Geo-Risk 2017 GSP 284 183

Reliability Analysis of a Himalayan Rock Slope Considering Uncertainty in

Post Peak Strength Parameters

Gaurav Tiwari1 and Madhavi Latha Gali2

Downloaded from ascelibrary.org by University of California, San Diego on 06/02/17. Copyright ASCE. For personal use only; all rights reserved.

1
Ph.D. Student, Dept. of Civil Engineering, Indian Institute of Science, Bangalore, India. E-mail:
[email protected]
2
Professor, Dept. of Civil Engineering, Indian Institute of Science, Bangalore, India. E-mail:
[email protected]

Abstract
Probabilistic analysis of a Himalayan rock slope supporting the piers of a railway bridge is
carried out by considering strength drop from peak to residual strength. Importance of
considering the uncertainty in post peak strength parameters for rock slopes in average quality
rock mass is discussed. Ignoring the strength drop in post-peak strength parameters from peak
strength parameters in the probabilistic analysis results in the overestimation of rock slope
stability. Equations are used in this article to provide the peak and post-peak parameters of rock
mass by considering peak and residual geological strength index (GSI). This method uses joint
alteration factor, joint large-scale waviness, small-scale smoothness factors and block volume to
calculate the peak and residual GSI values. Variability in the rock mass properties of a
Himalayan rock slope is evaluated through these equations using Monte-Carlo simulation.
Stability of the rock slope is analyzed through reliability analysis. Probability of failure was
calculated by considering and ignoring a drop in post-peak strength parameters for the slope.
Reliability analysis has been carried out using Hong’s point estimate method coupled with finite
element approach. Results from the analyses showed that the considering a drop in post-peak
strength parameters from peak-strength parameters have significant effect on the probability of
failure and hence it is important to take into account the variability of post-peak parameters
properly for evaluating performance level of rock slopes.

INTRODUCTION
A railway line is being laid in Jammu and Kashmir, India and this line is crossing the river
Chenab at a height of about 359 m. A bridge is proposed at the site and once completed; this will
be the world’s highest railway bridge at present. The piers of this bridge are supported by two
rock slopes named as S1 and S2. Among these slope S1 slope is selected for the probabilistic
analysis in the current study.
Knowledge of strength and deformation properties of the rock mass is important in
analyzing the stability of rock slopes. While substantial importance has been given for
determination of deformation modulus and peak strength parameters for rock mass, studies on

© ASCE

Geo-Risk 2017
 Geo-Risk 2017 GSP 284 184

post-peak strength behavior of rock mass are very limited (Ribacchi, 2000; Russo et al., 1998;
Cai et al., 2007). It has been concluded by Crawder and Bawden (2004) that the true nature of a
strain softening response is not known for large field-scale rock masses and it is necessary to
refine how post-failure characteristics are defined and this can be done only by back analyses of
a large number of well documented case histories. Hence in the absence of reliable empirical
equations relating post-peak strength parameters to some rock mass classification rating, some
Downloaded from ascelibrary.org by University of California, San Diego on 06/02/17. Copyright ASCE. For personal use only; all rights reserved.

random assumption is usually made in probabilistic stability analysis of rock slopes. Most
common assumption made in the stability analysis of rock slope is to assume rock mass as
elastic-perfectly plastic material which is not correct for an average quality rock mass
(35<GSI<65). For an average quality rock mass, a strength drop generally occurs after the stress
in the rock mass reaches a peak strength, which should be considered in the stability analysis
(Hoek and Brown, 1997). Hence, it becomes necessary to consider this strength drop in the
stability analysis of rock slopes.

Stability analysis of rock slopes requires precise values of various deformation and
strength parameters which are seldom available and these properties always contains some
uncertainty. Probabilistic approach has been extensively used to quantify the uncertainties in
rock mass properties and to provide a better insight into the performance of rock structures using
reliability based design methods. Various techniques have been developed over the years to
approximate the true value of the reliability index and probability of failure which include First
Order Second Moment (FOSM) method, Second Order Second Moment (SOSM) method, Point
Estimate methods (PEMs), Hasofer–Lind method, and Monte Carlo simulation. Among these
methods point estimate seems to be a good option for probabilistic analysis of rock slopes due to
some advantages. First, point-estimate methods can be coupled with any deterministic methods
like finite element method (FEM), limit equilibrium method (LEM) etc. for solving any
probabilistic problem. Secondly the computation time is much smaller than other methods like
Monte-Carlo simulation. Furthermore, PEMs overcome the difficulties associated with the lack
of knowledge of the probability functions of probabilistic variables, as these functions are
approximated using only their first few statistical moments.

In the current article a quantitative approach based on GSI is discussed which can be used
in probabilistic analysis of rock slopes. This approach takes into account the strength drop
occurring in the rock mass after reaching peak strength by using the method suggested by Cai et
al. (2007). The approach is applied to a Himalayan rock slope case study to demonstrate the
applicability and to make a comparison in the results by considering and ignoring strength drop.
Variability in rock mass properties are evaluated using Monte-Carlo simulation and then the
reliability analysis was carried out using Hong’s point estimate method.

© ASCE

Geo-Risk 2017
 Geo-Risk 2017 GSP 284 185

GENERAL DETAILS OF THE CASE STUDY

The area under investigation covers a part of Udhampur district of Jammu and Kashmir state. A
bridge of length 1.26 km is proposed at the site whose piers are situated on two rock slopes. This
bridge once constructed will be the world’s highest railway bridge at present at a height of about
359 m. Among the 18 piers of the bridge, 4 piers (S10-S40) are located on the left abutment and
the other 14 piers (S50-S180) on the right abutment. For the current article probabilistic analysis
Downloaded from ascelibrary.org by University of California, San Diego on 06/02/17. Copyright ASCE. For personal use only; all rights reserved.

of right slope has been carried out. Figure 1 shows the site location of the bridge.

Figure 1. Photograph of the Chenab Bridge site.

Major rock type at the bridge site is jointed dolomitic limestone. Joint mapping results
yielded that the strata are characterised by three major discontinuities. One sub-horizontal and
two sub-vertical joints were present at the site. The sub-horizontal joint set was named J1 while
the other two joint sets were named J2 and J3. The average joint spacing for joint sets J1, J2 and
J3 are 20 cm, 60 cm and 115 cm respectively and the corresponding standard deviations are 3.33
cm, 13.33 cm and 28.33 cm with a lognormal distribution. It was observed that the
discontinuities present at the site had generally small to moderate undulation with slightly rough
texture. It was observed that discontinuities present at the site have infill of poorly graded silty
sand whose thickness was varying along different joints. The average values for peak large-scale
waviness, peak small-scale smoothness and peak alteration parameters are 1.5, 1.5 and 4
respectively and a coefficient of variance was assumed as 8% for these parameters with a normal
distribution (Cai et al., 2004). For more details regarding geology at the site refer Tiwari and
Latha (2016). Statistical parameters of intact rock properties at the site are summarised in the
Table 1. Statistical parameters of the UCS of intact rock are estimated using laboratory testing
while the statistical parameters for mi are taken from the literature (Marinos and Hoek, 2000).

© ASCE

Geo-Risk 2017
 Geo-Risk 2017 GSP 284 186

Table 1. Statistical moments and distribution of the intact rock properties

Parameter Mean Standard Deviation Coefficient of Variation Distribution

UCS (MPa) 115 39.5 34.3% Normal
mi 9 1 11.1% Normal
γ (kN/m3) 27.62 - - -
Downloaded from ascelibrary.org by University of California, San Diego on 06/02/17. Copyright ASCE. For personal use only; all rights reserved.

CALCULATION OF VARIABILITY IN ROCK MASS PROPERTIES

Equations 1-7 given below are used for the calculation of probability distributions of the output
parameters from the known probability distributions of the input parameters using the Monte
Carlo method. Details regarding these equations can be seen in Cai et al. (2004) and Cai et al.
(2007).

/ /
/ / 26.5 + 8.79 + 0.9
/ , = / /
(1)
1 + 0.0151 − 0.0253

/ /
/
= /
(2)

= 0.21 − 0.61 + 1.41 (3)

= 0.16 − 0.25 + 0.82 (4)

= (5)
/
/
= ∗ (6)

/
/
= (7)

/
Where / is peak/residual GSI; is peak/residual large scale waviness (undulations from
/
1 to 10 m length of joints); is peak/residual small scale smoothness (asperities from 1 cm to
/
20 cm length of joints); is peak/ residual alteration parameter (depends on joint weathering
/ /
and infill); is peak/residual joint condition factor; is peak/residual block volume; is
/ /
frictional strength component of intact rock; , are peak/residual strength components
of rock mass respectively.

© ASCE

Geo-Risk 2017
 Geo-Risk 2017 GSP 284 187

As given above, the Hoek-Brown rock mass strength parameters can be determined from
equations linking mb and sb directly to GSI, and GSI is linked directly to block volume (Vb) and
joint condition factor (Jc), which are again linked to joint spacing, orientation, length, roughness,
etc. It can be seen that residual joint roughness and joint alteration parameters can be estimated
using peak joint roughness and joint alteration parameters from equations 3-5. These equations
are obtained by replacing conditional equations of Cai et al. (2007) by simple polynomial
Downloaded from ascelibrary.org by University of California, San Diego on 06/02/17. Copyright ASCE. For personal use only; all rights reserved.

equations. Residual GSI values obtained for the case study present in the literature using Cai et
al. (2007) and using the modified equations of the present study are given in Figure 2. These
equations can be adopted for finding the probability distributions of both peak and residual
strength parameters using Monte-Carlo simulation. Monte-Carlo simulation has been carried out
using @Risk (Palisade, 2016) and 10,000 iterations were carried out for each simulation. Firstly
variability in joint condition factor (Jc) was estimated which was used to find variability in GSI
and later they were used to find out variability in rock mass strength parameters. Table 2 shows
the estimated statistical parameters (Mean, Standard deviation and coefficient of variation COV
in percentage) for the strength of the rock mass present at the site.

30

25

20
Residual GSI

15 Cai et al. (2007)

Present Study
10

0
1 2

Figure 2. A comparison between residual GSI values estimated by Cai et al. (2007) and
present study for Kannagawa hydropower project at different locations.

Table 2. Statistical moments and distribution of peak and residual rock mass properties
using Monte-Carlo simulation

Parameter Mean Standard deviation COV% Distribution

Peak mb 1.31 0.1724 13.16 normal
Peak sb 0.0025 0.00056 22.4 lognormal
Residual mb 0.434 0.048 11.05 normal
Residual sb 0.00008 0.0000073 9.12 lognormal

© ASCE

Geo-Risk 2017
 Geo-Risk 2017 GSP 284 188

HONG’S POINT ESTIMATE METHOD (PEM) AND PROBABILITY OF FAILURE

Generally, PEMs consist of replacing a continuous density distribution function by specifically

defined discrete probabilities, which are intended to model the same low-order moments of that
distribution function. The determination of these moments is performed by adding weighted
discrete realizations (Ahmadabadi and Poisel, 2015a). There are many point estimate methods
Downloaded from ascelibrary.org by University of California, San Diego on 06/02/17. Copyright ASCE. For personal use only; all rights reserved.

available in the literature which have various advantages and disadvantages. For this study
Hong’s PEM (Hong, 1998) was used since it requires lesser number of evaluation points as
compared to some of the PEMs like Rosenblueth’s PEM method (Rosenblueth, 1975) and it can
consider the skewness in the probability density functions of random variables (Hong, 1998;
Ahmadabadi and Poisel, 2015a; Ahmadabadi and Poisel, 2015b). Summary of the steps involved
in Hong’s methods analysis:
Step 1. Evaluate the parameter

= + (8)
2

n = number of random variables; = skewness of the random variable

Step 2. Evaluate the two realization points for each random variable

, = μ + − . 9(a)
2

, = μ + + . 9(b)
2

,, , = realization points for each variable; μ = mean value of random variable; =

Standard deviation of random variable

Step 3. Calculate corresponding weights for each model output to compute statistical moments of
performance functions

, = 1+ 10(a)

, = 1− 10(b)

4. Compute the mth statistical moment of the performance function ( )

( )= ∑ ∑ , × μ ,… , …,μ (11)
Where G is the performance function which is factor of safety (FOS) for the present study.

© ASCE

Geo-Risk 2017
 Geo-Risk 2017 GSP 284 189

5. Calcullate the probability of faiilure

μ −
= − ( )

Where μ = mean factor of saffety; = standard deeviation of ffactor of safeety; ( ) deenotes

the cumu
ulative densitty function (CDF)
( of thee standard noormal distribbution.
Downloaded from ascelibrary.org by University of California, San Diego on 06/02/17. Copyright ASCE. For personal use only; all rights reserved.

FINITE ELEMENT
T MODEL DETAILS
D
Hong’s PEM
P was co oupled with finite elemeent method aand analysiss was carriedd out using finite
element method based software Phase2 (Ro ocscience, 22014). Rockk mass strenngth was deefined
using Hooek-Brown criterion
c forr both the slopes.
s Hoekk-Brown strrength criterrion was sellected
because Mohr-Coulo
M omb criterionn overestimaates the meaan factor of ssafety and unnderestimatees the
probabiliity of failuree in comparisson to Hoek--Brown criteerion (Li et aal., 2013). L
Lateral bounddaries
were fixeed in x-direcction and bo ottom boundaary was fixeed in both x and y direcctions. The sslopes
were disccretized, keeeping the num mber of 3-nooded trianguular finite eleements as 722,644 in sloppe S1
and nummber of elem ments were decided
d usin
ng H-converggence methood (Tiwari aand Latha, 22016;
Shen andd Karakus, 2013).
2 Absollute energy criterion
c waas used with 0.1 % tolerrance (Tiwarri and
Latha, 20016; Kanung go et al., 2013). Factors of safety w
were calculated using weell-accepted shear
strength reduction teechnique (SS ferent cases. Phase2 calcculates FOS
SR) for diffe S by reducinng the
original shear streng gth parameteer to bring the slope too the verge of failure ((Matsui and San,
1992). Fiigure 3 show ws finite elemment model ofo the slope for the preseent study.

Figure 3.
3 Finite eleement modeel of the slop
pe considereed for the sttudy

© ASCE

Geo-Risk 2017
 Geo-Risk 2017 GSP 284 190

RESULTS AND DISCUSSIONS

Both factor of safety and probability of failure are important to analyse the stability of slopes as
Dai and Wang (1992) concluded that factor of safety should not be abandoned in favour of
reliability analyses but it has to be used as a complementary tool to reliability analyses. Mean
factor of safety (FOS) and probability of failure are estimated by considering and ignoring
strength drop from peak strength to residual strength. Figure 4 shows the factor of safety and
Downloaded from ascelibrary.org by University of California, San Diego on 06/02/17. Copyright ASCE. For personal use only; all rights reserved.

probability of failure for the case study. It was observed that the FOS was 2.92 without
considering strength drop and 1.57 when strength drop was considered. Probability of failure was
approximately 5% when strength reduction was considered as compared to 0.1% when strength
drop is ignored. The analysis shows the expected performance level of slope to be good when the
strength reduction was ignored while the expected performance level was unsatisfactory when
strength reduction was considered (US army Corps of Engineers, 1999).
The difference between the values of the FOS and probability of failure is close to the
50% and 99% respectively. It can be seen that the amount of the overestimation in the FOS and
underestimation of probability of failure when strength drop is ignored in the analysis. It can also
be seen that how the expected performance changed from good to unsatisfactory and thus
overestimating the performance level of slope when strength drop was ignored. Hence, any
assumption regarding post-peak strength parameters for rock slopes without any proper
reasoning can lead to wrong results and can undermine the necessity of any stabilization methods
required to stabilize the slope. It becomes more important if the rock slope is bearing any
structure like for the current study the rock slope is supporting the piers of a bridge to consider
the variability in post-peak strength parameters correctly.
It should be noted that this study suffers from some limitations. Correlation between
various random variables is not taken into account and even the strength drop was assumed to be
instantaneous i.e. elastic-brittle-plastic behavior was assumed for strength drop case. However,
this simple approach is better to use for the rock slopes located in average quality rock mass.
3.5 4.8
4.4
probability of failure (%)

3
4
Factor of safety

2.5 3.6
3.2
2 2.8
1.5 2.4
2
1 1.6
1.2
0.5 0.8
0.4
0 0
strength drop perfectly plastic strength drop perfectly plastic

Figure. 4 Difference in the a) factor of safety b) probability of Failure of slope due to

consideration of different post-peak behavior

© ASCE

Geo-Risk 2017
 Geo-Risk 2017 GSP 284 191

CONCLUSIONS

Probability analysis of a rock slope with average quality rock mass has been carried out. A
simple quantitative GSI based approach is used to quantify the uncertainty in peak and post-peak
strength parameters using Monte-Carlo simulations, considering the drop in the shear strength
parameters of the rock mass. Probability of failure for the slope is evaluated using point estimate
Downloaded from ascelibrary.org by University of California, San Diego on 06/02/17. Copyright ASCE. For personal use only; all rights reserved.

method, with and without considering the drop in the strength. It was observed that the stability
of the slope was highly overestimated when the drop in strength was ignored. Probability of
failure was underestimated and factor of safety was overestimated when the strength drop was
not considered in the analysis. This study demonstrates and shows the effect of considering
uncertainty in peak and post-peak strength parameters on the stability of rock slope.

REFERENCES

Ahmadabadi, M. and Poisel, R. (2015a). “Probabilistic Analysis of Rock Slopes Involving

Correlated Non-normal Variables Using Point Estimate Methods.” Rock Mech Rock Eng.
Available online.

Ahmadabadi M. and Poisel, R. (2015b). “Assessment of the application of point estimate

methods in the probabilistic stability analysis of slopes.” Comput Geotech., 69, 540-550.

Cai M., Kaiser P.K., Tasaka Y. and Minami M. (2007). “Determination of residual strength
parameters of jointed rock masses using the GSI system.” Int J Rock Mech Min Sci, 44(2), 247–
265.

Cai M., Kaiser P.K., Uno H., Tasaka Y. and Minami M. (2004). “Estimation of rock mass
strength and deformation modulus of jointed hard rock masses using the GSI system.” Int J Rock
Mech Min Sci., 41(1), 3–19.

Crowder J.J. and Bawden W.F. (2004). “Review of post-peak parameters and behaviour of rock
masses: current trends and research.” RocNews 2004; Fall:13.

Dai S.H. and Wang M.O. (1992). Reliability Analysis in Engineering Applications. Van Nostrand
Reinhold, New York.

Hoek E. and Brown E.T. (1997). “Practical estimates of rock mass strength.” Int J Rock Mech
Min Sci., 34(8), 1165-1186.

Hong H.P. (1998). “An efficient point estimate method for probabilistic analysis.” Reliab Eng
System Saf., 59, 261–267.

© ASCE

Geo-Risk 2017
 Geo-Risk 2017 GSP 284 192

Kanungo D.P., Pain A. and Sharma S. (2013). “Finite element modeling approach to assess the
stability of debris and rock slopes – a case study from the Indian Himalayas.” Natural Hazards,
69, 1–24.

Li A.J., Cassidy M.J., Wang Y., Merifield R.S. and Lyamin A.V. (2012). “Parametric Monte
Carlo studies of rock slopes based on the Hoek–Brown failure criterion.” Computers and
Downloaded from ascelibrary.org by University of California, San Diego on 06/02/17. Copyright ASCE. For personal use only; all rights reserved.

Geotechnics, 45, 11–18.

Marinos, P. and Hoek, E. (2000). “GSI: a geologically friendly tool for rock mass strength
estimation.” Proceedings of the GeoEng2000 at the International Conference on Geotechnical
and Geological Engineering, Melbourne, Technomic Publishers, Lancaster, 1422–1446.

Matsui T. and Sam K.C. (1992). “Finite element slope stability analysis by shear strength
reduction technique.” Soils Found., 32, 59–70.

Palisade Corporation (2001) @RISK. Palisade Corporation, v.4.

Ribacchi R. (2000). “Mechanical tests on pervasively jointed rock material: insight into rock
mass behaviour.” Rock Mech Rock Eng, 33(4), 243–266.

Rocscience (2014). “Phase2 version 8.020: finite element analysis for excavations and slopes.”
Rocscience Inc., Ontario.

Rosenblueth E. (1975). “Point estimates for probability moments.” Proc Natl Acad Sci., USA,
72, 3812–3814.

Russo G., Kalamaras G.S. and Grasso P. (1998). “A discussion on the concepts of
geomechanical classes behavior categories and technical classes for an underground project.”
Gallerie e Grandi Opere Sotterranee; 54.

Shen J. and Karakus M. (2013). “Three-dimensional numerical analysis for rock slope stability
using shear strength reduction method.” Can. Geotech. J., 51, 164-172.

Tiwari G. and Latha G.M. (2016). “Design of rock slope reinforcement: An Himalayan case
study.” Rock Mech Rock Eng, Available online.

U.S. Army Corps of Engineers (1999). “Risk-based analysis in geotechnical engineering for
support of planning studies, engineering and design.” Rep. No. 20314-1000, Dept. of Army,
Washington, D.C.

© ASCE

Geo-Risk 2017