PROBABILISTIC STABILITY ANALYSIS OF EMBANKMENTS AND
SLOPES
By
Rossen A. Halatchev 1
Some attempts at probabilistic analysis of seismic
ABSTRACT stability are also known where the problem is more
complex and less adaptable to rigorous analysis.(Young,
This study introduces a new probabilistic approach to 1984; Hadj-Hamou and Kavazanjian (1985). These
seismic stability analysis of embankments and slopes. The investigations have application in a wide spectrum of
approach is based on Sarma's solution, which is modified dynamic impacts such as earthquakes, explosions, and in
through presentation of the seismic force by its horizontal loadings of transport and other kinds of machinery.
and vertical components; i.e. the force may have an The analogous modelling of the seismic stability
arbitrary inclination. The probabilistic modelling includes using Sarma's solution (1979) is presented in this paper. It
Monte Carlo simulation of the shear strength parameters of is modelled on a two-dimensional failure mechanism. The
the soil mass, which are assumed to have a normal probabilistic approach is based on the assumption that
distribution. The probability of failure is determined based . every lithological type in the zone of the investigated
on the seismic co-efficient, which is treated as a random engineering-geological cross section is homogeneous and
quantity. The probabilistic approach is illustrated through a the variability of the strength parameters has a stationary
solution of a numerical example from an opencast coal character within·its spatial borders.
mine. The deterministic stability model was modified with
INTRODUCTION the aim of taking into account the seismic force action with
an arbitrary inclination, rather than only a horizontal
From the literature on the probabilistic methods developed inclination as considered by Sarma.
at present, it is concluded that they are based on well- The probabilistic analysis included the Monte Carlo
known conventional deterministic methods of limit simulation, which was based on the assumed normal
equilibrium (Young, 1984; Chowdhury, 1986; Halatchev, distribution of the shear strength parameters. A criterion in
1987). In one way or another, these methods pre- the probabilistic model is the seismic co-efficient. Its
determine the sphere of their application and the adequacy distribution was determined and the results tested for
limits of the probabilistic modelling. most frequently conformity with normal, exponential and Rayleigh
modelled is the influence of the stochastic character of the distributions.
strength parameters' variance in the general stability
assessment. It is assumed that the main uncertainty in this DETERMINISTIC STABILITY ANALYSIS
assessment is due to the variability in the shear strength, Sarma's method is intended as a pseudo-static analysis,
which is mobilised on the failure surface. Usually, shear which leads to modelling the seismic stresses, varying with
strength parameters are described as having a normal time and space, through a set of horizontal forces that are
distribution, and useful in engineering pntctice. constant over time and have the same value at each point of
the soil mass. It is a simplified way of presenting the
University of Mining and Geology, Sofia, Bulgaria action of the seismic acceleration, and takes a single value
11th International Conference on Ground Control in Mining, The University of WoUongong, N.S.w., July 1992.
432
corresponding to the maximum, in order to guarantee the Ej + I = aj - pjK + Ejej (1)
stability in extreme conditions. Also, the acceleration is where:
considered only in a horizontal direction, whereas in aj and ej are co-efficients by Sarma
reality, the major manifestations such as earthquakes and Pj - co-efficient
explosives, exit in varying directions. Therefore the
question arises whether the use in the pseudo-static analysis Wjsin(9- cI>j + aj
Pj= ,
of a seismic acceleration in a horizontal plane, adopted by
cos(cI>j - aj + cI>j+1 - Sj+l)seccI>j+1 (2)
Sarma as critical, satisfies the specificity of this problem
fully and accurately.
In the absence of external forces (En + I = El = 0), the
following relationship for K is obtained about the whole
sliding body:
n-I n
an + L Ilk IT ej
K= k=1 j=k+1
n-I n
Pn+ L Pk j=k+1
k=1
IT ej (3)
The expression of the nonnal force on the slice
base, estimated after the determination of the forces acting
E.=E!+PW.
111
on the shear surfaces, is:
Xi,Ei-body force on slice boundaries
Wi - weight of the slice Nj = [Wj(1+Kcos8) + Xj+IcosSj+1 - XjcosSj-
xgi,ygi- coordinates of gravity-centre -Ej+lsinSj+1 + EjsinS~+ UjtancI>jsinaj -
xi'Yi - coordinates of p.A
-c;bitanaj -]coscI>;sec(cI>j - aj) (4)
Ni,T i - normal and resisting forces'in
the slice base
The verification of the condition for the equilibrium
of moments of the slice forces towards the lower left corner
Fig. 1 Geometry of, and forces acting on, a single slice. point A, is a compulsory element of the method and should
be made by using the formula:
In answer to this question, a modification of
Sarma's solution is presented, with the assumption that the Njljbjsecajcos(<Xj+Sj+I)-EjZj+Ej+1
seismic force, which acts as a vector in every slice, is not [~+I + sinbjsecajsin(aj+Sj+l)] - Wj[(xgj - Xj)
strictly fixed on a horizontal plane, (Fig. 1). This force (1 + sinKcos8) - Ksin8(ygj - yJ = 0 (5)
(Si=Sicos9) components, (K is the seismic co-efficient, 9 -
angle subtended by Si from vertical. The factor of safety F is calculated iteratively by
Between K and its horizontal (Kh) and vertical (Kv) reducing the shear strength parameters
components, the following relationship exists:
. ~
o./F, tancI>JF, c/F,tancI>JF in equation (3) until the
K = (K~ + K~)o.5.
calculating seismic co-efficient grows to zero or a given
Having solved the set of static equilibrium critical value, corresponding to the aims of the stability
equations of the forces on the slide base and shear surfaces analysis.
inside the sliding mass, a Coulomb-Mohr criterion is
obtained. For i th [Link] equation is:
11th International Conference on Ground Control in Mining, The University of Wollongong, N.S.W., July 1992.
433
PROBABILlSTIC STABILITY ANALYSIS value of the seismic co-efficient (Kcrid is used instead of
zero, for the seismic stability analysis. Analytically this
The factor of safety, in accordance with the computing probability, as a more common form of expression, may be
presented in the following way:
procedures of the method, is a fixed deterministic quantity.
Because of this, the seismic co-efficient is used as a
Prob (K > Km! I F = 1) (6)
criterion of the stability in the probabilistic model, as it is a
function of physical parameters. The seismic co-efficient in
The postulated probability can be assessed if the
reality can be considered as a continuous random quantity,
seismic co-efficient distribution is established. There are
which is characterised by an interval of variance, (Fig. 2).
some investigations modelling the seismic co-efficient with
K exponential (Augusti et al., 1984) and Rayleigh (Haji-
l. Distribution density
t\.
function Hamou et aI., 1985) distributions. Therefore the
..., Expectation consideration of a given distribution as absolute is incorrect
'"cv 1 and its verification is compulsory during the realisation of
'r<
tJ
2
the model (6) for every problem.
'r<
'H
- The following assumptions are made in the
'H
cv
K
crit ---" -
0 modelling:
tJ
tJ
The law of distribution of K is determined in
'r<
E I relation to the parameters cohesion and frictional co-
Ul
'r< 0 I efficient for reasons presented in the introduction; the shear
cv
UJ
1 F F strength parameters are modelled with a normal
s
Factor of safety distribution; the unit weight is considered as a deterministic
quantity because of its variance in very narrow limits; the
Fig. 2 Stochastic relationship between factor of safety and
geometrical quantities of the sliding body are also
seismic co-efficient.
deterministic.
MONTE CARLO SIMULATION
A proof of this statement are the samples, formed for the
The expression (3) represents a complex non-linear
physical parameters of equation (3), which is due to the
function of K with regard to the strength parameters. This
natural variability of the properties of the lithological types
circumstance excludes the possibility of an analytical
and to the errors associated with the selection and testing of
determination of the statistic characteristics of [Link] using
the soil samples.
the statistic characteristics of these parameters. A suitable
The condition for stability in the absence of a
alternative in this respect is the Monte Carlo simulation,
dynamic loading is F> 1 at K=O. This condition can be
whose realisation is reduced to the following procedures - a
represented as K>O at F= I, i.e. if the assessment of K is
computer generation of pseudo-random values distributed
bigger than zero. When the strength parameters are not
uniformly on a developed algorithm (Harbaugh et aI.,
reduced (F= I), at the following iterative process of
1970) and a transformation of the same ones into a new
computing, K will strive for zero. In parallel with it, the
empirical distribution on estimates of the statistical
factor of safety will take values bigger than unity, which is
characteristics of the strength parameters; alternative
used as a theoretical measure of the limit equilibrium.
assessment of the seismic co-efficient for forming a
Figure 2 is illustrative of this conclusion.
sample, characterising its stochastic nature of variance;
The probability of non-failure is identified with the
interpretation on the distribution of K and estimation of the
probability that a calculated value of the seismic co-efficient
probability of non-failure.
will be bigger than zero in the case that F takes an initial
IMPLEMENTATION
value in the computing process equal to unity. A critical
A computer program PROBSAR, written in Quick BASIC-
11 th International Conference on Ground Control in Mining, The University of Wollongong, N.S.w., July 1992.
434
4th version of Microsoft Corp. was developed for a requirements of this type of research progmrn. (Georgiev,
pmctical realisation of the approach described above. The 1972).
program is workable on 16 bits IBM compatible It It
computers. It was developed on a module principle and C=L ajcj and f =L b}j
includes 8 sub-routines. The main progmrn is intended for j=l j=l (7)
computing the seismic co-efficient and governing the sub-
routines. where: fj is the frictional co-efficient of the j-th lithological
type
Practical Example aj and bj are co-efficients:
The stability of a slope in Troyanovo opencast coal mine
It
(Bulgaria) was investigated. The slope height was 72m. aj=mlL mj
The overburden was presented by different types of clay, j=1 (8)
(Fig. 3). The statistical characteristic data of the physical
parameters are listed in Table 1. The stability was analysed
j-I
on the scheme of the wedge failure mechanism with a fixed
failure surface, passing through the border of the coal seam
(L ffip'Yp + O.5mJYj)mj
p=o
and grey-black clay. bj = - ' - - - - - - - - -
It j-I
The strength parameters of the shear surfaces inside L (L mp'Yp + O.5 mi'Yi)mj
the sliding masses were estimated as mean values in
;=1 p=o (9)
accordance with Sarma's method. Here they were
estimated as mean-weighted values in accordance with the
mj - thickness of the j-th lithological type.
Table l . Statistical data for physical parameters of the rock masses
2
Lythological Cohesion kN/m. Frict. coefficient [Link]
type 3
mean deviation mean deviation '{ , kN Im
1- Blue and
yellow clay 4.54 6.762 0.1584 0.0238 19.0 KEY:
2. Clay-sand I-prism of acti
complex 4.85 6.257 0.1228 0.0315 19.2 ve pressure
3. Black clay 4.60 5.234 0.0875 0.0177 17.4 2-central block
4. Coal 38.00 7.704 0.6009 0.0321 13.3 3-prism of pa-
5. Clay of inter- ssive presure
strata 4.72 6.433 0.0524 0.0103 17.5 H -critical
90
height
I
0
I
50 100
,
r
2 r .I.
m
failure surface type 1 1\ .7 1
rH90
~
11
£/ I type 2
tvne "'I type 4
.
\.
\. /
7
> v
.-Y
type 5
Fig.3 Slope of a spoil bank in Troyanovo coal opencast
11 th International Conference on Ground Control in Mining, The University of Wollongong, N.S.w., July 1992.
435
The expressions from (7) represent linear functions acceleration is proved as correct. The probability of non-
of random quantities (c', f'). Hence their mathematical failure for thi~ case is minimum (38%).
expectations and dispersions will be:
_ I _111111 Plot rf1f' k
Stitlstlcs
..
k k Iota ril.: [Link]
Mc = L ajMc'· and Mf =L bjM( " To~1 : 211
j=1 J j=1 J (10) " niss 8
P(Ie:. 0) -I
- "1" I
"Used : 2lIll
P(K:> 0.1 -I
k k 00%
p( It:> 0.2 6 1 % h. .225
~ =L L ahcrcj crc~Pcj.p .25 I i "~ I
Uu:ance:
std, Dev:
.1118
.812
j=1 p=1
~i~1
(11) x C,U, : 5.Z'15
Sllelmess: .fEj/
wtosis: 2.6'/5
,,J
~ "inhllll : .1'l4
.21 25th ~ : .215
t
llediaJI : .225
15th x : .233
llaxi_ : .253
.15
where: II 31 58 111 • 99
M' M'
Cj and ~ are respectively Ihe expectations of c'
e-Ioli.... '.lUIlt
and r of the j-th lithological Iype; Fig.4 Cumulative distribution of K
crC;, crc~, crrj , crr~ are respectively the mean square
Prob( >K crI't) ,%
deviations of c' and r of the j-th andp-th type
100
Pc;,,. Pr;" are the correlation factors.
F
s
The relationships (11) and (12) are worked out with
the assumption that there is a correlation between the
3.0
strength parameters with the aim of evaluating the degree of
independence based on the assumption of a normal
critical
distribution. _r--..!:.one
The slope stability was investigated at the force
incline e = 1500 and 200 trials were carried out by the 1.7
___ . =38%
ProbJ!!.ln _
Monte Carlo simulation technique. The probability of non-
failure was evaluated using the values of K: Og, [Link],
O.2g. The results are presented in the form of a cumulative
F . =0.77
distribution, (Fig.4). The sample was tested by the mln
Pearson's criterion and the hypothesis for a normal
o Loo-""'T'--4"T5---'---'9-0--r-1-3r-5-""'T'-e"::0:-'1
distribution was confirmed as most probable against the
alternatives for exponential and Rayleigh distributions. A
Fig. 5 Factor of safety and probability of non-failure
directional analysis was also carried out by assuming a
against seismic force inclination.
constant seismic co-efficient of O.2g and varying its
inclination from oo to l8oo, as indicated in Fig. 5. It
CONCLUSION
turned out that the range of variation in F is large enough.
The minimum factor of safety (0.77) corresponds to e= 1. The modification on Sarma's method allows
90 0 , [Link]' s postulate for the critical horizontal
investigation of the seismic stability with an arbitrary force
11 th International Conference on Ground Control in Mining, The University of Wollongong, N.S.w., July 1992.
436
inclination in the sliding mass. Geor!,'iev, G.K.,1972. Development of the slide processes
2. The probabilistic approach is developed for in the coal opencasts. 1. Vaglista, No. 6: 3-10, (in
modelling the stability in the conditions of uncertainty of Bulgarian).
the information about the shear strength of the soil mass. Hadj-Hamou, T. and Kavazanjian, E., 1985. Seismic
3. The application of the approach with the stability of gentle infinite slopes, I .. Geotech.
help of a computer program on an opencast slope confirms Engineering Div. ASCE (6) 3:681-697.
its effectiveness. Halatchev, R.A., 1987. A method for probabilistic
estimation of mine slope stability, I. Minno Del0,
REFERENCES No. 2:25-29, (in Bulgarian).
Harbaugh, I.W. and Bonham-Caner, G.,1970. Computer
Augusti, G., Baratta, A. and Casciati, F., 1984. Simulation in Geology, 1st ed., p: 253 (Wiley-
Probabilistic Methods in Structural Engineering, 1st Interseience: New York-London).
ed., p. 584 (Chapman and Hall: London-New Sarma, S., 1979. Stability analysis of embankments and
York). slopes, I. Geotech., Engineering Div. ASCE, (12)
Chowdhury, R.N., 1986. Geomechanics risk model for 105:1511-1523.
multiple failures along roc~ discontinuities, 1. Rock Young, D.S., 1984. Probabilistic analysis of blasting
Mech. Min. Sci. & Geomech. Abstr., (5) 23: 337- impact on open pit stability, Proceedings, 25th
346. Symposium on Rock Mechanics, Evanston,
Illinois: 1031-1041 (AIME: New York).
11 th International Conference on Ground Control in Mining, The University of WoIlongong, N.S.W., July 1992.
437