Seismic Performance Evaluation of Facilities:

Methodology and Implementation

T. Y. Yang1; J. Moehle2; B. Stojadinovic3; and A. Der Kiureghian4

Abstract: Performance-based earthquake engineering aims to quantify performance of facilities using metrics that are of immediate use
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 06/29/14. Copyright ASCE. For personal use only; all rights reserved.

to both engineers and stakeholders. A rigorous yet practical implementation of a performance-based earthquake engineering methodology
is developed and demonstrated for an idealized building. The methodology considers seismic hazard, structural response, resulting
damage, and repair costs associated with restoring the building to its original condition, using a fully consistent, probabilistic analysis of
the associated parts of the problem. The methodology can be generalized to consider other performance measures such as casualties and
down time, though these have not been pursued. The proposed procedure is consistent with common building design, construction, and
analysis practices such that it can be readily adopted in earthquake engineering practice today.
DOI: 10.1061/共ASCE兲0733-9445共2009兲135:10共1146兲
CE Database subject headings: Seismic analysis; Earthquake engineering; Evaluation; Facilities.

Introduction framework has not previously been accomplished.

Recognizing that earthquakes occur at random times and loca-
tions with unforeseen magnitudes and shaking frequencies, it is
rational to assess the performance of a facility in a probabilistic
manner. Moehle and Deierlein 共2004兲 presented a framework for
probabilistic seismic assessment that evaluates seismic hazard,
facility response, and the resulting damage and losses. The inher-
ent uncertainties associated with each of these distinct parts are
taken into account using the theorem of total probability. The
performance-assessment results in a quantitative and probabilistic
description of the seismic risk of the facility, expressed in terms
共such as cost, down time, and casualties兲 that stakeholders can use
to make risk management decisions. In addition to being useful to
stakeholders such as facility owners and public officials, the out-
come also is useful to engineers in that it can guide the design
toward optimal solutions considering both immediate and long-
term consequences.
Previous research has developed data, models, and analytical
tools to support the performance-assessment framework 共e.g.,
Moehle and Deierlein 2004兲. Additional studies 共e.g., May
2001, 2006兲 have defined useful performance metrics and societal
implications of performance-based earthquake engineering. These
advances notwithstanding, a complete implementation of the
1
Assistant Professor, Dept. of Civil Engineering, Univ. of British
Columbia, Vancouver, Canada 共corresponding author兲. E-mail:
yangtony2004@gmail.com
2
Professor, Dept. of Civil and Environmental Engineering, Univ. of
California, Berkeley, CA 94720.
3
Professor, Dept. of Civil and Environmental Engineering, Univ. of
California, Berkeley, CA 94720.
4
Professor, Dept. of Civil and Environmental Engineering, Univ. of
California, Berkeley, CA 94720.
Note. This manuscript was submitted on November 15, 2007; ap-
proved on June 3, 2009; published online on September 15, 2009. Dis-
cussion period open until March 1, 2010; separate discussions must be
submitted for individual papers. This paper is part of the Journal of
Structural Engineering, Vol. 135, No. 10, October 1, 2009. ©ASCE,
ISSN 0733-9445/2009/10-1146–1154/$25.00.
In the present study, a procedure to implement the perfor-
mance-assessment framework is developed and demonstrated.
The procedure rigorously considers the seismic hazard, facility
response, resulting damage, and repair costs associated with re-
storing the facility to its original condition. The procedure is
consistent with conventional design, construction, and analysis
practices, such that it can be readily incorporated in engineering
practice. Sample results, based on application of the procedure to
an idealized building, illustrate how expected repair costs and
their distribution among various building components can be used
to guide structural design, investment, and insurance decisions.
Performance-Assessment Framework
The performance assessment is divided into four steps: seismic
hazard analysis, facility response analysis, damage analysis, and
loss analysis as shown in Fig. 1 共Moehle and Deierlein 2004兲. The
term im denotes an intensity measure 共e.g., the peak ground ac-
celeration experienced by the facility兲, edp denotes an engineer-
ing demand parameter 共e.g., an interstory drift ratio兲, dm denotes
a damage measure 共e.g., the amount of cracks in a wall兲, dv
denotes a decision variable 共e.g., a monetary loss, down time, or
casualty count兲, and G共x 兩 y兲 = Pr共x ⬍ X 兩 Y = y兲 denotes the condi-
tional complementary cumulative distribution function of a ran-
dom variable X given a particular outcome Y = y of random
variable Y.
Seismic hazard analysis identifies the seismic hazard at a site.
The outcome of the seismic hazard analysis is a seismic hazard
curve, ␭共im兲, that quantifies the mean annual rate of exceeding a
given value of the seismic intensity measure 共im兲 共e.g., the rate at
which the peak ground acceleration will exceed a specified value
for a particular location in a given year兲. In addition, seismic
hazard analysis is used to identify which types of earthquakes
共e.g., fault mechanisms, magnitudes, and distances兲 are most im-
portant to the seismic hazard at the site; this, in turn, is used to
characterize and select appropriate ground motion records for the
subsequent response analysis.

1146 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / OCTOBER 2009

J. Struct. Eng. 2009.135:1146-1154.


Seismic Hazard Response Damage Loss
Analysis Analysis Analysis Analysis
λ ( im ) G ( edp | im ) G ( dm | edp ) G ( dv | dm )
im: intensity
measure
edp: engineering
demand parameter
dm: damage
measure
dv: decision
variable
Fig. 1. Performance-assessment framework
With ground motion records selected based on the seismic
hazard analysis, the responses of structural and nonstructural
components to seismic excitations are quantified in the response
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 06/29/14. Copyright ASCE. For personal use only; all rights reserved.
analysis step. The outcomes are response statistics that relate en-
gineering demand parameters, such as drift or stress, to the earth-
quake hazards experienced by the facility, measured using the
selected seismic intensity measure.
The third step of the performance-assessment is damage analy-
sis. Damage analysis uses experience gained from laboratory test-
ing, published data, post earthquake reconnaissance reports,
engineering judgment, or a combination of these to quantify the
structural and nonstructural component damage based on an ap-
propriate edp. The outcome of the damage analysis is usually
presented in fragility curves that represent the conditional prob-
ability of occurrence of different damage states given the value of
the edp. Using the edp value obtained from the response analysis,
building structural and nonstructural component damage mea-
sures can be quantified using the fragility curves.
Last, a loss analysis is performed to translate damage quanti-
ties to decision variables that can be used by building owners,
engineers, or other stakeholders to make risk management deci-
sions. This calculation is based on data such as material unit
quantities for repairs, labor costs, repair times, and casualties as-
sociated with damage states defined during damage analysis. The
outputs of the loss analysis can be, for example, the mean rate of
exceeding a certain threshold repair cost for a set period of time,
the cumulative expected monetary loss for repair of the facility
using continuous hazard scenarios, and the total monetary loss for
the facility for a specified hazard level.
The procedure to integrate the components of the
performance-assessment framework is based on the following
simplifying assumptions:
1. G共dm 兩 edp , im兲 = G共dm 兩 edp兲, i.e., for a given edp, dm is sta-
tistically independent of im. In other words, for measuring
damage, im does not provide information that is not already
contained in the edp.
2. G共dv 兩 dm , edp , im兲 = G共dv 兩 dm兲, i.e., for given dm, dv is sta-
tistically independent of edp and im. In other words, for
measuring losses, edp and im do not provide information that
is not already contained in the dm.
3. The facility is immediately restored to its original condition
after each damaging earthquake. Without this assumption,
one would have to account for the state of a damaged struc-
ture being subjected to subsequent earthquakes.
Based on the above assumptions, the mean annual rate of
events 兵dv ⬍ DV其 is derived as follows:
Let ␭共im兲 denote the mean rate of events 兵im ⬍ IM其 per year.
Noting that this function decreases with increasing im,
−d␭共im兲 / dim dim = −d␭共im兲 represents the mean rate of events
兵im ⬍ IM ⱕ im + dim其 per year. Let G共dv 兩 im兲 = Pr共dv ⬍ DV 兩 IM
= im兲 be the conditional probability of DV exceeding dv for given
IM = im. This probability represents the fraction of the events with
兵im ⬍ IM ⱕ im + dim其 that cause DV to be greater than dv. Thus,
the corresponding mean annual rate of events 兵dv ⬍ DV其 is
JOURNAL OF STRUCTURAL ENGINEERING © ASCE / OCTOBER 2009 / 1147
J. Struct. Eng. 2009.135:1146-1154.
␭共dv ⬍ DV 兩 IM = im兲 = −G共dv 兩 im兲d␭共im兲. Integrating over all in-
tensity values, we obtain
冕
␭共dv ⬍ DV兲 = G共dv兩im兲兩d␭共im兲兩 共1兲
im
where we have used the absolute value in account of the negative
value of the derivative of the hazard curve. The conditional prob-
ability G共dv 兩 im兲 is obtained by use of total probability as fol-
lows:
G共dv兩im兲
冕冕
= G共dv兩dm,edp,im兲dG共dm兩edp,im兲dG共edp兩im兲
dm edp
= 冕冕
dm edp
G共dv兩dm兲dG共dm兩edp兲dG共edp兩im兲 共2兲
where use has been made of the assumptions of conditional inde-
pendence described earlier. In the above expression, dG共x 兩 y兲 de-
notes the derivative of the conditional complementary cumulative
distribution function, which is identical to the negative of the
conditional probability density function. Substituting Eq. 共2兲 into
Eq. 共1兲, we obtain the mean annual rate of the decision variable
共e.g., total dollar loss or down time兲 exceeding a threshold value
␭共dv ⬍ DV兲
冕冕冕
= G共dv兩dm兲dG共dm兩edp兲dG共edp兩im兲兩d␭共im兲兩
im dm edp
共3兲
The four stages of performance assessment are evident in the
above formula. Specifically, ␭共im兲 requires hazard analysis,
G共edp 兩 im兲 requires response analysis, G共dm 兩 edp兲 requires dam-
age analysis, and G共dv 兩 dm兲 requires loss analysis. Because no
account is made of the state of a damaged or deteriorated struc-
ture being subjected to subsequent earthquakes, the derivation is
done assuming the facility is nondeteriorating and that it is re-
stored to its original condition immediately after a damaging
earthquake 共Der Kiureghian 2005兲. The result in Eq. 共3兲 is iden-
tical to the formulation presented earlier by researchers in the
Pacific Earthquake Engineering Research Center 共Moehle and
Deierlein 2004; Cornell and Krawinkler 2000兲. However, we be-
lieve the derivation here is clearer and more rigorous.
Implementation of the Performance-Assessment
Framework for Repair Cost Evaluation
The performance assessment framework can serve as the basis for
a rigorous performance assessment. The challenge is to imple-
ment the assessment in a manner that is suitable for practitioners
to use in a design office setting. Two issues must be addressed to
achieve this goal: 共1兲 the performance measures 共dv, dm, and
edp兲 and their conditional probabilities, G共x 兩 y兲, must be quanti-
fied and formulated in a straightforward way using data readily
available to practicing engineers; and 共2兲 the extensive computa-
tions required to integrate Eq. 共3兲 must be encapsulated in proce-
dures that are transparent and easy to implement. An
implementation that fulfills these two goals is presented in the
following six steps: 共1兲 Define structural and nonstructural com-
ponent PGs. 共2兲 Conduct seismic hazard analysis and ground mo-
 the facility. To keep the calculations to a reasonable level, only
those components whose expected damage is likely to contribute
significantly to performance should be considered; those compo-
nents whose contribution to performance is minimal need not be
included.
A sufficient number of damage states 共DSs兲 should be defined
for each PG to describe the range of damage to the components in
that group. Because the repair cost after an earthquake is related
to the repair actions that are required, the DSs should be defined
in relation to the different types of repair that might be required.
For example, DSs for a floor acceleration-sensitive PG may be:
DS1: no damage; DS2: slight damage requiring cosmetic repair;
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 06/29/14. Copyright ASCE. For personal use only; all rights reserved.

DS3: severe but repairable damage; and DS4: total replacement

Fig. 2. Example of fragility curves used to define four damage states
for a drift-sensitive PG. The vertical axis is the probability that the
damage state for the PG is equal to or greater than the damage state
for which a fragility curve is plotted. The probability of being in
damage state DS1 共no damage兲 or worse is not plotted, as that is
always 100%.
tion selection. 共3兲 Evaluate the response of the building. 共4兲
Generate additional correlated edp vectors. 共5兲 Compute the total
repair cost. 共6兲 Generate different expressions of the total repair
cost data.
Define Structural and Nonstructural Component PGs
In this step the various structural and nonstructural components of
the facility are classified into different performance groups 共PGs兲.
Each PG consists of one or more building components whose
performance is similarly affected by a particular engineering de-
mand parameter 共edp兲. For example, one PG might consist of
all similar nonstructural components whose performance is sensi-
tive to interstory drift between the second and third floors of
needed. For each DS, a damage model 共fragility curve兲 defines
the probability that, given an edp value, the damage in compo-
nents of a PG will be equal to or greater than the specified DS.
The differences between the fragility curves then define the prob-
abilities of being in specific DSs. Fig. 2 shows example fragility
curves defined for a drift-sensitive PG. If the peak interstory drift
equals 3%, the conditional probability that a component in this
PG is in DS DS2 or worse is essentially 1.00, in DS3 or worse is
0.72, and DS4 or worse is 0.30. Therefore, by the mathematics of
set theory the probability of being in DS1 共no damage兲, DS2,
DS3, and DS4 is 0.00, 0.28, 0.42, and 0.30, respectively.
After the PGs are identified, building construction data, ob-
tained from as-built documents, in-use surveys, or typical struc-
ture and occupancy categories, are used to quantify the
components of each PG in the facility. For example, square foot-
age of partition walls and number of pocket doors may be deter-
mined.
Because each DS is defined with respect to a repair action
required to correct it, the total repair quantities for each item in a
PG at different DSs can be defined according to the estimate of
damage incurred. Table 1 shows an example of the repair quanti-

Table 1. Example of Damage-Associated Repair Quantities for the PG Shown in Fig. 2

Repair items Units DS1 DS2 DS3 DS4
共a兲 Demolition/access
Finish protection sq ft 0 6,000 6,000 6,000
Ceiling system removal sq ft 0 2,000 3,000 5,000
Drywall assembly removal sq ft 0 800 800 6,000
Miscellaneous MEP locations 0 2 4 6
Remove exterior skin 共salvage兲 sq ft 0 0 0 5,600

共b兲 Repair
Welding protection sq ft 0 1,500 1,500 1,500
Shore beams below and remove locations 0 0 0 12
Cut slab at damaged connection sq ft 0 70 150 1,600
Carbon arc out weld lin. Ft 0 40 50 50
Remove damaged element sq ft 0 0 100 100
Replace weld—from above lin. ft 0 40 40 40
Remove/replace connection Lb 0 0 0 3,000
Replace slab sq ft 0 70 70 1,600

共c兲 Put-back
Misc. MEP and clean-up locations 0 2 4 6
Wall frame 共studs and drywall兲 sq ft 0 800 800 6,000
Replace exterior skin 共salvage兲 sq ft 0 0 0 5,600
Ceiling system sq ft 0 2,000 3,000 5,000

1148 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / OCTOBER 2009

J. Struct. Eng. 2009.135:1146-1154.

 Table 2. Sample edp Matrix
Filename edp 1 edp 2 ¯ edp N
GM 1 2.02 0.75 ¯ 1.05
GM 2 0.98 0.27 ¯ 0.35
] ] ] ] ]
GM M 1.44 0.88 ¯ 1.15
ties at different DSs for the sample PG shown in Fig. 2. Addi-
tional DSs and repair items can be added for different PGs.
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 06/29/14. Copyright ASCE. For personal use only; all rights reserved.
Conduct Seismic Hazard Analysis and Ground Motion
Selection
In this step, a conventional seismic hazard analysis is conducted,
taking into account the site and the layout of the facility. The
seismic hazard analysis quantifies the seismic hazard at the site
and guides selection of suites of ground motion records to be used
in the nonlinear dynamic analysis. For example, a suite of several
ground motion records representing the seismic hazard with a
10% probability of exceedance in 50 years at the site may be
selected. Generally, the records should be from soil types similar
to those at the building site, and should be from earthquakes
having magnitude, faulting mechanism, and distance similar to
those that dominate the shaking hazard at the site. The records
may be manipulated 共e.g., linearly scaled or spectrum matched兲
so that they represent a target seismic hazard level. Procedures for
selecting and manipulating ground motions to represent the seis-
mic hazard are the subject of ongoing research and are beyond the
scope of this work; for additional discussion, see Abrahamson
共2006兲 and Stewart et al. 共2001兲.
Evaluate the Response of the Building
In this step, a structural analysis computer model of the facility is
developed first. The complexity of the model 共inclusion of non-
structural elements in addition to the structural elements兲, the di-
mension of the model 共two- versus three-dimensional兲, and the
type of material and geometric nonlinear behavior considered
must be determined based on engineering judgment of how im-
portant each of these model aspects is to determining the demand
on the facility with sufficient accuracy. Then, the seismic re-
sponse of the facility is determined by conducting a dynamic
response analysis for each of the selected and scaled ground mo-
tion records. Values of edp associated with each PG 共defined in
Step 1兲 are monitored. The peak edp values obtained in the dy-
namic analyses for each ground motion record are summarized in
an edp matrix, as depicted in Table 2. One edp matrix should be
defined for each seismic hazard level considered, i.e., for each
ground motion suite representing the selected seismic hazard
level. The rows of this matrix represent different realizations of
the edp vector, X. Because the ground motions represent indepen-
dent earthquake events, the rows of the matrix represent statisti-
cally independent realizations of X. However, it is important to
realize that the different edp values for a given earthquake are
generally correlated. For example, in a first-mode dominated
structure, all interstory drifts are closely correlated because they
are all proportional to the corresponding interstory drifts of the
first-mode shape. This means that if drift-sensitive components
are damaged in a story, it is likely that similar components are
JOURNAL OF STRUCTURAL ENGINEERING © ASCE / OCTOBER 2009 / 1149
J. Struct. Eng. 2009.135:1146-1154.
ln
x y M Y , DY , R YY
u
exp
w z
Fig. 3. Process of generating the correlated edp vectors
also damaged in stories above or below. Proper account of this
correlation structure is essential for accurate estimation of dam-
age and loss due to earthquakes.
Generate Additional Correlated edp Vectors
Computing a sufficiently large number of edp vector realizations
using dynamic structural analysis is hampered by the paucity of
recorded strong ground motions and by the computational cost.
On the other hand, for accurate estimation of the cost of damage
a large sample of realizations is necessary. Therefore, in the pro-
cedure developed here additional correlated edp vectors are gen-
erated using a simulation method, as described below.
We assume that the vector X has the joint lognormal distribu-
tion. This is a common assumption and supported by available
evidence 共Cornell and Krawinkler 2000兲. It follows that the vec-
tor Y = ln X obtained by taking the natural logarithm of each ele-
ment of X has the joint normal distribution. The mean vector,
MY, and covariance matrix, ⌺YY, of Y are estimated by use of the
data in Table 2.
Using MATLAB 共MathWorks 2006兲, we first compute y
= log共x兲, the natural logarithm of the matrix of realizations of X.
It should be note that MATLAB uses the notation log rather than
ln for the natural logarithm. Once the matrix y has been com-
puted, we then compute the mean of each column, MY
= mean共y兲, where each entry of the mean column represents the
mean of each edp response. The process is then followed by
computing the standard deviation of each edp response and plac-
ing these values on the diagonal of a matrix, DY = diag共std共y兲兲.
This diagonal matrix containing the standard deviation quantities
is then combined with the correlation coefficient matrix of Y,
RYY = corrcoef共y兲, to compute the covariance matrix of Y, ⌺YY
= DYRYYDY.
Having computed the above statistics, we next generate an
artificial set of logarithmic edp values having the above mean and
covariance. We denote these additional vectors as z. One can
show that a jointly normal vector Z having means MY, standard
deviations DY, and correlation matrix RYY is related to a vector of
standard normal random variables U having zero mean and unit
covariance matrix by
Z = D YL YU + M Y 共4兲
where LY is the lower-triangular decomposition of the correlation
matrix RYY so that RYY = LYLYT. Because the correlation matrix is
positive definite, LY can be obtained by Choleski decomposition
of RYY. The MATLAB command to accomplish this operation is
LY = chol共RYY兲.
With the above relations established, artificial samples of Z are
generated by first generating samples of U and then computing
the corresponding realizations of Z from Eq. 共4兲. These simulated
values of Z will have the proper mean, variance, and correlations
 as described above. For completeness, we note that the MATLAB Unit repair cost
command u = normrnd共0 , 1兲 can be used to generate artificial
Max cost
samples of U. Finally, the generated joint normal logarithmic edp Uncertainty
vectors z are transformed into artificially generated edp values, w, $i
Min cost ∞
by taking the exponential of z. This is accomplished with the
MATLAB command w = exp共z兲.
The process of generating the additional correlated edp vectors Min quantities Qi Max quantities Total repair quantities
is illustrated in Fig. 3. Each row of the edp matrix, w, represents
a correlated edp vector from an artificial earthquake at the hazard Fig. 4. Example of the unit repair cost function
level considered in the demand analysis. The process enables the
efficient generation of large numbers of artificial edp vectors that
have the same statistical distribution as the seed data that were Once the total repair quantities are identified, the total repair
generated by the dynamic analyses of the building model 共e.g.,
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 06/29/14. Copyright ASCE. For personal use only; all rights reserved.

cost for the building is computed by multiplying the total repair

Table 2兲. It is recommended that a set of seed data be developed
at different hazard levels, from one that elicits near linear re-
sponse to one corresponding to the most severe shaking likely, so
that changes in the response statistics can be identified and incor-
porated in the cost evaluation process. Studies by the writers have
shown that stable cost estimates can be obtained with as few as
200 realizations at each hazard level, but the computation process
is so efficient that ten times as many realizations can be generated
without noticeable computational expense.
Compute the Total Repair Cost
For each edp realization in X and W, and for each PG 共defined in
Step 1兲, a unique DS is determined. To determine the DS for a
given PG, the fragility relation for the PG 共Fig. 2兲 is entered at an
abscissa equal to the associated edp and an ordinate selected by a
random number generator with a uniform distribution over the
interval 共0,1兲. For example, if the uniform random number gen-
erator produces a number of 0.6 and the edp value 共shown in Fig.
2兲 is 3%, the PG is in DS DS3. If the uniform random number
generator produces a number less than 0.3, the PG is in DS DS4.
Given the DS, the repair quantities for each item in the PG are
located from the repair quantity table 共Table 1兲. This process is
repeated for all PGs and the repair quantities for each item 共e.g.,
finish protection repair quantity in Table 1兲 are summed over all
PGs.
quantity by a unit repair cost and summing over all items. Fig. 4
shows an example of a unit repair cost function, which assumes a
discount for large quantities. Uncertainties in the unit repair cost
can also be considered by making the unit cost a random variable.
Note that the total building repair cost is computed based on
individual realizations of the edp vector, i.e., for one “earth-
quake,” at a selected seismic hazard level. The process is repeated
a sufficiently large number of times using the computed x and
generated w realizations of the edp vector to obtain a distribution
of the total repair cost at the selected hazard level. The process is
then repeated for a different seismic hazard level.
To be able to derive meaningful conclusions about the mean
annual frequency of exceeding a loss value, it is necessary to
analyze the facility for a relatively large number of earthquake
realizations, in each case developing damage and repair costs.
Generation of edp vectors using the procedure outlined in Step 4
instead of running a large number of nonlinear dynamic analyses
makes the proposed implementation feasible as a practical analy-
sis tool.
This implementation of the performance-assessment method-
ology has an embedded assumption that the damage experienced
by different PGs for given values of their respective edp values
are statistically independent. If these damage values are statisti-
cally dependent, the presented methodology can be modified to
account for the dependence by adopting a joint damage distribu-
tion model. These extensions have not been implemented at this
time.

Table 3. Summary of PG Assignment Generate Different Expressions of the Total Repair

No. Location edp a
Components Cost Data
1 Level 1 to Level 2 du1 Application of the performance-assessment methodology to
Structural: evaluate the risk exposure of a facility in a given seismic hazard
2 Level 2 to Level 3 du2
seismic-force-resisting system environment produces data that can be presented in several dif-
3 Level 3 to Level R du3
4 Level 1 to Level 2 du1 ferent ways. Examples include: the probability of exceeding a
Exterior enclosure: repair cost threshold given that ground shaking is at a level cor-
5 Level 2 to Level 3 du2
panels, glass, etc. responding to a specified return period; the mean annual rate of
6 Level 3 to Level R du3
repair costs exceeding a threshold value; and mean cumulative
7 Level 1 to Level 2 du1 Interior nonstructural annual total repair cost. Some of these and others are illustrated in
8 Level 2 to Level 3 du2 drift-sensitive:
an example that follows.
9 Level 3 to Level R du3 partitions, doors, glazing, etc.
10 below Level 2 a2 Interior nonstructural
11 below Level 3 a3 acceleration-sensitive:
Example of the Seismic Performance Evaluation
12 below Level R aR ceilings, lights, sprinkler heads, etc.
Procedure
13 at Level 1 ag Contents: general office on first
14 at Level 2 a3 and second floor, To illustrate the seismic performance evaluation procedure, a
15 at Level 3 aR computer center on third prototype office building located in Berkeley, California was
16 at Level R aR Equipment on roof designed and analyzed. The prototype building 共Fig. 5兲 has six
a bays of steel moment resisting frames in each principal direc-
The notation dui and ai represent the interstory drift ratio at the ith story
and the floor acceleration at the ith floor, respectively. tion 共shown with heavy line兲. Although designed as a three-

1150 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / OCTOBER 2009

J. Struct. Eng. 2009.135:1146-1154.

 edp, and components assigned to each PG. Multiple DSs were
defined for each PG and summarized as fragility relations. Fig. 2

28' ft
shows the fragility relation defined for PG 1, PG 2, and PG 3,

5 @28
which are drift-sensitive structural components; the fragility rela-

5@
tions for the other PGs are presented in Yang 共2006兲. Associated
Z
Y X
repair quantities for each PG at different DSs were prepared.
77@
@ 28'
28 ft Table 1 shows the repair quantities for PG 1, PG 2, and PG 3;
Steel moment resisting frames Steel gravity frames quantities for the other PGs are presented in Yang 共2006兲. Unit
a) Global view b) Plan view repair costs were idealized using the trilinear function shown in
Fig. 4, with parameters summarized in Table 4.
Fig. 5. View of the prototype building Three discrete hazard levels, 50% probability of exceedance
in 50 years 共50/50兲, 10% probability of exceedance in 50 years
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 06/29/14. Copyright ASCE. For personal use only; all rights reserved.

共10/50兲, and 5% probability of exceedance in 50 years 共5/50兲,

dimensional building according to the 2003 International Building
Code 共International Code Council 2003兲, only the response in the
principal X direction is presented here.
Some elements of the performance-assessment process require
engineering judgment. Therefore, the writers engaged experts
working on the Applied Technology Council 58 共2008兲 project
team to develop PGs, quantities, fragility relations, and repair
costs. A total of 16 PGs were defined. Table 3 lists the location,
were selected to study the seismic performance of the prototype
building. Nine representative ground motions were selected for
each hazard level 共Tables 5 and 6兲. The ground motions were then
scaled so the 5% damped spectral acceleration of the ground mo-
tion at the first-mode period of the building was equal to the target
spectral ordinate at the selected hazard level, using the design
spectra from the U.C. Berkeley Seismic Guidelines 共University of
California, Berkley 2003兲.

Table 4. Parameters Used to Define the Unit Repair Cost Function

Minimum Maximum Maximum Minimum
Units quantity cost quantity cost
共a兲 General clean-up
Office papers and books sq ft 1,000 $0.1 10,000 $0.06
Office equipment sq ft 1,000 $0.06 10,000 $0.04
Loose furniture/file drawers sq ft 1,000 $0.05 10,000 $0.03
Water damage sq ft 1,000 $0.15 20,000 $0.1
共b兲 Contents
Computer center sq ft 10,000 $100 50,000 $75
Conventional office sq ft 10,000 $25 50,000 $21
共c兲 Roof-top MEP
Repair in place lin. ft — $10,000 — $10,000
Remove and replace lin. ft — $200,000 — $200,000
共d兲 Structural demolition
Finish protection sq ft 1,000 $0.3 40,000 $0.15
Ceiling system removal sq ft 1,000 $2 10,000 $1.25
Drywall assembly removal sq ft 1,000 $2.5 20,000 $1.5
Miscellaneous MEP location 6 $200 24 $150
Remove exterior skin 共salvage兲 sq ft 3,000 $30 10,000 $25
共e兲 Structural repair
Welding protection sq ft 1,000 $1.5 10,000 $1
Shore beams below and remove location 6 $2,100 24 $1,600
Cut slab at damaged connection sq ft 10 $200 100 $150
Carbon arc out weld lin. ft 100 $15 1,000 $10
Remove damaged beam/column sq ft 100 $80 2,000 $50
Replace weld—from above lin. ft 100 $50 1,000 $40
Remove/replace connection lb 2,000 $6 20,000 $5
Replace slab sq ft 100 $20 1,000 $16
共f兲 Structural put-back
Miscellaneous MEP and clean-up location 6 $300 24 $200
Wall framing 共studs and drywall兲 sq ft 100 $12 1,000 $8
Replace exterior skin 共salvage兲 sq ft 1,000 $35 10,000 $30
Ceiling system sq ft 100 $8 60,000 $5
List continuesa…
a
A complete list of the unit repair cost parameters is presented in Yang 共2006兲.

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / OCTOBER 2009 / 1151

J. Struct. Eng. 2009.135:1146-1154.

 Table 5. Summary of the Ground Motions for 50/50 Hazard Level Table 7. Standard Deviation of the Peak Engineering Demand Param-
eters
Distance Soil
Earthquake Mw Station 共km兲 type Standard Standard Standard
edp Median deviation Median deviation Median deviation
Coyote Lake, 5.7 Coyote Lake, Dam Abutment 4.0 C
共units兲 共50/50兲 共50/50兲 共10/50兲 共10/50兲 共5/50兲 共5/50兲
June 8, 1979 Gilroy #6 1.2 C
Parkfield, 6.0 Temblor 4.4 C du1 共%兲 0.66 0.08 1.40 0.39 1.66 0.65
June 27, 1996 Cholome Array #5 3.7 D du2 共%兲 0.95 0.08 1.89 0.40 2.44 0.62
Cholome Array #8 8.0 D du3 共%兲 1.44 0.34 2.29 0.53 2.63 0.73
Livermore, 5.5 Morgan Territory Park 8.1 C ag 共g兲 0.53 0.24 0.59 0.25 0.76 0.33
January 27, 1980 a2 共g兲 0.90 0.37 0.77 0.35 0.99 0.41
Morgan Hill, 6.2 Coyote Lake, Dam Abutment 0.1 C a3 共g兲 0.76 0.26 0.74 0.30 0.92 0.35
April 24, 1984 aR 共g兲 0.60 0.13 0.64 0.16 0.74 0.17
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 06/29/14. Copyright ASCE. For personal use only; all rights reserved.

Anderson Dam, downstream 4.5 C

Hall Valley 2.5 C

Step 6, are plotted in Fig. 6. The results indicate the median total
An analytical model of the building incorporating inelastic re- repair costs for the prototype building are $0.72 million, $1.13
sponse characteristics for this framing type was implemented million, and $1.85 million dollars for ground shaking at the 50/50,
using OpenSees 共2007兲 and dynamic responses were calculated 10/50, and 5/50 hazard levels, respectively.
for the scaled ground motions. Table 7 lists median and standard The repair cost and the seismic hazard relations can be com-
deviation values of the peak edps. The symbols dui and ai repre- bined to determine the mean annual rate of the repair cost exceed-
sent the interstory drift ratio at the ith story and the total accel- ing a threshold value. This result is obtained by first computing
eration at the ith floor, respectively. the complement of the cumulative distribution function shown in
A MATLAB-based program was developed to carry out the Fig. 6, then multiplying it by the slope of the hazard curve at the
calculations involved in Step 3 through Step 6 of the performance corresponding ground motion intensity level, and finally integrat-
assessment 共Yang 2006兲. Input data include the PGs, the repair ing the resulting curves across the seismic hazard interval consid-
quantity table, the repair cost function, the edp matrices obtained ered in the seismic hazard analysis. Repeating this process for all
by running a limited number of response history analyses, and the repair cost values produces a loss curve that represents the mean
total number of repair cost simulations to compute the loss func- annual rate of the repair cost exceeding a threshold value. Fig. 7
tion. Given these input quantities, the program generates edp re- shows the result for the example building.
alizations, damage, and repair cost loss data, and presents it in a Following the derivation presented by Der Kiureghian 共2005兲,
variety of formats, including those described below. 关The ATC 58 the expected cumulative value of a cost X resulting from all earth-
project has implemented these routines, along with additional rou- quakes over the period of one year is
tines and a user interface in its computer program PACT 共Applied
冕 冕
⬁ ⬁
Technology Council 58 2008兲.兴 E关兺X兴 = x兩d␭共x兲兩 = ␭共x兲dx 共5兲
Table 8 presents a sample of the edp vectors 共realizations兲 0 0
generated by the MATLAB program for the 10/50 hazard level.
The complete set of generated edp vectors has the same statistical where the last expression is obtained from integration by parts.
distribution as the seed data obtained from nonlinear dynamic Thus, the area under the loss curve represents the mean cumula-
analyses. In this example, 10,000 edp vectors were generated in tive total repair cost for all earthquake events in one year time. In
0.070013 s using a 1.4 GHz Intel Pentium M processor with 521 the example presented in Fig. 7, the mean cumulative annual total
MB of RAM. repair cost for this facility is $31,300. This value has relevance to
The cumulative distribution functions of the total repair cost the premium one would be willing to pay to insure the building
for the prototype building, calculated following Step 4 through against the repair cost of future earthquakes.
The repair cost information can be dissected by PGs. This
information can be viewed in terms of bar charts as shown in
Table 6. Summary of the Ground Motions for 10/50 and 5/50 Hazard Fig. 8. Such breakdown provides insight into how different PGs
Levels contribute to the total repair cost and how these contributions
vary with the change of seismic hazard level. Such informa-
Distance Soil
Earthquake Mw Station 共km兲 type
Loma Prieta, 7.0 Los Gatos Present Center 3.5 C Table 8. Sample of Generated edp Vectors for 10/50 Seismic Hazard
October 17, 1989 Saratoga Aloha Ave. 8.3 C Level
Corralitos 3.4 C
edp Number Number Number Number Number
Gavilan College 9.5 C 共units兲 1 2 3 4 ... N
Gilroy Historic Building Unknown C
du1 共%兲 1.26 1.41 1.37 0.97 . . . 1.73
Lexington Dam Abutment 6.3 C
du2 共%兲 1.45 2.05 1.96 1.87 . . . 2.55
Kobe, Japan, 6.9 Kobe JM A 4.4 C
January 17, 1995 du3 共%兲 1.71 2.43 2.63 2.74 . . . 2.46
Tottori, Japan, 6.6 Hino 1.0 C ag 共g兲 0.54 0.55 0.75 0.55 . . . 0.45
October 6, 2000 a2 共g兲 0.87 0.87 1.04 0.92 . . . 0.57
Erzincan, Turkey, 6.7 Erzincan 1.8 C a3 共g兲 0.88 0.77 0.89 1.12 . . . 0.45
March 13, 1992 aR 共g兲 0.65 0.78 0.81 0.75 . . . 0.59

1152 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / OCTOBER 2009

J. Struct. Eng. 2009.135:1146-1154.

 1.0
0.9
P(Total repair cost ≤C) 0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 06/29/14. Copyright ASCE. For personal use only; all rights reserved.
0 1 2 3
Cost, C [dollar]
Fig. 6. CDF of the repair cost distribution
tion indicates what design modifications will be most effective in
reducing repair costs for a structure in a given seismic hazard
environment.
Summary and Conclusions
An evolving approach to performance-based earthquake engineer-
ing aims to express performance of a facility in terms of the risk
of repair cost, down time, and casualties. A performance-
assessment framework is presented that separates the risk-
assessment task into four analysis steps: seismic hazard analysis,
seismic demand analysis, damage analysis, and loss analysis. The
conceptual advantage of such approach is that each analysis step
can be handled by different groups of experts, and separately
developed analysis modules can be assembled as appropriate for
the performance-assessment of a facility. A challenge of the meth-
odology has been to develop a practical procedure for integrating
the four analysis steps to produce a quantitative performance as-
sessment that is both rigorous and practical.
A practical implementation of the performance-assessment
methodology is presented in this paper. The principal innovation
in the proposed implementation is the method for generation of
seismic demand information. Instead of conducting a prohibi-
tively large number of dynamic analyses, data from a smaller
number of analyses are used to establish the characteristics of
demand distribution functions and the correlation structure among
different demand random variables. These distribution and corre-
Loss curve
0.04
Annual rate of exceeding
total repair cost = C
0.03
Mean cumulative
0.02
annual total repair cost
= $ 31,300
0.01
0.00
0 1 2 3
Cost, C [dollar]
Fig. 7. Loss curve generated using the data presented in Fig. 6
1.0 1.0
Probability
Probability
0.5 0.5
0.0 0.0
15 15
0.5 0.5
6 10 10
x10 6
1.0 5 x10 1.0 5
1 1
Cost, C [dollar] Performance Performance
Cost, C [dollar]
group number group number
a) Hazard level = 50/50 b) Hazard level = 5/50
50/50
10/50 Fig. 8. Deaggregation of total repair cost at different hazard levels
5/50
4 5 6 7 lation characteristics are, in turn, used to generate correctly dis-
6
x 10 tributed and correlated damage outcomes using a random number
generator. This innovation enables an implementation of the
performance-assessment framework that is transparent, easily au-
tomated, and highly efficient. As such, it is suitable for use in
engineering practice.
Acknowledgments
This work was supported in part by the Earthquake Engineering
Research Centers Program of the National Science Foundation
under Award No. EEC-9701568 through the Pacific Earthquake
Engineering Research Center 共PEER兲. The performance-
assessment framework, developed by PEER researchers, was in-
troduced by A. Cornell 共Stanford University兲. Concepts of the
performance-assessment implementation were developed in col-
laboration with C. Comartin 共Comartin Associates兲 and M. Com-
erio 共University of California, Berkeley兲. A. Dutta and R.
Hamburger 共Simpson Gumpertz and Heger兲 designed the example
building per code minimum requirements, and R. Bachman, C.
Comartin, G. Hecksher, and A. Whittaker 共ATC-58 project team兲
provided associated fragility, repair method, and repair cost infor-
mation. Any opinions, findings, and conclusion or recommenda-
tions expressed in this work are those of the writers and do not
necessarily reflect those of the organizations or individuals noted
here.
References
Abrahamson, N. A. 共2006兲. “Selecting and scaling accelerograms for dy-
namic analysis.” Proc., 1st European Conference on Earthquake En-
gineering and Seismology 共CD-ROM兲, Swiss Society for Earthquake
Engineering and Structural Dynamics, Geneva.
Applied Technology Council 58. 共2008兲. Development of next-generation
performance-based seismic design procedures for new and existing
buildings, Applied Technology Council, 具http://www.atcouncil.org/
atc-58.shtml典.
Cornell, C. A., and Krawinkler, H. 共2000兲. “Progress and challenges in
seismic performance assessment.” PEER Center News, 具http://peer.
berkeley.edu/news/2000spring/index.html典.
Der Kiureghian, A. 共2005兲. “Non-ergodicity and PEER’s framework for-
mula.” Earthquake Eng. Struct. Dyn., 34, 1643–1652.
International Code Council. 共2003兲. International building code, Interna-
tional Code Council, Falls Church, Va.
MathWorks. 共2006兲. MATLAB—The language of technical computing
4 version—7.0.4, MathWorks, Natick, Mass.
x 10
6 May, P. 共2001兲. “Organizational and societal considerations for
performance-based earthquake engineering.” PEER Rep. No. 2001/04,
Pacific Earthquake Engineering Research Center, Univ. of California,
Berkeley, Calif.
JOURNAL OF STRUCTURAL ENGINEERING © ASCE / OCTOBER 2009 / 1153
J. Struct. Eng. 2009.135:1146-1154.
 May, P. 共2006兲. “Societal implications of performance-based earthquake
engineering.” PEER Rep. No. 2006/12, Pacific Earthquake Engineer-
ing Research Center, Univ. of California, Berkeley, Calif.
Moehle, J., and Deierlein, G. G. 共2004兲. “A framework methodology for
performance-based earthquake engineering.” Proc., 13th World Conf.
on Earthquake Engineering 共CD-ROM兲, Canadian Association for
Earthquake Engineering, Vancouver, Canada.
OpenSees. 共2007兲. “Open system for earthquake engineering simulation
共OpenSees兲 framework—Version 1.7.4.” Pacific Earthquake Engi-
neering Research Center, Univ. of California, Berkeley, 具http://
opensees.berkeley.edu/典.
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 06/29/14. Copyright ASCE. For personal use only; all rights reserved.
1154 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / OCTOBER 2009
J. Struct. Eng. 2009.135:1146-1154.
Stewart, J. P., Chiou, S. J., Bray, J. D., Graves, R. W., Somerville, P. G.,
and Abrahamson, N. A. 共2001兲. “Ground motion evaluation proce-
dures for performance-based design.” PEER Rep. No. 2001/09, Univ.
of California, Berkeley, Calif.
Univ. of California, Berkley. 共2003兲. U.C. Berkeley seismic guideline,
Univ. of California, Berkeley, Calif.
Yang, T. Y. 共2006兲. “Performance evaluation of innovative steel braced
frames.” Ph.D. dissertation, Graduate Division of the Univ. of Cali-
fornia, Berkeley.