You are on page 1of 19

Engineering Geology 59 (2001) 233±251

www.elsevier.nl/locate/enggeo

Development of a probabilistic approach for rock wedge failure


H. Park a,*, T.R. West b
a
Geotechnical Engineering Division, Korea Institute of Construction Technology, 2311 Taehwa-Dong, Ilsan-gu, Koyang, Kyonggi-do 411-712,
South Korea
b
Department of Earth and Atmospheric Sciences, Purdue University, West Lafayette, IN 47907, USA.
Received 14 March 2000; accepted for publication 13 October 2000

Abstract
For rock slope engineering, uncertainty and variability are inherent in data collected on orientation and strength of disconti-
nuities, yielding a range of results. Unfortunately, conventional deterministic analysis based on the factor of safety concept,
requires a ®xed representative value for each parameter without regard to the degree of uncertainty involved. Therefore, the
deterministic analysis fails to properly represent uncertainty and variability, so common in engineering geology studies. To
overcome this shortcoming, the probabilistic analysis method was proposed and used for more than a decade in rock slope
stability analysis. However, most probabilistic analyses included a deterministic model as part of the analysis procedure causing
subsequent problems, which went uncorrected. The objectives of this paper are to develop a solution for these dif®culties in
probabilistic analyses and to propose an appropriate simulation procedure for the probabilistic analysis of rock wedge failures.
As part of the solution, probability of kinematic instability and probability of kinetic instability are evaluated separately to
provide a proper, combined evaluation for failure probability. To evaluate the feasibility of this new probabilistic approach, the
procedure is applied to a practical example, a major, highway rock cut in North Carolina, USA. Results of the probabilistic
approach are compared to those of the deterministic analysis; ®ndings are signi®cantly different, indicating that the determi-
nistic analysis does not depict rock slope variations, particularly where signi®cant scatter in parameter data occurs. q 2001
Elsevier Science B.V. All rights reserved.
Keywords: Probabilistic analysis; Rock slopes stability; Wedge failure; Simulation

1. Introduction and geotechnical parameters, are distinctive charac-


teristics of engineering geology Ð geotechnical engi-
In rock slope engineering, an important aspect of neering studies dealing with natural materials Einstein
engineering geology studies, the natural materials and Baecher, 1982). In rock slope stability analysis,
comprising most slopes posses an innate variability the uncertainty and variability may be in the form of a
that is dif®cult to predict or calculate. Uncertainty large scatter in discontinuity attitude data, the geome-
also arises from insuf®cient information concerning try of jointing or laboratory test results, the latter
the site conditions and incomplete understanding or performed to characterize discontinuity parameters
simpli®cation of a failure mechanism. Therefore, and properties. Consequently, one of the most impor-
uncertainty and variability in geologic conditions tant and dif®cult challenges in rock slope engineering
is the selection of a representative value for stability
* Corresponding author. Fax: 182-31-910-0211. analysis from an array of widely scattered data.
E-mail address: hjpark@kict.re.kr (H. Park). Therefore, many engineers and researchers,
0013-7952/01/$ - see front matter q 2001 Elsevier Science B.V. All rights reserved.
PII: S 0013-795 2(00)00076-4
234 H. Park, T.R. West / Engineering Geology 59 (2001) 233±251

attempting to limit and quantify variation and uncer- equilibrium analysis. The deterministic method of
tainty in their data, have adopted various methods to limit equilibrium analysis for rock slope stability
select appropriate, representative values for disconti- was introduced by Jaeger (1971) and Kutter (1974),
nuity parameters. In an attempt to overcome uncer- and recent techniques, which are commonly used,
tainty and variability, the probability theory and were established by Hoek and Bray (1981) and Good-
statistical techniques have been applied to slope stabi- man (1976). Numerous applications of this analysis
lity analysis. Its application has provided an important and useful ®eld adaptations have been accomplished
tool for quantifying and modeling variability and in the past 20 years (Kovari and Fritz, 1984; Watts and
uncertainty. However, since most probabilistic West, 1985; Sakurai and Shimizu, 1987).
analyses include a deterministic model within their A slope stability analysis consists of a two step
probabilistic procedure, subsequent complications process. The ®rst is to determine if the orientation
develop. This is because the deterministic model for of the discontinuities could result in instability for a
wedge failure, which is commonly adopted as the rock slope de®ned by those discontinuities. This
basis of the probabilistic solution, is not really appro- determination is usually accomplished by means of
priate in the probabilistic approach. Another possible stereographic analysis of the structural fabric and is
problem arises because kinematic instability and commonly referred to as kinematic analysis (Piteau
kinetic instability are not clearly distinguished in and Peckover, 1978). If the kinematic analysis indi-
most probabilistic methods. Apparently, this occurs cates that the structural condition is potentially
because many researchers failed to understand the unstable, the kinetic stability of the body is assessed
relationship between kinematic and kinetic analysis using the limit equilibrium method; this provides the
and consequently did not consider probability of second step.
instability in the kinematic analysis. In the deterministic analysis, a representative orien-
Therefore, this paper proposes answers and tation (the mean value in many cases) for a particular
improvements to the problems mentioned above and joint set is determined and kinematic tests are carried
described later in detail when considering wedge fail- out using a stereographic projection involving a single
ure. Finally, the proposed probabilistic approach is representative value for each discontinuity set. Then a
applied to a practical example, stability analysis of a single representative strength parameter for each
highway rock cut in North Carolina, USA. The discontinuity set is determined and this value is used
purpose is to evaluate the validity of the probabilistic to calculate the FS in a deterministic analysis. The
approach and to compare results between probabilistic result of the deterministic analysis for limit equili-
and deterministic methods. brium is a single FS, based on the representative
values of discontinuity parameters, which are typi-
cally taken as the mean values.
2. Deterministic analysis method However, most input parameters (e.g. material
strength, joint geometry and pore water pressures) in
A deterministic analysis, because of its simple the FS calculation are not precisely known because of
calculation, is commonly used to evaluate the stability uncertainty and variations in testing, modeling, and
of a slope system, based on ®xed values of dis- spatial variation. Thus, each of these parameters
continuity parameters. In practice, the values for is a random variable and the analysis using
discontinuity parameters are determined from a repre- different values for each of these parameters can
sentative array of a large number of scattered data, result in a different FS value. Thus, safety factor
obtained from site investigation and laboratory, or in itself is a random variable, depending on many
situ testing. Traditionally, assessments of the risk of input parameters.
failure are made on the basis of allowable factors of However, the traditional concept of FS does not
safety, obtained through previous experience for the re¯ect the degree of uncertainty of these parameters.
system considered, under the existing environment. In In most cases, the mean value of each parameter is
slope stability analysis, the factor of safety (FS), assigned as a ®xed value for calculation of FS.
which is used to ensure stability, is based on limit However, some engineers tend to select values higher
H. Park, T.R. West / Engineering Geology 59 (2001) 233±251 235

or lower than the mean, due to uncertainty and varia- as a random variable and can be replaced by the prob-
tion in input parameters. This can yield completely ability of failure to measure the level of slope stability.
different FS values for the same project. The National The probability of failure is simply de®ned as the
Research Council (1995) and Harr (1987) pointed out probability of having FS # 1 given as a percentage
that the conventional factors of safety commonly used that is equal to the area that FS # 1 under the prob-
by engineers are based on nominal values of resis- ability density function (PDF) for FS.
tance and load. In addition, although the conventional In general, probabilistic analysis is performed by
FS depends on the physical model, the method of two procedures (Tabba, 1984): the ®rst step consists
calculation, and mostly on the choice of parameters, of analysis of available geotechnical data to determine
there is usually a lack of consensus for these choices. the basic statistical parameters (that is, mean and
Consequently, inconsistency is likely to exist among variance) and PDF in order to represent and predict
engineers and between applications by the same engi- the random property of these geotechnical parameters.
neer. Also, the same FS can be associated with a large The PDF is used to model the relative likelihood of a
range in reliability. For example, according to Tabba random variable. In cases where it is believed that a
(1984), when FS ˆ 1.5, the probability of failure can given set of measured data represents a set of repre-
increase from 10 25 to 10 22 when the standard devia- sentative sample values of the variable, and no other
tion increases from 0.15 to 0.25. Thus FS is not a information is available, a probability density distri-
consistent measure of safety or risk. bution is representative of the random variable. The
Another disadvantage in the deterministic analysis mean value of the PDF represents the best estimate of
is that FS values cannot be compared for different the random variable, and the standard deviation or
modes of failure. According to Einstein and Baecher coef®cient of variance (c.o.v.) of the PDF represents
(1982), the mechanically equivalent de®nitions of an assessment of the uncertainty.
failure have different FS values. In Low's (1997) In the second step, risk analysis of slope stability is
analysis of rock wedges, the same safety factors for accomplished using the basic statistical parameters
different sliding modes with the same slope geometry and the probability density distribution developed
and same joint sets have very different probabilities of from the previous step. That is, once the probabilistic
failure. Consequently, as Tabba (1984) pointed out, properties of input parameters are assumed, the prob-
the following shortcomings occur in deterministic ability of failure can be evaluated by many different
analyses: (1) inability to account for variations in risk analysis procedures. The Monte Carlo simulation
properties and conditions; (2) dif®culty in portraying method is commonly used to evaluate reliability of the
the relative importance of various sets of data in the slope system when direct integration of the system
overall stability condition. function is not practical, but the PDF of each compo-
nent variable is completely prescribed. In this proce-
dure, values of each component are generated
3. Probabilistic analysis method randomly by its respective PDFs and then these values
are used to evaluate the factors of safety. By repeating
In the deterministic approach of slope stability, the this calculation, the probability of failure can be esti-
FS and all input parameters take on ®xed values in mated by the proportion of results in which the safety
spite of the fact that these parameters and even the FS factor is less than one. This estimation is reasonably
shows a degree of uncertainty. Therefore, despite the accurate only if the number of simulations is very
simplicity of the traditional deterministic approach, it large. The advantage of this method is that the
does not properly represent the uncertainty and varia- complete probability distribution for the FS is
bility in parameters and in the analytical or empirical obtained if the PDF of input parameters is assessed
models. precisely and correlation between the input para-
As an alternative to the deterministic approach, the meters is estimated. The disadvantage is that large
probabilistic analysis has been introduced to quantify numbers of simulations are required when the failure
the uncertainty and variability in parameters and in probability is relatively small. In addition, this simu-
the analytical model. In the analysis, FS is considered lation may cause errors if the probability of failure is
236 H. Park, T.R. West / Engineering Geology 59 (2001) 233±251

Fig. 1. Stereoplot of data required for wedge stability analysis (after Hoek and Bray, 1981).

very low. When the PDFs of the component variables tion function. The disadvantage is that mathematical
are not available, but their mean and c.o.v. are known, calculations are dif®cult when the number of compo-
the ®rst-order second-moment method (FOSM) can nent variables is large.
be used to calculate approximately the probability of An alternative method for calculating the moments
failure. This method is based on the truncated Taylor for the FS is to use the point estimate method (PEM).
series expansion of the FS beyond the ®rst-order term. The FS is determined for all possible combinations of
It yields a good approximation if the uncertainties of one low and one high value (point estimate) of each
the variables are small. Inputs and outputs are component variable and then the combinations are
expressed as expected values and standard deviations. weighted by the product of their associated probabil-
Advantages of this method are: the calculation is ities (Harr, 1987; Wolff, 1996). This method also
simple and only information of moments (that is, requires only moment information for the variables.
mean and variance as ®rst moment and second An advantage of this approach is that correlation of
moment) are needed rather than a complete distribu- random variables can be easily considered in the
H. Park, T.R. West / Engineering Geology 59 (2001) 233±251 237

calculation of moments for FS. However, both FOSM and Bray (1981) (Fig. 1).
and PEM provide only estimates of mean and standard  
3 g
deviation for the FS and these methods do not provide FS ˆ …cA X 1 cB Y† 1 A 2 w tan fA
gH 2g
information regarding the distribution of FS.
 
g
1 B 2 w Y tan fB …1†
2g
4. Conditions for wedge failure
sin u24
Xˆ …2†
In the present paper, only the standard case of a sin u45 cos u2;na
wedge failure in a rock mass, de®ned by four planes
(slope face, top slope and two discontinuities) will be sin u13
Yˆ …3†
considered. As indicated previously, two conditions sin u35 cos u1;nb
should be satis®ed for rock slope failure: kinematic
and kinetic. cos c a 2 cos c b cos una;nb
Aˆ …4†
sin c 5 sin2 una;nb

4.1. Kinematic conditions for wedge failure cos c b 2 cos c a cos una;nb
Bˆ …5†
sin c 5 sin2 una;nb
The orientation of the line of intersection mainly
controls the kinematic analysis of wedge failures. The where cA and cB are the cohesive strengths of planes A
kinematic instability is based on the following three and B; f A and f B the angles of friction on planes A
conditions (Norrish and Wyllie, 1996). and B; g the unit weight of the rock; g w the unit
weight of water; H the total height of the wedge.
1. The trend of the line of intersection must be similar
to the dip direction of the slope face. 5. Development of probabilistic approach
2. The plunge of the line of intersection must be less
than the dip of the slope face. Several authors (Carter and Lajtai, 1992; Muralha
3. The plunge of the line of intersection must be and Trunk, 1993; Trunk, 1993; Leung and Quek,
greater than the angle of friction of the discontinu- 1995; Feng and Lajtai, 1998) have developed different
ity surface. probabilistic analysis methods for rock slope stability.
Basically these approaches used two different reliabil-
In most cases, stereographic analysis is a simple and ity analysis methods, such as Monte Carlo simulation
effective way to analyze kinematic stability. and FOSM, but these approaches were based on a
similar system failure function.
However, those probabilistic analysis methods
4.2. Kinetic conditions for wedge failure (that is, Monte Carlo and FOSM) used by previous
researchers are not entirely probabilistic procedures.
After kinematic analysis of wedge stability indi- This is because the system failure function, which
cates the possibility of a wedge failure, a kinetic Hoek and Bray (1981) established and other authors
analysis must be performed. Analysis of wedge stabi- have commonly used, is based on a deterministic
lity requires details of the wedge geometry as de®ned model and only some of the input parameters are
by the location and orientation of the bounding treated as random variables. For complete probabil-
surfaces. Wedge stability can be evaluated using the istic analysis, jointing should be described as a
FS concept by resolving the forces acting normal to random variable and the stochastic model for
the discontinuities and in the direction parallel to the discontinuities has to be considered for rock slope
line of intersection. The most common procedure used stability based on a random process. This is because
widely is the stereographic projection method of Hoek stochastic models present a unifying theory that
238 H. Park, T.R. West / Engineering Geology 59 (2001) 233±251

accounts for discontinuity properties observed in the The third problem is related to the probability of
®eld (Glynn, 1979). Therefore, some researchers system failure. Actually, the probability of failure,
(Call and Nicholas, 1978; Glynn, 1979) tried to evaluated by multiplication of the probability of
develop the stochastic model for discontinuity and kinematic instability and the probability of kinetic
rock slope stability. However, an appropriate instability, is a conditional probability that two and
stochastic model for rock slope stability has yet to only two joints intersect to form a wedge and the
be proposed. This is due to a limitation of under- wedge fails. Therefore, in order to analyze the stabi-
standing for the random joint generation process and lity of rock slopes, the probability of failure for the
a lack and dif®culty of observation for discontinuity entire slope should be evaluated. In all, these
parameters. problems must be solved to properly apply the prob-
Therefore, in the current research, the deterministic abilistic approach to a deterministic analysis model.
model for rock slope stability analysis, established by
Hoek and Bray (1981), is used. In addition, the Monte 5.1. Problems related to deterministic model
Carlo simulation method is utilized for reliability
analysis to evaluate the probability of failure. This is 5.1.1. Importance of deterministic analysis model
because the deterministic model for wedge failure of In most cases for the probabilistic approach, a
rock slopes is not easily solved by analytical means. deterministic failure model is adopted as the basis of
The Monte Carlo technique is a numerical simulation the probabilistic solution. This is one of the important
method that solves mathematical problems through factors affecting the probability of failure in a prob-
random sampling and repeated calculation. It is the abilistic analysis. According to Mostyn and Li (1993),
most complete method for risk analysis since all the a change in the deterministic model producing a 30%
random variables are represented by their joint PDF reduction in the FS has led to a 300% increase in the
and the probability of failure as the result of risk estimated probability of failure. However, the varia-
analysis is represented by its PDF (Mostyn and Li, tion in the FS in slope stability analysis is not really a
1993). problem because the acceptable FS can be condi-
However, when the deterministic failure model is tioned to a large extent by the choice of the determi-
adopted into a probabilistic analysis approach, several nistic model. Yet, there is no similar method to
problems require consideration and a solution. The accomplish this for the probability of failure (Mostyn
®rst is which deterministic failure model should be and Li, 1993). Another case showing the impact of the
used? The deterministic model is one of the important deterministic model is provided in the paper by Li and
factors affecting the probability of failure in a prob- White (1987). They indicated that a variation of two
abilistic analysis. Therefore, depending upon which orders of magnitude in the estimated probability of
model is selected, the probability of failure can be failure developed, entirely from the choice of which
different for a wide range of values. deterministic model to use.
The second problem is how to evaluate the prob-
ability of failure. In previous probabilistic analyses, 5.1.2. Selection of failure model in wedge failure
the concept of probability of failure was quite confus- As discussed previously, in a wedge failure, the
ing. The procedures for evaluating probability of fail- method of Hoek and Bray (1981) has been used
ure differ signi®cantly from one researcher to another, to analyze the stability of wedges, based on Eqs.
because kinematic instability and kinetic instability (1)±(5) above. However, this method requires a
were not clearly distinguished in most previous proce- complicated calculation procedure. In addition, the
dures. This problem is also related to the dif®culty angles (c5, u na;nb; u24 , u 45, u2;na; u13, u35, and u1;nb ) in
involving kinematic analysis. Because kinematic Eq. (1)±(5) are needed to calculate the coef®cients
analyses are commonly performed on stereographic X, Y, A and B, and the angles can only be measured
projections, it is not easy to evaluate kinematic from a stereoplot. That means that if this equation is
instability using the probabilistic approach and adopted in probabilistic analysis and is used in Monte
therefore to calculate the probability of kinematic Carlo simulation, a stereograph is needed for each set
instability. of many parameter combinations and then the angles
H. Park, T.R. West / Engineering Geology 59 (2001) 233±251 239

Fig. 2. Slope geometry in wedge failure (after Low and Einstein, 1991).

must be measured from the stereoplot. Also, as the the discontinuities and in the direction parallel to the
Monte Carlo simulation requires more and more calcu- line of intersection are calculated and after examining
lations, more stereoplots are required. Consequently, the sliding modes, the FS is determined.
these equations are not appropriate for use in a prob-
abilistic approach. 5.2. Problems related to kinematic analysis
As alternative method, Hoek and Bray (1981)
presented the equations for computing the FS by a In order to check the stability of rock slope systems,
systematic procedure in the appendix of their text- both kinematic and kinetic analyses are required
book. This solution can be easily adopted to the prob- based on the geometry and strength of discontinuities.
abilistic analysis because of its simple and systematic This is accomplished for both the probabilistic as well
procedure. The other advantage of this procedure is as for the deterministic analysis. The procedures
that factors of safety for four different failure modes in required for kinematic stability and kinetic stability
wedge failure can be calculated. The four different for the probabilistic analysis methods have been
sliding modes are: (1) wedge sliding in contact with proposed by McMahon (1971). Later Glynn (1979)
both joint planes; (2) sliding in contact with plane 1 also advocated this method. McMahon estimated the
only; (3) sliding in contact with plane 2 only; (4) probability of failure from the stereoplot by establish-
contact is lost on both planes. Therefore, the probabil- ing two regions of kinematic and kinetic stability on
ities of failure for these four different sliding modes the equal-area projection. After that, using the concept
can be evaluated individually by using this procedure. of the composite model, Glynn considered the prob-
It also includes the in¯uence of a tension crack, but in ability of failure as the multiplication of probability of
the current research, a tension crack for simplicity, is kinematic instability and probability of kinetic
not considered. In this method, based on attitude (dip instability. However, most researchers who performed
and dip direction) and mechanical (friction angle and probabilistic analysis, calculated only probability of
cohesion) values of discontinuity planes, unit vectors kinetic instability and then considered this to be the
of each plane and vectors of the lines of intersection of probability of failure in rock slope analysis. The major
various planes are evaluated. Using these values, atti- reason why these researchers failed to perform the
tude of line of intersection of two discontinuities is kinematic analysis seems to be that they did not
calculated, and areas of the faces and wedge volumes understand the relationship between the kinematic
are evaluated based on the geometry of the slope and and kinetic analyses. That is, even if a rigid block in
discontinuities. Finally, the forces acting normal to a rock mass has an FS less than 1, the block will not
240 H. Park, T.R. West / Engineering Geology 59 (2001) 233±251

αapparent the intersection line. In some cases where Eq. (6) is


used and the angle between dip direction of the inter-
αi
section line and the dip direction of the slope face is
very large, it is possible that a stable rock block is
interpreted as being an unstable block. This develops
Plane A because the true dip of the intersection line is less than
the true dip of slope face, indicating that the block is
Direction of sliding unstable. However, in reality, the true dip of the inter-
Plane B section line is greater than the apparent dip of slope
face and the block is actually stable kinematically.
Therefore, in order to prevent this problem, the appar-
ent dip of the slope face should be evaluated and
Eq. (6) should be corrected accordingly.
Dip direction of slope The apparent dip can be determined using the
Slope face face following equation:
 
Fig. 3. Sliding along the line of intersection of plane A and B (after a apparent ˆ tan21 tan a cos…bi 2 bs † …7†
Hoek and Bray, 1981).
where b i and b s are the dip directions of the lines of
intersection and slope face, respectively. Therefore,
undergo failure if the block is kinematically stable. In Eq. (6) should be corrected as follows
addition, the dif®culty in performing kinematic analy-
sis is another reason. Because many kinematic V , e , a apparent …8†
analyses are performed based on stereographic In the current study, the probability of kinematic
projection, a calculation of probability of kinematic instability was evaluated using this apparent dip of
instability is not easy to accomplish, as mentioned the slope face.
previously. The stereographic projection method is After the probability of kinematic instability is
not suited as input for the computational, repeated evaluated, the probability of kinetic instability is eval-
calculations in the Monte Carlo simulation which is uated separately. This is the solution to the problem
commonly adopted in a probability analysis. noted above in previous research reports, where
Therefore, Low and Einstein (1991) suggested a analysis procedures for the two conditions were not
simple equation for checking a kinematic instability clearly divided. Therefore, in the current research, the
for the formation of a wedge (Fig. 2). When e is the probability of kinematic instability and kinetic
dip angle of the line of intersection which can be instability were evaluated separately and later multi-
evaluated from the geometry of slope face and discon- plied to determine the probability of failure.
tinuities, they considered that the condition of kine-
matic instability was 5.3. Problem related to probability of slope system
V,e,a …6† failure

where V and a are given dip angles of the upper The probability of failure, Pf, is the probability that
ground surface and slope face, respectively. If Eq. (6) any two intersecting joints will form an unstable wedge
is not satis®ed, a wedge is not possible and a wedge and the wedge will fail. It can be computed as the multi-
failure does not occur. This equation is simple and can plication of the probability of kinematic instability and
be easily used in repeated calculations of probabilistic the probability of kinetic instability presented in the
analysis, especially, for the Monte Carlo simulation. previous section. However, Pf is actually a conditional
However, the authors found that this equation had a probability based on the premise that two and only two
problem. As can be observed in Fig. 3, the true dip of joints intersect to form a wedge (Glynn, 1979). In a real
the line of intersection should be compared with the situation, two or more wedges can form and fail at the
apparent dip of the slope face in the dip direction of same time. This is because there is a possibility that
H. Park, T.R. West / Engineering Geology 59 (2001) 233±251 241

several joints intersect each other and produce several Table 1


wedges. Therefore, P 0f ; the probability of failure for the Relation between types of system and correlation of failure surfaces
entire slope system, can be either less than, or larger than Type of Correlation of system
Pf. If the rock mass has widely spaced joints relative to system
the dimensions of the slope, the probability of failure of Independent system Perfectly correlated
the entire slope system will be lower than the probability system
of failure of the rock mass with closely spaced joints. Q
Series 12
Q (12Pfi) Max Pfi
Therefore, P 0f should be evaluated to analyze the stabi- Parallel Pfi Max Pfi
lity of the slope system because the design engineer is
concerned more about stability of the entire slope
system. This problem has been discussed by several mately equal, we will have
authors, but many probabilistic analyses consider only
Pf as the probability of rock slope system failure. P 0f ˆ 1 2 …1 2 Pf †N for independent wedges …11†
The probability of failure for the entire slope system,
P 0f ; is a function of Pf and the number of wedges formed P 0f ˆ Pf for perfectly correlated wedges …12†
in the slope. In addition, P 0f is in¯uenced by the correla- By contrast, if the system is parallel, the probability of
tion between possible failure surfaces. system failure is
In order to determine the relationship among the
Y
N
number of wedges, the probability of a wedge failure P 0f ˆ Pfi for independent wedges …13†
(Pf) and the correlation between failure surfaces, we iˆ1
need to understand system reliability theory. Accord-
ing to the theory of system reliability, if a system P 0f ˆ Max Pfi for perfectly correlated wedges
consists of several different components, the probabil- …14†
ity of system failure depends on the relationship
between the components composing the system. Again, if all Pfi are approximately equal,
Systems that consist of components connected in paral- P 0f ˆ PNf for independent wedges …15†
lel are such that the total failure of the system requires
failures of all the components. By contrast, systems P 0f ˆ Pfi for perfectly correlated wedges …16†
that are composed of components connected in series
are such that the failure of any one or more of these These relationships are summarized in Table 1.
components constitutes the failure of the system. In Therefore, if we consider that the rock slope
other words, if one component fails, the system is system, which is a type of series system, involves n
unable to complete the operation. A rock slope system wedges, the probability of slope system failure would
is a type of series system. When evaluating system be
reliability, probabilities of system failure are evaluated P 0f ˆ 1 2 …1 2 Pf †n
in a totally different way based on the relationship …17†
between components. That is, if the system is a series, in the case of independent wedges
the probability of system failure is
Pf ˆ Pf in the case of perfect correlation …18†
Y
N
P 0f ˆ12 …1 2 Pfi † if wedges are independent However, at the present time, there is no rational
iˆ1 method for assessing the correlation between possible
…9† failure surfaces. Therefore, the best we can do
currently is to recognize that the probability of system
failure is approximated using two extreme cases,
P 0f ˆ Max Pfi if wedges are perfectly correlated
which are the perfectly correlated and the independent
…10† cases:
If the probabilities of failure Pf1 to PfN are approxi- Pf , P 0f , 1 2 …1 2 Pf †n …19†
242 H. Park, T.R. West / Engineering Geology 59 (2001) 233±251

Table 2
Comparison with probability of failure and distribution for parameters (after Kulatilake et al., 1985)

Used distribution for probabilistic parameters Best distribution For FS with Probability of failure
(mean and c.o.v)
Orientation of plane A Orientation of plane B fA fB

Mean Mean Mean Mean FS ˆ 0.86 ±


Mean Mean Uniform Uniform Normal, lognormal (0.86, 0.10) 0.96
Mean Mean Normal Normal Normal (0.86, 0.10) 0.96
Mean Mean Lognormal Lognormal Normal, lognormal (0.86, 0.10) 0.95
Bivariate normal Bivariate normal Mean Mean Lognormal (0.88, 0.33) 0.72
Bivariate normal Bivariate normal Uniform Uniform Gamma, lognormal (0.88, 0.35) 0.70
Bivariate normal Bivariate normal Normal Normal Normal, gamma (0.89, 0.33) 0.68
Bivariate normal Bivariate normal Lognormal Lognormal Normal, gamma (0.89, 0.33) 0.68
Fisher Fisher Mean Mean Lognormal (0.94, 0.53) 0.67
Fisher Fisher Uniform Uniform Lognormal (0.94, 0.54) 0.64
Fisher Fisher Normal Normal Lognormal, gamma (0.94, 0.54) 0.62
Fisher Fisher lognormal lognormal lognormal (0.94, 0.54) 0.62

6. Statistical considerations for discontinuity and of experience and good engineering judgement are
slope parameters always needed (Muralha and Trunk, 1993).
The parameters used for probabilistic analysis can
In the process of developing the probabilistic be subdivided into two groups by their scatter and
approach, one of the important considerations is a randomness: deterministic parameters and probabilis-
statistical model of the geometrical and mechanical tic parameters. However, various researchers have
parameters of discontinuities. The stability of a rock designated different parameters as random variables,
slope is in¯uenced by different parameters of which leaving the remainder to be treated deterministically.
the most important are: Thus some have adopted the geometric parameters of
discontinuity and groundwater conditions as random
² orientation of discontinuities; variables only, others also treat the strength para-
² geometrical parameters of discontinuities such as meters as random variables. Therefore, a clear expla-
length, spacing and persistence; nation of the parameters and their properties used in
² shear strength of discontinuities. the current research is required before providing a
description of the simulation procedures.

Most of these parameters should be considered as 6.1. Deterministic parameters


random variables and types of distribution functions
for each random variable should be selected carefully Deterministic parameters are considered as known
in a probabilistic analysis. However, there is a lack of quantities (hence, deterministic) and have the same
consensus on these choices, which could lead to very values for all sliding blocks. Therefore, a ®xed single
different analysis results. As Kulatilake et al. (1985) value is chosen for this type of parameter and used in
pointed out, a different choice of density distribution probabilistic analysis. Slope orientation, slope dimen-
affects signi®cantly the probability of rock slope fail- sion and material properties belong to this class of
ure (Table 2). The distribution functions are chosen parameter.
according to physical reasons and values of their para- The slope is assumed to be straight and consisting
meters based upon research results and evaluation of of two planes, the upper and lower slope. Slope dip
the data. In addition, since the number of tests direction and dip angle of the lower plane and the
performed and data measured are generally insuf®- upper plane are required for stability analysis and
cient for a sound statistical analysis, a certain amount considered to be deterministic. Slope dimensions are
H. Park, T.R. West / Engineering Geology 59 (2001) 233±251 243

also assumed to be deterministic parameters. In plan direction of the joint plane. A Fisher distribution is
view, the slope is taken as a rectangular, so the hori- considered as the appropriate PDF to represent the
zontal length and height of the slope are considered as distribution of joint orientation. Among joint dimen-
deterministic parameters. Rock density or unit weight sion parameters, spacing, length and persistence are
is assumed as a ®xed value in this study. However, in major factors that control slope stability and these are
some papers such as that by Feng and Lajtai (1998), considered as random parameters. Spacing of discon-
density is considered as a random variable, but in tinuities controls the number of sliding blocks in the
current study the number of data was too small to slope system. Therefore, it affects the probability of
analyze or consider density to be a random property. overall system failure. A lognormal distribution is
selected as the PDF of this parameter based on
6.2. Probabilistic parameters measurements by the authors and the Chi-square
goodness-of-®t tests (Park, 1999). It is widely
Water pressure, joint orientation, joint dimension, accepted that trace lengths of discontinuities follow
and joint shear strength are considered as probabilistic an exponential distribution because strong physical
parameters because the tested and measured values of and logical reasons support this concept. In addition,
these parameters are widely scattered and therefore joint length is a major parameter used for evaluating
true values of these parameters cannot be evaluated joint persistence and it should be considered as a
precisely. random variable. In the current study, persistence is
Water pressure is a signi®cant parameter in slope considered as a function of the length of a disconti-
stability analysis, especially, in kinetic analysis nuity and the dimensions of the sliding block.
because water ®lls the sliding surface and affects the A bilinear strength envelope is used to approximate
driving and the resisting forces. Since the water level the true, non-linear behavior of discontinuity shear
¯uctuates according to weather, precipitation and strength. The distribution of friction angles is assumed
season, it is not easily measured or predicted. There- to be a normal distribution based on the experimental
fore, the water level is considered as a random vari- test data on the study areas and their distributions
able. In wedge failure analysis, the water pressure (Park, 1999). Hoek (1997) suggested a truncated
distribution is considered to be a pyramid shape and normal distribution for this variable because in some
the average water pressure on a discontinuity surface cases, the complete normal distribution provides
acting on the discontinuity surfaces can be evaluated unreasonably low or high values for the friction angle.
by
gw hw
uˆ …20† 7. Proposed probabilistic stability analysis
6
where g w is the unit weight of water and hw the height 7.1. Simulation procedure
of groundwater surface. In order to consider the water
pressure parameter as a probabilistic parameter, the The Monte Carlo simulation is frequently used to
height of groundwater surface is considered to be a evaluate the failure probability of a mechanical
random variable. The PDF of groundwater height is system, in particular, when direct integration is not
considered to have a uniform distribution in this practical or when the equation is dif®cult to integrate.
study. In this research, the Monte Carlo method was
Joint orientation, joint dimension, and joint shear employed because the deterministic model for rock
strength are the most important parameters regarding slope failure is not easy to solve by analytical
slope stability. They are de®ned separately for each means. The simplest Monte Carlo simulation
joint set and may be correlated among different joint approach is to assume that for a given stability analy-
parameters within the set and between sets. However, sis, each variable takes a single value selected
in this study, it is assumed that the different para- randomly from its measured distribution, independent
meters are independent among joint sets. of the other variables. The group of randomly selected
Joint orientation is expressed by the dip and dip parameters is combined with ®xed input data to
244 H. Park, T.R. West / Engineering Geology 59 (2001) 233±251

Start

Enter input data


(Slope and Discontinuity parameters)

Number of simulation, NT

Randomly pick dip and dip direction of discontinuity

Calculate slope geometry dimensions

Generate position of the sliding block and


height of groundwater surface

Kinematic analysis

Yes
Stable?
Increase # of iteration
No

Number of kinetic simulation, NK

Generate shear strength parameters

Calculate factor of safety

Check the number of kinetic iterations, NS

No
NS = NK ?

Yes

Check the number of kinematic iterations, NM

No
NM = NT ?
Increase # of iteration
Yes

Calculate probability of failure

Stop

Fig. 4. Flowchart of simulation procedure in probabilistic analysis.


H. Park, T.R. West / Engineering Geology 59 (2001) 233±251 245

generate a single random value for FS. This process is uated by a simple calculation from a variable u as
repeated many times to generate a large number of proposed in Leung and Quek (1995). Consequently,
different FS values, which can be plotted in histogram randomly generated dips and dip directions of the
form. The important point in this simulation process is discontinuity are obtained in this step of the proce-
that the values in the random sequence must re¯ect the dure. In the second step, the position of the sliding
distribution pattern and this indicates that a plot of the block and the height of groundwater surface are
sequence of random values as a histogram, should generated. For the wedge failure case, the intersection
provide the same shape as the parent distribution point of the intersection line of the two joints with the
(Priest, 1993). That is, a single value or a sequence slope face, which is a daylighting position, is de®ned
of values, which re¯ect the properties of the speci®ed randomly. Distribution of this position is considered
PDF, is generated. The simulation procedure of this as a uniform probability distribution along the height
study is discussed below and also expressed as a ¯ow- of the slope. Therefore, a new daylighting position on
chart in Fig. 4. the block is generated for each iteration. In addition,
The ®rst step of the simulation procedure is the the height of groundwater level is also assumed as a
generation of joint orientation. As mentioned uniform probability distribution along the line of
previously, the Fisher distribution is employed to intersection for the wedge failure case. Then, dimen-
represent the distribution for the discontinuity orienta- sions of the sliding wedge block are evaluated using
tion data. Therefore, in order to generate the joint the systematic procedure suggested by Hoek and Bray
orientation whose parent distribution is the Fisher (1981) for the wedge failure. The trend and plunge of
distribution, a variable u , which is an angle between the line of intersection, areas of the joint faces and
measured orientation and true orientation, obeys the volumes of wedges are evaluated based on the infor-
Fisher distribution. It has the following density mation for the orientation of two intersecting joints
function. and the intersection point for the line of intersection.
Based on these parameters, which are generated and
K sin u eK cos u evaluated in previous steps, the kinematic stability of
f … u† ˆ …21†
eK 2 e2K a wedge is checked. If sliding is possible kinemati-
where u is the angular deviation from the mean orien- cally, the next step of simulation for kinetic analysis
tation which ranges between 0 and p/2, and K is Fish- begins. However, if sliding is impossible, the proce-
er's constant. However, to simplify Eq. (21), the dure moves to the ®rst step of simulation, and then
following approximate expression for the cumulative another parameter value for discontinuity orientation
probability distribution is used. and daylighting position are generated. Then the
previous steps are repeated in the simulation proce-
P…u† < 1 2 eK…cos u21† …22† dure. In the next step, for the kinematically unstable
wedge, the shear strength parameters are generated
For the purposes of simulation of the Fisher distribu-
randomly using statistical parameters and the prob-
tion, P(u ) in Eq. (22) is replaced by the uniform
ability density distribution of shear strength para-
random value RiU and the variable u is replaced by
meters. The internal friction angle of the
the required random values RiF : Therefore, Eq. (22)
discontinuity is only considered as a shear strength
can be rearrange to give the desired form as follows:
! parameter and the PDF of this parameter is assumed
i ln…1 2 RiU † to be a normal distribution on the basis of experimen-
RF ˆ Arc cos 11 …23† tal test data. In this step, for the same wedge geometry
K
that occurs in the previous steps, many different shear
Next, a random number generated from the uniform strength parameters are generated and used to calcu-
distribution is substituted in RiU and RiF is calculated late many different factors of safety.
using Eq. (23). Repeating this calculation, a sequence After performing a suf®ciently large number of
of the variable u is generated and this generated set of iterations, the probability of failure is determined.
u values shows a Fisher distribution. The dip and dip The probability of failure consists of two failure prob-
direction of the generated discontinuity will be eval- abilities (or instability); the probability of kinematic
246 H. Park, T.R. West / Engineering Geology 59 (2001) 233±251

instability and the probability of kinetic instability. where P[A] is the probability that a wedge will
The probability of kinematic instability is evaluated daylight, that is, the wedge is kinematically unstable,
as the ratio of the number of iterations (or the number and P[B] is the probability that the driving force
of wedges formed in each iteration) that are deter- exceeds the resisting forces. Therefore in this study,
mined as kinematically unstable to the number of P[A] in Eq. (25) is similar to the probability of kine-
total iterations. The probability of kinetic instability matic instability and P[B] is similar to the probability
is the ratio of the number of the iterations that the FS of kinetic instability.
is less than 1 to the number of total iterations that As mentioned previously, the probability of kine-
the FS is calculated. Then the probability of slope matic instability is de®ned as
failure is evaluated by multiplying the probability of
kinematic instability by the probability of kinetic Nm
instability Pfm ˆ …26†
NT

7.2. Assessment of probability of failure where Nm is the number of iterations that are kinema-
tically unstable for the wedge and NT is the total
The de®nition of the probability of failure is some- number of iterations. Since the kinetic analysis is
times presented in a vague fashion in previous performed only when the wedge is kinematically
research publications. According to Quek and Leung unstable, the probability of kinetic instability is
(1995), the probability of failure is calculated: de®ned as
NF
Pf ˆ …24† Nf
NT Pfk ˆ …27†
Nm
where NF is the number of iteration that the wedges
failed, that is, FS is less than 1, and NT is the total
where Nf is the number of iterations that a wedge has a
number of iterations that wedges were analyzed.
FS less than 1. That is, the probability of kinetic
However, as Feng and Lajtai (1997) pointed out, NT
instability is a conditional probability based on the
can be interpreted in two different ways; NT is either
premise that the wedge is kinematically unstable.
the total number of iterations performed or only those
Therefore, on the basis of probabilistic theory, the
iterations that form kinematically unstable wedges.
multiplication of the probability of kinematically
Depending on this de®nition, the probability of failure
instability and the kinetically instability produces
will be different. Lajtai and Carter (1989) de®ne NT as
the probability of failure. That is,
the number of the iterations that form kinematically
unstable wedges and Feng and Lajtai (1998) used both
Nm N N
de®nitions without a clear distinction between them. Pf ˆ Pfm Pfk ˆ £ f ˆ f …28†
However, a clear de®nition and the results of prob- NT Nm NT
ability of failure are provided in the current paper. The
probability of failure is de®ned as the multiplication From this, the probability of failure should be de®ned
of the probability of kinematic instability by the prob- as the ratio of the number of iterations that the FS is
ability of kinetic instability. This multiplication is less than 1, which is based on premise that the wedge
based on the concept of composite models suggested is kinematically unstable, and the number of total
by Marek and Savely (1978) and Serrano and Castillo iterations. Hence, the multiplication of Pfm and Pfk
(1974). The concept of composite models states that provides a clear de®nition based on probability theory
all the individual components that affect probability of and simpli®es the evaluation of FS, thereby removing
failure must be compiled to calculate probability of any confusion. Based on this, the probability of kine-
failure. Glynn (1979) also suggested the composite matic instability and the probability of kinetic
model for calculation of FS. That is, instability are evaluated separately using the simula-
X tion procedure and are easily applied to evaluate the
Pf ˆ P‰AŠP‰BŠ …25† probability of slope stability.
H. Park, T.R. West / Engineering Geology 59 (2001) 233±251 247

Table 3
Input values for probabilistic analysis

(a) Input for slope geometry


Orientation of slope Height of slope (m) Unit weight of rocks (kN/m 3)
(dip direction/dip)
069/45 30 25.7
(b) Input for discontinuity properties
Set no. J1 J2 J3 J4 J5 F1
Mean orientation (dip 257/25 174/69 086/60 125/80 232/88 092/46
direction/dip)
Fisher constant 57 180 50 34 14 16
Mean friction angle (8) 40 40 40 40 40 27.03
STD of friction angle 3.78 3.78 3.78 3.78 3.78 2.94
Mean length (m) 3.7 3.7 3.7 3.7 3.7 2.4
Roughness (8) 9 9 9 9 9 6
Spacing (m) 0.9 0.9 0.9 0.9 0.9 0.9

8. Application of probabilistic analysis respectively. Friction angles for foliation have a mean
and standard deviation of 27.03 and 2.948, respec-
This probabilistic approach was applied to a stabi- tively. As discussed previously, a truncated normal
lity analysis for a rock cut along Inter-state Highway distribution is suggested by Hoek (1997). However,
26 in North Carolina, USA. A 30 m high slope with an according to investigations by the authors and testing
orientation of 069/45 was excavated in a rock mass in simulations, there is a very low possibility (about
which the upper slope was essentilly horizontal. The 0.3%) that friction angle will be less than 308 and
rock slope is located within the Blue Ridge Belt, a greater than 508 based on the testing data acquired
geologic province bounded to the west by the Great from these sites. Cohesions for both discontinuities
Smoky thrust fault and to east by the Brevard Fault were assumed to equal 0. However, it was decided
zone. The major lithology of the rock mass is Precam- to assign the same friction angle, mean length and
brian aged, Cranberry granite gneiss. It is a medium- roughness value for all joint planes. This is because
grained, even-textured rock that varies from light to of the limited number of measurements and testing
dark gray in color. A total of 280 discontinuities were results obtained. However, the random properties of
measured by means of the scanline method on rock foliation were collected separately and assigned
outcrops and existing road cut slopes, as well as use of different values. Input data for the slope geometry
a borehole method providing oriented cores. The sets and discontinuity properties are listed in Table 3.
of measured discontinuity orientations were deter- Therefore, using the input values for each parameter
mined using the clustering algorithm proposed by and the simulation procedure proposed previously, the
Mahtab and Yegulalp (1982). As a results of the clus- probability of slope failure was evaluated.
tering procedure, six sets, consisting of ®ve joint sets In this study, in order to compare the probability of
and one foliation set, were identi®ed and the mean (or slope failure with the traditional deterministic
representative) directions and Fisher constants for method, the FS for each case was calculated. This
each set were determined. Length, spacing and rough- determination of safety factor uses the same determi-
ness of discontinuities were also measured in the ®eld nistic models and the same formula as used in the
and their probabilistic properties were decided using probabilistic method. All deterministic parameters
visual comparisons and the Chi-square goodness-of- used in the probabilistic simulation were assigned
®t test. Shear strengths of discontinuities in this area the same values when calculating the deterministic
were determined by direct shear tests on discontinu- FS. However, for random parameters used in the prob-
ities. The shear strengths for joints and foliations were abilistic approach, mean values of respective distribu-
measured separately and the mean and standard devia- tions were used as is commonly done by most design
tion of friction angles for joints were 40.0 and 3.788, engineers for deterministic analysis.
248 H. Park, T.R. West / Engineering Geology 59 (2001) 233±251

Table 4
Results of wedge failure for the deterministic analysis and the probabilistic analysis

Set no. 1 Set no. 2 Deterministic Probabilistic

Factor of safety Probability of failure

Kinematic Kinetic Total Pf (%)

No contact Plane 1 Plane 2 Both planes

J1 J2 Stable 0 0 0 0 0
J1 J3 Stable 0 0 0 0 0
J1 J4 Stable 0 0 0 0 0
J1 J5 Stable 0 0 0 0 0
J1 F1 Stable 0.049 0.001 0 0.0048 0 0.028
J2 J3 Stable 0.005 0.132 0 0.011 0 0.072
J2 J4 Stable 0 0 0 0 0
J2 J5 Stable 0.003 0.1428 0 0.0144 0 0.047
J2 F1 Stable 0.316 0.0495 0 0.053 0.0002 3.25
J3 J4 Stable 0.006 0.092 0 0.046 0.0002 0.083
J3 J5 Stable 0 0 0 0 0
J3 F1 Stable 0.2713 0.0034 0 0.04 0 1.18
J4 J5 Stable 0 0 0 0 0
J4 F1 FS4 ˆ 1.53 0.526 0.0183 0 0.0713 0 4.71
J5 F1 Stable 0.315 0.0132 0 0.0584 0 2.26

The results of stability analyses are listed in Table 4. F1, J4 and F1 and J5 and F1 have possibilities of
As stated previously, a total of six discontinuity sets kinematic instability. Among these eight, four combi-
were identi®ed and for stability analysis, all possible nations, J2 and F1, J3 and F1, J4 and F1 and J5 and F1,
discontinuity combinations were generated and show a high probability of kinematic instability, 31.6,
checked for stability. As Table 4 indicates, the prob- 27.1, 52.6 and 31.5%, respectively. Yet, except for the
abilities of kinematic instability and kinetic instability J4 and F1 combination, the other combinations do not
were evaluated separately. In addition, the probability show kinematic instability based on the deterministic
of kinetic instability is evaluated for each mode of analysis. Therefore, as discussed previously, a single
sliding for a wedge failure. representative orientation (commonly a mean value)
According to the deterministic analysis results for for a discontinuity set does not properly represent the
each discontinuity combination, all discontinuity widely scattered orientation data. In such cases, the
combinations except one (J4 and F1) were shown to deterministic approach gives a misleading conclusion
be kinematically stable. That is, only one discontinu- in the stability analysis, which could lead to a serious
ity combination was kinematically unstable and there- problem such as a large slope failure.
fore only its FS was evaluated. However, even though However, as can be seen in Table 4, the probabil-
the J4 and F1 combination has a kinematically ities of kinetic failure for the combinations yield
unstable condition, the FS for this combination was mostly low values, ranging from 0.02 to 14.3%.
calculated to be 1.53. Therefore, the wedge formed by Thus, the probabilities that any two intersecting
the J4 and F1 combination should not fail. Conse- discontinuities will form an unstable wedge, Pf, are
quently, based on the deterministic analysis results, expected to be quite small. The Pf values for combi-
this slope has no possibility of failure and is quite nation J1 and F1, J2 and J3, J2 and J5, J2 and F1, J3
stable. and J4, J3 and F1, J4 and F1 and J5 and F1 are 0.028,
By contrast, the results of a probabilistic analysis 0.072, 0.047, 3.25, 0.083, 1.18, 4.71 and 2.26%,
indicate that eight discontinuity combinations, J1 and respectively.
F1, J2 and J3, J2 and J5, J2 and F1, J3 and J4, J3 and However, as can be observed in Table 3, the
H. Park, T.R. West / Engineering Geology 59 (2001) 233±251 249

Table 5 because the probability of kinematic instability is


Results for evaluation of the probability of slope failure and average evaluated based on geometry of the slope and orienta-
volume of possible wedge
tion of discontinuities, and the probability of kinetic
Set Set Probability of Average volume of instability is based on the strength parameters in a
no. 1 no. 2 entire slope failure possible wedge (m 3) given discontinuity orientation. Therefore, on the
basis of the evaluated probabilities, the parameters
J1 J2 0 0
J1 J3 0 0 that critically affect slope stability in a speci®c area
J1 J4 0 0 can be determined. For the combination such as J1 and
J1 J5 0 0 F1, J2 and F1 and J3 and F1, the probabilities of
J1 F1 0.0003±0.0039 0.48 kinematic instability is greater than the probabilities
J2 J3 0.0007±0.0207 0.02
of kinetic instability. Therefore, geometry of the slope
J2 J4 0 0
J2 J5 0.0005±0.0154 0.03 and orientation of discontinuities play a major role in
J2 F1 0.0324±0.9999 7.98 these combinations.
J3 J4 0.0008±0.0229 0.0002
J3 J5 0 0
J3 F1 0.0118±0.2479 0.5
J4 J5 0 0
9. Summary and conclusions
J4 F1 0.0471±0.6859 3.61
J5 F1 0.0226±0.4222 4.29 Probabilistic analysis has been used to quantify and
model the variability and uncertainty of input data,
which are commonly involved in engineering geology
spacing of discontinuities, which affects the probabil- studies. In probabilistic analysis, input parameters are
ities of entire slope failure, P 0f ; in this area is as small considered as random variables and their random
as 0.9 m. Therefore, P 0f values are expected to have properties are characterized using statistical para-
high values due to such small spacing values. This is meters and the PDF. Since the rock slope stability is
because the small spacing value indicates that numer- mainly dependent upon the characteristics of discon-
ous discontinuities occur in a slope cut and a great tinuities, the random properties of discontinuities play
number of wedges could form. The P 0f values for a critical role in probabilistic analysis of rock slope
each combination are shown in Table 5. As expected, stability. However, stochastic properties of disconti-
some upper bounds have high probabilities compared nuities cannot be fully described as random variables,
to their lower bound values that are the probabilities and consequently most probabilistic approaches for
that one wedge is formed. Especially, for the combi- rock slope stability include a deterministic model as
nation of joint set 2 and foliation set 1, the upper part of the failure procedure. Therefore, the problems
bound of the probability of slope system failure is due to the inclusion of a deterministic model need to
quite high (99.9%) and the expected volume of a be considered and solved. Previous probabilistic
possible wedge is approximately 8 m 3. In addition, procedures do not provide a comprehensive treatment
the upper bound probability of the J3 and F1 combi- of these problems.
nation is 24.8%, but the average volume is small, The study presents improvements that eliminate
0.5 m 3. Also, the upper bound of J4 and F1 and J5 some of the previous limitations. Utilized are the
and F1 show 68.6 and 42.2%, respectively and their equations of systematic procedures proposed by
possible failure volumes are 3.6 and 4.3 m 3. There- Hoek and Bray (1981) because many deterministic
fore, wedge failures, whose volumes are relatively models used in previous research are not appropriate
small, are expected in this area. An interesting obser- for probabilistic analysis. Also, the proposed proce-
vation is that combinations with high probabilities of dures are simple and easy to apply in a probabilistic
failure involve foliation set 1. In addition, as the prob- approach. Another improvement provided in this
ability of kinematic instability and the probability of study is calculation of the probability of failure by
kinetic instability are evaluated separately in the prob- multiplying the probability of kinematic instability
abilistic approach for this study, we can decide which by the probability of kinetic instability. This is
parameters are more critical for slope stability. This is appropriate as probability of kinetic instability is a
250 H. Park, T.R. West / Engineering Geology 59 (2001) 233±251

conditional probability based on the premise that a Feng, P., Lajtai, E.Z., 1998. Probabilistic treatment of the sliding
wedge block is kinematically unstable. By contrast, wedge with EZSlide. Engineering Geology 50, 153±163.
Glynn, E.F., 1979. A Probabilistic approach to the stability of rock
previous probabilistic methods evaluating failure
slopes. PhD thesis, Massachusetts Institute of Technology,
probability did not consider kinematic aspects in a Cambridge, Massachusetts.
probabilistic manner. Finally, as the probability of Goodman, R.E., 1976. Methods of Geological Engineering in
failure is a conditional probability based on the Discontinuities Rocks, West, St. Paul, Minnesota.
premise that only one unstable wedge block occurs, Harr, M.E., 1987. Reliability-Based on Design in Civil Engineering.
McGraw-Hill, New York.
probability of failure for the entire slope must also
Hoek, E., 1997. Rock Engineering; Course notes by Evert Hoek
consider the discontinuity spacing, indicating the (Online), available on the Internet at http://www.rockeng.utor-
block sizes likely to occur. These new considerations onto.ca/Hoekcorner.htm.
provide improvements that can be easily applied using Hoek, E., Bray, J.W., 1981. Rock Slope Engineering. . 3rd ed.In-
the current probabilistic analysis. stitute of Mining and Metallurgy, London.
Jaeger, J.C., 1971. Friction of rocks and stability of rock slopes.
This new probabilistic approach was applied to a
Geotechnique 21, 97±134.
practical stability analysis of a rock cut slope in North Kovari, K., Fritz, P., 1984. Recent developments in the analysis and
Carolina, USA. Comparison of results for the deter- monitoring of rock slopes. Proceedings of Fourth International
ministic analysis and probabilistic analysis shows a Symposium on Landslides, Toronto, Canada.
signi®cant difference between them. Deterministic Kulatilake, P.H.S.W., Finley, R.E., Ghosh, A., 1985. Effect of varia-
analysis while based on a ®xed value for discontinuity bility of joint orientation and strength on factor of safety of
wedge stability. Proceedings of International Symposium on
and slope parameters failed to indicate the possibility Fundamentals of Rock Joints, Loen, Norway, pp. 25±33.
of slope failure. Consequently, the deterministic Kutter, H.K., 1974. Analytical methods for rock slope analysis. In:
analysis does not represent actual rock slope condi- Muller, L. (Ed.). Rock Mechanics. Springer, New York, pp.
tions but neglects the random nature of parameters 198±211.
and can provide simpli®ed and misleading results, Lajtai, E.Z., Carter, B.J., 1989. GEOSLIDE Ð A computer code on
the IBM PC for the analysis of rock slopes. Department of Civil
possibly causing serious problems. By contrast, the
and Geological Engineering, University of Manitoba, Winni-
probabilistic analysis is more representative of the peg, Canada.
actual behavior of parameters and provides results, Leung, C.F., Quek, S.T., 1995. Probabilistic stability analysis of
which most likely will more closely represent actual excavations in jointed rock. Canadian Geotechnical Journal
rock slope conditions. In all, the rock slope stability 32, 397±407.
affected by the discontinuity characteristics which are Li, K.S., White, W., 1987. Probabilistic approaches to slope design.
Research Report N. 20, Department of Civil Engineering,
widely scattered and variable, cannot be properly Australian Defensa Force Academy, Canberra, Australia, 54.
represented by single value input parameters and a Low, B.K., 1997. Reliability analysis of rock wedges. Journal of
single FS value. Therefore, it is recommended that Geotechnical and Geoenvironmental Engineering 123 (6),
the probabilistic analysis should be applied especially 498±505.
in cases where signi®cant scatter in the data of slope Low, B.K., Einstein, H.H., 1991. Simpli®ed reliability analysis for
wedge mechanisms in rock slopes. Proceedings of Sixth Inter-
parameters occur.
national Symposium on Landslides, Christchurch, New Zeal-
and, A.A. Balkema, pp. 499±507.
Mahatab, M.A., Yegulalp, T.M., 1982. A rejection criterion for
de®nition of clusters in orientation data. Proceedings of 22nd
References symposium on Rock Mechanics, Berkeley, American Institute
of Mining Metallurgical and Petroleum Engineers, New York,
Call, R.D., Nicholas, D.E., 1978. Prediction of step path failure pp. 116±123.
geometry for slope stability analysis. Proceedings of 19th US Marek, J.M., Savely, J.P., 1978. Probabilistic analysis of the plane
Symposium on Rock Mechanics, Stateline, Nevada. shear failure mode. Proceedings of 19th US Symposium on
Carter, B.J., Lajtai, E.Z., 1992. Rock slope stability and distributed Rock Mechanics. Stateline, Nevada, pp. 40±44.
joint systems. Canadian Geotechnical Engineering Journal 29, McMahon, B.K., 1971. Statistical methods for the design of rock
53±60. slopes. First Australian±New Zealand Conference on Geome-
Einstein, H.H., Baecher, G.B., 1982. Probabilistic and statistical chanics, pp. 314±321.
methods in engineering geology. Rock Mechanics 12, 47±61. Mostyn, G.R., Li, K.S., 1993. Probabilistic slope analysis Ð state
Feng, P., Lajtai, E.Z., 1997. Probabilistic treatment of the sliding of play. Proceedings of Conference on Probabilistic Methods in
wedge with EzSlide. Engineering Geology 50, 153±163. Geotechnical Engineering, Canberra, Australia, pp. 89±109.
H. Park, T.R. West / Engineering Geology 59 (2001) 233±251 251

Muralha, J., Trunk, U., 1993. Stability of rock blocks Ð Evaluation of rock excavations. International Journal of Rock Mechanics
of failure probabilities by the Monte Carlo and ®rst order relia- and Mining Sciences and Geomechanics Abstract 32 (6),
bility methods. International Symposium on Assessment and 617±620.
Prevention of Failure Phenomena in Rock Engineering, Istan- Sakurai, S., Shimizu, N., 1987. Assessment of rock slope stability
bul, Turkey, A.A. Balkema, pp. 759±765. by Fuzzy Set Theory. Proceeding of Sixth International
National Research Council, 1995. Probabilistic Methods in Congress on Rock Mechanics, Montreal, Canada, pp. 503±506.
Geotechnical Engineering. NRC. National Academy Press, Serrano, A.A., Castillo, E., 1974. A new concept about the stability
Washington, DC. of rock masses. Proceeding of Congress of Third International
Norrish, N.I., Wyllie, D.C., 1996. Rock slope stability analysis. In: Society for Rock Mechanics. Denver, Colorado, pp. 826±829.
Turner, A.K., Schuster, R.L. (Eds.), Landslides; Investigation Tabba, M.M., 1984. Deterministic versus risk analysis of slope
and Mitigation, Special Report 247, Transportation Research stability. Proceedings of Fourth International Symposium on
Board, National Science Council, pp. 391±424. Landslides. Toronto, Canada, pp. 491±498.
Park, H.J., 1999. Risk analysis of rock slope stability and stochastic Trunk, U., 1993. Probabilistic stability analysis of rock wedges.
properties of discontinuity parameters in western North Caro- Proceedings of Safety and Environmental Issues on Rock Engi-
lina. PhD thesis, Purdue University, West Lafayette, IN, USA. neering, Eurock'93, Lisbon, pp. 227±232.
Piteau, D.R., Peckover, F.L., 1978. Engineering of rock slope. In: Watts, C.F., West, T.R., 1985. Electronic notebook analysis of rock
Schuster, R.L., Krizek, R.J. (Eds.), Landslides: Analysis and slope stability of Cedar Bluff. Bulletin of the Association of
Control, Special Report 176, TRB, National Research Council, Engineering Geologists 22 (1), 67±85.
pp. 192±234. Wolff, T.F., 1996. Probabilistic slope stability in theory and prac-
Priest, S.D., 1993. Discontinuity Analysis for Rock Engineering. tice. ASCE Geotechnical Special Publication No. 58, Uncer-
Chapman & Hall, London. tainty in Geologic Environment: from Theory to Practice,
Quek, S.T., Leung, C.F., 1995. Reliability-based stability analysis ASCE Special Conference, ASCE, New York, NY.

You might also like