Professional Documents
Culture Documents
Abstract: In the present study, results of reliability analyses of four selected rehabilitated earth dam sections, i.e., Chang, Tapar,
Downloaded from ascelibrary.org by LIBRARY PERIODICALS on 11/06/14. Copyright ASCE. For personal use only; all rights reserved.
Rudramata, and Kaswati, under pseudostatic loading conditions, are presented. Using the response surface methodology, in combination
with first order reliability method and numerical analysis, the reliability index 共兲 values are obtained and results are interpreted in
conjunction with conventional factor of safety values. The influence of considering variability in the input soil shear strength parameters,
horizontal seismic coefficient 共␣h兲, and location of reservoir full level on the stability assessment of the earth dam sections is discussed
in the probabilistic framework. A comparison of results with those obtained from other method of reliability analysis, viz., Monte Carlo
simulations combined with limit equilibrium approach, provided a basis for discussing the stability of earth dams in probabilistic terms,
and the results of the analysis suggest that the considered earth dam sections are reliable and are expected to perform satisfactorily.
DOI: 10.1061/共ASCE兲GT.1943-5606.0000313
CE Database subject headings: Structural reliability; Dams, earth; Seismic effects; Dam safety.
Author keywords: Reliability; Earthen dams; Seismic effects; Dam safety.
using regression analysis based upon least-squares error ap- 8 − − − 6.70 18.852 23.565 1.15
proach, is established as
y = f共x1,x2,x3 . . .兲 + e 共1兲 surface model between input and output variables, 2n 共n
The symbol “e” represents the other sources of uncertainty, not = number of input variables兲 factorial design is used. The proce-
accounted for in “f ⬘.” Myers and Montgomery 共2002兲 presented dure involves the determination of output response, i.e., pseudo-
an excellent literature on the use of RSM. Using the developed static FS, for a combination of input parameters 共sample points兲,
response surface model, the reliability index 共兲 is calculated. and regression analysis, based upon the least-squares error ap-
proach, is performed to develop the response surface model. The
adequacy of the fitted model is examined on the basis of the
Reliability Analysis normal probability plot 共which should be approximately along a
Reliability index 共兲 is calculated on the basis of the matrix for- straight line兲, and the comparison of computed values of the co-
mulation for the second moment, Hasofer-Lind reliability index efficients of multiple determination 共R2兲 and adjusted coefficients
共兲, defined as of multiple determination 共R2adj兲.
The pseudostatic FS is evaluated using “strength reduction
 = min冑共X − M兲TC−1共X − M兲 共2兲 technique” available in the numerical code FLAC 共FLAC 5.0 ref-
X苸F
erence manual 2007兲. The present analyses are effective stress
where X = vector of random variables; M = vector of mean value of analyses using effective stress shear strength parameters, i.e., ef-
random variables; C = covariance matrix of random variables; and fective cohesion 共c⬘兲 and the effective friction angle 共⬘兲, which
F = failure region. Reliability index is defined as the minimum are obtained from consolidated-undrained 共CU兲 triaxial tests 共CU
distance from the transformed failure surface to the origin of the tests with pore pressure measurements兲 on soil samples collected
reduced variate space. Adopting the procedure explained by Low from the dam site. To take into account the influence of develop-
and Tang 共1997兲, in which the above Eq. 共2兲 is used, the value of ing pore pressure on the degradation of strength and stiffness of
 can be obtained with the help of “SOLVER,” an optimization the soil due to earthquake loading, the measured effective cohe-
tool available in Microsoft’s Excel. sion and the measured effective friction angle are reduced by 20%
before being used in the analyses. This reduction is similar to the
adjustment to the undrained shear strength suggested by Makdisi
Results of the Analysis and Discussion and Seed 共1978兲 and Seed 共1979兲. The following sections provide
a discussion on typical results obtained for the rehabilitated
After the January 26, 2001 Bhuj earthquake, remedial measures Chang dam section.
were taken to rehabilitate the damaged or failed earth dam sec- The typical values of soil properties used in the reliability
tions located in the Kachchh region of Gujarat, India. Following analysis of the Chang dam section are summarized in Table 1. In
the procedure explained in the previous sections, the reliability the absence of a large amount of test data from the sites consid-
analyses of the selected dam sections, i.e., Chang, Tapar, Rudra- ered, the coefficient of variation 共COV兲 共%兲 is chosen based on
mata, and Kaswati, are performed. To develop a linear response the values suggested in the literature 共Duncan 2000兲.
Table 3. Comparison of Results of the Reliability Analysis of Rehabilitated Dam Sections at RFL
 共MCS兲  共RSM兲
a
Dam sections Slope Mean FS  Mean FS 
Chang 共H = 15.50 m兲 U/S 关1V:2.5H兴 1.92 2.98 1.86 2.87
D/S 关1V:2.0H兴 1.86 2.86 1.73 2.76
Tapar 共H = 17.75 m兲 U/S 关1V:2.0H兴 5.67 6.31 4.27 5.88
D/S 关1V:2.25H兴 1.61 2.47 1.54 2.39
Kaswati 共H = 15.74 m兲 U/S 关1V:3.0H兴 1.92 3.21 1.85 2.97
D/S 关1V:2.5H兴 1.30 3.68 1.21 3.32
Rudramata 共H = 27.57 m兲 U/S 关1V:3.0H兴 1.74 3.25 1.64 3.15
D/S 关1V:4.5H兴 1.32 3.08 1.25 2.89
a
Values are adapted from Sivakumar Babu et al. 共2007兲.
Becker 共1996兲 suggested that the selection of the characteristic the procedure for the reliability analysis is followed for the other
values of geotechnical parameters and corresponding confidence three seismic zones 共II, III, and IV兲 in India, which is based upon
intervals should be incorporated in reliability based designs. In the seismic coefficients provided in the code. From the results
Downloaded from ascelibrary.org by LIBRARY PERIODICALS on 11/06/14. Copyright ASCE. For personal use only; all rights reserved.
the present study, a combination of lower 共 − 1.65兲 and upper presented in Table 4, it can be noted that the values of  are
limit 共 + 1.65兲 values of the input variables 共 = mean value; higher in these regions.
= standard deviation兲 are used for the selection of sample points
and the output response 共i.e., pseudostatic FS兲 for each sample
Effects of Correlation among Input Parameters
point is calculated. The upper limit and lower limit values are
based upon the assumption that soil parameters follow normal The values of correlation coefficient 共兲 between cohesion 共c⬘c 兲
distribution and upper and lower limit values have probabilities of and friction angle 共⬘c 兲 are taken as ⫺0.25, ⫺0.50, and ⫺0.75
being exceeded by 5 and 95%, respectively. The pseudostatic FSs 共Baecher and Christian 2003兲, and the results are summarized in
calculated for the eight combinations of input parameters 共or Table 5. It can be noted that the consideration of correlation be-
sample points兲 are provided in Table 2. Using these data, regres- tween the parameters marginally improves the reliability index
sion analysis is performed to obtain a linear response surface 共兲 values.
model as
FS = − 1.3762 + 0.0125c⬘c + 0.0454⬘c + 0.0674s⬘ Consideration of Lognormal Distribution for Input
Parameters
共R2 = 0.999;R2adj = 0.998兲 共3兲
Table 5 provides  values evaluated for the uncorrelated lognor-
where c⬘c , ⬘c = effective stress parameters, i.e., cohesion and fric- mally distributed input soil parameters, c⬘c , ⬘c , and s⬘. It can be
tional angles for core material, respectively, and s⬘ = effective observed that the values are relatively higher than the correspond-
frictional angle for the shell material. It can be noted that the ing values obtained for normally distributed input parameters.
values of R2 and R2adj are close to 1.0 and the normal probability Baecher and Christian 共2003兲 made similar observations.
plot is also approximately along a straight line 共not shown兲, which
ensures the adequacy of the fitted model. Using the linear re-
Uncertainties in the ␣h and RFL Values
sponse surface model, the reliability index value is obtained as
2.87. A summary of results obtained for four dam sections is Considering the case of the Chang dam section, the mean value of
presented in Table 3 and it is compared with previous results ␣h and depth of RFL are taken as 0.15 and 14.0 m 共H = 15.5 m
共Sivakumar Babu et al. 2007兲. It can be noted that the pseudo- and free board= 1.5 m兲, respectively. The COVs in the ␣h value
static FSs values for the rehabilitated dam sections are well above and RFL are taken as 25 and 5%, respectively, in accordance with
the acceptable limit, i.e., 1.0. However, based on  values, it can Yanmaz and Beser 共2005兲. The input parameters are taken as
be noted that the performance levels of the dam sections are just uncorrelated lognormally distributed continuous random vari-
“above average” as per USACE 共1997兲 guidelines. Taking into ables. The corresponding linear response surface model is given
consideration the effect of seismic location on dam safety issues, by