You are on page 1of 14

Computers and Geotechnics 34 (2007) 137150

www.elsevier.com/locate/compgeo

Two-dimensional slope stability analysis by limit equilibrium


and strength reduction methods
a,*
Y.M. Cheng , T. Lansivaara b, W.B. Wei a

a
Department of Civil and Structural Engineering, Hong Kong Polytechnic University, Hong Kong
b
Department of Civil Engineering, Tampere University of Technology, Finland

Received 16 February 2006; received in revised form 28 September 2006; accepted 18 October 2006
Available online 27 December 2006

Abstract

In this paper, the factors of safety and the locations of critical failure surfaces obtained by the limit equilibrium method and strength
reduction method are compared for various slopes. For simple homogenous soil slopes, it is found that the results from these two meth-
ods are generally in good agreement except when / 0 is zero. It is also found that the strength reduction method (SRM) is usually not
sensitive to the dilation angle, soil moduli or the solution domain size and the choice of these parameters is not critical in the analysis.
While the SRM may perform well in many cases, two major limitations of this method are found in this study: it is sensitive to the non-
linear solution algorithm for the case of a soft band with frictional material and it is incapable of determining other failure surfaces which
may be only slightly less critical than the SRM solution but still require treatment for good engineering practice.
2006 Elsevier Ltd. All rights reserved.

Keywords: Slope stability; Limit equilibrium method; Strength reduction method; Local minimum

1. Introduction ent methods of analyses are similar so that the assumptions


on the internal force distributions are not major issues for
For slope stability analysis, the limit equilibrium practical use except for some particular cases. Besides the
method (LEM) is widely used by engineers and researchers LEM, limit analysis has also been used for simple problems
and this is a traditional and well established method. [5], but its applications in complicated real problems is still
Although the LEM does not consider the stressstrain rela- limited, so this method is seldom adopted for routine anal-
tion of soil, it can provide an estimate of the factor of ysis and design. Upper bound limit analysis requires the
safety of a slope without the knowledge of the initial con- assumption of a ow rule (associated or non-associated)
ditions with the result that the LEM is favored by many and the application of energy balance on vertical/inclined
engineers. The LEM is well known to be a statically inde- sides between slices and along the failure surfaces. An
terminate problem and assumptions on the distributions energy balance applied along vertical/inclined sides is actu-
of internal forces are required for the solution of the factor ally equivalent to the assignment of a yield criterion to f(x).
of safety. The variational approach to determine the factor While the energy balance equation can be used along the
of safety proposed by Baker and Garber [2] does not failure surfaces without question, there is not a strong
require the assumption on the internal force distribution requirement that energy balance can be applied to the ver-
but is tedious to use even for a single failure surface. Mor- tical/inclined sides between slices. In this respect, limit
genstern [14], among others, has however pointed out that analysis is not better than the LEM in prescribing the inter-
for normal problems, the factors of safety FOS from dier- slice force function and it usually gives FOS slightly greater
than those from the LEMs. Both the LEMs and limit
*
Corresponding author. analysis methods require trial failure surfaces and optimi-
E-mail address: ceymchen@polyu.edu.hk (Y.M. Cheng). zation analysis to locate the critical failure surface, which

0266-352X/$ - see front matter 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compgeo.2006.10.011
138 Y.M. Cheng et al. / Computers and Geotechnics 34 (2007) 137150

is a dicult N-P type global optimization problem. Many denition can be used. The authors have found that these
dierent proposals have been suggested in the past and two denitions will give similar locations of the critical fail-
detailed discussions on various methods for locating the ure surface under most cases and maximum shear strain
critical failure surface have been provided by Cheng [6]. increment is chosen for the present study.
While most of these methods can work for normal prob- In the present study, the limit equilibrium method is
lems, they may be trapped by the presence of local minima considered using the MorgensternPrice method with
in the solution. The use of a modied simulated annealing f(x) = 1.0 (equivalent to Spencer method). Krahn [12]
method by Cheng [6] is one of the few successful methods and Abramson et al. [1] have pointed out that f(x) may
which can escape from a local minimum. It is important be critical for some special cases, but this is generally not
that the global minima from the LEM are used in the pres- the case unless the problem is highly complicated. Besides
ent study when compared with the results from the SRM, Spencers method, the authors have also tried the general-
otherwise the comparisons will not be meaningful. ized limit equilibrium (GLE) method [1] and the results are
The SRM was used for slope stability analysis as early as also similar to those by the MorgensternPrice method.
1975 by Zienkiewicz et al. [21] Later the SRM was applied The dierences of the FOS and the critical failure surfaces
by Naylor [15], Donald and Giam [9], Matsui and San [13], from f(x) = 1.0 and f(x) = sin(x) are also found to be small
Ugai and Leshchinsky [19], Dawson et al. [8], Griths and for both the MorgensternPrice method and the GLE
Lane [10], Song [18], Zheng et al. [20] and others. This tech- method in the present study. Hence, the discussions in
nique has also been adopted in several well-known com- the present paper are not sensitive to the type of LEM.
mercial geotechnical nite element programs. The main In performing the SRM analysis, many soil parameters
advantages of the SRM are as follows: (i) the critical failure and boundary conditions are required to be dened which
surface is found automatically from the shear strain arising are absent in the corresponding LEM analysis. The impor-
from the application of gravity loads and the reduction of tance of the various parameters and the applicability of the
shear strength; (ii) it requires no assumption on the inter- SRM under several special cases are considered in the fol-
slice shear force distribution; (iii) it is applicable to many lowing sections.
complex conditions and can give information such as stres-
ses, movements, and pore pressures which are not possible 2. Stability analysis for a simple and homogeneous soil slope
with the LEM. Griths and Lane [10] pointed out that the
widespread use of the SRM should be seriously considered Firstly, a homogeneous soil slope with a slope height
by geotechnical practitioners as a powerful alternative to equal to 6 m and slope angle equal to 45 (Fig. 1) is consid-
the traditional limit equilibrium methods. One of the main ered. For the three cases in which the friction angle is 0,
disadvantages of the SRM is the long solution time since the critical slip surface is a deep-seated surface with
required to set up the computer model and to perform a large horizontal extent, the models are larger than the
the analysis. With the development of computer hardware one as shown in Fig. 1 and have a width of 40 m and a
and software, the SRM can now be performed within a rea- height of 16 m. In the parametric study, dierent shear
sonable time span suitable for routine analysis and design. strength properties are used and the LEM, SRM1 and
Other limitations of the SRM include the choice of an SRM2 analyses are carried out. The cohesive strength c 0
appropriate constitutive model and parameters, boundary of the soil varies from 2, 5 and 1020 kPa while the friction
conditions and the denition of the failure condition/fail- angle / 0 varies from 5, 15, 25 and 3545. The density,
ure surface which will be considered in the present paper. elastic modulus and Poisson ratio of the soil are kept at
Many researchers have compared the results between 20 kN/m3, 14 MPa and 0.3, respectively, in all the analysis.
the SRM and LEM and have found that generally the As shown in Fig. 1, the size of the domain for the SRM
two methods will give similar FOS. Most of the studies analyses is 20 m in width and 10 m in height and there
are, however, limited to homogenous soil slopes and the
geometry of the problems is relatively regular with no spe-
cial features (e.g. the presence of a thin layer of soft mate-
rial or special geometry). Furthermore, there are only
limited studies, which compare the critical failure surfaces
from the LEM and SRM as the FOS appear to be the pri-
mary quantity of interest. In this paper, the two methods
are compared under dierent conditions and both the
FOS and the locations of the critical failure surfaces are
considered in the comparisons. In the present study, both
a non-associated ow rule (SRM1 and dilation angle = 0)
and an associated ow rule (SRM2 and dilation angle =
friction angle) are applied in the SRM analyses. To dene
the critical failure surface from the SRM, both the maxi-
mum shear strain and the maximum shear strain increment Fig. 1. Discretization of a simple slope model.
Table 1
Factors of safety FOS by LEM and SRM
Case c 0 (kPa) / 0 () FOS (LEM) FOS (SRM1, FOS (SRM2, FOS dierence with FOS dierence FOS dierence between
non-associated) associated) LEM (SRM1, %) with LEM (SRM2, %) SRM1 and SRM2

Y.M. Cheng et al. / Computers and Geotechnics 34 (2007) 137150


1 2 5 0.25 0.25 0.26 0 4.0 4.0
2 2 15 0.50 0.51 0.52 2 4.0 2.0
3 2 25 0.74 0.77 0.78 4.0 5.4 1.3
4 2 35 1.01 1.07 1.07 5.9 5.9 0
5 2 45 1.35 1.42 1.44 5.2 6.7 1.4
6 5 5 0.41 0.43 0.43 4.9 4.9 0
7 5 15 0.70 0.73 0.73 4.3 4.3 0
8 5 25 0.98 1.03 1.03 5.1 5.1 0
9 5 35 1.28 1.34 1.35 4.7 5.5 0.7
10 5 45 1.65 1.68 1.74 1.8 5.5 3.6
11 10 5 0.65 0.69 0.69 6.2 6.2 0
12 10 15 0.98 1.04 1.04 6.1 6.1 0
13 10 25 1.30 1.36 1.37 4.6 5.4 0.7
14 10 35 1.63 1.69 1.71 3.7 4.9 1.2
15 10 45 2.04 2.05 2.15 0.5 5.4 4.9
16 20 5 1.06 1.20 1.20 13.2 13.2 0
17 20 15 1.48 1.59 1.59 7.4 7.4 0
18 20 25 1.85 1.95 1.96 5.4 5.9 0.5
19 20 35 2.24 2.28 2.35 1.8 4.9 3.1
20 20 45 2.69 2.67 2.83 0.7 5.2 6.0
21 5 0 0.20 0.23 15
22 10 0 0.40 0.45 12.5
23 20 0 0.80 0.91 13.8

139
140 Y.M. Cheng et al. / Computers and Geotechnics 34 (2007) 137150

are 3520 zones and 7302 grid points in the mesh for analy- and force equilibrium, is adopted and the critical failure
sis. Based on limited mesh renement studies, it was found surface is evaluated by the modied simulated annealing
that the discretization as shown in Fig. 1 is suciently good technique as proposed by Cheng [6]. The tolerance for
so that the results of analyses are practically insensitive to a locating the critical failure surface by the simulated anneal-
further reduction in the element size. For the LEM, the ing method is 0.0001, which is suciently accurate for the
MorgensternPrice method, which satises both moment present study.

Fig. 2. Slip surface comparison with increasing friction angle (c 0 = 2 kPa): (a) c 0 = 2 kPa, / = 5; (b) c 0 = 2 kPa, / = 45.

Fig. 3. Slip surface comparison with increasing cohesion (/ = 5): (a) c 0 = 2 kPa, / = 5; (b) c 0 = 20 kPa, / = 5.

Fig. 4. Slip surface comparison with increasing cohesion (/ = 35): (a) c 0 = 2 kPa, / = 35; (b) c 0 = 20 kPa, / = 35.
Y.M. Cheng et al. / Computers and Geotechnics 34 (2007) 137150 141

From Table 1 and Figs. 2 and 3, it is found that the FOS those from LEM than those obtained from SRM1.
and critical failure surfaces determined by the SRM and The critical failure surfaces from SRM1, SRM2 and
LEM are very similar under dierent combinations of soil LEM are practically the same when the cohesive
parameters for most cases except when / 0 = 0. When the strength is small (it is dicult to dierentiate clearly in
friction angle is greater than 0, most of the FOS obtained Figs. 2a,b, 3a, 4a), but noticeable dierences in the crit-
by the SRM dier by less than 7.4% with respect to the ical failure surfaces are found when the cohesive
LEM results except for case 16 (c 0 = 20 kPa, / 0 = 5) strength is high (Figs. 3b, 4b, 5a,b).
where the dierence is up to 13.2%. When the friction angle 5. The right end of the failure surface moves closer to the
is very small or zero, there are relatively major dierences crest of the slope as the friction angle of the soil is
between the SRM and LEM for both the FOS and the crit- increased (which is a well known result). This behaviour
ical slip surface (Table 1 and Fig. 5). The dierences in the is more obvious for those failure surfaces obtained from
FOS between the LEM and SRM reported by Glamen SRM1. For example, for the ve cases where the cohe-
et al. [16] are greater than those found in the present study. sion of the soil is 2 kPa (Fig. 2), when the friction angle
The authors suspect that this is due to the manual location is 5, 15 and 25, the right end-point of the failure sur-
of critical failure surfaces by Glamen et al. [16], as opposed face derived from SRM1 is located to the right of the
to the global optimization method used here. Based on right end-point of the critical failure surface obtained
Table 1 and Figs. 2 and 3, some conclusions can be made from the LEM. When the friction angle is 35, the right
as follows:

1. Most of the FOS obtained from the SRM are slightly 20,15 28,15
larger than those obtained from the LEM with only
few exceptions.
Soil1
2. The FOS from an associated ow rule (SRM2) are 28,10
slightly greater than those from a non-associated ow 28,9.5
8,8 8,7.5
(SRM1), and this dierence increases with increasing Soil2
friction angle. These results are reasonable and are 0,5
8,7.1
5,5
expected. The dierences between the two set of results y Soil3
5,4.5
are, however, small because the problem has a low level
of kinematic constraint.
3. When the cohesive strength of the soil is small, the dif- 0,0
28,0
x
ferences in FOS between the LEM and SRM (SRM1
and SRM2) are greatest for higher friction angles. When Fig. 6. A slope with a thin soft band.
the cohesion of the soil is large, the dierences in FOS
are greatest for lower friction angles. This result is some- Table 2
what dierent from that of Dawson et al. [7], who con- Soil properties for Fig. 6
cluded that the dierences are greatest for higher friction Soil Cohesion Friction Density Elastic Poisson
angles when the results between SRM and limit analysis name (kPa) angle () (kN/m3) modulus ratio
are compared. (MPa)
4. The failure surfaces from the LEM, SRM1 and SRM2 Soil1 20 35 19 14 0.3
are similar in most cases. In particular, the critical fail- Soil2 0 25 19 14 0.3
ure surfaces obtained by SRM2 appear to be closer to Soil3 10 35 19 14 0.3

Fig. 5. Slip surface comparison with increasing cohesion (/ = 0): (a) c 0 = 5 kPa, / = 0; (b) c 0 = 20 kPa, / = 0.
142 Y.M. Cheng et al. / Computers and Geotechnics 34 (2007) 137150

end-point of the failure surface obtained by the SRM1 3. Stability analysis of a slope with a soft band
and LEM are nearly at the same location. When the fric-
tion angle is 45, the distance of the right end-point A special problem with a soft band has been constructed
derived from SRM1 is located to the left of the right by the present authors as it appears that similar problems
end-point derived from LEM.
6. For SRM analyses, when the friction angle of soil is Table 3A
small, the dierences between the slip surfaces for FOS by SRM from dierent programs when c 0 for soft band is 0
SRM1 and SRM2 are greatest for smaller cohesion Program/FOS 12 m domain 20 m domain 28 m domain
(Fig. 3). When the friction angle is large, the dierences Flac3D 1.03/1.03 1.30/1.28 1.64/1.61
between the slip surface for SRM1 and SRM2 are great- Phase 0.77/0.85 0.84/1.06 0.87/1.37
est for higher cohesion (Fig. 4). Plaxis 0.82/0.94 0.85/0.97 0.86/0.97
7. It can also be deduced from Figs. 25 that the potential Flac2D No solution No solution No solution
failure volume of the slope becomes smaller with The values in each cell are based on SRM1 and SRM2, respectively. (min.
increasing friction angle but increases with increasing FOS = 0.927 from MorgensternPrice analysis).
cohesion. This is also well known behaviour, as when
the cohesive strength is high the critical failure surface Table 3B
will be deeper. FOS by SRM from dierent programs when / 0 = 0 and c 0 = 10 kPa for
soft band
Although there are some minor dierences in the results Program/FOS 28 m domain
between the SRM and LEM in this example, the results Flac3D 1.06/1.06
from these two methods are generally in good agreement Phase 0.99/1.0
which suggests that the use of either the LEM or SRM is Plaxis 1.0/1.03
Flac2D No solution
satisfactory in general. The authors have, however, con-
structed an interesting case where the limitations of the The values in each cell are based on SRM1 and SRM2, respectively (min.
FOS = 1.03 from the MorgensternPrice analysis).
SRM are demonstrated.

Fig. 7. Mesh plot of the three numerical models with a soft band: (a) numerical simulation model 1; (b) numerical simulation model 1; (c) numerical
simulation model 1.
Y.M. Cheng et al. / Computers and Geotechnics 34 (2007) 137150 143

have not been considered previously. The geometry of the tried several well known commercial programs and
slope is shown in Fig. 6 and the soil properties are shown obtained very surprising results. The locations of the criti-
in Table 2. It is noted that c 0 is zero and / 0 is small for soil cal failure surfaces from SRM for solution domain width
layer 2, which has a thickness of just 0.5 m. The critical fail- of 12 m, 20 m and 28 m are virtually the same. The local
ure surface is obviously controlled by this soft band, and failures from the SRM, as shown in Fig. 8b, range from
slope failures in similar conditions have actually occurred x = 5 m to x = 8 m and the failure surfaces are virtually
in Hong Kong (for example, the Fei Tsui Road slope fail- the same for the three dierent solution domains. The
ure in Hong Kong). majorities of the critical failure surfaces lie within layer 2,
In order to consider the size eect (boundary eect) in which has low shear strength, and are far from the right
the SRM, three dierent numerical models are developed boundary. It is surprising to nd that dierent programs
to perform the SRM using MohrCoulomb analysis and produce drastically dierent results (Table 3A) for the
the widths of the domains are 28 m, 20 m and 12 m, respec- FOS even though the locations of the critical failure surface
tively (Fig. 7). In these three SRM models, various mesh from these programs are very similar. For the cases as
sizes were tried until the results were insensitive to the num- shown in Fig. 1, and other cases in a latter part of this
ber of elements used for the analysis. For example, when paper, the results are practically insensitive to the domain
the domain size is 28 m, the FOS was found to be 1.37 size, while the cases shown in Fig. 6 are very sensitive to
(Tables 3A and 3B) with 12,000 elements, 1.61 with 6000 the size of domain for the programs Flac3D (SRM1 and
elements and 1.77 with 3000 elements using SRM1 analysis SRM2) and Phase (SRM2). Results from the Plaxis pro-
and the program phase. gram appear to not be sensitive to the domain size but
Since the FOS for this special problem have great dier- are quite sensitive to the dilation angle (which is dierent
ences from those found using the LEM, the present authors from the previous example). The SRM1 results from

s oil1

14

12

10 s oil2
s oil3

0
0 5 10 15 20 25

Fig. 8. Critical failure surfaces from LEM and SRM for frictional soft band problem: (a) critical solution from LEM when soft band is frictional material
(FOS = 0.927); (b) critical solution from SRM for 12 m width domain.
144 Y.M. Cheng et al. / Computers and Geotechnics 34 (2007) 137150

program Phase are also not sensitive to the domain size for Table 4
SRM1, but results from SRM2 behave dierently. The FOS with non-associated ow rule for 12 m domain
FOS from Flac3D appear to be overestimated when the Element Tolerance Maximum number of FOS
soil parameters for the soft band are low, but the results number (stress iterations
analysis)
from this program are not sensitive to the dilation angle
which is similar to all the other examples in the present 1500 0.001 100 0.80
2000 0.001 100 No
study. For SRM1, the results from Phase and Plaxis appear result
to be more reasonable as the results are not sensitive to the 2000 0.003 100 No
domain sizes, while for SRM2, the authors take the view result
that results from Plaxis may be better. It is also surprising 2000 0.004 100 No
to nd that Flac2D cannot give any result for this problem, result
2000 0.005 100 No
even after many dierent trials, but the program worked result
properly for all the other examples in this study. 2000 0.008 100 0.81
There is another interesting and important issue when 2000 0.01 100 0.82
SRM is adopted for the present problems. For the problem 2000 0.001 500 0.74
with a 12 m domain, Phase cannot provide a result with the 2000 0.003 500 0.77
2000 0.004 500 0.77
default settings and the authors have varied these (includ- 2000 0.005 500 0.79
ing the tolerance and number of iterations allowed) until 3000 0.001 100 No
convergence is achieved. The results of analysis for a result
12 m domain with Phase are shown in Tables 4 and 5. It 3000 0.003 100 0.79
is observed that the number of elements used for the anal- 3000 0.004 100 0.8
3000 0.005 100 0.8
ysis has a very signicant eect on the factor of safety, 3000 0.01 100 0.84
which is not observed for the cases in Table 1. The toler- 3000 0.001 500 0.77
ance used in the nonlinear equation solution also has a
major impact on the results for this case. This is less obvi-
ous for other cases considered in the present study. of major dierences in the soil parameters is the actual
Besides the special results shown above, the FOS from cause for the diculties in the SRM analysis. Great care
the 28 m domain analysis appear to be large for Flac3D is required in the implementation of a robust nonlinear
and Phase when the strength parameters for the soil layer equation solver for the SRM.
2 are low. In fact, it is not easy to dene an appropriate fac-
tor of safety from SRM analysis for this problem. If the Table 5
cohesive strength of the top soil is reduced to zero, the fac- FOS with associated ow rule for 12 m domain
tor of safety can be estimated as 0.57 from the relation Element Tolerance Maximum number of FOS
tan //tan h, where h is the slope angle. It can be seen that number (stress iterations
for the LEM, the cohesive strength 20 kPa for soil 1 helps analysis)
to bring the factor of safety to 0.927 and a high factor of 1000 0.001 100 1.03
safety for this problem is not reasonable. Without the 1200 0.001 100 1.0
1500 0.001 100 No
results from the LEM for comparison, it may be unconser- result
vative to adopt the values of 1.64 (1.61) from the SRM 1500 0.003 100 No
based on Flac3D. result
When the soil properties of the soft band are changed to 1500 0.004 100 1
c 0 = 10 kPa and / 0 = 0, the results of analyses are as shown 1500 0.005 100 1.39
1500 0.01 100 2.09
in Table 3B. It is found that the critical failure will extend 1500 0.001 500 0.86
to a much greater distance so that a 28 m wide domain is 1500 0.003 500 0.98
necessary. The FOS from the dierent programs are 3000 0.001 100 No
virtually the same, which is drastically dierent from the result
results in Table 3A (the same meshes were used for Tables 3000 0.003 100 No
result
3A and 3B). 3000 0.004 100 No
If the soil properties of soils 2 and 3 are interchanged so result
that the third layer of soil is the weak soil, the FOS from 3000 0.005 100 No
SRM2 are 1.33 (with all programs) for all three dierent result
domain sizes. The corresponding factor of safety from 3000 0.01 100 No
result
the LEM is 1.29 from the MorgensternPrice analysis. 3000 0.001 500 0.85
The locations of the critical failure surface from the SRM 3000 0.003 500 0.89
and LEM for this case are also very close, except for the 3000 0.004 500 0.9
initial portion, as shown in Figs. 9a and b. It appears that 3000 0.005 500 1.5
the presence of a soft band with frictional material, instead 3000 0.01 500 2.09
Y.M. Cheng et al. / Computers and Geotechnics 34 (2007) 137150 145

The problems as shown in Table 3A.A may reect the past, but this actually a very important issue which is illus-
limitations of commercial programs rather than the limita- trated by the following examples. In the SRM, there is no
tions of the SRM, but they illustrate that it is not easy to local minimum as the formation of the shear band will
compute a reliable FOS for this type of problem using attract strain localization in the solution process. To inves-
the SRM. The results are highly sensitive to dierent non- tigate this issue, an 11 m height slope as shown in Fig. 10 is
linear solution algorithms which are not clearly explained considered. The slope angle for the lower part of the slope
in the commercial programs. Great care, eort and time is 45 while the slope angle for the upper part of the slope is
are required to achieve a reasonable result from SRM for 26.7. The cohesion and friction angle of the soil are 10 kPa
this special problem and comparisons with the LEM are
necessary. It is not easy to dene a proper factor of safety
from the SRM alone for the present problem as the results
are highly sensitive to the size of domain and the ow rule. 26,21 46,21
In this respect, the LEM appears to be a better approach
1V
for this type of problem. 16,16 Soil property:
2H
Cohesion=10kPa
1V
0,10 10,10 Friction angle=30degree
1H
4. Local minimum in LEM Density=20kN/m3
y Elastic modulus=14MPa
For the LEM, it is well known that many local minima Poisson ratio=0.3
may exist besides the global minimum. This makes it di-
0,0 46,0
cult to locate the critical failure surface by classical optimi-
x
zation methods. Comparisons of the LEM and SRM with
respect to local minima have not been considered in the Fig. 10. Slope geometry and soil property under study.

Fig. 9. Critical solutions from LEM and SRM when the bottom soil layer is weak: (a) critical failure surface from LEM when the bottom soil layer is weak
(FOS = 1.29); (b) critical failure surface from SRM2 and 12 m domain (FOS = 1.33).
146 Y.M. Cheng et al. / Computers and Geotechnics 34 (2007) 137150

and 30, respectively, and the density of the soil is form soil. The soil parameters are c 0 = 5 kPa and / 0 = 30
20 kN/m3. while the unit weight is 20 kN/m3. The global minimum
The failure mechanism by the SRM is shown in and local minima are determined in accordance with the
Fig. 11 and the FOS is 1.47 for both non-associated ow procedures of Cheng [6] and dierent boundaries for the
and associated ow. The right end-point of the failure left and right exit ends are specied in the study. Using
surface is located to the right of the crest of the slope. the LEM, the global minimum FOS is obtained as 1.33
The results derived from the LEM are presented in (Fig. 13a) but several local minima are found with
Fig. 12 (number of slices is 50). The global minimum fac- factors of safety in the range 1.381.42 as shown in
tor of safety is 1.383 but a local minimum FOS of 1.3848 Fig. 13be. From the SRM, only the factor of safety
is also found. The location of the failure surface for the 1.327 is found which is similar to the global minimum
local minimum 1.3848 is very close to that from the shown in Fig. 13a. If slope stabilization is only carried
SRM, and the failure surface for the global minimum out for this failure surface, the possible failure surfaces
from the LEM is not the critical failure surface from shown by Fig. 13d and e will not be considered. Baker
the SRM. Since the FOS for the two critical failure sur- and Leshchinsky [3] have proposed the concept of the
faces from the LEM are so close, both failure surfaces safety map, which enables the global minimum and local
are probable failure surfaces should be considered in minima from the LEM to be visualized easily, but it
slope stabilization. For the SRM, there is only one appears that construction of such a map using the SRM
unique failure surface from the analysis and another pos- is impossible. In this respect, the LEM is a better tool for
sible failure mechanism cannot be easily determined. slope stability analysis. It is possible that the use of
Thus, the SRM analysis may yield a local failure surface the SRM may miss the location of the next critical failure
of less importance while a more severe global failure sur- surface (with a very small dierence in the FOS but a
face remains undetected, as illustrated in the next exam- major dierence in the location of the critical failure sur-
ple. This is clearly a major drawback of the SRM as face) so that the slope stabilization measures may not be
compared with the LEM. adequate. This interesting case has illustrated a major lim-
The present authors have also constructed another itation of the SRM for the design of slope stabilization
interesting case which is worth discussion. Fig. 13 shows works.
a relatively simple slope with a total height of 55 m in a uni-

5. Inuence of elastic modulus for SRM analysis

In order to investigate the inuence of the soil elastic


modulus in slope stability analysis when the SRM is
used, the four dierent models shown in Figs. 1417
are considered. In all these models, the strength parame-
ters are the same for every soil so that there is actually
only one soil in the LEM analysis. For the SRM analy-
ses, dierent elastic moduli are used for dierent soils
and the results of the analyses, shown in Tables 68,
indicate that they are relatively insensitive to the elastic
modulus of soil. When the elastic moduli of soils 1
Fig. 11. Result derived by SRM. and 2 are interchanged, there is only a minor change

Fig. 12. Global and local minima by LEM.


Y.M. Cheng et al. / Computers and Geotechnics 34 (2007) 137150 147

in the factor of safety. In fact, the eect of ow rule is Results of a model with two dierent soils and a vertical
slightly more important than the elastic moduli in the boundary.
present example. In general, the dierences in both the Results of a model with two dierent soils and an
FOS and the locations of the critical failure surfaces inclined boundary.
are small between the LEM and SRM. For case 2 in Results of a model with three dierent soils and inclined
Table 9, the dierence in the FOS between the LEM boundary.
and SRM is signicant but the locations of the critical
failure surfaces from the two methods (Fig. 18) are close 6. Discussion and conclusions
to each other.
Results of a model with two soils and a horizontal In the present study, a number of interesting features of
boundary. the SRM were highlighted which are important for a

30
silt
30
25
silt
25
20
20
15
15
10
10

5 5

0 0
0 5 10 15 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55

30 30

silt silt
25 25

20 20

15 15

10 10

5 5

0 0
0 5 10 15 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55

30
silt
25

20

15

10

0
0 5 10 15 20 25 30 35 40 45 50 55

Fig. 13. Local minima from LEM and critical solution from SRM for a simple slope: (a) FOS = 1.33 (global minimum); (b) FOS = 1.375 (local
minimum); (c) FOS = 1.415 (local minimum); (d) FOS = 1.40 (local minimum); (e) FOS = 1.383 (local minimum); (f) FOS = 1.327 from SRM.
148 Y.M. Cheng et al. / Computers and Geotechnics 34 (2007) 137150

11,11 21,11 29,17

Soil1
8,8
21,8 11,11 Soil1
11,10
Soil2
5,5 11,9
0,5 Soil2 29,6
0,5 5,5
5,4 29,5
Y Y 0,4
0,3 Soil3
5,3
0,0 21,0 0,0 29,0
X X
Fig. 14. Slope geometry with a horizontal boundary. Fig. 17. Slope geometry with three soils and an inclined boundary.

11,11 21,11
choice of the dilation angle (which is important for the
adoption of the SRM in slope stability analysis). When
8,8 an associated ow rule is assumed, the critical slip surfaces
from SRM2 appear to be closer to those from the LEM
than those from SRM1.
5,5
0,5 For the SRM, the eects of the dilation angle, the toler-
Soil1 Soil2
ance for nonlinear equation analysis, the soil moduli and
Y the domain size (boundary eects) are usually small but
still noticeable. In most cases, these factors cause dier-
0,0
8,0 21,0 ences of just a few percent and are not critical for engineer-
X ing use of the SRM. Since the use of dierent LEM
Fig. 15. Slope geometry with a vertical boundary. methods will also give dierences in the factor of safety
of several percents, the LEM and SRM can be viewed as
similar in performance for normal cases.
Drastically dierent results are determined from dier-
11,11 21,11
ent computer programs for the problem with a soft band.
Soil1 For this special case, the factor of safety is very sensitive
11,10
8,8 to the size of the elements, the tolerance of the analysis
21,8
and the number of iteration allowed. It is strongly sug-
5,5
gested that the LEM be used to check the results from
0,5 the SRM. This is because the SRM is highly sensitive to
0,4 5,4 the nonlinear solution algorithms and ow rule for this spe-
Soil2
Y cial type of problem. Even though results from the pro-
gram Plaxis (and some results from the program Phase)
0,0 21,0 are insensitive to the domain size, they are very sensitive
X to the ow rule which is a result not found in all the other
Fig. 16. Slope geometry with an inclined boundary. examples. The SRM has to be used with great care for
problems with a soft band of this nature.
The two examples with local minima for the LEM illus-
proper analysis of a slope. While most research has concen- trate another limitation of the SRM in engineering use.
trated on the FOS between the LEM and SRM, the present With the SRM, there is strain localization during the solu-
works have compared the locations of the critical failure tion and the formation of local minima is unlikely. In the
surfaces from these two methods. In a simple and homog- LEM, the presence of local minima is a common phenom-
enous soil slope, the dierences in the FOS and locations of enon, and this is a major dierence between the two meth-
the critical failure surfaces from the SRM and LEM are ods. Thus, it is suggested that the LEM should be
small and both methods are satisfactory for engineering preformed in conjunction with the SRM as a routine check.
use. It is found that when the cohesion of the soil is small, Through the present study, two limitations of the SRM
the dierence in FOS from the two methods is greatest for have been established: (1) it is sensitive to nonlinear solu-
higher friction angles. When the cohesion of the soil is tion algorithms/ow rule for some special cases and (2) it
large, the dierence in FOS is greatest for lower friction is unable to determine other failure surfaces which may
angles. With regard to the ow rule, the FOS and locations be only slightly less critical than the SRM solution but still
of the critical failure surface are not greatly aected by the require treatment for good engineering practice. If the
Y.M. Cheng et al. / Computers and Geotechnics 34 (2007) 137150 149

Table 6
Comparison of factor of safety for Fig. 14
Case Cohesion / Density Poisson Elastic modulus of Elastic modulus of FOS FOS FOS
(kPa) () (kN/m3) ratio soil1 (kN/m2) soil2 (kN/m2) (LEM) (SRM1) (SRM2)
1 10 15 20 0.3 1.4e5 1.4e2 0.9826 0.954 0.983
2 10 15 20 0.3 1.4e2 1.4e5 0.9826 0.966 0.989
3 10 25 20 0.3 1.4e5 1.4e2 1.2951 1.235 1.297
4 10 25 20 0.3 1.4e2 1.4e5 1.2951 1.259 1.320

Table 7
Comparison of factor of safety for Fig. 15
Case Cohesion / Density Poisson Elastic modulus of Elastic modulus of FOS FOS FOS
(kPa) () (kN/m3) ratio soil1 (kN/m2) soil2 (kN/m2) (LEM) (SRM1) (SRM2)
1 10 15 20 0.3 1.4e5 1.4e2 0.9826 0.964 0.985
2 10 15 20 0.3 1.4e2 1.4e5 0.9826 0.963 0.976
3 10 25 20 0.3 1.4e5 1.4e2 1.2951 1.288 1.304
4 10 25 20 0.3 1.4e2 1.4e5 1.2951 1.282 1.273

Table 8
Comparison of factor of safety for Fig. 16
Case Cohesion / Density Poisson Elastic modulus of Elastic modulus of FOS FOS FOS
(kPa) () (kN/m3) ratio soil1 (kN/m2) soil2 (kN/m2) (LEM) (SRM1) (SRM2)
1 10 15 20 0.3 1.4e5 1.4e2 0.9826 0.951 0.980
2 10 15 20 0.3 1.4e2 1.4e5 0.9826 0.975 0.980
3 10 25 20 0.3 1.4e5 1.4e2 1.2951 1.240 1.292
4 10 25 20 0.3 1.4e2 1.4e5 1.2951 1.240 1.275

ing the SRM to a slope with a soft band. When the soft
band is very thin, the number of elements required to
achieve a good solution is extremely large so that very sig-
nicant computer memory and time are required. Cheng [6]
has tried a slope with a 1 mm soft band and has eectively
obtained the global minimum factor of safety by the simu-
lated annealing method. If the SRM is used for a problem
with a 1 mm thick soft band, it is extremely dicult to
dene a mesh with a good aspect ratio unless the number
of elements is huge. For the SRM with a 500 mm thick soft
band, about one hour of CPU time for a small problem
(several thousand elements) and several hours for a large
problem (over ten thousand elements) were required for
Fig. 18. Critical slip surface of case2 in Fig. 17 based on SRM and LEM.
the Phase program, while the program Flac3D required
13 days (for small to large meshes). If a problem with a
SRM is used for routine analysis and design of slope stabil- 1 mm thick soft band is to be modeled with the SRM,
ization measures, these two major limitations have to be the computer time and memory required will be huge and
overcome and it is suggested that the LEM should be car- the method is not applicable for this special case. The
ried out as a cross-reference. If there are great dierences LEM is perhaps better than the SRM for these cases.
between the results from the SRM and LEM, great care For the SRM, there are further limitations which are
and engineering judgment should be exercised in assessing worth observing. Shukha and Baker [17] have found that
a proper solution. There is one practical problem in apply- there are minor but noticeable dierences in the factors of
Table 9
Comparison of factor of safety for Fig. 17
Case Cohesion / Density Poisson Elastic modulus of Elastic modulus of Elastic modulus of FOS FOS FOS
(kPa) () (kN/m3) ratio soil1 (kN/m2) soil2 (kN/m2) soil3 (kN/m2) (LEM) (SRM1) (SRM2)
1 10 15 20 0.3 1.4e5 1.4e2 1.4e5 0.9400 0.932 0.951
2 10 15 20 0.3 1.4e2 1.4e5 1.4e2 0.9400 0.903 0.947
3 10 25 20 0.3 1.4e5 1.4e2 1.4e5 1.2611 1.210 1.275
4 10 25 20 0.3 1.4e2 1.4e5 1.4e2 1.2611 1.211 1.268
150 Y.M. Cheng et al. / Computers and Geotechnics 34 (2007) 137150

safety from Flac using square elements and distorted ele- [3] Baker R, Leshchinsky D. Spatial distribution of safety factors. J
ments. The use of distorted elements are however unavoid- Geotech Geoenviron Eng 2001;127(2):13545.
[5] Chen WF. Limit analysis and soil plasticity. USA: Elsevier; 1975.
able in many cases. Furthermore, when both the soil [6] Cheng YM. Locations of critical failure surface and some fur-
parameters c 0 and / 0 are very small, it is well known that ther studies on slope stability analysis. Comput Geotech 2003;30:
there are numerical problems with the SRM. The failure 25567.
surface in this case will be deep and wide and a large domain [7] Dawson E, Motamed F, Nesarajah S, Roth, W. Geotechnical stability
is required for analysis. It has been found that the solution analysis by strength reduction. In: Slope stability 2000, Proceedings of
sessions of Geo-Denver 2000, August 58, 2000, Denver, Colorado;
time is extremely long and a well-dened critical failure sur- 2000. p. 99113.
face is not well established from the SRM. For the LEM, [8] Dawson EM, Roth WH, Drescher A. Slope stability analysis by
there is no major diculty in estimating a factor of safety strength reduction. Geotechnique 1999;49(6):83540.
and the critical failure surface under these circumstances. [9] Donald IB, Giam, SK. Application of the nodal displacement method
The advantage of the SRM is the automatic location of to slope stability analysis. In: Proceedings of the fth AustraliaNew
Zealand conference on geomechanics, Sydney, Australia, 1988; p.
the critical failure surface without the need for a trial and 45660.
error search. With the use of modern global optimization [10] Griths DV, Lane PA. Slope stability analysis by nite elements.
techniques, the location of critical failure surfaces by a sim- Geotechnique 1999;49(3):387403.
ulated annealing method, a genetic algorithm or other [12] Krahn J. The 2001 R.M. Hardy Lecture; the limits of limit
methods is now possible and trial and error search with equilibrium analysis. Can Geotech J 2003;40(3):64360.
[13] Matsui T, San KC. Finite element slope stability analysis by shear
the LEM is no longer required. While the LEM suers strength reduction technique. Soils Found 1992;32(1):5970.
from the limitation of an interslice shear force assumption, [14] Morgenstern, NR. The evaluation of slope stability a 25 year
the SRM suers from being sensitive to the nonlinear solu- perspective, Stability and performance of slopes and embankments
tion algorithm/ow rule for some special cases. II, Geotechnical Special Publication No. 31, ASCE; 1992.
Griths and Lane [10] have suggested that a non-associ- [15] Naylor DJ. Finite elements and slope stability. Numer. Meth. In:
Geomech., Proceedings Of the NATO Advanced Study Institute,
ated ow rule should be adopted for slope stability analysis. Lisbon, Portugal; 1981. p. 22944.
As the eect of ow rule on the SRM is not negligible in some [16] Saeterbo Glamen MG, Nordal S, Emdal A. Slope stability evalua-
cases, such as those involving a soft band, the ow rule is tions using the nite element method. NGM 2004, XIV Nordic
indeed an issue for a proper slope stability analysis. It can Geotechnical Meeting, vol. 1; 2004. p. A-4961.
be concluded that both the LEM and SRM have their own [17] Shukha R, Baker, R. Mesh geometry eects on slope stability
calculation by FLAC strength reduction method linear and
merits and limitations, and the use of the SRM is not really nonlinear failure criteria. In: Third international conference on
superior to the use of the LEM in routine analysis and FLAC and numerical modeling in Geomechanics, Sudbury, Ontario,
design. Both methods should be viewed as providing an esti- Canada; 2003. p. 10916.
mation of the factor of safety and the probable failure mech- [18] Song E. Finite element analysis of safety factor for soil structures. Chi
anism, but engineers should also appreciate the limitations J Geotech Eng 1997;19(2):17. in Chinese.
[19] Ugai K, Leshchinsky D. Three-dimensional limit equilibrium and
of each method when assessing the results of their analyses. nite element analysis: a comparison of results. Soils Found
1995;35(4):17.
References [20] Zheng YR, Zhao SY, Kong WX, Deng CJ. Geotechnical engineering
limit analysis using nite element method. Rock Soil Mech
[1] Abramson LW, Lee TS, Sharma S, Boyce GM. Slope stability and 2005;26(1):1638.
stabilization methods. second ed. USA: Wiley; 2002. [21] Zienkiewicz OC, Humpheson C, Lewis RW. Associated and non-
[2] Baker R, Garber M. Theoretical analysis of the stability of slopes. associated visco-plasticity and plasticity in soil mechanics. Geotech-
Geotechnique 1978;28(4):395411. nique 1975;25(4):67189.

You might also like