Earthwork Project Management, Slope Stability Analysis, and Wave-Based 73

Testing Techniques GSP 252 © ASCE 2014

Probabilistic Slope Stability Analysis Using Morgenstern-Price Method

Bai Tao1, Tao Xiong2, Zhang Deyu3

1, Ph.D, School of Civil and Environmental Engineering, Wuhan Institute of Technology, Wuhan,
Hubei 430073, P.R.China; Email: [email protected]
2, M.S, School of Civil and Environmental Engineering, Wuhan Institute of Technology, Wuhan, Hubei
430073, P.R.China; Email: [email protected]
3, Ph.D, State Key Laboratory of High Performance Civil Engineering Materials, Jiangsu Research
Downloaded from ascelibrary.org by New York University on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

Institute of Building Science Co., Ltd., Nanjing 210008, China; Email: [email protected]

ABSTRACT: This paper presents a new method to analyze the slope stability that
considers the spatial variability of soil properties from the probabilistic point of view.
Orthogonal decomposition is performed to spatial covariance matrix of soil strength
by Karhunen-Loève expansion (K-L). Then the Latin Hypercube Sampling (LHS) and
the Cholesky decomposition method are used to optimize the generated spatial random
fields. In the proposed method, a readily available concise algorithm is utilized to
calculate the factor of safety for each preset slip surface. The optimization technique is
then adopted to search for the global critical slip surface. The generated samples are
examined to be rational for they are good enough to match the objective function.
Results show that the samples can reduce the simulation number greatly compared to
Monte Carlo method (MC), while the calculated failure probabilities from LHS
method and MC method are close to each other. Sensitivity analysis of geometric
parameters shows that the correlation length exerts minor influence on the slope
stability. The failure probability increases with the increasing of slope angle, slope
height and the coefficient of variability. For a given slope, the results show that only
the combined using of factor of safety and failure probability could be effective
enough for the slope safety evaluation.

INTRODUCTION

Slope soil is a complex engineering material. One of the main sources of

heterogeneity is inherent spatial soil variability due to natural geologic processes. As
the development of random field theory, the uniformity hypothesis of the soil
properties gains more and more challenge. The representative of random field used in
slope stability analyses is the Random Finite Element Method (RFEM) proposed by
Griffiths and Fenton (2004). It comprises the theories of random field and the
nonlinear finite element method, and the strength reduction method is adopted to
calculate the factor of safety of the spatially variable soil slope. Though the RFEM is
advantageous for it does not require a priori assumption relating to the shape or
location of the failure mechanism, the redundant variables such as elastic modulus and
poisson ratio, are introduced to the system.
The limit equilibrium method, combined with the Monte Carlo method, is also
available in the reliability analysis of the slope stability. Many researchers (El-ramly
2002; Li, 1987; Cho 2010) have made the attempts to perform the probabilistic study
on the slope stability. However, the method they used cannot take into account the
spatial variability over the whole space domain, as well as the natural tedious
computation produced by using the Monte Carlo method.

Earthwork Project Management, Slope Stability Analysis, and Wave-Based Testing Techniques
 Earthwork Project Management, Slope Stability Analysis, and Wave-Based 74
Testing Techniques GSP 252 © ASCE 2014

The traditional deterministic method only uses one index, the factor of safety, to
assess the slope stability. It cannot judge the risk level of soil slope. Therefore, this
paper combines Morgenstern-Price method and effective sampling to research the
probabilistic properties of the soil slope. The algorithm used in this study can generate
the random fields with the lowest computational cost. And the generated random fields
have the property of universal cover of the whole space domain without the correlation
between any two sampling cases.

RANDOM FIELD MODEL

Downloaded from ascelibrary.org by New York University on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

Spatial Variability of Soil

Random field can capture the features of spatial structure of soil properties. A
typical finite-element mesh for the slope cross section considered in the paper is
shown in Fig. 1(a). For simulations of the random field, the sampling algorithm is
implemented in Matlab. Then, the simulated random field is mapped onto the slope
mesh as shown in Fig. 1(b).
10 23

5 13

3
0
0 5 10 15 20 25 30

(a) Mesh used for analysis (b) Shear strength mapped onto mesh
Fig. 1 Mesh used for random field

Gaussian random field is stationary because its joint probability distribution that
governs the field is invariant when translated over the parameter space. Which implies
that its mean u  x  , variance  2  x  , and correlation function   x, x ' are constant
for any location within the random field and that the covariance of two random
variables depends on their relative locations, regardless of their absolute locations
within the random field. In this study, an exponential autocorrelation function is used.
Different autocorrelation distances in the vertical and horizontal directions are
assumed as follows:
   2   2  
 x y 
  x, y   exp         (1)
     
   x     
y

where  x and  y = autocorrelation distances in the horizontal and vertical directions,

respectively.
The midpoint method, the simplest method of discretization, has been used to
handle the spatial variation of soil properties (Low, 2003; Cho 2007). In this method,
the field within one cell is represented by the random value at the centroid position.
The random variables can be correlated to one another by adjusting the spatial
correlation length  x ( or  y ).

Earthwork Project Management, Slope Stability Analysis, and Wave-Based Testing Techniques
 Earthwork Project Management, Slope Stability Analysis, and Wave-Based 75
Testing Techniques GSP 252 © ASCE 2014

Simulation of Random Fields

Random field generation of normally distributed variable x is discussed step by
step as follows:
Step 1: Generation of covariance matrix C
   2   2  
 x y 
C  x, y       exp        
2 2
(2)
 
   x    y   
Downloaded from ascelibrary.org by New York University on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

Step 2: Orthogonal decomposition to C using Karhunen-Loève (K-L) expansion


H  x,      x    i i   i  x  , x   (3)
i 1

where i and  i denote the eigenvalues and eigenfunctions of the covariance matrix
C , respectively, i   is a set of orthogonal random coefficients (uncorrelated
random variables with zero mean and unit variance).
Step 3: Sampling optimization of random fields
Because i   is totally random, there will be lots of redundant calculations in the
N sim times realization of k random variables (cell number of slope section). The
simulation is the standard Monte Carlo method. Consequently, Latin Hypercube
Sampling (LHS) and Cholesky decomposition method are employed to optimize the
generated spatial stochastic fields. The focus of these two methods is on the sampling
of i   , which will determine the permutation of the whole generated random fields.
Good permutation can reduce the correlation between any two independent
realizations of the random fields. For detailed descriptions of the algorithm, please
refer to Olsson et al (2002).

SEARCH FOR CRITICAL FAILURE SURFACE

The circular trial slip surface is assumed in this study. And the searching technique
for the global critical slip surface is shown in Fig. 2. It is an optimization process until
the slip surface associated with the lowest factor of safety is found.

Fig. 2 Searching for the critical slip surface

Earthwork Project Management, Slope Stability Analysis, and Wave-Based Testing Techniques
 Earthwork Project Management, Slope Stability Analysis, and Wave-Based 76
Testing Techniques GSP 252 © ASCE 2014

The grid above the slope is the grid of rotation centers, as   O  shown in Fig. 2.
Each grid point is the circle center for the trial slips. Line L1 and L2 are the up and
low limit of bottom of the trail slip surfaces, respectively. The distance L between
these two lines is divided into n equal parts. Then the radius of ith trail for each center
of circle is
i
R i  RL1  L, i  0,1, 2, n . (4)
n
where the RL1 is the perpendicular distance from center of circle to line L1 .
Downloaded from ascelibrary.org by New York University on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

If   O  owns m intersections, a total of m   n  1 computation times are

needed. The slip surface with the lowest factor of safety is chosen to be the global
critical slip surface. A readily available concise algorithm, proposed by Zhu et al
(2005), is employed to calculate the factor of safety.

EVALUATION OF ALGORITHM

The sampling algorithm should assure the sample distribution accord with the target
distribution, and reduce the number of simulation under the condition of satisfying the
calculation precision. Fig. 3 shows the similarity between the sample and target
distribution, which indicates that the sampling method used in this study performs
well.

0.20
Target distribution
0.15 cT  10.00kPa
 cT  2.50 kPa
Sample distribution
Pdf

0.10 cS  9.89kPa

 cS  2.53kPa
0.05

0.00
0.0 5.0 10.0 15.0 20.0 25.0
Cohesion (kPa)
Fig. 3 Sampling verification

As shown in Fig. 3, cT and  cT are mean and standard deviation value of the
given target distribution of soil cohesion, respectively. And cS and  cS are the
corresponding value for the sample distribution. These two kinds of distribution have
almost the same probability density function (pdf).

Earthwork Project Management, Slope Stability Analysis, and Wave-Based Testing Techniques
 Earthwork Project Management, Slope Stability Analysis, and Wave-Based 77
Testing Techniques GSP 252 © ASCE 2014

0.18 0.18
MC simulation LHS + Cholesky
0.15 0.15
Frequence 0.12

Frequence
0.12
0.09 Probability p=0.071 0.09 Probability p=0.074
0.06 0.06
0.03
N=10000
0.03
N=2000
0.00 0 1 2 3 4 5 0.00 0 1 2 3 4
10 10 10 10 10 10 10 10 10 10 10
Number of simulation Number of simulation
(a) MC simulation (b) LHS simulation
Downloaded from ascelibrary.org by New York University on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 4 Comparison between two methods

Fig. 4 shows the iterations for LHS method and MC method, respectively. As can
be seen from Fig. 4, to obtain almost the same failure probability for the same slope,
2000 times are needed for LHS method, while 10000 times for MC method. It means
that the sampling algorithm employed in this study is advantageous over the MC
method for its high efficiency.

EXAMPLE ANALYSIS

In this section, the simulated random field is used through direct coupling with the
slope stability analysis routine written in Fortran.
Literature review (El-Ramly 2003) indicates that the correlation length is within a
range of 10~40m in the horizontal direction, while in the vertical direction it ranges
from 1 to 3m. For simplicity, the correlation length is chosen to be 10m, 20m, 30m
and 40m in this study.
The shear strength parameter cohesion c is assumed to be lognormally distributed,
defined by a mean c and a standard deviation  c . Then the underlying normally
distributed ln c are given by
 ln c  ln 1   c2 c2  (5)
1
ln c  ln c   ln2 c (6)
2

Correlation Length
The slope used in this section is 5m high and stands at an inclination of 1V: 1.25H.
The cohesion c , is assumed to be lognormally distributed with c  10kPa
and  c  4kPa . The value of friction angle is a constant,   10 , and the vertical
correlation length is equal to 2m. For  x and  y are at the equivalent position in the
sense of geometry in Eq. (1), only the variation of horizontal correlation length is
taken into consideration here.

Earthwork Project Management, Slope Stability Analysis, and Wave-Based Testing Techniques
 Earthwork Project Management, Slope Stability Analysis, and Wave-Based 78
Testing Techniques GSP 252 © ASCE 2014

1.051 0.429
Mean of Fs Failure Probability

Failure probability
1.049 0.427

Mean of Fs
1.047 0.425

1.045 0.423

1.043 0.421
Downloaded from ascelibrary.org by New York University on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

10 20 30 40
Correlation length
Fig. 5 Influence of correlation length

Fig. 5 shows the mean of Fs and the slope failure probability when the horizontal
correlation length changes. It is clear that the mean of Fs increases with the increase
of horizontal correlation. Accordingly, the failure probability obeys the opposite law.
In spite of it, the variation of these two indices is in a very narrow range, which means
that the correlation length has low impact on the slope stability.

Slope Angle and Slope Height

Fig. 6(a) shows the influence of slope angle on the slope stability. The geometric
and strength parameters are as follows: slope height 5m,  x  10m ,  y  2m ,
c  10kPa ,   10 . The slope inclination changes, and so does the variation of soil
cohesion. As shown in Fig. 6(a), the slope failure probability decreases with the
decrease of inclination. This tendency is more obvious when the variation of cohesion
is bigger. It indicates that compaction uniformity of the slope soil is very important in
the construction, which shows great significance in the steep slope. For the slope with
inclination of 1V: 1.25H, the difference of failure probability would be more than 40%
between slopes with the variation of cohesion to be 0.1 and 0.5, respectively. For the
slope with low value of variation of cohesion, it is meaningless to enhance the slope
stability just by decreasing the slope inclination, which would be costly.
56 56
Variation of cohesion Variation of cohesion
Failure Probability (%)
Failure probability (%)

0.1 0.2 0.3 0.1

42 42
0.4 0.5 0.2
0.3
28 28 0.4
0.5

14 14

0 0
1:1.25 1:1.5 1:1.75 1:2 5 10 15 20
Inclination Height (m)
(a) Influence of slope angle (b) Influence of slope height
Fig. 6 Influence of slope angle and slope height

Earthwork Project Management, Slope Stability Analysis, and Wave-Based Testing Techniques
 Earthwork Project Management, Slope Stability Analysis, and Wave-Based 79
Testing Techniques GSP 252 © ASCE 2014

Fig. 6(a) shows the influence of slope height on the slope stability. The geometric
and strength parameters are as follows: slope inclination 1V: 1.5H,  x  10m ,  y  2m ,
c  25kPa ,   15 . The variation of cohesion changes as the slope height does. As
shown in Fig. 6(b), the significance of variation of cohesion would be magnified when
the slope height turns bigger. The failure probabilities, with the slope height of 15m,
are 0.01% and 12.83% when the soil cohesion variation is equal to 0.1 and 0.5,
respectively. However, when the slope height is 20m, the failure probabilities grow to
be 12.05% and 50.85%. As the slope height turns smaller, the slope turns to be more
Downloaded from ascelibrary.org by New York University on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

stable. Then the variation of strength parameter plays little role on the slope stability.
It can be concluded that the compaction uniformity is very important for the slope
with steep inclination or great height. Meanwhile, it demonstrates that the losing sight
of the variation of strength parameter is totally unwise in the slope stability analysis.
The uniformity assumption, for the slope soil in the deterministic method evaluating
the slope stability, needs further study.

Variation of Cohesion
Fig. 7 shows the indices for evaluating slope stability, the mean value of Fs and
failure probability, varying with the change of cohesion. The numerical model is the
same as the one used in Fig. 6(a). The points A , B and C , as marked in Fig. 7,
represent three different cases. The mean value of Fs of A , B and C are 1.073,
1.185 and 1.265, respectively. Point A would be definitely the safest in the framework
of deterministic method, followed by point B , and point C is the most dangerous.
However, the failure probability just performs the opposite: point C is the safest for
its failure probability is the smallest and point A is the most dangerous. It is
interesting and instructive in assessing the slope stability. The results show that neither
the factor of safety nor the failure probability can assess the slope safety separately. In
the evaluation of slope stability, the combined using of these two indexes could be
more effective and reliable.

56
Variation
42 0.1
Failure probability (%)

0.2
28 0.3
0.4
A 0.5
14 B
C
0
1.029 1.148 1.265 1.376 1.490
1.043 1.168 1.288 1.406 1.525
-14 1.057 1.185 1.311 1.436 1.559
1.067 1.200 1.331 1.459 1.587
1.073 1.209 1.342 1.475 1.607
-28
10 12 14 16 18
Cohesion (kPa)
Fig. 7 Influence of cohesion

Earthwork Project Management, Slope Stability Analysis, and Wave-Based Testing Techniques
 Earthwork Project Management, Slope Stability Analysis, and Wave-Based 80
Testing Techniques GSP 252 © ASCE 2014

CONCLUSIONS

Numerical model is used to investigate the influence of sensitive parameters on the

slope stability from the probabilistic point of view. Spatial variability is taken into
consideration and LHS method is used to generate the spatially random field for
calculations.
The sampling verification demonstrates that the code for LHS method is good
enough to satisfy the precision. And more important is that it can enhance the
calculation efficiency greatly.
Downloaded from ascelibrary.org by New York University on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

Parametric study shows that the correlation length has low impact on the slope
stability. The failure probability increases with the increasing of slope angle, slope
height and the coefficient of variation. For a given slope, the results show that neither
the factor of safety nor the failure probability can assess the slope safety separately.
Only the combined using of these two indexes could be effective enough for the slope
safety evaluation.

REFERENCES:

Cho, S. E. (2010) “Probabilistic Assessment of slope stability that considers the spatial
variability of soil properties.” J. Geotechnical & Geoenv. Engrg., Vol.136(7): 975-
984.
El-Ramly, H., Morgenstern, N. R., Cruden D. M. (2002) “Probabilistic slope stability
analysis for practice.” Can. Geotech. J., Vol.39 (3): 665-683.
El-Ramly, H. (2003) “Probabilistic stability analysis of a tailings dyke on presheared
clay–shale.” Can. Geotech. J., 40: 192-208.
Fenton, G. A. and Griffiths, D. V. (2008) “Risk Assessment in geotechnical
engineering.” John Wiley & Sons, Inc. Hoboken, New Jersey.
Griffiths, D. V. and Fenton, G. A. (2004). “Probabilistic slope stability analysis by
finite elements.” J. Geotechnical & Geoenv. Engrg., Vol.130 (5): 507-518.
Li, K. S. and Lumb, P. (1987) “Probabilistic design of slopes.” Can. Geotech. J., Vol.
24(4): 520-535.
Low, B. K. (2003) “Practical probabilistic slope stability analysis. ” 12th Panamerican
Conference on Soil Mechanics and Geotechnical Engineering and 39th U.S. Rock
Mechanics Symposium. MIT, Cambridge, Massachusetts, vol. 2: 2777–2784.
Olsson, A. M. J. and Sandberg, G. E. (2002) “Latin hypercube sampling for stochastic
finite element analysis.” J. Engrg. Mech., Vol 128(1):121-125.
Zhu, D. Y., Lee, C. F., Qian, Q. H, et al. (2005) “A concise algorithm for computing
the factor of safety using the morgenstern–price method.” Can. Geotech. J., Vol
42(1): 272-278.

Earthwork Project Management, Slope Stability Analysis, and Wave-Based Testing Techniques