Sie sind auf Seite 1von 21

Rock Mech Rock Eng (2009) 42:729–749

DOI 10.1007/s00603-008-0011-3

ORIGINAL PAPER

Probabilistic Stability Evaluation of Oppstadhornet


Rock Slope, Norway

H. S. B. Duzgun Æ R. K. Bhasin

Received: 4 January 2007 / Accepted: 16 April 2008 / Published online: 5 July 2008
Ó Springer-Verlag 2008

Abstract Probabilistic analyses provide rational means to treat the uncertainties


associated with underlying parameters in a systematic manner. The stability of a 734-
m-high jointed rock slope in the west of Norway, the Oppstadhornet rock slope, is
investigated by using a probabilistic method. The first-order reliability method
(FORM) is used for probabilistic modeling of the plane failure problem in the rock
slope. The Barton–Bandis (BB) shear strength criterion is used for the limit state
equation. The statistical distributions of the BB criterion parameters, for which
comprehensive data were collected and statistically analyzed, are determined by using
distribution fitting algorithms. The sensitivity of the FORM model for the BB criterion
is also investigated. It is found that the model is most sensitive to the mean value of the
residual friction angle (/r) and least sensitive to the mean value of the slope angle (bf).
It is also found that the standard deviation of joint compressive strength (JCS) causes
the greatest difference in the reliability index, which has the least sensitivity to the
change in the mean and standard deviation of joint roughness coefficient (JRC).

Keywords Rock slope  Risk assessment  Reliability  Probabilistic slope stability

1 Introduction

Rock slope stability problems contain many uncertainties due to inadequate


information about site characteristics and inherent variability and measurement
errors in the geological and geotechnical parameters. Probabilistic modeling of the

H. S. B. Duzgun (&)
Mining Engineering Department, Middle East Technical University, 06531 Ankara, Turkey
e-mail: duzgun@metu.edu.tr

R. K. Bhasin
Norwegian Geotechnical Institute, P.O. Box 3930, Ullevaal Stadion, 0806 Oslo, Norway

123
730 H. S. B. Duzgun, R. K. Bhasin

rock slope stability problem allows the systematic treatment of these uncertainties.
In probabilistic slope stability analyses the uncertainties are taken into account
through the use of probability distributions, or the moments of the parameters.
Probabilistic rock slope stability analyses are also essential for quantitative risk
assessment. Risk is defined as the probability that a slope failure occurs within a
given period of time, multiplied by the consequences of the slope failure (IUGS
1997). Although there is still some reluctance among engineers to use probabilistic
methods, the need for probabilistic slope stability analysis for quantitative risk
assessment has led to their increased use in recent years.
One difficulty in probabilistic analyses is the lack of acceptable limits, i.e.,
probability of failure (Pf) or reliability index (b), so that the stability/instability
assessment can be made by the comparison of the computed reliability index of the
given slope and acceptable values. Such acceptable values are established for the
safety factor (SF) in deterministic analyses. Usually a SF of 1.3 for temporary and of
1.5 for permanent slopes are considered to be acceptable in engineering practice in
deterministic analyses (Hoek and Bray 1981; Hoek 1997; Pine and Roberds 2005).
However, deterministic analyses do not involve treatment of uncertainties, as only
characteristic values of the uncertain parameters are used. This may lead to conditions
where there are different safety margins for slopes with the same factor of safety
(Nadim et al. 2005). Hence, the outputs of deterministic analyses cannot be
incorporated into quantitative risk analyses. A discussion of the pros and cons of the
deterministic and probabilistic safety evaluation of slopes is given by Christian (2004).
The establishment of acceptable values for b and Pf requires probabilistic
stability evaluations of existing and failed slopes. For this purpose, the stability of
the 734-m-high Oppstadhornet rock slope in Norway is evaluated in this study using
the first-order reliability method (FORM). The Oppstadhornet rock slope, which has
been investigated by the Geological Survey of Norway (NGU), is located on the
west coast of Norway (Fig. 1) in the Møre and Romsdal County. The investigations
by NGU involved field studies to observe any recent movements in the slope and to

Fig. 1 Location of the Oppstadhornet rock slope on the West Coast of Norway (source: www.ngu.no)

123
Probabilistic Stability Evaluation 731

identify large-scale instability structures that could potentially collapse into the
fjord and result in generation of high waves or tsunamis (Blikra et al. 2001). NGU
conducted systematic mapping of rock slides and potential rock fall areas in the in
Møre and Romsdal County (west of Norway) to increase the knowledge and
awareness of the risk of such events. In this context, the Oppstadhornet rock slope is
reported to be one of the major potential rock slide areas by NGU (Blikra et al.
2001).
Rock slope stability problems are mostly governed by rock discontinuities, which
is the case in Oppstadhornet. For persistent rock discontinuities, the failure is
controlled by the shear strength of the discontinuities. It is widely acknowledged
that shear failure along rock discontinuities exhibits nonlinear behavior (Barton
1976). Hence limit equilibrium functions should be nonlinear. In this paper the
plane failure problem is formulated by considering the Barton–Bandis shear failure
criterion.

2 Oppstadhornet Rock Slope

The Oppstadhornet rock slope is composed of granitoid gneiss with zones of schist.
The rocks have well-developed foliations striking ENE-WSW and dipping
moderately to steeply to the south. The slope area is described by Robinson et al.
(1997) as an unstable mountainside with a height of about 734 m and width of
several kilometers, stretching from near shore to the peak of the slope. Figure 2
illustrates the location of rock blocks where movements have been observed on a
topographic map of the slope.
A considerable amount of large rock avalanches and bedrock failures were
noticed during NGU’s investigations on the southern slope of the mountain. A
picture of the steep mountainside towards the southwest slope is given in Fig. 3
and a cross-section of the slope is displayed in Fig. 4. According to the profile
shown in Fig. 4 the major sliding plane is following the foliation in the upper part
of the slope whereas it is cutting through the foliation in the lower part of the
slope. Some of these structural features could be observed from aerial photos, as
shown in Fig. 4, while others were interpreted from field observations. It is also
reported by Blikra et al. (2001) that large blocks of several tens of cubic meters
show internal fracturing and sliding both along the foliation joints and on the
cross-joints (Fig. 4).
The average slope angle between the top of the mountain and the road at the foot
of the slope is about 36°, but some sections of the slope are much steeper and some
have a shallower dip. In addition to joints along the foliation, there are also cross-
joints striking NW–SE (perpendicular to the plane of the cross-section shown in
Fig. 4). The rock mass is divided into blocks of various sizes that range from several
cubic meters to several tens of cubic meters. At some places block sizes of several
hundred cubic meters exist.
The discontinuities in the rock are mostly characterized as rough and undulating.
Many of the discontinuities are tightly healed with nonsoftening impermeable
filling. A number of the joint walls are slightly altered with only surface stains.

123
732 H. S. B. Duzgun, R. K. Bhasin

Fig. 2 Topographic map of the slope with the rock blocks where movements have been observed (Blikra
et al. 2001)

Some of the foliation joints are persistent in nature and could be followed for
hundreds of meters.
The volume of the entire potentially unstable mountain slope is estimated to be
10–20 million m3. This is based on the demarcated area shown in Fig. 2, in which
the thickness of the potentially unstable masses has been estimated to be 50–75 m in
the upper part and 25 m in the lower part of the slope. Three possible shear surfaces
(‘‘foliation surfaces’’ in Fig. 4) have been indicated in the profile by NGU. This

123
Probabilistic Stability Evaluation 733

Fig. 3 Photo of the slope (Bhasin and Kaynia 2004)

Fig. 4 Aerial photograph and cross-section of the slope with the foliation surfaces and potential blocks
prone to slide. The cross-section line (AB) is shown on the aerial photograph (Braathen et al. 2003)

profile is based on mapping carried out in the field and through aerial photographs
(Braathen et al. 2003; see Fig. 4).
The Geological Survey of Norway concludes that the age of the movements is
less than 11,500 years (i.e., after deglaciation) since the area shows no indication of
being affected by processes related to permafrost conditions. This indicates that

123
734 H. S. B. Duzgun, R. K. Bhasin

Fig. 5 Slope geometry and


potentially unstable blocks
shown in Fig. 4

sliding has taken place after the most unstable condition during deglaciation. There
are also some indications of recent movements, but no measurements of possible
movements have been carried out.
The identified large-scale instability has the potential to slide into the fjord and
result in generation of high waves or tsunamis (ICG 2004). Since a tsunami caused
by slope failure in Oppstadhornet can have an adverse effect on developed areas and
new development projects along the coastlines of the fjords, slope stability
evaluation is essential to assess the associated risk.
As can be seen from Figs. 2, 3, and 4, there are potentially three layers of rock
which can slide (Fig. 5). The first and the third layers in Fig. 5 are in the form of
single rock blocks (block 1 and block 3, respectively). The second layer is
composed of two blocks, namely block 2.1 and block 2.2 (Fig. 5). The slope
geometry given in Fig. 5 indicates a plane failure mode. Therefore, probabilistic
models are developed for the plane failure case.

3 Probabilistic Analysis of Stability

Stability analysis of a rock slope generally involves four steps: collection and
analyses of data, identification and analysis of failure mechanism, computation of
safety measures [factor of safety (SF) in deterministic analyses, reliability index
(b)/probability of failure (Pf) in probabilistic analyses], and evaluation of
computed safety indices. Data collection and analyses involve obtaining all the
relevant data for computation of resisting and driving forces on the slope.
Identification and analysis of failure mechanisms have two stages, namely
kinematic and kinetic. Kinematic analysis involves determination of potential rock
block instabilities in the rock mass for the considered study region based on the
discontinuity orientation. Usually kinematic analyses are followed by kinetic
analyses, where the potentially unstable rock blocks are investigated by limit
equilibrium or numerical analyses through computation of safety measures related
to each block.
In deterministic analyses, the safety measure is the SF defined by the ratio of the
sum of resisting forces to the sum of driving forces. In probabilistic analyses the
safety measure is either b or Pf. b is the minimum distance from the origin of
normalized basic variables to the limit state function (Fig. 6).

123
Probabilistic Stability Evaluation 735

Fig. 6 Graphical representation of the FORM approximation (Nadim et al. 2005)

A normalized basic variable has a mean of zero and a standard deviation of one.
Pf is obtained either by Monte Carlo simulation or reliability-based methods. In
Monte Carlo simulation, the SF is evaluated many times; each time a probable value
of the parameters in the SF is sampled from their probability distribution functions.
Then the probability distribution of SF is obtained, which allows one to compute the
probability that SF will be less than 1 or any other specified safety level, i.e., Pf. In
reliability-based methods the computed b is related to Pf under certain satisfied or
imposed conditions.
Finally the computed safety measures are evaluated by comparing them with
some acceptable safety levels.

3.1 Data Collection and Analysis

The data for probabilistic stability analysis of the Oppstadhornet slope were
obtained from previously performed studies on the slope, which are mainly
numerical stability analyses (Bhasin and Kaynia 2004; Bhasin et al. 2004; Dahle
2004). The collected data from field investigations involve geological mapping of
the slope, identification of joint roughness profiles to be used for the BB criterion,
Schmidt hammer testing, and characterization of the exposed joint walls in the
slope. The E modulus, Poisson’s ratio (m), and the uniaxial compressive strength (rc)
parameters were obtained by laboratory testing (Table 1). The joint input
parameters to be used in the Barton–Bandis shear failure criterion and the
mechanical properties of the intact rock used for the numerical modeling studies are
shown in Table 1.
Barton and Choubey (1977) have postulated that the reduction in JCS values due
to the scale effect roughly corresponds to the reduction of rc with increasing sample
size. Furthermore, they have concluded that there is a significant scale effect on
JRC. The larger the base length considered, the less steep the asperities, which
results in reduced JRC values. The scale correction for in situ block sizes (Ln) is
derived using the following scale correction equations (Barton and Bandis 1990):
 0:02JRC0
Ln
JRCn  JRC0 ð1Þ
L0

123
736 H. S. B. Duzgun, R. K. Bhasin

Table 1 Barton–Bandis joint


Parameters Estimated
parameters and intact rock
mean values
parameters
Joint roughness coefficient, JRC0 12
Joint compressive strength, JCS0 (MPa) 89
Laboratory scale length L0 (m) 0.1
In situ block size Ln (m) 1.0
Residual friction angle /r (°) 28
Uniaxial compressive strength rc (MPa) 100
Density (q) kN/m3 27.5
Poisson’s ratio (m) 0.25
Deformation modulus Ed (GPa) 40

 0:03JRC0
Ln
JCSn  JCS0 ð2Þ
L0
For the present case a JRC0 value of 10 when corrected to full scale (Ln) gives a
JRCn value of 6.3 (Eq. 1). Likewise a JCSo value of 100 MPa when corrected to full
scale gives a JCSn value of 50 MPa (Eq. 2).
In addition, probabilistic modeling of rock slopes requires information about the
probability distributions of the basic variables as well as the moments of the
distributions. There are two main types of basic variables: those related to strength
parameters and those related to geometrical parameters. In this study, the basic
variables related to the strength parameters involve joint roughness coefficient
(JRC), joint wall compressive strength (JCS), and residual friction angle (/r) for the
Barton–Bandis shear failure criterion. The previous numerical studies for the
Oppstadhornet slope basically contain data for the Barton–Bandis shear strength
criterion. In this paper the raw data collected for JRC, JCS, and /r (Barton and
Bandis 1990) are statistically analyzed and distribution fitting tests are carried out to
determine appropriate probability distributions. Distribution fitting to the parameters
of the Barton–Bandis shear failure criterion was carried out using Bestfit software
(version 4.5, Bestfit 2005) and the details of distribution fitting for the strength
related basic variables are given in Sect. 3.3. Table 2 lists the descriptive statistical
properties of the considered shear strength parameters.
Geometrical parameters are slope height (H), width (W), and angle (bf) as well as
the discontinuity dip (bs). The discontinuity dip and slope angle are considered to be

Table 2 Summary statistics for


Variable Min. Max. Mean ð
xÞ Standard Number of
Barton–Bandis and Coulomb
deviation (s) samples (n)
shear failure parameters
JRC0 6.00 20.00 11.93 4.18 29
JRCn 4.55 7.96 6.53 1.04 29
JCS0 (MPa) 16.00 190.00 88.87 50.30 91
JCSn (MPa) 7.02 83.34 38.86 22.05 91
/r (°) 22.00 34.00 28.00 2.27 26

123
Probabilistic Stability Evaluation 737

basic variables. Since the slope height and width have relatively less uncertainty
they are taken as deterministic variables.

3.2 Analysis of Failure Mechanism

The analysis of the failure mechanism has two basic stages, namely kinematic and
kinetics. Kinematic analysis involves determination of potential rock block
instabilities in the rock mass of the considered study region based on the
discontinuity orientation. Kinematic analyses can be performed either determinis-
tically or probabilistically.
In a typical deterministic approach, the orientation of a particular discontinuity
set is represented by a characteristic orientation value, which is usually the mean.
Then, kinematic tests are performed (e.g., using stereographic projections) in order
to determine potential instabilities as well as possible failure modes. This approach
is usually followed for investigating single block instability cases. For kinematic
analysis of multiple blocks, key block theory is widely implemented. Warburton
(1981) and Goodman and Shi (1985) developed the key block theory for movements
in translation, and Mauldon and Goodman (1996) adapted it for movements in
rotation. This theory has later been extended and applied by Mauldon and Ureta
(1996), Fulvio-Tonon (1998), and Sagaseta et al. (2001).
Generally, probabilistic approaches are derived from the stochastic treatment of
deterministic ones. Hence probabilistic kinematic instability approaches use the
same logic as deterministic kinematic methods; the only difference in probabilistic
methods is that they consider the underlying parameters in a stochastic manner.
There are two main types of approaches for assessing the probabilistic kinematic
instability in rock slopes. In the first type, which are also called lumped models, the
discontinuity properties such as spacing, trace length, and orientation are fitted to a
statistical or empirical distribution. Then Monte Carlo simulation is applied by
using the fitted distributions for discontinuity parameters, and kinematic tests are
performed to determine the probability of forming unstable blocks (e.g., McMahon
1971; Carter and Lajtai 1992). The basic shortcoming of the lumped models is that
discontinuity data obtained from different locations in the field are treated as if they
are the same throughout the study region. Hence lumped models do not consider the
spatial distribution and correlation of the discontinuity parameters (Nadim et al.
2005). To overcome this shortcoming, stochastic discontinuity models (e.g., Priest
and Hudson 1983; Einstein and Dershowitz 1996; Ivanova 1998) and geostatistical
methods (e.g., Carosso et al. 1987; Chiles 1988; Young 1993), which fall into the
second group of probabilistic kinematic analyses, are used.
The kinematic analysis is followed by kinetic (mechanical) stability analyses of
the slope. If the probabilistic approaches are used for the kinematic analyses, the
resultant probabilities are multiplied by the Pf values obtained from kinetic
analyses, since Pf is conditioned on the formation of unstable blocks (i.e., Pf given
that an unstable block forms, should be computed). Such conditioned approaches
are mainly used for rock slope design purposes as it is required to predict first the
probable formation of unstable rock blocks (e.g., Einstein et al. 1980; Feng and
Lajtai 1998) for a given slope geometry before combining this with the probability

123
738 H. S. B. Duzgun, R. K. Bhasin

of kinetic failure, Pf. For existing or natural slopes the slope geometry is usually
observed and measured in the field, and kinematic analysis is usually unnecessary
prior to kinetic analysis. In the Oppstadhornet rock slide case, as the geometry of
potential rock blocks were observed by NGU, kinematic analyses were not
performed and only the kinetic stability of the blocks is modeled based on
probabilistic approaches. Probabilistic computation of the kinetic instability of the
slope involves the formulation of the limit state equation, which is obtained by
equating the safety margin (subtraction of the driving forces from the resisting
forces) to zero, identification of parameters in the limit state equation to be treated
as random variables (basic variables), assessment of probability distributions with
their moments for each basic variable, and analysis of the uncertainties in the basic
variables by using statistical models for the systematic treatment of various sources
of uncertainties.
The computed value of the probability of slope failure, Pf is basically dependent
on the level of uncertainty in the basic variables, which is represented by the
standard deviation or coefficient of variation (c.o.v., a unit-less measure of
uncertainty, which is the ratio of the standard deviation to the mean value of the
basic variable). The higher the c.o.v. or standard deviation, the higher the Pf values.
Hence analyzing uncertainties associated with the basic variables in a systematic
way yields more realistic estimates of the slope’s safety. The uncertainties result
from insufficient information and inadequate knowledge about the properties of the
basic variables involved in rock slope stability, such as spatial and temporal
variation in rock properties, limited data collection and laboratory testing,
discrepancies between laboratory and in situ conditions, etc. Sources of uncertain-
ties affecting the rock slope stability parameters can be considered to be composed
of three components, namely, inherent variability, and statistical and systematic
uncertainties. The inherent variability is due to the fact that, even in a homogeneous
rock medium, the rock properties exhibit variability by nature. The limited sampling
and laboratory testing cause the statistics (i.e., mean, standard deviation) of a basic
variable to have uncertainty. This type of uncertainty is called statistical, because it
can decrease with increasing number of samples. The systematic uncertainties may
stem from the discrepancies between the laboratory and in situ conditions, due to
factors such as scale, anisotropy, and water saturation. Additional sampling may not
necessarily reduce this type of uncertainty, because the same test conditions are
likely to occur. Duzgun et al. (2002, 2003) presented a comprehensive statistical
model for quantification and analysis of the various uncertainties involved in
estimation of friction angle by aggregating inherent variability, and statistical and
systematic uncertainties within the first-order uncertainty analysis framework.
In the Oppstadhornet case, as there were not enough data to decompose the
uncertainties into inherent, systematic, and statistical uncertainty components, these
analyses cannot be performed. For a more detailed description of uncertainty analysis
on the basic variables see Ang and Tang (1984) and Duzgun et al. (2002, 2003).
The probable failure mode in the Oppstadhornet slope is plane failure (Figs. 2, 3,
4, and 5). Hence formulation of the limit state equation involves analysis of driving
and resisting forces for a plane failure case. The basic mechanism of plane failure is
best expressed by a sliding mass on an inclined plane. The mechanical principles

123
Probabilistic Stability Evaluation 739

Fig. 7 Forces on the rock block in the plane failure case (Priest 1993)

state that sliding occurs when resisting forces are smaller than the driving forces that
are parallel to the sliding plane. The mechanical principle of sliding is basically due
to gravitational loading on an inclined plane. Figure 7 illustrates the components of
typical forces acting on a rock block in the plane failure case.
In the analysis of the system of forces the stability of a unit slice of rock in Fig. 7,
is considered. It is convenient to analyze the forces G, U, and V in terms of their
components that lie parallel to the sliding plane, which form the driving forces, and
that are normal to the sliding plane, which contribute to the resisting forces. The
parallel and normal force components are listed in Table 3. Forces that tend to
activate sliding or compress the sliding plane are taken as positive. The details of
this formulation are given by Priest (1993).
The description of the forces listed in Table 3 is as follows:
G: weight of the sliding block
U: water force on the plane of sliding
V: water force in the tension crack
where;
bs: angle of the sliding plane
bc: angle of the tension crack

Table 3 Forces acting on the


Force Parallel component Normal component
sliding block shown in Fig. 7
G GP = G sin bs GN = G cos bs
U UP = 0 UN = -U
V VP ¼ V sinðbc  bs Þ VN ¼ V cosðbc  bs Þ

123
740 H. S. B. Duzgun, R. K. Bhasin

bf: angle of the slope


bt: angle of the top of the slope
In the plane failure case, the performance function, gðxÞ; can be defined as the
difference of the resisting forces, Rf and the driving forces, Df, as given in Eq. 3.
gðxÞ ¼ Rf  Df ; ð3Þ

where x is a vector of basic variables and


Df ¼ GP þ UP þ VP ; ð4Þ
where
GP: the component of the weight of the block (G) parallel to the sliding plane AD
in Fig. 7
UP: component of the water force (U) parallel to the sliding plane AD in Fig.7
VP : component of the water force (V) in the tension crack (CD) parallel to the
sliding plane AD in Fig. 7
LAD: length of the sliding plane AD in Fig. 7
GN: component of the weight of the block (G) normal to the sliding plane AD in
Fig. 7
UN: component of the water force (U) normal to the sliding plane AD in Fig. 7
VN: component of the water force (V) in the tension crack (CD) normal to the
sliding plane AD in Fig. 7
Rf for the Barton–Bandis criterion is defined by
Rf ¼ sp LAD ð5Þ
where
sp: peak shear strength of the foliation joint
The peak shear strength of foliation joint is computed based on the Barton–
Bandis nonlinear shear failure criterion as given in Eq. (6).
   
0 JCS
sp ¼ rn tan JRC log þ /r ð6Þ
rn
where
r0 n: effective normal stress (MPa)
JRC: joint roughness coefficient
JCS: joint wall compressive strength (MPa)
/r: residual friction angle

3.3 Computation of Safety Indices

The computation of safety indices requires the identification of basic variables


(parameters to be considered as random variables). For rock slopes, parameters can
be divided into geometrical parameters (e.g., slope height, width, slope angle,

123
Probabilistic Stability Evaluation 741

existence of joints and cracks and their geometry, discontinuity dip) and mechanical
parameters (e.g., strength, groundwater condition). If the discontinuities are through
going or persistent, mechanical parameters have relatively higher uncertainty than
geometrical parameters, since geometrical parameters can be directly measured
whereas mechanical parameters require prediction based on laboratory testing, in
situ investigations, and engineering judgment. However, if the discontinuities are
impersistent, the failure will be governed by both the intact rock failure between the
impersistent discontinuities and the failure along the discontinuities (Einstein et al.
1983), uncertainties in geometrical parameters are higher than those in mechanical
parameters.
In order to determine appropriate probability distribution functions for the basic
variables, data summarized in Table 2 are used for distribution fitting analyses, The
data collected by the Norwegian Geotechnical Institute (NGI) during field surveys
for the parameters of JRC, JCS, and /r, which are the parameters of the Barton–
Bandis shear strength criterion, were fed to distribution fitting procedure of Bestfit
software (version 4.5 2005). The results of distribution fitting analyses are given in
Table 4. As can be seen from Table 4, JRC fits best to triangular, beta, and uniform
distributions, while the best fitting distributions for JCS are beta, triangular,
uniform, Weibull, gamma, and log-normal. The best fitting distributions for the
angles of /r, bs, and bf are found to be triangular and uniform. Except for JCS, all of
the basic variables fit to bounded distributions. This is also consistent with the
nature of the parameters and the recent literature (Low 2007; Jimenez-Rodriguez
and Sitar 2007) as the values of JRC, /r, bs, and bf are defined for a range. JRC is
defined between 0 and 20. The theoretical value for /r is between 0° and 90°.
Although theoretically JCS can have values between zero and infinity, it takes on
values within a range for a given rock discontinuity. The triangular, beta, and
uniform distributions are the three best fitting distributions for JRC and JCS
(Table 4). The triangular and uniform distributions are found to be suitable for the
basic variable /r (Table 4).
In the literature Park and West (2001) assumed a normal distribution for friction
angle in probabilistic analysis of a rock slope. However, Park and West (2001) refer
to Hoek (1997) who suggested the use of a truncated normal distribution for friction
angle since using a normal distribution may give unreasonably low or high values.
Muralha and Trunk (1993) used the log-normal distribution in their probabilistic rock
slope stability analyses. Recently, Jimenez-Rodriguez and Sitar (2007) used the beta
distribution for friction angles in analyzing wedge stability, stating that the beta

Table 4 Fitted distribution for the basic variables


Basic variable Range Fitted distributions, decreasing degree of fit

JRC 4.6–8.0 Triangular, beta, uniform


JCS (MPa) 7–83 Beta, triangular, uniform, Weibull, gamma, log-normal
/r (°) 22–34 Triangular, uniform, beta
bs (°) 23–16 Uniform
bf (°) 35–21 Uniform

123
742 H. S. B. Duzgun, R. K. Bhasin

distribution prevents problems of unbounded distributions. Moreover, the uniform


distribution was used for the geometric variables (angles related to wedge geometry)
in the study of Jimenez-Rodriguez and Sitar (2007). Similarly, Low (2007) modeled
a rock slope stability problem in Hong Kong by using the beta distribution for friction
angle, cohesion, and horizontal distance of the tension crack behind the slope crest
and indicated that the beta distribution is advantageous over the normal distribution
due to its versatility and flexibility for bounded basic variables.
Consistent with the recent literature (Low 2007; Jimenez-Rodriguez and Sitar
2007), when the available data for the Oppstadhornet case are statistically analyzed
it was found that distributions which are defined for a range are suitable for the
strength parameters of the Barton–Bandis shear strength criterion. In fact this
finding is consistent with engineering practice as the strength of rock discontinuities
is usually defined for a range of values. In this paper, for ease of computation, the
triangular distribution is used for /r, JRC, and JCS.
For the parameters where the data are insufficient to fit a probability distribution,
assumptions are made. As the possible values for the parameters bs and bf are
known inside a range, the uniform distribution is found to be the most suitable.
The computation of Pf and b depends on parameters of the failure criterion (/r),
joint roughness coefficient (JRC), and joint compressive strength (JCS). The results
of the uncertainty analysis are used as input to the probabilistic slope stability
models. Then failure probability (Pf) for the considered rock slope as well as its
reliability index (b) can be evaluated based on Monte Carlo simulation (MCS) or the
first-order reliability method (FORM). In this study FORM is used for the
assessment of the probability of slope failure. In FORM, it is easy to perform
sensitivity analyses through the use of direction cosines of the basic variables. The
first step in FORM is the formulation of the performance function, g(X), where X is
the vector of basic variables, which are the stability analysis parameters defined as
random variables. The performance function (Eq. 3) is defined by the safety margin,
which is the difference between resisting forces, Rf (Eq. 5), and the driving forces,
Df (Eq. 4). Then Pf is defined by Eq. 7, if the joint density function of all basic
variables, Fx(X), is known (Nadim et al. 2005).
Z
Pf ¼ Fx ðXÞdx ð7Þ
gðXÞ\0

Since the analytical solution of the integral in Eq. 7 is generally impossible, Pf is


computed by an approximation. If the basic variables are not normally distributed
and/or correlated with other basic variables, the vector of the basic variables (X) is
transformed to the standard normal space U, where U is the vector of independent
Gaussian variables with mean and standard deviation of 0 and 1, respectively. The
function, g(U) becomes a linear function and takes the following form:
Xn
gðUÞ ¼ ai U i  b ð8Þ
i¼1

where ai is the direction cosine of the transformed basic variable Ui, b is the
distance between the origin and the hyperplane g(U) = 0, and n is the number of

123
Probabilistic Stability Evaluation 743

basic variables. The graphical representation of reliability index and the perfor-
mance function is given in Fig. 6. Then Pf is given by Eq. 9
" #
X
n
Pf ¼ P½gðUÞ\0 ¼ P ai Ui  b\0 ¼ 1  UðbÞ ð9Þ
i¼1

The basic steps of the FORM approximation algorithm are:


1. Transformation of the vector of basic variables into a standard Gaussian vector
2. Finding the most likely point or design point for the basic variables in the
failure domain
3. Estimation of Pf from Eq. (9)
As it is difficult to predict the design point, an initial guess is made for it, which is
usually the point formed by the mean values of the basic variables. The algorithm is
iterated until sufficient convergence is obtained. Because of the iterative nature of
the algorithm, the approximation may not converge, especially for highly nonlinear
performance functions. FORM also allows one to use correlated basic variables. In
this case, the algorithm involves an additional transformation of basic variables into
an uncorrelated space. Examples for the use of FORM models for rock slope
stability for various failure modes are given by Kimmance and Howe (1991), Trunk
(1993), Quek and Leung (1995), Low (1997), Duzgun et al. (2003), and Bafghi and
Verdel (2004).
For the Oppstadhornet rock slope, the probability of slope failure (Pf) and its
reliability index (b) are evaluated based on FORM for the Barton–Bandis shear
strength criterion, with uncorrelated basic variables. In FORM modeling of
Oppstadhornet rock slope, the basic variables are treated as independent and
distributions related to each basic variable are determined by fitting distribution
functions to the sampled data. The FORM analyses were carried out by using
Comrel software.
The probabilistic analyses for Oppstadhornet are performed by evaluating b for
the likelihood of various block failure scenarios. Then, the sensitivity of b to
changes in the basic variables in FORM is also investigated.
Since there is no information about the existence and measurement of water in
the tension crack and on the sliding plane, the effect of water force on the stability
cannot be investigated. However, any decrease in frictional properties of the
discontinuities, which can be caused by formation of gauge material during
movements, existence of water, weathering, etc., is investigated by calculating Pf
and b for the mean values of /r of 28°, 25°, and 20°.
The potentially unstable blocks in Fig. 5 can fail in various ways:
– Scenario 1 (AI). All the blocks (blocks 1, 2.1, 2.2, and 3) fail at the same time as
a single block, (1 + 2.1 + 2.2 + 3). Failure is on discontinuity plane 3.
– Scenario II (AII). Only block 1 fails and the rest are stable (1). Failure is on
discontinuity plane 1.
– Scenario III (AIII). Blocks 1, 2.1, and 2.2 fail at the same time as a single block,
block 3 remains stable (1, 2.1, 2.2). Failure is on discontinuity plane 2.

123
744 H. S. B. Duzgun, R. K. Bhasin

– Scenario IV (AIV). First block 1 fails, then blocks 2.1 and 2.2 fail at the same
time (1, 2.1 + 2.2). First failure is on discontinuity plane 1 and second failure is
on discontinuity plane 2.
– Scenario V (AV). First block 1, then sequentially block 2.1, block 2.2, and finally
block 3 fail (1, 2.1, 2.2, 3). The first, second, and third failures are on
discontinuity planes 1, 2, and 3, respectively.
Since the slope geometry in Fig. 5 for each block on the slope is different,
computations were carried out by using the geometrical properties of each block
illustrated in Fig. 7 (Table 5). The same distribution functions as discussed in the
previous paragraphs are used for the evaluation of each scenario. Moreover, the
distributions and moments of the basic variables (Table 2) related to the strength
parameters are kept the same in the evaluation of each scenario.
Among the five possible slope instability scenarios, Pf for AI–AIII is the direct
calculation of the slope failure probability. The scenarios AIV and AV on the other
hand, require the formulation of Pf based on the conditional probabilities, since the
occurrence of these scenarios are conditioned on first the failure of block 1, then the
failures of blocks 2.1 and 2.2 (scenario IV), and finally the failure of block 3
(scenario AV). Hence, Pf for scenarios AIV and AV are given in Eqs. (10) and (11),
respectively;
Pf ðIVÞ ¼ P½block 1 fails  P½block 2:1 and block 2:2 fail togetherjblock 1 failed
ð10Þ
Pf ðIVÞ ¼ P½block 1 fails  P½block 2:1 failsjblock 1 failed
 P½block 2:2 failsjblock 1 and block 2:1 failed
 P½block 3 failsjblock 1; block 2:1 and block 2:2 failed ð11Þ
In Table 6, the calculated Pf for the three mean values of /r values for the four
scenarios are given. The highest Pf value is obtained for AII, which involves the
failure of only block 1. AI has the second highest Pf value, indicating the failure of
all the potentially unstable blocks. The Pf value for scenario V is the lowest.
Therefore, it is recommended that in the Oppstadhornet rock slope the potential
failure of block 1 and the failure of the slope as a whole should be considered for the
evaluation of landslide risk. As can be seen from Table 6, there is approximately
one order of magnitude difference between the Pf values computed for /r of 28°,
25°, and 20°, respectively. Hence this suggests that a potential decrease in the
frictional properties of the discontinuities should be carefully investigated.

Table 5 Geometrical
Involved Geometrical parameters
parameters of rock blocks
blocks
Height Width Slope angle Discontinuity
H (m) W (m) bf (°) dip bs (°)

1 320 171 35 27
2.1 15 13 45 28
2.2 329 140 26 22
3 240 217 21 16

123
Probabilistic Stability Evaluation 745

Table 6 Pf values for the five scenarios


Block/scenario Individual block Pf Scenario Pf

/r = 28° /r = 25° /r = 20° /r = 28° /r = 25° /r = 20°

1 5 9 10-3 3 9 10-2 2 9 10-1


-4 -3
2.1 4 9 10 2 9 10 3 9 10-2
-5 -4
2.2 9 9 10 8 9 10 2 9 10-1
-5 -4
2.1 + 2.2 6 9 10 7 9 10 2 9 10-2
-7 -6
3 3 9 10 4 9 10 9 9 10-4
I 1 9 10-3 7 9 10-3 1 9 10-1
-3 -2
II 5 9 10 3 9 10 2 9 10-1
-5 -4
III 7 9 10 8 9 10 2 9 10-2
IV 3 9 10-7 2 9 10-5 4 9 10-3
V 5 9 10-17 2 9 10-13 1 9 10-7

FORM analyses allow one to investigate the sensitivity of the reliability index b
to the moments (mean and variance/standard deviation) of the basic variables using
the direction cosines. Figures 8 and 9 show the sensitivity of b to changes in mean
and standard deviation of the basic variables, respectively. When the effects of
mean values of the basic variables are examined, it can be seen that /r has the
strongest influence on b (Fig. 8). The second most influential basic variable on b is
the dip of the discontinuity (sliding) plane (bs), while b is least affected by the
change in the mean value of slope angle (bf) (Fig. 8). In FORM formulated for the
Barton–Bandis shear failure criterion, a change in the standard deviation of JCS
causes the greatest difference in b (Fig. 9). Figure 9 also illustrates that b has the
second greatest sensitivity to the changes in the standard deviations of /r and bf.

Fig. 8 Sensitivity of the reliability index to changes in the mean value of the basic variables

123
746 H. S. B. Duzgun, R. K. Bhasin

Fig. 9 Sensitivity of the reliability index to changes in the standard deviations of the basic variables

This indicates that a comprehensive uncertainty analysis (quantification and


analyses of inherent variability, statistical, and systematic uncertainties) of JCS
and /r would give better estimates of b The sensitivity analyses point out that b has
the least sensitivity to the change in the mean and standard deviation of JRC
(Figs. 8, 9).

3.4 Evaluation of Safety Indices

Probabilistic stability analyses of the Oppstadhornet rock slope indicates that failure
is most probable either on the first discontinuity plane or the third discontinuity
plane (Fig. 5; Table 6).
If the first block slides, which is scenario II, the estimated volume of rock to slide
in the fjord ranges between 5.4 and 11 million m3 with an average value of
8.2 million m3 (Bhasin and Kaynia 2004). When the whole slope on discontinuity
plane 3 slides (scenario I), the volume of the failed rock is predicted to be between
17.4 and 53 million m3 with an average value of 35 million m3.
The assessment of the slope safety requires comparison of computed Pf and b
values with ‘‘acceptable’’ values. Although there is limited work on acceptable
values of Pf for rock slopes, the studies of Hoek (1991), Chowdhury and Flentje
(2003), Duzgun et al. (2003), and Park et al. (2005) suggest an acceptable level of Pf
within the range of 0.10 and 0.15. Hence the upper most block (block I) in Fig. 5 for
/r less than 25° can be considered as unsafe. It is clear that for /r of 28°, the slope
seems stable as Pf values are small (0.001 for AI and 0.005 for AII).
The computed failure probabilities should be interpreted as a measure of hazard
and not as a measure of risk unless they are associated with the consequences of the
failure. The computed Pf value of a slope provides much more information about
the effects of the uncertainties involved in the calculation than the factor of safety
from a deterministic analysis. To use the calculated Pf value in risk analyses, it is
required to associate this Pf with frequency-based probabilities (i.e., Pf/year). This
can be done either by considering the time-dependent behavior of the basic

123
Probabilistic Stability Evaluation 747

variables in the probabilistic models or associating the computed Pf value with the
frequency of failures in the region (Nadim et al. 2003). As such data are not
available for the Oppstadhornet case the annual probability cannot be computed.
However, the approach proposed in this study can form the basis of quantitative
relative hazard assessment and a guide for decision makers investing in detailed
works.

4 Conclusions

Probabilistic analyses treat rationally and explicitly the uncertainties in all the
analysis parameters. The case study of Oppstadhornet illustrates that probabilistic
methods can easily be used for rock slope stability evaluations in engineering
practice.
From the sensitivity analyses performed for the Oppstadhornet rock slope, it is
found that the prediction of strength parameters plays a critical role in the safety
evaluation. Hence, more effort should be spent on the data collection for strength
parameters. In addition, comprehensive uncertainty analyses of the strength
parameters could provide more-realistic estimates and increase the reliability of
the probability estimates.
The calibration study for determining acceptable Pf and b values for rock slopes
is needed for improved stability assessments. Although there are some recommen-
dations for acceptable Pf in the literature, they are inconsistent as the conditions for
which they are computed are not indicated. As the evaluated values of failure
probabilities and reliability indices in this study apply to a real rock slope, they
could serve as a basis for future calibration studies.

Acknowledgments This paper was prepared during the first author’s postdoctoral fellowship at the
International Centre for Geohazards (ICG) at the Norwegian Geotechnical Institute (NGI) in Oslo. The
authors would like to thank their colleagues from ICG, NGI, and NGU for their valuable discussions and
contributions. Professor Kaare Høeg and Dr. Suzanne Laccase are thanked for their critical review of the
manuscript. The views expressed in this paper are those of the authors and not necessarily of the above-
mentioned organizations.

References

Ang HS, Tang WH (1984) Probability concepts in engineering planning and design. Wiley, New York
Bafghi ARY, Verdel T (2004) The probabilistic key-group method. Int J Numer Anal Methods Geomech
28:899–917
Barton N (1976) The Shear strength of rock and rock joints. Int J Rock Mech Min Sci 13:255–279
Barton N, Bandis S (1990) Review of predictive capabilities of JRC-JCS Model in engineering practice.
In: Proceedings of international symposium on rock joints. Loen, Norway, pp 603–610
Barton N, Choubey V (1977) The Shear strength of rock joints in theory and practice. Rock mechanics,
vol 1/2. Springer, Vienna, pp 1–54
BestFit (2005) Probability distribution fitting for windows, Paliside corporation, New York, 14867
Bhasin R, Kaynia AM (2004) Static and dynamic simulation of a 700-m high rock slope in western
Norway. Eng Geol 71:213–226
Bhasin R, Kaynia A, Blikra LH, Braathen A (2004) Insights into the deformation mechanisms of a jointed
rock slope subjected to dynamic loading. Int J Rock Mech Min Sci Geomech Abstr 41:470–471

123
748 H. S. B. Duzgun, R. K. Bhasin

Blikra LH, Braaten A, Skurtveit E (2001) Hazard evaluation of rock avalanches; the Baraldsnes—Oterøya
area, Geological Survey of Norway (NGU), Report no: 2001.108, 33pp
Braathen A, Blikra LH, Berg SS, Karlsen F (2003) Rock-slope failures of Norway; type, geometry and
hazard. Nor J Geol 84:67–88
Carter BJ, Lajtai EZ (1992) Rock slope stability and distributed joint systems. Can Geotech J 29:53–60
Carosso G, DelGreco O, Giani GP (1987) Some probabilistic approaches to stability analysis of open pit
exploitations. In: Proceedings of international symposium on engineering geology. Beijing, China,
pp 881–891
Chiles JP (1988) Fractal and geostatistical methods for modeling a fracture network. Math Geol 20:631–
654
Chowdhury R, Flentje P (2003) Role of slope reliability analysis in landslide risk management. Bull Eng
Geol Environ 62:41–46
Christian JT (2004) Geotechnical engineering reliability: how well do we know what we are doing? J
Geotech Geoenviron Eng 130:985–1003
Dahle H (2004) Slope stability analysis for Oppstadhornet (in Norwegian), Master of Science Thesis,
Norwegian University of Science and Technology, Trondheim, Norway
Duzgun HSB, Yucemen MS, Karpuz C (2002) A probabilistic model for the assessment of uncertainties
in the shear strength of rock discontinuities. Int J Rock Mech Min Sci Geomech Abstr 39:773–754
Duzgun HSB, Yucemen MS, Karpuz C (2003) A methodology for reliability-based design of rock slopes.
Rock Mech Rock Eng 36(2):95–120
Einstein HH, Baecher GB, Veneziano D (1980) Risk analysis for rock slopes in open pit mines. Final
Technical Report, Massachusetts Institute of Technology, Departmet of Civil Engineering,
Cambridge, USA, Publication No. R80-17
Einstein HH, Veneziano D, Baecher GB, O’Reilly KJ (1983) The effect of discontinuity persistence on
rock slope stability. Int J Rock Mech Mining Sci 20:227–236
Einstein HH, Dershowitz WS (1996) Characterizing rock joint geometry with joint system models. Rock
Mech Rock Eng 21:21–51
Feng P, Lajtai EZ (1998) Probabilistic treatment of the sliding wedge with EzSlide. Eng Geol 50:153–163
Fulvio-Tonon PE (1998) Generalization of Mouldon’s and Goodman’s vector analysis of key block
rotations. J Geotech Geoenviron Eng 124:913–922
Goodman RE, Shi GH (1985) Block theory and its application to rock engineering, Prentice Hall,
Englewood Cliffs
Hoek E (1991) When is a design engineering acceptable? In: Proceedings of seventh ISRM congress on
rock mechanics, vol 3, pp 1435–1497
Hoek E, Bray J (1981) Rock slope engineering, 3rd edn. Institute of Mining & Metall
Hoek E (1997) Rock engineering: course notes by evert hoek (online).
http://www.rockeng.utoronto.ca/Hoekcorner.htm
International Center For Geohazards (ICG) (2004) Rock slope failures-models and risks. Modelling the
Oppstadhornet slope with Plaxis ICG Report No: 20031093-2, 73pp
International Union of Geological Sciences (IUGS) (1997) Working group on landslides, Committee on
Risk Assessment. Quantitative risk assessment for slopes and landslides—the state of the art. In:
Cruden DM, Fell R (eds) Landslide risk assessment. Balkema, Rotterdam, pp 3–12
Ivanova VM (1998) Geologic stochastic modeling of fracture systems in rocks, Phd Dissertation.
Massachusetts Institute of Technology, Cambridge, USA
Jimenez-Rodriguez R, Sitar N (2007) Rock wedge stability analysis using system reliability methods.
Rock Mech Rock Eng 40(4):419–427
Kimmance JP, Howe JH (1991) A combined geostatistical and first order second moment reliability
analysis of slopes in kaolinsed granite. In: Proceedings of seventh ISRM congress on rock
mechanics, pp 905–911
Low J (1997) Reliability analysis of rock wedges. J Geotech Geoenviron Eng (ASCE) 123:498–505
Low BK (2007) Efficient probabilistic algorithm illustrated for a rock slope, Rock Mech Rock Eng
(online)
McMahon BK (1971) A statistical method for the design of rock slopes. In: Proceedings of first Australia–
New Zealand Conf. on Geomechanics: 314–321
Mauldon M, Goodman RE (1996) Vector analysis of key-block rotations. J Geotech Eng (ASCE)
122:976–987
Mauldon M, Ureta J (1996) Stability analysis of rock wedges with multiple sliding surfaces. Int J Rock
Mech Min Sci Geomech Abstr 33:51–66

123
Probabilistic Stability Evaluation 749

Muralha J, Trunk U (1993) Stability of rock slopes-evaluation of failure probabilities by the Monte Carlo
and first order reliability methods. In: Proceedings of assessment and prevention of failure
phenomena in rock engineering, pp 759–765
Nadim F, Krunic D, Jeanjean P (2003) Probabilistic slope stability analyses of the Sigsbee escarpment. In:
Proceedings of offshore technology conference, OTC Paper No.: 15203
Nadim F, Einstein H, Roberds W (2005) Probabilistic stability analysis for individual slopes in soil and
rock. In: Proceedings of international conference on landslide risk management, pp 63–98
Park H, West TR (2001) Development of a probabilistic approach for wedge failure. Eng Geol 59:233–
251
Park H-J, West TR, Woo I (2005) Probabilistic analysis of rock slope stability and random properties of
discontinuity parameters, Interstate Highway 40, Western North Carolina, USA. Eng Geol 79:230–
250
Pine RJ, Roberds WJ (2005) A Risk-based approach for the design of rock slopes subject to multiple
failure modes-illustrated by a case study in Hong Kong. Int J Rock Mech Min Sci Geomech Abstr
42:261–275
Priest SD (1993) Discontinuity analysis for rock engineering. Chapman & Hall, London
Priest SD, Hudson J (1983) Discontinuity frequency in rock masses. Int J Rock Mech Min Sci Geomech
Abstr 20:73–89
Quek ST, Leung CF (1995) Reliability-based stability analysis of rock excavations. Int J Rock Mech Min
Sci Geomech Abstr 32:617–620
Robinson P, Tveten E, Blikra LH (1997) A post-glacial bedrock failure at Oppstadhornet, Oteroya, More
og Romsdal: a potential major rock avalanche. NGU Bull 433:46–47
Sagaseta C, Sanchez JM, Canizal JA (2001) General analytical solution for the required anchor force in
rock slopes with toppling failure. Int J Rock Mech Min Sci Geomech Abstr 38:379–397
Trunk U (1993) Probabilistic stability analysis of rock wedges. In: Proceeding of safety and
environmental issues on rock engineering, Eurock’93 Symp, pp 227–231
Warburton PM (1981) Vector stability analysis of an arbitrary polyhedral rock block with any number of
free faces. Int J Rock Mech Min Sci Geomech Abstr 18:415–427
Young DS (1993) Probabilistic slope analysis for structural failure. Int J Rock Mech Min Sci Geomech
Abstr 30:1623–1629

123

Das könnte Ihnen auch gefallen