tie beam can be constructed at Improving the Connection Between Walls
the top of the upper floor walls. Usually, the The significance of the adequate connection
width of the RC tie beam is equal to the between walls was known to the constructors of
thickness of the walls. However, in case masonry buildings. This is proven by the use of
there are decorative elements on the facade better quality, large-size cut stones in the corners
(to be restored after interventions) or the of the buildings, even in the case of rather poor
accommodation of insulation is needed, the general construction quality (Fig. 25).
width of the RC tie beam can be adequately Nevertheless, the insufficiently stiff in their
reduced. The construction of the RC tie beam plane floors and roofs that are simply supported
is also possible after temporary shoring of the on the walls do not prevent damage due mainly
roof, if the roof is not to be replaced. The to out-of-plane bending of walls and piers
construction of the tie beam with the roof in (Fig. 16). S
its place is not encouraged, although it is It should be noted that, if enhancement of the
technically feasible. diaphragm action of floors and roofs and ade-
quate connection thereof with the walls is pro-
The tie beam may be connected to masonry vided, further improvement of the connection
using anchors. The diameter, the length, and the between walls may not be needed. Actually, in
spacing of the anchors should be adequately such a case, the walls do behave as vertical slabs
designed. It should be noted that the construction fixed along their perimeter, and their vulnerabil-
of RC tie beams in intermediate floors is not ity to out-of-plane actions is significantly
recommended because (a) it is a quite invasive reduced. Nevertheless, if further improvement
intervention and (b) the RC tie beam will inevi- of the connection between walls is needed, steel
tably be of limited width. Thus, its connection to ties can be used. The use of ties for the connection
the beams of the floor and to the walls should be between walls was quite common during the con-
ensured in a way similar to that described in the struction of the existing masonry buildings.
Strengthening Techniques: Masonry and Heritage Structures, Fig. 25 Elaborate connection between walls at the
corners of masonry buildings
Strengthening
Techniques: Masonry
Fig. 26 The metallic tie
(of limited length) was not
able to prevent the out-of-
plane collapse of the wall
In that case, the ties (of limited length) were position. The ties are dimensioned so that their
arranged within the thickness of one wall and tensile stress does not exceed 50 % of their yield
anchored on its transverse wall (Fig. 26). strength, in order to avoid excessive deforma-
Steel ties provided as an intervention to tions when they are mobilized. Special care
improve the connection between walls are not should be taken for their anchorage on masonry.
embedded in masonry. They are (preferably) The dimensions of the anchoring plates should
arranged at both sides of a wall or only along allow (a) for almost uniform distribution of the
the interior face of a wall. They may be arranged compressive stresses on masonry and (b) for suf-
in one single level or in more levels, provided that ficiently small compressive stresses on masonry,
the height of the story allows for more ties to be to avoid local crushing.
positioned. Ties may be provided along one or The use of FRP ties as an alternative to steel
both main axes of the building. ties is not encouraged. Although FRPs have
The ties are usually passive (i.e., a very high tensile strength (compared to normal
non-prestressed). Therefore, they are expected steel) and they may have very high modulus of
to be mobilized only in case differential move- elasticity (which could prevent the occurrence of
ment of the two connected walls takes place. In large deformations), they are characterized by
case wall A (Fig. 27) tends to separate from walls complete lack of ductility. Therefore, if deforma-
C and D, the ties should be able to develop tions larger than their deformation at failure
a tensile force sufficient to sustain wall A in its occur, their fracture will be imminent.
elements as well. This phenomenon is accentu-

B ated in the case of seismic events.
A typical example of the aforementioned
Total force to be taken by the four damage is presented in Fig. 30: The cupola itself
ties= (Self-weight of wall
“A”)x(seismic coefficient)
remained undamaged. However, all piers of the
C D drum failed (some of them in shear, others in
out-of-plane bending). Due to the out-of-plane
deformations of the piers, the supporting arches
have failed. The horizontal component of
the thrust of the arches has pushed the vertical
A
elements apart, resulting to out-of-plane dis-
placements of the walls exceeding 0.10 m
Strengthening Techniques: Masonry and Heritage (observe the deformed shape of the monument
Structures, Fig. 27 Arrangement of steel ties to improve
the connection between walls (schematic)
in Fig. 30).
It is, therefore, of major importance to take
measures for the reduction of the effects of thrust,
by applying the techniques described in the
Arches, Vaults, and Domes following.
In old masonry structures, arches (Fig. 28) were
constructed to bridge openings in buildings Ties and Struts
(Fig. 28 a, b, f ) and river bridges (Fig. 28g), to This is a typical efficient measure to reduce the
transfer loads from domes to masonry piers, and effect of thrust. The technique was applied in
so on, whereas vaults and domes (Fig. 28c, d, e, f ) masonry structures either during construction or
are frequently used to bridge large spans. In most as a remedial measure during previous interven-
of cases, stone or brick curved elements can tions (Fig. 31). The technique consists in provid-
safely bear the vertical loads for which they ing at the base of the curved element horizontal
were conceived and constructed. Damage may ties that can sustain the horizontal component of
be due to excessive vertical loading resulting, the thrust. The ties (made of steel or stainless
for example, from inadequate use (Fig. 29) or steel or titanium, depending on the case) are
due to the lack of maintenance and repair anchored to transverse elements. For the maxi-
(Fig. 28e). mum acceptable stress in the ties, as well as for
In cases of damage due to excessive loads or to their anchorage to masonry, the rules presented in
decay, simple repair is sufficient. Repair may section “Improving the Connection Between
include local reconstruction, grouting, or Walls” are valid.
stitching of cracks, as described in section When the structure is subjected to seismic S
“Repair and Strengthening Techniques for action, there is a cyclic (inwards/outwards)
Masonry Walls.” movement of the vertical supporting elements.
It should be noted, however, that most of the Therefore, in addition to the ties, elements func-
damages of curved elements are due to the effects tioning in compression (struts) should also be
of the thrust. It is well known that the loads are provided. In old structures, quite frequently, tim-
transferred from the curved elements to their ber elements were used, able to play the double
supports following a thrust line; the thrust line is role of strut and tie (due to their sufficient resis-
oblique when it reaches the supports. Therefore, tance in buckling). This solution can be applied in
there is a horizontal component of the thrust that modern interventions as well. Alternatively, the
causes horizontal displacements to the supporting solution shown in Fig. 32 (adopted for interven-
elements. In the (quite frequent) case of flexible tions to Dafni Monastery, Miltiadou et al. 2012)
supports, outwards moving vertical elements may can be chosen: A timber compression element is
be damaged and induce damage to the curved provided. A hole is drilled at the core of the
Strengthening Techniques: Masonry and Heritage Structures, Fig. 28 Arches, vaults, and domes
timber element and a steel tie is accommodated. drum (Fig. 33). The ties (made either of steel or
The tie is anchored on masonry. FRP) are designed to sustain the horizontal com-
ponent of the thrust. It has to be noted, however,
Steel or FRP Ties in the Perimeter of the Curved that they do not lead to any reduction in the
Elements flexibility of the system. Therefore, their applica-
This is a technique frequently applied at the tion is useful but, quite often, not sufficient for the
base of cupolas, as well as at the base of their protection of the curved elements.
Strengthening Techniques: Masonry and Heritage Structures, Fig. 29 Partial collapse of the bridge after the
passage of a heavy track
Strengthening Techniques: Masonry and Heritage Structures, Fig. 30 The main church of Dafni Monastery
(Greece) damaged due to the 1999 earthquake
Buttresses supporting a vault or a dome are stiffened by an

This is a technique applied also in the past external buttress. Thus, the out-of-plane defor-
(Fig. 34): The flexible vertical elements mations of the supporting elements are
Strengthening Techniques: Masonry and Heritage Structures, Fig. 31 Arches with timber ties
Strengthening
Techniques: Masonry
Fig. 32 The timber strut/
steel tie solution was tested
in-laboratory and adopted
for application to the
monument (Dafni
Monastery)
Strengthening Techniques: Masonry and Heritage Structures, Fig. 33 Dafni Monastery: existing steel tie at the
base of the cupola, to be replaced by a stainless steel tie
Strengthening Techniques: Masonry and Heritage during the extensive interventions of the early twentieth
Structures, Fig. 34 (a) Ossios Lucas Monastery: exter- century, a pair of stone masonry buttresses was
nal stone masonry buttresses were added against the constructed against the northern wall of the church. This
southern facade. Due to architectural constraints, the but- technique was not repeated against the southern wall
tresses do not reach the critical region of the system of (where the courtyard is located)
arches that support the cupola. (b) Dafni Monastery:
Strengthening
Techniques: Masonry
Fig. 35 Dafni Monastery:
steel diaphragms to be
constructed on top of the
system of vaults; they will
be covered by a timber
pavement
S
significantly reduced, and damage is avoided or arranged either under the curved elements or on
significantly reduced. top of them, when the curved elements are visible
It is obvious that in many cases, the construc- from the interior of the building. In this way, the
tion of external buttresses, although a mechani- effects of the thrust are minimized and both the
cally efficient solution, cannot be adopted either curved elements and their supporting vertical ele-
because there is no space for them to be accom- ments are protected. This solution was adopted
modated or because the appearance of the struc- also in the case of Dafni Monastery (Fig. 35).
ture is significantly altered.
Modification of the Structural System
Enhancing the Diaphragm Action of Vaults and If the existing bearing system proves to be inade-
Domes quate to resist seismic actions or if the new use of
This is an alternative solution applicable when the structure requires higher performance (e.g.,
diaphragms (metallic or wooden ones) can be a residential building is given the use of a cultural
removed new
stone stones
protection against stirrups

sliding soil
Increase of the dimensions of the footing
Strengthening Techniques: Masonry and Heritage Structures, Fig. 36 Interventions to foundation
Strengthening
Techniques: Masonry
Fig. 37 Micro-piles
cement grout
PLAN
SECTION
center), there may be a need for drastic interven- original one (due to pounding). Therefore, in the
tions, such as construction of additional bearing design of the new system, deformations (rather
elements, closing of openings with masonry, than forces) may be the governing parameter.
etc. In such cases, the construction of a new bear-
ing system in the interior of the existing structure Strengthening of the Foundation
may be envisaged. When this alternative is Interventions to the foundation are needed when
selected, usually the new bearing system is made the mass of the structure is significantly increased
of steel. It should be noted, however, that for the due to the interventions or due to the change of use
new added system to assist the existing one in (higher live loads or heavier pavements, etc.).
sustaining seismic actions, the former has to be There are also cases of interventions needed
connected with the original structural members. when there are damages in the building due to
Moreover, the two systems have to be compatible sliding or differential settlements. The interven-
from the deformation point of view; otherwise, the tions to the foundation consist usually in increas-
new added system may cause damage to the ing the dimensions of foundation walls using
masonry or reinforced concrete (Fig. 36), ▶ Post-Earthquake Diagnosis of Partially

depending on the case. Instrumented Building Structures
When the properties of the foundation soil are ▶ Seismic Analysis of Masonry Buildings:
such that a deep foundation is needed, micro- Numerical Modeling
piles (Fig. 37) constitute a good solution. In the ▶ Seismic Strengthening Strategies for Heritage
case of buildings in regions of archaeological Structures
interest, it should be made sure – before the ▶ Seismic Vulnerability Assessment: Masonry
application of micro-piles – that there are no Structures
ruins of older structures under the foundation of
the building under examination. References
Binda L, Fontana A, Mirabella Roberti G (1994) Mechan-

Summary ical and stress distribution in multiple-leaf walls. In:
Proceedings, 10th international brick block masonry
conference, Calgary, Canada, pp 51–59
Techniques that are commonly applied either to ElGawady MA (2005) Shear model for URM walls
reinstate or to enhance the bearing capacity of retrofitted with FRP, Annual NZSEE Conference,
masonry, and masonry components and masonry Wairakei, New Zealand (electronic source)
Faella C, Martinelli E, Nigro E, Paciello S (2004) Tuff
buildings are briefly presented and commented
masonry walls strengthened with a new kind of C-FRP
upon. This text comprises techniques aiming at sheet: experimental tests and analysis. In: 13th WCEE,
(a) reinstating the mechanical properties of Vancouver (Paper no, 923)
masonry (such as rejointing, grouting and Grillo VE (2003) FRP/steel strengthening of unreinforced
concrete masonry piers. MSc thesis. University of
stitching of cracks, etc.), (b) enhancing the
Florida. 222 pp
mechanical properties of masonry (such as Magenes G, Penna A, Rota M, Galasco A, Senaldi I (2012)
grouting of masonry, reinforced plaster), Shaking table teat of a full scale stone masonry build-
(c) improving the overall behavior of a masonry ing with stiffened floor and roof diaphragms. In: 15th
WCEE, Lisbon
building (such as enhancement of the diaphragm
Mazzon N, Valluzzi MR, Aoki T, Garbin E, De Canio G,
action of floors and roof, improved connection Ranieri N, Modena C (2009) Shaking table tests on
between horizontal and vertical elements, etc.), two multi-leaf stone masonry buildings. In: 11th Cana-
and (d) improving the foundation conditions of dian masonry symposium, Toronto
Miltiadou A, Delinicolas N, Vintzileou E, Mouzakis CH,
the building. The techniques presented herein are,
Dourakopoulos J, Yannopoulos P (2012) Retrofitting
in principle, applicable to existing masonry of the main church of Dafni Monastery. Hellenic Min-
buildings independently of their architectural, istry of Culture (in Greek)
historical, and cultural values. There are, how- Morbin A (2001) Strengthening of masonry elements with
FRP composites. Diploma Dissertation. Dipartimento
ever, some cases – indicated in the text – where
di Ingegneria Edile, Università di Padova, Italy, 190 pp
intervention techniques have to be applied with Mouzakis C, Vintzileou E, Adami C-E, Karapitta L (2012) S
caution, when a heritage building is to be repaired Dynamic tests on three leaf stone masonry building
or strengthened. without timber ties before and after interventions.
SAHC, Wroclaw
Rashadul I (2008) Inventory of FRP strengthening
methods in masonry structures. Master thesis, Techni-
Cross-References cal University of Catalunya. 131 pp
Toumbakari E-E (2002) Lime-pozzolan-cement grouts
and their structural effects on composite masonry
▶ Ancient Monuments Under Seismic Actions:
walls. Doctor thesis, Katholieke Universiteit Leuven
Modeling and Analysis Valluzzi M-R (2004) Consolidamento di murature in
▶ Damage to Ancient Buildings from pietra. Iniezioni di calce idraulica naturale. Collana
Earthquakes Scientifica REFICERE, GruppoEditoriale Faenza Edi-
triceS.p.a., 128 pp. (in Italian)
▶ Masonry Box Behavior
Vintzileou E (2001) The effect of deep rejointing on the
▶ Masonry Components compressive strength of brick masonry. Masonry Int
▶ Masonry Modeling 15(1):8–12
3604 Structural Design Codes of Australia and New Zealand: Seismic Actions
Vintzileou E, Miltiadou-Fezans A (2008) Mechanical located on the boundary between the Indo-
properties of three-leaf stone masonry grouted with Australian and Pacific Plate and suffers from
ternary or hydraulic lime based grouts. Eng Struct
30(8):2265–2276 frequent strong earthquakes. In fact, New
Vintzileou E, Tassios TP (1995) Three-leaf stone masonry Zealand is one of the first countries to account
strengthened by injecting cement-grouts. J Struct Eng for seismic actions in a building standard; as early
Struct Division ASCE 121(5):848–856 as 1935, and following the 1931 Hawke’s Bay
Zaopo N (2011) Valutazione sperimentale dell’efficacia di
interventi di miglioramento sismico di solai in legno earthquake that claimed 256 lives, the NZS
riforzati nel piano, Diploma dissertation (supervisor 95 provided seismic loads to be considered in
MR Calluzzi), University of Padua, 212 pp. (in Italian) design (McRae et al. 2011). Despite constant
advances and amendments in the seismic stan-
dards over the years, the recent 22 February 2011
Christchurch earthquake of magnitude M = 6.3
Structural Design Codes of Australia resulted in 185 casualties, numerous injuries, and
and New Zealand: Seismic Actions extensive damages in the central business district
of Christchurch and its eastern suburbs.
George P. Kouretzis, Mark J. Masia and The purpose of this entry is to provide an
Clive Allen outline of the seismic design actions currently
Faculty of Engineering and Built Environment, considered in the abovementioned Standards
The University of Newcastle, Callaghan, NSW, and a short discussion on the rationale behind
Australia their adaptation, in the light of other widely
used modern seismic codes provisions, such as
the Eurocode 8 EN-1998-1 and the American
Synonyms ASCE/SEI 7-10. It covers the determination of
elastic and inelastic design response spectra to be
Australia; Design; Earthquake; New Zealand; used together with static and dynamic analysis
Seismic actions; Structural methods, as well as scaling of strong motion
recordings according to NZS 1170.5:2004 to per-
form time history analyses. Emphasis is put on
Introduction the determination of the input rather than on the
provisions about the performance of equivalent
The design of structures to resist earthquake static, modal, and time-domain analyses, which
effects in Australia and New Zealand follows can be found elsewhere, including the Standards
the pertinent Australian Standard AS themselves.
1170.4:2007 and New Zealand Standard NZS
1170.5:2004 provisions. Both Standards are
based on the common Australian/New Zealand Elastic Response Spectra
Standard AS/NZS 1170:2002 on structural design
actions; and although they share elements such as The elastic response spectra for horizontal load-
the site subsoil classification system, they incor- ing, C(T), can be described by the following
porate certain provisions such as the near-fault generic equation, compatible with both
factor in NZS 1170.5:2004 to account for the Standards:
different seismotectonic regime of the two coun-
tries. Indeed, Australia is an area of generally low CðT Þ ¼ Ch ðT Þ ðZ RÞ N ðT, DÞ (1)
seismicity, with the most catastrophic recent
event being the 1989 Newcastle earthquake of where Ch(T) is the spectral shape factor, Z R is
magnitude M = 5.6, which resulted in 13 casual- the product of the hazard factor Z times the factor
ties and significant damages in the wider New- R that depends on the annual probability of
castle area. On the other hand, New Zealand is exceedance of the design earthquake and the
Structural Design Codes of Australia and New Zealand: Seismic Actions 3605
Structural Design Codes of Australia and New static analyses, (b) AS 1170.4:2007/dynamic analyses, (c)
Zealand: Seismic Actions, Fig. 1 Spectral shape factor NZS 1170.5:2004/equivalent static analyses, (d) NZS
(5 % damping) plots for (a) AS 1170.4:2007/equivalent 1170.5:2004/dynamic analyses
design limit state (ultimate or serviceability) a provision would be of value only for special
under consideration, and N(T, D) is the near- structures such as tanks containing liquids, which
fault factor, applicable only to the NZS are not within the scope of the specific Australian
1170.5:2004. The parameters comprising the and New Zealand Standards.
elastic response spectra equation are presented As mentioned in the Introduction, Australian S
in the following paragraphs. and New Zealand Standards share a common site
classification scheme, which is preferably based
Spectral Shape Factor for Different Subsoil on the predominant site period (or low amplitude
Classes natural period) of the site Ts, estimated as:
The spectral shape factors Ch(T) to be used for
equivalent static analyses and for dynamic T s ¼ 4H s =V s (2)
(modal and time history) analyses are plotted sep-
arately in Fig. 1, while the relevant functions are where Hs is the depth to the seismic bedrock of the
provided in Table 1 for the different site subsoil site and Vs is the shear wave velocity of the over-
classes. These spectral shape factors correspond to lying soil layer. In the absence of sufficient data to
5 % structural damping, and a correction factor to estimate the predominant site period, the site can
convert them to other damping values is not probe classified to one of the five subsoil classes (A to
vided as, e.g., in EN-1998-1. However, such E) using (in the order of most preferred to least
Structural Design Codes of Australia and New Zealand: Seismic Actions, Table 1 Spectral shape factor (5 %
damping) for different site subsoil classes
Site Equivalent static method Dynamic analyses methods

subsoil Structure AS NZS Structure AS NZS
class period, T (sec) 1170.4:2007 1170.5:2004 period, T (sec) 1170.4:2007 1170.5:2004
A 0 < T 0.1 2.35 1.89 0 < T 0.1 0.8 + 15.5 T 1.0 + 1.35(T/
0.1)
0.1 < T < 0.3 0.704/ 0.1 < T < 0.3 0.704/ 2.35
0.3 T < 0.4 T 2.35 0.3 T 1.5 T 2.35 1.60(0.5/T)0.75
0.4 T 1.5 1.60(0.5/T)0.75
1.5 < T 3.0 1.056/T 2
1.05/T 1.5 < T 3.0 1.056/T2 1.05/T
3<T 3.15/T2 3<T 3.15/T2
B 0 < T 0.1 2.94 1.89 0 < T 0.1 1.0 + 19.4 T 1.0 +
1.35(T/0.1)
0.1 < T < 0.3 0.88/ 0.1 < T < 0.3 0.88/ 2.35
0.3 T < 0.4 T 2.94 T 2.94
0.4 T 1.5 1.60(0.5/T)0.75 0.3 T 1.5 1.60(0.5/T)0.75
1.5 < T 3.0 1.32/T 2
1.05/T 1.5 < T 3.0 1.32/T 2
1.05/T
3<T 3.15/T2 3<T 3.15/T2
C 0 < T 0.1 3.68 2.36 0 < T 0.1 1.3 + 23.8 T 1.33 + 1.60(T/
0.1)
0.1 < T < 0.3 1.25/ 0.1 < T < 0.3 1.25/ 2.93
0.3 T < 0.4 T 3.68 T 3.68
0.4 T 1.5 2.0(0.5/T)0.75 0.3 T 1.5 2.0(0.5/T)0.75
1.5 < T 3.0 1.874/T 2
1.32/T 1.5 < T 3.0 1.874/T 2
1.32/T
3<T 3.96/T2 3<T 3.96/T2
D 0 < T 0.1 3.68 3.0 0 < T 0.1 1.1 + 25.8 T 1.12 +
1.88(T/0.1)
0.1 < T < 0.3 1.98/ 0.1 < T < 0.3 1.98/ 3.0
0.3 T < 0.56 T 3.68 0.3 T < 0.56 T 3.68
0.56 T 1.5 2.4(0.75/T)0.75 0.56 T 1.5 2.4(0.75/T)0.75
1.5 < T 3.0 2.97/T 2
2.14/T 1.5 < T 3.0 2.97/T 2
2.14/T
3<T 6.42/T2 3<T 6.42/T2
E 0 < T 0.1 3.68 3.0 0 < T 0.1 1.1 + 25.8 T 1.12 +
1.88(T/0.1)
0.1 < T < 0.3 3.08/ 0.1 < T < 0.3 3.08/ 3.0
0.3 T < 1.0 T 3.68 0.3 T < 1.0 T 3.68
1.0 T 1.5 3.0/T0.75 1.0 T 1.5 3.0/T0.75
1.5 < T 3.0 4.62/T 2
3.32/T 1.5 < T 3.0 4.62/T 2
3.32/T
3<T 9.96/T2 3<T 9.96/T2
preferred): borehole log data together with in situ The five different subsoil classes are defined in
and laboratory tests, seismological methods the Standards as follows (AS 1170.4:2007, NZS
including strong motion recordings (Nakamura 1170.5:2004):
1989; Bouckovalas et al. 2002), qualitative bore-
hole log descriptions, and geological information Class A: Strong rock. Strong to extremely
for the site. For layered sites, the site period can be strong rock, with:
estimated via an averaging method described in (a) Unconfined compressive strength greater
AS 1170.4:2007 and NZS 1170.5:2004. than 50 MPa, and
Structural Design Codes of Australia and New Zealand: Seismic Actions, Table 2 Maximum depth limits for site
subsoil class C (After AS 1170.4:2007, NZS 1170.5:2004)
Representative undrained shear Representative SPT Maximum depth of
Soil type and description strength, Su (KPa) N-values soil (m)
Cohesive Very soft <12.5 – 0
soils Soft 12.5–25 – 20
Firm 25–50 – 25
Stiff 50–100 – 40
Very stiff to 100–200 – 60
hard
Cohesionless Very loose – <6 0
soils Loose dry – 6–10 40
Medium – 10–30 45
dense
Dense – 30–50 55
Very dense – >50 60
Gravels – >50 100
(b) An average shear wave velocity over the Class C: Shallow soil sites. Sites that are not
top 30 m Vs,30 > 1500 m/s, and classified as class A, class B, or class E:
(c) Not underlain by materials having (a) The predominant site period estimated as
a compressive strength less than 18 MPa above is Ts 0.6 s, or
or a shear wave velocity less than 600 m/s (b) Soil depth does not exceed the maximum
Class B: Rock. Rock, with: values listed in Table 2.
(a) Unconfined compressive strength between Class D: Deep or soft soil sites. Sites that are not
1 MPa and 50 MPa, and classified as class A, class B, or class E; and:
(b) An average shear wave velocity over the (a) The predominant site period estimated as
top 30 m Vs,30 > 360 m/s, and above is Ts > 0.6 s, or
(c) Not underlain by materials having a com- (b) Soil depth exceeds the maximum values
pressive strength less than 0.8 MPa or listed in Table 2, or
a shear wave velocity less than 300 m/s (c) Sites that are underlain by less than 10 m
A surface layer of no more than 3 m depth of of cohesive soil with undrained shear
highly weathered or completely weathered rock strength Su < 12.5 KPa or cohesionless
or soil material may be present in a class B site. soil soils with SPT values N < 6.
It should be mentioned here that, based on the Class E: Very soft soil sites:
average shear wave velocity over the top 30 m (a) More than 10 m of very soft cohesive soil S
criterion, other seismic codes such as EN-1998- with undrained shear strength
1 or ASCE/SEI 7-10 would classify a site with Su < 12.5 KPa, or
Vs,30 > 360 m/s to dense/stiff soil deposits, (b) More than 10 m of very loose cohesionless
rather than rock (class B according to soils with SPT values N < 6, or
EN-1998-1, which corresponds to 360 < (c) More than 10 m depth of soft-loose soils
Vs,30 < 800 m/s, and class C according to with shear wave velocity values
ASCE/SEI 7-10, which corresponds to 366 < Vs < 150 m/s, or
Vs,30 < 762 m/s). This suggests that class B in (d) More than 10 m combined depth of soils
AS 1170.4:2007 and NZS 1170.5:2004 is rather with properties described in (a), (b), and
broad, covering a range of subsoil conditions (c) above.
that may not exhibit similar behavior during an
earthquake rich in high- to medium-frequency A comparison between the spectral shape fac-
content (Bouckovalas and kouretzis 2001). tors of AS 1170.4:2007 and NZS 1170.5:2004 is
Structural Design Codes

of Australia and New
Zealand: Seismic
Actions,
Fig. 2 Comparison of AS
1170.4:2007 and NZS
1170.5:2004 dynamic
spectral shape factor plots
for rock (A, A and B) and
soft soil (E) types (5 %
damping)
attempted in Fig. 2, for the two extreme cases of Western Australia, where the hazard factor
rock (A and ) and soft soil (Ε) sites. The New ranges between Z = 0.14 and Z = 0.22, and the
Zealand Standard, which generally applies to Macquarie Island, located halfway between New
earthquakes with higher magnitudes (M 6.5), Zealand and Antarctica, where the hazard factor
features lower spectral factor values in the low is equal to Z = 0.60.
structural period range (T < 1 s) and higher In NZS 1170.5:2004, the hazard factor
spectral factor values in the high period range Z corresponds to 0.5 times the spectral accelera-
(T > 1 s). This is compatible with the fact that tion (5 % damping) for a structure period T =
soil nonlinearity effects alter the spectral content 0.5 s and soil subclass C, considering a design
of stronger earthquakes with magnitude M > 5.5, earthquake with a return period of 500 years
which in addition have a richer low-frequency (Fig. 1c, d). The minimum value across New
content attributed also to near-fault effects. Zealand is Z = 0.13, to ensure no collapse of
structures even in areas of low seismicity, i.e.,
Hazard Factor and Annual Probability of every structure is designed to survive the
Exceedance 84-percentile strong motion of a magnitude M =
The hazard factor Z of a particular location cor- 6.5 normal-faulting earthquake at source-to-site
responds to the peak ground acceleration (in g’s) distance of 20 km, a seismic scenario
for site class B (AS 1170.4:2007) or classes A and corresponding to the low-seismicity areas of
B (NZS 1170.5:2004), considering a design New Zealand. The hazard factor at the high-
earthquake with return period of 500 years, i.e., seismicity major cities (e.g., Wellington, Napier,
a 10 % probability of exceedance during Hastings) is of the order of Z = 0.40, with the
a 50-year design life. The hazard factors in the maximum value considered in the Standard being
Australian Standard have not been updated since Z = 0.60.
its previous version AS 1170.4:1993 and are The probability factor (AS 1170.4:2007) or
based on the work of Gaull et al. (1990), who return period factor (NZS 1170.5:2004) R, listed
used data of the Australian Geological Survey in Table 3, is used to scale the hazard factor to the
Organisation dated back to 1856. All areas in required annual probability of exceedance, for
Australia are assumed to be seismically active the design limit state under consideration. In AS
for design purposes, with the hazard factor gen- 1170.4:2007, the annual probability of exceed-
erally ranging between Z = 0.05 and Z = 0.13, ance is correlated with the design working life
with the exception of the Meckering region in and the importance level of the structure, with the
Structural Design Codes of Australia and New 1170.0:2002 (Table 3.3 of AS/NZS
Zealand: Seismic Actions, Table 3 Probability factor 1170.0:2002); however, in the New Zealand
(AS 1170.4:2007) or return period factor (NZS
1170.5:2004) Standard, it is correlated with the design
(serviceability or ultimate) limit state: The annual
Annual probability of exceedance R
probability of exceedance for the common ser-
1/2,500 1.8
viceability limit state SLS1 (requirement for no
1/2,000 1.7
repairs on structural and nonstructural compo-
1/1,500 1.5
1/1,000 1.3
nents) is 1/25 for all structures. The special ser-
1/800 1.25 viceability limit state SLS2 applies to structures
1/500 1.0 carrying critical post-disaster functions only
1/250 0.75 (importance level 4) and is associated with an
1/200 0.70 annual probability of exceedance of 1/500 for
1/100 0.50 design working life of 50 years. A special hazard
1/50 0.35 study is required to define the SLS2 actions for
1/25 0.25 structures of importance level 4 with design
1/20 0.20 working life of 100 years or more.
As far as the ultimate limit state is concerned,
the design annual probability of exceedance is the
latter provided in AS/NZS 1170.0:2002 same as in AS 1170.4:2007, for structures classi-
(Appendix F). The design annual probability of fied to importance levels 2 and 3. Note that the
exceedance applies to the ultimate limit state product Z∙R in NZS 1170.5:2004 need not exceed
only: According to AS 1170.4:2007, structures Z∙R = 0.70 for ultimate limit state analyses but
of importance levels 1–3 (i.e., all structures must be higher than Z∙R = 0.20 when an annual
except ones carrying critical post-disaster func- probability of exceedance 1/2,500 is considered.
tions and exceptional structures) de facto satisfy These Z∙R bounds correspond to the higher and
the serviceability limit state requirements when lower seismicity regions of New Zealand, respec-
designed in accordance to the Standard. A special tively: The upper bound matches the
study needs to be carried out for critical post- 84-percentile near-fault strong motion due to
disaster structures only (importance level 4), to a M = 8.1 event from the major Alpine Fault
ensure that they remain operational after that runs along the South Island, divided by
a seismic event equivalent to the ultimate limit a margin of safety equal to 1.5 likely to result
state design event for ordinary importance level from applying the code design provisions.
2 structures. The design annual probability of
exceedance for structures and buildings Near-Fault Factor
(importance level 2) with a working life of The near-fault factor N(T,D) is introduced in NZS S
50 years is 1/500 and corresponds to a 10 % prob- 1170.5:2004 to account for near-source effects on
ability of exceedance during the design life of the the strong motion during the activation of a strike-
structure. For major structures affecting crowds slip fault: a) forward directivity, resulting in high
and associated with “very great” economic, peak velocities and displacements, and b) polar-
social, and environmental consequences of fail- ization of the long-period motions in the near-
ure (importance level 3), the annual probability of source region, where medium- to long-period
exceedance is reduced to 1/1,000 for structures pulses tend to be stronger in the direction perpen-
with a design working life of 50 years and to dicular to the strike of the fault. Near-fault effects
1/2,500 for structures with a design working life in NZS 1170.5:2004 are considered for eleven
of 100 years. (11) major strike-slip faults classified as “class
The required annual probability of exceedance A” faults according to the Californian Category
for structures designed according to NZS A fault criteria (Petersen et al. 2000), capable of
1170.5:2004 is provided again in AS/NZS producing earthquakes with magnitude M = 7.0
Structural Design Codes of Australia and New However, a provision is included stating that
Zealand: Seismic Actions, Table 4 Maximum values when the dynamic analysis requires the consider-
of the near-fault factor Nmax(T) for different structure
periods ation of vertical earthquake forces, both upward
and downward motions need to be considered and
Linear interpolation
NZS 1170.5:2004 functions
the vertical spectral shape factor shall be taken
Structure Structure equal to
period, T (sec) Nmax(T) period, T (sec) Nmax(T)
1.5 1.0 1.5 1.0 Cv ðT Þ ¼ 0:5 Ch ðT v Þ ðAS 1170:4 : 2007Þ (4)
2 1.12 1.5 < T < 4 0.24 T
3 1.36 + 0.64 where Tv is the vertical period of vibration.
4 1.60 4T<5 0.12 T In NZS 1170.5:2007 on the other hand, when
+ 1.12 vertical design actions need to be considered in
5 1.72 5 1.72 the analyses (e.g., time history analyses), the
elastic response spectrum for vertical loading is
taken equal to 70 % of the corresponding hori-
or greater and having slip rates of 5 mm/year of zontal spectrum, as
greater. The near-fault factor N(T,D) must be
considered during the derivation of the medium- Cv ðT Þ ¼ 0:7 Ch ðT v Þ ðNZS 1170:5 : 2004Þ
to long-period band of the elastic response spec- (5)
tra for sites located within a distance D less than
20 km from the traces of the major strike-slip where Tv = 0 for the design of the structure as
faults listed in NZS 1170.5:2004 and applies to a whole, or equal to the vertical period of the
annual probability of exceedance less that 1/250: element under consideration, for the design of
parts and components. Note that the vertical com-
N ðT, DÞ ¼ N max ðT Þ for D 2 km (3:1) ponent of strong earthquake motion is usually
rich in high-frequency content; thus, peak ground
N ðT, DÞ ¼ 1 þ ðN max ðT Þ1Þ ð20DÞ=18 acceleration and spectral values in the near-
for 2km < D 20 km source region may exceed the horizontal values
(3:2) (NZS 1170.5 Supp 1:2004, Niazi and Bozorgnia
1992; Ambrasseys and Simpson 1996). In light of
N ðT, DÞ ¼ 1:0 for D > 20 km (3:3) this, NZS 1170.5 Supp 1:2004 recommends that
when the distance of the site from the fault trace is
where the factor Nmax(T) is provided as a function less than D = 10 km, the vertical design actions
of the structure period T in Table 4 and Fig. 3a. be considered equal to the horizontal design
Linear interpolation functions are provided for actions for structure period T 0.30 s. Strong
intermediate period values, as proposed in NZS motion recordings from recent earthquakes, such
1170.5:2004. In addition, a comparison of the as the Christchurch 2011 earthquake, provided
effect of near-fault correction (Eqs. 3.1, 3.2, and further evidence in support of the above, as will
3.3) on the spectral shape factors for dynamic be discussed later in this entry.
analyses is illustrated in Fig. 3b, for different
distances from the fault.
Structural Ductility and Structural
Vertical Design Actions Performance Factor-Inelastic Design
According to AS 1170.4:2007, vertical earth- Spectrum
quake actions generally need not to be considered
in design. For the design of mechanical and elec- Derivation of the (inelastic) design spectrum
trical components, the vertical earthquake forces requires the correction of the elastic response
are taken equal to 50 % of the horizontal forces. spectrum to account for the ability of the
Structural Design Codes of Australia and New near-fault effects for a rock site (soil subclass A and B)
Zealand: Seismic Actions, Fig. 3 (a) Variation of the at a distance D 2 km, D = 10 km, and D > 20 km from
near-fault factor Nmax(T) with the structure period, T, and a major active fault (5 % damping)
(b) comparison of spectral shape factors corrected for
particular structure to dissipate seismic energy effects or the contribution of nonstructural

via nonlinear response. Apart from the structural elements (e.g., in fill walls) which is not
ductility factor, m, which is related to the level of taken directly into account in typical analysis
inelastic demand that can be reliably sustained by models.
the structure, AS 1170.4:2007 and NZS (d) The energy dissipation of the structure is
1170.5:2004 introduce the structural perfor- typically higher than assumed, due to
mance factor Sp to reduce the elastic design seis- damping introduced from nonstructural ele-
mic actions. The structural performance factor is ments and soil-foundation interaction effects.
employed to quantify a number of effects that are
not explicitly taken into account in state-of- The ductility factor m for the ultimate limit
practice structural analysis procedures by simply state is calculated in accordance with the appro-
scaling the design loads. Such effects include priate material standard, where such data are pro-
(NZS 1170.5 Supp 1:2004): vided. Contrary to NZS 1170.5:2004, where
reference is made to material standards only and S
(a) Excitation effects: The estimated seismic a special study is required otherwise, maximum
loads correspond to a peak ground accelera- ductility factors for typical structural systems are
tion value, which may be reached only during provided directly in AS 1170.4:2007 (Table 5).
a single loading cycle and therefore is Alternatively, for a specific structure, m and Sp
unlikely to lead to significant damage can be determined via a nonlinear static pushover
(“effective” ground acceleration concept). analysis. Note also that AS 1170.4:2007 applies
(b) Individual structural elements typically fea- only to structures with ductility factor m 3, and
ture a higher capacity than modeled during when a higher ductility factor is considered, the
the analysis of the structure, due to higher design must be performed in compliance with
material strength, strain hardening, strain NZS 1170.5:2004 provisions (Table 5).
rate effects, etc. While NZS 1170.5:2004 does not refer explic-
(c) The total structural capacity is typically itly to ductility factor values for the ultimate limit
higher than predicted, due to redundancy state, for serviceability limit state analyses of
Structural Design Codes of Australia and New Zealand: Seismic Actions, Table 5 Ductility factor m and
structural performance factor Sp for different structural systems and specific structure types (After AS 1170.4:2007)
Structural system Description m Sp
Steel structures Special moment-resisting frames (fully ductile)a 4 0.67
Intermediate moment-resisting frames (moderately ductile) 3 0.67
Ordinary moment-resisting frames (limited ductile) 2 0.77
Moderately ductile concentrically braced frames 3 0.67
Limited ductile concentrically braced frames 2 0.77
Fully ductile concentrically braced framesa 4 0.67
Other steel structures not defined above 2 0.77
Concrete Special moment-resisting frames (fully ductile)a 4 0.67
structures Intermediate moment-resisting frames (moderately ductile) 3 0.67
Ordinary moment-resisting frames 2 0.77
Ductile coupled walls (fully ductile)a 4 0.67
Ductile partially coupled wallsa 4 0.67
Ductile shear walls 3 0.67
Limited ductile shear walls 2 0.77
Ordinary moment-resisting frames in a combination with limited ductile shear 2 0.77
walls
Other concrete structures not listed above 2 0.77
Timber Shear walls 3 0.67
structures Braced frames (with ductile connections) 2 0.77
Moment-resisting frames 2 0.77
Other wood- or gypsum-based seismic-force-resisting systems not listed above 2 0.77
Masonry Close-spaced-reinforced masonryb 2 0.77
structures Wide-spaced-reinforced masonryb 1.5 0.77
Unreinforced masonryb 1.25 0.77
Other masonry structures not complying with AS 3700:2001 1 0.77
Specific structure Tanks, vessels, or pressurized spheres on braced on unbraced legs 2 1.0
types Cast-in-place concrete silos and chimneys having walls continuous to the 3 1.0
foundation
Distributed mass cantilever structures, such as stacks, chimneys, silos, and skirt- 3 1.0
supported vertical vessels
Trussed towers (freestanding or guyed), guyed stacks, and chimneys 3 1.0
Inverted pendulum-type structures 2 1.0
Cooling towers 3 1.0
Bins and hoppers on braced or unbraced walls 3 1.0
Storage racking 3 1.0
Signs and billboards 3 1.0
Amusement structures and monuments 2 1.0
All other self-supporting structures not otherwise covered 3 1.0
The design of structures with m > 3 should be in accordance to NZS 1170.5:2004
a
b
Values from AS 3700:2001
common structures (SLS1), it is provisioned that Note that in NZS 1170.5:2004, the inelastic
1.0 m 1.25, and for critical post-disaster horizontal design spectra are not derived by
structures (SLS2), it is provisioned that directly multiplying the elastic spectra by (1/m).
1.0 m 2.0. Furthermore, the ductility factor Instead, to account for the transition between
for vertical actions is always considered to be equal displacement theory (which is valid for
m = 1.0. longer structure periods) and equal energy theory
(valid for shorter structure periods), a transition and in NZS 1170.5:2004 as

point is defined at T1 = 0.7 s for soil subclasses
A to D and at T1 = 1.0 s for soil subclass E, where
CðT ÞSp
T1 is the largest translation period of vibration Cd ðT Þ ¼ ðfor horizontal actionsÞ (9:1)
km
along the direction under consideration. So, the
factor to derive the inelastic response spectra is
estimated as: Cd ðT Þ ¼ Cv ðT ÞSp ðfor vertical actionsÞ (9:2)
For subsoil classes A, B, C, and D:
Actions for Dynamic Time History
km ¼ m for T 1 0:7 s (6:1) Analyses
In addition to actions applicable to common

km ¼ 1 þ ðm 1ÞT 1 =0:7 for T 1 < 0:7 s equivalent static and modal dynamic analyses,
(6:2) AS 1170.4:2007 and, mainly, NZS 1170.5:2004
include certain provisions for the determination
For subsoil class E: of the appropriate design actions when dynamic
time-domain analyses are to be performed. While
in AS 1170.4:2007 there is a vague requirement
km ¼ m for T 1 1:0 s or m < 1:5 (6:3)
that the response spectra of the actual accelera-
tion time histories used shall “approximate” the
km ¼ 1:5 þ ðm 1:5ÞT 1 for T 1 < 1:0s and m 1:5 design spectrum of Eq. 8, NZS 1170.5:2004 pre-
(6:4) scribes a more elaborate and explicit procedure
for the selection and scaling of ground motion
For the purpose of calculating km, the largest records.
translation period of vibration T1 is not taken Ground motion accelerographs to be used for
less than 0.4 s. time-domain analyses according to NZS
The structural performance factor Sp in AS 1170.5:2004 shall consist of both horizontal com-
1170.4:2007 is provided in Table 5. In NZS ponents of the recording; the vertical component
1170.5:2004, the structural performance factor may have to be considered too, for the analysis of
is taken equal to Sp = 0.7 for ultimate limit state structures sensitive to vertical strong ground
analyses, except for low-ductility structures with motion. The above imply that the procedure
1.0 < m < 2.0, where Sp is calculated as: refers to three-dimensional analyses, yet an adap-
tion to two-dimensional models is feasible, by
Sp ¼ 1:30:3m (7) considering only one horizontal component of
each ground motion record as, e.g., in ASCE/ S
SEI 7-10. A “family” of no less than three records
For ultimate limit state verification of structures
must be employed in each time history analysis.
against sliding or overturning, Sp is taken equal to
The selected acceleration records must be con-
Sp = 1.0. For serviceability limit state SLS1 and
sistent with the local seismotectonic regime
SLS2 analyses, Sp is taken equal to Sp = 0.7,
(expected magnitude, fault rupture mechanism,
except otherwise dictated by the relevant material
source-to-site distance, etc.). When the site is
standard.
located near a major fault, i.e.,
In accordance to the above, the inelastic
N(T, D) > 1 (Eqs. 3.1, 3.2, and 3.3), then one
design response spectrum in AS 1170.4:2007 is
set of the records must have a forward directivity
defined as
component. Simulated, artificial accelerographs
may be used too, when appropriate real acceler-
CðT ÞSp ation recordings are not available. Each record
Cd ðT Þ ¼ (8)
m shall be scaled by a “record scale factor” k1 to
Structural Design Codes of Australia and New for the city of Christchurch and soil subclass D, (b) scaled
Zealand: Seismic Actions, Fig. 4 (a) Comparison of elastic response spectra of the same strong motion record-
the 2011 Christchurch strong motion recordings elastic ings to match the NZS 1170.5:2004 design spectra for
response spectra (5 % damping, horizontal components) largest structure translation period T1 = 1 s and Sp = 1
with the design NZS 1170.5:2004 elastic design spectra
match the target design spectrum over the period importance level 2 in Christchurch with
range of interest and by a “family scale factor” k2 a design life of 50 years, Z R = 0.22 g and
to ensure that the energy content of at least one N(T,D) = 1, i.e., no near-fault effects, need to
record in the family exceeds that of the design be taken into account due to possible rupture of
spectrum over the period range of interest. The the Port Hills Fault that is believed to have caused
target design spectrum is defined as the 2011 earthquake. Assuming for simplicity
that Sp = 1, the target spectrum is plotted in
1 þ Sp Fig. 4.
SAtarget ¼ C ðT Þ (10)
2 Strong motion recordings during the Christ-
church 2011 earthquake exceeded by far the
where the spectral shape factor C(T) is calculated design peak ground acceleration at the ground
from Eq. 1 and Sp is the structural performance surface of class D soft soils implied by NZS
factor defined in the section “Structural Ductility 1170.5:2004 (0.246 g) and reached maximum
and Structural Performance Factor-Inelastic values of 1.67 g in the horizontal direction and
Design Spectrum.” 2.20 g in the vertical direction at the Heathcote
In order to present the procedure of scaling Valley Primary School Station (HVSC) (www.
ground motion records according to NZS geonet.org.nz). Here a family of three near-field
1170.5:2004, the horizontal components of three recordings on soft to very soft soil are used, with
typical strong ground motion recordings from the peak horizontal ground accelerations ranging
2011 Christchurch M = 6.3 earthquake were from 0.34 to 0.67 g (Table 6).
used. These records were scaled to match the The elastic response spectra of the horizontal
target ultimate limit state spectrum of NZS components of these three recordings are plotted
1170.5:2004 for a common structure in Fig. 4a, in comparison with the design response
(importance level 2) located in Christchurch, spectra of NZS 1170.5:2004. A band-pass filter in
founded on subsoil class D (soft soil). According the frequency ranges of 0.10–0.25 Hz and
to NZS 1170.5:2004, for a structure of 24.5–25.5 Hz was applied to the time histories
Structural Design Codes of Australia and New Zealand: Seismic Actions, Table 6 Characteristics of the typical
strong motion recordings from the Christchurch 22 February 2011 M = 6.3 earthquake (Source: www.geonet.org.nz)
Site Subsoil class acc. to NZS Epicentral PGA-horiz.1 PGA-horiz.2 PGA-vert.
code 1170.5:2004 distance (g) (g) (g)
CBGS D 7 0.553 0.452 0.360
PRPC E 6 0.669 0.595 1.88
CHHC D 6 0.345 0.364 0.601
used to derive the spectra. It is clear that the The best-fit scale factor k1 of each horizontal
spectral values of these typical records are ground motion component was determined
above the design spectra across practically the as the value minimizing the function
whole range of important structural periods, an log(k1SAcomponent/SAtarget) over the period range
indication of the severity of the Christchurch of interest, in a least mean square sense. In other
earthquake. words, the aim was to find via an interactive
Scaling of ground motion records to match the procedure the k1 value that minimizes the sum:
target design spectrum via the scale factor k1 was
performed while taking into account a structure
ð1
1:3T
2
period range that depends on the largest transla- k1 SAcomponent
log dT ¼ min (11)
tional period of the structure in the direction of SAtarget
interest, T1. The period range of interest is defined 0:4T 1
as 0.4 T1 T 1.3 T1, with the product 0.4 T1 not

taken less than 0.4 s. Notice that this band of The selected recordings must satisfy
significant periods is considerably narrower com- a “similarity” criterion 0.33 < k1 < 3.0. More-
pared to the corresponding one prescribed in over, in order to verify that each selected record
EN-1998-1, which is 0.2 T1 T 2.0 T1, or reasonably matches the target spectrum over the
even ASCE/SEI 7-10, where the corresponding period range of interest, the following inequality
range is 0.2 T1 T 1.5 T1. For the case at hand, quantifying the mean (over the period of interest)
it was assumed that T1 = 1 s, and the period range square difference of the scaled over the target
of interest is denoted by the yellow band in Fig. 4. spectral values must be satisfied:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u ð 1

u 1:3T
u 1 k1 SAcomponent 2
D1 ¼ t log dT logð1:5Þ (12)
ð1:3 0:4ÞT 1 SAtarget
0:4T 1 S
Although not a normative criterion, a stricter fit The record family scale factor, k2, ensures that
D1 log(1.3) is proposed for most cases, the principal component of at least one record
according to NZS 1170.5 Supp 1:2004. spectrum (after being scaled by k1) exceeds the
The k1 and D1 factors for the records at hand target spectrum over the period range of interest.
are listed in Table 7, where it is depicted that the It is estimated as the maximum value of the ratio
specific records are compatible with NZS SAtarget/max(SAprincipal) 1.0 within the period
1170.5:2004 requirements. The component of range of interest, where max(SAprincipal) is the
each record with the lower k1 value was nomi- maximum spectral value between all the princi-
nated as “principal” (Fig. 4, Table 7), and pal components of the family, at each period step
this value of k1 was used as the record scale considered for the derivation of the spectra.
factor. To confirm the selection of the principal and
Structural Design Codes of Australia and New

Zealand: Seismic Actions, Table 7 Scaling factors for
the considered typical strong motion records
Site
code Component k1 D1 k2
CBGS N89W 0.608 0.070 1.05
(principal)
S01W 0.789 0.105
PRPC W (principal) 0.672 0.081
S 0.785 0.097
CHHC N01W 0.683 0.064
S89W 0.598 0.058
(principal)
secondary components, the record family scale

factor must be within the range 1.0 < k2 1.3.
If k2 > 1.3, then either a different record must be
Structural Design Codes of Australia and New
selected or the selection of the principal/second- Zealand: Seismic Actions, Fig. 5 Comparison of the
ary components may be reversed, aiming to min- mean response spectrum (5 % damping) of the three prin-
imize the product k1k2. This may be the case cipal components listed in Table 7 with the NZS
1170.5:2004 target spectrum for soil subclass D and
when all three principal components in a family
Sp = 1. The period of interest band according to NZS
feature low spectral values within a particular 1170.5:2004 (0.4 T1 T 1.3 T1) and according to
period band, while one of the secondary compo- ASCE/SEI 7-10 (0.2 T1 T 1.5 T1) is also drawn
nents is rich in frequency content within the same
band; thus, although the scale factor k1 may be the above, if one embraced the more conservative
greater for that particular strong motion compo- EN-1998-1 and ASCE/SEI 7-10 provisions
nent, the product k1k2 may be lower. In the case at regarding the average response spectra from the
hand however, the family factor k2 is within the three principal components, the specific scaling
desirable range (Table 7), a fact that can be con- factors applied would not be acceptable, as their
firmed visually from Fig. 4. average spectrum is below 100 % (ASCE/SEI
The above scaling procedure must be followed 7-10) and marginally below 90 % (EN-1998-1)
for all directions of interest, corresponding to of the target spectrum (Fig. 5).
different translational periods of the structure, Finally, note that NZS 1170.5:2004 (as also
T1, and thus a different period range of interest. ASCE/SEI 7-10) bases the accelerograph scaling
Although more elaborate than the analogous pro- procedure on spectral values within the range of
cedure of EN-1998-1 or ASCE/SEI 7-10 for the interest only and does not impose restrictions on
time history representation of the seismic action, the zero period spectral response acceleration
where it is simply required that no value of the values of the strong motion, as, e.g., EN-1998-1.
mean elastic spectrum of all time histories used is
less than 90 % (EN-1998-1) or 100 % (ASCE/SEI
7-10) of the corresponding value of the elastic Summary
response spectrum, the procedure prescribed in
NZS 1170.5:2004 results in a more rational scal- Seismic design action definitions in the Australian
ing of strong motion records, and in a closer AS 1170.4:2007 and New Zealand NZS
match to the design spectrum, as it does not 1170.5:2004 Standards were presented in
impose restrictions related only to lower-than- a concise way, focusing on the determination of
target spectral acceleration values found in the elastic and inelastic design response spectra to
EN-1998-1 and ASCE/SEI 7-10. As a result of be used together with static and dynamic methods
Structural Optimization Under Random Dynamic Seismic Excitation 3617
of analysis. A short discussion against provisions September 7, 1999 Athens, Greece Earthquake. Nat
of other modern seismic codes, such as EN-1998-1 Hazards 27:105–131
Bouckovalas G, Kouretzis GP (2001) Stiff soil amplifica-
and ASCE/SEI 7-10, was attempted to point out tion effects in the 7 september 1999 Athens (Greece)
certain key differences in particular clauses. earthquake. Soil Dyn Earthq Eng 21(8): 671–687
Although Australian and New Zealand Standards Department of Building and Housing (2011) Compliance
embrace most of the recent developments on earth- document for New Zealand building code, clause B1,
structures, Amendment 11, August. New Zealand
quake engineering, such as accounting for near- Government
source effects on strong motion (but not topogra- EN-1998-1. Eurocode 8: design of structures for earth-
phy effects as, e.g., EN-1998-1), any seismic code quake resistance – Part 1: general rules, seismic
is not a “static” document, but rather a “dynamic” actions and rules for buildings. European Committee
for Standardization
one, and evolves with lessons learned from recent Gaull BA, Michael-Leiba MO, Rynn JMW (1990) Proba-
major earthquakes that are included in future revi- bilistic earthquake risk maps of Australia. Aust J Earth
sions as, e.g., the increase of the hazard factor Z for Sci 37:169–187
the Christchurch area from Z =0.22 g to Z = 0.30 g GeoNet-Earthquake Commission-GNS Science. website:
www.geonet.org.nz. Accessed Sept 2013
in the latest compliance document of the New McRae G, Clifton C, Megget L (2011) Review of NZ
Zealand Building Code (Department of Building building codes of practice. Report to the royal com-
and Housing 2011). mission of inquiry into the building failure caused by
the Christchurch earthquakes ENG.ACA.0016.1
Nakamura Y (1989) A method for dynamic characteristics
estimation of subsurface using microtremors on the
Cross-References ground surface. RTRI Q Rep 30(1):25–33
Niazi M, Bozorgnia Y (1992) Behaviour of near-source
▶ Earthquake Response Spectra and Design vertical and horizontal response spectra at SMART-1
array. Earthq Eng Struct Dyn 21:37–50
Spectra NZS 1170.5:2004. Structural design actions. Part 5: earth-
▶ Earthquake Return Period and Its quake actions-New Zealand. Standards New Zealand
Incorporation into Seismic Actions NZS 1170.5 Supp 1:2004. Structural design actions. Part
▶ European Structural Design Codes: Seismic 5: earthquake actions. New Zealand-commentary.
Standards New Zealand
Actions NZS 95. New Zealand standard code of building by-laws.
▶ Response-Spectrum-Compatible Ground Wellington, Standards Association of NZ
Motion Processes Petersen MD, Toppozada TR, Cao T, Cramer CH, Reichle
▶ Seismic Actions due to Near-Fault Ground MS, Bryant WA (2000) Active fault near-source zones
within and bordering the state of California for the 1997
Motion Uniform Building Code. Earthq Spectra 16(1):69–83
History Analysis
Structural Optimization Under S

References
Random Dynamic Seismic Excitation
Ambrasseys NN, Simpson KA (1996) Prediction of verti-
cal response spectra in Europe. Earthq Eng Struct Dyn Christian Bucher
25:401–412 Institute of Building Construction and
AS 1170.4:2007. Structural design actions. Part 4: earth-
quake actions in Australia. Standards Australia
Technology, Vienna University of Technology,
AS 3700:2001. Masonry structures. Standards Australia Wien, Austria
AS/NZS 1170:2002. Structural design actions. Standards
Australia/standards New Zealand
ASCE/SEI 7-10. Minimum design loads for buildings and Synonyms
other structures. ASCE Standard, American Society of
Civil Engineers, Structural Engineering Institute
Bouckovalas GD, Kouretzis GP, Kalogeras IS (2002) Site- Optimal design; Random vibration; Response
specific analysis of strong motion data from the surface; Seismic isolation; Structural reliability
3618 Structural Optimization Under Random Dynamic Seismic Excitation
Introduction
In civil engineering, the design of structures has Bad

Compromises
always been governed by fulfilling the needs of
both safety and economy. Ideally, the designer
Objective 2
Pareto front
would want to minimize cost while simulta-
neously maximizing safety. To a large extent,
these objectives are in conflict; therefore, suitable
compromises need to be found. A well-
established tool for finding this set of best com- Good
Compromises
promises is Pareto optimization. Figure 1 illus-
trates the trade-off between two conflicting Objective 1
objectives, both of which should be minimized.
Reducing one objective automatically implies Structural Optimization Under Random Dynamic
Seismic Excitation, Fig. 1 Pareto front for the case of
increasing the other one. Good compromises are two conflicting objectives to be minimized
to be found on the Pareto front.
Mathematically, this can be expressed as fol-
lows. Given a multi-objective minimization prob- (as specified, e.g., by design codes) and aiming
lem with m scalar-valued objective functions fk(x) for the economically best solution satisfying the
safety requirements. Mathematically, this corre-
f k : Dx ℝn ! ℝ (1) sponds to a constrained optimization problem.
Due to many uncertainties involved in the con-
a point x is called nondominated if (Ngatchou struction process as well as in the environmental
et al. 2008) loads acting on a structure, the optimization prob-
lem cannot be treated deterministically. Figure 2
f k ðx Þ fk ðxÞ8k 8x Dx and shows a simple example of a stochastic optimi-
(2) zation problem in which the design variables
∃‘ : f‘ ðx Þ < f‘ ðxÞ8x Dx
contain uncertainties (e.g., due to the manufactur-
ing or construction tolerances) and the objective
The Pareto set, i.e., the set of all nondominated (cost) function contains uncertainties (e.g., due to
solutions of an m objective problem can be price fluctuations on the markets), and finally, the
constructed by assembling all solutions xa of all safety constraint function contains uncertainties
aggregated optimization problems of the form (e.g., due to randomness of environmental loads
such as earthquakes or storms).
X
m
fa ¼ ak f k ðxÞ ! Min:! (3)
k¼1
Optimization Methodology
in which ak are positive coefficients. For more
involved problems, alternative method based on In a deterministic setting, a constraint optimiza-
multi-objective genetic algorithms (MOGA) such tion problem involves an objective function f(x)
as non-dominated sorting genetic algorithms and several constraint functions gk(x), k = 1. . . m.
(NSGA) (Ngatchou et al. 2008) may be more Then the problem can be formulated as
appropriate. Applications of these approaches
can be found, e.g., in Liu and Frangopol (2005) x ¼ argminf ðxÞ; gk ðxÞ 0; k ¼ 1 . . . m (4)
and Furuta and Kameda (2006).
Traditionally, however, in structural engineer- Since the cost function f will typically be ran-
ing, the abovementioned compromise has been dom, any optimization will have to be formulated
realized by setting minimum target safety levels on the basis of expected values or probabilities
Design Design
Variable 2 Variable 2
Contour lines of
objective function
rgin
ty ma
Safe
Infeasible Domain Infeasible Domain
Design Design
Variable 1 Variable 1
Structural Optimization Under Random Dynamic

Structural Optimization Under Random Dynamic
Seismic Excitation, Fig. 3 Safety margin for constraints
Seismic Excitation, Fig. 2 Simple example for a sto-
in a stochastic optimization problem
chastic optimization problem
derived from the cost function. In this way, the In traditional structural design, this is realized
objective function R for the optimization can be by introducing a safety margin (usually expressed
formulated, e.g., in terms of the mean value f and in terms of partial safety factors) into the other-
standard deviation sf, as wise deterministic formulation of the constraints
(cf. Fig. 3).
R ¼ f þ ksf (5)
Computational Tools
In this form, a multiple of standard deviation of
the cost function is added to the mean values to The reduction of the large computational effort
form the objective function. The case k = 0 cor- can be tackled in various ways, such as:
responds to game theory (suitable for, e.g., large
portfolio management, fair gambling), whereas • Reduce and simplify the structural model, e.g.,
k > 0 implies a so-called risk-aversive strategy. by modal reduction (Clough and Penzien
Such a strategy is commonly employed when 1993) or proper orthogonal decomposition
large losses cannot be compensated by long-term (Bamer and Bucher 2012).
averaging (strong asymmetry). • Reduce the probabilistic description, e.g., by
Also, the constraint functions gk(x) will be equivalent linearization (Roberts and Spanos
random. Therefore, it is helpful to use the proba- 2003) or tail-equivalent linearization
bility of constraint violation as new constraint, (Fujimura and Kiureghian 2007). S
i.e., subject to the minimization of R to • Approximate the input-output relation of the
system by a meta-model (Bucher and Macke
P½gk ðx Þ > 0 e (6) 2005).
• Approximate the input-failure probability
In safety-related constraints, the probability relation by a meta-model (Gasser and
level e may be very small, e.g., e ¼ 105 . Schuëller 1997).
Hence, it is absolutely mandatory to provide • Compute the input-failure probability relation
computationally efficient methods for computing by efficient Monte Carlo methods (Bucher
such small probabilities. This need is even more 2009a).
pronounced as the application of optimization
algorithm required the repeated computation of In the following, the response surface method
small failure probabilities (whenever the algo- and a specific Monte Carlo method (asymptotic
rithm changes the design variables). sampling) will be explained in some detail.
Response Surface Method h

The response surface method has been a topic of x2
extensive research in many different application
areas since the influential paper by Box and Wil-
son in 1951 (Box and Wilson 1951). A fairly com-
plete review on existing techniques and research
directions of the response surface methodology
can be found in Hill and Hunter (1966), Mead
and Pike (1975), Myers et al. (1989), and Myers
(1999). One of the earliest suggestions to utilize x1
the response surface method for structural reliabil-
ity assessment was made in Rackwitz (1982). Here Structural Optimization Under Random Dynamic
Lagrangian interpolation surfaces and second- Seismic Excitation, Fig. 4 Response surface and data
points
order polynomials are rated as useful response
surfaces. Moreover, the importance of reducing
the number of basic variables and error checking the response surface q(u; x(k)) and the measured
is emphasized. Polynomials of different orders in response z(k) at the m points of experiment
combination with regression analysis are proposed 0
ðkÞ
in Faravelli (1989), in which fractional factorial xðkÞ ¼ x1 , . . . , xðnkÞ , k ¼ 1, 2, . . . , m (7)
designs are utilized to obtain a sufficiently large
number of support points. The absolutely essential becomes as small as possible. In other words, the
validation of the chosen response surface model is sum of squares function
done by means of analysis of variance.
In Ouypornprasert et al. (1989), it has been m
X 2
pointed out that for reliability analysis, it is sðuÞ ¼ zðkÞ q u; xðkÞ (8)
k¼1
most important to obtain support points for the
response surface very close to or exactly at the has to be minimized. This corresponds to a min-
limit state g(x) = 0. This finding has been further imization of the variance of the random error
extended in Kim and Na (1997) and Zheng and terms e. The minimizing choice of u is called a
Das (2000). In Brenner and Bucher (1995), the least squares estimate and is denoted by u. ^
response surface concept has been applied to The above regression problem becomes more
problems involving random fields and nonlinear simple to deal with when the response surface
structural dynamics. Besides polynomials of dif- model is linear in its parameters u. Let us assume
ferent orders, piecewise continuous functions that the response surface is given by (Fig. 4)
such as hyperplanes or simplexes can also be
utilized as response surface models. ¼ y1 q1 ðxÞ þ y2 q2 ðxÞ þ þ yp qp ðxÞ (9)
Assume that an appropriate response surface
model q() has been chosen to represent The least squares estimate of the parameter
the experimental data. Then, for estimating the vector is given by
values of the parameters u in the model, the
u^ ¼ ð Q0 QÞ Q0 z
1
method of maximum likelihood can be utilized. (10)
Under the assumptions of a Gaussian distribution
of the random error terms e, the method of max- This estimator is unbiased, i.e.,
imum likelihood can be replaced by the more h i
common method of least squares (Box and E u^ ¼ u (11)
Draper 1987). In the latter case, the parameters
u are determined in such a way that the sum of A detailed discussion is given by (Bucher and
squares of the differences between the value of Macke 2005).
Structural Optimization
Under Random Dynamic
Seismic Excitation, hk (X1.....Xk )
x
Fig. 5 Schematic sketch of
first-passage problem
Load, Response
XK XN
X1
Time
Monte Carlo Simulation principle of causality, the values of h at point k in

time can only depend on the basic variables X1 . . .
First-Passage Problem Xk (i.e., on those in the past put to the present as
Generally, the probability of failure PF in an expressed by the index k) and not on those in the
n-dimensional space of random variables X1,. . . future.
Xn can be computed as The generalized safety index (or reliability
index) b is defined by
ð ð
PF ¼ f X1 , ... Xn ðx1 , . . . xn Þdx1 . . . dxn (12)
DF b ¼ F1 ð1 PF Þ ¼ F1 ½FH ðhÞ (14)
In this equation, f X1 , ...Xn ðx1 , . . . xn Þ denotes the Here F1(.) is the inverse standardized Gaussian
joint probability function of the random variables distribution function. Without loss of generality,
X1,. . . Xn, and DF denotes the failure domain, i.e., it is assumed that the random variables Xi are
the region of the n-dimensional random variable Gaussian and that they are independent and iden-
space in which failure occurs. Typically, this is tically distributed (i.i.d.) with zero mean and unit
denoted in terms of a scalar limit state function standard deviation.
g(.) attaining negative values, i.e., DF = {(X1, . . . Since the failure probabilities to be computed
Xn) j g(X1, . . . Xn) 0}. For applications in time- are usually very small (e.g., PF = 105), it is not
dependent problems as those arising structural feasible to evaluate the integral in Eq. 12 using
dynamics, the appropriate formulation leads to standard numerical integration procedures. This S
the first-passage problem. Here the failure is due to the fact that the number of integration
domain is typically written as points required to perform, e.g., Gaussian inte-
gration in dimension n, grows exponentially with
n. Monte Carlo methods do not have this depen-
DF ¼ ðX1 , . . . Xn Þj max hk ðX1 , . . . Xn Þ x
1kN dence on the dimension; however, crude or plain
(13) Monte Carlo simulation requires a number of
samples roughly larger than 10 PF . This, again,
In this equation, h(.) denotes a response quantity may lead to prohibitively large computational
of interest, and x is a critical threshold value of effort.
this response. The random variables Xi usually
denote the random excitation (e.g., earthquake or Asymptotic Sampling Method
wind) which is discretized in time. A schematic The asymptotic sampling method as described in
sketch is shown in Fig. 5. Note that due to the Bucher (2009a, b) exploits the asymptotic
behavior of the failure probability expressed in This means that for a linear function of Gaussian
terms of the reliability index b when changing the variables, the scaled reliability index b/f is
standard deviation of the basic random variables. invariant with respect to the choice of f.
If the original standard Gaussian random vari- For the general case, this is not true. There is,
ables are replaced by variables with non-unit however, an asymptotic property which ensures
standard deviations s ¼ 1f , then the computed similar behavior for many nonlinear cases. As
reliability index will depend on the choice of f. stated by Breitung (1984) and Gollwitzer and
As a first simple case, consider a linear function Rackwitz (1988), the reliability index asymptot-
of the basic random variables Xk, say ically depends linearly on f or in scaled notation
X
N bðf Þ
gðXÞ ¼ ak X k (15) lim ¼ const: (20)
f !1 f
k¼1
In order to exploit this asymptotic relation, Bucher

with arbitrary real-valued coefficients ak. The
(2009b) suggested to utilize the formulation
random variable Y = g(X) will then be Gaussian
with a zero mean and a variance
b B
¼Aþ 2 (21)
f f
X
N
s2Y ¼ a2k (16)
k¼1 in which the coefficients A and B are determined
from a regression analysis using the values of b as
The distribution function FY (y) of this variable estimated from several Monte Carlo runs with
will therefore be given by different values of f. It should be mentioned that
the asymptotic sampling method is a representa-
0 1
tive of a very recent family of methods exploiting
B C the dependency of the failure probability on a
x B x C
FY ð x Þ ¼ F ¼ FB s
B X ffiffiffiffiffiffiffiffiffiffiffiffi
ffiC
C (17) scaling parameter of the standard deviation of
sY @ N 2A
ak the input variables. A different approach to this
k¼1 is given by Naess and Gaidai (2008) and Naess
et al. (2010). For a comparative review and syn-
Upon changing the standard deviation of all basic thesis of these two approaches, it is referred to
variables from unity to a value of 1/f, the standard Qin et al. (2012).
deviation of Y changes by the same amount. Thus,
the distribution function changes to
Optimal Design of Seismic Isolation
0 1
Systems
B C
B xf C
B
FY ðxÞ ¼ FBsffiffiffiffiffiffiffiffiffiffiffiffiffiC
C (18) Definition of Problem
@ X N
A In order to ensure structural safety and integrity
a2k
k¼1 in earthquake conditions, it may be useful or even
necessary to equip structures with protective
from which the generalized reliability index devices. One possible choice is seismic isolation
according to Eq. 14 is immediately found as devices.
The basic scenario in which such devices are
xf b x utilized is shown in Fig. 6.
b ¼ sffiffiffiffiffiffiffiffiffiffiffiffiffi ! ¼ sffiffiffiffiffiffiffiffiffiffiffiffiffi (19) In this scenario, a structure is subjected to an
X N f X N
a2k a2k earthquake described by the ground acceleration
k¼1 k¼1 a(t). Its effect on the structure is to be mitigated
Structural Optimization
Under Random Dynamic F
Seismic Excitation,
Fig. 6 Structure and
friction-based seismic
isolation device
X
by a device which limits the transfer of forces

from the ground to the structure. One such device
consists of a combined friction and spring ele-
ment, in which the spring can also be replaced by
a re-centering force due to gravity effects
(so-called friction pendulum systems, see, e.g.,
Roussis and Constantinou (2006)). The device
has two characteristic parameters, one is the max-
R, m
imum force transmitted by friction (described by
the friction coefficient m) and the other is the Structural Optimization Under Random Dynamic
re-centering (or restoring) spring stiffness con- Seismic Excitation, Fig. 7 Simple sketch for friction
stant (described by the effective radius of curva- pendulum system
ture R) (Fig. 7). Both parameters should be
chosen such as to ensure the desired protective
effect for the structure and the integrity of the isolator will be good for that purpose. On the
isolation device itself. other hand, the isolator itself can have only lim-
The question of optimal design for stiffness ited displacements due to constructive reasons.
and friction of seismic isolation systems has been This means that the isolator should be “stiff”
addressed previously in the literature (e.g., enough. Hence, the targets (objectives) of the
Constantinou and Tadjbakhsh (1983), Iemura optimal design of a seismic isolation device are
et al. (2007), Jangid (2005)). The method of in conflict with each other. As mentioned above, S
equivalent linearization in conjunction with a the determination of the Pareto set allows the
random process model for the earthquake excita- preselection of suitable candidates for the final
tion tuned to the El Centro record was used in decision-making process.
Jangid (1996). Here, a full nonlinear dynamic The relevant quantities for the Pareto optimi-
analysis will be used as a basis for the computa- zation are the maximum displacement xD of the
tion of the structural response. The ground friction device (this should be small in order to
motion is modeled as a nonstationary (i.e., evo- ensure safety of the supports) and the maximum
lutionary) random process; hence, the response structural displacement xS (measured relative to
should to be characterized in suitable probabilis- the device, this should be small to ensure seismic
tic terms. isolation). As the quantities are random, charac-
As the main function, if the seismic isolation is teristic values (i.e., mean value plus three times
to keep the effect of the ground acceleration from the respective standard deviation) are chosen for
the superstructure, it is obvious that a “soft” the objectives.
Maximum device displacement XD [m] Maximum structural displacement Xs [m]
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.05 0.07 0.09 0.11 0.13 0.15
0.30 0.30
0.27 0.27
0.24 0.24
Friction coefficient m [-]

0.21 0.21
0.18 0.18
0.14 0.14
0.11 0.11
0.08 0.08
0.05 0.05
0.02 0.02
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
Radius R [m] Radius R [m]
Structural Optimization Under Random Dynamic structural displacement (right) depending on design
Seismic Excitation, Fig. 8 Characteristic value of max- parameters R and m
imum device displacement depending (left) and maximum
Structural Optimization 0.30

Seismic Excitation, 0.27 Swarm 0
Fig. 9 Convergence of
particle swarm 0.24
optimization method to one

0.21
point in the Pareto set
0.18 Swarm 9
0.14
Swarm 2
0.11
0.08
0.05
0.02
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
Radius R [m]
Selected Results discrete values are utilized to form response

In order to prepare the optimization, the charac- surface approximations which by inherent
teristic values for xD and xD are computed for smoothing largely eliminate the sampling
100 randomly selected combinations of the uncertainty inherent in Monte Carlo methods.
design variables R and m. At each sample point, The resulting response surfaces are shown in
the structural system is analyzed for 100 ran- Fig. 8.
domly generated earthquakes (using a An aggregated objective function R is then
nonstationary Kanai-Tajimi model (Bucher generated as
2009c)), and the characteristic values for xD and
xS are computed using sample statistics. These R ¼ axD þ ð1 aÞxS (22)
Structural Optimization 0.30

Seismic Excitation, 0.27
Fig. 10 Final Pareto set
0.24

0.21
0.18
0.14
0.11
0.08
0.05
0.02
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
Radius R [m]
in which a is a real number in the interval [0, 1]. Box GEP, Draper NR (1987) Empirical model-building
Solving the aggregated problem results in one and response surfaces. Wiley, New York
Box GEP, Wilson KB (1951) On the experimental attain-
point of the Pareto set. The convergence of this ment of optimum conditions. J R Stat Soc Ser B
optimization using a particle swarm optimizer is 13:1–45
shown in Fig. 9. Breitung KW (1984) Asymptotic approximations for
Finally, the results of aggregated optimization multinormal integrals. J Eng Mech 110(3):357–366
Brenner CE, Bucher C (1995) A contribution to the
runs for different values of a yield the Pareto set SFE-based reliability assessment of nonlinear struc-
as shown in Fig. 10. tures under dynamic loading. Probabilist Eng Mech
It turns out that the Pareto set contains very 10:265–273
large friction coefficients which are very difficult Bucher C (2009a) Computational analysis of randomness
in structural mechanics. In: Frangopol DM
to realize in practice. Hence, the optimal solu- (ed) Structures and infrastructures book series,
tions may not be accessible. vol 3. Taylor & Francis, London
Bucher C (2009b) Asymptotic sampling for high-
dimensional reliability analysis. Probabilist Eng
Mech 24:504–510
Cross-References Bucher C (2009c) Probability-based optimization of fric-
tion damping devices. Struct Saf 31:500–507
▶ Code-Based Design: Seismic Isolation of Bucher C, Macke M (2005) Response surface methodol-
Buildings ogy. In: Nikolaidis E, Ghiocel DM, Singhal S (eds) S
Structural reliability handbook. CRC Press, Boca
▶ Nonlinear Dynamic Seismic Analysis Raton, p 19, 1–19, 23
▶ Reliability Estimation and Analysis for Clough RW, Penzien J (1993) Dynamics of structures,
Dynamical Systems 2nd edn. McGraw-Hill, New York
▶ Robust Design Optimization for Earthquake Constantinou MC, Tadjbakhsh IG (1983) Probabilistic
optimum base isolation of structures. J Struct Eng
Loads 109(3):676–689
▶ Stochastic Analysis of Nonlinear Systems Faravelli L (1989) Response-surface approach for reliabil-
ity analysis. J Eng Mech 115:2763–2781
Fujimura K, Kiureghian AD (2007) Tail-equivalent line-
References arization method for nonlinear random vibration.
Probabilist Eng Mech 22:63–76
Bamer F, Bucher C (2012) Application of the proper Furuta H, Kameda T (2006) Application of multi-
orthogonal decomposition for linear and nonlinear objective genetic algorithm to bridge maintenance.
structures under transient excitations. Acta Mech In: Ceragioli F, Dontchev A, Futura H, Marti K,
223:2549–2563 Pandolfi L (eds) System modeling and optimization.
3626 Structural Reliability Estimation for Seismic Loading
IFIP International Federation for Information Roussis PC, Constantinou MC (2006) Uplift-restraining
Processing, Boston, pp 139–148 friction pendulum seismic isolation system. Earth-
Gasser M, Schuëller GI (1997) Reliability-based optimi- quake Eng Struct Dyn 35(5):577–593
zation of structural systems. Math Method Oper Res Zheng Y, Das PK (2000) Improved response surface
46:287–307 method and its application to stiffened plate reliability
Gollwitzer R, Rackwitz R (1988) An efficient numerical analysis. Eng Struct 22:544–551
solution to the multinormal integral. Probabilist Eng
Mech 3(2):98–101
Hill WJ, Hunter WG (1966) A review of response surface
methodology: a literature survey. Technometrics
8:571–590
Iemura H, Taghikhany T, Jain S (2007) Optimum design Structural Reliability Estimation for
of resilient sliding isolation system for seismic protec- Seismic Loading
tion of equipments. Bull Earthquake Eng 5(1):85–103
Jangid RS (1996) Optimum damping in a non-linear base
Achintya Haldar
isolation system. J Sound Vibration 189(4):477–487
Jangid RS (2005) Optimum friction pendulum system for Department of Civil Engineering and
near-fault motions. Eng Struct 27(3):349–359 Engineering Mechanics, University of Arizona,
Kim S-H, Na S-W (1997) Response surface method using Tucson, AZ, USA
vector projected sampling points. Struct Saf 19:3–19
Liu M, Frangopol DM (2005) Bridge annual maintenance
prioritization under uncertainty by multiobjective
combinatorial optimization. Comp-Aided Civil Inf Synonyms
Eng 20(5):343–353
Mead R, Pike DJ (1975) A review of response surface
Energy dissipation in dynamic analysis; Finite
methodology from a biometric view-point. Biometrics
31:803–851 element analysis; Implicit performance function;
Myers RH (1999) Response surface methodology – Nonlinear analysis; Risk management; Seismic
current status and future directions. J Qual Technol loading; Structural reliability; Time-domain
31:30–44
dynamic analysis; Time-domain reliability
Myers RH, Khuri AI, Carter JWH (1989) Response sur-
face methodology: 1966–1988. Technometrics analysis
31:137–157
Naess A, Gaidai O (2008) Monte Carlo methods for esti-
mating the extreme response of dynamical systems.
Introduction
J Eng Mech 134(8):628–636
Naess A, Gaidai O, Batsevych O (2010) Prediction of
extreme response statistics of narrow-band random The estimation of structural reliability has matured
vibrations. J Eng Mech 136(3):290–298 over the years. It has been accepted by the profes-
Ngatchou PN, Zarei A, Fox WLJ, El-Sharkawi MA
sion that if the risk in engineering design cannot
(2008) Pareto multiobjective optimization. In: Lee
KY, El-Sharkawi MA (eds) Modern heuristic optimi- completely be eliminated, it needs to be managed
zation techniques. Wiley-Interscience, Piscataway, appropriately. The issue has attracted added sig-
pp 189–207 nificance since the basic design philosophy has
Ouypornprasert W, Bucher C, Schuëller GI (1989) On the
changed from human safety to structural safety.
application of conditional integration in structural reli-
ability analysis. In: Ang AHS, Shinozuka M, Schuëller This is in response to limit enormous amount of
GI (eds) Proceeding 5th international conference on damage caused to infrastructures during recent
structural safety and reliability. San Francisco, earthquakes in China, Chile, Haiti, India, Iran,
pp 1683–1689
Japan, the USA, and other parts of the world. The
Qin J, Nishijima K, Faber MH (2012) Extrapolation
method for system reliability assessment: a new word “structure” is used here in a generic sense. It
scheme. Adv Struct Eng 15(11):1893–1909 represents real engineered systems including
Rackwitz R (1982) Response surfaces in structural reli- structures in a nuclear power plant, multistory
ability. Berichte zur Zuverlässigkeitstheorie der
buildings, onshore and offshore structures,
Bauwerke. Heft 67, M€ unchen
Roberts JB, Spanos P (2003) Random vibration and sta- soil–pile interaction problems in offshore mooring
tistical linearization. Dover Publications, Mineola systems, aerospace structures, computer
Structural Reliability Estimation for Seismic Loading 3627
packaging systems, systems represented by finite dynamic nonlinear problems, they are also to be a
elements, etc. Most of the discussions made here function of time. In response to questions on
are appropriate for different forms of dynamic future research directions in the risk evaluation
including seismic loadings. techniques, Rackwitz (2000) commented that the
The estimation of the probability of failure, use of FORM/SORM complemented by Monte
expressed in terms of risk or reliability, implied Carlo simulations (MCS) would be the next fron-
that it needs to be estimated just before failure tier. Several recent studies reflected this idea and
indicating a very complicated process a structure simulation has become an integral part of reli-
has to go through before failure. The appropriate ability evaluation studies.
risk management requires an acceptable reliability
method considering all major loads and load com-
binations that may act on the structures during Challenges in the Reliability Estimation
their lifetime, modeling the structures appropri- for Seismic Loading
ately satisfying the underlying physics at the time
of failure and analyzing their nonlinear behavior as Reliability estimation for dynamic or seismic
realistically as possible just before failure. loading is evolving. The probability of failure
If the performance functions or limit state implies that it needs to be estimated just before
functions are explicit, i.e., if they can be failure developing various sources of nonlinear-
expressed as explicit functions of all the random ities. The finite element method (FEM) is com-
variables present in the formulation and the per- monly used by the deterministic community to
formance requirements, there are several study nonlinear problems where dynamic load-
methods with various degrees of sophistication ings are applied in time domain. This clearly
that are available (Haldar and Mahadevan indicates that an FEM-based general purpose reli-
2000a). The most commonly used procedures ability evaluation method, parallel to the deter-
are the first-order and second-order reliability ministic analysis, is necessary.
method commonly known as FORM/SOR- The most sophisticated deterministic dynamic
M. They are also referred to as the Rackwitz- analysis of structures requires that the load should
Fiessler (1976) algorithm. Although not always be applied in time domain. This is one of the
necessary, for the most efficient implementation major challenges in the reliability analysis for
of the procedures require the performance func- seismic loading. The classical random vibration-
tions are available in explicit forms. The algo- based approaches were used in the past for this
rithms are iterative in nature and the gradients of purpose; however, they did not provide informa-
performance or limit state functions are required tion acceptable to the deterministic community.
to estimate the coordinates of the most probable The classical random vibration-based approaches
failure point (MPFP) and the corresponding reli- have numerous limitations including the loads S
ability index and failure probability. However, which are applied in the form of power spectral
their applicability in estimating the reliability of density functions essentially appropriate for lin-
real structural systems could be very limited. For ear structural behavior, the uncertainty in the
routine applications of FORM/SORM, the deriv- linear or nonlinear structural behavior which
atives of the performance functions are difficult may need to be incorporated in approximate
to obtain even for a simple structure like a ways, several performance-enhancing features
two-dimensional small frame. Furthermore, risk currently introduced in structures which cannot
analysis methods currently available were devel- be incorporated in appropriate ways, etc. The
oped for relatively simple systems with numerous most severe weakness is that the seismic loading
assumptions which cannot be satisfied for large cannot be applied in time domain.
realistic structural systems. For such systems, the Structural elements (beams, columns, connec-
limit states are expected to be implicit, and for tions, etc.) are generally designed first for
strength according to design guidelines given in Under dynamic or seismic loading, the structural
codes satisfying some underlying reliability response will depend on many factors including
requirements, although unknown to most the changes in the structural dynamic properties
designers. However, the overall system reliability (stiffness, frequency, damping, etc.) and the
in strength considering the reliabilities of all the corresponding response amplification as the load-
elements remains unknown. It is not addressed in ing progresses from linear to nonlinear stages; the
codes since it is difficult to estimate and can only sophistication in the physics-based structural
be assessed based on numerous idealistic modeling, i.e., FR versus PR connections and
assumptions (brittle or ductile behavior, depen- support conditions; the various sources of energy
dency of failure of elements, consideration of dissipation intentionally introduced in the system
nonlinear limit states, etc.) (Haldar and to improve its behavior; etc.
Mahadevan 2000a). Moreover, the design satis- One basic limitation of reliability analysis
fying strength requirements may not satisfy the under seismic loading is the incapability of han-
global performance or serviceability require- dling structural behavior appropriately. Because
ment, e.g., excessive lateral defection, a major the analysis is based on tracking the uncertainty
cause of structural failure for seismic loading. propagating through the steps of deterministic
Serviceability requirements also may control the analysis (Haldar and Mahadevan 2000b), the effi-
seismic design in most cases and should not be ciency of the deterministic FEM is extremely
ignored (Huh and Haldar 2002). important. It has been reported in the literature
(Kondoh and Atluri 1987; Shi and Atluri 1988;
Physics-Based Dynamic Behavior: Connection Haldar and Nee 1989) that the assumed stress-
Conditions based FEM has many advantages over the com-
Structural members are connected to each other monly used displacement-based FEM, particu-
using various types and forms of connections, larly for nonlinear analysis of frame structures
specifically for steel structures. These connec- capable of producing large deformation. In the
tions are routinely modeled as fully restrained assumed stress-based FEM, the tangent stiffness
(FR). However, extensive experimental studies can be expressed in an explicit form, the stresses
indicate that they are essentially partially of an element can be obtained directly, fewer
restrained (PR) connections with different rigid- elements are required to describe a large defor-
ities. In a deterministic analysis, PR connections mation configuration, and integration is not
add a major source of nonlinearity by decreasing required to obtain the tangent stiffness. Different
the overall stiffness of a frame. In a dynamic sources of nonlinearity, especially due to the
analysis, it changes the dynamic properties of presence of partially restrained (PR) connections,
the structures and adds a major source of energy can be incorporated without losing the basic sim-
dissipation. In reliability analysis, it also adds a plicity. It is very accurate and efficient in analyz-
major source of uncertainty. Thus, the reliability ing nonlinear responses. Some of its essential
of steel frames with PR connections and features in the context of the dynamic problems
subjected to seismic excitation is expected to be are discussed very briefly in the following
quite different than that of frames with FR con- sections.
nections. The implications of the presence of PR
connections and the uncertainty in modeling Unified Stochastic Finite Element Method
them are essential for the reliability analysis. (SFEM)
As discussed earlier, to capture a realistic struc-
Static Versus Dynamic or Seismic Analyses: tural behavior, an FEM-based formulation is
Improved FEM-Based Formulation desirable. When uncertainty in the variables in
The modeling of structures to extract response the formulation is considered, it leads to the con-
information under the static and dynamic appli- cept of stochastic finite element method (SFEM)
cation of loadings is expected to be very different. (Haldar and Mahadevan 2000b).
Without losing any generality, the limit state element models. The following iteration scheme
function can be expressed in terms of the set of can be used for finding the coordinates of the
basic random variables x (e.g., loads, material design point:
properties, and structural geometry), the set of
displacements u, and the set of load effects

gð yi Þ
s (except the displacements) such as internal yiþ1 ¼ yti ai þ ai (1)
j∇gðyi Þj
forces, stresses, etc. The displacement u = QD,
where D is the global displacement vector and
Q is a transformation matrix. The limit state where
function can be expressed as gðx, u, sÞ ¼ 0. For
reliability computation, it is convenient to trans- h it ∇gðyi Þ
@gðyÞ @gðyÞ
form x into the standard normal space y = y(x) ∇gðyÞ ¼ @y1 , . . . , @yn and ai ¼
∇g
j ðyi Þj
such that the elements of y are statistically inde-
pendent and have a standard normal distribution. (2)
An iterative algorithm can be used to locate the
design point or MPFP on the limit state function To implement the algorithm, the gradient ∇gðyÞ
using the first-order approximation. During each of the limit state function in the standard normal
iteration, the structural response and the response space can be derived as (Haldar and Mahadevan
gradient vectors are calculated using finite 2000b):

@gðyÞ @gðyÞ @gðyÞ @gðyÞ 1
∇gðyÞ ¼ Js , x þ Q þ Js, D JD, x þ J y, x (3)
@s @u @s @x
Pf ¼ FðbÞ ¼ 1:0 FðbÞ (5)

where Ji,j’s are the Jacobians of transformation
(e.g., Js,x = @s/@x) and yi’s are statistically inde-
where F is the standard normal cumulative dis-
pendent random variables in the standard normal
tribution function. Equation 5 can be considered
space. The evaluation of the quantities in Eq. 3
as a notational failure probability. When the reli-
will depend on the problem under consideration
ability index is larger, the probability of failure
(linear or nonlinear, two or three dimensional,
will be smaller. The author and his team
etc.) and the performance functions used. The
published numerous papers to validate the above
essential numerical aspect of SFEM is the evalu-
procedure for static application of the loading.
ation of three partial derivatives, @g/@s, @g/@u,
and @g/@x and four Jacobians, Js,x, Js,D, JD,x, and S
Jy,x. They can be evaluated by procedures
suggested by Haldar and Mahadevan (2000b) for
Dynamic Governing Equation for SFEM
linear and nonlinear, two- or three-dimensional
The nonlinear dynamic equilibrium equation can
structures. Once the coordinates of the design
be expressed at time t + Dt as:
point y* are evaluated with a preselected conver-
gence criterion, the reliability index b can be eval-
ðkÞ ðkÞ
uated as: €
M tþDt D þ t C tþDt D_ þ t Kðk Þ
k tþDt
DDðkÞ
ðkÞ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi €
¼ tþDt FðkÞ tþDt Rðk1Þ M tþDt D (6)
g
b ¼ ð y Þ t ð y Þ (4)
where M is the mass matrix, tC is the viscous
The probability of failure, Pf, can be calculated damping coefficient matrix at time t, tK(k) is the
as: global tangent stiffness matrix of the kth iteration
at time t, tþDt DDðkÞ is the incremental displace- and tKD is the dynamic tangent stiffness matrix
ment vector of the kth iteration at time t + Dt, and can be shown to be:
tþDt ðkÞ
F is the external load vector of the kth
iteration at time t + Dt, tþDt Rðk1Þ is the internal
t
KD ¼ f1 M þ f2 t K (10)
force vector of the (k1)th iteration at time
€ðkÞ is the seismic ground accel-
t + Dt, and tþDt D tþDt ðk1Þ
F
ðk Þ
and tþDt DFD are the modified external
g
eration vector of the kth iteration at time t + Dt. force and its incremental vector, respectively.
The damping matrix tC in Eq. 6 is considered The modified external force vector can be
to be viscous. In a realistic seismic analysis of expressed as:
frames, the amount of damping energy to be
generated will depend on the non-yielding and tþDt ðk1Þ
F € ðk1Þ
¼ tþDt Fðk1Þ þ tþDt Pðk1Þ M tþDt D
yielding state of the material and the hysteretic (11)
behavior if the material yields. For mathematical
simplicity, the effect of non-yielding energy dis- And tþDt Rðk1Þ is the internal force vector of the
sipation is generally represented by equivalent system. The term tþDt Pðk1Þ in Eq. 11 is the
viscous damping varying between 0.1 % and modified force vector contributed by the displace-
7 % of the critical damping (Leger and Dussault ment, velocity, and acceleration at time t and the
1992). Among many alternatives, Rayleigh-type displacement at time t + Dt and can be written as
damping is utilized in this study. It can be h i
represented as: tþDt ðk1Þ
P ¼ M f1 t D þ f3 t D_ þ f4 t D
€ f1 tþDt Dðk1Þ
h i
t
C ¼ aM þ g t K (7) þt K f5 t D þ f6 t D_ þ f7 t D
€ f5 tþDt Dðk1Þ
(12)
where tK is the tangent stiffness matrix and a and
ðkÞ
g are proportional constants which can be evalu- The incremental external force term tþDt DFD can
ated from the natural frequencies of the structure. be shown to be:
Since Rayleigh-type damping is proportional to
ðk Þ
mass and stiffness matrixes, it is more appropri- tþDt ðkÞ €
DFD ¼ tþDt DFðkÞ M tþDt DD (13)
g
ate for nonlinear behavior as it incorporates infor-
mation on changes in the stiffness values. The coefficients, fi’s, are constants and can be
evaluated (Haldar and Nee 1989) in terms of Z,
a, bN, g, and Dt as:
Numerical Procedures
1 Za Zg
f1 ¼ 2
þ , f2 ¼ þ 1,
The following numerical procedures can be bN Dt bN Dt bN Dt
followed to implement SFEM concept for seis- 1 Za
f3 ¼ þ a,
mic or dynamic loading. The step-by-step direct bN Dt bN
integration numerical procedure using the
1 1 1
Newmark-b method, with parameters Z = 1/2 f4 ¼ 1 þ Za Dt,
2bN 2bN Z
and bN = 1/4, can be used to solve Eq. 6. The
Za Zg Zg
iterative scheme can be expressed as: f5 ¼ , f6 ¼ g, f7 ¼ g Dt
bN Dt bN 2bN
(14)
t
KD tþDt DDðkÞ ¼ tþDt FðkÞ tþDt Rðk1Þ (8)
Equation 8 now can be solved using the modified
where
Newton–Raphson method. Once the displace-
ðkÞ
ments are obtained, the member forces can be
tþDt ðkÞ
F ¼ tþDt Fðk1Þ þ tþDt DFD (9) calculated accordingly.
Consideration of PR Connections in the The Richard model discussed up to now repre-

Finite Element Formulation sents only the monotonically increasing loading
portion of the M-y curves. However, the
The presence of partial connection rigidity needs unloading and reloading behavior of the M-y
to be incorporated into the deterministic analysis curves is also essential for any nonlinear seismic
of structures to capture their behavior. In general, analysis. This subject was extensively addressed
the relationship between the moment M, trans- in the literature (Colson 1991; El-Salti 1992).
mitted by the connection, and the relative rotation They theoretically developed the unloading and
angle y is used to represent the flexible behavior. reloading parts of the M-y curves using the
Among the many alternatives (Richard model, Masing rule. A general class of Masing models
piecewise linear model, polynomial model, expo- can be defined with a virgin loading curve
nential model, B-spline model, etc.), the Richard expressed as:
four-parameter moment–rotation model is cho-
sen here to represent the flexible behavior of a f ðM, yÞ ¼ 0 (18)
connection. It can be expressed as (Richard and
Abbott 1975): and its unloading and reloading curve can be
described by the following equation:

k kp y
M¼ !1 =N þ kp y (15) M M a y ya
ðkkp ÞyN f , ¼0 (19)
1 þ M0 2 2
where (Ma, ya) is the load reversal point as shown

where M is the connection moment, y is the in Fig. 1.
relative rotation between the connecting ele- Using the Masing rule and the Richard model
ments, k is the initial stiffness, kp is the plastic represented by Eqs. 15 and 17, the unloading and
stiffness, M0 is the reference moment, and N is reloading behavior of a PR connection can be
the curve shape parameter. These parameters are generated as:
identified in Fig. 1.

Although an ordinary beam-column element k k p ð ya yÞ
is used to represent a PR connection element for M ¼ Ma !1 =N kp ðya yÞ
ðk kp Þðya yÞN
numerical analyses, its stiffness needs to be 1 þ 2M0

updated at each iteration since the stiffness
representing the partial rigidity depends on y. (20)
This can be accomplished by updating the
Young’s modulus as: and S

lC lC @MðyÞ dM k kp
EC ðyÞ ¼ K C ð yÞ ¼ (16) K C ð yÞ ¼ ¼ þ kp
IC I C @y dy !Nþ1
ðk kp Þðya yÞN N
1 þ 2M0

where lC, IC, and KC(y) are the length, the
moment of inertia, and the tangent stiffness of (21)
the connection element, respectively. KC(y) is
calculated using Eq. 15 and can be shown to be: If (Mb, yb) is the next load reversal point, as
shown in Fig. 1, the reloading relation between
dM k kp M and y can be obtained by simply replacing (Ma,
K C ð yÞ ¼ ¼ þ kp (17)
dy !Nþ1 ya) with (Mb, yb) in Eqs. 20

Tsavdaridis 2015

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Tsavdaridis 2015

Uploaded by

Copyright:

Available Formats

S

Pgravity Pgravity + Vinertial Pgravity + Vinertial Pgravity + Vinertial

Hinertial Hinertial Hinertial

Loose Liquefied Liquefied

Stage I Stage II Stage III Stage IV

Figure 2 shows the concept of the two lengths. Step-by-Step Methodology

p2 EI i.e., Pfailure < Pcr. Equally, the growth of plastic

will be generated due to P-D moment, leading to Determination of “Critical Depth” HC

Summary ▶ Seismic Reliability Assessment, Alternative

Bhattacharya S, Adhikari S, Alexander NA liquefaction-induced lateral spreading in 1995

References Cosmo-Skymed; Differential interferometry;

Introduction imaging process returns only a 2D projection of

SAR Tomography for 3D

represent two coordinates of the natural radar 4p 4p

Gernhardt S, Adam N, Eineder M, Bamler R (2010) Poten-

School Seismic Safety and Risk Mitigation, Table 1 (continued)

School Seismic Safety and Risk Mitigation, Table 2 (continued)

School Seismic Safety and Risk Mitigation, Table 2 (continued)

hoped to enroll 24 million of these children in Cost-Effectiveness Arguments

safer traditional construction forms and practices Overview of Global Experiences in

References OECD (2004) Keeping schools safe in earthquakes.

Mitigation. They tell the stories of a variety of

Columbia (FSSS) identified and tackled concerns China

India: Uttar Pradesh Italy

graduate engineers of traditional and informal Turkey

importance factor of 1.5, public schools are • Implementation of a comprehensive

EQUIPMENT ANCHOR BOLT Equipment jumps Force fails

load (transferred from the PS to them, including Secondary Structural Elements

Secondary Structures and

If no flexible joints are considered (in between Modeling and Analysis

Secondary Structures and

Secondary Structures and 8

decisive eigenfrequency of the principal system ðlog43Þ

Secondary Structures and 50

Secondary Structures and

Secondary Structures and 0.6

Modeling and Analysis, 0.4

Secondary Structures and a

Secondary Structures and a

FRSs. Acceleration versus

References Taghavi S, Miranda E (2003) Response assessment of

Due to instrumental and practical reasons, the

parasitic vibrations (instrument vibration modes

z€ðtÞ þ 2ho0  z_ðtÞ þ o20  z ¼ €

Seismic Accelerometers, AMPLITUDE RESPONSE

response. The free period is

0,1 1 10 100 1.000

Displacement Transducer difference, and the phase depends on the sign of

a VARIABLE GAP b VARIABLE AREA

Seismic Accelerometers, CAPACITIVE TRANSDUCERS

Seismic Accelerometers, Fig. 6 Block schematic of an FBA

a-mechanical accelerometer response mm/(m/s2)

Seismic Accelerometers, 1.5

where kB is the Boltzmann constant The two capacitive techniques described

mounted in a printed circuit board (PCB) with com/productfile.html?productid=10319). And

by adding the recorded self- 4

Seismic Accelerometers, Fig. 14 The Etna

Seismic Accelerometers, Fig. 16 Nanometrics Titan

is a class C accelerograph with only event record-

Two kinds of accelerometer or accelerograph

ANSS Working Group on Instrumentation, Siting, Instal-

a Synthetic Ground Motions (High-Frequency)

Time (s) Time (s) Time (s) T (s)

z€ðtÞ þ 2ho0 z_ðtÞ þ o20 z ¼ €

mode1: T1 =0.77s, M1/Mtot =63.4% mode2: T2 =0.65s, M2 /Mtot =31.6%