Seismic Performance Sensitivity and Uncertainty Analysis of Gravity Dams

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

EARTHQUAKE ENGINEERING & STRUCTURAL DYNAMICS

Earthquake Engng Struct. Dyn. 2015; 44:41–58


Published online 3 July 2014 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/eqe.2457

Seismic performance sensitivity and uncertainty analysis of


gravity dams

M. Alembagheri*,† and M. Seyedkazemi


Department of Civil Engineering, Sharif University of Technology, Tehran, Iran

SUMMARY
Uncertainties in structural engineering are often arising from the modeling assumptions and errors, or from
variability in input loadings. A practical approach for dealing with them is to perform sensitivity and uncer-
tainty analysis in the framework of stochastic and probabilistic methods. These analyses can be statically and
dynamically performed through nonlinear static pushover and IDA techniques, respectively. Of the existing
structures, concrete gravity dams are infrastructures which may encounter many uncertainties. In this re-
search, probabilistic analysis of the seismic performance of gravity dams is presented. The main character-
istics of the nonlinear tensile behavior of mass concrete, along with the intensity of earthquake excitations
are considered as random variables in the probabilistic analysis. Using the tallest non-overflow monolith
of the Pine Flat gravity dam as a case study, its response under static and dynamic situations is reliably ex-
amined utilizing different combinations of parameters in the material and the seismic loading. The sensitiv-
ity analysis reveals the relative importance of each parameter independently. It will be shown that the
undamaged modulus of elasticity and tensile strength of mass concrete have more significant roles on the
seismic resistance of the dam than the ultimate inelastic tensile strain. In order to propagate the parametric
uncertainty to the actual seismic performance of the dam, probabilistic simulation methods such as Monte
Carlo simulation with Latin hypercube sampling, and approximate moment estimation techniques will be
used. The final results illustrate the possibility of using a mean-parameter dam model to estimate the mean
seismic performance of the dam. Copyright © 2014 John Wiley & Sons, Ltd.

Received 21 November 2013; Revised 25 May 2014; Accepted 12 June 2014

KEY WORDS: concrete gravity dams; nonlinear static pushover; incremental dynamic analysis; sensitivity
analysis; uncertainty; Monte Carlo

1. INTRODUCTION

One of the important aspects of the performance-based earthquake engineering is accurate estimation
of the seismic demand and capacity of structures [1]. There are two sources of uncertainty in
seismic performance of structures: (a) aleatory randomness, for example, due to natural ground
motion record variability and (b) epistemic uncertainty stemming from modeling assumptions,
omissions or errors [2]. Adequate understanding about the expected range of structural response can
be obtained considering the effects of uncertainties in probabilistic and stochastic analysis of
structures [3]. Computing the variability in the seismic performance due to epistemic uncertainty is
usually performed using safety factors or standard dispersion values [4]. The calculations are
performed using a deterministic model of the structure assuming specific numerical values for the
loads and the strength parameters. But these methods do not lead to actual response. The required
values for a safety factor are defined to ensure a satisfactory structural performance considering

*Correspondence to: Mohammad Alembagheri, Department of Civil Engineering, Sharif University of Technology,
Tehran, Iran.

E-mail: alembagheri@alum.sharif.ir

Copyright © 2014 John Wiley & Sons, Ltd.


42 M. ALEMBAGHERI AND M. SEYEDKAZEMI

uncertainties in three basic aspects: (i) the applied loads; (ii) the strength parameters; and (iii) the limits
and assumptions inherent to the structural analysis method selected. The probabilistic analysis
considers explicitly the uncertainties in the loading and strength parameters that are considered as
random variables. The uncertainties in input parameters are then transformed in probability of limit-
states of a structure. A probabilistic analysis requires more information than a deterministic analysis
[5]. Several investigations have been performed to isolate some useful cases and gain insight into
the effect of the properties of a model to its estimated seismic performance [6–8]. However, these
attempts have been partly made in case of concrete dams. In a research, the probabilistic analysis of
gravity dams has been performed using artificially generated seismic excitation and random
variations of concrete properties over the spatial domains [9]. A robust method of considering both
sources of uncertainty (loadings and strength parameters) in structural engineering is the use of
Monte Carlo simulation [10] within the framework of incremental dynamic analysis (IDA) [11].
Although this method has been extensively used to evaluate limit-state uncertainty of steel and RC
frames [2, 12, 13], it is fairly generalizable and applicable to a variety of structures and limit-states.
Concrete dams are among the huge infrastructures, which their seismic response can be influenced
by the presence of various uncertainties [5]. Because limited data is available on the field observations
and laboratory tests of major dams, theoretical and numerical methods are preferred for evaluation of
limit-states of concrete dams in light of uncertainties. In this context, two well-known methods for
calculating the seismic demand and capacity are nonlinear static pushover (SPO) and IDA. In SPO
analysis, a monotonic intensifying lateral load is statically applied to the structure, and its response
is tracked across various limit-states [14]. In the other approach, IDA, the earthquake ground
motion’s intensity applied to the structure is incrementally increased and shifts the structure to its
severer limit-states [11]. However, it should be noted that the seismic loads are not the only loads
that can lead to the failure of concrete gravity dams; based on the historical records, the static
hydraulic loadings have prominent effect in their failure [15]. Many researches have focused on
reliability analysis of dams under hydraulic loadings. Altarejos-Garcia et al. [16] estimated the
probability of failure of a gravity dam as a case study considering the sliding failure mode under
hydraulic loadings. Carvajal et al. [17] proposed a procedure for probabilistic assessment of
hydraulic loads and shear strengths of roller compacted concrete (RCC) gravity dams. A case study
demonstrated the applicability of their procedure on an existing RCC gravity dam. Ellingwood and
Tekie [18] presented the basic fragility concepts and databases required to support the fragility
assessment of gravity dams under seismic and hydraulic loadings. They illustrated their method
using a concrete monolith from the Bluestone Dam in USA. Fell et al. [19] used quantitative risk
assessment techniques to assess the dam safety management utilizing methods available for
estimating the probability of failure of embankment and concrete dams, for normal operating loads,
flood, and earthquake.
In this research, the effects of aleatory randomness and epistemic (modeling) uncertainty to the
response of concrete gravity dams will be assessed first using sensitivity analysis and then
uncertainty evaluation using (a) Monte Carlo simulation with Latin hypercube sampling (LHS); (b)
point-estimation; and (c) first-order second-moment techniques. The purpose of this study is the
application of the methodology already proposed in Reference [13], which is about the steel frames,
to concrete gravity dams in order to examine the peculiarities of their response when compared with
that of the frame structures studied in the mentioned reference. Efficient Monte Carlo
simulation and moment estimation techniques are used to propagate the uncertainty from
parameters to the IDA-evaluated seismic performance offering different compromises in speed
and accuracy [13]. Using the Pine Flat gravity dam as a case study and focusing on the mass
concrete tensile behavior parameters, the strength parameters uncertainties will be examined
under the randomness of various earthquakes intensities. The Pine Flat dam is a well-shaped
gravity dam, which is two-dimensionally modeled along with its compressible reservoir on
rigid foundation, in a plane-stress analysis. For SPO analysis, the method developed by
Alembagheri and Ghaemian [20] is employed, and for IDA, series of 10 earthquake ground
motions is utilized for dynamic analysis of the dam–reservoir system. Sensitivity analysis is
performed by shifting the basic characteristics of concrete tensile constitutive behavior, that is,
undamaged modulus of elasticity, tensile strength, and ultimate inelastic tensile strain, which

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe
SEISMIC SENSITIVITY AND UNCERTAINTY ANALYSIS OF GRAVITY DAMS 43

are assumed to be random variables, from their assigned mean value. However, for uncertainty
analysis, normal distribution is assumed for the random variables and a set of 30 instances of
the dam model is selected using LHS method within Monte Carlo simulation.
From the sensitivity analysis, it will be shown that the modulus of elasticity influences the
deformation demands and does not influence the final capacity. The tensile strength affects both
the deformation demand and loading capacity of the dam only at the final failure. However,
ultimate inelastic tensile strain is of minor importance. Finally, the uncertainty analysis concludes
that one can reliably use the mean-parameter dam model to estimate the mean seismic
performance of the dam.

2. NUMERICAL MODELING

The gravity dam selected is the tallest non-overflow monolith of Pine Flat gravity dam constructed on
the Kings River, California, as case study. The geometric properties of the selected cross section
(monolith) of the dam are described in Figure 1(a). This concrete dam with the height of 122 m has
been constructed to control the seasonal flood and supply water for agricultural and recreational
areas. Large reservoir with a capacity of 1.2 e + 9 m3 is located on the upstream of the dam [21]. The
dam along with its reservoir is modeled using finite elements in 2D plane-stress analysis. The depth
of the reservoir is considered 116 m in seismic analyses, which means that the reservoir is full. The
reservoir’s length is assumed 5 times the dam height, and the non-reflective boundary condition has
been assigned to its far-end to model the radiation of the pressure waves into infinity. Zero
hydrodynamic pressure is assigned to the reservoir’s free surface, and there is no absorbing
boundary condition at the reservoir’s bottom. The foundation is assumed rigid in all of analyses.
The base of the dam has been tied to the rigid foundation; hence, there is no sliding along the dam–
foundation interface. For a gravity dam, the sliding may occur along the dam–foundation interface,
which is generally the weakest zone, or within the dam body or rock foundation. However, the
present research focuses on the cracking behavior of the dam body during seismic excitations, and it
does not take into account the effects of sliding. The water compressibility and dam–water
interaction are considered in the dynamic analyses. The finite element mesh of the selected monolith
of the dam and part of its reservoir, which is used for nonlinear static and dynamic analyses, is
shown in the Figure 1(b). The mesh has been refined in crack-prone areas especially at the base and
neck, compared to the mesh size required for the linear analysis, because of limitations on the
element size in a nonlinear analysis.

(a) (b)

Figure 1. (a) Configuration of the tallest non-overflow monolith of the Pine Flat dam (all dimensions are in
meters); (b) Finite element mesh of the dam and the truncated reservoir.

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe
44 M. ALEMBAGHERI AND M. SEYEDKAZEMI

Material nonlinearity of mass concrete in static and dynamic analyses is modeled using plastic–
damage approach [22], which is in fact a homogenous damage mechanics method. In this approach,
under uniaxial cyclic tension, the concrete behaves linearly before the tensile strength (failure stress
and elastic limit), σt0. At this stress level, fine cracks are formed within the material matrix. After
that, the combination of these micro-cracks is modeled macroscopically by strain softening. By
unloading on each point after the elastic limit on the softening branch, the response is weakened and
deterioration of material’s elastic stiffness is characterized by a scalar tensile damage variable, dt,
which is assumed to be a function of the plastic strains. The reduction of the elastic modulus
regarding the tensile damage is determined by

E ¼ ð1  d t ÞE 0 (1)

in which, E0 is the initial undamaged elastic modulus, either in tension or compression. Damage
variable, dt, varies between zero, which shows no damage, and one, indicating total loss of stiffness.
The uplift pressures are not modeled. In this research, the constitutive tensile behavior of mass
concrete is considered using a bilinear stress–strain backbone as shown in Figure 2. Based on the
considered model, three main parameters are governing the tensile behavior of mass concrete. These
parameters are the undamaged modulus of elasticity (E0), tensile strength (σt0), and ultimate
inelastic tensile strain (εu).
Although this model is somewhat simplified, it is versatile enough to simulate the main
characteristic behavior of mass concrete from brittle to ductile tensile fracturing [23]. A ‘base’
concrete tensile behavior is defined using the properties listed in Table I, which are assumed to be
the mean values for all tensile behavior parameters. The tensile damage variable, dt, is assumed to
be its maximum possible value on each point of the softening branch of the constitutive model. This
means that by progressing along the softening branch, inelastic strain, εin, shown in Figure 2, should
have minimum possible value for each point on the softening branch. Thus, there is a base-case
model that will serve as the basis for comparing all other modified models. The compression
behavior is assumed linear. It should be noted that previous studies have identified the nonlinear
tensile behavior of the mass concrete as the most influential parameter in nonlinear seismic response
of gravity dams; the compression damage is not usually observed, and tensile failure mode
dominates the dam structural model [20, 24–26].
The fundamental period of the base-case dam model is T base
1 = 0.462 s and accounts for more than
85% of the dam body’s total mass. Essentially, this is a first-mode dominated structure that still
allows for significant sensitivity to higher modes. The material damping for mass concrete is
considered using modified Rayleigh damping [27] with the mass proportional damping coefficient of
1.57 and the stiffness proportional damping coefficient of 0.0014, which produces about 5% critical
damping in the range of first five vibration modes of the coupled base-case dam–reservoir system.
The same coefficients are assumed for other modified models.

Figure 2. Concrete constitutive behavior backbone in uniaxial tension and its three controlling parameters.

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe
SEISMIC SENSITIVITY AND UNCERTAINTY ANALYSIS OF GRAVITY DAMS 45

Table I. The ‘base’ concrete properties both for static and dynamic situations.
Material parameter Value
Undamaged Young’s modulus, E0 (GPa) 27
Poisson’s ratio 0.2
Density (Kg/m3) 2400
Tensile strength, σt0 (MPa) 3.0
Ultimate tensile strain, εu 0.00023

3. ASSESSMENT OF SEISMIC PERFORMANCE

The seismic performance of the gravity dam is statically and dynamically evaluated using SPO and
IDA techniques, respectively. For the SPO analysis, the method described by Alembagheri and
Ghaemian [20] will be used. In this method, the lateral pushover load is considered as combination
of the inertial load, which is proportional to the fundamental mode shape of the base-case model
and dam’s width; and the hydrodynamic load, which its distribution is modeled with Westergaard’s
added mass approach. In this approach, the inertial and hydrodynamic loads have the same direction
during earthquakes. The loads are scaled appropriately with respect to each other and applied
incrementally on the upstream face of the dam. The same lateral load distribution is assumed for all
models. Because of the unsymmetric shape of the dam, the lateral pushover load should be
separately applied into the upstream (U/S) and downstream (D/S) directions. This load will be
applied after exerting the weight and hydrostatic loads to the dam. By drawing the total lateral load
(base shear) against the relative crest displacement, the pushover curve will be generated. Increasing
the lateral load pushes the dam into its severer limit-states in two opposite directions.
The IDA is a parametric analysis that more comprehensively estimates the structural performance
under seismic loads. For IDA study, a suite of 10 ordinary ground motions is selected (Table II).
They have large magnitudes of 6.2–7.3, recorded on firm soil (type B or C of USGS) at moderate
distances, and derived from Pacific Earthquake Engineering Research Center (PEER) strong motion
database. The longitudinal component of earthquake records is scaled into multiple increasing levels
of intensity, and a nonlinear dynamic analysis of the dam–reservoir system after application of static
loads is performed under each of them. The records are uniformly applied to the dam’s base. The
desirable response of the system, which is called engineering demand parameter (EDP), will be the
relative crest displacement. It can be used to categorize various limit-states of the dam model [20].
For intensity measure (IM), the 5% damping  spectral
 acceleration at the fundamental vibration
period of the base-case model, that is Sa T base
1 ; 5% has been shown to be an appropriate choice for
measuring the earthquake loading intensity [11, 20]. The naming is according to current PEER
terminology. Even under stiffness and mass uncertainties, the fundamental period of the bas-case
 base
dam model, T 1 , can still serve as a reliable reference point [13]. Therefore, Sa T 1 ; 5% can be
base

Table II. Ground motions selected for IDA of the Pine Flat dam.

USGS Sa(5%)
Record Component Moment PGV R soil at T1 = 0.462s
No Earthquake name name (deg) magnitude PGA (g) (cm/s) (km) type (g)

1 Imperial Valley, 1940 IVELC 180 7.0 0.313 29.8 8.3 C 0.635
2 Imperial Valley, 1979 IVPTS 315 6.5 0.204 16.1 14.2 B 0.274
3 San Fernando, 1971 SFPAS 90 6.6 0.11 13.3 31.7 B 0.362
4 San Fernando, 1971 SFPPP 21 6.6 0.136 5.6 38.9 B 0.193
5 Landers, 1992 LADSP 0 7.3 0.171 20.2 23.2 B 0.582
6 Loma Prieta, 1989 LPSTG 0 6.9 0.512 41.2 13 B 0.684
7 Loma Prieta, 1989 LPGIL 67 6.9 0.357 28.6 11.6 B 0.852
8 Loma Prieta, 1989 LPAND 270 6.9 0.244 20.3 21.4 B 0.442
9 Morgan Hill, 1984 MHG06 90 6.2 0.292 36.7 11.8 B 0.546
10 Northridge, 1994 NRORR 360 6.7 0.514 52.2 22.6 B 1.252

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe
46 M. ALEMBAGHERI AND M. SEYEDKAZEMI

recommended for general use avoiding simpler but less efficient IMs, such as PGA. However,
renormalizing to another IM is actually trivial and only matter  of post-processing [13]. All
earthquake records are scaled to multiple levels of Sa T base
1 ; 5% from zero to 1.0 g in the regular
steps of 0.1 g. An IDA curve is the plot of IM versus EDP for each model under each scaled ground
motion. Using spline interpolation, the IDA curve is generated from discrete points. Such results are
in turn summarized to produce the mean, median, 16% and 84% fractile IDA curves. The
distribution of the seismic demand and capacity of the structure for rare ground motion intensities,
therefore, can be accurately estimated [11].
Now alternate models of the dam should be selected for evaluation. There is large number of
variations which can be chosen considering the three parameters of adopted tensile behavior of mass
concrete. The three backbone parameters, that is, E0, σt0, and εu, vary independently from each other
but uniformly over the dam body. So the mass concrete is homogenous and isotropic in entire the
dam. The varying spatial distribution of the parameters over the dam body can be the subject of
future studies. In the following sections, the effects of the three parameters are evaluated in a
probabilistic framework using sensitivity and uncertainty analyses.

4. EVALUATION OF THE MODEL SENSITIVITY

The sensitivity analysis is used to evaluate the sensitivity of the structural response of the dam model to
the variations of tensile behavior of mass concrete. This analysis is performed by shifting each of the
three backbone parameters independently just one at a time, above and below its central base-case
value. Both SPO and IDA are utilized for sensitivity analysis for the following pairs of
modifications: E0 (GPa) = {23, 31}; σt0 (MPa) = {2.5, 3.5}; and εu = {0.00016, 0.00030}. The
sensitivity analysis performed in this section follows the procedure adopted in Reference [13].
However, the IDA is a powerful method for more thoroughly predicting the structural seismic
performance with regard to the input parameters, but due to the complexity of the dynamic
nonlinear response of the structures, it is beneficial to firstly perform a SPO analysis. The SPO is
simple method that presents helpful insight on the expected response of a model, even though it has
its own limitations to predict the model’s dynamic behavior.

4.1. Sensitivity of the static pushover analysis


The SPO analysis is performed as a preliminary evaluation of the performance sensitivity of the base-
case versus the modified models. Because of the unsymmetric shape of the gravity dam, the increasing
lateral load must be separately applied into each opposite directions of U/S and D/S.

4.1.1. Application of the pushover load into the downstream direction. After exerting the
gravitational and hydrostatic loads, the lateral pushover load is statically and incrementally applied
to the U/S face into the D/S direction. In general, in this direction, by increasing the lateral load, the
tensile softening is firstly appeared at the heel of the dam (the ‘yielding state’), which is followed by
complete tensile damage (dt = 1.0) at the heel (the ‘cracking initiation’ state). Then, this crack is
stably propagated along the dam base until the situation at which by further increasing the lateral
load, an unstable crack is formed at the place of slope changes on the U/S face of the dam and
rapidly propagated to the D/S face. Thus, a major crack appeared at the dam’s neck. This stage is
considered as the ‘ultimate state’ (collapse) [20]. Table III summarizes the models responses in the
three mentioned limit-states for the base-case and the other six modified models. The SPO curves of
relative crest displacement against the total lateral load (base shear) are plotted in Figures 3(a–c)
where the base-case’s one is plotted in solid line.
The sensitivity of structural performance considering various elastic moduli (E0) is shown in
Figure 3(a). With an increase in elastic modulus of the mass concrete, there is an increase in the
slope of the SPO curve, as expected. About 15% variation in the elastic modulus of concrete causes
minor change in limit-states’ loading capacities. However, considerable change is observed in the
corresponding crest displacements, especially at the ‘ultimate state’. Therefore, the main effect of

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe
SEISMIC SENSITIVITY AND UNCERTAINTY ANALYSIS OF GRAVITY DAMS 47

Table III. Models’ responses at three limit-states from SPO analysis into the D/S direction.
Model Yielding state Cracking initiation Ultimate state
Base-case Total lateral load (MN) 20.1 24.8 65.7
Crest displacement (cm) 1.6 2.0 7.6
E0 = 23 GPa Total lateral load (MN) 20.4 23.6 65.7
Crest displacement (cm) 1.9 2.1 9.0
E0 = 31 GPa Total lateral load (MN) 20.6 26.1 65.8
Crest displacement (cm) 1.4 1.9 6.9
σt0 = 2.5 MPa Total lateral Load (MN) 13.9 23.0 57.7
Crest displacement (cm) 1.1 1.8 6.6
σt0 = 3.5 MPa Total lateral load (MN) 26.6 27.1 73.6
Crest displacement (cm) 2.0 2.1 9.1
εu = 0.00016 Total lateral load (MN) 20.1 20.7 65.7
Crest displacement (cm) 1.6 1.8 7.7
εu = 0.00030 Total lateral load (MN) 20.1 29.5 65.7
Crest displacement (cm) 1.6 2.2 7.7

elastic moduli is on the crest displacement demands of various limit-states. This result was expected
because elastic modulus only affects the elastic stiffness of the dam and does not change the tensile
strength of the dam.
Clearly shown in Figure 3(b), the tensile strength (σt0) is directly related to the global strength and
deformation capacity of the dam. The curves coincide in the elastic part (pre-yielding range), which
was expected because elastic moduli are the same for all cases. As shown in Table III, significant
changes occur both in force and displacement of various limit-states and global resistance of the

Figure 3. Sensitivity of the SPO curves to the tensile backbone parameters, D/S direction.

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe
48 M. ALEMBAGHERI AND M. SEYEDKAZEMI

dam. Increasing σt0 by 17% provides beneficial effect practically near failure, improving the maximum
base shear by about 12%, and increasing the maximum attainable crest displacement by 20%, which
are the first highest base shear and deformation among all models. Decreasing σt0 has reverse
effects. Totally, variation of the σt0 makes the most effects near final failure range of the response.
Variation of the ultimate inelastic tensile strain changes the behavior of the mass concrete from
brittle to ductile. The sensitivity analysis conducted shows that 30% variation in the εu, without
changing the σt0, totally does not change the behavior of the dam. However, careful examination of
Table III and damage profiles shows some marginal variation in the ‘cracking initiation’ state.
4.1.2. Application of the pushover load into the U/S direction. By application of the lateral load into
the U/S direction, the dam elastically responds until appearance of unstable cracking from the place of
slope change on the D/S face, which is quickly propagated into the U/S face. The dam continues to
deform under constant load, which indicates failure or ‘ultimate state’ in the U/S direction. So the
dam’s behavior consists of two parts: elastic and failure. There is no stable crack growth within the
dam’s body [20]. Hence, in this direction, the dam’s behavior can be simulated by a single-degree-
of-freedom (SDOF) system with perfectly elastic–plastic behavior. The results of the SPO analysis
into the U/S direction are summarized in Table IV and Figure 4.
Generally, the variation of the models’ response is the same as the D/S direction. The change in
elastic moduli only affects the slope of the elastic part of the SPO curves; increase in the elastic
moduli increases the slope and vice versa. This slope is the stiffness of the equivalent SDOF system,
that is, 12.45, 10.75, and 14.78 MN/cm for E0 = 27 (base-case), 23 and 31GPa models, respectively.
So, the 15% change in the elastic modulus causes about 14% decrease and 19% increase in the
stiffness of the equivalent SDOF system for E0 = 23 and 31GPa models, respectively. The elastic
moduli do not make noticeable effect on the ultimate lateral load the dam can resist (Figure 4(a)).
The change of the tensile strength of concrete considerably affects the carrying load and
deformation capacity of the dam (Figure 4(b)), but does not affect its equivalent SDOF stiffness.
Variation only in ultimate inelastic tensile strain of the mass concrete, that is, making the concrete
more brittle or ductile, does not affect the dam behavior.
All in all, variation of elastic modulus mostly influences the deformation demands of the dam in the
entire loading domain; and variation of tensile strength makes significant changes to the post-failure
part of the behavior. The effect of ultimate tensile strain is more apparent when loading is applied
into the D/S, but from practical point of view, its effect is not important.

4.2. Sensitivity of the IDA


The sensitivity of the dam performance to the three main parameters of tensile behavior is dynamically
investigated using the IDA. By perturbing the tensile properties of the concrete from its ‘base’ model,
one at a time, the influence of each parameter on the behavior of the dam is assessed using the three

Table IV. Models’ responses at one limit-state from SPO analysis into the U/S direction.
Model Ultimate state
Base-case Total lateral load (MN) 47.3
Crest displacement (cm) 3.8
E0 = 23 GPa Total lateral load (MN) 47.3
Crest displacement (cm) 4.4
E0 = 31 GPa Total lateral load (MN) 47.3
Crest displacement (cm) 3.2
σt0 = 2.5 MPa Total lateral load (MN) 40.3
Crest displacement (cm) 3.2
σt0 = 3.5 MPa Total lateral load (MN) 54.3
Crest displacement (cm) 4.3
εu = 0.00016 Total lateral load (MN) 47.3
Crest displacement (cm) 4.0
εu = 0.00030 Total lateral load (MN) 47.3
Crest displacement (cm) 3.9

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe
SEISMIC SENSITIVITY AND UNCERTAINTY ANALYSIS OF GRAVITY DAMS 49

Figure 4. Sensitivity of the SPO curves to the tensile backbone parameters, U/S direction.

pairs as previously mentioned. Only 10 records are used to trace the mean IDA curves so small differences
are statically insignificant. Like SPO analysis, the dam’s maximum motion into the U/S and D/S directions
should be separately examined because of the cyclic nature of the dynamic loading and unpredictability of
the direction of maximum displacement. Each mean IDA curve is the result of 10 different input records
(Table II), each scaled into 10 increasing intensity levels. Therefore, each mean or median IDA curve is
the result of 100 nonlinear dynamic analyses of Pine Flat dam.
Figures 5(a–c)
 show mean IDA curves of maximum D/S crest displacement versus the seismic
1 ; 5% . Obviously, increasing the elastic modulus causes more hardened response
intensity, Sa T base
of the dam in the entire intensity range, as shown in Figure 5(a), like the corresponding SPO
analysis. The local hardening is observed in some parts of the curves. The IM corresponding to each
limit-state does not significantly change by changing the E0. Figure 5(b) shows that the variation of
tensile strength does not change the elastic part of the response. Increasing the tensile strength
decreases the maximum crest displacement at each intensity level in post-yielding region. The IM
corresponding to each limit-state increases by increasing the σt0. The change in ultimate inelastic
tensile strain does not considerably affect the mean IDA curves, as shown in Figure 5(c). This is in
agreement with the previous observations from SPO analysis. It seems that the strength loss caused
by brittle fracturing of concrete will not dominate the response of the dam.
The mean IDA curves in the U/S direction are shown in Figures 6(a–c). The responses are not
straightforward like theD/S direction. From Figure 6(a), the same results as the D/S direction can be
 base
seen until Sa T 1 ; 5% = 0.6g, after that increasing E0 increases the crest displacement with respect
to the base-case model, but not as high as decreasing the E0 affecting the response. Investigation of
the damage profiles shows more imposed damage on the dam’s body by increasing the E0. From
Figure 6(b), changing the σt0 does not alter the mean IDA curve in pre-yielding range, but it causes
strange effects on the response in the post-yielding region. Decreasing the σt0 increases the response
after ‘yielding state’. This issue can be seen for increasing the σt0 at some intensity
 levels. More brittle
or ductile concrete, does not affect the maximum crest displacement until Sa T base 1 ; 5% = 0.5 g, as

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe
50 M. ALEMBAGHERI AND M. SEYEDKAZEMI

Figure 5. Sensitivity of the mean IDA curves to the tensile behavior backbone parameters, D/S direction.

shown in Figure 6(c). After that, making the concrete more brittle, decreases the maximum crest
displacement in each intensity level. However, making the concrete more ductile, changes the behavior
so that no clear conclusion can be drawn. It should
 be mentioned that the diffused cracking pattern
1 ; 5% = 0.7 g makes impossible the accurate interpretation of
imposed to the dam structure after Sa T base
the results. So, the results at the high shaking intensities should be cautiously accounted for.
A noteworthy observation in all figures is that the tensile characteristics of the mass concrete can
cause uncertainties in the behavior and performance of the dam under seismic motions. These
uncertainties are not easily assessable and predictable. However, these uncertainties may be
manageable using probabilistic methods.

5. PROBABILISTIC ANALYSIS

In order to examine the effects of uncertainties on the seismic performance of the selected gravity dam,
the base-case’s tensile constitutive backbone of the mass concrete is varied by assigning probabilistic
distributions to its three random parameters, that is, the undamaged elastic modulus (E0), the tensile
strength (σt0), and ultimate inelastic tensile strain (εu). The probabilistic analysis performed in this
section follows the procedure adopted in Reference [13]. By the knowledge of the authors, the
existing literature does not provide adequate guidance on the probabilistic distributions of the
mentioned parameters, so they are appropriately defined arbitrarily. They are assumed to
independently vary and normally distributed with a mean equal to the corresponding base-case’s
value and a coefficient of variation (standard deviation divided by mean), cv, of 20% for all of them.
The normal distribution assigns nonzero probability even for physically impossible values of the
parameters, for example, εu < σEt00 , so appropriate cut-off values should be defined to avoid assigning
unrealistic values to the random variables. The distribution properties of the uncertain parameters are

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe
SEISMIC SENSITIVITY AND UNCERTAINTY ANALYSIS OF GRAVITY DAMS 51

Figure 6. Sensitivity of the mean IDA curves to the tensile behavior backbone parameters, U/S direction.

listed in Table V. All distributions are appropriately rescaled so that the concentration of high
probabilities at the cut-off minimum and maximum points is avoided [28].

5.1. Monte Carlo simulation with Latin hypercube sampling


There are few ways for estimating the variability and uncertainty of complex nonlinear structural
models, such as concrete gravity dams. The most comprehensive but time-consuming route is the
Monte Carlo simulation [10]. This method is used to model phenomena with significant uncertainty
in inputs, such as those that can be expected in earthquake analysis of structures.
Here, by sampling N times from the parameter distributions, this procedure creates a population of N
possible instances of the structure, each of which needs to be analyzed [13]. The full distribution of the
seismic response of the structure, therefore, can be reliably predicted assuming a large enough number
of sampled structures. The functionality of Monte Carlo simulation can be further improved by
replacing the classic random sampling of the population with Latin hypercube sampling (LHS). The
LHS is a statistical method for generating a sample of plausible collections of parameter values from a
multidimensional distribution. The LHS was described by McKay et al. in 1979 [29]. It was further
elaborated by Iman and others [30, 31]. This route has been used by many researchers to handle
parameter uncertainties within IDA framework [11, 13]. The LHS is a special case of stratified

Table V. The distribution properties of the random parameters.


Parameter Mean cv (%) Min. Max. Type

E0 27 GPa 20 21 GPa 33 GPa Truncated normal


σt0 3.0 MPa 20 2.2 MPa 3.8 MPa Truncated normal
εu 0.00023 20 0.00014 0.00032 Truncated normal

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe
52 M. ALEMBAGHERI AND M. SEYEDKAZEMI

sampling that allows efficient estimation of the quantity of interest by reducing the variance of classic
Monte Carlo [13].
There is no specified value for sample size N to achieve a certain confidence level in LHS [30].
Melchers [32] presented different formulas to estimate the required number of simulations to ensure
proper convergence to an accurate estimate of the probability of failure of the system analyzed. The
simplest formula is from Broding et al. [33] that suggested:

 lnð1  DÞ
N> (2)
Pf

where D is the given confidence level and Pf is the probability of failure. So more than 30 simulations
are required for a 95% confidence level and Pf = 101. This value for the probability of failure of
gravity dams seems to be large, but this choice is purely illustrative, for reducing the number of
simulations. However, the intensity of considered earthquakes is incrementally increased so that the
dam is forcedly shifted to its higher performance levels.
A relatively high value of N that is dependent on the structure type and number of random parameters
and substantially larger than the number of random parameters, will always result to reasonably accurate
estimates for practical purposes. Here, Monte Carlo simulation with LHS is carried out for N = 30
realizations of the dam–reservoir system. This number is selected based on the Equation 2, and the
performed sensitivity analysis and the observed variability of the dam’s response to variation of the
random parameters. It can pinpoint accuracy in estimation of the whole range of the dam’s responses.
The investigation of the uncertainties is firstly started by evaluating the SPO curves of the samples,
as plotted in Figures 7(a, b). Their SPO curves are generated in two opposite directions of U/S and D/S.
The samples show various slopes in the elastic section in two directions, which is due to their varying
elastic modulus. In the D/S direction, the final relative crest displacement, ranges from about 5 to
10 cm, which shows significant various ductility of the samples. The maximum loading capacities
(base shears) are varying from about 56 to 75 MN. In comparison with the results of the sensitivity
analysis, it can be said that these differences are mostly because of variation in tensile strength of
the mass concrete. However, the obtained range of ultimate crest displacement here is wider than the
corresponding range in sensitivity analysis, but the range of the base shear is approximately the same.
In the U/S direction, the final crest displacement varies between about 3.2 to 4.5 cm, which shows
lower variability with respect to the D/S direction. The ultimate strengths are from about 40 to
55 MN. The results, in general, illustrate lower ductility and strength with respect to the D/S
direction. The elastic range in the U/S direction is more extended than the D/S direction. Again, the
final failure mechanism of the dam model in the D/S and U/S directions is abrupt neck cracking
from U/S and D/S faces, respectively. The sample-to-sample variation of the results increases as the
samples approach to their final failure.

Figure 7. The collection of 30 SPO curves of the dam samples for motion into the (a) D/S direction; (b)
U/S direction.

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe
SEISMIC SENSITIVITY AND UNCERTAINTY ANALYSIS OF GRAVITY DAMS 53

The base-case’s pushover curve also appears in Figures 7(a, b) in solid line. In both directions, the
base-case’s curve is approximately located in the middle of the curves population both for
displacement and lateral force. This shows that the actual mean response of the dam is
approximately same as the response of the mean model, in two opposite directions. So the dam can
be accurately modeled utilizing the mean values of concrete tensile behavior parameters, which
tends to respond as the mean response values of the dam samples.
The IDA study is performed for each sample using the selected earthquake records (Table II) so
there are 10 × N = 300 IDA curves and the N = 30 corresponding mean IDAs, as shown in Figure 8
along with their mean and mean ± one standard deviation curves. The variability in the results is
apparent specifically in inelastic post-yielding section. The response of the dam in D/S direction is
generally hardened, however, in U/S direction is softened.
The mean curves of the mean IDAs in two opposite directions along with the corresponding base-
case’s IDA curves are plotted against each other in Figure 9. These curves show the overall mean
response of the dam with respect to the response of the mean model. High accuracy can be seen for
entire D/S direction and elastic range of U/S direction. Although the maximum 20% difference at
high intensities in U/S direction might not be significant for engineering purposes. So one can
reliably use the mean model to analyze the mean response of the dam with probabilistic distributed

Figure 8. The collection of 30 mean IDA curves of the dam samples for motion into the (a) D/S direction; (b)
U/S direction.

Figure 9. The overall mean and base-case’s IDA curves of the dam model.

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe
54 M. ALEMBAGHERI AND M. SEYEDKAZEMI

parameters for tensile behavior of mass concrete, which is the most influencing factor controlling the
nonlinear behavior and final failure mechanism of concrete gravity dams [20, 22, 25, 26].

5.2. Moment estimation


For making useful conclusions from the obtained results, the probabilistic nature of the curves should
be quantified and simplified by estimating their moments. In this way, a central value and dispersion
for the demanded
 response
 values (maximum relative crest displacement) at each intensity level
1 ; 5% should be computed.
given by Sa T base
Using the results of the IDA study in framework of the Monte Carlo simulation with LHS (Figure 8),
the median of the demanded responses at each intensity level is defined as central value, and the dispersion
caused by the uncertainty in the median response will be defined by its β-value [34], that is, the standard

deviation of the natural logarithm of the mean demanded responses conditioned on Sa T base 1 ; 5% .
Suppose that ln Δi;jmax be the natural logarithm of the demanded response for a given intensity
 base 
Sa T 1 ; 5% , instance j (j = 1,…,N) and record i (i = 1,…,Q); AVG (lnΔmax ) is the overall mean of
the logarithmic demands. Then, ln Δjmax; 50% is the logarithm of their median value for structure j,
and AVG (lnΔmax, 50 %) is the mean of the corresponding natural logarithm of the medians over all
realizations. The central value (overall median), CΔ, and its dispersion, βU, due to parameter
uncertainty can be obtained as

 
C Δ ¼ med j Δjmax; 50% (3)

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u    2
u∑N lnΔj 
t j¼1 max; 50% AVG lnΔ max; 50%
βU ¼ (4)
N1

where ‘AVG’ and ‘medj’ are the mean and median operators for the overall samples j.
The two other approaches used for moment estimation are point-estimate (PEM) and first-order
second-moment (FOSM) methods. They refer to the process of estimating a parameter from a
probability distribution, based on observed data from the distribution. They involve the use of
sample data to calculate a single value, which is to serve as a ‘best guess’ or ‘best estimate’ of a
random unknown population parameter. Both methods require only 2K + 1 simulations, where K is
the number of random parameters, to produce a central value (overall median, C  Δ) and its

corresponding dispersion of the demanded responses (Δmax) at every intensity level Sa T base1 ; 5% .
This dispersion (βU) is due to the strength parameters uncertainty. The detailed description of these
methods is omitted, and the readers are referred to References [13, 35–38] for their detailed
description and formulation.

5.3. Discussion
The results of all three methods (Monte Carlo with LHS, PEM, and FOSM) for the overall median IDA
curve in the U/S and D/S directions appear in Figure 10 along with the simple overall mean and base-
case IDA curves. It is noted that these results are based on independent probability distributions for
parameters of tensile constitutive model of mass concrete; however, the random nature of the
seismic loadings is taken into account by theanalysis of 10 earthquakes. The FOSM and PEM show
1 ; 5% = 0.5 g for both directions. For higher limit-
excellent consistency with the LHS up to Sa T base
states in the D/S direction, the prediction is more accurate than the U/S direction. The LHS, mean
and base-case curves are approximately the same for entire intensity range. This shows that the
median-parameter model will produce the median seismic performance. Thus, the gravity dam,
which is analyzed utilizing the central values (means or medians) for the parameters, accurately
estimates the central value (mean or median) response. The level of similarity between the results is

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe
SEISMIC SENSITIVITY AND UNCERTAINTY ANALYSIS OF GRAVITY DAMS 55

Figure 10. Median IDA curves along with the overall mean and base-case IDA curves in (a) D/S direction;
(b) U/S direction.

such that it ensures reliable usage of the conclusions in practice. The similarity of the overall mean and
LHS IDA curves shows that the results are evenly distributed.
Thus, on average, the overall median/mean capacity will be approximately the same as the capacity
of the median/mean model in two opposite directions. So, it would be advisable to use the base-case or,
even better, Monte Carlo simulation with LHS for a reasonable estimate of the mean/median when
performing a limited performance estimation using dynamic runs at a single IM-level. It should be
kept in mind that here only an example of a gravity dam is presented that does not include all
possible sources of nonlinearity and uncertainty. Therefore, any drawn conclusions should be
viewed in light of these limitations.  
Because there is single value of Δmax-demand for each value of Sa T base ; 5% , the approximate PEM
  1
and FOSM-derived median Δmax-given-Sa T base 1 ; 5% results in Figure 10 can be thought of as IDA
curves in the classical form [11].
The estimates of βU obtained by the three methods for three random parameters into the D/S and U/S
directions appear in Figure 11. The strength parameters uncertainties smoothly changes in entire
intensity range. It has a near zero value in the elastic range for U/S direction; however, it is not the
case for D/S direction. It is because of the change in elastic properties of the model’s material
whose effect is more prominent for motion into the D/S direction. For both directions, the PEM
method has more dispersion, and the FOSM method has less dispersion with respect to the Monte
Carlo method with LHS. In general, the dispersions are less than 0.3, which shows a usable estimate
of βU even using only 2 × 3 + 1 = 7 sample instances rather than 30 for LHS. That is almost an order-
of-magnitude reduction in computations at the cost of a reasonable error.
The dispersion due to seismic loading uncertainty,
 which is measured based on the record-to-record
variability of Δmax given the Sa T base
1 ; 5% , can be represented by its β-value, that is, by the standard
deviation of the natural logarithm of the EDP given the IM. This can be directly estimated from the
sample IDA curves of the base-case, or it can be approximated from the corresponding fractile IDAs
as [13]:

1 
βR ≈ max  lnΔmax
lnΔ84% 16%
(5)
2

max and Δmax are the 84% and 16% fractiles of the Δmax-demand.
where the Δ84% 16%

The total dispersion, βRU, which includes all considered sources of uncertainty, here is affected by
the strength parameters uncertainty βU and the seismic loading uncertainty βR [34]. It can be
approximated as the square root of the sum of the squares of βU and βR [13, 39]:

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe
56 M. ALEMBAGHERI AND M. SEYEDKAZEMI

 
1 ; 5% using LHS, PEM, and
Figure 11. Strength parameters uncertainties at each level of intensity Sa T base
FOSM for motion into (a) D/S direction; (b) U/S direction.

Figure 12. Strength parameters, seismic loading, and total dispersion of the results of the Monte Carlo
simulation with LHS in (a) D/S direction; (b) U/S direction.

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
βSRSS
RU ¼ β2R þ β2U (6)

As shown in Figure 12, the total dispersion enables one to compare the relative contribution of
strength parameters and seismic loadings uncertainties to the total dispersion.
The seismic loading variability has the same trend as the strength parameters uncertainty. Both the
βU and βR generally decrease and increase at higher limit-states in D/S and U/S directions, respectively,
as the dam approaches global dynamic instability. It shows that the uncertainty caused by all
parameters have different variation in two opposite directions.

6. CONCLUDING REMARKS

The seismic performance sensitivity and uncertainty analysis of gravity dams is performed selecting
the tallest non-overflow monolith of Pine Flat gravity dam as case study. The dam is modeled along
with its reservoir using finite elements in 2D plane-stress analysis. The nonlinear tensile behavior of

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe
SEISMIC SENSITIVITY AND UNCERTAINTY ANALYSIS OF GRAVITY DAMS 57

concrete is modeled using plastic–damage approach assuming a bilinear constitutive stress–strain


curve, which is controlled by the three main parameters: undamaged elastic modulus, tensile
strength, and ultimate inelastic tensile strain. The sensitivity and uncertainty analysis are performed
within the framework of SPO and IDA. For SPO analysis, a prescribed method and for IDA, a
series of 10 earthquake ground motions are utilized. By perturbing each of the three main
parameters of the concrete tensile constitutive behavior, one at a time, above and below a
preassigned mean value, the sensitivity of the dam’s response to each parameter is examined.
Because of unsymmetric shape of the dam, these examinations should be separately performed in
two opposite directions of upstream (U/S) and downstream (D/S). It was found that the variation of
elastic modulus mostly affects the behavior of the dam during the elastic region. It changes the
deformation demands but does not affect the loading capacity. Variation of tensile strength makes
significant changes to the post-yielding part of the behavior, both on deformation demands and
loading capacities. The effect of ultimate inelastic tensile strain is more apparent when loading is
applied to the D/S, but from practical point of view, its effect is not important.
The three main parameters of the tensile behavior create the source of the strength parameters
uncertainties, while the spectral acceleration of the selected earthquakes forms the source of seismic
loading uncertainty in the probabilistic analysis. The Monte Carlo simulation with LHS was used to
generate the population N = 30 instances of the dam–reservoir system. The results showed that
overall median/mean response of the dam models using Monte Carlo with LHS will approximately
same as the response of the median/mean model in two opposite directions. The point-estimation
and first-order second-moment techniques also provide relatively accurate estimation of the mean
response at low to moderate shaking intensities. Low dispersion of the results is another indicator of
good correlation of the response results. So, one can reliably use the mean model to analyze the
mean response of the dam with probabilistic distributed parameters for tensile behavior of mass
concrete, which is the most influencing factor controlling the nonlinear behavior and final failure
mechanism of concrete gravity dams.
This research could be extended considering the following perspectives: (i) the correlation between
different strength parameters; (ii) hydromechanical analysis of crack opening; (iii) the uplift pressures
generated within the cracks; (iv) varying spatial distribution of the strength parameters over the
structure; and (v) studying the behavior of cracked (jointed) dams.

REFERENCES
1. Ghobarah A. Performance-based design in earthquake engineering: state of development. Engineering Structures
2001; 23:878–884.
2. Dolsek M. Incremental dynamic analysis with consideration of modeling uncertainties. Earthquake Engineering and
Structural Dynamics 2009; 38:805–825.
3. Hwang HHM, Jaw JW. Probabilistic damage analysis of structures. ASCE Journal of Structural Engineering 1990;
116:1992–2007.
4. USACE. Gravity dam design. United States Army Corps of Engineers, Engineering Manual 1110-2-2200, 1995.
5. Leclerc M, Leger P, Tinawi R. CADAM user’s manual. Department of Civil, Geological and Mining Engineering.
Ecole Polytechnique de Montreal, Quebec, 2001.
6. Luco N, Cornell CA. Effects of connection fractures on SMRF seismic drift demands. ASCE Journal of Structural
Engineering 2000; 126:127–136.
7. Ibarra LF. Global collapse of frame structures under seismic excitations. PhD Dissertation, Department of Civil and
Environmental Engineering, Stanford University, Stanford, CA, 2003.
8. Porter KA, Beck JL, Shaikhutdinov RV. Sensitivity of building loss estimates to major uncertain variables.
Earthquake Spectra 2002; 18:719–743.
9. de Araujo JM, Awruch AM. Probabilistic finite element analysis of concrete gravity dams. Advances in Engineering
Software 1998; 29:97–104.
10. Rubinstein RY. Simulation and the Monte Carlo Method. John Wiley & Sons: New York, 1981.
11. Vamvatsikos D, Cornell CA. Incremental dynamic analysis. Earthquake Engineering and Structural Dynamics 2002;
31:491–514.
12. Liel AB, Haselton CB, Deierlein GG, Baker JW. Incorporating modeling uncertainties in the assessment of seismic
collapse risk of buildings. Structural Safety 2009; 31:197–211.
13. Vamvatsikos D, Fragiadakis M. Incremental dynamic analysis for estimating seismic performance sensitivity and
uncertainty. Earthquake Engineering and Structural Dynamics 2010; 39:141–163.
14. Chopra AK, Goel RK. A modal pushover analysis procedure for estimating seismic demands for buildings.
Earthquake Engineering and Structural Dynamics 2002; 31:561–582.

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe
58 M. ALEMBAGHERI AND M. SEYEDKAZEMI

15. ICOLD. Dam failures - statistical analysis. Bulletin 99, International Commission on Large Dams,1995.
16. Altarejos-Garcia L, Escuder-Bueno I, Serrano-Lombillo A. Estimation of the probability of failure of a gravity dam
for the sliding failure mode. 11th ICOLD Benchmark Workshop on Numerical Analysis of Dams, Vol. 1, Theme C,
Valencia, 2011.
17. Carvajal C, Peyras L, Bacconnet J, Becue P. Probability modeling of shear strength parameters of RCC gravity dams for
reliability analysis of structural safety. European Journal of Environmental and Civil Engineering 2009; 13:91–119.
18. Ellingwood B, Tekie PB. Fragility analysis of concrete gravity dams. ASCE Journal of Infrastructure Systems 2001;
7:41–48.
19. Fell R, Bowles DS, Anderson LR, Bell G. The status of methods for estimation of the probability of failure of dams
for use in quantitative risk management. 20th ICOLD Congress, Beijing, China, 2000.
20. Alembagheri M, Ghaemian M. Seismic assessment of concrete gravity dams using capacity estimation and damage
indexes. Earthquake Engineering and Structural Dynamics 2013; 42: 123–144.
21. http://en.wikipedia.org/wiki/Pine_Flat_Dam [10 November 2013].
22. Lee J, Fenves GL. A plastic-damage concrete model for earthquake analysis of dams. Earthquake Engineering and
Structural Dynamics 1998; 27:937–956.
23. Lee J, Fenves GL. Plastic-damage model for cyclic loading of concrete structures. ASCE Journal of Engineering
Mechanics 1998; 124:892–900.
24. Chopra AK, Chakrabarti P. Earthquake analysis of concrete gravity dams including dam-water- foundation rock
interaction. Earthquake Engineering and Structural Dynamics 1981; 9:363–383.
25. El-Aidi B, Hall JF. Non-linear earthquake response of concrete gravity dam part 1: modeling. Earthquake
Engineering and Structural Dynamics 1989; 18:837–851.
26. Ghrib F, Tinawi R. An application of damage mechanics for seismic analysis of concrete gravity dams. Earthquake
Engineering and Structural Dynamics 1995; 24: 157–173.
27. Hall JF. Problems encountered from the use (or misuse) of Rayleigh damping. Earthquake Engineering and
Structural Dynamics 2006; 35:525–545.
28. Benjamin JR, Cornell CA. Probability, Statistics, and Decision for Civil Engineers. McGraw-Hill: New York, 1970.
29. McKay MD, Conover WJ, Beckman R. A comparison of three methods for selecting values of input variables in the
analysis of output from a computer code. Technometrics 1979; 21:239–245.
30. Iman R. Latin hypercube sampling. In: Encyclopedia of Statistical Sciences. Wiley: New York. DOI: 10.1002/
0471667196.ess1084.pub2.
31. Iman RL, Helton JC, Campbell JE. An approach to sensitivity analysis of computer models, Part 1. Introduction,
input variable selection and preliminary variable assessment. Journal of Quality Technology 1981; 13:174–183.
32. Melchers RE. Structural Reliability Analysis and Prediction (2nd edn). John Wiley & Sons. ISBN 0-471-98771-9, 1999.
33. Broding WC, Diederich FW, Parker PS. Structural optimization and design based on a reliability design criterion.
Journal of Spacecraft 1964; 1:56–61.
34. Cornell CA, Jalayer F, Hamburger RO, Foutch DA. The probabilistic basis for the 2000 SAC/FEMA steel moment
frame guidelines. ASCE Journal of Structural Engineering 2002; 128:526–533.
35. Lehmann EL, Casella G. Theory of Point Estimation (2nd edn). Springer. ISBN 0-387-98502-6, 1998.
36. Rosenblueth E. Point estimates for probability. Applied Mathematical Modeling 1981; 5:329–335.
37. Lee TH, Mosalam KM. Seismic demand sensitivity of reinforced concrete shear-wall building using FOSM method.
Earthquake Engineering and Structural Dynamics 2005; 34:1719–1736.
38. Pinto PE, Giannini R, Franchin P. Seismic Reliability Analysis of Structures. IUSS Press: Pavia, Italy, 2004.
39. SAC/FEMA. Recommended seismic design criteria for new steel moment-frame buildings. Report No. FEMA-350,
SAC Joint Venture, Federal Emergency Management Agency, Washington, DC, 2000.

Copyright © 2014 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2015; 44:41–58
DOI: 10.1002/eqe

You might also like