Soil Dynamics and Earthquake Engineering 31 (2011) 830–840

Contents lists available at ScienceDirect

Soil Dynamics and Earthquake Engineering

journal homepage: www.elsevier.com/locate/soildyn

Developing fragility curves for a pile-supported wharf

Jiunn-Shyang Chiou a,n, Chi-Han Chiang a, Ho-Hsiung Yang b, Shang-Yi Hsu a
a
National Center for Research on Earthquake Engineering, 200, Section 3, Xinhai Rd., Taipei, Taiwan 10668, ROC
b
Taiwan Ocean Research Institute, 3F, 106, Ho-Ping E. Road, Section 2, Taipei, Taiwan 106, ROC

a r t i c l e i n f o a b s t r a c t

Article history: This study proposes a procedure for developing seismic fragility curves for a pile-supported wharf.
Received 31 August 2010 A typical pile-supported wharf, as commonly used in the ports of Taiwan, is chosen for demonstration.
Received in revised form For a structural model of the wharf, the deck is modeled by shell elements and the Winkler model is
20 January 2011
used for the pile–soil system, in which the piles and soils are represented by beam elements and
Accepted 25 January 2011
springs, respectively. A pushover analysis with lateral loads distributed according to the fundamental
Available online 19 February 2011
modal shape of the wharf structure is conducted to deduce the capacity curve of the wharf. The
procedure for developing fragility curves can be explicitly performed using the spreadsheet platform in
Microsoft EXCEL. First, quantitative criteria for damage states are established from the sequence of
development of plastic zones. Then a nonlinear static procedure called the Spectrum Capacity Method
(CSM) is used to efficiently construct a response matrix of the wharf to 24 earthquake events with
differing levels of peak ground acceleration (PGA). Based on the damage criteria and the response
matrix, the fragility curves of the wharf can be thus constructed through simple statistical analysis.
Shifted lognormal cumulative distribution functions are also employed to better approximate the
fragility curves for practical applications.
& 2011 Elsevier Ltd. All rights reserved.

1. Introduction building structures [7–9] and bridge structures [10–12], as well as

Wharf structures that accommodate import and export activ-
ities are essential components of a port transportation system. For
a port in an area of high seismicity, wharf structures are
susceptible to earthquake damage and may adversely affect the
normal operation of the port system. For example, the 1995
Hyogoken-Nanbu earthquake damaged many wharf structures,
such as caissons and pile-supported wharfs, in Kobe Port,
Japan [1], and the 1999 Chi-Chi earthquake caused damage to
some caisson quay walls in Taichung Port, Taiwan [2]. These ports
suffered enormous economic losses due to the downtime of the
wharfs.
A great deal of research using either model tests or numerical
simulations or both was carried out to investigate the behavior of
wharf structures during an earthquake [3–6]; however, less effort
was made in determining the seismic vulnerabilities of wharf
structures. The use of fragility curves has been deemed an
effective means for evaluating the seismic vulnerability of a
structural member or system. Fragility curves express the condi-
tional probability of exceeding a certain damage state for a given
ground motion intensity. This approach is commonly adopted in
n
Corresponding author.
E-mail address: jschiou@ncree.narl.org.tw (J.-S. Chiou).
in some geotechnical structures such as expressway embank-
ments [13], dams [14], and so on. It is also a useful tool in current
performance-based earthquake engineering to link economic
losses, like repair costs, repair durations and loss of life, to the
damage of structures or systems [15–16]. With the aid of the
fragility curves of wharf structures, Na and Shinozuka [17]
proposed a simulation-based framework for evaluating the eco-
nomic loss of a damaged seaport system.
Some well-known earthquake loss analysis systems, such as
Hazards U.S. (HAZUS) [18] and the Taiwan Earthquake Loss
Estimation System (TELES) [19], also use fragility curves to
estimate the damage probabilities of the components in a system.
In HAZUS, four classes of fragility curves are recommended for
items in a port system: waterfront structures, cranes and cargo
handling equipment, fuel facilities, and warehouses. In this
system, wharf structures are categorized as a waterfront struc-
ture. Wharf structures actually have three main structural types:
gravity quay wall, sheet pile quay wall, and pile-supported wharf.
However, in HAZUS, they are assigned the same set of fragility
parameters, regardless of the differences between the structural
types. Since these structural types have different structural
characteristics, they should have different fragility curves to
express their own seismic vulnerabilities properly.
Therefore, the objective of this study is to propose a procedure
for a pile-supported wharf to develop its specific fragility curves

0267-7261/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.soildyn.2011.01.011
 J.-S. Chiou et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 830–840
efficiently. To illustrate this, a typical pile-supported wharf used
in Taiwan is chosen as an example. In the following sections, first,
a framework for generating fragility curves is introduced and then
the items in it are explained explicitly.
2. Framework for developing fragility curves
As defined in HAZUS, fragility curves can be ground motion-
related or permanent ground displacement-related. The former
is for the damage due to the ground shaking. The latter is for
the damage due to the permanent displacement induced by
ground failure such as liquefaction or a landslide. For waterfront
structures, HAZUS only provides permanent ground displace-
ment-related fragility curves, without curves for ground motion.
Therefore, this study aims to derive the ground motion-related
fragility curves for wharf structures.
Fragility curves can be derived through professional judg-
ment [20], damage data associated with past earthquakes [13,21],
or numerical simulations based on quasi-static or dynamic
analysis [10–12]. Fragility data from professional judgment may
be subjective because the results may be dominated by a few
experts. On the other hand, damage data from past earthquakes
may be better, but it is usually too scarce to provide sufficient
samples and cannot cover a wide range of shaking intensities.
Since there is little damage data associated with past earthquakes
for pile-supported wharfs in Taiwan, this study develops fragility
curves by numerical analysis.
Various methods have been proposed for generating fragility
curves, which can be broadly categorized in two ways. The first is
to pre-assign the form of the fragility curves, usually in the form
of two-parameter (median and logarithmic standard deviation)
lognormal distribution functions. Shinozuka et al. [21] used
lognormal functions to develop empirical and analytical fragility
curves for bridges. For a particular damage state, they used the
maximum likelihood method to determine the unknown para-
meters of the fragility function by analyzing the fraction of
damaged samples under different excitation levels. Similarly,
Lignos et al. [22] used several types of fragility functions to
analyze experimental data for fragility curves of reduced beam
section moment connections.
Without prescribing the form of the fragility curves, the other
type of method is to directly compare the demand and capacity of
a structure. Either the demand or the capacity or both of the
structure are deemed to be stochastic for analyzing the damage
probabilities of the structure through statistic analysis for differ-
ent levels of seismic loading. In these analyses, major differences
are often made in the different damage models that are adopted
to characterize the damage of structures. For instance, Hwang and
Jaw [7] used the story ductility ratio as a damage measure for
buildings. It seems to be a direct method since structural damage
due to an earthquake is generally attributed to excessive inter-
story displacement. From another point of view, Hwang and
Huo [23] used a dissipation energy-based damage model pro-
posed by Park and Ang [24] for generating fragility curves of
structures. The Park and Ang model has been widely used because
its damage index is simple [8,11]; however, it normally needs to
be applied with a nonlinear time-history analysis to compute
the dissipated energy of the structural members. In computing
damage probability, a direct but burdensome way is doing Monte
Carlo simulations [8,10,11,23]. Also, reliability methods, like
FORM or Subset Simulation, can be alternative ways to obtain
the damage probability more efficiently, especially for a system
with multiple damage types or criteria [25–27]. Moreover, for the
analytical fragility curves generated, they can be further updated
using the Bayesian method when available actual damage data is
831
incorporated [9,28]. The above discussion mainly regards time-
invariant fragility curves. Recently, an advanced concept that
considers age effects was proposed for developing time-variant
fragility curves [29], but the approach needs more information
about the deterioration of each component in a system.
In this study, the second type of method is used. A framework
for developing fragility curves for a wharf structure is proposed as
follows:
(1) Construct the numerical model of a target wharf structure
and analyze its capacity curve. The structural properties of
the wharf are assumed to be determinate in this study.
(2) Choose a pool of representative ground motions near the site
as earthquake events for fragility analysis.
(3) Calculate the seismic responses of the wharf subject to
selected earthquake events, which are scaled to different
levels of ground motion intensity.
(4) Define the criteria for damage states and generate fragility
curves through damage probability analysis based on the
seismic response data and the set damage criteria.
(5) Use continuous functions to further approximate the above
discrete fragility data for practical applications.
This whole process is a cumbersome task. Based on this
framework, the following sections will consider the characteris-
tics of wharf structures to develop a specific procedure to
construct the fragility curves of a pile-supported wharf efficiently.
3. Illustrative example
A pile-supported wharf consists of a deck supported by a
number of piles beneath it. A typical wharf in the Kaohsiung Port
in Taiwan, commonly used in Taiwan, is selected for illustration.
The profile of the example wharf is shown in Fig. 1. In this port, a
total of 16 wharfs of combined length 2700 m were constructed
with this profile. Since the pile-supported wharf is separated from
the sheet pile wall behind it by expansion joints, this study
assumes that the sheet pile wall is stable and does not interact
with the wharf during an earthquake. The selected wharf unit has
a length of 37 m and a width of 12 m. The water level is 0.8 m
below the bottom of the deck. The deck is a reinforced concrete
slab that is supported by a group of 66 vertical precast prestressed
hollow piles, each with a diameter of 0.5 m and a thickness of
9 cm. The pile spacing is 3.7 m in the longitudinal direction and
2 m in the transverse direction.
3.1. Numerical model
The program SAP2000 [30] is used to construct the 3D
numerical model of the wharf, as shown in Fig. 2. The deck is
modeled by the shell elements and the Winkler beam model is
adopted to model the pile–soil system. Piles are modeled by beam
elements and rigidly connected to the deck. The soil is modeled
by spring elements with nonlinear p–y relations for horizontal
reactions and fs–z curves for skin friction. The horizontal soil
reaction p is defined as
p ¼ kh D y ð1Þ
where kh is the horizontal subgrade reaction coefficient at the
lateral displacement y and D is the pile diameter. In this study, the
initial horizontal subgrade reaction coefficient kh0 is set to be
1500 Ns (kN/m3), in which Ns is the blow count of the Standard
Penetration Test [31]. For the nonlinear part of the p–y curve, a
832 J.-S. Chiou et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 830–840

1200
EL +2.20 m 200 200 200 200 200 150 GL +0.00 m

EL +0.80 m EL +1.10 m
L.W.L ±0.00 m Expansion joint
Anchor

SM

GL -6.20 m

1.7
1 ML
GL -9.20 m
Rubble

1
4
EL -10.50 m PC Pile SM

PC Pile EL -12.85 m
φ = 50cm , L = 23.5m GL -15.20 m

PC Pile
φ = 50cm , L = 22.5m

PC Pile
φ = 50cm , L = 20.5m
SM

EL -19.50 m

EL -21.0 m
Unit : cm
EL -22.50 m

Fig. 1. Typical profile of pile-supported wharf in Kaohsiung Port.

Fig. 2. Numerical model of the wharf.

 J.-S. Chiou et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 830–840 833

reduced subgrade reaction coefficient kh is adopted through
0:5
kh ¼ kh0 y
where y is the lateral displacement in centimeters. The reference
displacement for kh0 is set to be 1 cm. This equation is similar to
that commonly used in Japanese practice [32–33]. Since the piles
are partly buried in the slope, a reduction in the subgrade reaction
is considered for the slope side. A multiplier f is adopted based on
the criteria used in Japan [34] as follows:
f ¼0 0 o Z o0:5
f ¼ 0:3 log Z þ 0:7 0:5 r Z r 10
f ¼1 10 o Z
where Z is the ratio of the horizontal pile–slope distance to the
pile diameter, as shown in Fig. 3.
The unit skin friction fs is defined as
fs ¼ ks z
where ks is the side subgrade reaction coefficient at a vertical
movement of z. In this study, ks is assumed to be 0.3kh at the same
magnitude of displacement [34].
The R/C piles used in the target wharf have cross-sections as
shown in Fig. 4; their moment–curvature curves can be calculated
by section analysis using the program UC Fyber [35]. The curve is
further simplified as a trilinear line that passes through the crack
point, the first yield point, the effective yield point, and the
ultimate point, as in Fig. 4. The effective yield point is determined
using the equal-area method so that the integral of the trilinear
line is equal to that of the original curve. The plastic hinge model
is applied to simulate the plasticity that develops in piles. Since
ð2Þ
the locations of plastic zones are unknown before the analysis, the
distributed plastic hinge model [36] is used in this study. As
shown in Fig. 5, this model distributes a series of plastic hinges
along potential plastic zones and thus can effectively model the
development of plasticity in the pile during a loading process.
Based on the sectional moment–curvature relationship shown
in Fig. 4, the properties of the distributed plastic hinges are
calculated as a curve displayed in Fig. 6. According to [36], the
plastic curvature is defined as
ð3Þ
M
kp ¼ k ð5Þ
EI
where k is the total curvature of the pile section due to the
applied moment M, and EI is the elastic flexural rigidity. The
ð4Þ subtraction of M/EI from the total curvature indicates that the real
plastic curvature should exclude the part of the curvature of
elastic rebound at the moment M.
In Fig. 6, four points define the curve: A is the origin, B is the
crack point, C is the effective yield point and D is the
ultimate point.

Slope surface

D

Fig. 5. Distributed plastic hinge.

Fig. 3. Pile to slope distance.

400
400
350 350
D. Ultimate point
C. Yield point
300
Effective yield point Ultimate point 300
Moment (kNm)
Moment (kNm)

First yield point 250

250
φ 16mm steel bar
200 200 B. Crack
Crack point
150 150
100 Effective prestress: 100
7.84 MPa
50 Section analysis 50
50 cm Simplified A. Origin
0 0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Curvature (1/m) Plastic curvature (1/m)

Fig. 4. Moment–curvature curve of pile section. Fig. 6. Property of distributed hinges.

834 J.-S. Chiou et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 830–840

3.2. Pushover curve
Once the structural model of the wharf is constructed, the next
step is to perform a pushover analysis with SAP2000 to obtain its
capacity curve. The modal analysis is performed first to get the
fundamental modal shape of the wharf. Fig. 7 shows the funda-
mental modal shape of the wharf structure with a fundamental
period of 0.73 s. In addition to a seaward movement, the deck also
shows a rocking motion towards the sea. To model the inertial
loads on the piles and on the deck, the pushover forces are applied
in proportion to the fundamental mode shape based on a given
total lateral force V as [37]
!
X N
Fi ¼ wi fi = wi fi V ð6Þ
i¼1
where Fi is the lateral force on node i (i¼1, 2, 3, y, N), N is the
number of nodes, wi is the dead weight associated with each node i,
fi is the amplitude of the fundamental mode at node i, and V is the
total lateral force (equivalent to the resultant of all the pushover
forces acting on the wharf).
The monitored point for the deck displacement is set at the
center of the deck. Through pushover analysis, the total lateral
force is increased from zero to the ultimate state and the capacity
curve in terms of the total lateral force versus the corresponding
deck displacement is deduced as shown in Fig. 8.
3.3. Damage sequence
On the pushover curve in Fig. 8, six damage points are
identified: Ct, Cd, Yt, Yd, Ut and Ud. The point of Ct represents
the state where a pile initially cracks at the pile top, Cd is the state
where a pile initially cracks below the water, Yt is the state where
a pile yields at the pile top, Yd is the state where a pile yields
below the water, Ut is the state where a pile reaches its ultimate
state at the top and Ud is the state where a pile section below the
obtained from nearby seismic stations with similar geologic
conditions to that site; their magnitudes and shaking levels are
large enough, and the sources of the earthquakes will be at
different positions to reflect their spatial variability. Under the
above considerations, 12 sets of earthquake records from past
earthquakes (1995–2010) that happened in Taiwan are selected.
These ground motions were recorded by seismic stations within
10 km of Kaohsiung Port. They all had a local magnitude ML above
4.0 and a PGA level of above 0.015 g. Their basic information and
properties are shown in Table 1. This study utilizes the E–W and
N–S directions of the acceleration records to produce 24 earth-
quake events for fragility analysis.
5. Analysis method for the seismic response
To calculate the seismic response of a structure, either
dynamic or static analysis methods can be applied, but it is
preferable to apply the static analysis because dynamic analysis
is time-consuming. Some static approaches have been proposed
to obtain the dynamic responses of a structure. They include the
equal energy method [38], the inelastic spectrum method [39–41],
the capacity spectrum method [37], and so on. This study uses the
Capacity Spectrum Method (CSM), which is a systematic non-
linear static procedure. The method was proposed by ATC-40 [37]
for the seismic evaluation and retrofit of buildings. Further, a
modal pushover analysis procedure was developed to extend the
capacity of the CSM to building structures with several modes of
vibration [42]. And the same concept is currently applied to
bridges [10,43–44], even for developing their fragility curves [10].
6000
a pile section below
5000 the water reaches
its ultimate state
Total lateral force (kN)

water reaches its ultimate state. The corresponding responses of a pile reaches its ultimate state at the top
a pile yields below the water
the deformed wharf are shown in Fig. 9. It can be seen that during 4000
a pile yields at the top
the pushover process, the plasticity occurs first at the pile tops, at
the row nearest to the land side, and gradually moves downward 3000
Capacity
and seaward. The damage in the piles in the row closest to the a pile initially cracks below the water Ct
land side is extensive. When the deck displacement reaches 2000 Cd
Yt
around 0.22 m, the capacity reaches the maximum. Afterwards, a pile initially cracks at the top
Yd
the capacity begins to decrease. 1000 Ut
Ud

0
4. Ground motions 0 0.05 0.1 0.15 0.2 0.25 0.3
Deck displacement (m)
To select representative ground motions for the fragility
analysis of wharf structures at a particular site, records are better Fig. 8. Pushover curve of the wharf.

Fig. 7. Fundamental modal shape of the wharf structure.

 J.-S. Chiou et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 830–840 835

Fig. 9. Damage development during pushover process.

Table 1
Earthquake records for fragility analyses. Sa

No. Time PGA (gal) Source Local

depth magnitude Initial demand spectrum
(km) (ML)
UD NS EW

1 1995/04/25/16/29 39.96 39.18 38.22 43.7 5.5

2 1996/07/06/09/27 13.40 28.23 22.49 27.0 5.2
3 1998/11/18/06/27 9.57 22.97 23.45 16.5 5.5
4 1999/09/21/01/47 12.62 28.54 32.36 8.0 7.3
5 1999/09/26/07/52 7.42 16.62 16.38 12.1 6.8 ap
6 2000/05/08/00/18 18.90 45.58 42.16 39.8 5.0
7a 2003/12/10/12/38 10.82 38.34 35.30 17.7 6.4
8a 2004/05/19/15/04 7.24 24.58 16.32 27.1 6.0 Performance point
9 2006/04/01/18/02 10.46 69.38 44.92 7.2 6.2
10a 2006/06/17/05/33 9.44 20.12 27.02 31.1 4.1
11a 2006/12/26/20/34 50.32 128.84 122.91 50.2 7.0
Capacity Reduced demand
12 2010/03/04/08/18 20.83 35.58 53.48 22.6 6.4
spectrum spectrum
a
The epicenter was located at the sea area off the Taiwan island.

dp Sd
In the CSM, the acceleration history of an earthquake event
and the capacity curve of the wharf structure are transformed into
Fig. 10. Capacity Spectrum Method.
the demand spectrum and the capacity spectrum, respectively, in
the form of an acceleration–displacement response spectrum,
when adopting the Type A model [40,41]. Accordingly, an
Sa versus Sd (ADRS format), as shown in Fig. 10. When the structure
improved method using inelastic spectra instead of damped
enters the nonlinear state, the earthquake demand may be
elastic spectra was proposed [40,41]. However, in actual applica-
damped down due to the hysteretic response of the structure
tion, for existing structures, the ATC-40 suggests using Type B and
and thus a modifying factor of less than one is included to further
C model which correspond to smaller damping ratios. In this study,
reduce the initial demand spectrum. The modifying factor is a
the Type C model is adopted because the example wharf is old
function of the effective damping associated with the maximum
and most earthquake events selected are long duration of shaking
response of the structure. The intersection of the reduced demand
which may increase the potential for degradation of the wharf
spectrum and the capacity spectrum gives the performance point
structure.
(dp, ap), which represents the maximum seismic response of
the structure to the earthquake event. Since the location of the
performance point is unknown beforehand, iterative trials are
required to determine the performance point. The ATC-40 suggests 6. Fragility curves by a spreadsheet platform
three different procedures to simplify the iteration process. The
details can be found in ATC-40. Fragility analysis can be easily performed with spreadsheet
In ATC-40, three hysteretic damping models Type A, B, and C calculations using Microsoft EXCEL, as shown in Fig. 11. There are
that correspond to the full, moderate, and poor hysteretic loop, four parts in the spreadsheet. The first is for introducing the
respectively, are defined. Among them, the ATC-40 procedure bounds of the damage states. The second is for constructing a
significantly underestimates the displacement demand of systems response matrix for the wharf subjected to earthquake events
836 J.-S. Chiou et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 830–840

Fig. 11. Spreadsheet platform for fragility analysis.

with different PGA levels. The third is for performing statistical
analysis to deduce the fragility curves. The final step is to simplify
the obtained fragility curves for easy application. These steps are
as follows.
6.1. Bounds of the damage states
On the pushover curve obtained, several important damage
points have been identified to show the degree of damage to the
wharf, as shown in Fig. 8. The International Navigation Associa-
tion (PIANC) [38] proposed qualitative criteria for judging the
degree of damage to a pile-supported wharf based on the peak
responses of the piles, as shown in Table 2. In this table, four
damage states, I, II, III and IV, correspond to serviceable, repair-
able, near collapse and collapse levels of a wharf structure,
respectively. At the serviceable level, the structure should have
minor or no structural damage and the structure should continue
to function. At the repairable level, the structural damage is
controllable and repairable. At the level of near collapse, the
structural damage is extensive. At the level of collapse, the
structural strength is completely lost. However, PIANC did not
quantitatively provide the bound for each damage state. Accord-
ingly, based on the sequence of plasticity development in the
pushover process, this study sets the points Cd, Yd and Ud to be
the upper bounds of the damage states I, II and III, respectively, as
shown in Fig. 12. The reasons for this decision are as follows.
Since the cracking of a pile in the water will cause erosion by the
sea water and affect the functionality of the wharf, the upper
bound of the damage state I is defined at the point Cd that the pile
cracks in the water. When the wharf reaches the point Yd, some
of the piles yield at the pile tops. As in-ground yielding is
 J.-S. Chiou et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 830–840 837

Table 2
Damage criteria for pile-supported wharfs [38].

Damage state

I Serviceable II Repairable III Near collapse IV Collapse

General Minor or no structural
damage
Piles (peak response) Essentially elastic response
with minor or no residual
deformation
Controlled structural Extensive structural Complete loss of structure
damage damage in near collapse
Controlled limited inelastic Ductile response near Beyond the State III
ductile response and collapse (double plastic
residual deformation hinges may occur at one or
intending to keep the limited number of piles)
structure repairable

6000 1
Capacity spectrum
0.9 Original demeand spectrum (5% inherent damping) -- EQ7 (NS)
5000 Reduced demand spectrum (6% effective damping)
0.8
Total lateral force (kN)

Performance Point

Spectral Acceleration (g)

4000 0.7
I II III IV
0.6
3000 Capacity
Ct 0.5
Cd
2000 Yt
Yd 0.4 (dp, ap) = (0.146m, 0.42g)
Ut
Ud 0.3
1000 I / II
II / III
III / IV 0.2
0 0.1
0 0.05 0.1 0.15 0.2 0.25 0.3
Deck displacement (m) 0
0 0.1 0.2 0.3 0.4 0.5
Fig. 12. Damage bounds on capacity curve. Spectral Displacement (m)

Fig. 13. Performance point for Event 14 (EQ7-NS) by CSM (PGA ¼0.3 g).
inaccessible for repairs, the point Yd is defined as the upper bound
of the damage state II. When the wharf is at the point Ud, a pile
calculated from the information on the two parameters of the
has two plastic zones where the pile sections reach the ultimate
normal distribution: the mean (m) and the standard deviation (s)
state. Therefore, the upper bound of the damage state III is set at
of the sample population [45] as shown below:
the point Ud.
l ¼ lnm12B2 ð8  1Þ
6.2. Response matrix
B2 ¼ ln½1 þ d2  ð8  2Þ
The PGA is chosen as a parameter to represent the ground where d is s/m.
shaking intensity. The selected earthquake events are normalized
to 0.l g and then scaled to eight PGA levels of 0.1–0.8 g. For each
6.3. Fragility analysis
PGA level, the performance points of the structure due to the
selected earthquake events are calculated. For example, for Event
According to the displacement bound of each defined damage
14 (EQ7-NS) with a PGA of 0.3 g, the CSM is employed to obtain
state, the fragility curve for the damage state si is the conditional
the performance point of the wharf, which is the intersection of
probability that the wharf has a state of damage exceeding the
the capacity spectrum and reduced demand spectrum, as shown
damage state si at a specific PGA level, as shown below:
in Fig. 13. In this figure, through several iterations, when a trial  
performance point (dp, ap) is selected to be (0.146 m, 0.42 g), the     lnðxi Þl
P S 4s9PGA ¼ P X 4xi 9PGA ¼ 1F ð9Þ
reduced demand spectrum associated with this trial point B
(6% effective damping) intersects the capacity spectrum at the where F(  ) is the standard normal cumulative distribution
same position (the circular hollow point) and thus the perfor- function, xi is the upper bound for si (i¼I, II, III), and l and z are
mance point is found. Collecting all the deck displacements of the as defined above and are dependent on the PGA level.
24 selected events, each with eight PGA levels, produces a 24  8 In this way, the fragility curves can be obtained as shown in
response matrix. Fig. 14. From this figure, it is easy to read the damage probability
For each column in the matrix (i.e., being at the same PGA for each damage state at a specific PGA level. For example, when
level), the displacement responses can be further assumed to be a the PGA is at 0.3 g, the fragility probabilities for the damage states
lognormal distribution with the probability density function I, II and III are 75%, 28% and 8%, respectively.
(PDF) as follows:
"  2 #
1 1 ln xl 6.4. Simplified fragility curves
fX ðxÞ ¼ pffiffiffiffiffiffi exp  0rX o1 ð7Þ
2pBx 2 B
For convenient application, the fragility curves are usually
where l and z are the two parameters of the lognormal distribu- expressed as lognormal cumulative distribution functions
tion of the random displacement variable X. They can be (lognormal CDF) [7,18–19,27]. In this way, the fragility curves
838 J.-S. Chiou et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 830–840

1.0 1.0
I I
0.9 I 0.9 II I
II III
0.8 III 0.8 I (lognormal)
II (lognormal)
0.7 II 0.7 III (lognormal)
II

P [S> s |PGA]
P [S> s |PGA]

0.6 0.6
0.5 III 0.5 III
0.4 0.4
0.3 0.3
0.2 IV 0.2 IV
0.1 0.1
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
PGA (g) PGA (g)
Fig. 14. Fragility curves of the example wharf. Fig. 15. Simplified lognormal fragility curves.

can be represented by only two parameters, as follows: Table 3

Z a "  2 # Fragility parameters for the example wharf.
1 1 ln aln mA
FA ðaÞ ¼ pffiffiffiffiffiffi exp  da ð10Þ
0 2pBA a 2 BA Damage state Median (g) Standard deviation

where A is the random variable of the PGA, mA is the median of A, I 0.21 0.72
II 0.46 0.83
and zA is the logarithmic standard deviation of A.
III 0.89 1.09
Eq. (10) can be related to the standard normal cumulative
distribution function with an argument Z, as follows:
FA ðaÞ ¼ F½Z ð11Þ 1.0 I
0.9 II I
where Z is the standard normal variable, which is defined as III
0.8 I (log)
[ln(a) ln(mA)]/zA. II (log-shifted)
On the fragility curve, for the fragility probability FA(a) at a 0.7 III (log-shifted) II
P [S > s |PGA]

PGA level of a, the associated normal variable can be computed 0.6

using the following equation: 0.5 III
Z ¼ F1 ½FA ðaÞ ð12Þ 0.4
0.3
where F  1(  ) is the inverse function of the standard normal
cumulative distribution.
0.2 IV
According to the definition of a standard normal variable, for a 0.1
fragility curve with a lognormal distribution, the relationship of 0.0
Z versus ln(a) is linear. The intercept and slope of the linear 0.0 0.2 0.4 0.6 0.8 1.0
relation will be ln(mA) and zA, respectively. PGA (g)
Therefore, to fit the fragility curve to the lognormal cumulative
Fig. 16. Simplified fragility curves with shifted lognormal CDF.
distribution function, the above idea can be applied. First, for each
fragility curve, the standard normal variables Z associated with
To decrease the discrepancy of the poor fit in this range, a lower
the fragility probabilities for the eight PGA levels are calculated
bound of the PGA is set to shift the lognormal CDF off the range of
by Eq. (12). Then, the relationship of the standard normal variable
the PGA where the fragility probability is negligible. Therefore,
Z versus the associated ln(a) value is constructed to retrieve the
the fragility curve is expressed in two parts:
values of ln(mA) and zA.
The simplified curves are displayed as dashed lines in Fig. 15
FA ðaÞ ¼ 0 for a oat ð13  1Þ
and the corresponding fragility parameters are listed in Table 3.
The medians for the damage states I, II and III are 0.21, 0.46 and
Z "  2 #
0.89 g, respectively, and the logarithmic standard deviations are aat
1 1 lnðaat ÞlnmA
0.7, 0.83 and 1.09, respectively. The overall trends of the simpli- FA ðaÞ ¼ pffiffiffiffiffiffi exp  da for a Z at
0 2pBA ðaat Þ 2 BA
fied curves and the originals are fairly good; however, for the
ð13  2Þ
damage states II and III the simplified curves overestimate the
fragility probabilities for the PGA levels of 0.1–0.3 g (the max- where at is the lower bound of the PGA for a lognormal CDF.
imum difference in probability reaches 11% for the damage state II In applying Eq. (13), the lower bound of the PGA for each
when PGA¼0.2 g). damage state is required. Therefore, a finer fragility analysis is
performed by adding more PGA levels in the range of 0.1–0.3 g.
6.5. Simplified fragility curves with a shifted CDF From the trend of the fragility probabilities in this range, the
lower bound values of the PGA for the damage states I, II and III
In the previous section, the simplified fragility curves for the are determined to be 0, 0.12 and 0.18 g, respectively.
damage states II and III overestimate the fragility probabilities at Once the lower bound of the PGA for each damage state is set,
the PGA levels of 0.1–0.3 g. This is because the original fragility a fitting approach similar to that mentioned previously can also
curves for the damage states II and III show very low fragility be applied. Fig. 16 shows the modified curves with the set lower
probabilities (close to zero) in the range of PGA of 0.1–0.2 g. bounds and the corresponding parameters are listed in Table 4.
 J.-S. Chiou et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 830–840
Table 4
Fragility parameters for shifted lognormal CDF.
Damage state Lower bound (g) Median (g) Standard deviation
I 0 0.21 0.72
II 0.12 0.34 0.96
III 0.18 0.67 1.31
It can be seen that the shifted CDF greatly improves upon the
drawback of the original lognormal CDF with no shift.
7. Conclusions
This paper proposes a feasible procedure for developing the
fragility curves of a pile-supported wharf. The main components
and methods of analysis in the process are addressed and an
example wharf is examined for demonstration. The general
conclusions are deduced as follows:
1. In modeling a wharf structure, the nonlinearity of the soil and
the piles should be considered to account for their effects on
the lateral response of the structure. In this study, the non-
linearity of the soil is simulated by applying nonlinear soil
springs, while the nonlinearity of a pile is modeled by placing
distributed plastic hinges in beam elements. The distributed
plastic hinge model can completely capture the development
process of plasticity in piles.
2. The damage sequence of a wharf structure can be clearly
traced from the locations, ranges and degrees of yielded plastic
hinges. The damage criteria of the wharf for different damage
states can be quantified accordingly.
3. The development of fragility curves can be easily performed
with spreadsheet calculations using Microsoft EXCEL. In this
spreadsheet platform, based on the constructed response
matrix and the set damage criteria, the fragility curves of a
wharf structure can be obtained through simple statistical
analysis.
4. For higher damage states, such as II and III, the simplified
fragility curves with the conventional lognormal CDF may
overestimate the fragility probability at the low PGA level. To
improve this issue, the shifted CDF with a lower bound of PGA
is adopted to give a better fit.
Acknowledgements
The authors would like to thank the National Center for
Research on Earthquake Engineering and the Harbor & Marine
Technology Center, Institute of Transportation, Ministry of Trans-
portation and Communication of Taiwan for financial support.
References
[1] Nozu A, Ichii K, Sugano T. Seismic design of port structures. Journal of Japan
Association for Earthquake Engineering 2004;4(3):195–208.
[2] Hwang JH, Yang CW, Chen CH. Investigations on soil liquefaction during the
Chi-Chi earthquake. Soils and Foundations 2003;43(6):107–23.
[3] Donahue MJ, Dickenson SE, Miller TH, Yim SC. Implications of the observed
seismic performance of a pile-supported wharf for numerical modeling.
Earthquake Spectra 2005;21(3):617–34.
[4] McCullough NJ, Dickenson SE, Schlechter SM, Boland JC. Centrifuge seismic
modeling of pile-supported wharves. Geotechnical Testing Journal 2007;
30(5):349–59.
[5] Chang WJ, Chen JF, Ho HC, Chiu YF. In situ dynamic model test for pile-
supported wharf in liquefied sand. Geotechnical Testing Journal 2010;33(3):
212–24.
839
[6] Na UJ, Chaudhuri SR, Shinozuka M. Probabilistic assessment for seismic
performance of port structures. Soil Dynamics and Earthquake Engineering
2008;28:147–58.
[7] Hwang HHM, Jaw JW. Probabilistic damage analysis of structures. Journal of
Structural Engineering 1990;116(7):1992–2007.
[8] Singhal A, Kiremidjian AS. Method for probabilistic evaluation of
seismic structural damage. Journal of Structural Engineering 1996;122(12):
1459–67.
[9] Ramamoorthy SK, Gardoni P, Bracci JM. Probabilistic demand models and
fragility curves for reinforced concrete frames. Journal of Structural Engineer-
ing 2006;132(10):1563–72.
[10] Shinozuka M, Feng MQ, Kim HK, Kim SH. Nonlinear static procedure for
fragility curve development. Journal of Engineering Mechanics 2000;126(12):
1287–95.
[11] Karim KR, Yamazaki F. A simplified method of constructing fragility curves
for highway bridges. Earthquake Engineering and Structural Dynamics 2003;
32:1603–26.
[12] Padgett JE, DesRoches R. Methodology for the development of analytical
fragility curves for retrofitted bridges. Earthquake Engineering and Structural
Dynamics 2008;37:1157–74.
[13] Maruyama Y, Yamazaki F, Mizuno K, Tsuchiya Y, Yogai H. Fragility curves for
expressway embankments based on damage datasets after recent earth-
quakes in Japan. Soil Dynamics and Earthquake Engineering 2010;30:
1158–67.
[14] Tekie PB, Ellingwood BR. Seismic fragility assessment of concrete gravity
dams. Earthquake Engineering and Structural Dynamics 2003;32:2221–40.
[15] Goulet, et al. Evaluation of the seismic performance of a code-conforming
reinforced-concrete frame building-from seismic hazard to collapse safety
and economic losses. Earthquake Engineering and Structural Dynamics
2007;36:1973–97.
[16] Bradley BA, Cubrinovski M, Dhakal RP, MacRae GA. Probabilistic seismic
performance and loss assessment of a bridge–foundation–soil system. Soil
Dynamics and Earthquake Engineering 2010;30:395–411.
[17] Na UJ, Shinozuka M. Simulation-based seismic loss estimation of seaport
transportation system. Reliability Engineering and System Safety 2009;94:
722–31.
[18] Federal Emergency Management Agency (FEMA). HAZUS99 Technical man-
ual, Washington, DC, 1999.
[19] Yeh CH. Taiwan Earthquake Loss Estimation System—TELES. Technical
report, NCREE 03-002, National Center for Research on Earthquake Engineer-
ing, Taipei, Taiwan, 2003 [in Chinese].
[20] Applied Technology Council (ATC). Earthquake damage evaluation data for
California. ATC-13, Redwood City, California, 1985.
[21] Shinozuka M, Feng MQ, Lee J, Naganuma T. Statistical analysis of fragility
curves. Journal of Engineering Mechanics 2000;126(12):1224–31.
[22] Lignos DG, Kolios D, Miranda E. Fragility assessment of reduced beam section
moment connections. Journal of Structural Engineering 2010;136(9):
1140–50.
[23] Hwang HHM, Huo JR. Generation of hazard-consistent fragility curves. Soil
Dynamics and Earthquake Engineering 1994;13:345–54.
[24] Park YJ, Ang AHS. Mechanistic seismic damage model for reinforced concrete.
Journal of Structural Engineering 1985;111(4):722–39.
[25] Au SK, Beck JL. Subset simulation and its application to seismic risk
based on dynamic analysis. Journal of Engineering Mechanics 2003;129(8):
901–17.
[26] Pinto PE, Franchin P, Lupoi A, Lupoi, G. Seismic fragility analysis of structural
systems. In: Fajfar P, Krawinkler H. editors, Proceedings of an international
workshop of performance-based seismic design concepts and implementa-
tions, PEER Rep. no 2004-005, UC Berkeley, 2004.
[27] Cimellaro GP, Reinhorn AM. Multidimensional performance limit state for
hazard fragility functions. Journal of Engineering Mechanics 2011;137(1):
47–60.
[28] Singhai A, Kiremidjian AS. Bayesian updating of fragilities with application to
RC frames. Journal of Structural Engineering 1998;124(8):922–9.
[29] Ghosh J, Padgett JE. Aging considerations in the development of time-
dependent seismic fragility curves. Journal of Structural Engineering
2010;136(12):1497–511.
[30] SAP2000: V11.0-integrated software for structural analysis and design.
Berkeley, CA, USA: Computers and Structures, 2004.
[31] Ministry of Transportation and Communication (MOTC). Design guidelines
for port structures. Taipei, Taiwan, 2000 [in Chinese].
[32] Architecture Institute of Japan (AIJ). Recommendation for the design of
building foundations, 1988 [in Japanese].
[33] Ministry of Transport (MOT). Technical standards for port and harbour
facilities and commentaries. Japan Port and Harbour Association, Japan,
1989 [in Japanese].
[34] Japan Road Association (JRA). Specifications for Highway Bridges, 1990 [in
Japanese].
[35] Chadwell CUC. Fyber: Cross Section Analysis Structural Software, version 2.2.
Department of Civil Engineering. Berkeley: University of California, 1999.
[36] Chiou JS, Yang HH, Chen CH. Use of plastic hinge model in nonlinear
pushover analysis of a pile. Journal of Geotechnical and Geoenvironmental
Engineering 2009;135(9):1341–6.
[37] Applied Technology Council (ATC). Seismic evaluation and retrofit of concrete
buildings. Rep. no. SSC 96-01: ATC-40, 1, Redwood City, California, 1996.
840 J.-S. Chiou et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 830–840
[38] International Navigation Association (PIANC). Seismic design guidelines for
port structures. A.A. Balkema Publishers, 2001.
[39] Luo X, Murono Y, Nishimura A. Verifying adequacy of the seismic deforma-
tion method by using real examples of earthquake damage. Soil Dynamics
and Earthquake Engineering 2002;22:17–28.
[40] Fajfar P. Capacity spectrum method based on inelastic demand
spectra. Earthquake Engineering and Structural Dynamics 1999;28:
979–93.
[41] Chopra AK, Goel RK. Evaluation of NSP to estimate seismic deformation: SDF
systems. Journal of Structural Engineering 2000;126(4):482–90.
[42] Chopra AK, Goel RK. A modal pushover analysis procedure for estimating
seismic demands for buildings. Earthquake Engineering and Structural
Dynamics 2002;31:561–82.
[43] Tian Y, Chaturvedi SK. A seismic retrofit design methodology for R/C bridge
columns using fiber composites. Earthquake Spectra 2004;20(2):483–502.
[44] Casarotti C, Pinho R. An adaptive capacity spectrum method for assessment
of bridges subjected to earthquake action. Bulletin of Earthquake Engineering
2007;5:377–90.
[45] Haldar A, Mahadevan S. Reliability assessment using stochastic finite ele-
ment analysis. John Wiley & Sons, Inc, 2000.