Griffiths, D. V. & Marquez, R. M. (2007). Géotechnique 57, No. 6, 537–546 doi: 10.1680/geot.2007.57.6.

537

Three-dimensional slope stability analysis by elasto-plastic finite

elements
D. V. G R I F F I T H S * a n d R . M . M A R Q U E Z *

Slope stability analysis is one of the oldest applications in L’analyse de la stabilité des versants représente l’une des
geotechnical engineering, yet it remains one of the most plus anciennes applications en ingénierie géotechnique.
active areas of study in both research and practice. The Elle reste pourtant l’une des disciplines d’étude les plus
vast majority of slope stability analyses are performed in actives, en recherche et en pratique. La grande majorité
two dimensions under the assumption of plane strain des analyses de stabilité des versants est réalisée en deux
conditions. Even when two-dimensional (2D) conditions dimensions, en considérant des conditions de déformation
are not appropriate, three-dimensional (3D) analysis is plane. Même lorsque des conditions à deux dimensions
rarely performed. There are a number of reasons for ne sont pas adaptées, il est rare que des analyses à trois
this. The majority of work on this subject strongly sug- dimensions soient effectuées, et ce pour un certain nom-
gests that the 2D factor of safety is conservative (i.e. bre de raisons. La majorité des travaux à ce sujet
lower than the ‘true’ 3D factor of safety). Even when 3D suggère fortement que le coefficient de sécurité 2D est
may be justified on geometric grounds, the available conservateur (c’est-à-dire inférieur au « vrai » coefficient
methods, being often based on extrapolations of 2D de sécurité 3D). Même lorsque des raisons géométriques
‘methods of slices’ to 3D ‘methods of columns’, are justifieraient la 3D, les méthodes disponibles, souvent
complex, involve numerous assumptions, and are not basées sur des extrapolations de « Méthode de coupes »
readily modified to account for realistic boundary condi- 2D en « Méthode de colonnes » 3D, sont complexes,
tions in the third dimension such as sloping abutments. s’appuient sur de nombreuses hypothèses et ne peuvent
The power and versatility of the elasto-plastic finite être aisément modifiées pour tenir compte de conditions
element approach to slope stability analysis in 2D are de frontière réalistes en 3D, telles que des culées incli-
well known, and these advantages are even more attrac- nées. La puissance et la versatilité de l’approche des
tive in 3D. The paper demonstrates some 3D slope éléments finis en comportement élasto-plastique pour
stability analyses by finite elements, placing the results in l’analyse de la stabilité des versants en 2D sont bien
context with 2D solutions and validating the results where connues, et ces avantages sont encore plus prometteurs
possible against alternative methods. en 3D. Cet article présente certaines analyses de stabilité
de versants en 3D en adoptant la méthode des éléments
finis, mettant les résultats en contexte avec les solutions
KEYWORDS: failure; limit equilibrium methods; numerical 2D et les validant contre des méthodes alternatives lors-
modelling; plasticity; pore pressures; slopes qu’il est possible.

INTRODUCTION initially unknown, it is difficult to set up general algorithms

Slope stability analysis remains an active and important area that would search for the critical failure surface, especially
of study for geotechnical engineers, both in practice and in in cases where it may not be spherical.
academia. The vast majority of slope stability analysis is The assumption that 2D analyses lead to conservative
performed in two dimensions, without much consideration factors of safety needs some qualification, however. First,
being given to the appropriateness of that assumption, or the a conservative result will be obtained only if the most
impact that a more realistic three-dimensional (3D) analysis pessimistic section in the 3D problem is selected for 2D
would have. analysis (e.g. Duncan, 1996a). In a slope that contains
There are a number of reasons for this apparent omission. layering and strength variability in the third dimension,
Most importantly, the two-dimensional (2D) factor of safety this ‘most pessimistic’ section may not be intuitively
is generally considered to be conservative (i.e. lower than obvious. Second, as pointed out by Arellano & Stark
the ‘true’ 3D factor of safety), so practitioners are reluctant (2000), the corollary of a conservative 2D slope stability
to invest in the more time-consuming 3D approaches. A analysis is that back-analysis of a failed slope will lead to
further disadvantage of some 3D methods is that, being an unconservative overestimation of the soil shear strength.
based on extrapolations of 2D ‘methods of slices’ to 3D Although Hutchinson & Sarma (1985) and Hungr (1987)
‘methods of columns’ (e.g. Stark & Eid, 1998, Chen et al., have both asserted that the factor of safety in 3D is
2005), they contain many assumptions relating to side forces always greater than in 2D, it cannot be ruled out that an
that are not easily justified, and the methods are not readily unusual combination of soil properties and geometry could
modified to account for realistic geometries and boundary lead to a 3D mechanism that is more critical. Bromhead
conditions in the third dimension such as sloping abutments. (2004) argued that some landslide configurations with
Furthermore, in cases where the critical failure surface is highly variable cross-sections could lead to failure modes
in which the 3D mechanism was the most critical. Other
Manuscript received 8 February 2006; revised manuscript accepted investigators have indicated more critical 3D factors of
5 March 2007. safety (e.g. Chen & Chameau, 1982; Seed et al., 1990),
Discussion on this paper closes on 1 February 2008, for further although this remains a controversial topic (see Table 1
details see p. ii. later for a summary of recent papers).
* Division of Engineering, Colorado School of Mines, Golden, The potential benefits of the elasto-plastic finite element
USA. (FE) approach to slope stability analysis in 2D are well

537
538 GRIFFITHS AND MARQUEZ
Table 1. 3D slope stability methods been significant activity in recent years on 3D analysis
techniques, however, with many of the methods based on
Authors Method extrapolations of 2D analyses. Using a similar format to
Duncan (1996b), Table 1 provides a list of 3D slope stability
Chen et al. (2001) Upper-bound plasticity publications that have appeared in the last five years.
Huang et al. (2002) Limit equilibrium Although direct comparisons with 2D results were not
Chang (2002) Limit equilibrium
Chugh (2003) Finite differences
directly presented in all these papers, the overwhelming
J. Chen et al. (2003) Limit equilibrium majority either stated or implied that 3D analysis gave
Z. Chen et al. (2003) Upper-bound rigid FE higher factors of safety than 2D analysis, provided the most
Farzaneh & Askari (2003) Upper-bound analysis critical cross-section was selected for the 2D analysis.
Xie et al. (2003) Limit equilibrium
Bromhead & Martin (2004) Limit equilibrium
Sainak (2004) Finite elements
Loehr et al. (2004) Limit equilibrium BRIEF REVIEW OF FINITE ELEMENTS IN SLOPE
Jiang & Yamagami (2004) Spencer’s method STABILITY
Chen et al. (2005) Upper-bound rigid FE The elasto-plastic FE method has been shown to be a
Zhang et al. (2005a, 2005b) Extended Janbu powerful alternative to conventional slope stability analysis
Chang (2005) Sliding blocks techniques (e.g. Smith & Hobbs, 1974; Zienkiewicz et al.,
Silvestri (2006) Analytical 1975; Griffiths, 1980). The first published FE slope stability
Xie et al. (2006) GIS methods software was reported in the second edition of the text by
Smith & Griffiths (1988), and increased use of the method
by other researchers over the years (e.g. Kidger, 1990; Potts
et al., 1990; Matsui & San, 1992; Jeremic, 2000; Sainak,
known, and were summarised by Griffiths & Lane (1999) as
2004) has now led to its inclusion in several proprietary
follows:
geotechnical software packages.
(a) No assumption needs to be made in advance about the The FE program used in the present paper employs a 3D
shape or location of the failure surface. Failure occurs analysis of elastic-perfectly plastic soils with a Mohr–Cou-
‘naturally’ through the zones within the soil mass in lomb failure criterion assuming zero dilation. Although any
which the soil shear strength is unable to sustain the 3D FE could be used in principle, the current work utilises
gravitationally generated shear stresses. 20-node hexahedral elements with ‘reduced integration’
(b) Since there is no concept of slices or columns in the (eight Gauss points per element: see e.g. Zienkiewicz, 1977;
FE approach, there is no need for assumptions about Hughes, 1987) in the stiffness matrix generation and stress
side forces and the consequent implications for local redistribution phases of the algorithm. This element was
and global equilibrium. The finite element method chosen because it is the 3D counterpart of the 8-node plane
preserves global equilibrium until ‘failure’ is reached. element used successfully by the authors and other investiga-
(c) If realistic soil compressibility data are available, the tors in the past to model 2D collapse problems.
FE solutions will give information about deformations It is well documented that the factor of safety of slopes
at working stress levels. assuming elastic-perfectly plastic constitutive models is in-
(d ) The FE method is able to monitor progressive failure sensitive to the construction sequence (e.g. Smith & Grif-
up to and including overall shear failure in, for fiths, 2004). In the current study, therefore, stresses are
example, an analysis involving sequential construction applied to an initially weightless FE mesh through the
of an excavation or embankment. generation of gravity loads that are applied in a single
increment. The stresses developed from the addition of
Thanks to the remarkable increase in computational power gravity are then compared with the Mohr–Coulomb failure
and falling costs in recent years, meaningful 3D analysis can criterion. If the resulting stresses at a particular Gauss point
now be performed on a conventional desktop or laptop lie within the Mohr–Coulomb failure envelope then that
computer. A free 3D FE slope stability analysis program is point is assumed to remain elastic. Alternatively, if the
described in detail in the text by Smith & Griffiths (2004), stresses lie outside the failure envelope, yielding of that
and can be downloaded from the web.y point has occurred, and the non-linear parts of the algorithm
While 2D and 3D slope stability analyses are not expected are activated. The resulting stresses in the yielding regions
to give significantly different results in many cases, the are redistributed to neighbouring elements that still have
availability of accurate and inexpensive software for 3D reserves of strength, using a viscoplastic algorithm (e.g.
analysis makes the use of 2D approaches for all cases harder Perzyna, 1966; Zienkiewicz & Cormeau, 1974). The algo-
to defend. Perhaps the best justification for promoting 3D rithm is iterative, since redistribution of stresses in one
analysis at this stage lies mainly in advancing the state of yielding region may initiate yielding in neighbouring regions
the art. Three-dimensional analysis is more realistic in being that were initially elastic. This iterative process continues
able to account properly for the fixity and geometry of until the formation of a failure mechanism consisting of a
abutments in the third dimension. This leads not only to contiguous zone of soil at failure.
improved accuracy, but also to a better understanding of the
fundamental nature of slope failure mechanisms.
Properties of soil model
To model the soil mass during the FE analysis, the
RECENT ACTIVITY program utilises the six parameters shown in Table 2. The
It is fair to say that, at the time of writing, 3D slope key parameters are the total unit weight ª and the shear
stability analysis is performed so rarely in practice that there strength parameters
9 and c9 (or
u ¼ 0 and cu in un-
is no ‘standard’ method (analogous to Bishop’s method, say) drained analysis). For the examples presented in this paper,
that is widely accepted by geotechnical engineers. There has the elastic parameters Young’s modulus and Poisson’s ratio
were assigned nominal values of 105 kN/m2 and 0:3 respec-
† www.mines.edu/
vgriffit/4th_ed/Software tively, for both drained and undrained analyses, as they have
 SLOPE STABILITY ANALYSIS BY ELASTO-PLASTIC FINITE ELEMENTS 539
SRF
Table 2. Six-parameter soil model 0 0·5 1·0 1·5 2·0
0

9 (
u ) Friction angle
c9 (cu ) Cohesion 0·01
ł Dilation angle Iterations
13 FS ⫽ 1·73
E9 (Eu ) Young’s modulus 0·02
25
9 (u ) Poisson’s ratio 48

Eδmax/γH2
ª Unit weight 0·03 80

0·04 218
676
little influence on the computed factor of safety (e.g. 0·05
Hammah et al., 2005). 1000⫹
In this work, the dilation angle ł was set to zero, 0·06
implying no volume change during yield. The role of the
0·07
dilation angle has been discussed in detail elsewhere (e.g.
Griffiths & Lane, 1999); however, the practical consideration Fig. 1. The rapid increase in the dimensionless displacement
for all geotechnical limit analysis is that this parameter, along with non-convergence signifies slope failure, at which FS
which affects volume change during plastic yielding, has  SRF
relatively little influence on collapse load predictions in
unconfined drained problems.
a distance L/2 (assuming symmetry) in the z direction. Fig.
2(b) shows typical coarser and finer meshes of 20-node
Obtaining a factor of safety elements as used in this study. In both cases, the depth of
In traditional geotechnical practice the factor of safety is the mesh in the z direction was altered by simply adding or
defined as the ratio of the average shear strength of the soil removing ‘slices’ of elements in that direction.
to the average shear stress developed along the critical The first slope analysed consisted of ‘undrained clay’ with
failure surface. Typically, a factor of safety of about 1.5 is shear strength given by
required for design. cu
Based on the above definition, the current approach is to
u ¼ 08, ¼ 0:20 (3)
ªH
use a shear strength reduction technique in which factored
shear strength parameters c9f and
9f , given by and elastic properties as indicated above.
c9 The slope is inclined at an angle of 26.578 to the
c9f ¼ (1) horizontal (2:1 slope), and the boundary conditions are given
SRF as ‘rough-smooth’ for the 3D analysis. Table 3 explains the
 
tan
9 meaning of the various boundary conditions that can be

9f ¼ tan 1 (2)
SRF specified by the user. The ‘rough-smooth’ boundary condi-
tion implies a symmetric analysis about the plane z ¼ L/2:
are used in the analysis, where SRF is a ‘strength reduction thus only half of the actual depth L of the slope is analysed.
factor’. In line with conventional slope stability analysis The bottom (y ¼ D ) and far side (z ¼ 0) of the slope are
methods, the strength reduction factor is assumed to apply fully fixed, while the back (x ¼ 0) and front side (z ¼ L/2)
equally to both c9 and tan
9. In order to obtain the ‘true’ of the slope are constrained by vertical rollers. The dimen-
factor of safety the strength reduction factor is gradually sions of the slope analysed are given in Table 4. The depth
increased until failure of the slope, as described in the next L of the slope is to be varied in the range H , L , 14H
section, occurs. When this critical value has been found, the (because of symmetry the actual mesh depth varied by half
factor of safety of the slope is equal to the strength reduc- this amount), enabling an investigation to be made of the
tion factor and FS  SRF. influence of three-dimensionality. Both the coarser and finer
meshes indicated in Fig. 2(b) were run to illustrate the
sensitivity of results to mesh refinement.
Failure of the slope Table 5 shows results for the specific case of L ¼ 2H
In the program used in this study, slope failure is said to (coarser mesh) as SRF was gradually increased. The table
have occurred when the algorithm cannot converge within a shows seven trial strength reduction factors, ranging from
user-specified iteration ceiling (typically set to 1000). If the 0.5 to 1.734.
algorithm reaches the iteration ceiling it means that the The ‘Iterations’ column displays the number of iterations
algorithm is unable to find a stress redistribution that will needed for convergence. As the factor of safety is ap-
simultaneously satisfy both global equilibrium and the proached, the algorithm has to work harder to reach conver-
Mohr–Coulomb failure criterion with reduced strength para- gence, as seen by the increase in the number of iterations.
meters. At this point slope failure occurs, resulting in rapidly When SRF ¼ 1.73 the analysis was unable to converge
increasing nodal displacements in the mesh. Fig. 1 shows a within 1000 iterations, and a sudden increase in the dimen-
typical graph of SRF against E9max /ªH2 (a dimensionless sionless displacement was observed. At this point FS 
displacement), where max is the maximum nodal displace- SRF, and the factor of safety is given by FS  1.73. The
ment component at convergence, and H is the slope height. results in Table 5 were the actual values plotted in Fig. 1.
Several 3D analyses were performed using both the
meshes indicated in Fig. 2(b). In addition, a conventional
VALIDATION AGAINST 2D ANALYSIS limit equilibrium analysis was performed on the same cross-
An initial step in validating the results was to compare the section, giving a 2D factor of safety of FS ¼ 1.25. A
results from the 3D analyses with those obtained by conven- comparison of the 3D FE and 2D limit equilibrium analyses
tional 2D limit equilibrium analysis. The example geometry is given in Fig. 3. The factor of safety in 3D was always
shown in Fig. 2(a) is of a homogeneous slope in which the higher than in 2D but tended to the plane strain solution for
geometry and dimensions in the x–y plane are extended by depth ratios of the order of L/H  10. It should be noted
540 GRIFFITHS AND MARQUEZ
y

W1

S x
z
W2

D
H

L/2

(a)

(b)

Fig. 2. (a) 3D slope dimensions with uniform section and properties in the z direction;
(b) typical coarser and finer 3D meshes of 20-node hexahedral elements (L/H = 2)

Table 3. Description of 3D boundary condition

Plane Smooth-smooth Rough-smooth Rough-rough

x ¼ 0 Vertical rollers Vertical rollers Vertical rollers

x ¼ W1 + S + W2 Vertical rollers Vertical rollers Vertical rollers
y ¼ D Fixed Fixed Fixed
z ¼ 0 Vertical rollers Fixed Fixed
z ¼ L Vertical rollers Vertical rollers Fixed
(z ¼ L/2 for symmetry)

Table 4. Dimensions of slope for 2D and 3D analyses Table 5. 3D results for L = 2H (coarser mesh) with an iteration
ceiling of 1000
D W1 S W2 L
SRF E9max /(ªH2 ) Iterations
1.5H H 2H H H ! 14H
0.500 0.701 13
1.000 1.004 25
1.500 1.470 48
that the finer FE mesh always gave slightly lower factors of 1.625 1.645 80
safety than the coarser mesh, but the difference never 1.6875 1.845 218
exceeded 2%. 1.7188 2.217 676
1.7344 2.855 1000+

OTHER VALIDATION EXAMPLES

For further verification of the present method, a series of quite simple, geometrically, and include relatively homoge-
3D analyses is now presented for comparison with results neous material properties. It should be emphasised that these
obtained by various investigators using other 3D slope simplifications are merely to validate the present approach
stability techniques. The examples described are typically and facilitate comparison with other published results. The
 SLOPE STABILITY ANALYSIS BY ELASTO-PLASTIC FINITE ELEMENTS 541
2·4
region. In view of the symmetry, ‘rough-smooth’ boundary
2·3
conditions were used, as described in Table 3. It should be
2·2
noted that the spherical failure surface does not necessarily
2·1
represent the critical failure mechanism: thus in a less
2·0 confined mesh the FE approach might find a more critical
1·9 failure path. The results from the FE analysis in the present
1·8 3D FE (coarser mesh) study, together with results obtained by several other investi-
3D FE (finer mesh)
FS

1·7 gators who considered the same slope, are summarised in

2D limit equilibrium
1·6 Table 6. The deformed mesh corresponding to the uncon-
1·5 verged solution when SRF ¼ 1.39 is shown in Fig. 5.
1·4
1·3 1·25
1·2
1·1 Zhang (1988)
1·0 The example shown in Fig. 6 is taken from Zhang (1988)
0 2·0 4·0 6·0 8·0 10·0 12·0 14·0 16·0 and has been used by various investigators as part of the
L/H
validation of their particular 3D slope stability methods (e.g.
Fig. 3. Comparison of 3D FE and 2D limit analyses for
u = 08 Lam & Fredlund, 1993; Huang & Tsai, 2000; J. Chen et al.,
slope with cu /ªH = 0.20 2003). The proposed critical slip surface from limit equili-
brium considerations was circular in the x–y plane and
FE approach described herein requires no such simplifica- ellipsoidal in the out-of-plane direction, which indicated
tions, since complex geometries and material variability are suitable 3D mesh dimensions. Symmetry implied the use
easily implemented. once more of ‘rough-smooth’ boundary conditions.
The problem was analysed for two different cases. In Case
1, the slope is composed of a drained homogeneous soil.

Baligh & Azzouz (1975)

In this example a 3D slope was considered as shown in
Fig. 4, in which Baligh & Azzouz (1975) suggested a
R ⫽ 24·4 m
spherical failure surface. From the radius of the sphere, a 20
Case 1 Circular
suitable out-of-plane dimension for the FE mesh was de-
slip surface 2
duced that would force a mechanism in the same general
1 15
2
1 ° γ ⫽ 18·8 kN/m3
0·5 m 10 (m)
φ⬘ ⫽ 20°
c⬘ ⫽ 29 kN/m
R ⫽ 1·0 m γ ⫽ 1·0 kN/m3 5
cu/γR ⫽ 0·1 Weak layer
φu ⫽ 0° c⬘ ⫽ 0, φ⬘ ⫽ 10°
0
50 40 30 20 10 0
(m)
Fig. 4. Plan and section of 3D failure surface in purely cohesive
slope as reported by Baligh & Azzouz (1975) Fig. 6. Cross-section of 3D slope reported by Zhang (1988)

Table 6. Comparison of 3D results (FS) for the slope in Fig. 4

Baligh & Azzouz (1975) Hungr et al. (1989) Huang & Tsai (2000) Present study

1.402 1.422 1.399 1.39

Rough

Smooth

Fig. 5. Undeformed and ‘failed’ FE meshes for example shown in Fig. 4

542 GRIFFITHS AND MARQUEZ
The resulting factor of safety of 2.17 from the present study
is comparable to those obtained by other investigators, as
seen in Table 7. The results found using the package
Rough
CLARA (Hungr, 1988) were obtained by the authors. Fig. 7
shows the undeformed and ‘failed’ meshes for this analysis.
It can be seen that the failure mechanism is similar to that
described in Fig. 6.
Case 2 has the same slope geometry, but includes a
weak layer in the foundation of the slope (e.g. geosyn-
thetic layer). Chen et al. (2001) used their upper-bound
approach to analyse the slope, while a limit equilibrium Smooth
analysis was used by Chen et al. (2003a). The present
study yielded a factor of safety of 1.58, which again is in
the vicinity of those provided by other investigators, as Weak layer
shown in Table 8. Fig. 8 shows that the failure mechanism
concentrated itself along the weak layer, giving a lower
Fig. 8. ‘Failed’ FE mesh for Case 2 of Zhang (1988) example
factor of safety, as would be expected. Fig. 9 shows the
curves of dimensionless displacement against SRF for both
5
cases.
For the relatively simple validation examples shown here,
the computed FE results lay well within the quite narrow
range of values obtained by other investigators.
The main power of the FE method, however, lies in its
10
ability to deal with more complicated geometries, boundary

E⬘δmax
γH 2
conditions and property variability. With the confidence FS FS
gained through the validation examples, the final part of
this paper introduces more realistic boundary conditions
and examines their influence on the computed factor of
15
safety.

Case 2 Case 1

0 0·5 1·0 1·5 2·0 2·5

INFLUENCE OF SLOPING SIDES SRF
All the examples considered so far had vertical boundaries
in the out-of-plane direction. In geotechnical practice it is Fig. 9. Dimensionless displacements for Cases 1 and 2 of Zhang
clearly more likely that these boundaries will be inclined to (1988) example

Table 7. Comparison of 3D results (FS) from various investigators for Case 1 of Zhang
(1988) example

Zhang (1988) Chen et al. Chen et al. CLARA Present study

(2001) (2003)

2.122 2.262 2.187 2.167 2.17

Rough

Smooth

Fig. 7. Undeformed and ‘failed’ FE meshes for Case 1 of Zhang (1988) example

Table 8. Comparison of 3D results (FS) from various investigators for Case 2 of Zhang
(1988) example

Zhang (1988) Chen et al. Chen et al. CLARA Present study

(2001) (2003)

1.553 1.717 1.603 1.620 1.58

 SLOPE STABILITY ANALYSIS BY ELASTO-PLASTIC FINITE ELEMENTS 543
the horizontal, as would be the case at the abutments of an Bottom depth l2 held constant
earth dam. Another comparison was completed by holding l2 constant
In order to accommodate these changes, a geometry and varying Æ. This time, as Æ is increased, it can be
subroutine was developed (Marquez, 2004) to allow the expected that the calculated factor of safety will decrease.
inclination of the side slopes to be controlled through the When l1 is increased, the soil mass becomes less confined
input data. With this modification, the vertical sides consid- and a failure mechanism can develop more easily. Increasing
ered earlier can be considered as a special case. Æ in this case results in a less stable slope, with a summary
of results shown in Fig. 12.
The 2D result is included in both sets of analyses,
Parametric studies emphasising the conservativeness of a ‘plane strain’ analysis
To help understand the influence of sloping sides a series performed at the centreline.
of tests and comparisons was made. Fig. 10 shows the Comparing the results obtained from various trials, it can
parameters involved in defining the sloping geometry. In the be seen that the depth of the slope (out-of-plane) and the
examples that follow either l1 or l2 was held constant while side slope inclination both have a significant influence on
Æ was varied, with other dimensions fixed to the values the calculated factor of safety.
given in Table 9. The boundary conditions in this case were
‘rough-smooth’, with the sloping boundary fully restrained
(rough) and the vertical boundary allowed to move freely in
the vertical plane (smooth). This analysis implies symmetry, INFLUENCE OF VARIABLE STRENGTH
so the actual depth of the embankment modelled was twice The first example of variable strength involves an un-
the depth of the mesh in the third dimension. drained clay (
u ¼ 08) with dimensions given in Table 10.
The slope is constructed with a ‘strong’ soil (cu /ªH ¼
0.263) near the abutments surrounding a ‘weak’ soil (cu /ªH
Top depth l1 held constant ¼ 0.132) in the central parts of the slope. Sloping sides
When l1 is held constant and Æ is increased, it can be were included in the analysis with l1 /l2 ¼ 2 and Æ ¼ 298.
expected that the calculated factor of safety will increase. Since the slope was no longer symmetrical (the side slope
As Æ is increased, the soil mass begins to sit more on the angle on either side differs, as seen in Fig. 13(a)), the
abutment or canyon wall. The slope also becomes more boundary conditions used are now ‘rough-rough’. The analy-
confined, resulting in added stability. Two examples were sis was performed, and the corresponding deformed mesh at
developed, one with l1 /H ¼ 0.833 and the other with l1 /H ¼ failure is shown in Fig. 13(b). The 3D factor of safety was
1.67. The results for the calculated factor of safety for the computed as 1.42, while the 2D analysis based on the
two problems can be seen in Fig. 11. It can be seen that the weaker soil in the centre of the slope gave a factor of safety
factor of safety for the narrower (l1 /H ¼ 0.833) slope is equal to 0.81.
greater than that of the deeper slope (l1 /H ¼ 1.67). The data The second example is similar to that in Fig. 13. The
also show that the influence of Æ on the slope decreases as example again assumes undrained soils (
u ¼ 0) with
the slope becomes wider. This indicates that 3D effects are dimensions given in Table 11. The strong soil (cu /ªH ¼
much more important in narrow slopes than in deep slopes. 0.263) is once again surrounding the weaker soil (cu /ªH ¼
0.132). The top of the slope l1 was held constant, with the

W1 1·20
1·15 l2/H ⫽ 0·83
l1 S 1·10 l2/H ⫽ 1·67
O 1·05
W2 2D
α 1·00
FS 0·95
H 0·90
D
0·85
y 0·80
0·75
z 0·70
l2 0 5 10 15 20 25 30 35 40 45
α
O
x Fig. 12. Results for constant l2 with
9 = 208, c9/ªH  0.127 and
rough-smooth boundary conditions
Fig. 10. 3D sloping side geometry

1·5 Table 9. Dimensions for analyses with sloping sides

1·4
1·3 l1/H ⫽ 0·83 D W1 S W2
1·2 l1/H ⫽ 1·67
1·1 2D 1.5H H 2H 0.66H
FS
1·0
0·9
0·8
0·7 Table 10. Dimensions for the slope in Fig. 13 (see Fig. 10)
0·6
0 5 10 15 20 25 30 35 40 45 D W1 S W2 l1 l2
α

Fig. 11. Results for constant l1 with
9 = 208, c9/ªH  0.127 and 1.5H H 2H 0.667H 2.5H 1.25H
rough-smooth boundary conditions
544 GRIFFITHS AND MARQUEZ
Weaker Weaker
0 0
⫺4 ⫺4
⫺8
y ⫺8
⫺12 y
⫺12
⫺16
⫺16
0 5 10 15 20 25 30
z 0 5 10 15 20 25 30
(a) z
(a)

Rough
Rough Rough
0 Rough
0
⫺5
⫺5
y ⫺10
l1/l2 ⫽ 2 y ⫺10 l1/l2 ⫽ 3
⫺15 0 α ⫽ 29°
⫺15 0 α ⫽ 35·8°
10 φu ⫽ 0°
0 10 φu ⫽ 0°
5 20 0
10 x 5 20
15 30
z 20 10 x
25 15 30
40 z 20
30 25 40
(b) 30
(b)
Fig. 13. Finite element mesh displaying a non-symmetric slope
with weak soil (c9/ªH = 0.132) surrounded by stronger soil (c9/ Fig. 14. Finite element mesh showing a more confined slope
ªH = 0.263) with weak soil (c9/ªH = 0.132) surrounded by stronger soil (c9/
ªH = 0.263)

Table 11. Dimensions for the slope in Fig. 14

D W1 S W2 l1 l2 Decreasing strength
0
1.5H H 2H 0.667H 2.5H 0.833H ⫺4
⫺8
y
⫺12
⫺16
bottom of the slope l2 decreased to give a ratio of l1 /l2 ¼ 3
0 5 10 15 20 25
and Æ ¼ 35.88, as shown in Fig. 14(a). The boundary z
conditions used were again ‘rough-rough’. The 3D factor of (a)
safety in this case was computed as 1.50, which should be
compared with the unchanged 2D factor of safety of 0.81.
The deformed mesh at failure is given in Fig. 14(b). A
higher 3D factor of safety in this case is to be expected
because of the greater confinement of the soil mass provided Rough
by the increase in the l1 /l2 ratio. This is a similar effect to
that demonstrated in Fig. 11. 0 l1/l2 ⫽ 2·5
The final example, shown in Fig. 15, utilises symmetry ⫺5 α ⫽ 29°
(‘rough-smooth’ boundary conditions): thus only half the φu ⫽ 0°
y ⫺10
problem is analysed. The slope consists of undrained clay 0
(
u ¼ 0) with dimensions given in Table 12. In this case, ⫺15
10 Smooth
the strength decreases linearly from cu /ªH ¼ 0.219 at the 0 20
abutments to cu /ªH ¼ 0.132 at the centreline. The 3D factor 5
10 30 x
of safety in this case is computed as 1.30, and should be z 15 20 40
compared with the usual 2D factor of safety of 0.81 at the 25
centreline. The 3D deformed mesh at failure is given in Fig. (b)
15(b). The difference between the 2D and 3D results is still
Fig. 15. Finite element mesh with linearly decreasing soil
significant but not as pronounced as when there was a step strength
change in strength.

CONCLUDING REMARKS
The FE technique for slope stability analysis has grown
Table 12. Dimensions for the slope in Fig. 15
significantly in popularity in recent years, owing to its power
and versatility. The benefits of the FE approach to 2D slope D W1 S W2 l1 l2
stability analysis are well documented; however, these ad-
vantages over traditional limit equilibrium approaches are 1.5H H 2H 0.667H 2.08H 1.25H
even more important in 3D owing to the ease with which
 SLOPE STABILITY ANALYSIS BY ELASTO-PLASTIC FINITE ELEMENTS 545
complex geometries, boundary conditions and property var- element analysis of slopes. J. Geotech. Engng 122, No. 7, 577–
iations in the out-of-plane direction can be introduced. The 596.
paper has presented results from several 3D slope examples Farzaneh, O. & Askari, F. (2003). Three-dimensional analysis of
using an elasto-plastic FE approach. Results were validated nonhomogeneous slopes. J. Geotech. Geoenviron. Engng 129,
against conventional 2D limit equilibrium analyses of a No. 2, 137–145.
Griffiths, D. V. (1980). Finite element analysis of walls, footings
homogeneous slope and demonstrated the convergence of the and slopes. Proceedings of the symposium on computer applica-
3D factor of safety on the 2D result as the out-of-plane tions to geotechnical problems in highway engineering (ed. M.
dimension was increased. Further examples demonstrated the F. Randolph), pp.122–146. Cambridge: PM Geotechnical Ana-
influence of boundary conditions and confinement in the lysts Ltd.
form of sloping abutments and embankment depth. Finally, Griffiths, D. V. & Lane, P. A. (1999). Slope stability analysis by
some examples were presented that introduced variable finite elements. Géotechnique 49, No. 3, 387–403.
strength parameters across the slope in the out-of-plane Hammah, R. E., Yacoub, T. E., Corkum, B. & Curran, J. H. (2005).
direction. A comparison of finite element slope stability analysis with
While it seems unlikely that 3D slope stability will conventional limit-equilibrium investigation. Proc. 58th Cana-
dian Geotechnical and 6th Joint IAH-CNC and CGS Ground-
become a routine approach in geotechnical practice any time
water Specialty Conferences – GeoSask 2005, Saskatoon, 480–
soon, the increased speed and falling costs of computers 487.
mean that 3D non-linear FE analyses can now be performed Huang, C. C. & Tsai, C. C. (2000). New method for 3D and
routinely on a desktop or laptop computer. Perhaps the best asymmetric slope stability analysis. J. Geotech. Geoenviron.
justification for promoting 3D analysis at this stage lies Engng ASCE 126, No. 10, 917–927.
mainly in advancing the state of the art. Three-dimensional Huang, C. C., Tsai, C. C. & Chen, Y. H. (2002). Generalized
analysis is simply more realistic, and leads not only to method for three-dimensional slope stability analysis. J. Geo-
improved accuracy but also to a better understanding of the tech. Geoenviron. Engng, ASCE 128, No. 10, 836–848.
nature of slope failure mechanisms. Hughes, T. J. R. (1987). The finite element method. Englewood
Cliffs, NJ: Prentice Hall.
Hungr, O. (1987). An extension of Bishop’s simplified method of
slope stability analysis to three dimensions. Géotechnique 37,
ACKNOWLEDGEMENT No. 1, 113–117.
The authors wish to acknowledge the support of NSF Hungr, O. (1988). CLARA 2.31: Slope stability in two or three
grant CMS-0408150 on ‘Advanced probabilistic analysis of dimensions for IBM compatible microcomputers. Vancouver: O.
stability problems in geotechnical engineering’. Hungr Geotechnical Research Inc.
Hungr, O., Salgado, F. M. & Byrne, P. M. (1989). Evaluation of
three-dimensional method of slope stability analysis. Can. Geo-
tech. J. 26, No. 4, 679–686.
REFERENCES Hutchinson, J. N. & Sarma, S. K. (1985). Discussion on ‘Three-
Arellano, D. & Stark, T. D. (2000). Importance of three-dimensional dimensional limit equilibrium analysis of slopes’. Géotechnique
slope stability analysis in practice. In Slope Stability 2000, GSP 35, No. 2, 215.
no. 101 (eds D. V. Griffiths et al.), Reston, VA: ASCE, pp. 18– Jeremic, B. (2000). Finite element methods for three-dimensional
32. slope stability analysis. In Slope Stability 2000, GSP no. 101
Baligh, M. M. & Azzouz, A. S. (1975). End effects on the stability (eds D. V. Griffiths et al.), Reston, VA: ASCE, pp. 224–238.
of cohesive slopes. J. Geotech. Engng, ASCE 101, No. GT11, Jiang, J. C. & Yamagami, T. (2004). Three-dimensional slope
1105–1117. stability analysis using an extended Spencer method. Soils
Bromhead, E. N. & Martin, P. L. (2004). Three-dimensional limit Found. 44, No. 4, 127–135.
equilibrium analysis of the Taren landslide. In Advances in Kidger, D. J. (1990). Visualisation of finite element eigenvalues and
geotechnical engineering (Skempton Conference), Vol. 2, pp. three-dimensional plasticity. PhD thesis, Department of Engi-
789–802. London: Thomas Telford. neering, University of Manchester.
Chang, M. (2002). A 3D slope stability analysis method assuming Lam, L. & Fredlund, D. G. (1993). A general limit equilibrium
parallel lines of intersection and differential straining of block model for three-dimensional slope stability analysis. Can. Geo-
contacts. Can. Geotech. J. 39, No. 4, 799–811. tech. J. 30, No. 6, 905–919.
Chang, M. (2005). Three-dimensional stability analysis of the Loehr, J. E., McCoy, B. F. & Wright, S. G. (2004). Quasi-three-
Kettleman Hills landfill slope failure based on observed sliding- dimensional slope stability analysis method for general sliding
block mechanism. Comput. Geotech. 32, No. 8, 587–599. bodies. J. Geotech. Geoenviron. Engng, ASCE 130, No. 6, 551–
Chen, J., Yin, J. & Lee, C. F. (2003a). Upper bound limit analysis 560.
of slope stability using rigid finite elements and nonlinear Marquez, R. M. (2004). Three-dimensional slope stability analysis
programming. Can. Geotech. J. 40, No. 4, 742–752. using finite elements. Masters thesis, Division of Engineering,
Chen, J., Yin, J. H. & Lee, C. F. (2005). A three-dimensional Colorado School of Mines.
upper-bound approach to slope stability analysis based on Matsui, T. & San, K. C. (1992). Finite element slope stability
RFEM. Géotechnique 55, No. 7, 549–556. analysis by shear strength reduction technique. Soils Found. 32,
Chen, R. H. & Chameau, J. L. (1982). Three-dimensional limit No. 1, 59–70.
equilibrium analysis of slopes. Géotechnique 32, No. 1, 31–40. Perzyna, P. (1966). Fundamental problems in viscoplasticity. Ad-
Chen, Z., Wang, J., Wang, Y., Yin, J. H. & Haberfield, C. (2001). A vances in Applied Mechanics 9, 243–377.
three-dimensional slope stability analysis method using the Potts, D. M., Dounias, G. T. & Vaughan, P. R. (1990). Finite
upper bound theorem. Int. J. Rock Mech. Mining Sci. 38, No. 3, element analysis of progressive failure of Carsington embank-
379–397. ment. Géotechnique 40, No. 1, 79–102.
Chen, Z., Mi, H., Zhang, F. & Wang, X. (2003b). A simplified Sainak, A. N. (2004). Application of three-dimensional finite ele-
method for 3D slope stability analysis. Can. Geotech. J. 40, No. ment method in parametric and geometric studies of slope
3, 675–683. stability. In Advances in geotechnical engineering (Skempton
Chugh, A. K. (2003). On the boundary conditions in slope stability Conference), Vol. 2, pp. 933–942. London: Thomas Telford.
analysis. Int. J. Numer. Anal. Methods Geomech. 27, No. 11 Seed, R. B., Mitchell, J. K. & Seed, H. B. (1990). Kettleman Hills
905–926. waste landfill slope failure. II: Stability analysis. J. Geotech.
Duncan, J. M. (1996a). Soil slope stability analysis. In Landslides: Engng 116, No. 4, 669–689.
Investigation and mitigation, Special Report 247 (eds A. K. Silvestri, V. (2006). A three-dimensional slope stability problem in
Turner et al.), Chapter 13. Washington, DC: Transportation clay. Can. Geotech. J. 43, No. 2, 224–228.
Research Board. Smith, I. M. & Griffiths, D. V. (1988). Programming the finite
Duncan, J. M. (1996b). State of the art: limit equilibrium and finite- element method, 2nd edn. Chichester: John Wiley & Sons.
546 GRIFFITHS AND MARQUEZ
Smith, I. M. & Griffiths, D. V. (2004). Programming the finite analysis I: Basic theory. Science in China, Series E-Engineering
element method, 4th edn. Chichester: John Wiley & Sons. & Materials Science 48, 171–183.
Smith, I. M. & Hobbs, R. (1974). Finite element analysis of Zhang, J. F., Li, Z. G. & Qi, T. (2005b). An extension of
centrifuged and built-up slopes. Géotechnique 24, No. 4, 531– 2D Janbu’s generalized procedure of slices for 3D slope
559. stability analysis I: Numerical method and applications. Science
Stark, T. D. & Eid, H. T. (1998). Performance of three-dimensional in China, Series E-Engineering & Materials Science 48,
slope stability methods. J. Geotech. Geoenviron. Engng 124, No. 184–195.
11, 1049–1060. Zhang, X. (1988). Three-dimensional stability analysis of concave
Xie, M., Esaki, T., Zhou, G. & Mitani, Y. (2003). Geographic slopes in plan view. J. Geotech. Engng, ASCE 114, No. 6, 658–
information systems-based three-dimensional critical slope stabi- 671.
lity analysis and landslide hazard assessment. J. Geotech. Zienkiewicz, O. C. (1977). The finite element method, 3rd edn.
Geoenviron. Engng 129, No. 12, 1109–1118. London, New York: McGraw-Hill.
Xie, M. W., Esaki, T. and Cai, M. F. (2006). GIS-based implemen- Zienkiewicz, O. C. & Cormeau, I. C. (1974). Viscoplasticity—
tation of three-dimensional limit equilibrium approach of plasticity and creep in elastic solids: a unified numerical solu-
slope stability. J. Geotech. Geoenviron. Engng 132, No. 5, 656– tion approach. Int. J. Numer. Methods Engng 8, No. 4, 821–845.
660. Zienkiewicz, O. C., Humpheson, C. & Lewis, R. W. (1975).
Zhang, J. F., Qi, T. & Li, Z. G. (2005a). An extension of 2D Associated and non-associated viscoplasticity and plasticity in
Janbu’s generalized procedure of slices for 3D slope stability soil mechanics. Géotechnique 25, No. 4, 671–689.