Vous êtes sur la page 1sur 9

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/237378922

Rock slope stability and distributed joint


systems

Article in Canadian Geotechnical Journal · January 2011


DOI: 10.1139/t92-006

CITATIONS READS

11 227

2 authors, including:

Bruce J Carter
Cornell University
35 PUBLICATIONS 748 CITATIONS

SEE PROFILE

All content following this page was uploaded by Bruce J Carter on 05 December 2015.

The user has requested enhancement of the downloaded file.


Rock slope stability and distributed joint systems
1

BRUCEJ. CARTER
AND EMERY
Z. LAJTAI
Departments of Civil and Geological Engineering, The University of Manitoba, Winnipeg, Man.,
Canada R3 T 2N2
Received February 5, 1991
Accepted September 30, 1991

A deterministic (GEOSLIDE) and a probabilistic (PROSLIDE) microcomputer code are introduced to aid in performing
rock wedge analyses based on the limit equilibrium method. The deterministic code evaluates the stability of a single
rock wedge formed by discontinuities in rock through three-dimensional vector algebra. GEOSLIDE undertakes a full
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Guangzhou Jinan University on 06/03/13

kinematic analysis (daylighting and obstruction), analyzes both wedge and plane sliding, and provides for anchor designs
and sensitivity analyses (cohesion, friction, and water forces). Through multiple stability analyses, PROSLIDE evaluates
the probability of failure for a rock slope by examining the distribution of the factors of safety from all the potential
sliding wedges formed by the discontinuities of the rock mass. The probability of failure is expressed as the ratio of
kinematically free wedges that have a factor of safety less than unity to the total number of wedges. PROSLIDE can
form and analyze as many as 2000 different pairs of discontinuities in less than 30 min using a 25 MHz 486 IBM-
compatible computer. In a worked example, the probability of failure for a fixed slope strike and loading condition
is shown to vary with the slope angle, following the characteristic 'S' shape of a cumulative distribution function. The
effect of an anchor force is to spread the distribution over a wider range of the factor of safety (SF), pushing many
wedges into a potential upslide situation and splitting the distribution about the failure zone of the stability diagram
(-1 < SF < 1).
Key words: rock slope, rock wedge, stability analysis, factor of safety, probability of failure, Monte Carlo simulation.

Un code dkterministique (GEOSLIDE) et un code probabilistique (PROSLIDE) pour les micro-ordinateurs sont introduits
pour aider rCaliser des analyses de coins de roc basCes sur 1'Cquilibre limite. Le code dkterministique Cvalue la stabilitk
d'un simple coin de roc, form6 par des discontinuitCs dans la roche, au moyen d'algkbre vectorielle tridimensionnelle.
GEOSLIDE prockde 2 une analyse cinkmatique complkte (espace libre et obstruction), analyse le glissement du coin et
For personal use only.

du plan et fournit la possibilitC de calculer des ancrages et de faire des analyses de sensibiliti (cohision, friction et
forces de pression d'eau). Au moyen d'analyses multiples de stabilitC, PROSLIDE Cvalue la probabilitk de rupture d'un
talus rocheux en examinant la distribution des coefficients de sCcuritC de tous les coins formCs par des discontinuitCs du
massif rocheux et pouvant potentiellement glisser. La probabilitk de rupture est exprimCe par le rapport des coins cinCmati-
quement libres qui ont un coefficient de sCcuritC infkrieur 2 I'unitC, sur le nombre total de coins. PROSLIDE peut former
et analyser jusqu'a 2000 paires differentes de discontinuitCs en moins de 30 min en utilisant un micro-ordinateur compatible
a I'IBM 486 de 25 MHz. Dans un exemple de calcul, l'on montre que la probabilitC de rupture pour une direction
de pente et une condition de chargement donnCes varie avec l'angle de la pente, se conformant a la forme en
S caractkristique d'une fonction de distribution cumulative. L'effet de la force d'un ancrage est de rCpartir la distribu-
tion sur une plage plus large du coefficient de sCcuritC (SF), repoussant plusieurs coins dans une situation potentielle
vers le haut du talus et coupant la distribution du diagramme de stabilitC autour de la zone de rupture (- 1 < SF > 1).
Mots clPs : talus rocheux, coin de roc, analyse de stabilitC, coefficient de stabilitC, probabilitk de rupture, simulation
de Monte Carlo.
[Traduit par la redaction]

Can. Geotech. J. 29, 53-60 (1992)

Introduction approach to slope design is appropriate ( ~ o a t e s1977;


One of the most difficult jobs in rock slope engineering Einstein and Baecher 1983). Probabilistic techniques require
is selecting design data for a deterministic stability analysis statistical assessments of multiple analyses. Such analyses
from a site investigation report prepared by an engineering can now be conducted relatively quickly using microcom-
or structural geologist. The availability of a contoured puters. The purpose of this paper is (i) to introduce a
stereonet plot of joints displaying hundreds of careful field microcomputer tool that performs multiple analyses and
measurements is not necessarily comforting, since the con- (ii) to show how this tool can solve problems associated with
ventional design routine for wedge failure (Hoek and Bray the design of rock slopes.
1977) requires only two planes of weakness. The median
orientations of planes form the representative wedge, but
perhaps the design should be for the worst condition. Alter- Deterministic analyses of slope stability
natively, why not analyze all the possible wedges that could Rock slope failure may occur in a variety of ways, involv-
be formed by all possible combinations of the discontinuities ing rigid body movement by translational sliding or rotation
in the rock mass? of single or multiple rock slices or wedges (Varnes 1978;
The latter technique falls into the general category of Martin and Kaiser 1984) or through toppling (Goodman and
probabilistic design. Since the geological domain from which Bray 1976; Aydan et al. 1989; Scavia et al. 1990). This paper
the design data must be selected is probabilistic and the geo- is concerned with only one type of movement, translational
logical data consequently widely distributed, a probabilistic wedge or planar sliding (Hoek and Bray 1977).
Printcd in Canada / Imprime au Canada
54 CAN. GEOTECH. J. VOL. 29, 1992

The general deterministic method of limit equilibrium If any of the parameters are correlated, however, the Monte
analysis for wedge or planar stability dates back to the 1970s Carlo solution may give incorrect'resultS, in which case the
(Jaeger 1971; Kutter 1974), and recent techniques are based proper solution lies in the reliability index or first-order
predominantly on the work of Hoek and Bray (1977) and second moment (level 11) method (Glynn and Ghosh 1982;
Goodman (1976). A typical slope-stability analysis consists Nguyen 1985). An additional advantage of this method over
of collecting structural data, selecting the representative Monte Carlo simulations is a reduction in computer time
wedge, and a kinematic analysis of this wedge for daylighting (Barbosa et al. 1989). The Monte Carlo simulation method,
and obstruction. The loads are then introduced, i.e., gravity, however, is very easily adapted to existing limit equilibrium
water forces, earthquake loading, surcharge, and other computer codes.
external forces, and the factor of safety for the most likely All Monte Carlo simulations produce not a single factor
or critical sliding mode, either planar or wedge slide along of safety, but a distribution of values. The probability of
dip vectors or along the line of intersection for wedge slide, failure may then be defined by dividing the number of values
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Guangzhou Jinan University on 06/03/13

is computed by limit equilibrium methods. Sensitivity anal- less than one by the total number (Savely 1985; Dershowitz
yses evaluating the influence of such parameters as cohesion, and Einstein 1984). Alternatively, the probability of failure
friction, and water condition may lead to redesign by can be found from the fitted normal probability density dis-
anchoring or by changing the geometry or loading condition. tribution diagram by calculating the area under the curve
The final result of the deterministic analysis is a single for factors of safety less than unity (Savely 1985).
factor of safety, the ratio of the available resistance and the
driving force (the component of the total force that points Software description
in the sliding direction). The stability analysis codes described in this paper have
been developed to enable deterministic wedge or plane slid-
Statistical analyses of slope stability ing using GEOSLIDE and a probabilistic analysis with
In rock mechanics, practically all the parameters are PROSLIDE. GEOSLIDE is an interactive, computer implemen-
statistical; the joint set characteristics (Call et al. 1976; Glynn tation of the conventional, vector-based, wedge analysis
and Einstein 1979), the rock strength properties (Chowdhury method, and it includes numerous graphical displays of the
1986), and the loading conditions (Glass et al. 1978; Tao stereonet, wedge, and force vectors. It has been designed
and Hong 1987). Since the introduction of probabilistic tech- to analyze a single wedge or plane geometry. PROSLIDE is
niques for rock slope stability (McMahon 1971), site-specific a multiple-analysis version of GEOSLIDE and will run unat-
For personal use only.

applications have been published (Piteau and Martin 1977; tended until either a prescribed number or all the possible
Savely 1985). combinations of discontinuities have been examined. Both
In any analysis of rock stability, the first step is always codes were written in True BASIC and will run on any type
the collection of field data on joint orientation. The rock of IBM PC or compatible equipped with a graphics card
mass is usually divided into structural domains and within and monitor. The codes were tested with monochrome and
these the discontinuities are separated into sets, with several color monitors, on PC, XT, AT-286, AT-386, AT-486, and
sets forming a system. The statistical distributions of the PS/2 computers.
orientations and possibly the strength-related structural data For a complete slope-stability analysis using the codes
can then be established for each set (Coates 1977; Herget developed by the authors, three different programs may be
1978; Savely 1985). The interpretation of the field data is required: GEOSLIDE, the solution for the sliding wedge or
not necessarily a simple process. For instance, joint orien- plane; PROSLIDE, the discontinuity analysis code (Monte
tation is subject to error and bias (Barton 1978; Herget 1978; Carlo simulation); and some type of a statistics-graphics
Baecher and Einstein 1981). However, the bias in orienta- package. Another routine, DISIT,also written by the
tion and in other parameters, such as trace length, persis- authors, can be used to turn the single-variable True BASIC
tence, and spacing (Baecher and Lanney 1978), can be record files of PROSLIDE into histograms and cumulative
treated quantitatively to correct for errors in measuring distributions or to fit Gaussian and Weibull distributions
(Terzaghi 1965; Wathugala et al. 1990). to the factors of safety. DISIT can only do graphics screen
Once the correct parameter distributions are known, dumps. For professional-quality figures, a commercial soft-
stability analyses may proceed in several ways: (i) conven- ware that does x-y plots and can recover the ASCII (text)
tional limit equilibrium analysis using a subjectively selected files of PROSLIDE would be a better choice.
critical wedge (Hoek and Bray 1977); (ii) entirely probabi- There have been a number of other computer codes devel-
listic techniques, either Monte Carlo simulations (Einstein oped to analyze rock slope stability (ISRM 1988; Gibbs
et al. 1983; Scavia et al. 1990) or reliability index methods 1989), some of which include probabilistic analyses. Other
(Nguyen 1985; Ramachandran and Hosking 1985; Barbosa programs that have been referenced in papers include
et al. 1989); or (iii) a combination of deterministic and prob- FRACWEDG (Kendorski and Bindokas 1987) and an unnamed
abilistic techniques (Piteau et al. 1985; Martin et al. 1986). Fortran code written for the PC (Ghosh and Haupt 1989).
Neither the deterministic nor the probabilistic methods are Most of the codes have quite similar descriptions, although
entirely complete by themselves, as the concepts of cost and actual similarities and differences can only be found by using
risk assessment (Coates 1977; Call 1985; Kirsten and Moss all of the codes. A comparison of these codes with each other
1985; Einstein 1988) also merit consideration. and GEOSLIDE-PROSLIDE is beyond the scope of this paper,
Probabilistic analyses can take many forms. A truly prob- however.
abilistic method should accept statistical distributions (cor- GEOSLIDE and PROSLIDE were originally developed as
rected for bias) for every parameter that is subject to spatial learning tools for the undergraduate rock mechanics stu-
variability. Monte Carlo simulations can be performed using dents. This is the main reason for the two-part program,
random values of the parameter distributions (Coates 1977). as the actual mechanics of wedge and planar sliding are the
CARTER AND LAJTAI

main focus, whereas statistics are not taught in detail.


Because the programs are learning tools, there are a great
number of explanatory screens showing intermediate results
as the analyses proceed. Both GEOSLIDE and PROSLIDE have
been run successfully for more than a year by both gradu- NO SLIDING POSSIBLE

ate and undergraduate rock mechanics students at the


authors' institution. The codes have been verified using the
examples of Hoek and Bray (1977). PROSLIDE, GEOSLIDE,
and DISIT may be obtained from the authors at the cost of
reproduction and postage. A manual may not be necessary
C DISPLAY WEDGE

k
TRUNCATE EDGES
but it can also be supplied at extra cost.
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Guangzhou Jinan University on 06/03/13

Deterministic analysis using GEOSLIDE


The sliding wedge problem of the Hoek and Bray (1977)
slope-stability analysis is implemented by this code. The
wedge is assumed to be continuous and rigid. The driving
and resisting forces are calculated as vectors, and the factor
of safety is found using three-dimensional vector algebra.
The code works with a single set of geometry and loading,
although a sensitivity-analysis option allows the loading and
strength parameters to be varied once the initial factor of
safety is calculated. The numerator in the factor of safety I
CALCULATE FACTOR O F SAFETY
term is the sum of the total cohesive and frictional resisting
I
forces. The denominator is the sliding-direction-parallel
component of the resultant force which includes weight, sur-
charge, water, earthquake, external, and anchor forces. The
For personal use only.

sliding-direction-parallel component is positive when it


points downslope, negative when it points upslope.
A negative factor of safety signals a potential upslide
(a safety factor between 0 and - 1 is unstable and will slide
upwards).
The program flow is shown in Fig. 1. The third step, < / CHANGE LOADS
kinematic analysis, consists of two parts: a check for SENSITIVITY ANALYSES
QUIT
SENSITIVITY ANALYSES

daylighting of the line of intersection and the two dip vec-


tors (Hoek and Bray 1977) and an obstruction analysis .I. <
(Goodman 1976) if either of the two dip vectors daylights.
This process eliminates those dip slides (wedge sliding along
the dip vector of plane A or B) where the other inactive plane
would block the sliding motion. The daylighting and c7->
obstruction analyses are accompanied by explanatory FIG. 1. The flow chart for the interactive, deterministic, limit
graphic displays. If both the daylighting and the obstruc- equilibrium, microcomputer code GEOSLIDE.
tion conditions are passed, i.e., sliding is kinematically possi-
ble, the stability analysis begins.
The wedge is displayed in vector format and in a three- condition, can be varied, and the effect of the selected
dimensional view which can be rotated. The shape of the parameter on the factor of safety is displayed on the screen
wedge can be modified by truncating in the lateral direc- in x-y plots. An anchor can be installed to increase the factor
tion and by inserting a tension cutoff at the back of the of safety to a desired level. The cohesion subroutine can han-
wedge. The code then accepts a series of loading conditions: dle back analysis, a procedure to find the cohesive force for
external forces (anchors, earthquake loading, etc.) and an actual slide.
several options for the water condition.
For every wedge, five different sliding modes can be Probabilistic analysis using PROSLIDE
examined: (i) wedge sliding along the line of intersection GEOSLIDE requires the full attention of the user, whereas
between plane A and plane B; (ii) wedge sliding along the PROSLIDE is designed to work unattended. The user must
dip vector of plane A; (iii) wedge sliding along the dip vec- enter the slope geometry, a system of discontinuities, and
tor for plane B; (iv) plane sliding on plane A; and (v) plane the loading condition. After that, the code will take over
sliding on plane B. The corresponding factor of safety is by doing a Monte Carlo simulation. The simulation is done
calculated for each active sliding direction. The last two by randomly selecting the discontinuities rather than ran-
(plane slide) cases represent unobstructed sliding of a block dom values of the individual parameters; the discontinuity
of rock along the dip vector of either joint plane. listing includes the trend, plunge, cohesion, and friction.
The influence of the loading condition and the strength The actual code is a simplified version of GEOSLIDE
parameters can be evaluated through the sensitivity sub- (Fig. 2). For efficiency, all the graphics and displays of inter-
routines within GEOSLIDE. All the forces, including the water mediate results are eliminated. The sensitivity subroutines,
CAN. GEOTECH. J. VOL. 29, 1992

C? START

DISCONTINUITIES
LOADING
lf OF RUNS

PAIR OF DISCONTINUITIES
n - E
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Guangzhou Jinan University on 06/03/13

slope

KINEMATIC ANALYSIS

OBSTRUCTION ANALYSIS

FIG. 3. A stereonet showing the slope geometry at a slope angle


of 45" and the poles to the discontinuities in the rock mass.
matched with all other planes. Any pair that has already
been analyzed is bypassed in the selection routine; no
FlND THE FACTOR OF SAFETY
FOR ALL SLIDING MODES duplication of wedges is allowed. After picking a pair of
discontinuities, the usual deterministic, limit equilibrium
For personal use only.

analysis is undertaken. If sliding is possible, the code can


calculate as many as five safety factors. The five safety fac-
I FlND MINIMUM FACTOR O F SAFETY
STORE DATA IN USER FILES
tors represent the three potential wedge sliding directions
and the two potential plane sliding directions. Only four of
the safety factors are independent because sliding on plane A
and plane B should give identical distributions, since any
discontinuity may appear as either plane A or plane B. The
case is slightly different for wedge sliding on the dip vectors,
since the resulting safety factor will depend on the full
geometry of the wedge. Still, the cumulative distributions
for wedge sliding on the dip vectors of plane A and plane B
should be very similar. For each sliding wedge, the code will
DETERMINE CUMULATIVE PROBABILITIES
MAKE TEMPORARY DATA FILES label the lowest of the safety factors as the minimum factor
PLOT SF vs CUMULATIVE PROBABILITY of safety (MSF).
The output of PROSLIDE is the listing of the factors of
safety in several files. The first file is the largest; the num-
ber of rows is the total number of potentially sliding
FIG.2. The flow chart for the multiple-analysis code PROSLIDE. (kinematically free) wedges, and the seven columns are the
identification numbers for plane of weakness A and B and
anchoring subroutine, and the ability to modify the loads the five safety factors, namely, sliding along the line of
are also dispensed. The discontinuity listing is the new ele- intersection, wedge slide on plane A, wedge slide on plane B,
ment in the analysis. This can be entered from the keyboard plane slide on A, and plane slide on B. The second file con-
or from previously stored True BASIC record files. The dis- tains the discontinuity numbers (A and B) and the minimum
continuities need not be grouped into sets. The only require- factor of safety, and the third file carries the discontinuity
ment is to enter the (i) trend of the pole, (ii) plunge of the numbers and the minimum wedge factor of safety (MWSF).
pole and two strength parameters, (iii) unit cohesion, and The latter file ignores the factors of safety produced through
(iv) friction angle. Admittedly, estimating the unit cohesion plane analysis, so MWSF is the smallest of the three possible
and friction angle for each discontinuity is a difficult if not wedge sliding modes (intersection, dip of A, or dip of B).
impossible task. Friction angle could be estimated by using The three ASCII files described above are optionally
published surface profiles (Barton 1974), and unit cohesion created permanent files. There are also several temporary
used as a weighting parameter. The geologist could weight files that are rewritten during each run of PROSLIDE. Single-
the significance of a particular discontinuity by using a single variable (factor of safety for one sliding mode) True BASIC
figure for such descriptive terms as major and minor joint, record files are mainly for use in DISIT to make histograms
joint persistence, length, continuity, and spacing. and cumulative distribution curves. A set of two-variable
Every plane has the same chance of being selected for ASCII files are for x-y plotting of cumulative probability
either plane A or plane B, and each plane is eventually against factor of safety.
CARTER A?qD LAJTAI 57
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Guangzhou Jinan University on 06/03/13

factor of safety factor of safety


FIG. 4. The cumulative probability distribution of the mini- FIG. 5. The cumulative probability distribution for the mini-
mum of the five factors of safety, from a PROSLIDE run for a slope mum factor of safety (total) includes the values for wedge sliding
angle of 45". along a dip vector (dip slide) or along the line of intersection
(intersection slide) and for unobstructed sliding on a single plane
(plane slide). Plane slide is considered only if the discontinuity strike
Probabilistic assessments is within 15" of the strike of the slope. In this example, only dis-
The Monte Carlo simulation performed by PROSLIDE continuity 63 falls into this category.
yields a list of factors of safety for every possible, kine-
matically free wedge. Absolute values of the factor of safety but could cause upslide conditions for others. In fact, several
of less than one indicate wedges that will fail. The distribu- runs of PROSLIDE may be required to find the best slope
tion of values can be shown by both frequency and cumula- design. In a typical slope-stability exercise, the user may have
For personal use only.

tive probability diagrams using either PROSLIDE, DISIT,or to run both GEOSLIDE and PROSLIDE several times.
some other statistics program. As mentioned previously, the
probability of failure is taken as (Dershowitz and Einstein
1984) Application of probabilistic analysis
Unstable wedges The nature and the advantage of a probabilistic analysis
P(f) Total possible wedges using PROSLIDE are best demonstrated through an example.
Although this is not the true probability of failure, unless A 30 m high slope, striking at S70°E and capped by a top
all the input parameters have been expressed as corrected slope (strike at S60°E and dip of lo0), is to be excavated
probability distributions, it does serve to represent the in a rock mass that contains three major and one minor set
stability of the slope. of discontinuities. Eighty-seven measurements of disconti-
PROSLIDE is not a fully probabilistic analysis package. nuities are displayed in Fig. 3. For this example, the first
Only the discontinuities (orientation, and the associated run has no loading other than the weight and the water
strength parameters) are allowed to vary, thus the Monte forces arising from a water condition, where the piezometric
Carlo simulation method is well suited to the analysis. The surface at the slope crest has a head that is 75% of the depth
code otherwise expects the input of a single set of param- to the slide plane below. The latter fixes the position of the
eters describing the slope geometry and the loading condi- piezometric surface at the crest in the vertical plane of the
tion. It is the authors' opinion that the effect of the line of intersection (or dip vector for plane slide). The head
variability of the loading conditions is better treated through at the recharge and discharge points is taken as zero. Linear
the sensitivity analysis option of GEOSLIDE. Ideally, PROSLIDE variation is assumed for intermediate positions. Because esti-
would evaluate the influence of the probabilistic nature of mates of the unit cohesion are often unreliable, it is possible
the geological condition by searching for the critical wedge at this stage to exclude cohesion for all discontinuities, a
geometry. This geometry is then used in CEOSLIDE to test feature that will be used later. The user may limit the num-
for the sensitivity of this wedge to variations in the loading ber of simulations or let the code examine all possible com-
conditions and perhaps to find the appropriate support sys- binations of planes; N discontinuities can form N ( N - 1)/2
tem. A new run of PROSLIDE would indicate the effect of different pairs of discontinuities. It is not necessary to
the new loading conditions on the rest of the wedges. examine all the possible combinations. The shape of the
A 100% safe design would move all the wedges out of the cumulative distribution curve changes very little once the
unsafe region; no factors of safety should fall between - 1 number of determinations (measured factors of safety)
and + 1. This may not always be practical or even possible, exceeds about 100. For this particular discontinuity distri-
and typically a certain percentage of wedges would be bution and slope geometry, 500 pairs of discontinuities will
allowed to fail. yield between 100 and 150 potential sliding wedges. The
Another use for PROSLIDE is to find the ideal slope angle figures of this paper show cumulative probability plots based
for the given set of discontinuities, or if a slope higher than on factor of safety data from a 500 pair run.
this is required, to find the proper orientation and magnitude To determine the ideal slope angle, several runs of
of an anchor force. An anchor installed in a particular orien- PROSLIDE using different angles will be required. A good
tation may improve the factor of safety for some wedges starting point is a dip of 45". PROSLIDE will automatically
CAN. GEOTECH. J. VOL. 29, 1992

$.:-!.,
45'Slope
100 MN Anchor f
.--
-______-- /

/. -- - - - -I;."'

45" Slope
W~thoutAnchor
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Guangzhou Jinan University on 06/03/13

I S t , I , , ,

factor of safety factor of safety


FIG. 6. Cumulative probability distribution of the factor of FIG. 8. An anchor force widens the distribution of factors of
safety for slope angles between 30 and 90". safety and splits the distribution about the unsafe region of - 1 to
+ 1.

percent of wedges ..
.-.X- kinematically free to slide .
-3
C
.-
C
50 -

40 :
0
0
.!? . . wedges
percent of
For personal use only.

30
u failing
'4-
0
,-
C
20 : . , (-1 < SF < I )
a, 10 -
2

factor of safety
slope angle
FIG. 9. Changing the magnitude and (or) the plunge of the
FIG.7. The percentage of wedges that are kinematically free to anchor force changes the distribution of factors of safety. Although
slide and the percentage of failing-sliding wedges (- 1 < SF < 1) it may be impractical, an anchor can usually be designed to move
as a function of the slope angle. The percentage terms are based all the safety factors out of the unsafe region (- 1 to + I). 0,100
on the number of discontinuity pairs examined in a single PRO- MN at 15"; A, 150 MN at 15"; 0 , 150 MN at 20".
SLIDE run.

display the cumulative probability of the minimum factor angle (Fig. 6). Only the 30" and the 40" slopes have no
of safety at the end of the simulations (Fig. 4). The distri- wedges with factors of safety less than unity. The 45" slope,
bution does not include all the factors of safety computed not shown in this figure, has about 6% of the wedges fail-
during the run, only the minimum of the five sliding modes. ing. As expected, both the number of wedges free to slide
One of the files will have the discontinuities identified with increases and the number of failures among these increases
the minimum factors of safety. A check of the listings shows with the slope angle (Fig. 7). Both relationships have the
the small vertical segment at SF = 0.994 is due to multiple characteristic shape of the cumulative distribution shown
plane analysis of plane sliding on discontinuity 63, which by others (Young and Hoerger 1988; Piteau and Martin
happens to run subparallel with the slope. In the postprocess- 1977). The effect of the slope angle is relatively minor at
ing routine of PROSLIDE, there are several ways to display low and at high angles. The largest effect is indicated for
the factor of safety distributions for individual sliding slopes between 40 and 60".
modes. It is also possible to decompose the total distribu- Anchoring the potentially unsafe slope is always an alter-
tion into its components. Figure 5, for example, shows that native. The effect of an installed anchor is relatively simple
the lowest values are coming from the wedge sliding along to establish for a single wedge using GEOSLIDE. The
the dip vectors of the discontinuities rather than from sliding influence of an anchor, designed for one particular wedge,
along the line of intersection between them. Only the dis- on the rest of the wedges is more problematic. When the
tribution of the minimum wedge factors of safety will be data are widely distributed, as in this case, a force pointing
considered now, even though the plane-failure case may be upslope will spread the distribution even wider, possibly into
more important in many cases. negative values. The benefit of an anchor force is seen in
While keeping the external loading condition the same, the splitting of the distribution by moving some or all the
the minimum wedge factors of safety will vary with the slope failing wedges out of the - 1 to + 1 failure zone (Fig. 8).
CARTER AND LAJTAI 59
1 0 .
Summary and conclusions

X 0 8 r
E
,.) ... ...
... . . .. .. 8'-
f

*--
&u-
The stability of a rock slope depends greatly on the dis-
continuities in the rock mass. For a deterministic slope-
--
Z
----
~l
m
no cohes~on /./" * /%--@- e-#- 0 *-
stability analysis, specific values for such discontinuity
~l parameters as orientation (trend and plunge), unit cohesion,
06
8° and friction angle are required. All the parameters are sub-
I"
C )
a
ject to statistical variation. The usual procedure in a deter-
:
a,
f-heson ministic design is to select a representative wedge with prop-
- 0 4
3

$0 0 2 ,I
f gU
/*
45" Slope
erties that are judged to be appropriate and then calculate
a factor of safety. Probabilistic techniques, on the other
hand, incorporate the variability of design parameters and
I ..., /
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Guangzhou Jinan University on 06/03/13

express stability in probabilistic terms. The probabilistic


E , ,, , , , method described in this paper expresses the probability of
0 0
o 1 2 3 4 5 failure as the ratio of wedges that fail under the given loading
conditions to the total number of wedges. The wedges are
factor of safety defined by randomly selecting a pair of discontinuities from
FIG. 10. Eliminating the cohesion, a parameter that is difficult the entered list.
to evaluate, for all the discontinuities shifts the cumulative proba- Designs of rock slopes with representative values of the
bility distribution curve toward a lower factor of safety. discontinuities, as done in a deterministic analysis, could be
in significant error. The 50" slope of Fig. 6, for example,
would suggest a safety factor of 1.7 (median value) for the
A successful design would move all the wedges out of the representative wedge. However, the probability of failure
danger zone using the smallest possible force. This is best for this slope is quite high at 16%.
done by trial and error. There are two parameters that can For the same loading condition and slope strike, the prob-
be adjusted in PROSLIDE, the magnitude and the orientation ability of failure varies with the slope angle. The relation-
of the anchor. The orientation itself involves two param- ship, however, is not linear but follows the characteristic
eters, the trend and the plunge. Increasing the anchor S shape of the cumulative distribution function. The prob-
For personal use only.

magnitude will generally increase the upslide potential. This ability of failure for the presented example shows little
can be counteracted by increasing the plunge of the anchor. change for the low and high slope angles but varies rapidly
Changing the trend of the anchor leads to results that are in the 40"-60" range. Interestingly, the percentage of wedges
harder to anticipate. The distribution shown in Fig. 8 sug- that have the kinematic freedom to slide varies with the slope
gests too many upslope failures for the 100 MN anchor at angle in a similar manner.
a 5" plunge and a trend along that of the normal to the slope. The effect of an upslope-pointing external force, an
Using the same magnitude, the upslide potential can be elim- anchor, is to spread the factor of safety distribution wider,
inated completely by increasing the plunge to 15" (Fig. 9). possibly into negative values, and to split the distribution
This condition leaves a single failing wedge. This may be around the SF = - 1 to + 1 failure zone.
a successful design, since the probability of failure is less
than 1%. Furthermore, the failing wedge is identified as a Acknowledgements
sliver: an overlapping wedge formed by discontinuities 52 The code GEOSLIDE was developed from the ideas and
and 15. Both discontinuities come from the same set. concepts found in Rock slope engineering by Hoek and Bray
Analyzing this pair in GEOSLIDE would help in evaluating (1977) and in Methods in geological engineering in discon-
the risk of ignoring this unstable wedge. Alternatively, a few tinuous rocks by Goodman (1976). Earlier versions of
tries with a higher anchor force and a steeper plunge could GEOSLIDE were prepared by graduate students Neil Chandler
clear the failure zone. and Alan Wan of the University of Manitoba. Debugging
Anchors were examined in this case, simply to indicate of both GEOSLIDE and PROSLIDE would not have been possi-
the general applications of probability to slope-stability ble without the input of undergraduates who had to use both
design. For actual slope designs, costs and risk assessment codes in assignments. Discussions on the concepts of prob-
would have to be included as well. In this particular example, abilistic analysis with Brian Stimpson and Alan Woodbury
the failing wedges were mostly slivers and it is possible that of the University of Manitoba have been of great help.
some small failures would be acceptable. Alternatively,
anchors could be installed on a local basis in areas of high
risk. Aydan, O., Shimizu, Y., and Ichikawa, Y. 1989. The effective fail-
After the design of the anchor, one practical problem still ure modes and stability of slopes in rock mass with two discon-
remains. The discontinuity listing requires the input of spe- tinuity sets. Rock Mechanics and Rock Engineering, 22: 163-188.
cific values of unit cohesion and friction angle. The con- Baecher, G.B., and Einstein, H.H. 1981. Statistical and proba-
ventional attitude is that the friction angle can be estimated bilistic methods in rock engineering. Workshop Notes,
with acceptable accuracy, but estimates of the unit cohesion Massachusetts Institute of Technology, Cambridge.
Baecher, G.B., and Lanney, N.A. 1978. Trace length biases in joint
will always be open to significant error. Cohesion may be surveys. Proceedings, 19th U.S. Symposium on Rock Mechanics,
ignored and PROSLIDE allows setting cohesion for all planes Lake Tahoe, Nev., pp. 56-65.
equal to zero. This shifts the distribution to the left toward Barbosa, M.R., Moriss, D.V., and Sarma, S.K. 1989. Factor of
lower factors of safety (Fig. lo), indicating a greater prob- safety and probability of failure of rockfill embankments.
ability of failure. Geotechnique, 39: 47 1-483.
60 CAN. GEOTECH. J. VOL. 29, 1992

Barton, C.M. 1978. Analysis of joint traces. Proceedings, 19th U.S. Jaeger, J.C. 1971. Friction of rocks and stability of rock slopes.
Symposium on Rock Mechanics, Lake Tahoe, Nev., pp. 38-41. GCotechnique, 21: 97-134. \
Barton, N.R. 1974. Estimating the shear strength of rock joints. Kendorski, F.S., and Bindokas, A. 1987. Fracture geometry
Proceedings, 3rd Congress, International Society Rock characterization for use in rock mechanics design. Proceedings,
Mechanics, Denver, Colo., pp. 219-220. 28th U.S. Symposium on Rock Mechanics, Tucson, Ariz.,
Call, R.D. 1985. Probability of stability design of open pit slopes. pp. 89-97.
In Rock masses: modeling of underground openings/probability Kirsten, H.A.D., and Moss, A.S.E. 1985. Probability applied to
of slope failure/fracture of intact rock. Editedby C.H. Dowding. slope design-case histories. In Rock masses: modeling of under-
Proceedings of the Geotechnical Engineering Division, American ground openings/probability of slope failure/fracture of intact
Society of Civil Engineers, Denver, Colo., pp. 56-71. rock. Edited by C.H. Dowding. Proceedings of Geotechnical
Call, R.D., Savely, J.P., and Nicholas, D.E. 1976. Estimation of Engineering Division, American Society of Civil Engineers,
joint set characteristics from surface mapping data. Proceedings, Denver, Colo., pp. 56-71.
17th U.S. Symposium on Rock Mechanics, Snowbird, Utah, Kutter, H.K. 1974. Analytical methods for rock slope analysis. In
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by Guangzhou Jinan University on 06/03/13

pp. 65-73. Rock mechanics. Edited by L. Muller. Springer-Verlag,


Chowdhury, R.N. 1986. Geomechanics risk model for multiple New York, pp. 198-211.
failures along rock discontinuities. International Journal of Rock Martin, C.D., and Kaiser, P.K. 1984. Analysis of a rock slope with
Mechanics and Mining Sciences and Geomechanics Abstracts, internal dilation. Canadian Geotechnical Journal, 21: 605-620.
23: 337-346. Martin, D.C., Steenkamp, N.S.L., and Lill, J.W. 1986. Applica-
Coates, D.F. 1977. Pit slope manual. Chap. 5 Design. Canada Cen- tion of a statistical analysis technique for design of high rock
tre for Mineral and Energy Technology (CANMET), Report slopes at Palbora mine, South Africa. Mining Latin America
77-5. IMM Conference, Santiago, Chile, pp. 241-255.
Dershowitz, W.S., and Einstein, H.H. 1984. Application of McMahon, B.K. 1971. A statistical method for the design of rock
artificial intelligence to problems of rock mechanics. Pro- slopes. Proceedings, 1st Australia - New Zealand Conference
ceedings, 25th U.S. Symposium on Rock Mechanics, Northwest- on Geo-Mechanics, Melbourne, Australia, pp. 314-321.
ern University, Evanston, Ill., p. 489. Nguyen, V.U. 1985. Reliability index in geotechnics. Computers
Einstein, H.H. 1988. Special lecture: landslide risk assessment pro- and Geotechnics, 1: 117-138.
cedure. In Landslides, Proceedings of the 5th International Sym- Piteau, D.R., and Martin, D.C. 1977. Slope stability analysis and
posium on Landslides, Lausanne, Switzerland, A.A. Balkema, design based on probability techniques at Cassiar mine. CIM
Rotterdam, vol. 2, pp. 1075-1090. Bulletin, 70(No. 799): 139-150.
Einstein, H.H., and Baecher, G.B. 1983. Probabilistic and Piteau, D.R., Stewart, A.F., Martin, D.C., and Trenholme, B.S.
For personal use only.

statistical methods in engineering geology. Specific models and 1985. A combined limit equilibrium and statistical analysis of
examples, part I: exploration. Rock Mechanics and Rock wedges for design of high rock slopes. In Rock masses: modeling
Engineering, 16: 39-72. of underground openings/probability of slope failure/fracture
Einstein, H.H., Veneziano, D., Baecher, G.B., and O'Reilly, K.J. of intact rock. Edited by C.D. Dowding. Proceedings of the
1983. The effect of discontinuity resistance on rock slope Geotechnical Engineering Division, American Society of Civil
stability. International Journal of Rock Mechanics and Mining Engineers, Denver, Colo., pp. 93-121.
Sciences and Geomechanics Abstract, 20: 227-236. Ramachandran, K., and Hosking, I.A. 1985. Reliability approach
Ghosh, A., and Haupt, W. 1989. Computation of the seismic to stability analysis of soil/rock slopes. Proceedings, 5th Inter-
stability of rock wedges. Rock Mechanics and Rock Engineer- national Conference on Numerical Methods in Geomechanics,
ing, 22: 109-125. Nagoya, Japan, pp. 1019-1028.
Gibbs, B.L. 1989. Directory of mining programs. Gibbs Associates, Savely, J.P. 1985. An application of probability of rock slope
Boulder, Colo. design. In Rock masses: modeling of underground openings/
Glass, C.E., Savely, J.P., and Call, R.D. 1978. Determining seismic probability of slope failure/fracture of intact rock. Edited by
risk for economic optimum slope design. Proceedings, 19th U.S. C.D. Dowding. Proceedings of Geotechnical Engineering Divi-
Symposium on Rock Mechanics, Lake Tahoe, Nev., pp. 89-95. sion, American Society of Civil Engineers, Denver, Colo.,
Glynn, E.F., and Einstein, H.H. 1979. Probability of kinematic pp. 73-91.
instability in rock slopes-a numerical approach. Proceedings, Scavia, C., Barla, G., and Bernaudo, V. 1990. Probabilistic stability
20th U.S. Symposium on Rock Mechanics, Austin, Tex., analysis of block toppling failure in rock slopes. International
pp. 317-325. Journal of Rock Mechanics and Mining Sciences and Geo-
Glynn, E.F., and Ghosh, S. 1982. Effect of correlation on rock mechanics Abstracts, 27: 465-478.
slope stability analyses. Proceedings, 23rd U.S. Symposium on Tao, Z.Y., and Hong, W. 1987. The effects of joint cohesion and
Rock Mechanics, Berkeley, Calif., pp. 95-115. water saturation on rock slope stability. Reliability and Risk
Goodman, R.E. 1976. Methods of geological engineering in discon- Analysis in Civil Engineering 2, Proceedings, 5th International
tinuous rocks. West Publishing Company, St. Paul, Minn. Conference on Applications of Statistics and Probability in Soil
Goodman, R.E., and Bray, J.W. 1976. Toppling of rock slopes. and Structural Engineering, Vancouver, B.C., vol. 2, pp. 791-796.
Proceedings of a Specialty Conference on Rock Engineering for Terzaghi, R.D. 1965. Sources of error in joint surveys. GCotechnique,
Foundations and Slopes, American Society of Civil Engineers, 15: 287-304.
Boulder, Colo., vol. 2, pp. 201-234. Varnes, D.J. 1978. Slope movement types and processes. In Land-
Herget, G. 1978. Analysis of discontinuity orientation for a prob- slides: analysis and control. Special Report 176, Transportation
abilistic slope stability design. Proceedings, 19th U.S. Sym- Research Board, National Academy of Sciences, pp. 11-33.
posium on Rock Mechanics, Lake Tahoe, Nev., pp. 42-50. Wathugala, D.N., Kulatilake, P.H.S. W., Wathugala, G. W., and
Hoek, E., and Bray, J.W. 1977. Rock slope engineering. 2nd ed. Stephansson, 0. 1990. A general procedure to correct sampling
The Institution of Mining and Metallurgy, London. bias on joint orientation using a vector approach. Computers
ISRM (International Society of Rock Mechanics) 1988. List of com- and Geotechnics, 10: 1-31.
puter programs in rock mechanics. International Journal of Rock Young, D.S., and Hoerger, S.F. 1988. Geostatistics applications
Mechanics Mining Sciences and Geomechanics Abstracts, 25: to rock mechanics. Proceedings, 28th U.S. Symposium on Rock
183-252. Mechanics, University of Arizona, Tucson, Ariz., pp. 271-282.

View publication stats

Vous aimerez peut-être aussi