Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

International Journal of Mining Science and Technology 29 (2019) 357–370

Contents lists available at ScienceDirect

International Journal of Mining Science and Technology


journal homepage: www.elsevier.com/locate/ijmst

Probabilistic assessment of rock slopes stability using the response


surface approach – A case study
Hassan Basahel ⇑, Hani Mitri
Department of Mining and Material Engineering, McGill University, Montreal H3A 0E8, Canada

a r t i c l e i n f o a b s t r a c t

Article history: This paper examines the stability condition of a jointed rock slope in the south western region of Saudi
Received 16 August 2018 Arabia using deterministic and probabilistic approaches, under both dry and wet conditions. The study
Received in revised form 16 October 2018 area is characterized by complex geology in rugged terrains. The stability analysis is carried out using
Accepted 12 November 2018
the code FLAC3D to generate a 3-dimensional, ubiquitous joint model, to determine the influence of
Available online 22 November 2018
the dominant, unfavourable discontinuity orientation with respect to the slope face. The deterministic
analysis is first implemented using the mean values of the selected random variables, namely the dip,
Keywords:
dip direction and friction angle of the dominant discontinuity set, and the stability condition is assessed
Rock slope stability
Box-Behnken design
with a factor of safety based on the classical frictional joint constitutive model. A Box-Behnken design
Probabilistic analysis of slopes (BBD) approach is then adopted to create the surface response function as a second order polynomial
Response surface for the factor of safety. To do so, fifteen FLAC3D models are generated in accordance with the BBD.
Based on this, 10,000 simulations of different slope realizations are carried out using Monte-Carlo simu-
lation technique, and the probability of unsatisfactory of performance of the rock slope is assessed. It is
shown that the probabilistic approach provides more insight and confidence in the stability condition of
the rock slope, both under dry and steady state heavy rainfall conditions. A discussion is presented on the
significance of accepting lower safety factors when heavy rainfall conditions are encountered.
Ó 2018 Published by Elsevier B.V. on behalf of China University of Mining & Technology. This is an open
access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction masses onto the stability conditions of rock slopes [14]. This type
of studies relies on determining a range of stability factors (factor
Discontinuities are a considerable source of uncertainty and of safety) to describe the stability behaviour of a rock slope by
variability in engineering design of rock slopes. Rock slopes are incorporating different combinations of controlling factors, such
often found in complex and rugged terrains such as mountainous as cohesion and friction angle to be used as input parameters
areas, where strength and geometric properties of the discontinu- (independent variables), with numerical modelling, and then iden-
ities are subjected to change within a rock slope [1–8]. tifying the most influential factors that play a significant role on
Due to these variations in both the strength and geometric the output or factor of safety. This type of method is known as
properties of discontinuities in rock slopes, the field measurements the probabilistic approach. It is more efficient than the determinis-
of their geometric and mechanical characteristic are subject to tic approach, which is based on the analysis of the mean value of
variation in terms of accuracy and in the error ratio [5,9–11]. As the input variables to derive a single value of the output or factor
a result, rock slope design studies involve a lot of uncertainties. of safety.
The origin of these uncertainties in the geotechnical data and anal- Numerous probabilistic methods have been proposed in the
ysis can be from several sources, for instance, the methods used in field of geotechnical engineering specifically in slope stability stud-
the data acquisition, the analysis techniques, the judgment that ies. Most of these methods have been developed to address the
depends on the experience of the engineer, the methodology used influence of the inherent variability of strength properties for slope
in the study and other sources, all of which could influence the out- stability assessment in a realistic manner [8,15–18].
put results [12,13]. Some studies have also discussed the influence of geometric
In the past few decades, many studies have considered the parameters, particularly, the characteristics of joint networks,
effect of the inherent variability of geotechnical properties in rock which have been successfully incorporated in the stability assess-
ment of rock slopes [19]. In addition, the impact of different joint
surface properties on the overall stability condition of rock slopes
⇑ Corresponding author.
have been captured and analyzed [20].
E-mail address: hassan.basahel@mail.mcgill.ca (H. Basahel).

https://doi.org/10.1016/j.ijmst.2018.11.002
2095-2686/Ó 2018 Published by Elsevier B.V. on behalf of China University of Mining & Technology.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
358 H. Basahel, H. Mitri / International Journal of Mining Science and Technology 29 (2019) 357–370

Hammah et al. [5] discussed the application of probabilistic of the variability of the three independent variables on the result-
approach with numerical modelling. Two probabilistic approaches ing output variable i.e. factor of safety. The BBD is a three-level
have been adopted in the analysis, the point estimate method design used to establish an equation to predict the factor of safety
(PEM) and Monte-Carlo technique, to evaluate the probability of in the form of a second order polynomial. This approach is appro-
failure of a rock slope. This analysis takes into considerations the priate for the case that involves three factors and usually gives
influence of joint network on different stability conditions. The accurate results like the CCD but with less computational time
study concluded that the PEM gives good results for probability [32,33].
assessment with reasonable computational cost, but the method
suffers from curse of dimensionality (increased number of vari-
ables). On the other hand, the Monte-Carlo method is a powerful 2. Location and geological sitting of the study area
tool and more flexible to incorporate a wide range of probability
distributions without being affected by the curse of The study area is in the south-western part of Saudi Arabia. It is
dimensionality. situated between latitude 17°120 0000 and 17°150 0000 N and between
Brideau et al. [21] investigated the natural variability of the geo- longitude 43°020 0000 and 43°040 0000 E (Fig. 1). The selected case
metric parameters in joint surfaces such as persistence, spacing study is a road cut located on Road 12, which is one of the main
and orientation on the stability conditions of a rock slope. The and vital roads linking the cities and villages in the region.
authors used a 3D distinct element code for stability assessment. The study region is characterized by mountainous and difficult
The results show that the joint surfaces, with unfavorable persis- terrains. As a result, roads in this region were constructed across
tence (fully extended) and both fixed spacing and orientation val- very rugged terrain, where, most of the time, it resulted in steep rock
ues, do not always show a conservative solution. slopes with sharp slope angles and discontinuities that are almost
A response surface approach was also presented as an efficient parallel to the slope face. These excavated road cuts experience seri-
tool to identify the likelihood of the instability behaviour of a rock ous instability problems, especially, during and after rainstorms.
slope [22]. This approach has been applied using the central compos- Consequently, yearly rock slopes failures cause closure and damage
ite design (CCD), which is one of the statistical design methods used to these roads. Clearly, these situations threaten public safety of road
to implement experiments to examine the effects of the main inter- users and result in injuries and even deaths at times.
actions of different levels of the independent variables on the result- Geologically, the selected rock slope is in Fayfa Mountain (Jabal
ing response (dependent variable). As a result, an equation for the Fayfa) in Jazan region, which is covered by fractured syenite of Pre-
response, i.e. factor of safety, is established as a function of the cambrian age; Jabal Fayfa is part of the Arabian shield that contains
design variables (independent variables) of the response surface. Proterozoic layers rocks. Proterozoic layers rocks are originally
The resulting mathematical model or the response equation is used metamorphosed to greenschist facies and locally metamorphosed
to estimate the probability of the unsatisfactory performance in the to amphibolite or granulite facies [34]. These are known as Sabya
rock slope [23]. This technique is efficient because it can be used to formation (Fig. 2). The Sabya formation houses the oldest rocks;
generate a wide range of output values (dependent variables). i.e., they have been subjected to many tectonic movements that
factor of safety along with Monte-Carlo technique, with low compu- resulted in deformation, metamorphism and alteration processes
tational effort. This is particularly useful when the input variables [35]. It consists of many types of rocks such as, greenschist,
include geometric parameters, such as, the orientation of disconti- quartz-biotite schist, quartz-sericite schist, amphibole plagioclase
nuities, where the technique allows for the change of the inputs hornfels, hornfels, phyllite, black slate quartzite and marble [34].
without the need to re-construct the model itself many times. Structurally, many faults have been detected and observed in
In general, the stability condition in the rock slopes with joint the study area to tend north-west (NW) and west (W).
network is significantly controlled by rock structures. This kind
of failure mechanism is known as structurally controlled failure. 3. Methodology
These rock structures are mainly represented by discontinuities,
joints, bedding planes, faults, etc. The geometric aspects, such as The study involves several phases of analysis. It begins by defin-
dip and dip direction, and the strength properties, such as friction ing the suitable case study and identifying the parameters that
angle and cohesion of the discontinuous surfaces, are the critical control the instability condition. These parameters are the dip,
parameters that play the important role in the instability of rock the dip direction and the friction angle of the discontinuity surface.
slopes on road construction projects [24–30]. First, we compute the three statistical moments (mean, stan-
In this paper, the stability condition of an excavated rock slope dard deviation and skewness) for each parameter from which we
in mountainous areas using the response surface methodology is calculate the point estimates. Then, we use these estimate points
assessed. The analysis of a rock slope model with a single joint with the mean value for each parameter to create realizations of
set as a representative of the dominant joint is adopted. Three
parameters, dip, dip direction and friction angle for the rock joint
surface are considered probabilistically, the extent of the joint is
assumed to be infinite (persistence = 1), and the cohesion is
assumed to be zero with no infilling materials. The study area is
exposed to very highly intense seasonal rainstorms during certain
time periods of the year and most of the rock failures occur in rain
times [31]. Thus, we evaluate the probability of unsatisfactory of
performance for the rock slope in the case of dry condition and
then consider the effect of pore water pressure. We analyze both
conditions of the rock slopes and compare their stability beha-
viours to identify the critical situations. Consequently, the reme-
dial measures and means of mitigations are suggested based on
the estimate of the critical factor of safety calculated in different
weather conditions in the study region. The Box-Behnken design
(BBD) is selected as a response surface design to study the effect Fig. 1. Location map.
H. Basahel, H. Mitri / International Journal of Mining Science and Technology 29 (2019) 357–370 359

Fig. 3. Case study rock slope.

ables. The selected parameters are considered probabilistically in


the analysis. Other data such as rock mass properties and slope
Fig. 2. Geological map of the study area.
dimensions have been kept with their deterministic values.
the road cut by means of Box-Behnken design. Subsequently, we The friction angle is estimated according to the relation of Bar-
compute the shear strength reduction factor (SRF) value for each ton and Choubey [37], as follows:
realization using FLAC3D code. We identify the contribution and  
^
significance of each parameter via the Analysis of Variance £r ¼ £b  20A þ 20ðr=RÞ ð1Þ
(ANOVA). Then, we generate an adequate prediction model for esti-
mating the response (SRF). Finally, Monte-Carlo sampling tech- where £r is the residual friction angle, £b is the basic friction angle,
nique is used to run 10,000 slope realizations and the probability r is the rebound on weathered rock surface and R is the rebound on
of unsatisfactory of performance for the rock slope is estimated un-weathered rock surface. The values of r and R are estimated
for both dry and wet rock slopes. using the Schmidt Hammar (L-type) by measuring the hardness of
the rock.
4. Field measurements Forty readings of dip and dip direction of different joint sets sur-
faces were measured manually using Brunton compass. These
The selected case study, as mentioned earlier, is a rock slope readings have been plotted using the stereographic projection
located on Road 12 in Jazan region, south-west of Saudi Arabia. method using a software DIPS v. 6.0 for kinematic analysis. The
Basically, the rock slope consists of Amphibolites rock with moder- analysis has been done in a lower hemisphere projection. All poles
ate state of weathering. The rock slope contains three main joint and great circles that represent the different orientations of discon-
sets and the potential failure mechanism is structurally controlled. tinuities sets are shown in Fig. 4.
A detailed field investigation at the selected site has included the As shown in Fig. 4, fourteen values of dip and dip direction that
following. belong to set #1 are clustered within the critical zone of planar fail-
ure region (red area). Hence, these critical values of dip, dip direc-
(1) Determining the geometry of the rock slope, slope height, tion and friction angle, estimated from Eq. (1), are considered as
width and inclination angle of the slope face. random variables (independent variables or inputs) in the mod-
(2) Measuring the dips and dip directions of 40 joints. elling and the probabilistic analysis. Table 1 shows the data for
(3) Collecting the required measurements of rock mass parame- joint set #1.
ters for the rock mass classification (RMR system).
(4) Collecting all measurements needed for determining the
6. Point estimate method
strength properties for different joint set surfaces, such as,
joint roughness coefficient and Joint wall compressive
The point estimate method (PEM), proposed by Rosenbluth
strength.
[38], is one of the most popular statistical methods used in the field
(5) Collecting rock samples for laboratory testing to examine
of geotechnical engineering problems because it is simple, accurate
the strength of intact rock using the point load test.
[39,40] and it takes the numerical analysis uncertainty into
account, [41–44]. For a given set of data and for each input param-
The rock slope extends 50 m wide, approximately 20 m in
eter, the best fit distribution curve is determined. Then, the first
height and the slope face tilts at an angle of 60° with an azimuth
two statistical moments, the mean value (l) and the amount of
angle of 040° (north-east). The roadway is 10 m wide and is located
dispersion before and after the mean (standard deviation, r), are
directly below the rock slope with no ditch at the toe of the slope
extracted from this distribution. Sometimes the third moment
(Fig. 3). Moreover, the slope face is left without any support.
(skewness, s), which describes the shape of the distribution is con-
The rock mass of the selected rock slope is evaluated and cate-
sidered, if found. The resulting statistical moments for a known
gorized according to RMR system developed by Bieniawski [36].
distribution are then used in a specified mathematical function
The RMR contains five parameters known as a basic RMR; Intact
to calculate the point estimates (High ‘‘+” and low ‘‘”) for each
rock strength (UCS), rock quality designation (RQD), discontinu-
parameter. Then, the point estimates are used in conjunction with
ities spacing, conditions of discontinuities and groundwater.
another technique (i.e. Box-Behnken Design) to establish a second
order polynomial expression. Such polynomial is used to calculate
5. Random variables the response function (Y). In this study, Y represents the factor of
safety (FOS) or, alternatively, the shear reduction factor (SRF) and
Three parameters, namely dip, dip direction and friction angle, considered as the dependant variable or the output in the
of the most vulnerable surface, have been chosen as random vari- modelling.
360 H. Basahel, H. Mitri / International Journal of Mining Science and Technology 29 (2019) 357–370

Fig. 4. Stereographic projection of structural data for discontinuity sets of the case study.

Table 1 variables. In this method, the effects of each variable and their
Data of dip, dip direction and friction angle for joint set #1. interactions on the resulting response are determined. As a result,
No. Dip (°) Dip direction (°) Friction angle (°) a mathematical function to estimate the response function (Y) is
1 51 040 28
derived. For example, BBD is used for the estimation of the factor
2 52 041 23 of safety (FOS) for a wide range of independent variables and to
3 50 045 26 estimate the probability of unsatisfactory performance for the
4 49 046 26 response [5,46]. Also, this kind of statistical experiments are used
5 49 045 28
usually for the purposes of the optimization of the response char-
6 50 043 21
7 54 042 31 acterization to get an optimal design [33].
8 50 044 31 BBD is a rotatable design that involves three-level of design
9 51 043 25 points 3k, where k is the number of random variables. Unlike the
10 55 040 28 two-level design, 2k, that produces a linear response function and
11 52 043 30
12 53 046 26
cannot capture the nonlinear effect of input variables on the
13 49 045 25 resulted response Y, BBD is considered a highly efficient design
14 57 044 29 for estimating the second-order response surface that is able to
capture the curvature in the system for a given independent vari-
In this study, the PEM is applied in two scenarios based on the able. The design is shown geometrically in Fig. 5 [33]. A typical
statistical moments of the random variables xi , wherei is the vari- Box-Behnken design table for three-random variables is shown in
able index; the scenarios are described as follows. Table 2. The number of runs is thirteen, which covers all variable
(a) If the response function (Y) is a function of a random variable level treatment combinations. Besides, the center point is treated
xi (the input) and the three statistical moments (mean ‘‘l”, stan- three times. This could be advantageous because the design is
dard variation ‘‘r” and skewness ‘‘s”) of xi are known, the point esti-
mates, that represent xi , are calculated using the following
equation [45].
" rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi#
sxi s 2
¼ lxi þ rxi
xi
xi  1þ ð2Þ
2 2

(b) If the distribution of xi is symmetrical and approximately


Gaussian, the point estimates are calculated using the following
equation [38]:

xi ¼ lxi  rxi ð3Þ

7. Box-Behnken design (BBD)

BBD is one of the statistical experiments used for the analysis of


the problems that are influenced by several independent random Fig. 5. Geometric representation of a three-variable Box-Behnken design.
H. Basahel, H. Mitri / International Journal of Mining Science and Technology 29 (2019) 357–370 361

Table 2 with uncertainty. It generates a random number for each of the


A typical Box-Behnken design for three random variables. random variables in the problem and it makes combinations
Run X1 X2 X3 amongst all these random variables to perform several determinis-
1 1 1 0 tic computations [51]. The accuracy of the MCS depends on the
2 1 1 0 number of simulations performed, and it increases with the
3 1 1 0 increase of the number of simulations.
4 1 1 0 The MCS consists of sampling a set of materials from their joint
5 1 0 1
6 1 0 1
probability distributions function (PDF) and introducing them into
7 1 0 1 the model. As a result, a set of values represent the response (i.e.
8 1 0 1 factor of safety) is generated from MCS analysis and the informa-
9 0 1 1 tion on the distribution and moments of the response variable
10 0 1 1
are then obtained [52].
11 0 1 1
12 0 1 1 There are six essential elements that form the Monte-Carlo sim-
13 0 0 0 ulation (MCS) technique, which are described as follows [50].
14 0 0 0
15 0 0 0  Defining the problem in terms of all random variables.
 Quantifying the probabilistic characteristics of all the random
variables and the corresponding parameters.
rotatable by nature, and the repetition of the center point makes  Generating the values of these random variables.
the variance of predicted response constant throughout the region  Evaluating the problem deterministically for each set of realiza-
of interest and provides a good prediction [47]. tion of all the random variables.
BBD can be summarized as follows. A set of input variables is  Determining probabilistic information from such realizations.
given into the design frame, each variable is defined by three levels  Determining the accuracy and efficiency of the simulation.
of weighting values, (+1) represents the maximum level of the vari-
able, (1) represents the minimum level and (0) represents the The MCS has some advantages such as, flexibility in incorporat-
mean level [48]. The number of experiments, that are required to ing a wide variety of probability distributions without much
develop a BBD, are calculated by the following expression. approximation, and the ability to readily model correlations among
variables. In contrast, as the accuracy of MCS increases by increas-
N ¼ 2kðk  1Þ þ C 0 ð4Þ
ing the number of the simulations this can be disadvantageous
where k is the number of random variables and C0 is the number of because it becomes computationally expensive [50]. However, in
center points. this study, the MCS is used only to generate values for the random
As mentioned above, in the simplest case, which is the two- variables, while, the response has been found using BBD and the
level design (2k) with each variable treated in two levels of design response surface approach. This greatly reduces the computational
points (represented by +1 and 1, high level and low level, respec- cost and allows to increase the number of observations and hence
tively), the response is approximated to be linear over the range of the accuracy of the results.
variable levels chosen. This induces a first-order response surface
that is defined using the following relation:
9. Shear strength reduction (SSR)
Y ¼ b0 þ b1 x1 þ b2 x2 þ    þ bk xk þ  ð5Þ

where Y is the response, bij is the regression coefficient, xk is the Factor of safety (FOS) considers the failure criterion most com-
monly used to assist the stability behaviour of rock slopes. It can be
independent variable and €Il is a random error. On the other hand,
determined for any slope parameter of interest, such as water level,
the three-level design 3k, that is like Box-Behnken design, allows
slope height and strength property, by taking the ratio of the
to model the possible curvature in the response function and to
parameter’s actual value under a condition to its value under the
handle the case of random variables at three levels. The third level
critical condition. The critical condition is identified as the initia-
in this design facilitates the investigation of a quadratic relationship
tion of a hazardous state such as a slope failure.
between the response and each of the random variable [49], this
The acceptable FOS value depends on the nature of the problem
relationship is described as following:
being investigated. Service life, importance and complexity of the
Y ¼ b0 þ b1 x1 þ b2 x2 þ b12 x1 x2 þ b11 x21 þ b22 x22 þ  ð6Þ project are some of the factors that dictate the choice of an appro-
priate FOS. For example, acceptable FOS in civil engineering pro-
The input variables are treated randomly in the design, where jects are usually higher than those of mining engineering, and
the main and interaction effects between different levels of all this is because the anticipated service life of civil engineering
input variables and their influences on the resulting response (Y) works is usually longer than mining ones. For example, an inter-
are examined and identified. At the end, a polynomial function city roadway is likely to last over 100 years, whereas an open pit
for the response (Y) is obtained. This function contains multiple mine lasts generally between 5 and 50 years. The present study
terms of the random variables to estimate the interested response, is of a rock slope found in complex terrains and is subjected to
which in turn, used for the objective of prediction purpose of the heavy seasonal rainstorm. A slope failure along an inter-city road-
probability of the risk of failure. way like Road 12 of the case study, would create much concern to
traffic safety. In light of these points, a threshold FOS of 1.5 was
8. Monte-Carlo simulation technique (MCS) adopted, below which the stability of the rock slope would be con-
sidered ‘‘unsatisfactory”.
The Monte-Carlo simulation technique is considered as a very FOS can be calculated numerically using a variety of methods;
powerful tool for the engineer who has only a basic working the most popular methods are limit equilibrium (LEM), limit state
knowledge of probability and statistics for evaluating the risk or and shear strength reduction (SSR). In this paper, the shear
reliability of complex engineering systems [50]. Monte-Carlo strength reduction (SSR) is used because it is commonly used in
method is used in a wide range of engineering projects that deal conjunction with numerical modelling in the field of geomechanics
362 H. Basahel, H. Mitri / International Journal of Mining Science and Technology 29 (2019) 357–370

particularly in slope stability problems [53–56]. SSR calculates the H 0 ¼ s1 ¼ s2 ¼    ¼ sa ¼ 0 ð10Þ


SRF value, which represent the FOS value in the analysis by pro-
gressively reducing the shear strength of the materials until the H1 ¼ si –0 for at least one i ð11Þ
failure state is found.
Eq. (10) represents a null hypothesis that assumes the treat-
The stability condition of the case study is represented by the
ment effect or the interaction with the other parameter at all levels
structurally controlled failure mechanism. Therefore, the geomet-
is equal to zero, which means that this parameter does not con-
ric and strength properties for the potential failure surface should
tribute to the variability of the response. These hypotheses are
be considered. Based on that, the ubiquitous – joint material is
examined for all the random variables (parameters) included in
selected in this study, where the strength values for intact mate-
the design.
rial, C and Ø, and for the ubiquitous joint C j and £j are used in
Eq. (11) represents an alternative hypothesis, which assumes at
the analysis. The equation to calculate the reduction factor for
least one treatment effect at a certain level or interaction is not
the ubiquitous –joint material will be as follows:
equal to zero. This means that this parameter or its interaction
1 has a significant effect on the variability of the response; in this
C trial
j ¼ Cj ð7Þ case, the null hypothesis is not true and should be rejected.
F trial
The percentage contribution for all random variables on the
  obtained response can also be calculated. To do so, it is necessary
1
£trial
j ¼ arctan trial tan£j ð8Þ to compute the values of the sum of squares ðSSÞ of each random
F
variable and their interactions as follows:
where F is the factor of safety, C j is the cohesion for the joint surface
ContrastABCK ¼ ða  1Þðb  1Þ þ    ðk  1Þ ð12Þ
and £j is the friction angle for the joint surface.
In the ubiquitous joint model, additional factors can be con-
1
trolled, by the user, besides the strength parameters, (C j ; £j Þ, these Effect estimatesABCK ¼ ð13Þ
n2k1
are the geometric parameters of the potential failure surface, the
dip and dip direction. The change in these geometric parameters
1
could play a significant role on the results of the SRF values, espe- SSABCK ¼ contrastABCK ð14Þ
cially, if the analysis implemented in three-dimension numerical n2k
model code (i.e. FLAC3D). The percentage contribution for each random variable is calcu-
lated using the following equation:
10. Analysis of variance (ANOVA) SSABCK
Contribution percentage ¼  100 ð15Þ
SStotal
When all numerical modelling simulations for rock slope real-
izations needed for BBD are completed and their corresponding where ABC    K are the parameters, k is the number of parameters,
responses (SRF) are obtained, the treatment combinations for all n is the number of replications.
levels of random variables along with their responses are per-
formed to establish the functional relationship between the inde- 11. Model setup
pendent variables and the response. This has led to the
approximate proper response surface, which is used for predicting An initial three-dimensional numerical model has been con-
the probability of failure, as stated in Section 7. structed using FLAC3D code. This model represents the rock mass
Through the BBD analysis, the main effect of all random vari- model without an excavation, the height is 120 m from the datum
ables, their interactions and their contributions on the variability in z direction, the width is 50 m in x direction and it extends deeply
of the obtained response are determined using a statistical tech- 120 m in y direction, as shown in Fig. 6.
nique called the Analysis of variance (ANOVA). This technique The state of in-situ stress (r0 ) is assumed as follows: the verti-
detects the differences between the levels of the random variables cal stress (r0v ) in the z direction is given by the overburden pres-
and measures the amount of deviation of the response, resulted sure as described in the relationship:
from each treatment combination in the analysis, from the overall
mean of the responses. In other words, the Analysis of Variance is r0v ¼ cz ð16Þ
used in this study to identify the considerable (or most influential)
where c is the unit weight of the rock mass and z is the depth below
random variables that impact the response (i.e. SRF).
the ground surface. The horizontal stress, (r0x ; r0y Þ; for the numerical
Each treatment combination between different levels of param-
eters in BBD is called an observation. Any statistical experiment model is also assumed to be a function in cz. It increases below the
performed in the BBD involves several observations and this set ground surface towards the model datum, with gradient of c. Since
of observations are known as a model, and can be expressed using the rock material is homogenous, the horizontal stress in x and y is
the equation below: assumed to be equal:
 r0y ¼ r0x ¼ cz ð17Þ
i ¼ 1; 2;    ::; a
yij ¼ l þ si þ ij ð9Þ
j ¼ 1; 2;    ::; b The boundary conditions for all the sides of the model are
restricted horizontally and allowed to move only along the z direc-
th
where yij is the ij observation, a and b are the treatments, l is the tion. The model base can move horizontally only, and the ground
overall mean, si is the effect of the i parameter level and ij is a
th
surface is kept free without any restrictions. The model is exca-
random error; this model is also known as the effect model [33]. vated from the top left side to construct a rock slope with specific
The ANOVA technique works on testing the effect of each treat- dimensions, as show in Fig. 7. The slope is at an angle of 60°, with a
ment combination of parameter levels on the resulted response by slope height of 20 m and road width of 20 m.
means of the application of statistical hypotheses. Two hypotheses As can be seen in Fig. 7, the model size has been extended 3
are examined to test the treatment effect si . The Eqs. (10) and (11) times the slope height below the level of the roadway in the neg-
show the two hypotheses tests for one factor: ative direction of z axis, and 3 times the roadway width behind the
H. Basahel, H. Mitri / International Journal of Mining Science and Technology 29 (2019) 357–370 363

tests are performed by running the simulation and changing the


geometric ratios of the successive zone sizes in y direction. The
default ratio is 1 for all zones, but this ratio is changed to 1.1 from
the slope face to the back of the slope, where the zones sizes are
decreased around the slope and are increased towards the back
of the slope by this ratio (1.1). Then, the ratio is changed to 1.15
for the second run and 1.2 for the third and so on. The best mesh
was found at grid ratio of 1.4, where the displacement and the
SRF values appear to converge (Fig. 8).
The numerical analysis is done in both dry and saturated condi-
tions. For the groundwater analysis, a steady state fluid-flow is
adopted as the state of the groundwater within the numerical
model. Two properties related to the fluid-flow are important to
conduct the analysis; the first property is the hydraulic conductiv-
ity of the joint kh , which can be calculated using the following
Fig. 6. Initial model of the rock slope before excavation. expression [57].

ge2
kh ¼ ð18Þ
12v
where kh is the hydraulic conductivity of the joint in m/s, g is the
gravitational acceleration (9.81 m/s2), e is the joint aperture (m)
and v is coefficient of kinematic viscosity of the fluid (m2/s).
The second property is the permeability coefficient k which can
be obtained from:

kh
k¼ ð19Þ
qf g
where k is the permeability coefficient (m2/(Pa/s)), kh is the hydrau-
lic conductivity (m/s), qf is the fluid mass density and g is the grav-
itational acceleration. In case of model with single joint set, an
anisotropic flow is assumed, where the three permeability values
are specified as k1 , k2 , and k3 , and the flow orientation is assumed
to follow the joint attitude (dip/dip dir.). The directions k1 and k2
Fig. 7. The rock slope model after excavation.
are defined at the same plane of the joint surface, however, k3 is
perpendicular to the joint plane and assumed to be zero. Table 4
shows the necessary properties for the steady state fluid-flow anal-
excavated slope in the positive direction of y axis to avoid the ysis in FLAC3D numerical analysis.
effect of boundary conditions on the numerical analysis results. The groundwater surface extends from the slope face towards
The rock mass properties and the properties of the potential the slope surface; this extension has been chosen to be 2 times
failure surface that are used in the analysis are presented in Table 3. the slope height behind the toe of the slope [30], as shown in
The mechanical properties of the rock mass are estimated from the Fig. 9. The pore water pressure is applied on the upper-stream
Hoek Brown classification investigation for the rock slope site. The and the down-stream of the slope model with pressure gradient
results of this investigation have shown that the rock strength is based on the concept of cw z, where cw is the unit weight of water
133 MPa as obtained from the point load test, the density of the (Fig. 10).
rock sample is 2.9 g/cm3, the rock slope surface condition is
described as very blocky and the surface quality is fair, the geolog-
ical strength index (GSI) for the rock mass is estimated to be 55,
and the mi constant estimation, that is related to the amphibolite
rock, is 26. In addition, the disturbance factor has been taken as
1 for poor blasting and the effect of slope height has also been con-
sidered in the calculation of the rock mass properties, where the
height used is 20 m.
A sensitivity analysis for mesh quality of the model has been
conducted. Some test runs have been performed to refine the grid
size of the mesh and obtain an optimum grid for the analysis. These Fig. 8. Mesh size corresponding to the conversions of displacement.

Table 3
Rock mass and joint surface properties used in the numerical analysis.

Rock mass properties


Rock density (kg/m3) Bulk modulus (MPa) Shear modulus (MPa) Friction angle (°) Cohesion (kPa) Tensile strength (kPa)
2900 2.852 1.316 45 332 37
Rock joint surface properties
Dip (°) Dip dir. (°) Friction angle (°) Cohesion (Pa) Tensile strength (Pa) Dilation angle (°)
52 043 27 0 0 0
364 H. Basahel, H. Mitri / International Journal of Mining Science and Technology 29 (2019) 357–370

Table 4
Groundwater properties for steady state fluid-flow analysis in FLAC3D analysis.

Permeability coefficient (k1 = k2) Flow dip (°) Flow dip dir. (°) Porosity Water density (kg/m3) Water bulk modulus Water tension limit
9
1.2  10 52 043 0.3 1000 1000 0

first three statistical moments for each distribution (mean ‘l’, stan-
dard deviation ‘r’ and skewness ‘s’) have been identified. Figs. 11–
13 illustrate the best fitted distribution for each set of random vari-
ables with the corresponding statistical moments; Fig. 11 shows
the histogram and the exponential fit distribution for dip values,
Fig. 12 shows that the best fitted distribution curve is a triangle
and Fig. 13 shows the Normal distribution fit for the given values
of the friction angles. All statistical moments that have been deter-
mined from the fitted distributions are shown in Table 5.
After that, the point estimates for each set of variables are cal-
culated using Eq. (2) for the dip and dip direction variables, and Eq.
(3) for the friction angle variable. The choice of the appropriate
equation to calculate the point estimates is according to the exis-
tence of the third moment (the skewness) in the fitted distribution.
Fig. 9. Ground water flow models [30]. Table 6 shows the point estimates calculated based on the statisti-

Fig. 11. Histogram and exponential fit distribution for dip values.
Fig. 10. Sectional view shows the pore water pressure distribution within the rock
slope.

12. Results and discussion

12.1. Kinematic analysis

The results of the kinematic analysis of the rock discontinuities


using DIPs software indicate that a cluster of fourteen poles are
concentrated in an area that is almost parallel to the strike of the
slope face. This in turn allows a potential of plane failure to occur
(the red region), the mean attitude (dip/dip direction) of this clus-
ter has been determined as set #1 and its value is 52/040. Hence,
the joint set #1 has been selected as the critical joint, because a Fig. 12. Histogram and triangle fit distribution for dip direction values.
potential of plane failure is most likely to occur along this joint sur-
face (Fig. 4).
As shown earlier in Section 1, most of the parameters that con-
tribute to the rock slope failures are related to the rock discontinu-
ities. Thus, set #1 has been selected for further analysis and its
parameters represented by dip, dip direction and friction angle
have been chosen as random variables (see Table 1) and as a major
source of uncertainties in the stability analysis.

12.2. Point estimates

According to the order shown in the methodology, after defin-


ing the set of all random variables, the next step is finding the loca-
tion of the sampling points within the experimental design of BBD
by calculating the point estimates. To do so, the best fit distribution
curve has been assigned to each group of random variables and the Fig. 13. Histogram and normal fit distribution for friction angle values.
H. Basahel, H. Mitri / International Journal of Mining Science and Technology 29 (2019) 357–370 365

Table 5
The best fitted distributions and their corresponding statistical moments for each set of random variables.

Variable Best fit distribution Mean (l) Standard deviation (r) Skenwess (s)
Dip Exponential 52 2.6 2
Dip direction Triangle 43 1.8 0.6
Friction angle Normal 27 2.9 0

Table 6
The calculated point estimates for each set of variables.

Variable Mean Point estimate (xþ


i
) Point estimate (x
i )

Dip 52 57 47
Dip direction 43 46 40
Friction angle 27 30 24

cal moments of the fitted distribution for each set of variables.


These calculated point estimates are used as the sampling points
for each random variable with known statistical distribution for
the further analysis in BBD to properly estimate the statistical Fig. 14. Result of displacement in the reference model before an excavation.
moments for the response (factor of safety) and construct the
response surface that approximately describes the relationship
between the random variables (inputs) with the response (output). the saturated condition and the corresponding factor of safety
(FOS) are recorded as 1.64 and 1.31, respectively (Figs. 15 and
16). This in turn reflects the role of groundwater pressure in
12.3. Numerical modelling
increasing the amount of displacement, which led to a decrease
in the safety factor to below the threshold safety limit 1.5.
A three-dimensional numerical model using FLAC3D software is
For dry analysis, the highest value of the factor of safety is
constructed and the material properties for the rock and the single
recorded in realization #4 and it is 2.95 (Fig. 17), as shown in
joint set (set #1) are assigned to the model with the mean values;
Table 7. While, the lowest value of the factor of safety was found
this is the reference model in this study. A sensitivity analysis for
in realization #5 and it is 0.91 (Fig. 18). From these results, it is
the model grid is performed to identify the optimal mesh which
obvious that the factor of safety value is highly affected by the
is adopted for the analysis. Several runs have been conducted
dip angle of the joint; FOS values drop down by decreasing the
and the mesh size of the grid was changed in each run based on
angle of dip and vice versa, as noticed in Table 7.
the procedure mentioned in Section 11 to refine the mesh until
the values of the maximum Displacement and SRF are converged
(convergence < 10%). At the end, the suitable mesh size was found
to be 1.4 in Y direction and the SRF value to be 1.64, these values
are selected for the analysis (see Fig. 8).
Thirteen numerical realizations for rock slopes have been simu-
lated. These different realizations represent the variability of the
different levels of the input variables involved in the BBD experi-
mental frame. The SRF values have been computed as dependant
variables (responses) for each run in both the dry and the saturated
conditions and are presented in Table 7.
The reference model shows that the amount of displacement in
the initial model before executing the excavation is approximately
zero in the entire model as expected (Fig. 14). When the model is
excavated, the maximum displacement is computed; it is found Fig. 15. Shows displacement contours and simulated factor of safety for the
that the displacement is 15 cm in dry condition and is 21.1 cm in reference model in dry condition.

Table 7
Shows results of the factor of safety of dry and saturated conditions for all treatments of slope realizations.

Slope realizations Patterns Dip (°) Dip direction (°) Friction angle (°) SRF (dry) SRF (saturated)
1 0 47 40 27 1.30 1.12
2 +0 47 46 27 1.37 1.08
3 +0 57 40 27 2.64 2.20
4 ++0 57 46 27 2.95 2.49
5 0 47 43 24 0.91 0.71
6 0+ 47 43 30 1.31 1.14
7 +0 57 43 24 2.78 2.32
8 +0+ 57 43 30 2.85 2.35
9 0 52 40 24 1.57 1.30
10 0+ 52 40 30 1.65 1.44
11 0+ 52 46 24 1.78 1.38
12 0++ 52 46 30 1.87 1.48
13 000 52 43 27 1.64 1.31
366 H. Basahel, H. Mitri / International Journal of Mining Science and Technology 29 (2019) 357–370

angle is equal to the mean value and both the dip and the dip direc-
tion values are in the low levels, the FOS value increased to 1.3. But,
when the dip is at mean level and the other variables (the friction
angle and dip direction) values are in the low levels, the computed
factor of safety increased to 1.57. This confirms that the dip vari-
able plays an active role on the computed FOS and that the friction
angle slightly influences the resulted factor of safety, as seen in
treatment #1.
The numerical modelling results for the analysis of the influ-
ence of the groundwater pressure on the stability conditions, for
all slope realization treatments, have revealed that all values of
Fig. 16. Shows displacement contours and simulated factor of safety for the
reference model in saturated condition.
the factor of safety in general have decreased as compared to their
counterpart values in the dry state. The reference model shows a
clear drop in the value of FOS from 1.64 to 1.31 (20% drop) and
the stability condition transformed from a satisfactory condition
(>1.5) in the dry state to a critical condition (<1.5) in the saturated
state.
Furthermore, in slope realization #11, as another example, the
factor of safety has dropped by 22.5%, where the FOS changed from
1.78 in the dry state to 1.38 in the saturated state, as shown in
Figs. 19 and 20. Here the dip angle is equal to the mean value
(52°), whereas, the friction angle value of the joint plane is in the
low level (24°). Also, in slope realization #12, although the friction
angle and the dip direction values are in the high level, the factor of
safety decreased by 21% and the FOS value changed from 1.87 to
1.48. This reduction in the FOS happened because the dip angle
Fig. 17. Factor of safety for slope realization # 4 in dry condition. is equal to the mean (52°), (Figs. 21 and 22).
As seen in Table 7, the results of all numerical simulations for
the groundwater analysis give FOS values below the limit of safety,
1.5, in the range (0.71–1.48), except the cases with the dip variable
located in the high level (+), where the FOS values remain above 2,
despite the presence of the water pressure.
Therefore, this gives an indication that the dip angle of the joint
is very influential on the variability of the factor of safety values,
and that the high value of the dip angle could enhance the stability

Fig. 18. Factor of safety for slope realization #5 in dry condition.

Also, from Table 7, we note that the factor of safety takes high
values (>2.5) when the value of joint dip is in the high level (+),
as presented in slope realizations #3, 4, 7 and 8. Whereas, the fac-
tor of safety drops down to below (<1.4), which is below the
threshold safety limit, when the joint dip value is in the low level
(), as shown in realizations 1, 2, 5, and 6. Moreover, the factor of
safety values remain above the safety limit (>1.5), when the dip
Fig. 19. Factor of safety for slope realization #11 in dry condition.
angle of the joint has values that are equal to the mean values
(0), as shown in realizations from 9 to 13 in Table 7, regardless
of the levels of the combinations of the two other factors.
However, we find that when the dip direction is in the high level
(+) and the friction angle is equal to the mean value (0), the factor
of safety is below 1.5, as seen in realization #2, where the FOS
value is 1.37. In addition, when the friction angle is in the high
level (+) and the dip direction of the joint is equal to the mean
value (0), the amount of FOS remains below the safety limit 1.5,
as shown in realization #6 (FOS = 1.31).
The results of the factor of safety in the low-level treatment
combinations are presented in realizations 1, 5 and 9. The FOS
value is below 1 in treatment #5, where the dip and the friction
angle values are in the low levels, while the dip direction is kept
at the mean value. However, in treatment #1, when the friction Fig. 20. Factor of safety for slope realization #11 in saturated condition.
H. Basahel, H. Mitri / International Journal of Mining Science and Technology 29 (2019) 357–370 367

of the rock slope. For instance, the analysis shows that the high Table 8
level of the dip angle is 57° and it is almost located at the border Results of ANOVA for the rock slope in dry condition.

of the daylight envelop towards the stereonet center (see Fig. 4), Source of variability Sum of squares P-value Contribution (%)
this is in turn reduces the likelihood of the failure and increases Dip 5.0086125 <0.0001 88.5
the stability. Dip direction 0.0820125 0.0256 1.45
Friction angle 0.0512000 0.0557 0.90
Dip* dip direction 0.0144000 0.2453 0.25
12.4. Box-Behnken design Dip* friction angle 0.0272250 0.1302 0.481
Dip direction* friction angle 0.0000250 0.9584 0.0004
In order to predict the probability of unsatisfactory perfor- Dip2 0.4143692 0.0009 7.32
mance for the rock slope of the case study, a response function Dip direction2 0.0299077 0.1164 0.53
Friction angle2 0.0005769 0.8027 0.01
has been established through the design of experiments approach
using the Box-Behnken design. As explained earlier in Section 7,
the BBD is used to establish a quadratic relation between the
inputs (random variables) and the output (response variable). Table 9
To establish this relation, the regression coefficient for the lin- Results of ANOVA for the rock slope in saturated condition.
ear, the quadratic and the interactions should be established to
Source of variability Sum of squares P-value Contribution
identify the contribution of each random variable. To do so, the (%)
Analysis of Variance has been used and the contribution of each
Dip 3.5245125 <0.0001 86.1
random variable has been identified. Dip direction 0.0171125 0.1043 0.42
Figs. 21 and 22 show the plots of the simulated values of SRF Friction angle 0.0612500 0.0133 1.50
against the predicted SRF values for both the dry and the saturated Dip* dip direction 0.0272250 0.0545 0.67
conditions, respectively. Form these plots, we see that all the SRF Dip* friction angle 0.0400000 0.0291 0.98
Dip direction* friction 0.0004000 0.7740 0.0098
predicted points fall within the confidence curve (red-dotted lines)
angle
and close to the perfect fit (the diagonal line) and the results of the Dip2 0.3810519 0.0002 9.31
goodness of fit, R2 , are 0.99 and both models (dry and saturated) Dip direction2 0.0307442 0.0307442 0.75
Friction angle2 0.0000058 0.0000058 0.00014
are significant.
Tables 8 and 9 show the results of ANOVA, they indicate that
the dip variable is a significant source of variability on the resulted
response (FOS), where the percentage of its effect is 88.6% in the and the response (SRF), as presented in Eqs. (20) and (21) for both
case of the dry condition and 86.1% in the saturated condition. dry and saturated conditions, respectively. These relationships are
When we compare the numerically derived SRF, from FLAC3D, used to construct the response surfaces that used to predict the
and the probabilistically predicted SRF values from BBD, we can values of SRF for any slope realizations based on the variability
clearly see the consistency between the results in terms of the sig- of the random variables, these functions are as follows:
nificant effect of the dip variable on the stability behaviour of the
SRF ¼ 46:49535 þ ð1:25885Þ  D þ ð1:04175Þ  DD
rock slope, which is represented by the variability of the factor of
safety, see Tables 10 and 11. þ ð0:3757222222Þ  £ þ ð0:0134Þ  D2 þ ð0:01Þ
From the BBD experiment, mathematical expressions have been
 DD2 þ ð0:001388889Þ  £2 þ ð0:004Þ  D  DD
established to describe the relations between the random variables
þ ð0:0055Þ  D  £ þ ð0:0002777778Þ  DD  £ ð20Þ

SRF ¼ 47:996538889 þ ð1:26015Þ  D þ ð1:002527778Þ


 DD þ ð0:4311111111Þ  £ þ ð0:01285Þ  D2
þ ð0:0101388889Þ  DD2 þ ð0:000138889Þ  £2
þ ð0:0055Þ  D  DD þ ð0:006666667Þ  D  £
þ ð0:001111111Þ  DD  £ ð21Þ

where D is the dip angle, DD is the dip direction, £ is the friction


angle and SRF is the shear strength reduction.
3D response surfaces have been constructed using the previous
Fig. 21. Plot of simulated by predicted SRFs for slope realizations in dry condition.
mathematical expressions, where each response surface plot
explains the relation between two random variables and the SRF,
while the third variable is fixed at its mean value.
The plots in Fig. 23 illustrate the response surfaces of the slope
realizations in the dry condition. Plot (a) shows the response sur-
face for two variables, the dip and the dip direction against the
SRF, while the third variable (friction angle) is fixed at its mean
value of 27°. The SRF values are directly proportional to the dip
when the dip angle is higher than 50°, whereas the dip direction
does not have a clear impact on the SRF value, which takes the val-
ues in the range (1.2–1.4).
Plot (b) in Fig. 23 shows the constructed response surface for
the dip and the friction angle variables against the SRF values,
Fig. 22. Plot of simulated by predicted SRFs for slope realizations in saturated while the dip direction is fixed at its mean value of 043°. The dip
condition. variable strongly affects the variability of SRF values, where it
368 H. Basahel, H. Mitri / International Journal of Mining Science and Technology 29 (2019) 357–370

Table 10
Simulated and predicted SRFs for slope realization in dry condition.

Slope realizations Patterns Dip (°) Dip direction (°) Friction angle (°) Simulated SRF Predicted SRF
1 0 47 40 27 1.30 1.23
2 +0 47 46 27 1.37 1.32
3 +0 57 40 27 2.64 2.69
4 ++0 57 46 27 2.95 3.01
5 0 47 43 24 0.91 1.01
6 0+ 47 43 30 1.31 1.33
7 +0 57 43 24 2.78 2.76
8 +0+ 57 43 30 2.85 2.75
9 0 52 40 24 1.57 1.54
10 0+ 52 40 30 1.65 1.69
11 0+ 52 46 24 1.78 1.74
12 0++ 52 46 30 1.87 1.90
13 000 52 43 27 1.64 1.64

Table 11
Simulated and predicted SRFs for slope realizations in saturated condition.

Slope realizations Patterns Dip (°) Dip direction (°) Friction angle (°) Simulated SRF Predicted SRF
1 0 47 40 27 1.12 1.10
2 +0 47 46 27 1.08 1.02
3 +0 57 40 27 2.20 2.25
4 ++0 57 46 27 2.49 2.51
5 0 47 43 24 0.71 0.78
6 0+ 47 43 30 1.14 1.15
7 +0 57 43 24 2.32 2.31
8 +0+ 57 43 30 2.35 2.28
9 0 52 40 24 1.30 1.26
10 0+ 52 40 30 1.44 1.45
11 0+ 52 46 24 1.38 1.37
12 0++ 52 46 30 1.48 1.52
13 000 52 43 27 1.31 1.31

Fig. 23. The response surfaces of the slope realizations in the dry condition.

changes from 0.9 at dip angle of 47° to 3 at dip angle of 57°, this increases from 24 to 30; the amount of change in SRF is approxi-
means that a change of 10° in the dip angle causes a dramatic vari- mately 0.15.
ation in SRF. However, the friction angle affects the SRF only Fig. 24 shows the response surface plot of the dip angle and fric-
slightly, where the SRF values positively increase with the increase tion angle variables against the SRF values in case of the saturation
in the angle of friction. condition. The dip direction is fixed at the mean value (043°). It is
Plot (c) in Fig. 23 shows the response surface for the dip direc- clearly seen that the SRF values are significantly affected by the
tion and friction angle variables against the SRF values, while the angle of the joint dip and the relation is linear. Whereas, the plot
dip is fixed at its mean value of 52°. This relation shows that both also shows a limited effect of the change in the friction angle on
variables have relatively equal influence on the response. The dip the SRF values, where SRF value changes from 0.7 to 1.2 when
direction gives low SRF values when the joint strike is almost par- the angle changes in the range (24–30). When we investigated
allel to the strike of the slope face and its value enhances when the the dip and the dip direction variables against the SRF values, while
value of the dip direction increases; the SRF changes from 1.5 at keeping the friction angle fixed at the mean value of 27°, we found
dip direction value of 043° to nearly 1.77 at dip direction value that the effect of the dip on SRF values is pronounced, especially,
of 046°. Similarly, the effect of the friction angle is limited, where when the dip angle exceeds 50°, where the SRF increases sharply.
the SRF values changes from 1.50 to 1.65 when the angle of friction On the other hand, the variability of the dip direction does not
H. Basahel, H. Mitri / International Journal of Mining Science and Technology 29 (2019) 357–370 369

Fig. 26. Histograms and distribution fit for the calculated factor of safety for the
slope in the dry condition.
Fig. 24. Response surface plot for dip angle and dip direction against SRF values in
the case of dry condition.

the inverse Gaussian distribution fit for the calculated factor of


safety; from this distribution curve, the probability of unsatisfac-
tory performance for slope in saturated condition is found to be
64.2%.

13. Conclusions

The stability of rock slopes in mountainous areas, which are


subjected to heavy seasonal rains is examined in this study. Due
to the complexity of these areas, the rock mass discontinuity prop-
erties could be a source of uncertainties. Thus, slope stability stud-
ies will not give a clear answer if the deterministic approach is
Fig. 25. Histograms and distribution fit for the calculated factor of safety for the used. Consequently, the probabilistic methods are adopted in this
slope in the dry condition. work to find out the main sources of the uncertainty and to take
into consideration the variability of several parameters and the
effect of the groundwater pressure on the slopes.
Three parameters namely dip angle, dip direction, and friction
make any influence on the SRF results as they are almost constant angle of the potential failure surface were chosen as random vari-
values. We also observe that the variabilities of the dip direction ables. The factor of safety was selected as the indicator for the
and the friction angle, do not have a pronounced effect on the instability condition (response). The required measurements were
SRF values. collected for these variables, and the proper statistical moments
Through previous observations of the BBD results, we conclude and distributions were defined. Box-Behnken design (BBD) was
that the dip angle of the joint surface has a significant and clear used to incorporate the statistical and numerical methods together
role and control on the stability behaviour of the rock slopes, in in the analysis. A 3D numerical model was constructed; it includes
contrast to the role of the other variables, the dip direction and one joint set, which represents the potential failure surface (set #1)
the friction angle, which have a minor impact on the stability of that was determined through the stereographic projection
slopes. methods.
Based on the BBD design, fifteen treatments are generated to
12.5. Probability of unsatisfactory performance examine the effect of different level combinations amongst the
three random variables on the factor of safety. Using the BBD
The probability of the unsatisfactory performance to evaluate experiments, polynomial functions for two cases of dry and satu-
the stability behaviour of the case study has been done. This eval- rated slopes were established for calculating the factor of safety
uation has been made in the two conditions, the dry and the satu- as a function of the random variables. The results of BBD indicate
rated slopes. The Monte-Carlo technique has been implemented to that the key factor, which significantly influences the factor of
generate 10,000 random values for the input variables, the dip, the safety is the angle of dip of the failure surface. The other factors,
dip direction and the friction angle, for the critical joint set (set #1) dip direction and friction angle did not show clear effect on the
based on the statistical moments and the distribution that has variability of the factor of safety values. The results of the numer-
been determined for each variable. The SRF value is calculated ical modelling agree with the results obtained from the Box-
for each combination of these three variables using Eq. (20) for Behnken design, where both analyses show that the effect of the
the dry slope and Eq. (21) for the saturated slope. Thus, 10,000 val- dip angle of the failure plane is dominant on the instability of
ues have been generated for the SRF and the best fit distribution is the slope.
found for both cases, the dry and the saturated, which was found to Finally, the probability of unsatisfactory performance was
be an inverse Gaussian distribution. As aforementioned, the limit determined for the dry and the saturated slopes. The analysis
of satisfactory performance for the stability is defined to be, was done through Monte-Carlo simulation and the polynomial
SRF = 1.5. Consequently, the probability of unsatisfactory perfor- expression was established from the BBD analysis. The results indi-
mance for the dry and the saturated slopes are identified. Fig. 25 cate that the probability of instability increases by 27.8% in the
shows the histograms and the inverse Gaussian distribution fit case of saturated slopes. It is concluded that the situation of the
for the calculated factor of safety (SRF) for the slope in the dry con- saturated rock slope must be taken into account during the design
dition, the probability of unsatisfactory performance is found to be or in the recommendations for remedial measurements, especially
36.4% from the probability curve. Fig. 26 shows the histograms and in regions similar to the study area of this paper.
370 H. Basahel, H. Mitri / International Journal of Mining Science and Technology 29 (2019) 357–370

Acknowledgments [24] Hoek E, Bray JW. Rock slope engineering. London: Inst. Mining and metallurgy;
1981.
[25] Kostić S. Analytical models for estimation of slope stability in homogeneous
This work is financially supported by the Saudi Geological Sur- intact and jointed rock masses with a single joint. Int J Geomech 2017;17
vey through a doctoral fellowship at McGill University. The authors (10):04017089.
[26] Priest SD, Hudson JA. Discontinuity spacings in rock. Int J Rock Mech Mining
are grateful for their financial as well as technical support. Special
Sci Geomech Abstr 1976;13(5):135–48.
thanks are due to the Saudi Arabian Cultural Bureau in Canada. [27] Park HJ, West TR, Woo I. Probabilistic analysis of rock slope stability and
Special thanks are due to Mr. Mazen Abu Abdullah from Saudi Geo- random properties of discontinuity parameters, Interstate Highway 40,
Western North Carolina, USA. Eng Geol 2005;79(3):230–50.
logical Survey for his help with the maps.
[28] Sturzenegger M, Stead D. Quantifying discontinuity orientation and
persistence on high mountain rock slopes and large landslides using
terrestrial remote sensing techniques. Nat Hazards Earth Syst Sci 2009;9
References (2):267–87.
[29] Varnes DJ. Slope movement types and processes. Special Report
[1] Duzgun HS, Yucemen MS, Karpuz C. A methodology for reliability-based design 1978;176:11–33.
of rock slopes. Rock Mech Rock Eng 2003;36(2):95–120. [30] Wyllie DC, Mah C. Rock slope engineering. 4th ed. Florida: CRC Press; 2004.
[2] Einstein HH. Uncertainty in rock mechanics and rock engineering-Then and [31] Basahel H, Mitri H. Application of rock mass classification systems to rock
now. Proceedings of the 10th ISRM congress on international society for rock slope stability assessment: a case study. J Rock Mech Geotech Eng 2017;9
mechanics, 2003. (6):993–1009.
[3] Fisher BR, Eberhardt E. Assessment of parameter uncertainty associated with [32] Box GE, Hunter JS, Hunter WG. Statistics for experimenters: design,
dip slope stability analyses as a means to improve site investigations. J Geotech innovation, and discovery. New York: Wiley-Interscience; 2005.
Geoenviron Eng 2012;138(2):166–73. [33] Montgomery DC. Design and analysis of experiments. New York: John Wiley &
[4] Hadjigeorgiou J, Harrison JP. Uncertainty and sources of error in rock Sons; 2013.
engineering. Proceedings of the 12th ISRM congress on international society [34] Baamer WS, Youssef AM, Zabramawi YA, Al-Katheeri MM, Baamer MF, Basahel
for rock mechanics, 2011. HM, et al. Rock slope instability analysis along Jabal Fayfa main roads, Jazan
[5] Hammah RE, Yacoub TE, Curran JH. January. Probabilistic slope analysis with area, Kingdom of Saudi Arabia: Saudi geological survey technical report. SGS-
the finite element method. Proceedings of the 43rd US rock mechanics TR-2009-1 2010:65.
symposium & 4th US-Canada rock mechanics symposium on American rock [35] Fairer GM. Reconnaissance geology of the Jabal Fayfa quadrangle, sheet 17/
mechanics association, 2009. 43C. Kingdom of Saudi Arabia: Saudi Arabian deputy ministry for mineral
[6] Jimenez-Rodriguez R, Sitar N, Chacon J. System reliability approach to rock resource open-file report USGS-OF-02-87; 1981.
slope stability. Int J Rock Mech Min Sci 2006;43(6):847–59. [36] Bieniawski ZT, Bieniawski ZT. Engineering rock mass classifications: a
[7] Jaeger JC. Friction of rocks and stability of rock slopes. Geotechnique 1971;21 complete manual for engineers and geologists in mining, civil, and
(2):97–134. petroleum engineering. New York: John Wiley & Sons; 1989.
[8] Park HJ, Um JG, Woo I, Kim JW. Application of fuzzy set theory to evaluate the [37] Barton N, Choubey V. The shear strength of rock joints in theory and practice.
probability of failure in rock slopes. Eng Geol 2012;125(1):92–101. Rock Mech 1977;10(1–2):1–54.
[9] Johari A, Fazeli A, Javadi AA. An investigation into application of jointly [38] Rosenblueth E. Point estimates for probability moments. Proc Natl Acad Sci
distributed random variables method in reliability assessment of rock slope 1975;72(10):3812–4.
stability. Comput Geotech 2013;47(47):42–7. [39] Christian JT, Baecher GB. The point-estimate method with large numbers of
[10] Koyama T, Jing L. Effects of model scale and particle size on micro-mechanical variables. Int J Numer Anal Meth Geomech 2002;26(15):1515–29.
properties and failure processes of rocks - a particle mechanics approach. Eng [40] Christian JT, Baecher GB. Point-estimate method as numerical quadrature. J
Anal Boundary Elem 2007;31(5):458–72. Geotech Geoenviron Eng 1999;125(9):779–86.
[11] Shen H, Abbas SM. Rock slope reliability analysis based on distinct element [41] Harr ME. Reliability-based design in civil engineering. McGraw-Hill Book;
method and random set theory. Int J Rock Mech Min Sci 2013;61(10):15–22. 1987. 126–126.
[12] Baecher GB, Christian JT. Reliability and statistics in geotechnical [42] Miller SM, Whyatt JK, McHugh EL. Applications of the point estimation method
engineering. Chichester, U.K: Wiley; 2003. for stochastic rock slope engineering. Proceedings of the 6th north America
[13] Morgenstern NR. Managing risk in geotechnical engineering. Proceedings of rock mechanics symposium (NARMS) on American rock mechanics
the 10th pan American conference on soil mechanics and foundation association, 2004.
engineering, 1995. [43] Morales JM, Perez-Ruiz J. Point estimate schemes to solve the probabilistic
[14] Griffiths DV, Huang J, Fenton GA. Influence of spatial variability on slope power flow. IEEE Trans Power Syst 2007;22(4):1594–601.
reliability using 2-D random fields. J Geotech Geoenviron Eng 2009;135 [44] Wolff TF. Probabilistic slope stability in theory and practice. Uncertainty in the
(10):1367–78. geologic environment: from theory to practice. ASCE; 1996.
[15] Daftaribesheli A, Ataei M, Sereshki F. Assessment of rock slope stability using [45] Hong HP. An efficient point estimate method for probabilistic analysis. Reliab
the fuzzy slope mass rating (FSMR) system. Appl Soft Comput 2011;11 Eng Syst Saf 1998;59(3):261–7.
(8):4465–73. [46] Wong FS. First-order, second-moment methods. Comput Struct 1985;20
[16] Duncan JM. Factors of safety and reliability in geotechnical engineering. J (4):779–91.
Geotech Geoenviron Eng 2000;126(4):307–16. [47] Box GE, Hunter JS. Multi-factor experimental designs for exploring response
[17] Davis TJ, Keller CP. Modelling uncertainty in natural resource analysis using surfaces. Ann Math Stat 1957;28(1):195–241.
fuzzy sets and Monte Carlo simulation: slope stability prediction. Int J [48] Box GE, Behnken DW. Some new three level designs for the study of
Geographical Inf Sci 1997;11(5):409–34. quantitative variables. Technometrics 1960;2(4):455–75.
[18] Griffiths DV, Fenton GA. Probabilistic slope stability analysis by finite [49] Myers RH, Montgomery DC. Response surface methodology. New
elements. J Geotech Geoenviron Eng 2004;130(5):507–18. York: Willey; 2001.
[19] Karami A, Greer S, Beddoes R. Numerical assessment of step-path failure of [50] Haldar A, Mahadevan S. Probability, reliability, and statistical methods in
northwest wall of A154 Pit, Diavik diamond mines. Proceedings of the 2007 engineering design. New York: Wiley; 2000.
international symposium on rock slope stability in open pit mining and civil [51] Hammah RE, Yacoub TE, Curran JH. Probabilistic slope analysis with the finite
engineering, Perth, Australia, 2007. element method. Proceedings of the 43rd US rock mechanics symposium and
[20] Dershowitz WS, Gordon BM, Kafritsas JC, Herda H. A new three-dimensional 4th US-Canada rock mechanics symposium, 2008.
model for flow in fractured rock. Proceedings of the 17th international [52] Mellah R, Auvinet G, Masrouri F. Stochastic finite element method applied to
congress of the international association of hydrogeologists, Tucson, Arizona, non-linear analysis of embankments. Probab Eng Mech 2000;15(3):251–9.
1985. [53] Dawson E, You K, Park Y. Strength-reduction stability analysis of rock slopes
[21] Brideau MA, Chauvin S, Andrieux P, Stead D. Influence of 3D statistical using the Hoek-Brown failure criterion. Trends in Rock Mechanics, 2000.
discontinuity variability on slope stability conditions. In: Landslides and [54] Dawson EM, Roth WH, Drescher A. Slope stability analysis by strength
engineered slopes: Protecting Society through improved understanding, Banff. reduction. Géotechnique 2015;49(6):835–40.
p. 587–93. [55] Griffiths DV, Lane PA. Slope stability analysis by finite elements. Geotechnique
[22] Shamekhi E, Tannant DD. Probabilistic assessment of rock slope stability using 1999;49(3):387–403.
response surfaces determined from finite element models of geometric [56] Hammah RE, Yacoub TE, Brent C, Curran JH. The shear strength reduction
realizations. Comput Geotech 2015;69:70–81. method for the generalized Hoek-Brown criterion. ARMA/USRMS 2005;5:810.
[23] Hill WJ, Hunter WG. A review of response surface methodology: a literature [57] Barton N, de Quadros EF. Joint aperture and roughness in the prediction of flow
survey. Technometrics 1966;8(4):571–90. and groutability of rock masses. Int J Rock Mech Min Sci 1997;34(3–4):e1–252.

You might also like