1 s2.0 S014102961630654X Main

Engineering Structures 128 (2016) 374399
Contents lists available at ScienceDirect
Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct
Seismic fragility analysis of concrete dams: A state-of-the-art review

Mohammad Amin Hariri-Ardebili, Victor E. Saouma
Department of Civil Engineering, University of Colorado, Boulder, CO UBC 80309-0428, USA
a r t i c l e i n f o a b s t r a c t
Article history: Given the recent impetus for probabilistic based analyses of dams, and the limited previous attempts to
Received 21 April 2016 address this timely question, there is a need for a comprehensive assessment of the disparate previous
Revised 15 September 2016 work. Hence, this paper provides a comprehensive and comparative review of major (over twenty) pub-
Accepted 18 September 2016
lications addressing seismic fragility analyses of concrete dams.
First, fundamental concepts are reviewed and clarified to facilitate comprehension of the later part.
Then, papers are individually scrutinized, key figures redrawn to provide a uniform basis for comparison.
Keywords:
When deemed necessary, additional clarifications and cross referencing with equations in the first part
Fragility function
Concrete dam
are provided.
Earthquake Next, tables summarizing the various methods are presented, on the basis of which the authors provide
Uncertainty quantification a set of minimum requirements for seismic fragility curve/surface development. It is noted that the vast
Damage majority of the papers still relied on linear analyses with simplified limit states. On the other hands few
papers pursued a nonlinear approach and addressed the collapse mechanism and/or hybrid limit state
definitions.
Finally, the contextual framework within which fragility curves are used is presented within the scope
of a performance based earthquake engineering analysis of a concrete dam.
2016 Elsevier Ltd. All rights reserved.
1. Introduction high-hazard dam is reported to have increased by nearly 40%

over the past decade [2]. The Association of State Dam Safety
Dams, a most critical component of our energy generating Officials (ASDSO) reports that by 2020, 70% of the US dams will
infrastructure are aging and their deterioration levels are reach- be over 50 years old [3] and most of them are unlikely to safely
ing critical values. Indeed the American Society of Civil Engineers withstand current design guidelines for potential maximum
(ASCE) 2013 report card for Americas infrastructure [1] catego- floods (PMF) and maximum credible earthquakes (MCE).
rized the US dam hazards as (1) high (potentially causing loss Although few damages/failures have been reported for the con-
of life), (2) significant (economic losses), (3) low, and (4) unde- crete dams, Appendix A, this remains nevertheless a critical soci-
termined. About 14,700, 12,400, 59,000, and 1300 dams fall in etal concern.
each one of those four categories. The geographical distribution Furthermore, due to environmental constraints, few new dams
of the hazardous dams is shown in Fig. 1 and the number of are built, and older ones are expected to have a longer life
Abbreviations: 2D, two-dimensional; 3D, three-dimensional; AD, Anderson-Darling test; AIC, Akaike information criterion; ASCE, American Society of Civil Engineers;
ASDSO, Association of State Dam Safety Officials; ASI, acceleration spectral intensity; CDF, cumulative distribution function; CIA, cumulative inelastic area; CID, cumulative
inelastic duration; CLA, cloud analysis; COV, coefficient of variation; CP, collapse prevention; CS, conditional spectrum; DBL, design base level; DC, damage control; DCR,
demand capacity ratio; DI, damage index; DPM, damage probability matrix; DS, damage state; DSA, double stripe analysis; DSDR, damage spatial distribution ratio; EDP,
engineering demand parameter; EIDA, extended incremental dynamic analysis; EPA, effective peak acceleration; ETA, endurance time analysis; ETAF, endurance time
acceleration function; FEMA, Federal Emergency Management Agency; FFT, fast Fourier transform; FSI, fluid-structure interaction; FSS, factor of safety against sliding; GSI,
geological strength index; H, horizontal; IDA, incremental dynamic analysis; IM, intensity measure; Lg, very large data; LHS, Latin hypercube sampling; LS, limit state; MCE,
maximum credible earthquakes; MCS, Monte Carlo simulation; MDL, maximum design level; MLE, maximum likelihood estimation; MM, material/modeling uncertainty;
MMI, modified Mercalli intensity; MOM, method of moments; MSA, multiple stripe analysis; NLg, no very large data; PBEE, performance based earthquake engineering; PBEE-
2, second generation performance based earthquake engineering; PDF, probability density function; PFMA, potential failure mode analysis; PGA, peak ground acceleration;
PGAH, Horizontal peak ground acceleration; PGV, peak ground velocity; PMF, potential maximum floods; PSA, pseudo-spectral acceleration; PSDA, probabilistic seismic
demand analysis; PSHA, probabilistic seismic hazard analysis; RV, random variable; S, serviceability; SED, specific energy density; SIL, seismic intensity level; SSA, single
stripe analysis; SSE, sum of squared error; UHS, uniform hazard spectrum; V, vertical.
Corresponding author.
E-mail address: saouma@colorado.edu (V.E. Saouma).
http://dx.doi.org/10.1016/j.engstruct.2016.09.034
0141-0296/ 2016 Elsevier Ltd. All rights reserved.
M.A. Hariri-Ardebili, V.E. Saouma / Engineering Structures 128 (2016) 374399 375
Nomenclature
a vector of autocorrelated RVs with zero mean Sa T spectral acceleration at period T

a linear regression constant in CLA Sv T spectral velocity at period T
axconc fluctuations around M 0 Sd T spectral displacement at period T
Aacc Gaussian RV of the amplitude X acc t Sa T; t spectral acceleration at period T and time t of ETAF
A%OS percentage of the overstressed area Sd T; t spectral displacement at period T and time t of ETAF
AC cracked area on dam face S1toN
a combined acceleration response spectra including the
AT total area of dam face effective mass
b linear regression constant in CLA t time
c constant parameter in scattering of artificial ground t trg target time
motion t tot total duration of signal (ground motion)
C collapse trg target value of the considered quantity
C LS capacity parameter associated with the given LS TR return period
d correlation length for spatially random material V Mconc coefficient of variation of material property
dkIM im slope of the hazard curve wj jth vector of the RVs
D demand parameter x structural uncertainty
DIopening joint opening damage index xconc position vector of M conc
DIsliding joint sliding damage index xconc i coordinates of the elements center for the spatially
DIcracking crack-based damage index random material
edp a specific (known) value of EDP X an uncertain random variable
Ec elasticity modulus in concrete X acc t artificially generated seismic excitation using a
Ef elasticity modulus in rock non-stationary stochastic process
EK absolute kinetic energy y randomness of the external actions
ED viscous damping energy Z a vector containing Nele uncorrelated Gaussian RV
ER nonlinear resorting work b logarithmic standard deviation (dispersion)
EP work per seismic applied forces b ^ estimated standard deviation value
EQ absolute seismic input energy bcom logarithmic standard deviation due to combined
EH work done by hydrodynamic pressure uncertainties
ft concrete tensile strength bRTR logarithmic standard deviation due to only ground
fc concrete compressive strength motion record-to-record variability
F 1
i inverse of the cumulative distribution function of the ith bMM logarithmic standard deviation due to material/
RV modeling uncertainty
g i x; y LS function for the ith component dopening joint opening displacement
ht time-dependent deterministic (envelope) function dsliding joint sliding displacement
Hd dam height ed error term in drift capacity model
im a specific (known) value of IM ef error term in capacity model for interface joint tensile
IA arias intensity strength
I wj indicator of safety or failure based on MCS eC errors associate with capacity model
IC j index set for the modes belonging to the jth cut-set g median of the fragility function
L lower triangular matrix obtained by Choleskys g^ estimated median value
decomposition of the covariance matrix gRTR median of the fragility function due to only ground
LC cracked length at the dam base motion record-to-record variability
LT total length at the dam base gcom median of the fragility function due to combined
mDS downstream face slope of the dam uncertainties
M0 mean value of Mconc hSF generic safety factor
M conc random concrete properties j shape parameter of Weibull distribution
M bin ; Rbin magnitude and distance associate with a bin k scale parameter of Weibull distribution
NT arbitrary data points of artificial ground motion k0 scale factor of artificial ground motion
Nele number of elements in the finite element mesh kIM annual rate of ground motion exceedance
Nobs number of observations for curve fitting kLS mean annual frequency of exceeding a specific limit
N1 very large number for MCS state
Nsim total number of simulations pi Nsim random permutation of Nsim
Nfail total number of failed models rt;D tensile normal stress demand
NRV number of (basic) random variables rt;C tensile normal stress capacity
NGM number of (scaled) ground motions sD tangential stress demand
NWL number of pool elevations sC tangential stress capacity
NBSRV number of basic structural random parameters /n phase angles in the interval 0; 2p
NC number of cut-sets in failure evaluation U standard normal CDF
Pf probability of failure xn equally spaced frequencies at the interval 0; xu
PE probability of occurrence of at least one earthquake xu maximum excitation frequency of artificial ground
during the life time motion
PAjB conditional probability that A is true given that B is true jT condition event symbol (given)
PLS limit state probability S intersect symbol
Rck characteristics strength union symbol
Resp response of the system 2 membership symbol
S0 distance between each two points of spatially random
material
376 M.A. Hariri-Ardebili, V.E. Saouma / Engineering Structures 128 (2016) 374399
proposed a hybrid method in which a critical measure of the

resulting probabilistic safety assessment is the fragility curve.
Hence, this paper will focus exclusively on surveying the
reported fragility analysis for concrete dams. But first the funda-
mental underpinnings of the concept will be presented to facilitate
subsequent comprehension of the extensive and comparative sur-
vey. Finally, tables summarizing the various publications will be
presented.
2. Fundamentals
This section begins by providing a succinct explanation of the

general framework within which fragility curves are defined/used,
and then will focus on them subsequently.
Fig. 1. Geographical distribution of hazardous dams in the US [4]. 2.1. Next generation performance based earthquake engineering
Starting with the general concept of next generation perfor-

expectancy. A catalyst for these concerns is the improved under- mance based earthquake engineering (also known as PBEE-2)
standing of natural risks to dams are exposed: hydrological and [12], there are four distinct steps, Fig. 3:
seismic. Hazard Analysis is the first step of PBEE-2. Probabilistic seismic
Last but not least, limited financial resources compel society to hazard analysis (PSHA) in terms of location identification (with
develop new methodologies which better quantify risk in order to respect to a fault), geotechnical conditions (shear wave velocity),
optimize the limited resource distribution for dams not to undergo magnitude of previously recorded earthquakes, size of the rupture
drastic failure as speculated by mainstream literature [5], Fig. 2. area, fault mechanism, crustal rock damping characteristics, and
A premise of this paper is that the current approach based on rock properties. From the corresponding analysis one can deter-
Potential Failure Mode Analysis (PFMA) of dams [79], may not mine annual rate of exceedance of the ground motion amplitude
be sufficient and needs to be broadened. Fortunately, tools are kIM vs. intensity measure (IM) a metric for the ground motion char-
available to address this pressing problem in dams. For the most acteristic. k (inverse of return period T R ) for specific seismic inten-
parts, those have been developed in the building/bridge industry sity level (SIL) is determined from a Poissons probability model:
in California through a new probabilistic paradigm was which
came to be known as Performance Based Earthquake Engineering ln 1 PE
kIM 1
(PBEE) and adopted by Federal Emergency Management Agency t
(FEMA) [10]. Hariri-Ardebili [11] merged PFMA with PBEE and
where P E is the probability of occurrence of at least one event (i.e.
an earthquake) during the life time t.
Structural Analysis From the previous step, seismic excitations
are obtained and dynamic analysis with epistemic (material and
modeling) and/or aleatory (ground motion record-to-record vari-
ability) uncertainties can now be performed. Fig. 4 summarizes
the various levels of complexities for such an analysis. For fragility
analysis, multiple seismic excitations are needed, and this require-
ment can be met through four different approaches:
Multiple stripe analysis (MSA) where first m SILs (or stripes)
are identified and then n ground motion are selected and scaled
for each stripe. The selected ground motions in a stripe may or
may not be identical. However, in general, due to different
response spectra in different stripes the selected ground motions
are different [13]. Typically m P 3 corresponds to MSA, while
m 1 and m 2 are called single- and double- stripe analysis
(SSA and DSA), respectively. Results of a stripe analysis can be plot-
Fig. 2. Popular view of perceived dam safety concern [6]. ted in a IM-EDP coordinate system using discrete data points
Hazard analysis Structural analysis Damage analysis Loss analysis
Hazard model Structural model Fragility model Loss model
Facility
Site hazard Structural response Damage response Loss response
definition D is OK?
IM (im) EDP (edp) DM (dm) DV (dv)
D
Fig. 3. General framework for PBEE-2 [12].

of probabilistic analysis of nuclear power plants [19], in order to
Based Earthquake Engineering

Performance- and Probability-
High
Wide-range Analyses
High
distinguish the task of the structural engineers from the one of
Multiple-stripe Analysis
Very High
Low Incremental Dynamic Analysis
Very High
High
the seismologist. In the general form, the seismic fragility is
Endurance Time Analysis defined as the probability of exceedance of a certain limit state
(LS) for specific level of ground motion IM. Examples of LSs include
Computational Effort
Narrow-range Analyses
Overall Capability
Moderate Single-stripe Analysis Very Good minor damage, moderate and collapse. Those in turn are assessed
Moderate Very Good
Moderate
Double-stripe Analysis
Very Good from either EDP such as stresses, drifts, or from damage index
Cloud Analysis
(DI) such as crack length ratio or plastic deformations. IMs in turn
Nonlinear Static Analysis
Moderate
Moderate Displacement Ductility Method
Very Good
Very Good
are not limited to the usual PGA, as they may include Sa T, specific
Low
Pushover Analysis
Very Good
energy density (SED) or Arias intensities (IA). Hence, fragility is
Nonlinear Time-History Analysis defined as [20]:
Deterministic Analysis
Low Good
Linear Time-History Analysis
Low
Response Spectrum Method
Moderate Fragility PLSjIM im 3
Low Moderate
Equivalent Lateral Force Method
Low Low where P AjB is the conditional probability that A is true given that B
Seismic Coefficient Method
is true, and im refers to a specific value of IM.
Alternatively, fragility can be defined as [21]:
Fig. 4. Progressive analysis methodology for concrete dams [18].
Fragility PD P C LS jIM im 4

(where EDP refers to engineering demand parameter). Within each
stripe the results follow the log-normal distribution assumption. where D is the demand parameter and C LS is the capacity associated
Cloud Analysis (CLA) is a numerical procedure in which first a with the given LS.
dam is subjected to a set of (un-scaled or as-recorded) ground Seismic fragility curve (outcome of the structural analyses) can
motions and is analyzed numerically. If the ground motion records then be combined with the seismic hazard curve, kIM , to obtain the
are taken from a bin, they can represent an earthquake scenario mean annual frequency of exceeding a specific LS, kLS , as [22]:
defined by Mbin ; Rbin , the magnitude and distance representative Z
of the bin [14]. Then, from the results, EDP vs. IM are determined kLS P LSjIM im:jdkIM imj 5
and form the so-called cloud response. CLA method usually is used im
in conjunction with probabilistic seismic demand analysis (PSDA)
[15]. where dkIM im is the slope of hazard curve and the absolute value
Incremental Dynamic Analysis (IDA) considers n different is used because the slope is negative.
ground motions, each of which will be incrementally scaled m Fragility curves can be derived based on one of following four
times until failure occurs. a priori m is unknown and each ground approaches [23]:
motion n will result in a corresponding failure at a different inten-
sity level [16]. Following the analysis, the IDA curve connects the Empirical methods are based on post-earthquake surveys
resulting m demand parameters for each of the n ground motions. which are usually the most reliable source. They are specific
Each one of those curve will be asymptotic to the corresponding to particular sites and seismotectonic/geotechnical conditions
failure. Capture of the overall response by a single measurable and thus, are lack of generality [24].
quantity at a given IM (IM = imi ) can be determined through the Heuristic methods are based on expert elicitation and is useful
corresponding probability distribution. where the empirical information about damage data is very lim-
Endurance Time Analysis (ETA) method starts with a single ited. [25] is an examples for the expert opinion-based fragility
synthetic acceleration function (ETAF) whose amplitude increases curves (in fact damage probability matrix - DPM) proposed for
over time. The increasing amplitude in a single motion substitutes the California infrastructure system (including concrete dam).
for suites of ground motions (as in IDA) where each subsequent Analytical methods are based on structural analyses. Both the
suite has a higher amplitude. Thus, ETA can be interpreted as a static and dynamic analysis methods can be used. This method
dynamic pushover procedure. ETAF is generated such that it satis- is more reliable than two previous ones and reduce the bias.
fies the following relations for any time interval: Hybrid methods are based on combination of the above sources
to reduce the computational efforts and to compensate the sub-
t jective bias [26].
PGAt PGAtrg
ttrg
t trg More details about fragility curves can be found in [2731], and
Sa T; t S T 2 their application in structures (excluding dams) are found in [32].
ttrg a
t trg The next section will indeed provide a detailed and critical litera-
Sd T; t S T ture survey on the applications of fragility curves/surfaces for con-
ttrg d
crete dams.
where, ttrg is target time, usually taken to be 10 s [17], PGA is peak
ground acceleration, Sa T; t and Sd T; t are spectral acceleration 3. Review of the existing applications
and displacement at period T and time t, respectively. Note that
trg refers to the target value of the considered quantity. 3.1. de Arajo and Awruch (1998) [33]
Damage Analysis is discussed in more details below.
Loss Analysis where monetary losses associated with damage To the best of the authors knowledge, [33] is the first published
and loss of life are quantified based on the preceding step. research addressing probabilistic seismic analysis of concrete
dams. Both the concrete properties and the seismic excitation are
2.2. Fragility assessment assumed to be random variable (RV). Monte Carlo Simulation
(MCS) method is used to handle the simultaneous impact of both
Damage analysis seeks to determine fragility curves (or sur- uncertainties. The seismic excitation is artificially generated using
faces). The concept of fragility was first introduced in the context a non-stationary stochastic process, X acc t, for the free-field
motion. It is characterized by two RVs: (1) the amplitude, and (2) 1

the phase angles:
Cumulative probability
X NT 0.8
Aacc atT p
X acc t eaT p
e|{z} cos xn t /n 6
k0 n1
ht 0.6
where Aacc is a Gaussian RV (with a confidence interval of 98%) of the

0.4
amplitude X acc t; k0 is a scaling factor, xn are equally spaced frequen-
cies at the interval 0; xu , where xu is the maximum excitation fre- Observations
quency. /n are the phase angles which are assumed to be 0.2
CDF (normal)
independent RV with a uniform distribution in the interval 0; 2p, CDF (lognormal)
and NT is set to two hundred. Finally, ht is a time-dependent deter- 0
0.1 0.2 0.3 0.4 0.5 0.6 0.7
ministic function (also called envelope function) which includes a Safety factor
single shock; T p ttot 2
; t tot is the total duration of signal. Scattering
of X acc t in time is defined by a ttot c
where c is constant. Fig. 5. CDF of the safety factor against cracking (adapted from [33]).
The random concrete properties, M conc , have a mean M 0 , and a

position vector in the structural domain, xconc :
where X is the uncertain excitation. It is usually represented as an
M conc M 0 1 axconc 7 IM (Eqs. (3) and (4)) however, it may have another form such as
safety factor in this specific example, Fig. 5. U is standard normal
where axconc represents the fluctuations around M0 . It has a mean
CDF, g and b are the median of the fragility function (the X level
value equal to zero and an auto-correlation function as:
with 50% probability of exceedance) and the logarithmic standard
2
RS0 V 2Mconc eS0 =d 8 deviation (also referred to as the dispersion of X). From the method
of moments (MOM), the estimated g and b are [28]:
where S0 is distance between each two points (usually center of ele- s
! PNobs
ments), d correlation length (represents the distance over which the 1 X ^ 2
Nobs
^ i1 ln X i ln g
values of the property exhibit strong correlation), and V Mconc coeffi- g^ exp lnX i ; b 10
Nobs i1 Nobs 1
cient of variation (COV) of material property. Note that if there are
N ele elements in the finite element mesh, it requires to generate N ele
where N obs is number of observations.
different axconc i , where xconc i represents the coordinates at the cen-
T The Anderson-Darling (AD) test is used to determine the most
ter of the element. Then, a vector a a1 ; . . . ; aNele of autocorre- appropriate distribution type [35]. Application of this method
lated RVs with zero mean can be generated as a LZ, where L is requires conversion of the statistical results to a uniform variable,
the lower triangular matrix obtained by Choleskys decomposition p-value 2 [0,1) [36]. The larger p-values for a given distribution, the
of the covariance matrix and Z is a vector containing N ele uncorre- better the fit. Based on the results, the p-value for normal and log-
lated Gaussian RV. normal distributions are 0.38 and 0.81, respectively. Hence, the
The following properties were considered as RV: (1) concrete log-normal CDF is preferable.
compressive strength, (2) concrete tensile strength, (3) concrete
Youngs modulus, and (4) adhesion at the dam-foundation inter-
3.2. Ellingwood and Tekie (2003) [37]
face. However, only the concrete compressive strength was
assumed as basic RV and the others were obtained through deter-
In this study, both material/modeling (MM) and seismic uncer-
ministic correlations with the compressive strength.
tainties are considered and Latin hypercube sampling (LHS) is used
The factor of safety against sliding (FSS), concrete crushing at
to handle the concurrent impact of all uncertainties and to reduce
the toe and concrete cracking at the heel of the dam was computed
the computational effort [38]. In a very general classification, the
for 50 simulations (based on MCS). Defining hSF as a generic safety
sampling can be divided into two groups:
factor, the LS is defined as hSF < 1. Then, if phSF is the probability
density function (PDF), the probability to reach the LS (failure
R1 Crude MCS is a computational algorithm that relies on repeated
probability) is pf 0 phSF dhSF . random sampling. In reliability analysis, an unbiased estimator
Tucuru gravity dam in Brazil was selected as a case study. The of the LS probability can be written as:
height of the dam and pool elevation are 86 and 81 m, respectively. (
The two-dimensional (2D) finite element model (which models 1 X
N1
1 if LS wj P 0
stage construction) accounts for the dam-foundation-reservoir PLS I wj ; I wj 11
N1 j1 0 if LS wj < 0
interaction and a smeared crack model is adopted for cracking.
Only the stiffness-proportional part of damping is considered. where N 1 is a very large number, wj is the jth vector of the RVs,
Safety of the dam is evaluated for an earthquake of only 5 s, and
and I wj is indicator of safety or failure in the system. An esti-
frequencies bracketed by the interval 0; 5p, and a mean amplitude
mation of P LS can be obtained using adequate number of simula-
of 0.l g with a COV of 0.2.
tions, Nsim , where Nfail simulations are failed the LS:
The cumulative distribution function (CDF) of the safety factor
against cracking at the heel are determined and compared with Nfail
the cumulative distribution corresponding to a Gaussian variable. PLS 12
Nsim
Fig. 5 shows the 50 discrete data points and the fitted Gaussian dis-
tribution. Also shown is the log-normal CDF which is fitted to the There are two major issues in crude MCS: (1) A very large num-
empirical data points by the authors of this paper. A fragility curve ber of simulations are necessary to yield reliable results, and (2)
with the log-normal CDF is presented in the form of [34]: sampling is completely random. Fig. 6(a) illustrates this proce-
dure with two RVs (both are assumed to be uniform) and N sim
ln x=g
PLSjX x U 9 equal to eight. As seen, some of the eight horizontal and vertical
b
strata have more than one sample while some others are blank.
Uniform Random Variable 1 Uniform Random Variable 1 ground motion record-to-record variability) uncertainties.
Fig. 7 illustrates two of these methods and the difference:
Method I: This is a simultaneous method, where the number of

Uniform Random Variable 2
Uniform Random Variable 2

ground motion records in each SIL, n should be equal to the
number of models (already sampled and prepared based on
LHS). Then, each of the ground motions are randomly paired
with one of the models and n nonlinear transient analyses are
performed for a specific SIL. Thus, a total of n k analyses are
required, where k is the number of SILs. This method reduces
the computational efforts; however, for small number of n the
(a) Crude MCS application of this method is questionable. In the resulted fragi-
(b) LHS
lity curves (assuming a log-normal CDF), the median and the
Fig. 6. Comparison of MCS and LHS methods for a system with two RVs. logarithmic standard deviation have the combined nature gcom
and bcom , respectively [37].
Method (II): This is a simultaneous method, where the number
To address this potential preference/exclusion of the MCS, the of ground motion records, n, is independent from the number of
LHS has been devised [39]. model, m. Thus, for each SIL, n m analyses are required which
LHS guarantees samples to be drawn over the whole range of results a total of n m k analyses. This method is similar to
the distribution [40]. Having a system with N RV basic RVs, first extended incremental dynamic analysis (EIDA) [41], if it is per-
the range of each variable is split into N sim non-overlapping formed for all the SILs up to failure. This approach is computa-
intervals of equal marginal probability 1=N sim . Then, sampling tionally demanding but the most comprehensive method to
starts with the random selection of an interval followed by combine the uncertainties. Again, the resulted log-normal fragi-
another random selection of a point inside it. The procedure is lity curves have the combined median and the logarithmic stan-
repeated for all RVs: dard deviation, gcom and bcom , respectively.
Method (III): This is a non-simultaneous method, where uncer-
tainties are treated separately and then combined. Ground
pi j 0:5
xj;i F 1
i ; i 1; . . . ; NRV ; j 1; . . . ; Nsim 13 motion record-to-record variability is computed assuming
Nsim m 1 in Method (II) (it means a deterministic model with
all RVs at the mean values). Performing the nonlinear analyses
where pi 1, . . ., pi Nsim is a random permutation of lead to gRTR and bRTR . On the other hand, the logarithmic stan-
dard deviation due to material/modeling uncertainty, bMM , can
1,. . .,N sim ; F 1
i is the inverse of the cumulative distribution func-
be obtained from a set of nonlinear analyses, literature survey,
tion of the ith RV. Note that the resulting samples are uncorre-
or engineering judgment [11]. Finally, the combined dispersion
lated. Fig. 6(b) shows the previous example with LHS: none of
can be obtained as [42]:
the raws and columns have one, and no more than one, selec-
q
tion.In general there are three approaches in order to combine
bcom b2RTR b2MM 14
the epistemic (due to material/modeling) and aleatory (due to
Aleatory Epistemic Aleatory Epistemic

Uncertainty Uncertainty Uncertainty Uncertainty
n ground moons n model n ground moons m model
Seismic Intensity Level (1)
Seismic Intensity Level (1)

...
...
...
...
Seismic Intensity Level (2) Seismic Intensity Level (2)

...
...
1 2
Seismic Intensity Level (k) Seismic Intensity Level (k)

Fig. 7. Comparison of two simultaneous approaches in combination of aleatory and epistemic uncertainties.
Two major assumptions in Eq. (14) are: (1) two sources of uncer- LS2: Foundation material compressive failure at the toe: (1) a
tainties are independent, and (2) the uncertainty only propa- plastic strains of 104 is assumed to be the onset of nonlinearity
gates on the dispersion, b, while the median is kept constant. around the toe, and (2) a plastic strains of 103 dominates the
Note that some recent studies have shown that the epistemic nonlinearity of the foundation.
uncertainty not only increases dispersion but also affects the LS3: Sliding at the dam-foundation interface: (1) slippage of
median response [43,44]. 3 mm is considered to be the onset of sliding, (2) slippage of
13 mm may affects the performance of the drainage system,
The Bluestone dam in West Virginia [45], US was selected as
and (3) slippage of 152 mm may cause differential movements
case study. The height of the non-overflow section is 53 m and
between adjacent monoliths.
the overall crest length is 629 m. The dam consists of rigid blocks
LS4: Deflection at the crest with respect to the heel: two values
situated side-by-side and the shear transfer between them occurs
are considered 8 and 15 mm which are correspond to 0.014%
by friction. The dam is originally designed for 0.1 g horizontal
and 0.028% of the dam height, respectively.
and vertical accelerations. Commercial finite element program
[46] is used for all the simulations. Dam-foundation interface is
Fig. 8 shows the fragility curves at different LSs. The curves are
modeled by a perfectly plastic Mohr-Coulomb friction law. The
based on log-normal CDF, Eq. (9). However, it is important to note
contact surface is allowed to have elastic slip (0.005 in). The
that the b includes both uncertainty types. It is apparent that LS1
recorded free-field ground motions are de-convolueded before
and LS2 are less likely to occur than LS3 and LS4.
they are applied to the base of the foundation. The uplift pressure
is modeled by piecewise linear distribution: heel to drain, drain to
toe. A simplified two-parameter Darbres model (nodal masses in 3.3. Lin and Adams (20072008) [47]
series with dampers) is used for fluid-structure interaction (FSI).
The uncertainties are: The authors presented a set of empirical seismic fragility curves
for concrete dams located in eastern and western Canada [47] [48]
Uniform distribution: drain effectiveness, grout curtain effec- based on [25] (which is originally focused on the damage probabil-
tiveness, tail water elevation, effective uplift area, angle of fric- ities for California). Damage probabilities are expressed in terms of
tion, cohesion, dilation angle of foundation, mass and stiffness a mean DPM. It describes the probability of the facility having a
proportional Rayleigh damping. certain damage state (DS) at a given ground shaking intensity (in
Normal distribution: compressive strength of concrete. term of Modified Mercalli Intensities - MMI), Table 1.
Random Earthquake ground motions: 12 records (based on seis- Given the similarity of design codes, construction methods, and
mic hazard analysis). seismic conditions between California and western Canada, [25]s
DPM is applied. Then, the fragility curves are derived by fitting
A total of N sim 12 samples are drawn based on LHS. Each one the optimal log-normal CDF to discrete points of DPM in Table 1.
is analyzed by one of the 12 selected ground motions at six differ- Fig. 9 illustrates the relation between a group of fragility curves
ent SILs, this leads to totally 72 simulations, Method (I) in Fig. 7. (with three DSs) and a column of DPM for intensity im [49]. Finally,
The intensity levels correspond to the spectral accelerations of the IM parameter is converted from MMI to PGA [50]. The resulting
0.3, 0.5, 0.7, 0.9, 1.0, and 1.2 g. Four LSs, are considered as: fragility curves for western Canada are shown in Fig. 10(a).
Characteristics of the seismic motions in eastern Canada are dif-
LS1: Cracking of the neck: if the tensile stress at the neck ferent from the western one. Thus, the previously derived fragility
exceeds the desired tensile strength. Sa 0:2
curves must be modified. This is achieved through the PGA
ratio,
1 1
LS1: Tensile cracking LS3: Sliding > 0.1 in
Limit state probability
LS1: S22>350 psi LS3: Sliding > 0.5 in

0.8 LS2: Dominant nonlinearity
0.8 LS3: Sliding > 6.0 in
LS2: Onset of nonlinearity LS4: Rel. def. > 0.3 in
0.6 0.6 LS4: Rel. def. > 0.6 in
0.4 0.4
0.2 0.2
0 0
0.4 0.6 0.8 1 1.2 1.4 1.6 0.2 0.4 0.6 0.8 1 1.2 1.4
Sa(T1) [g] Sa(T1) [g]
(a) LSs 1 and 2 (b) LSs 3 and 4
Fig. 8. Seismic fragility curves with four LSs (adapted from [37]).
Table 1
Damage probability matrix for concrete dams (adapted from [25]).
Damage State MMI

VI VII VIII IX X XI XII
None 100 57.5
Slight 42.8 42.5 3.9 0.3
Light 57.5 95.8 88.5 19.3 0.5
Moderate 0.3 11.2 75.2 52.9
Heavy 6.5 46.4
Major 0.2
Destroyed
PGA = 0.10.7 g (7 SILs). Thus, a total of 42 nonlinear transient

analyses are performed.
P (DS >= dsi | IM = im)
P (DS = dsi | im)

Two LSs are defined: (1) crack length at the base, and (2) total
ds0
cracked area on the dam face. To quantify the LSs, energy error bal-
ds1 ance concept is used. This procedure can be summarized as
follows:
ds2
ds3 Find the minimum PGA, PGAmin, for each of the selected ground
im IM 0 1 2 3 dsi motions which causes 5% energy balance error:
Fig. 9. Relationship between continuous fragility curves and DPM. EP EQ EH EK ED ER
Energy balance error %
EQ EH
where T = 0.2 s was arbitrary selected. Median values for PGA and 15
Sa 0:2 for a probability of 2% in 50 years (corresponds to 2475 year where EK is absolute kinetic energy, ED viscous damping energy,
0:2
return period) are used. It was found that the average SaPGA ratio is ER nonlinear resorting work, EP the work of pre-seismic applied
1.94 and 1.66 for the western and eastern regions, respectively. forces, EQ absolute seismic input energy, EH the work done by
PGAE
Thus, for the same Sa 0:2 value, the PGAW
ratio is 1.17 (or roughly hydrodynamic pressure.
1.2). Finally, the fragility curves for the new region (eastern Determine the base crack length and also the total cracked area
Canada) are obtained by simply multiplying the PGA of the previ- of dam for each ground motion at PGAmin.
ous fragility curves by 1.2, Fig. 10(a). Hence, the impact of new Find the minimum value among all the ground motions.
region is considered by simply shifting the fragility curves left or Multiply this value in 0.75 (this is to ensure that the dam will be
right. safe).
Age of the dam (in terms of the different design/construction
periods) was also considered for the fragility curves. The previously Based on their calculations, these two LSs yield 26% of the base
derived fragility curves are thus considered as standard (or refer- length and 1.95% of the total area for this specific dam. Fig. 11
ence) representative of the design/ construction practice in 1950s. shows the resulting fragility curves (labeled as original). It is men-
Old hydro-power components are those designed/ constructed in tioned that these fragility curve are based on a log-normal CDF;
the 1900s. The fragility curves are then obtained by simply divid- however, a closer study reveals that this is not possible, and two
p different log-normal CDFs which have the minimum differences
ing the PGA values of the standard fragility curves by 2. New
with the original ones (labeled as best logN) are determined.
hydro-power components are those designed/built in the 2000s
and their fragility curves are obtained by multiplying the standard
p 3.5. Lupoi and Callari (20112012) [52]
fragility curves by 2. This is shown for light DS in Fig. 10(b).
Fragility curves for any other specific year can be obtained by
interpolation. This new procedure accounts for two uncertainties: material
properties and the external actions (ground motion and reservoir
level) [52,53]. In this method, the randomness of the external
3.4. Mirzahosseinkashani and Ghaemian (2009) [51] actions, y, is treated separately from the structural uncertainty, x.
The state of a ith component is characterized by the following LS
In this study only the ground motion record-to-record variabil- function:
ity is considered and the epistemic uncertainty is neglected in the
analysis of PineFlat dam. The height of which is 122 m and the g i x; y C i x maxDi x; y; t 16
t
reservoir water level is 117 m. A concrete smeared crack model
is used, while the foundation is linear elastic. No interface joint ele- where Di is the structural demand and C i is the corresponding
ment is inserted between dam and the massless foundation, capacity. Failure corresponds to the exceedance of a predefined
whereas pressure-based fluid elements are used for the reservoir. LS and thus, the failure occurs when g i < 0.
For the seismic analyses, 6 near-field ground motions are Since the RVs are separated, the variability in Di induced by y a
selected (with R < 12 km) and only the ground motion horizontal limited number of finite element analyses (for a set of ground
component is applied. Each of the 6 records is scaled from motion records and reservoir water levels) are performed and the
1 1
Damage state probability
Damage state probability
0.8 0.8
East
0.6 0.6
West
0.4 0.4
Slight
Light Old
0.2 0.2 Standard
Moderate
Heavy New
0 0
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
PGA [g] PGA [g]
Fig. 10. Empirical seismic fragility curves for different DSs, locations and construction ages (adapted from [47]).
1 Crest displacement with respect to heel (no threshold value is

LS1: original assigned to this LS).
LS1: best logN
0.8 Cracking at the base or neck (horizontal crack path) governed by
LS2: original
tensile normal stress, rt , at specific node(s) based on: rt;D > rt;C .
LS2: best logN
0.6 Two options are considered:
Single node (also called as series arrangement): which corre-
0.4
sponds to onset of cracking.
Group of nodes on a straight horizontal line (also called par-
allel arrangement): in which the LS is exceeded if this length
0.2
is reached to one third of the dam width at each specific
elevation.
0
0 0.1 0.2 0.3 0.4 0.5 0.6 Sliding at the base which is governed by tangential stress, s, at
PGA [g] specific node(s) based on: sD > sC . Both the above mentioned
conditions should be checked.
Fig. 11. Seismic fragility curves with two LSs (adapted from [51]). Cracking at the upstream face. Identical procedure as for crack-
ing of base/neck.
other random structural properties are kept constant at their mean Two basic RVs, two others for external actions, and two error
value. On the other hand, the uncertainty in the demand due to terms were used, Table 2.
structural parameters, x, is approximated by a linear expansion Ten ground motions (scaled from 0.1 to 0.9 g) and three water
of the demand vector with respect to mean of x: levels (HW 31, 36 and 41 m) were considered. First, the transient
X@Di x; y; t
analyses are performed at PGA = 0.1 g (as a base SIL). Then, assum-
; y; t
Di x; y; t Di x
xj xj 17 ing a linear behavior, the dam response at higher seismic levels are
@x
j j x obtained by scaling up the results calculated for PGA = 0.1 g. The

resulting fragility curves for HW 36 m are shown in Fig. 12. Based
@Di x;y;t
on this figure, series arrangement of components increases the
where @xj
is the Jacobian.

x
probability of failure compared to parallel arrangement.
In this study, there are N GM (scaled) ground motions, N WL pool
The effect of the structural uncertainties is shown in Fig. 12(a)
elevations, and N BSRV basic structural random parameters, thus a
for the system with parallel components. Although the authors,
total of N sim N GM N WL N ST dynamic analysis are required.
stated that the compound curves have lower slope, this is not
The capacity terms are expressed as:
clearly evidenced. Furthermore, no distributional model was fitted
to the empirical data points. Finally, as expected higher pool eleva-
C i x; eC i C i xeCi 18
tions increase the probability of failure at the system-level and
drift-based component fragility curves, but it does not have an
where C i x is determined using semi-empirical formulas (based on
impact on the component-level failure mechanism.
the literature) and eC i refers to the model error and its distributional
type is based on expert opinion.
3.6. Yao, Elnashai and Jiang (2012) [54]
Having the demand and capacity, the system-reliability prob-
lem is characterized in a general cut-set (set of components that
Only ground motion record-to-record variability is considered
leads to system failure) formulation, as:
in this research where the case study is a 305 m high arch dam
8 9 with 72% of filled reservoir. The dam originally included 25 vertical
<[NC \
=
contraction joints; however, only 3 are modeled using commercial
Pf y Pr C i x; eC i 6 Di xjy 19
:j1i2I ; finite element code [46] with nonlinear contact models, while the
Cj
concrete is assumed to be linear elastic. Finally, the hydrodynamic
pressure is modeled using Westergaards added mass approach
where N C is the number of cut-sets and IC j is the index set for the
[55]. A total of 18 ground motions are selected and categorized
modes belonging to the jth cut-set. Finally, the probability of failure
in 3 groups (6 records in each one):
is evaluated by a standard MCS due to its simplicity.
The 46.4 m high Kasho gravity dam is selected, water modeled
Group I: 2 sub-groups where the ground motions are scaled to
by displacement-based fluid elements and the foundation mass
PGA = 0.1 and 0.2 g.
considered along with a bi-linear uplift pressure. All the analyses
Group II: 3 sub-groups where the ground motions are scaled to
being linear elastic, only operational limit states are sought from:
PGA = 0.3, 0.4 and 0.5 g.
Group III: 3 sub-groups where the ground motions are scaled to
PGA = 0.6, 0.7 and 0.8 g.
Table 2
Random variables used by [52]. Thus, a total of 48 scaled ground motions at 8 SILs are used. The
reported maximum sliding and opening at the crest joint are given
Definition (variable) Model type Parameters
in Table 3. The dynamic analysis procedure is somehow analogous
x Characteristics strength (Rck ) Log-normal Rmean ; Rstd
ck ck to both MSA [56] and IDA [57].
Geological strength index (GSI) Uniform (GSImin, GSImax)
Unfortunately, no LS or fragility curves are given. Yet, from the
y Earthquake ground motion Seismic hazard published data a fragility surface could be derived either from Eq.
Reservoir water level Hydrological hazard
(9) (IM-based) or the following equation for an EDP-based
Err Error term in drift capacity Log-normal (1, 0.2) approach [58]:
model (ed )
Error term in capacity model for Log-normal (1, 0.2) lnedp lng
interface joint tensile strength (ef )
PEDP P edpjIM im 1 U 20
b
1 1
System
Probability of failure
Drift deformation
0.8 Base cracking 0.8
Base sliding With Struc. Uncer.
0.6 US face cracking 0.6
Neck cracking
0.4 0.4
Without
Struc. Uncer.
0.2 0.2
0 0
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.2 0.3 0.4 0.5 0.6 0.7 0.8
PGA [g] PGA [g]
(a) Parallel arrangement for components (b) Series arrangement for components
Fig. 12. Seismic fragility curves for HW 36 m (adapted from [52]).
Table 3
Maximum joint sliding/opening at different SILs (in mm) [54].
0.1 g 0.2 g 0.3 g 0.4 g 0.5 g 0.6 g 0.7 g 0.8 g

23/0 24/3 72/32 137/61 252/94 138/44 153/61 161/85
5/0 5/0 30/8 54/13 80/25 37/6 43/9 55/14
5/0 25/1 65/21 110/41 153/63 29/18 40/22 49/25
26/0 26/0 211/82 543/207 1325/418 25/3 35/5 46/8
27/0 28/0 33/10 36/17 62/22 943/279 1733/487 3673/876
5/0 27/0 35/8 58/15 89/20
P[EDP edp|IM] 58/16 81/32 121/60
P[EDP edp|IM]
1 1
0.5 0.5
0 0
1.5 0 1.5 0
1 1
0.5 50 0.5 50
0 100 0 100
IM edp IM edp
Fig. 13. Seismic fragility surfaces for joint response based on raw data from [54].
where g and b are median and the logarithmic standard deviation of Select n ground motion records.
the EDP conditioned on the IM. Assuming a range of edp (not spec- Determine the acceleration response spectra of the selected
ified by the authors) one can determine a fragility surface, Fig. 13, ground motions (Sa vs. T).
from which it is noted that the probability of exceedance of a speci- Analyze the dam under a unit impulse with the magnitude 1
fic im is higher for joint sliding than for opening (mostly at smaller [m/s2] (the dam with a full reservoir is considered and effects
edp values). of static loads are excluded).
Compute the horizontal crest displacement response in the
angular frequency domain using fast Fourier transform (FFT)
3.7. Ycel and Bichini (2013) [59] of the calculated time domain response.
Convert the displacement response to acceleration using
A set of parametric linear elastic analyses of a typical gravity x2 u.
u
dam are conducted and analytical fragility curves computed. 2D Determine the fundamental angular frequency from the first
finite element models were analyzed by the EAGD-84 finite ele- peak and convert it to the vibration period using T 1 x
2p
.
1
ment code [60]. The parametric values are:
Determine the spectral acceleration of n ground motion at the
computed vibration period, Sna T 1 .
Dam height (Hd ): 50, 75, 100, 125 and 150 m.
Perform linear elastic dynamic analyses of the system for n
Downstream face slope (mDS ): 0.70, 0.85 and 1.00.
selected ground motions.
Ratio of elasticity modulus in Concrete to rock (EEc ): 0.02, 1.00
f Extract the response, Respi , corresponds to ground motion i at
and 2.00. the spectral level Sia T 1 . Considering that the system behaves
Concrete tensile strength (f t ): 1.0, 1.5 and 2.0. linearly, find the results for any other spectral level j using
j
S T
Thus, a total of 5 3 3 3 = 135 different finite element Respj Sia T 1 Respi .
a 1
models are generated. In all cases the reservoir is assumed to be Determine the probability of observing visible damage as a
full and the foundation is massless. The following procedure was ratio of number of failed cases to total. The author of the thesis
used to generate fragility curves: did not mention what is the exact definition of the visible
Probability of visible damage
Probability of visible damage

1 1
E /E = 0.02
c f
0.8 E /E = 1.00 0.8
c f
E /E = 2.00 f = 1 MPa f = 2 MPa
c f t t
0.6 0.6
0.4 ft = 1 MPa 0.4

Ec/Ef = 0.02
0.2 0.2 E /E = 1.00

c f
ft = 2 MPa E /E = 2.00
c f
0 0
0 0.2 0.4 0.6 0.8 1 0 1 2 3 4
Sa(T) [g] Sa(T) [g]
(a) mDS =0.7 (b) mDS =1.0
Fig. 14. Seismic fragility curve for a dam with Hd 50 m (adapted from [59]).
damage (it is only referred to significantly exceeding the lin- 1

ear response limits). We suspect that he refers to the demand Minor
Probability of exceedance
capacity ratio (DCR) or cumulative inelastic duration (CID); Medium
0.8
Serious
however, again, the quantitative threshold is not clear. Collapse
Fit a Weibull distribution to the empirical data points: 0.6
Sa j
Pf 1 e k 21 0.4
where k and j 2 0; 1 are scale and shape parameters. It is

0.2
noteworthy that the median for Weibull distribution is
1=j
kln2 . 0
0.2 0.4 0.6 0.8 1 1.2
Twenty real ground motions (excluding vertical component) are PGA [g]
selected and applied to each of the models. Thus, a total of
Fig. 15. Seismic fragility curve for arch dams (adapted from [61]).
20 135 = 2700 linear elastic dynamic analyses are performed
and the fragility curves are derived. Fig. 14 only shows the fragility
curves for a generic dam with Hd 50 m. All results are indeed
intuitive and could have been predicted without such an extensive
loads. A five-level DS for concrete arch dams was proposed, Table 4.
investigation.
The resulting fragility curves are shown in Fig. 15. Log-normal CDF
are used to fit the data points, Eq. (9).
3.8. Zhong, Li and Bao (2013) [61]
Both the uncertainties in ground motion record and concrete 3.9. Abdelhamid et al. (2013) [62]
material are considered using MCS. However, no details about
the number of ground motions, SIL, type of distributional models Both the aleatory and epistemic uncertainties are considered
for RVs (modulus of elasticity, Poissons ratio and concrete damp- simultaneously. The methodology proposed by [37] is adapted to
ing) are given. A 296 m high arch dam is analyzed though a 3D derive the seismic fragility curves of an old gravity dam. A 2D
finite element model with nonlinear concrete model. Considered model is used, concrete is linear elastic, dam-rock interface gov-
loads are: hydrostatic pressure, silt, temperature drop, and seismic erned by Coulombs friction law. The foundation material is
assumed to be a Mohr-Coulomb with the perfectly plastic nonlin-
Table 4
ear behavior. Ground motion is applied at the foundation bottom
Damage state for arch dams [61]. through a deconvolution. LHS is used for sampling of the six uncor-
related RVs:
Level State Description
I No damage The dam remains intact after earthquake with only minor Uniform distribution: angle of friction; cohesion; dilation angle
cracks or spalling of concrete at un-important local areas
which do not affect the functionality of dam
of foundation; elasticity modulus of concrete; and Youngs
II Minor Localized surface cracks occur, the crack lengths are small modulus of soil.
damage and do not daylight. The dam can be restore to its normal Normal distribution: compressive strength of concrete.
functionality after minor repair
III Medium Aside from localized surface cracks, localized penetrating
Four LSs are defined and three threshold (or performance mea-
damage cracks occur at the dam abutment or near the crest. The
dam can restore the normal functionality with medium sures) are assigned to each one:
repair
IV Serious Penetrating cracks propagate in a vast area in vicinity of LS1: Tensile stress at the neck; thresholds: 1.0, 1.5 and 2.0 MPa.
damage abutment and the crest. Cracking is extensive and LS2: Sliding at the dam-foundation interface; thresholds: 5, 10
seriously affect the impounding of the water. Repairing is
difficult
and 20 mm.
V Collapse Cracks extend over 2/3 of the dam with penetrating cracks LS3: Crest relative horizontal displacement to heel; thresholds:
from crest to abutments or even bottom of the dam. The 5, 20 and 40 mm.
dam is collapsed and reservoir water is released LS4: Compressive stress at the dam heel; thresholds: 1, 3 and
6 MPa.
1 1
0.8 0.8
LS4 LS1
0.6 0.6
0.4 0.4
LS3 LS2
0.2 0.2
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
PSA [g] PSA [g]
Fig. 16. Seismic fragility curve (adapted from [62]).
On the other hand, six near-fault ground motions are selected Probability of exceeding each LS (denoted as failure in the orig-
and scaled based on pseudo-spectral acceleration (PSA) from 0.2 inal paper) at different SILs are computed and a log-normal CDF is
to 2.0 g with the increment of 0.2. Thus, a total of 60 nonlinear fitted to the empirical data points. The resulted fragility curves are
transient analyses are performed. The resulting fragility curves shown in Fig. 17. As seen, for most LSs, the near-field ground
are shown in Fig. 16. motions lead to higher probability of failure. Also, comparing the
LSs of the weir (i.e. LS1 and LS2) with those for concrete block
(i.e. LS3 and LS4) show that the failure probability of concrete block
3.10. Ju and Jung (2015) [63]
is higher.
As for [54], only record-to-record variability is considered in

3.11. Ghanaat et al. Series (20112015)
this research applied to a 933.5 m long and 11 m high weir over-
flow with steel reinforcement. A 2D model of the weir and founda-
Ghanaat and his co-authors developed a set of fragility curves
tion is analyzed by a commercial finite element code [46] using
for different concrete dams. Their approach is fundamentally sim-
linear elastic material. The applied loads are: self-weight, hydro-
ilar to the one proposed by [37]; however, there are some subtle
static and hydrodynamic pressures (based on [55] approach), uplift
differences which are addressed below. Both material/modeling
pressure, silt pressure and earthquake loading. For the dynamic
and seismic uncertainties are considered and LHS are used in all
analyses, 30 near-field and 30 far-field ground motion records
cases.
are selected and each one is scaled to 7 PGA levels as: 0.1, 0.2,
0.4, 0.6, 0.8, 1.0 and 1.5 g. Thus, a total of 420 linear time history
3.11.1. Ghanaat et al. (2011) [64]
analyses are performed and five LSs are considered:
In this first publication, seismic performance of Mhlebergs
gravity dam (which includes a weir structure, turbine building,
LS1: Compressive stress at the weir and stilling basin; thresh-
weir
and an administration building) is considered. Only one of the
old = 0.25 f c . eight nearly identical structural parts (11.5 m-wide) of the turbine
LS2: Tensile stress at the weir body and stilling basin; thresh- building was analyzed by LS-DYNA 3D model with sufficient
q
weir
old = 0.42 f c . details to include inlet/outlet, water-retaining gravity section,
LS3: Compressive stress at the mass concrete block; thresh- bridge, turbine building superstructure, foundation rock, headwa-
block ter and tailwater, and a constraint in the cross-stream direction.
old = 0.25 f c .
Nonlinearity stems from nonlinear tiebreak (surface-to-surface)
LS4: Tensile stress at the mass concrete block; threshold = 0.42
q contact elements between dam and foundation. Stress time histo-
block
fc . ries are used as seismic input rather than acceleration time
LS5: Crest displacement; threshold = 10 mm. histories.
Based on a series of deterministic nonlinear analyses it was
determined that the turbine building section controlled the analy-
sis as the median PGA capacities of the weir section and the turbine
1 building are 1.06 and 0.85 g, respectively.
LS1 Combined uncertainty due to record-to-record variability and
LS2 the material randomness was used with 30 trails. Material RVs
0.8 LS3
are: concrete elastic modulus, concrete damping, rock elastic mod-
LS4 Farfield
LS5 ulus, rock cohesion, and rock angle of friction. Except for the last
0.6
Nearfield two RVs, log-normal distribution is assumed. A logic tree analysis
is performed to determine the cumulative distributions of the
0.4
cohesion and friction angle (this method and the corresponding
values were not explained in the original paper).
0.2
For the time history analysis, method (I) in Fig. 7 with 30 three-
component ground motions (n) was used with an initial point
0 PGAH = 0.85 g (capacity of the system from deterministic analysis).
0 0.5 1 1.5 2
PGA [g] Subsequently, two higher intensities 0.95 and 1.10 g and five lower
intensities 0.75, 0.675, 0.60, 0.52, and 0.3872 g were selected. A
Fig. 17. Seismic fragility curves of the weir structure (adapted from [63]). total of 240 nonlinear analyses (k = 8 SILs and 30 trails for each)
1 1
Probability of sliding
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
Base joint
Lift joint
0 0
0 0.5 1 1.5 2 2.5 0 0.2 0.4 0.6 0.8 1
H
PGA [g] PGA [g]
(a) Adapted from [64] (b) Adapted from [66]
1
Probability of sliding failure
0.8
0.6
0.4
Dam (Composite)
Dam (RTR)
0.2
Pier (Composite)
Pier (RTR)
0
0 0.2 0.4 0.6 0.8 1
PGA [g]
(c) Adapted from [67]
Fig. 18. Seismic fragility curves for gravity dams (Ghanaat et al. series).
should have been performed. Given the computational cost of such 3.11.3. Ghanaat et al. (2015) [67]
an endeavor, analyses were terminated when the first convergence In this paper, the authors applied a similar procedure to the
failure (i.e. collapse) occurred, and the concept of possible resur- 332 ft reinforced concrete overflow section of the previous exam-
rection [16,65] at a higher intensity discarded. Though this ple [66]. A detailed 3D finite element model is used with only
approach greatly reduced the computational efforts, starting with the overflow section being nonlinear, and joint elements were
a relatively high initial PGAH only collapse fragility curves are applied at the base of the dam.
derived. Thirty ground motions are selected and scaled to five SILs (0.2,
i
Discrete Pif for each intensity (im ) were first determined 0.4, 0.5, 0.6, and 0.8 g). Each analysis is performed twice: with and
D i i1 E without the following epistemic RVs:
through a moving average approach im im 2
; P if . A log-normal
CDF is then fitted through the data points using least-square Concrete: elastic modulus (Log-normal), compressive strength
approach, Fig. 18(a). It would have been better if additional SILs (Log-normal), tensile strength (Log-normal), maximum aggre-
were considered beyond 1.1 g so as the discrete analyses would gate size (Log-normal).
yield probabilities of failure close to unity. Dam-foundation joint: rock modulus (Log-normal), tensile
strength (Log-normal), cohesion (Triangular), friction angle
(Triangular).
3.11.2. Ghanaat et al. (2012) [66]
Pier reinforcing steel: yield strength (Triangular), Strain-
The previous method was simplified by: (1) reducing the num-
hardening slope (Log-normal), rapture strain (Log-normal).
ber of trials from 30 to 10, (2) ground motions are scaled from 0.05
Others: Drain efficiency (Triangular), damping (Log-normal)
to 1.10 g and also are modified by directionality factors to account
for variability of horizontal and vertical components of the earth-
Simultaneous allocation of the RVs (Method I, Fig. 7) was used,
quake, and (3) using a Weibull distribution (Eq. (21)) in lieu of
and results fitted through a log-normal CDF. Fig. 18(c) shows the
the log-normal one [64]. Then this modified approach was applied
sliding probability with/without epistemic uncertainties. This
again to the tallest non-overflow section of a gravity dam (geomet-
example highlighted the importance of the material uncertainty
rically similar to the Folsom Dam) 336.50 ft high and 50 ft wide. A
on the seismic fragility curves.
constrained (in the cross-stream direction) 3D model was used,
dam-reservoir-foundation modeled as in the previous paper, and
3.12. kadkhodayan et al. (2016) [68]
again the nonlinearity stems from horizontal joint at the dam-
foundation interface and an upper lift joint at the neck.
A nonlinear (induced solely by the contraction joints) analysis
RVs for concrete are: elastic modulus, damping and for rock is
with IDA was performed while the DI was the percentage of over-
simply elastic modulus. For the base and lift joints the RVs are: ten-
sile strength, friction angle, cohesion; and finally drain efficiency. stressed area (A%
OS ) (originally proposed by [69]) on dam faces
Fig. 18(b) shows the probability of sliding based on two failure (upstream and downstream separately considered). Evolution of
modes, i.e. local and global. For the lower seismic intensities, slid- A%
OS vs. IM parameter is explained in Fig. 19. This approach yields
ing probability of the base joint is higher than upper lift joint and a brittle response, in that damage is not incrementally captured
vice versa. for IM < im1 and IM > im2 (as would be the case for a displacement
Fig. 19. Evolution of the overstressed area on dam face under increasing IM.
based one). This problem may not be the best criterion for an IDA, were used and three potential IMs considered: PGA, PGV and
but would be better suited for a MSA [70]. Sa T 1 . On the other hand, three performance levels were iden-
Surprisingly, rather than adopting the usual logarithmic model tified and quantified based on the 5th order regression model,
[71] or cubic polynomial splines [57] for the regression analysis Table 5.
(IM in terms of A%
OS ), a 5th order polynomial equation is used. Such
The potential error in using a normal distribution for fragility
an unusual choice should have been first assessed by Akaike infor- curves is highlighted in Fig. 20 where a non-zero probability of
mation criterion (AIC) [72]. This criterion rewards goodness-of- exceedance at zero IM (e.g. for PGV) is clearly evidenced. This
fitting and penalizes spurious parameters [70]. handicap could be easily alleviated by adoption of a log-normal
A 203 m high arch dam was selected as case study and a distribution model.
3D finite element model of the dam-foundation-reservoir sys- The choice of LS may have also resulted in another idiosyncrasy:
tem was developed [73] while nonlinearity was limited to con- the intersection of two fragility curves. This is not unusual, and was
traction and peripheral joints. Foundation is massless and first reported by [34] who proposed two methods to address it,
linear elastic. Nine three-component ground motion record Fig. 21:
Table 5
Limit state definition for arch dams in [68].
Limit States AUS% ADS% Damage level

OS OS
LS1 Serviceability (S) [012%] <16% Negligible

LS2 Damage control (DC) [1230%] [1636%] Minor
LS3 Collapse prevention (CP) [30100%] [36100%] Major
1 1
0.8 0.8
0.6 0.6
0.4 0.4
LS1: S LS1: S
0.2 0.2
LS2: DC LS2: DC
LS3: CP LS3: CP
0 0
0 0.2 0.4 0.6 0.8 1 0 20 40 60 80 100
PGA [g] PGV [m/s]
Fig. 20. Seismic fragility curve for dam upstream face (adapted from [68]).
Crossing fragility curves Modied; method (1) Modied; method (2)

LS1 LS1 LS1

LS2
LS2 LS2
LS3 LS3 LS3
Intensity measure Intensity measure Intensity measure
Fig. 21. Crossing fragility functions and their modifications (adapted from [34]).
Method (1): For any two log-normal fragility functions i and j Nonlinear analysis: displacements; joint opening damage index

with medians gj > gi and logarithmic standard deviations - DIopening max dopening ; joint sliding damage index -
bi bj which are crossing, Eq. (9) can be replaced with
8 0 19 DIsliding max dsliding ; crack-based damage index -
< ln x=gj =
DIcracking AACT (ratio of the cracked to total area).
Pi LSjX x max U@ A for all j P i 22
j : bj ;
A 203.5 m high arch dam was selected as a case study, the
Method (2): Select only fragility function i and modify its mean model accounted for dam, pressure based reservoir model, mass-
and dispersion: less foundation and analysis performed with [73]. In the nonlinear
analyses, rotating smeared cracks and Mohr-Coulomb based joint
1X N
elements were used. However, the material (epistemic) uncer-
bi 0 b for all i
N i1 i 23 tainty was neglected.
0 The purpose of the linear analysis was to assess the need for a
gi 0 exp 1:28b bi ln gi
nonlinear analysis at a given SIL [74,69]. Proposed criteria for the
limit of acceptability are shown in Table 6 for different dams.
3.13. Hariri-Ardebili and Saouma Series (20152016) Fig. 22(a) illustrates the probability of exceedance of the linear
analysis threshold for three indices (CID = 0.4 s; and DSDR = 20%).
Using an as much comprehensive approach as possible (in Clearly, nonlinear analyses are needed for intensities higher than
terms of nonlinear analysis and probabilistic methods) the authors 0.45 g. The spread of the fragility curves can be explained by the
have developed modern fragility curves/surfaces for concrete dams different spatial coverage of the CID (local) and DSDR (global).
in this series of publications. In Fig. 22(b) results of linear (DSDR) and nonlinear (DIcracking ) are
contrasted for SIL-3. Considering that only nine analyses were per-
3.13.1. Hariri-Ardebili et al. (2015) [70] formed for the SIL-3 stripe, and that many more are needed to
For the first time, MSA was used for concrete dams by the develop a fragility curve, the approach proposed by [75] was pur-
authors. The ground motion record-to-record variability was based sued. This method uses the original limited set of analysis to gen-
on m 3 SILs (DBL: design base level, MDL: maximum design erate multiple EDPs (N sim > 1000) such that the two sets have
level, and intermediary ones) and n 9 ground motion for each identical moments. Based on the original and generated data sets,
resulting in 27 nonlinear transient analyses. For comparison pur- a log-normal CDF was fitted. It is apparent that the two fragility
poses another set of 27 linear elastic analyses were also performed. curves (linear and nonlinear) are nearly identical (since their stan-
Fragility curves for both sets were derived and compared. The fol- dard deviations were originally identical) but one is shifted with
lowing were extracted from each analysis: respect to the other. Hence, one can conclude that for this particu-
lar dam a gross approximation of the nonlinear fragility curve may
Linear elastic analysis: stress; strain; displacements; demand be estimated from the one based on linear analysis by simply
capacity ratio, DCR frt ; cumulative inelastic duration (CID) that dividing the mean value of DSDRrby 2.
is the total duration of stress or strain excursions beyond a
threshold value associated with a DCR; cumulative inelastic
3.13.2. Hariri-Ardebili and Saouma (2016-a) [76]
area (CIA) that is the integral of stress time history for a given
To the best of the authors knowledge this was the first applica-
DCR; Damage spatial distribution ratio (DSDR), ratio of the over-
tion of CLA to the probabilistic analysis of concrete dams. CLA
stressed region to total dam area for a given DCR.
method is particular applicable in the context of such analysis
because the discrete data points are linear in the logarithmic scale
Table 6
Proposed index criteria for linear analysis. [71,77,78]:
Index Quantity ln g IM b: lnIM lna 24

Arch Gravity Buttressed
DCRcr [1, 2] where lna and b are the linear regression constants and g is the
CIDcr 0.4 s 0.3 s 0.4 s median value of EDP given IM.
CIAcr [0.2, 0.4] CLA have been previously used to develop fragility function [42]
DSDRcr 20% 15% 15%
based on Eq. (20) for framed structures. This general equation
1
1
0.9
Probability of Exceedance
Probability of Exceedance
CID
0.8
0.8 DSDR
0.7
DSDR
0.6 0.6
0.5
0.4 0.4
0.3
0.2 0.2
Cracked area
0.1
Overstressed area
0 0
0 0.1 0.2 0.3 0.4 0.5 0 5 10 15 20 25 30
PGA [g] Damage measure parameter [%]
Fig. 22. Seismic fragility curve for an arch dam (adapted from [70]).
(which did not account for lack of convergence, i.e. failure) was the increments, failure may not be properly captured. The
modified by [21] as follows: approach followed in this paper is different. First a ground motion
is properly selected, then scaled to 1.0 g, and then it is this refer-
P EDP P edp j IM P EDP P edp j IM; NLg:1 P Lg j IM ence scaled record which is incrementally multiplied by small
P Lg j IM 25 increments until failure occurs. However, to capture possible res-
urrection [16] an additional analysis is performed. It is believed
where P Lg j IM is the probability of having very large (Lg) EDP for that this fine-grained approach would yield more accurate results.
a given IM. In addition, P EDP P edp j IM; NLg is the fragility func- Two types of ground motion combinations are considered: hor-
tion given no very large (NLg) EDP and IM parameter. izontal only, horizontal and vertical. Furthermore, for each case
The reported study analyzed the tallest non-overflow monolith two loading scenarios are further considered: full and empty reser-
of a 122 m high gravity dam using Merlin [79]. 2D mesh of the dam voirs. Finally, each one of these four cases is subjected to 21 site-
and foundation was provided while the only source of nonlinearity specific actual ground motions scaled by 14 SILs resulting in a
being the interface between the two [80]. It is noteworthy that potential total of 4 21 14 1176 analyses. However, in many
beside gravity and hydrostatic loads, the uplift pressure is auto- cases failure did occur prior to the 13th SIL, and fewer analyses
matically adjusted in terms of crack length. had been performed. For each of the four cases a log-normal CDF
The aleatory uncertainty included by ground motion record-to- is fitted through the discrete data points to determine g ^
^ and b.
record variability only, and the epistemic one ignored. A large
number of ground motions (n 100) was selected using a Method of moments (MOM) where the functional relationship
Matlab-based code [81]. Fragility curves and surfaces were derived and the discrete data points have same moments.
using 70 different IMs and the optimal one was then identified. The Sum of squared error (SSE) based on the minimization of the
optimal ones, as defined by [22], were found to be: (1) structure- sum of the squared errors between functional relationship
dependent spectral IMs (i.e. Sa T 1 ; Sv T 1 and Sd T 1 ), (2) ground and discrete data points.
motion-dependent scalar IM (i.e. PGA and PGV), and (3) Maximum likelihood estimation (MLE) based on the to maxi-
structure-independent spectral IMs (i.e. ASI and EPA) are the best mization of the likelihood function [28].
IMs. However, from a practical and proficiency point of view
[22], the combined acceleration response spectra including the Based on the evaluation procedure suggested by [83], the worst
effective mass, Sa1toN , is the best. fit turned out to be the one based on the simplest method, i.e. MM.
Fig. 23(a) shows the fragility curves for joint opening and slid- On the other hand, the other two approaches (i.e. SSE and MLE)
ing at the dam-rock interface for three different LSs: 2 mm (initia- lead to nearly identical. Fig. 24(a) shows the probability of collapse
tion of opening/sliding), 5 mm (propagation of the opening/ for the four cases. As seen, H + V scenario increases PCjIM with
sliding), and 8 mm (near collapse condition). In all cases, joint slid- respect to H only case. Also, probability of collapse for the full
ing has the highest probability of exceedance. As the LS increases, reservoir is higher than the empty one. Though intuitive, these
the differences between those two fragility curves diminishes. analysis not only confirmed the expectations but also quantified
In so far all the fragility curves are derived for a finite values of them.
the EDP (such as sliding of 3, 5 and 8 mm). However, this can be Whereas most fragility curves are expressed in terms of IMs
generalized for continuous values of EDPs and a fragility surface (specific to a site and a structure), it is preferable to express them
could then be generated, Fig. 23(b). The more challenging problem in terms of EDPs. The advantage being twofolds: (1) curve is less
of developing a fragility curve in terms of two IMs was reported by site specific, and more generic thus potentially applicable to other
[82]. similar structures; and (2) it ties with the PBEE paradigm, Fig. 3.
This is illustrated in Fig. 24(b) where the EDP in this case corre-
sponds to the crest horizontal displacement and the LS have been
3.13.3. Hariri-Ardebili and Saouma (2016-b) [65]
smoothed.
IDA for collapse fragility curves was applied for the first time to
concrete dams. The finite element model, similar to a previous one
[76], with the difference that beside joint nonlinearity, concrete 3.13.4. Hariri-Ardebili and Saouma (2016-c) [84]
nonlinearity is also accounted for through smeared crack model. To the best of the authors knowledge, ETA was for the first time
A word of caution: IDA has been interpreted by some used to derive fragility curves. Since a single ETAF is used, there is
[37,64,66,67] as a preset number of amplifiers of a given earth- no record-to-record variability and thus fragility curves with alea-
quake record, and then analyses are performed on the basis of each tory uncertainty can not be determined. Hence, fragility curves for
one of them. Should failure occur prior to the last one, analyses are epistemic (material/modeling) uncertainty only can be computed.
terminated, on the other hand given the coarse grained nature of The procedure is simple:
1
P[EDP edp|IM]
1
P[EDP edp|IM]
0.8
0.6
Opening = 2 mm 0.5
0.4 Opening = 5 mm
Opening = 8 mm
Sliding = 2 mm 0
0.2 Sliding = 5 mm 0.4 0
Sliding = 8 mm 0.01
0 0.2 0.02
0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.03
IM edp
Sa(T1) [g]
Fig. 23. CLA-based seismic fragility curve and surface for a gravity dam (adapted from [76]).
Probability of collapse
H only (H)
0.8 H + V (H)
0.6
0.4
0.2
0
0.05 0.1 0.15 0.2 0.25 0.3
PGA [g]
Fig. 24. IDA-based seismic fragility curves and surface for a gravity dam (adapted from [65]).
Develop the structural model, select the random variables (with Eq. (14) to obtain a global comprehensive fragility model (Method
possible correlations), and the ETAF. III in page 10), Fig. 25(b).
Select the sampling procedure, i.e. MCS, LHS.
Prepare n inputs for analysis. 3.14. Bernier et al. Series (20152016)
Perform nonlinear analysis of the n models up to failure.
Define appropriate DIs and LSs and extract the fragility curves. 3.14.1. Bernier et al. (2015) [85,86]
The case study is again similar to the previous ones (2D gravity Fragility curves based on 3D finite element analyses of a 78 m
dam). The nonlinearity is originated only from zero-thickness gravity dam-foundation-reservoir system are developed following
interface joint element between the dam and foundation. Random LS-DYNA analyses where cross-canyon displacements are
variables associated with the joint, concrete and rock are listed as restrained. Nonlinearity stems from horizontal joints at dam-
follows: foundation interface and a single lift joint at the neck. Non-
reflecting boundary conditions are applied at the far-end of the
Interface joint [80]: tangential stiffness, normal stiffness, tensile massed foundation and reservoir is modeled using displacement-
strength, cohesion, friction angle, dilatancy angle, specific mode based fluid elements (Lagrangian formulation). Finally, uplift pres-
I fracture energy, and specific mode II fracture energy. sure is modeled by a bilinear distribution and is kept constant dur-
Concrete and rock: concrete modulus of elasticity and Poissons ing the analysis.
ratio, foundation modulus of elasticity. Many synthetic ground motion records were first generated,
and ultimately only 20 were retained: those with smallest sum
In all cases a truncated ([0.5, 1.5] times the mean) normal dis- of squared errors with respect to the National Building Code of
tributional assigned to the RVs. Mean values are based on engi- Canada. Vertical components were also applied by simply scaling
neering judgment, and COV were arbitrarily set at 1020%. LHS the corresponding horizontal ones by a random factor between
was used with 100 simultaneous samples. Two cases were consid- [0.5, 0.8]. Material uncertainties were assigned only to the inter-
ered with and without correlation resulting in a total 200 ETA faces (uniform distribution): cohesion, tensile strength, and angle
based simulations. Finally, fragility curves were develop based on of friction. Finally, the damping ratio was also a RV with log-
log-normal CDF assumption and MLE approach is used for curve normal distribution.
fitting. Fig. 25(a) shows the fragility curves with correlated epis- Using method I (Fig. 7) LHS was adopted and each of those 20
simulations were performed for first mode spectral accelerations
temic uncertainties, the corresponding DI is LLCT and the four selected
at intensities of 0.1, 0.2, 0.3, 0.4, 0.5, 0.7, 0.9, and 1.1 g. Thus, a total
bCorr
LSs are 0.1, 0.3, 0.6 and 0.99. For the uncorrelated RVs bUncorr
varies of 20 8 = 160 simulations were performed.
from 0.85 to 0.94 for different LSs implying that whenever possible Two LSs were considered: sliding at the concrete-to-rock inter-
correlations should be used. face and sliding at the concrete lift joint. Four damage levels,
Finally, the epistemic uncertainty bMM was combined with the adapted from [37], were used with the following damage thresh-
record-to-record variability bRTR , determined from IDA [65] using olds, Fig. 26(a):
1 1
Probability of collapse
0.8 0.8
P[DI LS | IM]
0.6 0.6
0.4 0.4
LS = 0.10
LS = 0.30 0.2
RTR Uncert.
0.2 LS = 0.60 RTR+Mat.(COV=0.1) Uncert.
RTR+Mat.(COV=0.2) Uncert.
LS = 0.99 0
0 0 0.1 0.2 0.3 0.4 0.5 0.6
0 0.05 0.1 0.15 0.2 0.25
PGA [g] S (T ) [g]
a 1
Fig. 25. ETA-based seismic fragility curves with epistemic uncertainty for a gravity dam (adapted from [84]).
Fig. 26. Seismic fragility curve for a gravity dam (adapted from [85,87]).
Minor or slight damage: incipient sliding. foundation-reservoir model using LHS. Probabilistic seismic hazard
Moderate damage: 25 mm sliding (moderate damage to the analyses are performed for 7 SILs (stripes) associated with Sa T 1
drain system with an increase of the uplift pressure). values of 0.1, 0.2, 0.3, 0.4, 0.5, 0.7 and 0.9 g (which corresponds
Severe (extensive) damage: 50 mm sliding (inefficiency of to return period ranges of 35020,000 years). In each level, 20
drains due to severe damage). ground motions are selected, and conditional spectrum (CS) is used
Major damage: 150 mm at the base or 100 mm at the neck instead of the uniform hazard spectrum (UHS). CS represents the
(causing unacceptable differential movements between the expected (mean and variability) response spectrum that has a
adjacent monoliths and loss of reservoir control). specified spectral acceleration at a conditioning period, T 1 in this
case. For any Sa T 1 , each of the 20 samples of the dam-reservoir-
Three types of distributional models, e.g. normal, log-normal foundation model is randomly paired with one of the 20 sets of
and Weibull were used on the raw data based on two fitting meth- horizontal and vertical ground motions selected with the CS
ods: SSE and MLE. Results indicted that SSE-based fitting with log- method for the same Sa T 1 .
normal model was the most suitable one. Finally, a log-normal CDF is used for fitting the discrete data
Also investigated was the variability of the RVs which arises points with the MLE. Fig. 26(c) shows the fragility curves for the
from the heterogeneity of the concrete, rock materials and con- base and neck sliding using MSA. All analyses were subsequently
struction methods. Only the spatial variation of the angle of friction repeated using IDA (selecting one set of ground motions based
in concrete-rock interfaces was scrutinized. Random field theory on UHS and incremental scale factors). Resulting fragility curves
was used to represent the spatial variability and correlation of a for four LSs with and without material/modeling uncertainty are
parameter within the system. An auto-covariance function similar contrasted, Fig. 26(d). A sample comparison for the incipiency of
to Eq. (8) was used. The correlation length for the dam-foundation sliding at the dam-foundation interface. It was determined that
interface was assumed to be equal to base length, half and quarter the median and dispersion parameters of the curves are signifi-
of base length; while for the lift joint at the neck, it is neck length cantly higher using MSA, as this will result in lower probability
and half of it. Fig. 26(b) shows the impact of spatially variation of of exceedance for identical Sa T 1 . Finally, it was shown that the
the angle of friction only on the fragility curves. It is clear that uncertainty related to material/modeling becomes less significant
the effect of spatial variation of angle of friction is minimal for when using the CS method and the fragility curves could be rea-
the three lower damage levels, while a slight difference can be seen sonably estimated by considering the RTR variability as the only
for the most severe one. source of uncertainty.
3.14.2. Bernier et al. (2016) [87] 3.15. Ansari and Agarwal (2016) [88]
Repeating the previous investigations, the authors in here
applied the MSA method. Uncertainties are propagated by again Ignoring material/modeling uncertainties and accounting for
generating 20 statistically significant samples of the dam- only 17 records (horizontal components only), this reported
analyze Long Term Field Monitoring data, and possibly perform

Forced Vibration Testing.
Finite Element Model based on the previous data and Material
Characteristics, determine desirable Features of the analyses and of
the Constitutive Models. Finally, select appropriate F.E. Package.
Initial Deterministic analysis to assess the model, to be fol-
lowed by a set of Parametric Analyses with initially N1 identifiable
unknown parameters, which may be reduced to N2 through cali-
bration with Forced Vibration Testing and Long Term Field
Monitoring.
Sensitivity Analysis Assuming that each of the remaining N2
variables has a minimum and maximum, then 2 N2 1 sensitiv-
ity analyses are performed. The first has all variables set to their
mean value, and then each variable is assigned the minimum or
maximum, one at a time. Results are then displayed in the so-
called Tornado Diagram from which the most sensitive N3 random
variables are selected and kept in further studies.
Epistemic Uncertainties are those associated with quantities
that should be known, but practically impossible to quantify, such
as material property Spatial Distribution [90] or Temporal Uncertain-
ties (e.g. time dependent material degradation). Either one of those
Fig. 27. FSS-based damage state matrix for gravity dams, based on [88]. two may have two or more random variables (such as tensile
strength and compressive strength) and a possible Variable Correla-
tion matrix.
investigation used IDA approach for the structural analyses of a
Aleatory Uncertainties are those due to unsurmountable
122 m high gravity dam. 170 2D finite element (with massless
lack of knowledge such as seismicity at a given site and at a
foundation) analyses were performed using [46]. A concrete dam-
given time. First, a Probabilistic Seismic Hazard Analysis is per-
age plasticity model was used and the weakness of the dam-
formed to determine the hazard curves. Then, corresponding
foundation interface ignored. Westergaard added mass was
Ground Motion Records are selected [91], and finally the Optimal
adopted, while uplift pressure was ignored. Two DIs are used:
Intensity Measure parameter is determined (such as PGA, Sa T 1 )
[71].
DI based on crest displacement and cumulative damage energy
Toward Fragility Finally, outcome of both epistemic and alea-
dissipation.
tory uncertainties are combined, Fig. 7 by performing Monte Carlo
DI based on FSS.
Structural Analyses, Fig. 6. This is by far the most computationally
Note that a comprehensive list of potential DIs for concrete expensive step. Limit States and Potential Failure Modes are then
(arch) dams is provided in [89]. Four DSs were considered: (1) extracted through data mining. Examples of limit states include
slight, (2) moderate, (3) extensive, and (4) severe, Fig. 27. Log- crest displacements and joint opening/sliding. In the last step, Fra-
normal distribution models were used for fragility curves based gility Curves/Surface are derived from the previous through a statis-
on Eq. (9). Fig. 28(a) and (b) shows the derived fragility curves tical interpretation and a cumulative distribution function defined
vs. ground motion IM and crest displacement, respectively. by Eq. (9).
4. Contextual framework 5. Summary
As mentioned in the introduction, Section 1, fragility curves This study sought to better understand and contrast the various
play an essential role in the modern probabilistic risk assessment seismic fragility analyses of concrete dams. Despite the review of
of engineering structures. This is further explained with reference over twenty publications, this is still a nascent research field. As
to Fig. 29. More specifically: such, prior to further growth, it is important to cross-correlate
Site and Dam Characteristics Determine Dam Site Seismicity them. These findings are best summarized in the following three
Map, examine the dam (Physical Model), select Instrumentation, Tables 79.
1 1
0.8 0.8
0.6 0.6
0.4 0.4
Slight Slight
Moderate Moderate
0.2 0.2 Extensive
Extensive
Severe Severe
0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.5 1 1.5 2
PGA [g] Crest displacement [m]
Fig. 28. Seismic fragility curve for a gravity dam (adapted from [88]).
Fig. 29. Outline of the concrete dams fragility analysis.

394
Table 7
Summary of the fragility analysis of concrete dams: Part I.
No Research information Modeling features Analysis type

Research team Year published Dam type Software used Joint NL Material NL FSI SSI Uplift Linear or Name of the
M.A. Hariri-Ardebili, V.E. Saouma / Engineering Structures 128 (2016) 374399

nonlinear method
1 de Araujo and Awruch 1998 Gravity Not specified No Smeared Yes Yes No NL Random
crack
2 Tekie and Ellingwood 2003 Gravity Abaqus Yes No Yes Yes Yes NL IDA
3 Lin and Adams 20072008 All No No No No No Empirical
4 Mirzahosseinkashani 2009 Gravity NSAG-DRI No Smeared Yes Yes No NL IDA
and Ghaemian crack
5 Lupoi and Callari 20112012 Gravity FEAP No No Yes Yes Yes LE Random
6 Yao et al. 2012 Arch Abaqus Yes No Yes Yes No NL MSA-IDA
7 Yucel 2013 Gravity EAGD-84 No No Yes Yes No LE Single dynamic
8 Zhong et al. 2013 Arch Not specified No Yes Yes Yes No NL Random
9 Abdelhamid et al. 2013 Gravity Not specified Yes No Yes Yes No NL Random
10 Ju and Jung 2015 Overflow weir Abaqus No No Yes Yes Yes LE IDA
11 Ghanaat et al. 2011 Turbine building of LS-Dyna Yes No Yes Yes Yes NL Truncated IDA
gravity dam
12 Ghanaat et al. 2012 Non-overflow section LS-Dyna Yes No Yes Yes Yes NL Truncated IDA
(gravity dam)
13 Ghanaat et al. 2015 Overflow section LS-Dyna Yes Yes Yes Yes Yes NL Truncated IDA
(gravity dam)
14 kadkhodayan 2015 Arch Ansys Yes No Yes Yes No NL IDA
15 Hariri-Ardebili et al. 2016 Arch Ansys Yes Yes Yes Yes No LE, NL MSA
16 Hariri-Ardebili and Saouma 2016 Gravity Merlin Yes No Yes Yes Yes NL CLA
17 Hariri-Ardebili and Saouma 2016 Gravity Merlin Yes Yes Yes Yes Yes NL Full IDA
18 Hariri-Ardebili and Saouma 2016 Gravity Merlin Yes No Yes Yes Yes NL ETA
19 bernier et al. 2015 Gravity LS-Dyna Yes Yes Yes Yes Yes NL Truncated IDA
20 bernier et al. 2016 Gravity LS-Dyna Yes Yes Yes Yes Yes NL MSA - IDA
21 Ansari and Agarwal 2016 Gravity Abaqus No Damage Yes Yes Yes NL IDA
plasticity
Table 8
Summary of the fragility analysis of concrete dams: Part II.
No Aleatory (ground motion record-to-record) uncertainty

Multi-component GM directionality GM spatial Number of Number of seismic Post-processing Optimal IM Type of response used Post-processing on EDPs
GM effect effect variability GMs used intensity on IMs selection

effect levels used
1 No No No 50 (artificial) Random, mean = 0.1 g, No No Sliding, Cracking, Crushing Safety factor
COV = 0.2
2 No No No 12 (recorded) 6 No No Deflection, sliding, Tensile and No
compressive stresses
3 No No No Qualitative damage states
4 No No No 6 (near-field) 7 No No Crack length and area No
5 No No No 10 (recorded) 9 No No Deformation, tensile stress Equivalent cracking
6 No No No 18 (recorded) 8 No No Joint opening and sliding No
7 No No No 20 (recorded) No No Principal stresses DCR, CID
8 No No No Not specified Not specified No No Cracked area No
9 No No No 6 (near-field) 10 No No Tensile and compressive No
stresses, deformation, sliding
10 No No No 60 (recorded) 7 No No Tensile and compressive No
stresses, deformation
11 Yes No No 30 (artificial) 8 No No Stresses, displacement, sliding Safety factor
12 Yes Yes No 10 (recorded) Not specified, No No Sliding, displacement
[0.05, 1.10]g
13 Yes No No 30 (recorded) 5 No No Sliding, displacement
14 Yes No No 9 (recorded) Not specified Yes No Stresses Overstressed area
clearly
15 Yes No No 9 (recorded) 3 Yes Yes Stress, strain, displacement, DCR, CID, DSDR
joint opening/sliding, cracked area
16 No No No 100 (recorded) Not applicable Yes Yes Displacement, sliding, joint opening DI
17 Yes No No 21 (recorded) 14 Yes Yes Displacement, sliding, joint opening DI
18 No No No 1 (artificial) Not applicable No No Displacement, sliding, joint opening DI
19 Yes No No 20 (synthetic) 8 No No Base/neck sliding
20 Yes No No 20 (recorded) 7 No No Base/neck sliding
21 No No No 17 (recorded) 10 No No Displacement, energy dissipation Safety factor, DI
395
396
Table 9
Summary of the fragility analysis of concrete dams: Part III.
No Epistemic uncertainty Special features Post-processing

Material/modeling Sampling Type of RVs selected Accounting Accounting Others Total Data Fragility Fragility Function
uncertainty type for water for aging number of mining curve or model fitting
level analyses capability surface method
1 Yes MCS Concrete compressive strength No No 50 No Curve Gaussian Not
specified
2 Yes LHS drain effectiveness, grout curtain effectiveness, tail water elevation, effective uplift area, angle No No 72 No Curve LN Not
of friction, cohesion, dilation angle of foundation, Rayleigh damping, Compressive strength of specified
concrete

3 No No Yes No Curve LN Not
specified
4 No No No 42 No Curve LN Not
specified
5 Yes MCS Characteristics strength, Geological strength index Yes No 60 No Curve Empirical
6 No No No 48 No Not
reported
7 Yes Parametric Dam height, Downstream face slope, Ratio Ec/Ef, Concrete tensile strength No No 2700 Yes Curve Weibull Not
specified
8 Yes MCS Modulus of elasticity, Poissons ratio, concrete damping No No Not No Curve LN Not
specified specified
9 Yes LHS angle of friction, cohesion, dilation angle of foundation, Young modulus of concrete, Young No No 60 No Curve LN Not
modulus of soil, compressive strength of concrete specified
10 No No No RC 420 Yes Curve LN Not
structure specified
11 Yes LHS concrete elastic modulus, concrete damping, rock elastic modulus, rock, cohesion, and rock No No 150 No Curve LN LSE
angle of friction
12 Yes LHS elastic modulus, damping, tensile strength, friction angle, cohesion, drain efficiency No No 100 No Curve Weibull Not
specified
13 Yes LHS elastic modulus, damping, tensile/compressive strength, max aggregate size, rock modulus, No No RC Pier 300 Yes Curve LN Not
friction angle, cohesion, drain efficiency, steel yield strength, strain hardening slope, rupture specified
strain
14 No No No 80 No Curve N Not
specified
15 No No No 54 No Curve LN SSE
16 No No No 100 Yes Curve/ LN MM
Surface
17 No Yes No 1200 Yes Curve/ LN MM, SSE,
Surface MLE
18 Yes LHS For interface joint: tangential stiffness, normal stiffness, tensile strength, cohesion, friction No No Corr RVs 200 No Curve LN MLE
angle, dilatancy angle, specific mode I and II fracture energies; Others: concrete modulus of
elasticity, Poissons ratio, mass density; Foundation modulus of elasticity
19 Yes LHS Cohesion, tensile strength, angle of friction (for C-to-C and C-to-R) No No SV RV 160 No Curve N, LN, LSE, MLE
Weibull
20 Yes LHS Cohesion, tensile strength, angle of friction (for C-to-C and C-to-R) No No CS vs. 140 No Curve LN MLE
UHS
21 No No No 170 No Curve LN Not
specified
Table A.10
Concrete dams shaken by significant earthquake (adapted from [92]).
ID Dam (completed) Type Height (m) Crest (m) Earthquake (date) R (km) Mag. PHGA (g) Notes
1 Lower Crystal Gravity 47 183 San Francisco (Apr 18, 0.4 8.3 0.520.68 (Est.) Not the slightest crack
Springs (1890) 1906)
2 Koyna (1963) Gravity 103 807 Koynanagar (Dec 11, 3.0 6.5 0.5 Cracks in both faces
1967)
3 Williams (1895) Gravity 21 27 Loma Prieta (Oct 17, 9.7 7.1 0.60 (Est.) No damage
1989)
4 Bear Valley (1912, Gravity 28 110 Big Bear (Jun 29, 1992) 14.5 6.6 0.57 No structural damage
1988)
5 Shih Kang (1977) Gravity 21.4 357 Chi Chi (Sep 21, 1999) 0 7.6 0.51 Vertical displ. of 9 m, concrete
rapture
6 Mingtan (1990) Gravity 82 Chi Chi (Sep 21, 1999) 12 7.6 0.40.5 (Est.) No damage
7 Kasho (1989) Gravity 46.4 174 Western Tottori (Oct 6, 38 7.3 0.54 Cracks in control building
2000)
8 Takou (2007) Gravity 77 322 Tohoku (Mar 11, 2011) 109 9.0 0.38 Cracking of gatehouse
9 Miyatoko (1993) Gravity 157 Tohoku (Mar 11, 2011) 135 9.0 0.32 No damage
10 Gibraltar (1920) Arch 52 183 Santa Barbara (Jun 29, 6.3 >0.3 (Est.) No damage
1925)
11 Pacoima (1929) Arch 113 180 San Fernando (Feb 9, 5 6.6 0.60.8 (Est.) Joint opening near thrust block
1971)
Northridge (Jan 17, 18 6.8 0.53 2 Joint opening between arch
1994) and thrust block
12 Ambiesta (1956) Arch 59 145 Gemona-Friuli (May 6, 20 6.5 0.36 (abutment) No damage
1976)
13 Rapel (1968) Arch 111 270 Santiago (Mar 3, 1985) 45 7.8 0.31 Damage to spillway and intake
tower, cracked pavement
Maule (Feb 27, 2010) 232 8.8 0.302
14 Techi (1974) Arch 185 290 Chi-Chi (Sep 21, 1999) 85 7.6 0.5 Local cracking of curb at crest
15 Shapai RCC (2003) Arch 132 250 Wenchuan (May 12, 32 8.0 0.250.5 (Est.) No damage
2008)
16 Hsinfengkiang Buttress 105 440 Reservoir (May 19, 1.1 6.1 0.54 Horiz. cracks in top of dam
(1959) 1962)
17 Sefidrud (1962) Buttress 106 414 Manjil (Jun 21, 1990) 7.7 0.71 (Est.) Horiz. cracks near crest, minor
displ. of blocks
Based on a closer study of these reported investigations, the References

authors feel that the following features should and could be con-
sidered for subsequent studies: [1] ASCE. 2013 Report Card for Americas Infrastructure; Dams, American Society
of Civil Engineers; 2013. <http://www.infrastructurereportcard.org/a/
#p/dams/overview> [last viewed March 2014].
1. Fluid- and soil-structure interaction. [2] US Army Corps of Engineers. CorpsMap; National Inventory of Dams; 2015.
2. structural nonlinearity should capture failure modes (usually a <http://nid.usace.army.mil/cm_apex/f?p=838:5:0::NO> [last viewed August
2015].
discrete joint crack is best). [3] ASDSO. State and federal oversight of dam safety must be improved, Magazine
3. Combination of material and joint nonlinearities is necessary to of Association of State Dam Safety Officials (ASDSO); 2011.
capture possible resurrection in the capacity curves. [4] FEMA-956, Living with dams: know your risks, FEMA P-956. Tech. rep.,
Washington, DC: Federal Emergency Management Agency; 2013.
4. Linear analysis may be adequate for a serviceability fragility
[5] Flynn S. America the resilient; defying terrorism and mitigating natural
curve; however, nonlinear ones should be used for progressive disasters. Foreign Aff 2008;87(2).
failure and collapse fragility curves. [6] Flynn SE. 5 disasters coming soon if we dont rebuild us infrastructure. Pop
5. Material and modeling uncertainties (epistemic) should be Mech; October 2007.
[7] FERC-PFMA. FERC guidance document: poteintial failure mode analysis. Tech.
accounted for within a Monte Carlo simulation family tech- rep., Federal Emergency Regulatory Committee; 2005.
nique (Latin Hypercube sampling are preferable). [8] FEMA-PFM. Selecting analytic tools for concrete dams to address key events
6. Combined temporal and spatial uncertainties have not been along potential failure mode paths. Tech. rep., Colorado: Federal Emergency
Management Agency Denver; 2011.
investigated, and should be studied. [9] USBR-manual. Dam safety risk analysis, best practices training manual, version
7. Ground motion record-to-record variability (aleatory) should 2.2. Tech. rep., Denver, Colorado: U.S. Department of the Interior Bureau of
be accounted for through cloud analysis, incremental dynamic Reclamation in corporation with the U.S. Army Corps of Engineers; 2011.
[10] Applied Technology Council. Seismic performance assessment of buildings
analysis, multiple stripe analysis, or endurance time analysis volume 1 methodology. Tech. Rep. FEMA P-58-1, Federal Emergency
techniques. Management Agency; September 2012.
[11] Hariri-Ardebili M. Performance based earthquake engineering for concrete
dams [Ph.D. thesis]. Boulder, CO: University of Colorado; 2015.
Though the computational cost of such analyses may be [12] Porter K. An overview of PEERs performance-based earthquake engineering
prohibitive, it is certainly achievable with modern computer methodology. In: Proceedings of the 9th international conference on
codes. applications of statistics and probability in civil engineering (ICASP9), San
Francisco, CA.
[13] Cornell A, Jalayer F. Factored nonlinear displacement demand estimation
Appendix A. Reported partial dam damages methods for probability-based safety assessment. In: Annual meeting research
digest No. 2002-7. A Publication of the Pacific Earthquake Engineering
Research Center; 2002.
Given their mass (for gravity dams) and their statical indetermi- [14] Jalayer F. Direct probabilistic seismic analysis: implementing non-linear
nacy (for arch dams) most of them have performed relatively well dynamic assessments [Ph.D. thesis]. Stanford, Palo-Alto, CA: Stanford
when subjected to earthquakes [92], Table A.10. University; 2003.
[15] Shome N. Probabilistic seismic demand analysis of nonlinear structures [Ph.D. [48] Lin L, Adams J. Seismic vulnerability and prioritization ranking of dams in
thesis]. Stanford: Stanford University; 1999. canada. In: Proceedings of the 14th world conference on earthquake
[16] Vamvatsikos D, Cornell C. Incremental dynamic analysis. Earthq Eng Struct engineering, Beijing, China.
Dynam 2002;31:491514. [49] Ioannou I, Rossetto T. Empirical fragility. In: Beer M, Kougioumtzoglou I, Patelli
[17] Nozari A, Estekanchi H. Optimization of endurance time acceleration functions E, Au I, editors. Encyclopedia of earthquake engineering. Berlin,
for seismic assessment of structures. Int J Optim Civ Eng 2011;1:25777. Heidelberg: Springer; 2014. p. 112.
[18] Hariri-Ardebili M, Sattar S, Estekanchi H. Performance-based seismic [50] Worden C, Gerstenberger M, Rhoades D, Wald D. Probabilistic relationships
assessment of steel frames using endurance time analysis. Eng Struct between ground-motion parameters and modified Mercalli intensity in
2014;69:21634. california. Bull Seismol Soc Am 2012;102(1):20421.
[19] Kennedy R, Cornell C, Campbell R, Kaplan S, Perla H. Probabilistic seismic [51] Mirzahosseinkashani S, Ghaemian M. Seismic fragility assessment of concrete
safety study of an existing nuclear power plant. Nucl Eng Des 1980;59:31538. gravity dams. In: Proceedings of the 29th annual USSD conference, Nashville,
[20] Ellingwood B, Kinali K. Quantifying and communicating uncertainty in seismic Tennessee, US.
risk assessment. Struct Saf 2009;31(2):17987. Risk Acceptance and Risk [52] Lupoi A, Callari C. The role of probabilistic methods in evaluating the seismic
Communication Risk Acceptance and Risk Communication. risk of concrete dams. In: Dolek M, editor. Protection of built environment
[21] Jalayer F, Franchin P, Pinto P. A scalar damage measure for seismic reliability against earthquakes. p. 30929.
analysis of rc frames. Earthq Eng Struct Dynam 2007;36:205979. [53] Lupoi A, Callari C. A probabilistic method for the seismic assessment of
[22] Tothong P, Luco N. Probabilistic seismic demand analysis using advanced existing concrete gravity dams. Struct Infrastruct Eng 2012;8:98598.
ground motion intensity measures. Earthq Eng Struct Dynam 2007;36:183760. [54] Yao X, Elnashai A, Jiang J. Analytical seismic fragility analysis of concrete arch
[23] Rossetto T, DAyala D, Ioannou I, Meslem A. Evaluation of existing fragility dams. In: Proceedings of the 15th world conference on earthquake
curves. In: SYNER-G: typology definition and fragility functions for physical engineering, Lisbon, Portugal.
elements at seismic risk. Springer; 2014. p. 4793. [55] Westergaard H. Water pressures on dams during earthquakes. Trans Am Soc
[24] Muntasir Billah A, Shahria Alam M. Seismic fragility assessment of highway Civ Eng 1933;98:41833.
bridges: a state-of-the-art review. Struct Infrastruct Eng 2015;11(6):80432. [56] Jalayer F, Cornell C. Alternative non-linear demand estimation methods for
[25] ATC-13. Earthquake damage evaluation data for california. report no.atc-13. probability-based seismic assessments. Earthq Eng Struct Dynam
Tech. rep., Redwood City, CA: Applied Technology Council; 1985. 2009;38:95172.
[26] Kappos A, Panagopoulos G, Panagiotopoulos C, Penelis G. A hybrid method for [57] Vamvatsikos D, Cornell C. Applied incremental dynamic analysis. Earthq
the vulnerability assessment of R/C and URM buildings. Bull Earthq Eng Spectra 2004;20:52353.
2006;4(4):391413. [58] Jalayer F, De Risi R, Manfredi G. Bayesian cloud analysis: efficient structural
[27] Shinozuka M, Feng M, Lee J, Naganuma T. Statistical analysis of fragility curves. fragility assessment using linear regression. Bull Earthq Eng 2015;13
J Eng Mech 2000;126:122431. (4):1183203.
[28] Baker J. Efficient analytical fragility function fitting using dynamic structural [59] Ycel A. Seismic analysis of concrete gravity dams including dam-foundation-
analysis. Earthq Spectra 2015;31(1):57999. reservoir interaction [Masters thesis]. Turkey: Middel East technical
[29] Lallemant D, Kiremidjian A, Burton H. Statistical procedures for developing University; 2013.
earthquake damage fragility curves. Earthq Eng Struct Dynam [60] Fenves G, Chopra A. EAGD-84: A computer program for earthquake analysis of
2015;44:137389. http://dx.doi.org/10.1002/eqe.2522. concrete gravity dams. University of California, Earthquake Engineering
[30] Porter K. A beginners guide to fragility, vulnerability, and risk. In: Beer M, Research Center; 1984.
Kougioumtzoglou IA, Patelli E, Au IS-K, editors. Encyclopedia of earthquake [61] Zhong H, Li H, Bao Y. Seismic risk analysis of an arch dam. Applied mechanics
engineering. Berlin, Heidelberg: Springer; 2015. p. 129. http://dx.doi.org/ and materials, vol. 353356. Trans Tech Publ; 2013. p. 20203.
10.1007/978-3-642-36197-5_256-1. [62] Abdelhamid H, Mahmoud B, Hussein M. Seismic fragility and uncertainty
[31] Sudret B, Mai C, Konakli K. Assessment of the lognormality assumption of analysis of concrete gravity dams under near-fault ground motions. Civ
seismic fragility curves using non-parametric representations. Struct Saf Environ Res 2013;5:1239.
2015:132. [63] Ju B, Jung W. Evaluation of seismic fragility of weir structures in South Korea.
[32] Pitilakis K, Crowley H, Kaynia A. SYNER-G: typology definition and fragility Math Prob Eng 2015;2015:10. http://dx.doi.org/10.1155/2015/391569.
functions for physical elements at seismic risk. Geotech Geol Earthq Eng [64] Ghanaat Y, Hashimoto P, Zuchuat O, Kennedy R. Seismic fragility of Mhleberg
2014;27. dam using nonlinear analysis with latin hypercube simulation. In: Proceeding
[33] de Arajo J, Awruch A. Probabilistic finite element analysis of concrete gravity of the 2011 USSD annual conference. p. 1197212.
dams. Adv Eng Softw 1998;29:97104. [65] Hariri-Ardebili M, Saouma V. Collapse fragility curves for concrete dams:
[34] Porter K, Kennedy R, Bachman R. Creating fragility functions for performance- comprehensive study. ASCE Struct Eng 2016;142(10). http://dx.doi.org/
based earthquake engineering. Earthq Spectra 2007;23:47189. 10.1061/(ASCE)ST.1943-541X.0001541.
[35] Anderson T, Darling D. A test of goodness of fit. J Am Stat Assoc [66] Ghanaat Y, Patev R, Chudgar A. Seismic fragility analysis of concrete gravity
1954;49:7659. dams. In: Proceedings of the 15th world conference on earthquake
[36] Marsaglia G, Marsaglia J. Evaluating the Anderson-Darling distribution. J Stat engineering, Lisbon, Portugal.
Softw 2004;9:15. [67] Ghanaat Y, Patev R, Chudgar A. Seismic fragility for risk assessment of concrete
[37] Tekie P, Ellingwood B. Seismic fragility assessment of concrete gravity dams. gravity dams. In: Proceeding of the 2015 USSD annual conference. p. 64560.
Earthq Eng Struct Dynam 2003;32:222140. [68] Kadkhodayan V, Aghajanzadeh M, Mirzabozorg H. Seismic assessment of arch
[38] Tekie P, Ellingwood B. Fragility analysis of concrete gravity dams. Tech. rep., dams using fragility curves. Civ Eng J 2015;1(2):1420.
Washington, DC: Georgia Institute of Technology, Prepared for U.S. Army [69] Ghanaat Y. Failure modes approach to safety evaluation of dams. In:
Corps of Engineers; 2002. Proceedings of the 13th world conference on earthquake engineering,
[39] McKay M, Beckman R, Conover W. A comparison of three methods for selecting Vancouver, BC, Canada.
values of input variables in the analysis of output from a computer code. [70] Hariri-Ardebili M, Saouma V, Porter K. Quantification of seismic potential
Technometrics 2000;42(1):5561. failure modes in concrete dams. Earthq Eng Struct Dynam 2016;45:97997.
[40] Iman R, Conover W. A distribution-free approach to inducing rank correlation [71] Padgett J, Nielson B, DesRoches R. Selection of optimal intensity measures in
among input variables. Commun Stat Simul Comput 1982;11(3):31134. probabilistic seismic demand models of highway bridge portfolios. Earthq Eng
[41] Dolek M. Estimation of seismic response parameters through extended Struct Dynam 2008;37:71125.
incremental dynamic analysis. In: Papadrakakis M, Fragiadakis M, Lagaros ND, [72] Akaike H. A new look at the statistical model identification. IEEE Trans Autom
editors. Computational methods in earthquake engineering. Computational Control 1974;19(6):71623.
methods in applied sciences, vol. 21. Netherlands: Springer; 2011. p. 285304. [73] ANSYS. Ansys software reference manuals, release notes, mechanical apdl,
[42] Cornell A, Jalayer F, Hamburger R. Probabilistic basis for 2000 SAC federal elements reference, commands reference and theory reference, version release
emergency management agency steel moment frame guidelines. J Struct Eng 11; 2007.
2002;128:52632. [74] USACE. Earthquake design and evaluation of concrete hydraulic structures.
[43] Vamvatsikos D, Fragiadakis M. Incremental dynamic analysis for estimating Tech. Rep. EM 1110-2-6053. Washington D.C., USA: Department of the Army,
seismic performance sensitivity and uncertainty. Earthq Eng Struct Dynam U.S. Army Corps of Engineers; 2007.
2010;39:14163. [75] Yang T, Moehle J, Stojadinovic B, DerKiureghian A. Seismic performance
[44] Dolek M. Simplified method for seismic risk assessment of buildings with evaluation of facilities: methodology and implementation. ASCE Struct Eng
consideration of aleatory and epistemic uncertainty. Struct Infrastruct Eng 2009;135:114654.
2012;8:93953. [76] Hariri-Ardebili M, Saouma V. Probabilistic seismic demand model and optimal
[45] Anon., Dam safety assurance program evaluation report for bluestone lake. intensity measure for concrete dams. Struct Saf 2016;59:6785.
Tech. rep., Huntington District: US Army Corps of Engineers, vol. 3; September [77] Jankovic S, Stojadinovic B. Probabilistic performance-based seismic demand
1994. model for {R/C} frame buildings. In: Proceeding of the 13th world conference
[46] ABAQUS. Abaqus theory manual, version 6.7 ed., Providence, RI, USA: ABAQUS, on earthquake engineering, Vancouver, B.C., Canada.
Inc.; 2007. [78] Ramamoorthy S, Gardoni P, Bracci J. Probabilistic demand models and fragility
[47] Lin L, Adams J. Lessons for the fragility of canadian hydropower components curves for reinforced concrete frames. J Struct Eng 2006;132:156372.
under seismic loading. In: Proceedings of the 9th Canadian conference on [79] Saouma V, Cervenka J, Reich R. Merlin finite element users manual; 2010.
earthquake engineering, Ottawa, Ontario, Canada. p. 176271. <http://civil.colorado.edu/saouma/pdf/users.pdf>.
[80] Cervenka J, Chandra J, Saouma V. Mixed mode fracture of cementitious [87] Bernier C, Monteiro R, Paultre P. Using the conditional spectrum method for
bimaterial interfaces; Part II: Numerical simulation. Eng Fract Mech 1998;60 improved fragility assessment of concrete gravity dams in eastern canada.
(1):95107. Earthq Spectra 2016;0:null. http://dx.doi.org/10.1193/072015EQS116M.
[81] Jayaram N, Lin T, Baker J. A computationally efficient ground-motion selection [88] Ansari MI, Agarwal P. Categorization of damage index of concrete gravity dam
algorithm for matching a target response spectrum mean and variance. Earthq for the health monitoring after earthquake. J Earthq Eng 2016(ja):null. http://
Spectra 2011;27:797815. dx.doi.org/10.1080/13632469.2016.1138167.
[82] Gehl P, Seyedi DM, Douglas J. Vector-valued fragility functions for seismic risk [89] Hariri-Ardebili M, Furgani L, Meghella M, Saouma V. A new class of seismic
evaluation. Bull Earthq Eng 2012;11(2):36584. http://dx.doi.org/10.1007/ damage and performance indices for arch dams via eta method. Eng Struct
s10518-012-9402-7. 2016;110:14560.
[83] Lilliefors H. On the Kolmogorov-Smirnov test for normality with mean and [90] Olsson A, Sandberg G. Latin hypercube sampling for stochastic finite element
variance unknown. J Am Stat Assoc 1967;62(318):399402. analysis. J Eng Mech 2002;128(1):1215.
[84] Hariri-Ardebili M, Saouma V. Sensitivity and uncertainty quantification of the [91] PEER. Ground motion database; 2014. <http://ngawest2.berkeley.edu/> [last
cohesive crack model. Eng Fract Mech 2016;155:1835. viewed November 2014].
[85] Bernier C, Padgett JE, Proulx J, Paultre P. Seismic fragility of concrete gravity [92] Nuss L, Matsumoto N, Hansen K. Shaken, but not stirred - earthquake
dams with spatial variation of angle of friction: case study. J Struct Eng performance of concrete dams. In: Proceedings of the 32nd USSD annual
2015:05015002. meeting and conference: innovative dam and levee design and construction
[86] Bernier C. Courbes de fragilit pour la vulnrabilit sismique de barrages-poids for sustainable water management, New Orleans, Louisiana.
en bton; March 2015.

1 s2.0 S014102961630654X Main

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

1 s2.0 S014102961630654X Main

Caricato da

Copyright:

Formati disponibili

Engineering Structures 128 (2016) 374399

Contents lists available at ScienceDirect

Seismic fragility analysis of concrete dams: A state-of-the-art review

1. Introduction high-hazard dam is reported to have increased by nearly 40%

a vector of autocorrelated RVs with zero mean Sa T spectral acceleration at period T

proposed a hybrid method in which a critical measure of the

This section begins by providing a succinct explanation of the

Starting with the general concept of next generation perfor-

Hazard analysis Structural analysis Damage analysis Loss analysis

Hazard model Structural model Fragility model Loss model

Fig. 3. General framework for PBEE-2 [12].

of probabilistic analysis of nuclear power plants [19], in order to

Based Earthquake Engineering

Fragility PD P C LS jIM im 4

motion. It is characterized by two RVs: (1) the amplitude, and (2) 1

where Aacc is a Gaussian RV (with a confidence interval of 98%) of the

The random concrete properties, M conc , have a mean M 0 , and a

Method I: This is a simultaneous method, where the number of

Uniform Random Variable 2

Aleatory Epistemic Aleatory Epistemic

Seismic Intensity Level (1)

Seismic Intensity Level (2) Seismic Intensity Level (2)

Seismic Intensity Level (k) Seismic Intensity Level (k)

LS1: S22>350 psi LS3: Sliding > 0.5 in

(a) LSs 1 and 2 (b) LSs 3 and 4

Damage State MMI

PGA = 0.10.7 g (7 SILs). Thus, a total of 42 nonlinear transient

P (DS = dsi | im)

Damage state probability

1 Crest displacement with respect to heel (no threshold value is

j j x obtained by scaling up the results calculated for PGA = 0.1 g. The

Fig. 12. Seismic fragility curves for HW 36 m (adapted from [52]).

0.1 g 0.2 g 0.3 g 0.4 g 0.5 g 0.6 g 0.7 g 0.8 g

Probability of visible damage

Probability of visible damage

0.4 ft = 1 MPa 0.4

0.2 0.2 E /E = 1.00

damage (it is only referred to significantly exceeding the lin- 1

where k and j 2 0; 1 are scale and shape parameters. It is

Fig. 16. Seismic fragility curve (adapted from [62]).

As for [54], only record-to-record variability is considered in

Limit States AUS% ADS% Damage level

LS1 Serviceability (S) [012%] <16% Negligible

Limit state probability

Crossing fragility curves Modied; method (1) Modied; method (2)

LS1 LS1 LS1

LS3 LS3 LS3

Intensity measure Intensity measure Intensity measure

Index Quantity ln g IM b: lnIM lna 24

analyze Long Term Field Monitoring data, and possibly perform

4. Contextual framework 5. Summary

Fig. 29. Outline of the concrete dams fragility analysis.

No Research information Modeling features Analysis type

M.A. Hariri-Ardebili, V.E. Saouma / Engineering Structures 128 (2016) 374399

No Aleatory (ground motion record-to-record) uncertainty

M.A. Hariri-Ardebili, V.E. Saouma / Engineering Structures 128 (2016) 374399

No Epistemic uncertainty Special features Post-processing

M.A. Hariri-Ardebili, V.E. Saouma / Engineering Structures 128 (2016) 374399

Based on a closer study of these reported investigations, the References

Potrebbero piacerti anche