Professional Documents
Culture Documents
231 - 242
© RAMS Consultants
Printed in India
1. Introduction
In the hilly terrains of India, including Himalayan mountains and Western Ghats, landslides
have been a major and widely spread form of natural disasters that occupy a position of
major concern. These landslides, year after year, bring about untold misery to human
settlements apart from causing devastating damages to transportation and communication
network. The assessment of slope stability is a complex geotechnical problem because of
the uncertainties involved. Due to the need to treat uncertainties and the inability of the
conventional deterministic methods of analysis to incorporate the variability of geologic
materials ,, there have been numerous attempts in recent years to use a probabilistic
approach complementary to the conventional approach for analyzing the slopes [2,7,9].
________________________________________
*
Corresponding author’s email: gvee@civil.iitb.ac.in 231
232 Vageesha S. Mathada, G. Venkatachalam and A. Srividya
Probability theory has been considered for many years as the only way to take account
of the uncertainties in the definition of the input variables. It has been recently underlined
that probability theory can be used when the variables are random in nature and when the
information required to define their probability density functions is available[5]. In the
following two cases at least the variables cannot be considered as random variables, since
the available body of information is small and their definitions come from subjective
judgment. The first case typically arises in geotechnical engineering design, where
statistical information on uncertain parameters may be scarce or even absent and the
definition of probabilistic quantities can be made only on the basis of normative
requirements. The second case arises when some variables, described with qualitative words
or phrases must be translated into linguistic variables and, finally, expressed in a numerical
way. The problems involving uncertain or subjective non-random variables especially, those
based on incomplete information can be treated in a consistent way in the context of the
theories of possibilities and fuzzy sets [4,5,7,15]. Possibility of an event is weaker
information than probability, but can be evaluated on the basis of available information
only. The advantage of possibility theory is that it is also able to point out those situations
where the body of information is scarce and no useful estimates of the occurrence of the
event can be obtained [5].
In the first part of this paper, the fuzzy set theory, development of fuzzy model,
procedure for possibilistic reliability analysis and the proposed hybrid method are
presented. The main objective of this paper is to incorporate different types of uncertainty in
the reliability analysis of slopes. Emphasis will be given to comparison of different
methods. The well known probabilistic methods are not discussed. In the second part a
comparison between probabilistic, possibilistic and hybrid reliability analysis is done for a
simple case of a homogeneous slope.
(a − x)
(a h for a ≤ x ≤ b
− b)
h for b ≤ x ≤ c
α A (x ) = (1)
(c − x)
h for c ≤ x ≤ d
(c − d)
0 otherwise
αA(x)
b c
1
Aα
α
0
a d X
Fig. 1: Trapezoidal Fuzzy Number with an α-cut
Some times nonlinear functions of the form of Gaussian fuzzy numbers are also used. The
normalized Gaussian fuzzy number is shown in Fig. 2. In that case the membership function
is given by
2
1 x−µ
−
2 σ
α A (x) = e (2)
1 1 x−µ 2
−
2 σ
α A ( x) = e
α A(X )
0
X
In this case the construction of fuzzy set depends on two things viz. the identification of
a suitable universe of discourse and the specification of an appropriate membership
function. For the TrFN as shown in Fig.1, a, b, c and d are calculated as:
a = E [ X i ] − k 1σ [ X i ]
b = E [ X i ] − k 2σ [ X i ]
(3)
c = E [ X i ] + k 2σ [ X i ]
d = E [ X i ] + k 1σ [ X i ]
ki being the number of standard deviations between the expected value of the parameter and
the fuzzy set vertexes. The values of k1 and k2 in (3) allow for the judgment or experience of
the expert to consider the soil behaviors that cannot be formalized or tested. Some
guidelines for the proper choice of k1 and k2 are available in [4,10,13]. Suggested values of
k1 and k2 are in the ranges of 1-3 and 0-1, respectively depending on the data available and
the accuracy of the results desired. The values of k1 and k2 of 1.5 and 0.5, respectively to get
the fuzzy variable were used in [7]. The fuzzy information can also be better modeled using
Gaussian fuzzy numbers as shown in Fig. 2. In this paper different sets of k1 and k2, and
Gaussian fuzzy numbers are used to model fuzzy information.
There are several methods by which reliability of geotechnical structures can be found.
According to the Mean Value First Order Second Moment (MFOSM) method, which is
widely used in geotechnical engineering., taking the performance function and the limit
state as ( FS − 1 = 0 ), as applicable to slopes, the reliability index based on MFOSM
model is given by:
E ( FS ) − 1
β = (4a)
σ ( FS )
FS ( µ X i ) − 1
= (4b)
2
m
∂FS m
∂FS ∂FS
∑ ∂X
i =1
σ [X i
i
]2 + 2 ∑ ∂X
i , j =1 i
∂X
j
[ j]
ρ ij σ [ X i ]σ X
i≠ j
The partial derivatives in the above equation are calculated using the relation
∂FS FS + − FS −
= (5)
∂X i 2σ [ X i ]
where FS + and FS – are the values of FS obtained using the parameter values greater than
and less than the mean by an increment σ [Xi], respectively as per [9].
σ [ FS ]
2
where σ [ln ( FS )] = ln 1 + (7)
E [ FS ]
and E [ln ( FS )] = ln ( E [ FS ]) −
(σ [ln ( FS )]) 2 (8)
2
In both the cases the probability of failure p f is calculated as:
236 Vageesha S. Mathada, G. Venkatachalam and A. Srividya
p f = Φ −1 ( − β ) (9)
where Φ is the standard normal variate.
Fuzzy sets based approaches for the computation of reliability index include Fuzzy Point
Estimate Method (FPEM) [4], Fuzzy FOSM [7] and Possibilistic method [11]. In the first
two methods the α-cuts near 1 are not considered. In FPEM and Possibilistic method, the
correlation coefficients are given equal rating while computing the weighting factors. The
method presented here removes the weighting factors from the equations presented in [11].
Its application to uncorrelated variables is given below.
parameters are defined by fuzzy numbers. The fuzzy numbers may be either in the form of
TrFN of the form a , b , c , d (refer Fig. 1) or Gaussian fuzzy numbers (refer Fig. 2). Now
to decide upon the position of α-cuts, select the number of α-cuts (N), depending on the
accuracy required. Detailed trials by the authors (not presented here) with various values of
N has shown that a value of 10 for N is sufficient for most of the practical problems in
geotechnical engineering. The positions of these N α-cuts is thus given by:
i
αi = , i = 1, 2 , 3 , L L , N (10)
N
This step removes the ambiguity of selection and position of α-cuts, and does not miss
the information content at higher membership grades. Let W = g ( x ) is the performance
function. In this case g(x) is nothing but the factor of safety expression from any of the limit
equilibrium methods. The ultimate goal is to get fuzzy factor of safety so that its mean and
standard deviation can be computed. To get the resulting fuzzy set of factor of safety,
initially, the worst and best combination of these uncertain variables is to be established.
These variables can be classified as pro-variables or opposing-variables depending on
whether there is increase or decrease in FS, respectively with the increase in the value of
that variable. Usually strength variables such as c’ and φ’ are pro-variables and unit weight
γ, pore pressure ratio ru are opposing variables.
At each α-cut, the lower bound and upper bound values for all uncertain parameters are
obtained as:
xα i − = inf
α ( xi ) ≥α i
( )
X αi
x α i + = sup X α i ( ) for pro - variables
α ( xi ) ≥α i (11)
x α i − = sup X α i (
)
α ( xi ) ≥α i
for opposing variables
xα i + = inf (X αi )
α ( x i ) ≥α i
Slope Stability Assessment – A Comparison of Probabilistic, Possibilistic and Hybrid Approaches 237
g r ( X αi+ ) + g r ( X αi− )
r
Wα = for i = 1, 2 , K , N (12)
i 2
The rth moment and the standard deviation are then obtained by using the relations:
1 r
N −1
EWr = [ ] N
Wα + 2
N ∑
α i Wαr ;
i
r = 1, 2 (13 a)
i =1
σ [W ] = [ ]
E W 2 − ( E [W ]) 2 (13 b)
E [ FS ] or β = f ( P1 , P2 , L , Pk , F1 , F2 , L , Fm ) (14)
where f is the function; P1, P2, …….., Pk are k uncertain parameters each being represented
by a p.d.f.; and F1, F2, …….., Fm are m uncertain parameters that are represented by fuzzy
numbers. FOSM method is combined with the method of α-cuts of fuzzy calculus for the
calculation of E [ FS ] or β . The stepwise procedure is given in detail in [14]. The
important points are summarized below:
238 Vageesha S. Mathada, G. Venkatachalam and A. Srividya
a) An α-cut level is selected. The 2m end points of each α-cut of all the fuzzy variables
would lead to 2m well defined combinations for m fuzzy uncertainties. These can now
be considered as deterministic inputs in the function evaluation.
b) With one of these fuzzy combinations as the input for fuzzy variables, the problem now
reduces to a problem with only k random uncertainties. Any of the standard methods
(FOSM, RPEM, etc.) can be used to compute 2m values of E [FS], σ [FS] and β ,after
assuming the type of distribution for FS.
c) 2m combinations of fuzzy variables will lead to 2m values of E [FS], σ [FS] and β at
each α-cut level. Select the minimum and maximum for E [FS], σ [FS] and β from the
respective 2m values for each of these at each α-cut level -using the guidelines given in
section 4.2.
d) Repeat this procedure for all α-cut levels and finally construct the fuzzy sets of E[FS],
σ [FS] and β..FS will have fuzzy-random distribution with fuzzy mean and fuzzy
standard deviations, and the resulting reliability index is fuzzy in nature.
e) To get an equivalent single value of β the following procedure is suggested:
Var [ FS ] = σ 2 + σ µ2 (15)
iv. Now, β H for the hybrid approach is calculated using the relations given in
equations (4) or (6) depending upon the assumption of distribution of FS.
The hybrid approaches take advantage of the rich information provided by PDFs, but retains
the conservative character of fuzzy calculus to account for these parameters for which
representation by unique PDFs is not justified by the available data [8] and the proposed
method is also along those lines.
5. Illustrative Example
The relationship between the performance function and the conventional factor of safety can
be written simply as:
In slope stability analysis X is a vector of soil properties and pore water pressures. The
error associated with the method of analysis, amounting to about 5% in computed FS for
even the relatively better techniques, is small compared to uncertainties arising in the
selection of these properties. In the following examples, the Bishop’s simplified method is
used. However, the method developed can be used with any other well known limit
equilibrium method of slope stability analysis. A homogeneous slope from [3], with height
Slope Stability Assessment – A Comparison of Probabilistic, Possibilistic and Hybrid Approaches 239
H=10m and the slope inclination 1V:3H, is considered for the examples discussed in this
section (Fig. 3). The properties considered are given in Table 1.
10 m 1
Table 2 gives the standard deviations of the different fuzzy numbers ,calculated using
the approach given in section 4.2. While doing the hybrid analysis, c’ and φ’ are considered
as fuzzy variables and γ and ru are considered as random variables. Ten α-cuts are
considered in fuzzy and hybrid analysis.
Probabilistic reliability analysis was done using standard MFOSM method as given in
section 4.1. Similarly, the possibilistic (fuzzy) reliability analysis is carried out using the
method proposed in section 4.2, while the hybrid approach follows the steps of section 4.3.
All these analyses were done on the critical deterministic surface as critical reliability
240 Vageesha S. Mathada, G. Venkatachalam and A. Srividya
surface and critical deterministic surface almost coincide for homogeneous slopes [13]. The
results are tabulated in Tables 3, 4 and 5 for probabilistic, possibilistic and hybrid analysis.
6. Discussions
From the results it is clear that the least failure probability is obtained from the probabilistic
analysis, highest failure probability (possibility) is obtained from fuzzy analysis and the
hybrid analysis gives the failure probability in between the two. The differences in
reliabilities are significant even though the uncertain material properties are modeled to
have similar standard deviations in both probabilistic and fuzzy modeling. The results
confirm the fact that force fitting a probabilistic distribution with insufficient information
may lead to non-conservative risk. It is also true that ignoring the available information
content may lead to uneconomical designs, which is clear in hybrid analysis. The trend in
the results is the same in all the analyses. The results in Tables 3-5 show that although the
difference between the expected values of FS by different approaches for a particular case is
small, the standard deviation of FS is very different in all, and hence the difference in
reliability index. In a real case of layered slope, if the floating surface is used instead of a
fixed surface, one may get a band indicating the failure zone in the case of fuzzy analysis
and in hybrid analysis. There are conflicting views on the usage of floating surfaces as it
does not give any physical meaning. For the problem analyzed this will not arise as the
minimum reliability surface and critical deterministic surface almost coincide.
7. Conclusions
A simple and clear deductive procedure has been proposed for computing the safety of
slopes when fuzzy uncertainties are present as well as when some of the uncertainties are
Slope Stability Assessment – A Comparison of Probabilistic, Possibilistic and Hybrid Approaches 241
fuzzy and others are random. The possibility of an event being greater than probability [14]
as per Zadeh’s Consistency Principle, the proposed PoRAM answers the need for an
appropriate measure where a probabilistic estimate would be on the unsafe side. In practical
situations where some variables are random and others fuzzy, the proposed Hybrid method
is recommended. It enables a realistic estimate of the reliability index which lies between
those of probabilistic and possibilistic values. Given a target β it is possible to evaluate the
degree of safety accomplishment in a slope.
References
[1] Chao Ru-Jen and B. M. Ayyub., Distributions with Fuzziness and Randomness, Proc.
of ISUMA-NAFIPS ’95, IEEE, pp. 668-673, 1995.
[2] Christian, J. T., C. C. Ladd and G. B. Baecher, Reliability Applied to Slope Stability
Analysis. J. Geotech. and Geoenv. Engrg., ASCE, Vol. 120, pp. 2180-2207, 1994.
[3] Chowdhury, R. N. and D. W. Xu, Reliability Index for Slope Stability Assessment –
Two Methods Compared, Reliability Engrg. and Sys. Safety, Vol. 37, pp. 99-108,
1992.
[4] Dodagoudar, G. R. and G. Venkatachalam, Reliability Analysis of Slopes using Fuzzy
Sets Theory, Computers and Geotechnics, Vol. 27, pp. 101-115, 2000.
[5] Ferarri, P. and M. Savoia, Fuzzy Number Theory to Obtain Conservative results with
respect to Probability, Computer Methods Appl. Mech. Engrg., Vol. 160, pp. 205-222,
1998.
[6] Fuller, R. and P. Majlender, On Weighted Possibilistic Mean and Variance of Fuzzy
Numbers, Fuzzy Sets and Systems, Vol. 136, pp. 363-374, 2003.
[7] Giasi, C. I., P. Masi, and C. Cherubini, Probabilistic and Fuzzy Reliability Analysis of
a Sample Slope near Aliano. Engrg. Geol., Vol. 67, pp. 391-402, 2003.
[8] Guyonnet, D., B. Bourgine, , D. Dubios, H. Fargier, B. Côme, and J. Chilès, J.,
Hybrid Approach for Addressing Uncertainty in Risk Assessments, Journal of
Environmental Engineering, ASCE, Vol. 129, pp. 68-78, 2003.
[9] Hassan, A. M. and T. F. Wolff, Search Algorithm for Minimum Reliability Index of
Earth Slopes, J. Geotech. and Geoenv. Engrg., ASCE, Vol. 125, pp. 301-308, 1999.
[10] Juang, C. H., Y.-Y.Jhi, and D. H. Lee, Stability Analysis of Existing Slopes
Considering Uncertainty., Engrg. Geol., Vol. 49, pp. 111-112, 1998.
[11] Mathada, V. S., G. Venkatachalam, and A. Srividya,, A method for Possibilistic
Reliability Analysis of Slopes, Proc. of the 9th Int. Conf. on Struct. Safety and
Reliability (ICOSSAR ‘2005), CD-ROM Version, pp 993-998, 2005.
[12] Misra, K. B. (Ed.), New Trends in System Reliability Evaluation, Elsevier, 1993.
[13] Nawari, N. O. and R. Liang, Fuzzy-based approach for determination of characteristic
values of measured geotechnical parameters, Can. Geotech. J., Vol. 37, pp. 1131-
1140, 2000.
[14] Venkatachalam, G., A. Srividya, V. S. Mathada, and D. Niranjani, A Hybrid
Approach for Uncertainty Modelling in Slope Stability Evaluation, NSAGE 2004,
Indian Inst. of Science, Bangalore, pp. 139-144, 2004.
[15] Zadeh, L. A., Probability Measures of Fuzzy Events, J. of Math. Anal. And
Applications, Vol. 23, pp.421-427, 1968.
India. He has more than 12 years of teaching experience in India and abroad. His research
interests included Reliability and Risk Estimation in Geotechnical Engineering, Ground
Improvement and Soil-Structure Interactions.
A. Srividya joined IIT Bombay in 1988 and is currently Associate Professor in Reliability
Engineering. She has supervised 3 Ph.D.s and 34 Masters in the area of Reliability in
Engineering Design, Structural Reliability, Quality Management, Modelling and
Simulation. She has over 62 publications in various journals and conferences. She has
served as Guest Editor of special issue of IETE Technical Review on Quality Management
in Electronics, Telecommunications & IT, 2001. She has organized and has been conference
co-chairperson of various international conferences like ICQRC 2001 and ICQRIT 2003.
She has received the award for “Leadership in Reliability Engineering Education &
Research” (2003) by Society of Reliability Engineering, Quality & Operational
Management at ICQRIT – 2003.