Advances in Engineeriq SoNare Vol. 29, No. 2, pp. 97-104.

1998
C 1998 Elsevier Science Ltd
Printed in Great Britain. All rights reserved
PII:SO965-9978(98)00052-O 096%9978/98/$19X42 + 0.00
ELSEVIER

Probabilistic finite element analysis of concrete

gravity dams
Jo& Milton de Araiijo”?* & Armando Miguel Awruchb
aDepartment of Materials and Construction, University of Rio Grande-FURG, Campus Carreiros, 96200-000 Rio Grande, RS, Brazil
‘Graduate Program in Civil Engineering, UFRGS, Porto Alegre, RS, Brazil

A methodology for the probabilistic analysis of concrete gravity dams is presented.

Concrete properties and seismic excitation are considered as random variables. The
seismic excitation is considered as a non-stationary stochastic process which is
artificially generated. Concrete properties have random variations over the spatial
domain. Structural response is obtained employing the finite element method to solve
the equations of motion of the coupled system dam-reservoir-foundation. Structural
safety is evaluated with respect to the main failure modes (cracking, concrete crushing
and sliding at the dam-foundation interface) using the Monte Carlo method. 0 1998
Elsevier Science Ltd. All rights reserved

Key wordA: concrete gravity dams, probabilistic, structural analysis, fluid-structure-

foundation interaction, earthquake response

1 INTRODUCTION structural safety must be obtained by analysing the dynamic

behaviour of the coupled system dam-foundation-
Loads due to seismic excitation are one of the most reservoir.
important actions that must be considered in concrete dam Seismic actions have essentially stochastic characteris-
design, even in regions of low seismic activity. Traditional tics, therefore they should be considered as probabilistic
procedures for concrete dam design consider forces due to loads. Moreover, material properties and structural dimen-
seismic excitation as equivalent static forces which are sions have random variations in the spatial domain. Thus, a
obtained by the addition of part of the reservoir to the struc- probabilistic dynamic analysis of the coupled system
tural mass. However, this pseudo-static formulation is not dam-reservoir-foundation is necessary.
satisfactory to reproduce the dynamic behaviour of the Several methods for the probabilistic structural analysis
coupled system dam -foundation-reservoir.’ had been studied in the last years, such as the Direct Monte
In fluid-structure interaction problems, it is necessary to Carlo Method, Monte Carlo with Neumann expansion and
analyse the fluid motion due to structural vibration and vice perturbation techniques. ’ i
versa. In the case of dam-reservoir interaction, the fluid The Direct Monte Carlo Method is the most simple and
may be considered as inviscid and the motion is assumed clear way to accomplish a probabilistic analysis, and for that
to be irrotational with small amplitude. reason is widely used.‘*-17 In this method, material proper-
Several fluid-structure interaction models have been ties, loads and other random variables are introduced by
developed in the last years, and they differ with respect to digital simulation, with any significative modification of
the fluid formulation. The basic unknowns in the models can the algorithm used in the deterministic analysis. Moreover,
be the fluid displacements,273but a mixed formulation using the Monte Carlo Method is statistically consistent and may
pressure and displacements components as unknowns may be be employed to test other techniques.““’
also employed.4.5 Scalar formulations with the hydrodynamic The main drawback of the Direct Monte Carlo Method is
pressures as basic unknowns have been widely used.6-9 that a structural stiffness matrix must be defined for each
As the hypothesis of rigid foundation may lead to an sample to be simulated, and this may lead to a high compu-
erroneous evaluation of the structural safety, it is necessary tational cost (depending on the problem size). In these cases,
to consider foundation flexibility. lo Consequently, the method can be improved by working with the mean
stiffness matrix and introducing deviations with respect to
*Author to whom all correspondence should be addressed the mean value in the load vector.‘”
98 J. Milton de Aralijo, A. M. Awruch

An alternative to reduce the computational effort representing the accelerations of the free-field motion of the
demanded by the Direct Monte Carlo Method is to use ground surface, is given by the expression
simulations with Neumann expansions.i6 In this case, simu-
lations are employed to generate the fluctuating part of the X(t) = 3(t) f cos(w,t - 9,) (1)
stiffness matrix, but back-substitution is accomplished using II=1
a constant stiffness matrix evaluated with mean values of where h(t) is a time-dependent deterministic function, A is a
the material properties; this matrix is decomposed only Gaussian random variable representing the amplitude of
once, at the beginning of the process. However, if the X(t) and h is a factor introduced in order to obtain an
variation coefficients of the properties are not small, this absolute maximum value of X(t) equal to A.
method requires higher order expansions, increasing the Frequencies o, are equally spaced at the interval [O,w,],
computational cost. 17,*0 with o, being the maximum value of the excitation
Another solution is obtained by expanding the random frequency. The phase angles r$, are independent random
variables of the problem using Taylor series. In this case, variables and with a uniform distribution in the interval
mass, stiffness and damping matrices, together with the load [0,27r], i.e. q5, = 2nb,, where b, is a random number with
vector, displacements, velocities and accelerations, are uniform distribution in the interval [O,l]. The number of
expanded in Taylor series. Introducing these expansions in terms in the series is 200.
the equations of motion, expected values and standard Values of the Gaussian random variable A can be
deviations of the structural response can be obtained without obtained through random numbers with uniform distribution
simulations. ‘7,18,21-26 These methods, in spite of their as is usual.31 However, in order to avoid peak values of A
efficacy in terms of computational effort, may be very with a small probability of occurrence, a confidence interval
inaccurate when the variation coefficients of the random of 98% is defined to disregard these extreme values.
variables have high values. Moreover, difficulties in deter- The function h(t) may be selected in order to obtain a
mining the terms of the series increase with the complexity desirable number of shocks in the artificial earthquake
of the constitutive model and loads. record. The function adopted in this work, which reproduces
Safety verification of concrete gravity dams can be based a single shock, is given by
on three criteria: cracking, concrete crushing and sliding at
the dam-foundation interface. A consistent evaluation of h(t) = e - at-Tp -e- olTP (21
the structural safety level requires the determination of the In eqn (2), TP = Td/2 is the time where the peak value of
probability that these failure modes will be reached or, in X(t) occurs, with Td representing the duration time of the
other words, the failure probability. This probability can be
estimated if the probability distribution of some random
variable representing a given safety limit for the structure 1.2 'I

is known. Unfortunately, it is not always possible or prac-

- 0.8-
tical to obtain this probability distribution. However, using
"u,
the Monte Carlo Method an approximated distribution can - OA-
be obtained, allowing a reliable evaluation with respect to .E.
defined failure modes. - 0.0
=:
In this work, the Direct Monte Carlo Method is employed x
0.4-
to evaluate the safety of gravity concrete dams under
seismic actions. The seismic action is considered as a 0.8 I ' I - I . r . 1
0 4 :me 12 16 20
non-stationary stochastic process, while concrete properties
are taken as random variables. A pressure formulation is (s)
adopted for the fluid and eight-node isoparametric finite
elements are used for space discretization.

2 THE RANDOM VARIABLES GENERATION

2.1 The earthquake artificial generation

An earthquake is essentially a non-stationary stochastic pro-

cess (with time dependent characteristics). Several methods
for the digital generation of this process have been pre-
sented.27-30Generally, a non-stationary process is obtained a (r) ” 20
by the product of a stationary process and a deterministic time
function of time (called the envelope function).
In this work, the non-stationary stochastic process X(t), Fig. 1. Samplesof the non-stationary process.
 Probabilistic analysis of concrete gravity dams 99

event. Scattering of X(t) in time is defined by cx = c/Td, 3 VERIFICATION OF THE STRUCTURAL SAFETY
where c is a constant.
Thus, the artificial seismic excitation is characterized by Quantilication of dam safety under seismic actions is still an
two random components: the amplitude and the phase open question to be discussed. A safety factor less than 1.0
angles. during a short time period is not enough to say that the
In Fig. 1, two samples of the process X(t) are shown. The structure is not stable.’ Moreover, local failure does not
mean value of the amplitude is O.lg, where g = 9.81 m s-’ mean the global failure of the structure. However, safety
is the gravity acceleration, and a variation coefficient equal factors are good indicators of the structure’s capability to
to 0.20 is adopted. Two frequency contents are considered support external loads.
and Td = 20 s. The analysis of the minimum values of safety factors,
together with the cracking intensity of the dam, gives
2.2 Random generation of the material properties general information about the damage caused by the seismic
action. The safety concept is, in fact, directly related to the
Concrete properties have random variations over the damage provoked by external action on the structure.9
structural domain. A generic property n can be written as In this work, structural safety verification is based on the
three criteria: cracking, concrete crushing and sliding at the
rl = %(l +4x)) (3) dam-foundation interface. Safety factors against these
failure modes are time varying due to the dynamic charac-
where no is the mean value of the property and a(x) repre- teristic of the excitation. Only smaller values during dam
sents the fluctuations around the mean value, with x being vibration are considered for each simulation.
the position vector on the structural domain. Thus, if 6 represents a generic safety factor, the limit state
The fluctuating part a(x) has a mean value equal to zero is defined by the condition 8 < 1. Then, if p(0) is the prob-
and an autocorrelation function given by ability density function, the probability to reach the limit
state (failure probability) is defined by-
(4)

where S is the distance between two points x and x + s on (3

the structural domain, V, is the variation coefficient of the
property and d is a scale factor. noting that 0 is always a positive number.
Using the finite method, the material properties may be Then, if p(0) is known, eqn (5) may be integrated to
taken at the centre af the element and considered as being obtain the failure probability of the structure subject to a
constants over the element domain. Thus, if there are N seismic action with assumed characteristics. The structural
elements in the finite element mesh, it is necessary to gen- reliability, given by the probability that the limit state will
erate N values of the variable a(xi), where xi represents the not be reached, is obtained by calculating R = 1 - pf.
coordinates at the centre of the element. The probability density function of the safety factors may
A vector a = (at,.. .,a,,,]T of autocorrelated random be estimated using the Monte Carlo Method, setting the
variables with mean values equal to zero can be generated values of 8 (obtained through simulations) in increasing
as a = LZ, where L is the lower triangular matrix obtained order. The cumulative probability of the factor of order n
by Cholesky’s decomposition of the covariance matrix and is estimated as nI(NS + l), where NS is the adopted number
Z is a vector containing N uncorrelated Gaussian random of simulations.33 This probability distribution is approxi-
variables. mated by some theoretical distribution that allows us to
After the vector a has been generated, eqn (3) is evaluate eqn (5).
employed to obtain the properties in each finite element. The result obtained is only an estimation of the structural
Peak values of the properties are avoided establishing a failure probability. In fact, several uncertainties are present
confidence interval of 98% to disregard the simulation. in the problem as inherent uncertainties in material and
The random properties adopted in this work are the loading representation, including material parameters, as
compressive strength, the tensile strength and the Young’s well as mathematical model uncertainties. Thus, this failure
modulus of the concrete and the adhesion at the dam- must be viewed as a probability which is conditioned by the
foundation interface. The concrete compressive strength is formulated hypothesis.
the basic random variable whose distribution over the
structural domain is obtained with the procedure described
previously. Others random properties are obtained through 4 CONCRETE MODEL AND SAFETY FACTOR
deterministic correlations with the compressive DETERMINATION
strength.‘0.32 The effects of strain rates in concrete proper-
ties are taken into account32 and the Poisson’s coefficient is Compressive stresses in concrete gravity dams have, gener-
taken as being equal to 0.2 and is assumed as a deterministic ally, relatively low values. Therefore, in these structures a
property. linear elastic behaviour of the concrete in compression can
loo J. Milton de Aratijo, A. M. Awruch

be assumed. Cracking is the major source of non-linearity

which must be considered.
A constitutive model for concrete, with smeared cracks, is
used in this work.34 It is assumed that a plane state of stress
and a linear elastic behaviour exist before the onset of the
cracking process. The model is based on the equivalent
strength concept to obtain mesh independent results. The
equivalent strength is obtained considering the stress field
on the finite element domain. This stress field is defined by
means of the known stresses in the integration points
through an interpolation process (employing the least
squares method). Using 2 X 2 integration points it is Fig. 2. Dam-reservoir-foundation system.
possible to define the equivalent tensile strength and the
equivalent compressive strength for concrete. Details of
the following system of differential equations is obtained
this formulation are given in Refs ‘“*34.
at the time interval [tm,tm+,]
The safety factor against cracking in any point of a
generic finite element is given by clu$ul, where uI is the EAii + AAi, + HAP = AFf @I
principal tensile stress and ulue is the equivalent biaxial
tensile strength. Similarly, the safety factor against crushing where AP is a vector with nodal values of the pressure
is given by uzuJuz, where uzueis the equivalent biaxial com- increments and a dot indicates a derivative with respect
pressive strength and u2 is the biggest principal compressive to time.
stress. Matrices for the fluid are given by8
The safety factor against sliding, S,, is given by
E= -$ nNrNdfl+ INTN dr (9)
s rs nTg
(6)
A=;c n NTN dr (10)
where r is the shear stress at the dam-foundation interface
aNT dN dNT 6’N
and integrals are computed along the contact line, taking H= (11)
two integration points. anax+ayYg da
The shear strength, rr, is evaluated by means of the The vector AFf, in which the coupling terms of the dam-
Mohr-Coulomb criterion and is given by foundation system are included, is given by
l7,l=v-p(u,+u) (7) AFf =
where Y is the adhesion, p is the friction coefficient, u
is the uplift pressure and u, is the normal stress on the
- LNTnT(fim + Al$)Ap + p,NTnTAw) dI’
interface.
Stress r and u, are obtained using the interpolation pro- (12)
cess previously described. The safety factor against sliding
is obtained assuming a triangular uplift pressure diagram In these equations, pm and W, are the specific mass of the
along the foundation contact and p = 1.0. The cohesion in fluid and the
.. total acceleration vector of the interface at time
an integration point is disregarded if 171> lr,l. t = t,, AW is the vector of total acceleration increments at
the time interval, N is a matrix with the shape functions of
the fluid elements, Ap is the increment of the hydrodynamic
5 DAM-RESERVOIR-FOUNDATION DYNAMIC pressure at the time interval, C is the velocity of the acoustic
INTERACTION wave, n is the outflow normal vector of the fluid boundary
and g includes the components of the gravity acceleration.
The dam-reservoir-foundation system to be analysed is In order to simplify the model and due to the lack of
indicated in Fig. 2. experimental results on the damping mechanisms of mass
Boundary conditions for the fluid are obtained assuming concrete dams under seismic loading, for the dam-
the presence of gravity waves in rs (free surface), the foundation system it is assumed that the damping is
Sommerfeld radiation condition in I’r(reservoir boundary proportional to the stiffness. Furthermore, the same damp-
at the far field) and applying the Newton’s law in rt ing proportionality factor is used for both the concrete dam
(fluid-dam and fluid-foundation interfaces). and the foundation, although this is not necessary. The
Considering a slightly compressible inviscid fluid and damping matrix is evaluated with mean values of the
neglecting convective terms in the momentum equations, material properties and, consequently, it is considered as a
 Probabilistic analysis of concrete gravity dams 101

deterministic matrix. Then, it can be shown” that the solved to obtain increments of relative displacements,
equation of the motion of the dam-foundation system can denoted by AU,‘. Applying Newmark’s method, AC, may
be written as be obtained and the increments of relative accelerations at
the solid-fluid interface are separated in a vector A\jir;.
MAti; + , + yK,dJ:,, + , + K,AU; + 1 = AF (13) With A\jti, = Al%; + AtiT the vector AFr, given in eqn
(12), may be assembled and eqn (8) is solved to obtain
where M and K, are the mass and elastic stiffness matrices nodal pressure increments API in the fluid domain. After
of the coupled dam-foundation system and Auk+, , API is known, we go to the next time step. At the end of
AfJ; + I and AU,,, r are the vectors of relative acceleration, each time interval it is necessary to define the unbalanced
velocity and displacement increments at the current time force vector pm.
interval. This procedure uses a single loop to deal with both the
Matrix K, is evaluated with mean values of the material non-linearity and the coupling. Although Newmark’s
properties and y is the damping proportionality factor. method is known as an unconditionally stable algorithm,
The vector AF is given by the adopted procedure may still be conditionally stable.
Thus, a preliminary test is necessary to define the time
AF = - mRA$? - rI@TiiAp dP + (o, (14) step which preserves accuracy and stability in the current
problem. For the system analysed in this work, it was found
where m is the dam mass matrix, R is a rigid body trans- that the time step must be smaller that a/( low,), where w, is
lational matrix corresponding to the degrees of freedom of the maximum frequency of the seismic excitation.
the dam, ii is the outflow normal of the structure and
foundation at the interface with the fluid and N is a
matrix with the shape functions of the structural and 6 ANALYSIS OF A DAM-RESERVOIR-
foundations elements. FOUNDATION SYSTEM
In this work, the same shape functions are adopted for
fluid and solid elements, i.e. N = fi. The analysis of the right transition wall of the Tucurui Dam
The vector Al%’ contains the stochastic acceleration in Brazil is presented in this work. This wall is a gravity
increments of the free-field motion at the ground surface, structure of mass concrete with two blocks of approximately
as defined in Section 2, and pin is the unbalanced force 30 m in length, separated by a contraction joint.
vector in t,. Vector IO,,,takes into account the non-linearity The dam-reservoir-foundation system is shown in
due to the cracking and deviations of the mean values of the Fig. 3.
concrete properties. This vector is evaluated with random The mean value of the concrete uniaxial compressive
properties obtained in each simulation. Using eqn (13), strength was taken as fC = 16.70 MPa. For the rock, the
matrix K, remains constant and non-linearities and devia- adopted Young’s modulus and Poisson’s coefficient were
tion of the mean values of the properties are included in the taken as 20000 MPa and 0.25, respectively. Although
load vector. previous cracking may introduce non-linear behaviour, the
A lumped mass matrix35 of the dam-foundation system is rock at the foundation system was assumed to have linear
used in this work. This lumped mass matrix is obtained from behaviour and a plane state of strain was adopted to analyse
the consistent mass matrix in order to preserve the total this part. For the dam-foundation system, 636 iso-
mass of the system. The consistent mass matrix is obtained parametric elements of eight nodes were used in the finite
using 3 X 3 integration points but, for the stiffness matrix, element mesh. The total number of nodes and total number
2 X 2 integrations points are employed. of displacement degrees of freedom were 2049 and 4098,
Eqns (8) and ( 13) are integrated in time using Newmark’s respectively. For the reservoir, 330 isoparametric elements
method. At time t, = 0 the fluid pressure is taken as the
same value of the hydrostatic pressure, while its first and
second derivatives with respect to time are taken as P=O
and P = 0, respectively. For the dam-foundation system,
initial displacements are obtained from the static analysis,
whereas velocities and accelerations are taken as equal to
zero for all the finite element domain. In the initial static
analysis, the dam is loaded with the self-weight and with the
hydrostatic pressure. Only this last effect is considered to be
acting over the foundation.
The following procedure is adopted to solve the system of
coupled equations: at time t i = 0, the pressure increment AP
is taken equal to zero and an increment of prescribed accel-
erations A%‘: is applied to the dam-foundation system. The
load vector AF is assembled using eqn (14) and eqn (13) is Fig. 3. Analysed system.
102 J. Milton de Arakjo, A. M. Awruch

with eight nodes were used. The total number of nodes was
1077 and, in this case, it is equal to the total number of
pressure degrees of freedom.
The first undamped natural frequency of the dam-
foundation system is 18.30 rad s-l. The damping matrix iii
was obtained specifying a typical damping ratio of 5% in g 3 Trig - 2.78
this vibration mode.’ The damping ratio of cracked finite
0 i
elements might be reduced, but this procedure was not con-
sidered. The speed of the acoustic wave was taken as 5 2
Tii -II
1434 m s-l.
The vector of rigid body accelerations induced by the -0
5

@1
.A

earthquake ground motion was taken as t2 cracking 0.42

~ = {Xi(t), 0.5Xj(t)JT (15)

2
W I--
00
0 10 20 30 40 50
where Xi(t) and Xj(t) are the non-stationary stochastic Number of simulations
processes described in Section 2.1.
Deterministic analysis has indicated that several para- Fig. 5. Expected values of the safety factors as a function of the
meters, such as the seismic characteristics, the flexibility number of simulations.
of the foundation, the adopted uplift pressure diagram and
7 RESULTS
the fluid interaction, may influence structural safety. For the
seismic excitation, the main parameters are the frequency
The safety factors against sliding, concrete crushing at the
content and the amplitude. Previous studies have indi-
toe and concrete cracking at the heel of the dam for 50
cated that for the Tucurui Dam a critical earthquake is
simulations are indicated in Fig. 4.
characterized by a maximum frequency equal to
Variations of expected values and standard deviation of
5a rad s-l approximately, which is very close to the first
these safety factors as a function of the sample size are
natural frequency. shown in Figs. 5 and 6. It can be observed in these figures
Thus, safety of the transition wall in the Tucurui Dam was
that expected values and standard deviations have decreas-
analysed considering an earthquake which lasts 5 s, with
ing fluctuations as the number of simulations increases.
frequencies contained in the interval [0,57r] rad SC’ and a The convergence of the process can be verified by
mean amplitude A, = O.lg. The variation coefficient for
analysing fluctuations of the index /3 = ~&e~, where PLg
the amplitude was taken equal to 0.2. and IJO represent the expected value and the standard
For the uniaxial compressive strength of the concrete, a deviation of the safety factor, respectively. As the failure
variation coefficient of 0.2 was adopted, and a high value for probability is related to the index 0, this index can be used to
the scale factor was taken to ensure a total correlation into determine the number of simulations which are necessary to
the same concrete layer. Thirty-five concrete layers were obtain the desired convergence. As is shown in Fig. 7, 50
placed during the construction stage.

E
.$ 0.6
3 crushing

z!

0 so
0 10 20 30 40 50 Number of simulations
Number of simulations
Fig. 6. Standard deviations of the safety factors as a function of
Fig. 4. Safety factors obtained in each simulation. the number of simulations.
 Probabilistic analysis of concrete gravity dams 103

16 safety factor, the probabilities of occurrence of the failure

modes can be estimated with eqn (5).
The analysis performed for the right transition wall of the
Tucurui Dam indicated that the probabilities of sliding at the
dam-foundation interface and concrete crushing in every
point of the dam are less than 10e4. This shows a satisfac-
tory reliability against these failure modes, although the
38 model uncertainties must be considered.
-0
C
6 On the other hand, reliability with respect to cracking on
the upstream face is very small, indicating a high probability
of cracking due to an earthquake with the characteristics
considered here. However, this is only verified at the heel
of the dam. In the other regions of the dam, the reliability is
0 ' I ' I ' I ' I significantly higher. These same conclusions were obtained
0 ‘IO 20 30 40 50 with the deterministic analysis (considering mean material
Number of simulations properties).

Fig. 7. Variations of the index @as a function of the number of

simulations.
8 CONCLUSIONS
simulations are enough to estimate the failure probability
with a reasonable precision. A methodology to verify the safety of concrete gravity dams
With the values obtained from 50 simulations, the prob- subject to seismic actions was developed in this work. The
ability distributions of the safety factors can be estimated as formulation takes into account the dynamic interaction of
was indicated in Section 3. The cumulative distribution of the system dam-reservoir-foundation. Safety factors
the safety factor against cracking at the heel of the dam is against sliding, cracking and concrete crushing were used
given in Fig. 8 and compared with the cumulative distribu- to define the structural performance. The stochastic charac-
tion corresponding to a Gaussian variable. In this figure, the teristics of the seismic excitation as well as the random
points were obtained through simulations, while the line is variations of the concrete properties were considered
referred to a Gaussian distribution. using the Monte Carlo Method. Through simulations it
Fig. 8 shows good agreement between the cumulative was possible to evaluate the structural safety with respect
probabilities obtained by simulations and that correspond- to the main failure modes.
ing to a Gaussian variable. The Kolmogorov-Smimov test The Monte Carlo Method, in spite of its greater require-
showed a good approximation with a significance level of ments in terms of computational effort with respect to semi-
5%. Similar results were obtained for the probability analytical methods, can be easily employed to solve any
distribution of the others safety factors. problem with probabilistic characteristics. The main draw-
Therefore, considering a Gaussian distribution for the back of this method is the fact that the stiffness structural
matrix must be defined for each simulation.
To overcome this drawback, an alternative procedure was
1.0 -
adopted in this work. In this procedure, the stiffness matrix
with mean values of the material properties was used.
B 0.8- Fluctuations around these mean values are introduced as
23 unbalanced forces, improving the performance of the
0 method in terms of computing time.
8 0.6-
h The formulation developed here was used to analyse the
F
stability of the right transition wall of the Tucurui Dam.
! 0.4-
Results obtained in this work have shown satisfactory
E reliability with respect to the main failure modes of the
structure, even for an earthquake with mean accelerations
3
0.2 - of O.lg, which is a reasonably high value for the region
where the dam was built.

0.0 0.2 0.4 0.6 0.8 1.0

Safety factor (cracking) ACKNOWLEDGEMENTS

Fig. 8. Cumulative probability of the safety factor against The authors would like to thank the CNPq (Brazilian
cracking. National Research Council) for its financial support.
104 J. Milton de Aralijo, A. M. Awruch

REFERENCES 461-469.
18. Chang, T. P. and Chang, H. C., Stochastic dynamic finite
1. LCger, P. and Katsouli, M., Seismic stability of concrete element analysis of a nonuniform beam. Znt. J. Solids
gravity dams. Earthquake Eng. Struct. Dyn., 1989, 18, Struct., 1994, 31, 587-597.
889-902. 19. Liu, M. L. and To, C. W. S., Adaptive time schemesfor
2. Hamdi, M. A., Ousset,Y. and Verchery, G., A displacement responses of non-linear multi-degree-of-freedom systems
method for the analysis of vibrations of coupled fluid- under random excitations. Comput. Struct., 1994, 52, 563-
structure systems.ht. J. Numer. Methods Eng., 1978, 13, 571.
139-150. 20. Shinozuka, M. and Deodatis, G., Response variability of
3. Olson, L. G. and Bathe, K. J., A study of displacement-based stochastic finite element systems.J. Eng. Mech., 1988, 114,
fluid finite elements for calculating frequencies of fluid- 499-5 19.
structure systems.Nuclear Eng. Design, 1983, 76, 137- 151. 21. Baecher, G. B. and Ingra, S., Stochastic FEM in settlement
4. Fenves, G. and Vargas-Loli, L. M., Nonlinear dynamic predictions. J. Geotech. Eng. Division, ASCE, 1981, 107,
analysis of fluid-structure systems. J. Eng. Mech., 1988, 449-463.
114, 219-240. 22. Liu, W. K., Belytschko, T. and Mani, A., Probabilistic finite
5. Vargas-Loli, L. M. and Fenves, G. L., Effects of concrete elements for nonlinear structural dynamics. Comput.
cracking on the earthquakeresponseof gravity dams.Earth- Methods Applied Mech. Eng., 1986, 56, 61-81.
quake Eng. Struct. Dyn., 1989, 18, 575-592. 23. Liu, W. K., Belytschko, T. and Mani, A., Randomfield finite
6. Sharan, S. K., Time-domain analysis of infinite fluid elements.Znt. J. Numer. Methods Eng., 1986,23, 1831- 1845.
vibration. Int. J. Numer. Methods Eng., 1987, 24, 945-958. 24. Liu, W. K., Besterfield, G. H. and Belytschko, T., Transient
7. Sharan,S. K. and Gladwell, G. M. L., A general method for probabilistic systems.Comput. Methods Applied Mech. Eng.,
the dynamic response analysis of fluid-structure systems. 1988, 67, 27-54.
Comput. Struct., 1985, 21, 937-943. 25. Liu, W. K., Bestertield, G. H. and Belytschko, T., Variational
8. Zienkiewicz, 0. C. and Bettess,P., Fluid-structure dynamic approach to probabilistic finite elements. J. Eng. Mech.,
interaction and wave forces. An introduction to numerical 1988, 114, 2115-2133.
treatment.Int. J. Numer. Methods Eng., 1987, 13, l-16. 26. Vanmarcke, E. and Grigoriu, M., Stochastic finite element
9. Cervera, M., Oliver, J. and Faria, R., Seismic evaluation of analysis of simple beams.J. Eng. Mech., 1983, 109, 1203-
concrete dams via continuum damage models. Earthquake 1214.
Eng. Struct. Dyn., 1995, 24, 1225-1245. 27. Kimura, M. and Izumi, M., A method of artificial generation
10. Araijo, J. M., Analysis of concrete gravity dam considering of earthquakeground motion. Earthquake Eng. Struct. Dyn.,
the constructive stage and the dynamic dam-reservoir- 1989, 18, 867-874.
foundation interaction. Doctoral Thesis, Graduate Program 28. Newland, D. E., An Introduction to Random Vibrations and
in Civil Engineering, Federal University of Rio Grande do Spectral Analysis. Longman, New York, 1981.
Sul, Port0 Alegre, RS, Brazil, 1995 (in Portuguese). 29. To, C. W. S. and Liu, M. L., Randomresponsesof discretized
11. Benaroya, H. and Rehak, M., Finite element methods in beamsand plates by the stochasticcentral difference method
probabilistic structural analysis: a selective review. Applied with time co-ordinate transformation. Comput. Struct., 1994,
Mech. Rev., 1988, 41, 201-213. 53,727-738.
12. Shinozuka, M., Monte Carlo solution of structural dynamics. 30. Yang, T. Y., Random Vibration of Structures. Wiley, New
Comput. Struct., 1972, 2, 855-874. York, 1986.
13. Shinozuka, M., Probabilistic modeling of concretestructures. 31. Clough, R. W. and Penzien, J., Dynamics of Structures.
J. Eng. Mech. Division, ASCE, 1972, 98, 1433-1451. McGraw-Hill, New York, 1975.
14. Shinozuka, M. and Astill, C. J., Random eigenvalue pro- 32. Comite Euro-International du Beton, CEB FIP Model Code
blems in structural analysis. AZAA Journal, 1972, 10, 456- 1990. Switzerland, 1993.
462. 33. Ang, A. H. S. and Tang, W. H., Probability Concepts in
15. Shinozuka, M. and Wen, Y. K., Monte Carlo solution of Engineering Planning and Design, vol. 1. Wiley, New
nonlinear vibrations. AIAA Journal, 1972, 10, 37-40. York, 1975.
16. Yamazaki, F., Shinozuka, M. and Dasgupta, G., Neumann 34. Aratijo, M. and Awruch, A. M., An objective cracking
expansion for stochastic finite element analysis. J. Eng. criterion for the analysis of concrete dams. Compt. Struct.,
Mech., 1988, 114, 1335-1355. 1996, 59,751-756.
17. Araujo, J. M. and Awruch, A. M., On stochastic finite 35. Hinton, E., Rock, A. and Zienkiewicz, 0. C., A note on a
elements for structural analysis. Comput. Struct., 1994, 52, masslumping in related processin the finite element method.
Earthquake Eng. Struct. Dyn., 1976, 4, 245-249