Vous êtes sur la page 1sur 16

Jiang, G.-J. & Magnan, J.-P. (1997). GeÂotechnique 47, No.

4, 857±872

Stability analysis of embankments: comparison of limit analysis with


methods of slices

G . - L . J I A N G  a n d J. - P. M AG NA N {

Limit analysis has been ®rmly established over L'analyse limite a recËu des bases solides au
the past decades and is now considered to be an cours des dernieÁres deÂcennies et elle est main-
accurate method for predicting failure mechan- tenant consideÂreÂe comme une meÂthode preÂcise
isms and for calculating limit loads. A new ®nite de preÂvision des meÂcanismes de rupture et de
element procedure has been developed, and is calcul des charges limites. Une nouvelle proceÂ-
brie¯y described in this paper, to determine dure de calcul en eÂleÂments ®nis est brieÁvement
automatically the optimal failure mechanism for deÂcrite dans cet article. Elle a eÂte deÂveloppeÂe
geometrically complicated structures under ap- a®n de deÂterminer automatiquement le meÂca-
plied loads. In limit analysis, the objective is nisme de rupture optimal pour les structures aÁ
generally to determine the collapse load, where- geÂometrie compliqueÂe sous les charges qui leur
as the factor of safety is calculated in methods sont appliqueÂes. Dans l'analyse limite, l'objectif
of slices. This paper also shows how the factor est en geÂneÂral de deÂterminer la charge de ruine,
of safety introduced in the limit equilibrium tandis qu'un facteur de seÂcurite est calcule dans
analysis can be systematically incorporated into les meÂthodes de tranches. Cet article montre
limit analysis. Several computations of stability aussi comment le facteur de seÂcurite introduit
of embankments were carried out to compare dans l'analyse de l'eÂquilibre limite peut eÃtre
the limit analysis results with those obtained by incorpore systeÂmatiquement dans l'analyse lim-
methods of slices based on the limit equilibrium ite. Plusieurs calculs de stabilite de remblais ont
analysis. Through this investigation, some differ- eÂte reÂaliseÂs pour comparer les reÂsultats de
ences and similarities between these two theories l'analyse limite avec ceux obtenus par la meÂth-
can be demonstrated. The numerical results ode des tranches, c'est aÁ dire par l'analyse de
demonstrate that limit analysis by the present l'eÂquilibre limite. Cette eÂtude a permis de
®nite element method generally gives factors of mettre en eÂvidence quelques diffeÂrences et res-
safety more accurate than or as accurate as semblances entre ces deux theÂories. Les reÂsultats
those obtained by methods of slices. numeÂriques deÂmontrent que l'analyse limite par
la meÂthode des eÂleÂments ®nis donne geÂneÂrale-
ment des facteurs de seÂcurite qui sont aussi
KEYWORDS: embankments; failure; landslides; nu- preÂcis, sinon plus, que ceux qui sont calculeÂs
merical modelling and analysis; plasticity; slopes. par les meÂthodes de tranches.

INTRODUCTION of earth slopes. The development of stability analy-


The stability of soil slopes and embankments is a sis and the theory of limit analysis was accelerated
basic problem in soil mechanics. In 1773 Coulomb from the 1950s because of the complete establish-
made the earliest attempt to solve the problem of ment of plasticity theory and the later advent of
the stability of vertical cuts. In the 1920s, much computers.
progress in this domain was achieved owing to the Two approaches were developed in limit analy-
works by Prandtl in plasticity and by Fellenius in sis: static and kinematic (Chen, 1975; SalencËon,
the method of slices. In 1937, Taylor used the 1983). Applications of these approaches in the past
logarithmic spiral method to determine the stability decades were generally focused on some commonly
encountered problems, such as the bearing capacity
of foundations. The general procedure is to assume
a kinematically admissible failure mechanism or a
Manuscript received 22 April 1996; revised manuscript
statically admissible stress ®eld and then to opti-
accepted 18 September 1996.
Discussion on this paper closes 1 December 1997; for mize the objective function with respect to a very
further details see p. ii. limited number of variable parameters. In this way,

Parsons Brinckerhoff Quade & Douglas, Inc., Florida. an upper or lower bound value of the limit load
{ Laboratoire Central des Ponts et ChausseÂes, Paris. can be found.

857
858 JIANG AND MAGNAN

The methods of slices are based on the limit structure. In the kinematic approach, we search for
equilibrium theory. Various methods of slices were a kinematically admissible velocity ®eld; we then
proposed in the 1950s and 1960s. These methods calculate the corresponding internal and external
were proposed with different assumptions on the plastic power dissipations. If the external power
shape of slip surfaces and the interslice forces. The dissipation is higher than the internal one, the load
most commonly used methods of slices comprised can be said to be greater than the limit load. In
the Fellenius method, the methods of Bishop this way, the limit load can be de®ned as the load
(1955), Janbu (1957), Morgenstern & Price (1965) under which there exists a statically admissible
and Spencer (1967). Other efforts (Chen & Mor- stress ®eld, yet a free plastic ¯ow (incipient plastic
genstern, 1983; Chen & Shao, 1988; Chen, 1992) ¯ow) can occur.
were made in the 1980s to develop ef®cient and The essential problem in limit analysis is how to
accurate optimizing techniques for the solution of determine the optimum among a series of statically
the critical slip surface and the minimum factor of admissible stress ®elds or kinematically admissible
safety. velocity ®elds so that the lower or upper bound is
Among differences between limit analysis and as close as possible to the limit load. In the past
methods of slices, the most important one is perhaps three decades, this problem was often treated by
that limit analysis is more versatile and can be used assuming a mode of stress distribution or a mode
to analyse the limit behaviour of all perfectly plastic of failure which was de®ned by some geometrical
structures while methods of slices were developed or other variable parameters, and then establishing
essentially for the stability analysis of earth slopes an objective function and optimizing this function
and embankments. Although limit analysis has been (manually or by computer) with respect to these
considered to be more rigorous than the methods of parameters. The de®nition of statically admissible
slices, its coherent application in engineering prac- stress ®elds or kinematically admissible velocity
tice has been rather scarce compared with its capa- ®elds for a particular structure often requires intui-
city of applicability. In practice, the Fellenius and tion of the calculator. Fortunately, the commonly
Bishop methods enjoy a larger popularity in stability used yield functions such as the Mohr±Coulomb
analysis of slopes and embankments among geotech- criterion are of linear form in the principal stress
nical researchers as well as engineers. space; the discontinuous slip surfaces are then
To enlarge the application of limit analysis in logarithmic, circular and=or planar surfaces ac-
geotechnical engineering, this paper presents a cording to the normality ¯ow rule of plasticity.
versatile ®nite element method and demonstrates Nevertheless, advanced computation schemes are
the validity of the upper bound method and some necessary for accurate determination of the limit
advantages of limit analysis over methods of slices. load if a complicated problem is to be tackled. As
for many other engineering problems, we resort to
numerical methods, especially the ®nite element
PLASTICITY AND LIMIT ANALYSIS method, for this purpose. Many attempts have been
Contrary to the method of slices, which was ®rst made in this direction (Anderheggen & KnoÈpfel,
developed by Fellenius for the speci®c problem of 1972). However, the ®nite element method in limit
the stability of slopes, limit analysis is based on analysis is still a research subject and has not been
plasticity theory, which is a general theory of widely used by the geotechnical engineer. In prac-
mechanics. In consequence, limit analysis can be tice, the methods of slices such as the Bishop and
applied to structures of arbitrary geometry, compli- Fellenius methods are more frequently applied than
cated loading conditions and homogeneous as well limit analysis in assessing the stability of slopes
as heterogeneous plastic materials. Many applica- and embankments. In this paper, the traditional
tions of limit analysis can be found in frame de®nition of factor of safety is introduced in limit
structures, the bearing capacity of foundations and analysis so that the results from limit analysis can
the stability of tunnels and slopes. be directly compared with those from methods of
In limit analysis, two approaches serve to brack- slices. By making a suf®ciently large number of
et the limit road. The ®rst one is the static comparisons, the upper bound limit analysis using
approach, which gives a lower bound of the limit ®nite elements proved to be a viable tool, since it
load. The second one is the kinematic approach, gave accurate limit loads or factors of safety and
which gives an upper bound of the limit load. In automatically provided rational failure mechanisms.
the static approach, we determine whether there In perfect plasticity theory, a plasticity admissi-
exists an equilibrium stress ®eld which is in equili- ble stress ®eld is de®ned by a convex yield func-
brium with the applied load and with which the tion F(ó):
plastic yield condition is nowhere violated in the F(ó) < 0 (1)
structure. If such a stress ®eld exists, it can be
ascertained that the applied load is less than the The failure yield surface is de®ned by the
limit load and no plastic failure will occur in the following equation:
STABILITY ANALYSIS OF EMBANKMENTS 859
F(ó) ˆ 0 (2) is the limit load. If
Within the framework of rigid plasticity, no min J(u0 ) ˆ 0 and u0 ˆ 0,
deformation is produced under the stress ó if u
F(ó) , 0. Plastic strains may occur only when
equation (2) is satis®ed and the strain rates are the load is less than the limit load. In this case, the
given by the associated normality ¯ow rule structure may be in a complete elastic state. It may
also be in a partial plastic state; however, the
@F plastic zone does not extend so far that a free
åˆë (3) plastic ¯ow can occur since rigid plasticity is

where å represents the strain rate tensor and ë assumed. For a load higher than the limit load,
denotes the plastic multiplier. The strain rate tensor there exists no stress ®eld which can simulta-
å is related to the velocity u by the following neously satisfy equation (1) and the equilibrium
equation: conditions; the plastic zone extends to such an
extent that a collapse will certainly occur under the
 
@u i @u j load. Between these two cases, the structure under
å ij ˆ 12 ‡ , i, j ˆ 1, 2 (4) the limit load is in an incipient plastic ¯ow state
@x j @xi
yet a limit equilibrium condition is satis®ed.
The strain rate is said to be plastically admissible In many engineering problems, the external
if equation (3) is veri®ed at each point of the loads ( f , T) consist of two types of loads. Some
structure. The velocity is subject to the boundary of them are dead loads which are represented in
conditions this paper by ( f 0 , T0 ); others are changeable loads
noted by ( f 1 , T1 ). Limit analysis or stability analy-
u ˆ 0 over à u (5) sis is applied to determine a multiplier m for
where à u is the boundary surface on which the which the sum of ( f 1 , T0 ) and m( f 1 , T1 ) repre-
velocity is given. sents the limit load point in the loading history.
In the following formulation, a ®xed rectangular This de®nition of loading conditions is complex
Cartesian coordinate system with coordinate axes for hand calculation, but is more realistic for
xi (i ˆ 1, 2) is used. The components of the stress practical problems and provides much convenience
tensor and strain tensor are denoted by ó ij and å ij, in the ®nite element formulation.
respectively and the components of the velocity With these de®nitions, the functional J in equa-
vector are denoted by u i. tion (7) can be rewritten as
The rate of plastic work per unit volume is … … …
de®ned by J (u) ˆ D(å) dÙ ÿ f 0 : u dÙ ÿ T0 : u dÃ
Ù Ù ÃT
D ˆ ó ij å ij (6) … …
This function can be written as D(å ij ) since the ÿm f 1 : u dÙ ÿ m T1 : u dà (8a)
Ù ÃT
plastic strain rates are produced only when condi-
tion (2) is satis®ed. D(å ij ) is then deduced by or
combining equations (2), (3) and (6). With all J (u) ˆ Ei (u) ÿ E0 (u) ÿ mEe (u) (8b)
above de®nitions, the upper bound theorem can be
stated as follows. where Ei (u), E0 (u) and mEe (u) represent respec-
If there exists a kinematically admissible velo- tively the internal work dissipation rate, the work
city ®eld and the associated strain rate is plastically rates done by the loads ( f 0 , T0 ) and by the loads
admissible, which makes the following functional m( f 1 , T1 ).
less than zero, the load is then higher than the Corresponding to the limit loading condition, a
limit load: kinematically admissible velocity ®eld u0 can be
… … … sought by letting the minimum of equation (8) be
J (u) ˆ D(å) dÙ ÿ f u dÙ ÿ: T : u dà zero. Therefore, the limit load factor can be deter-
Ù Ù ÃT mined by
(7) Ei (u0 ) ÿ E0 (u0 )
The ®rst term on the right-hand side of equation mˆ (9)
Ee (u0 )
(7) represents the work dissipation rate in the
structure. The last two terms represent the work The meaning of the multiplier m is as follows. In
rates of the body force f and the surface force T limit analysis problems, we know the loading mode
respectively. or loading process, and then determine the limit
The applied load corresponding to load point in the loading process. The prescribed
loads ( f 1 , T1 ) multiplied by m represent a loading
min J (u0 ) ˆ 0 and u0 6ˆ 0
u mode and the loads ( f 0 , T0 ) can be regarded as an
860 JIANG AND MAGNAN

initial loading condition. With m available, the tion of the functional L r with respect to the func-
limit loads can then be represented by the combi- tion ë and min(u,w) represents the minimization
nation of ( f 0 , T0 ) and m( f 1 , T1 ). with respect to the velocity function u and the
function w.
It has been proved that the following relation
holds (e.g. Jiang, 1992)
FINITE ELEMENT SOLUTION FOR THE LIMIT
år ÿ wr ˆ 0 (14)
MULTIPLIER
The basic properties of the functional J were and (u r , w r ) is the solution which minimizes func-
discussed by Jiang (1992, 1994). Different solution tional (11) with constraint equation (10). Therefore,
schemes have been proposed to obtain the kinema- u r is the solution which minimizes functional (8).
tically admissible velocity which minimizes the We can obtain the saddle point by the following
functional in considering these properties. In this algorithm.
paper, we use a scheme directly to minimize the
functional of equation (8) for an initial value of m Set n ˆ 0, give the scalar r a positive value;
and to adjust the value of m according to the value initialize u 0 and ë0 arbitrarily: (15)
of J so that ®nally we have min J ˆ 0. To this
Minimize L r f(u n , w), ë n g,
end, we employed the augmented Lagrangian tech-
nique which was commonly used to solve con- let w n be the solution: (16)
strained minimization problems. n
Minimize L r f(u, w ), ë g, n
The following supplementary constrained equa-
tion can be introduced for equation (8) let u n‡1 be the solution: (17)
n‡1 n n n
åÿwˆ0 (10) Update ë ˆ ë ‡ â(å ÿ w ): (18)
The functional J de®ned by expression (8) is then Repeat procedure (16)ÿ(18) until convergence:
transformed to
… In step (18), â is a positive scalar which in¯uences
J (u, w) ˆ D(w) dÙ ÿ E0 (u) ÿ mEe (u) (11) the convergence performance of the algorithm. In
Ù general, it is better to take a value between 0´1 and
The problem to be solved is thus to minimize 0´8 times r to ensure rapid and stable convergence.
J (u, w) with respect to the ®eld functions u and w The initial velocity u 0 is not necessarily kine-
and with the linear constrained equation (10). We matically admissible. In our program, this initial
can solve this problem by further transforming velocity vector is calculated by procedure (17) with
J (u, w) into its augmented Lagrangian form. The ë0 ˆ 0 and w ˆ 0. The minimization of the aug-
augmented Lagrangian functional can then be mented Lagrangian in (17) leads to solving a sys-
solved by ®nite elements using the Uzawa algo- tem of linear equations; the solution is the vector
rithm (Jiang, 1992). of all nodal velocities. The global system matrix is
The augmented Lagrangian functional can be similar to that in the linear elastic problem.
written as The supplementary variables w are assumed
… constant within each element. In this way, the
r
L r f(u, w), ëg ˆ J (u, w) ‡ (å ÿ w):(å ÿ w) dÙ minimization problem in (16) becomes very sim-
2 Ù ple. The corresponding objective function can be
…
considered as a sum of independent functions over
‡ ë:(å ÿ w) dÙ (12)
Ù each element so that the variables w can be
obtained element by element (Jiang, 1995). Particu-
with u ˆ 0 over à u. larly, the normality ¯ow rule can be directly intro-
The second term on the right-hand side of equa- duced in this level. For example, the normality
tion (12) represents the augmented term and the ¯ow rule can be translated to the following relation
third represents the ordinary Lagrangian term. for plane strain problems of a Mohr±Coulomb
After this transformation, our objective is to ®nd material:
the solution f(u r , w r ), ë r ), which is called the
saddle point, by solving 1 ‡ sin ö
å1 ˆ ÿå2 (å1 > ÿå2 > 0) (19)
1 ÿ sin ö
max L r f(u r , w r ), ëg ˆ L r f(u r , w r ), ë r g
ë
ˆ min L r f(u, w), ë r g where å1 and å2 are the principal strain rates and
(u,w) ö is the internal friction angle of Mohr±Coulomb
(13) materials. The plastic dissipation function of the
Mohr±Coulomb yield criterion can then be written
where the notation maxë represents the maximiza- as follows:
STABILITY ANALYSIS OF EMBANKMENTS 861
p 2
2cå2 cos ö (Fs ‡ tan2 ö) ‡ tan ö
Dˆÿ (å2 < 0) (20) å1 ˆ ÿå2 p 2
1 ÿ sin ö (Fs ‡ tan2 ö) ÿ tan ö
where c is the cohesive strength of Mohr±Coulomb (å1 > ÿå2 > 0) (27)
material.
In considering the constrained relation (10), The corresponding dissipation function then takes
equation (19) can equally be translated to a relation the form
between the eigenvalues of w: ÿ2cå2
D(å, Fs ) ˆ p (å2 < 0)
1 ‡ sin ö (F 2s ‡ tan2 ö) ÿ tan ö
w1 ˆ ÿw2 (w1 > ÿw2 > 0) (21)
1 ÿ sin ö (28)
This relation can directly be introduced in the
minimization problem (16). For the same reason, The problem to be solved can thus be stated as:
the incompressibility condition for von Mises ma- for the prescribed loading and geometrical condi-
terials or purely cohesive soils is introduced in tions, to ®nd the critical factor of safety Fs by
(16) by the following equation: minimizing the following functional and letting the
minimum be equal to zero:
w1 ‡ w2 ˆ 0 (22) … … …
Finally, it should be noted that ®nite elements J (u) ˆ D(å, Fs ) dÙ ÿ f : u dÙ ÿ T : u dÃ
cannot be totally arbitrary for incompressible pro- Ù Ù ÃT

blems, in order to avoid unrealistic failure mech- (29)


anisms (e.g. Jiang, 1992, 1995). For problems of
complex geometry and boundary conditions, we where f and T are the prescribed external body
can choose four constant triangular elements as and surface loads.
basic elements which form a quadrilateral and its The minimization of the functional (29) requires
diagonals. the use of an asymptotic procedure. In this proce-
dure, we assign an estimated value to Fs and then
minimize J (u), adjust the value of Fs and re-
minimize J (u) until we have min u J (u) ˆ 0; the
FACTOR OF SAFETY corresponding factor is the critical factor of safety,
The numerical procedure described in the pre- designated here by F cs. The asymptotic procedure
vious section is a general procedure which can be is inevitable for problems with the Mohr±Coulomb
used for both geotechnical and structural problems yield criterion, but it is simple for problems of
in order to determine the limit load factor. How- purely cohesive soils whether the soils are homo-
ever, the factor of safety de®ned in geotechnical geneous or not, because the plastic dissipation
engineering is based on the conception of mobil- function D is related linearly to the inverse of the
ized strength of the soil: factor of safety:
ÿ2cå2
ô ˆ cm ‡ ó tan öm (23) D(å, Fs ) ˆ (å2 < 0) (30)
Fs
tan ö and the minimization of (29) is equivalent to the
tan öm ˆ (24)
Fs minimization of
c … …
cm ˆ (25) J (u) ˆ ÿ2 cå2 dÙ ÿ Fs f : u dÙ
Fs Ù Ù
where cm and öm are the mobilized cohesion and …
the frictional angle respectively of the Mohr±Cou- ÿ Fs T : u dà (å2 < 0) (31)
lomb yield criterion, and Fs is the factor of safety ÃT

of the structure against failure. By specifying the constrained linear equation


The factor of safety Fs can be readily incorpo- … …
rated into the numerical procedure of limit analy- f : u dÙ ‡ T : u dà ˆ á (á . 0) (32)
sis. From equation (24), the following relation can Ù ÃT
be derived:
tan ö the critical value of the factor of safety is then
sin öm ˆ p (26) approached by
(F 2s ‡ tan2 ö) …
2
The principal strain rates å1 and å2 satisfy the F cs ˆ min (ÿcå2 ) dÙ (å2 < 0) (33)
following equation for the mobilized frictional á u Ù
angle öm according to the normality ¯ow rule (or To minimize (31) or (33) with the constraint equa-
substituting equation (26) into equation (19)): tion (32), the general procedure (15)±(18) is
862 JIANG AND MAGNAN

applicable. It can be seen from equations (31)± Spencer (1967), the slip surface can be circular or
(33) that, for a structure of purely cohesive soils, non-circular.
the factor of safety is identical to the limit load To illustrate the basic principle of the methods
multiplier m, provided the loading process is con- of slices, we describe brie¯y Bishop's simpli®ed
sidered to be a proportional one. The physical method in this section. For more detailed descrip-
meaning of this relation is that the structure will tion of different methods of slices and comparison
collapse either when the strengths of the soil are between them, refer to Nash (1987) and Fredlund
divided by the factor of safety or when the load is & Krahn (1977).
multiplied by it. In Bishop's simpli®ed method, the slip surface is
represented by a circular arc; the overturning mo-
ment M 0 and the resisting moment M r correspond-
METHODS OF SLICES ing to this slip circle (Fig. 1) can be assumed to be
The conventional stability analysis of slopes is two factors which respectively cause and resist
based on the limit equilibrium method. In this meth- against the failure. If the overturning moment is
od, it is postulated that the slope might fail by a mass higher than the resisting moment, the soil mass
of soil sliding on a failure surface. At the moment of within the slip circle will collapse along the slip
failure, the shear strength is fully mobilized all the circle. The overturning moment is calculated by
way along the failure surface, and the overall slope P
M 0 ˆ R w sin á (34)
and each part of it are in static equilibrium. On the P
basis of these postulates, many methods were de- where indicates summation over all slices and
vised. Most of these methods are characterized by R represents the radius of the slip circle. The
the division of the sliding mass into a number of resisting moment is calculated from
slices and are thus called methods of slices. The  
forces acting on a slice comprise the normal and P c tan ö
Mr ˆ R L‡ N9 (35)
shear forces on the base of the slice, the interslice Fs Fs
horizontal and shear forces, and the self-weight of
the soil. These forces must be in equilibrium on each where the coef®cient N9 is given by
slice as well as on the overall embankment. However,  
cL
many of them and their lines of action are unknown N9 ˆ w ÿ sin á ÿ uL cos á
at the outset of the analysis. It has been shown (Nash, Fs
 
1987) that for an embankment divided into n slices tan ö sin á
there are in general 5n ÿ 2 unknowns while there are cos á ‡ (36)
Fs
only 3n equations of statics available. Thus 2n ÿ 2
assumptions must be made for the problem to be
rendered statically determinate. Different types of
assumptions were made and led to various methods O
of slices. A summary of the assumptions of some
α
commonly used methods about the forces can be R c f b
found in Nash (1987).
In general, the number of equations to which an
assumption about forces leads is not equal to e
2n ÿ 2 as required for all equilibrium conditions
a
to be satis®ed. For example, the assumption in d

Bishop's simpli®ed method involves a total of


2n ÿ 1 equations. Thus the problem is overspeci- α
®ed, and in general overall horizontal equilibrium
is not satis®ed. Fortunately, it was found that the
factor of safety is not very sensitive to the value of
the interslice shear forces (Bishop, 1955). However,
in the Fellenius method, the factor of safety may c f
be underestimated by as much as 60% (Whitman
& Bailey, 1967) because the assumptions about the E2 w E1
interslice forces do not satisfy statics.
Besides the assumption about the forces, an L
e
α
assumption is required to be made about the slip
surface. In the Fellenius and Bishop methods, it is T
N′
assumed that failure occurs by rotation of a block d uL
of soil on a cylindrical slip surface. In the methods
of Janbu (1957), Morgenstern & Price (1965) and Fig 1. Bishop's simpli®ed method of slices
STABILITY ANALYSIS OF EMBANKMENTS 863
In the limit equilibrium condition, the overturning described by Lade (1977) for cohesionless soil,
moment is equal to the resisting moment. Let the methods of slice would not be ef®cient for the
moment of equation (34) be equal to that of (35); stability analysis of soil structures.
the factor of safety can be obtained by solving the
following non-linear equation:
P NUMERICAL EXAMPLES AND COMPARISON OF
(cL ‡ tan öN 9) THE RESULTS
Fs ˆ P (37)
w sin á Four numerical examples are presented in this
section. The ®nite element procedure of kinematic
The minimum factor of safety is obtained by trying analysis was implemented in CESAR-LCPC, a ®nite
different slip circles and taking partial derivatives element software package developed by the Central
of Fs as zero. Laboratory of Bridges and Highways, France. Cal-
Although the methods of slices have known a culations were carried out using an IBM RISC=
great success in the stability analysis of slopes and 6000 system. The computation for a plane strain
embankments, the following points should be worth problem (Example 4) with 1036 degrees of free-
noting when a method of slices is used. dom and 968 triangular elements took 78 s of CPU
time. With the fast microprocessors now available,
(a) As pointed out by Nash (1987), none of these
such a typical calculation can be completed in a
methods will yield the correct value of the
mere several minutes in a PC equipped with 16MB
factor of safety because of the assumptions
memory or higher. When the program is totally
involved about the forces
optimized, it can be hoped that the computation
(b) There is no proof that the stresses outside the
time can be further reduced considerably.
tested mobilized mass do not violate the yield
criterion. In fact, it can be observed from the
examples in this paper that plastic zones
Example 1
obtained by limit analysis often extended
Figure 2 shows an embankment of clay consid-
outside the tested mobilized mass given by
ered by Low (1989) for a semi-analytical method
the limit equilibrium method, even if the
of stability analysis of embankments on soft
corresponding factor of safety by limit analysis
ground. The undrained strengths of the upper, mid-
is less than that by the method of slices
dle and lower strata are respectively 30, 20 and
(c) The assumption of circular slip interfaces in
150 kN=m2 . The unit weight of the three layers is
the Fellenius and the Bishop methods is not
㠈 18 kN=m3 . Each layer has a thickness of
suitable for structures constituted of materials
4´5 m. The height of the slope is 6 m and the slope
of strong heterogeneity. The Morgenstern and
angle á is given by tan á ˆ 13. Fig. 3 illustrates the
Price method can be more useful for such
problems
(d) It is not certain that the load corresponding to
a factor of safety of unity is higher or lower 4.5 m c 5 30 kPa
3
1
than the actual limit load 6m
(e) It is virtually assumed in methods of slices that
4.5 m c 5 20 kPa
soil masses in the ultimate limit state fail in
shear and the linear Mohr±Coulomb criterion
4.5 m c 5 150 kPa
can be used. For soils that are modelled by
more complex criterion functions of non-linear
form, or curved yield surfaces such as the one Fig 2. An embankment of three layers (Example 1)

Fixed boundary

Fig 3. Finite element mesh of the embankment, and boundary conditions (Example 1)
864 JIANG AND MAGNAN

®nite element mesh and the boundary conditions. slices and Bishop's simpli®ed method. The present
The failure mechanism is represented by the nodal numerical result is 2´8% higher than this upper
velocities in Fig. 4 and the principal strain rates in bound solution.
Fig. 5. From these two ®gures, it can be observed The closeness of these results may be explained
that the failure mechanism agrees well with the by the fact that the optimal slip surface obtained
critical slip circle given by Low since no failure by the methods of slices is similar to the failure
occurs in the lower stratum and the failure zone mechanism provided by limit analysis since the slip
passes through the base of the middle layer. surface completely passes through a very narrow
Low's semi-analytical method gave a factor of shear strain band (Fig. 5) and no distortion occurs
safety of Fs ˆ 1:45, compared with a value of 1´44 in the slide mass. This example demonstrates that
by the ordinary method of slices and Bishop's for purely cohesive soils both limit analysis, and
simpli®ed method. The present numerical calcula- the Fellenius method and Bishop's simpli®ed meth-
tion gave a factor of safety Fs ˆ 1:48. It should be od can give very accurate factors of safety if the
noted that the limit loads approached by the pre- failure mechanism is a simple rotational mechan-
sent method are approached from the upper bound ism.
side, so it is certain that the actual limit load is not
higher than the numerical result. In terms of the
factor of safety, this means a collapse will happen Example 2
if the strengths of the soils are reduced by 1=1´48 Another example can also be found in the paper
times. A hand calculation by the upper bound of Low (1989). Fig. 6 shows an embankment of
method was undertaken using Low's optimized slip stiff clay which has been constructed on soft clay.
circle. The value of the corresponding factor of The height of the embankment and the thickness
safety is 1´44, which is identical to the value of the foundation layer are both 8 m. Other mater-
obtained by Low using the ordinary method of ial characteristics are shown in Fig. 6. The em-

Fig 4. Collapse velocities

Compression

Tension

Fig 5. Principal strain rates at collapse


STABILITY ANALYSIS OF EMBANKMENTS 865

c 5 95 kPa principal strain rates of the embankment and the


8 m φ 5 158 2 ®rst layer of the foundation soil. The failure strain
γ 5 20 kN/m3 1
15 kPa mechanism in the embankment is very complicated
4m
and there are very large horizontal shear deforma-
8m cu tions immediately under the embankment which
4m represent the effect of a strong horizontal thrust of
30 kPa the embankment to the foundation. In general,
zones of relatively important deformation represent
Fig 6. An embankment constructed on soft clay the zones where a plasti®ed state is reached earlier
(Example 2)
than other zones if a proportionally increased load-
ing condition is introduced.
Figure 9(a) shows that there is a large zone of
bankment is assumed to be a drained clay and the plasticity outside Low's slip circle even though the
foundation clay undrained. The ®nite element corresponding factor of safety by limit analysis is
mesh, with 1660 nodes and 3176 constant strain less than that given by Low. This demonstrates that
triangles, and the boundary conditions are shown it is not certain from the limit equilibrium analysis
in Fig. 7. The failure mechanism is illustrated by whether the zones outside the critical slip circle
the nodal velocities in Fig. 8 and the principal obtained by methods of slices are plasti®ed.
strain rates in Fig. 9. A value of 1´14 of the factor of safety was
Low's semi-analytical method gave a factor of obtained by Low using Bishop's simpli®ed method.
safety equal to 1´378. The ordinary method of The discrepancy between the results given by the
slices indicated a factor of safety of Fs ˆ 1:36; the Fellenius and Bishop's simpli®ed methods for ef-
present limit analysis gave a factor of safety fective analyses is well known (Whitman & Bailey,
Fs ˆ 1:25. In this example, limit analysis gave a 1967). The two methods are equivalent for a ö ˆ 0
limit load less than that obtained by Low. The analysis when the embankment strength is neglec-
reasons may be explained as follows. From Fig. ted, but they result in some differences when the
9(a), it can be seen that the failure deformation embankment contributes to resisting the failure and
mechanism is characterized by a very wide band of when ö 6ˆ 0 for the embankment material. The
deformation which cannot be modelled by a simple results by Low con®rm this conclusion since both
slip circle. Fig. 9(b) presents on a larger scale the methods gave a factor of safety of 1´31 when

Fixed boundary

Fig 7. Finite element mesh and boundary conditions of the embankment (Example 2)

Fig 8. Failure velocities by limit analysis and the critical slip circle by limit equilibrium
866 JIANG AND MAGNAN

0.014

(a)

0.005

(b)

Fig 9. Principal strain rates: (a) complete view; (b) zoom

ö ˆ 0 was speci®ed for the embankment. However, tangential to the top of the Lower Peat and to the
the present limit analysis method gave nearly the bottom of the Buttery Clay. The analyses were
same factor of safety for the embankment with a performed on total stress with the internal friction
friction angle ö ˆ 158. This may be explained by angle equal to 08. The factor of safety was
the fact that the cohesion strength of the embank- Fs ˆ 1:0 for a berm width of 6´1 m. A berm width
ment is so high compared with those of the of 12´2 m was employed in the ®nal design and the
foundation clays that the variation of its friction corresponding factor of safety Fs ˆ 1:2 was pre-
angle from 158 to 08 has relatively little effect on dicted by Skempton.
the stability. The mesh shown in Fig. 11 was used for the
limit analysis of the bank with a berm width of
6´1 m. Fig. 12 illustrates the failure velocities. Fig.
Example 3 13 presents the failure strain rates, which show that
Figure 10 shows the trial pro®les for a channel a very regular shear band is formed and accompa-
(Skempton, 1946; Henry, 1986). Berm widths of nied by a sub-failure zone in the ®ll. A substantial
6´1, 9´1, 12´2 and 15´2 m were investigated by portion of the slip zones is completely planar at
Skempton with various radii of slip circles, which the base of the Buttery Clay layer. Furthermore,
were assumed to pass through the toe and to be the shear band signi®cantly deviates from the
STABILITY ANALYSIS OF EMBANKMENTS 867
1 in
2
6
Berm
5 1 in .
4 25
3
2
1
Sand and gravel

Stratum No. Soil Thickness: m cu: kN/m2 γ: kN/m3


6 Fill 5.8 12.0 17.3
5 Brown silt 1.5 23.9 18.5
4 Grey silt 1.5 22.7 18.1
3 Upper Peat 0.6 33.5 11.0
2 Buttery Clay 4.3 22.7 16.0
1 Lower Peat 0.6 33.5 11.0

Fig 10. Design of a bank (after Skempton, 1946)

Fixed boundary

Fig 11. Mesh for the bank and boundary conditions

A B C

C
A
B

Fig 12. Failure velocities and slip surfaces: A, non-circular surface by the Morgenstern±Price method
(Chen); B, circular surface by the slip circle method (Skempton); C, circular surface by the Bishop
method (Chen)

critical slip circle given by Skempton. Particularly, tion of Skempton that the failure does not pene-
Fig. 13 shows that the main plastic failure zone is trate the Lower Peat on account of its relatively
located outside the critical slip circle given by high strength. The limit analysis gave a factor of
Skempton. The calculation con®rmed the assump- safety of only Fs ˆ 0:905, which shows signi®cant
868 JIANG AND MAGNAN

A B C

C
A
B

Fig 13. Principal strain rates and slip surfaces

variation over Skempton's factor Fs ˆ 1:0. How- since he obtained a minimum factor of safety of
ever, the present result agrees very well with the 1´068 using the Bishop method and 0´974 using
factors of safety obtained by Z. Y. Chen using the Morgenstern & Price method.
Bishop's simpli®ed method and the Morgenstern &
Price method, which are 0´899 and 0´864 respec-
tively. In Chen's calculations, the simplex method
of optimization was used to obtain the minimum Example 4: bearing capacity of a strip foundation
factor of safety, and optimal cubic splines (Chen & on weightless soil
Shao, 1988) were adopted to construct smooth The problem of the bearing capacity has been
non-circular slip surfaces in the Morgenstern & investigated by a large number of researchers using
Price method. The non-circular slip surface ob- various approaches. An interesting comparison
tained by Chen is totally located inside the failure study was reported by Brinch Hansen (1966) (see
band of plasticity and can be said to be approxi- also Nash, 1987). The problem with which the
mately parallel to the shear sliding direction. The former author dealt is a strip foundation on weight-
optimal slip circle obtained by Chen is, in a sub- less and cohesionless soil with a friction angle of
stantial portion, nearer to the shear band than 308. A uniform unit surcharge was prescribed and
Skempton's but does not pass through the toe. the exact solution was given by Prandtl using
The limit analysis gave a factor of safety plasticity theory. The limit load qf can be ex-
Fs ˆ 1:007 for the bank with a 12´2 m berm width; pressed by
this is signi®cantly less than the factor of safety qf ˆ N q s (38)
Fs ˆ 1:2 given by Skempton. Once again, the re-
sults of Chen are close to the limit analysis result where s is the uniform surface load (Fig. 14) and

q
s

Fig 14. Mesh for a strip foundation, and boundary conditions


STABILITY ANALYSIS OF EMBANKMENTS 869
N q is the bearing capacity factor. With the above DISCUSSION
soil strength properties and loading conditions, the In the past, various de®nitions of factor of safety
exact solution for N q is 18´4 while the numerical have been proposed and there has often been con-
calculation gave a result of 19´7. Note that the troversy about these de®nitions. Two categories of
associated ¯ow rule of plasticity cannot be applied de®nitions can be described as follows:
to kinematical analysis of cohesionless soils be-
cause the plastic dissipation function vanishes. (a) The de®nition using the mobilized shear
However, an upper bound to the limit load can be strength, by equation (23). Most methods of
obtained easily by arti®cially assigning a positive slices use this de®nition and it is therefore the
value slightly higher than zero to the cohesion. In most common one used in stability analysis of
the present calculation, a cohesion of 0´01 was slopes.
speci®ed so that the numerical procedure could be (b) The de®nition using the weight of soils. In
used. Chen & Snitbhan (1975), the weight was
It can be observed from Table 1 that there is no de®ned as that of sliding mass encompassed
evident relationship between the bearing capacity by free surfaces and the sliding surface. Such a
factor N q and the conventional factor of safety Fs. de®nition was also given by other authors. In
Another calculation was then made to obtain the limit analysis, the ratio of the collapse load to
factor of safety, which is equal to 1´03. the actual load can be considered as this type
These two calculations permit the conclusion that of de®nition.
both the bearing capacity factor and the safety factor
are much closer to the exact solution than the re- Another possibility to assess the stability of
sults by other methods. Moreover, the failure mech- slopes is to de®ne a stability factor such as that
anism obtained by the present analysis is in good proposed by Taylor (1937). Such de®nition has the
agreement with that of Prandtl (Figs 15 and 16). drawback that it is only suitable in situations of

Table 1. Comparison of bearing capacity factors determined by different methods


Author Method Rupture line Nq Percentage of Fs
N q Prandtl
Fellenius Slices Circle 5´6 30 0´58
Krey Friction circle Circle 9´0 49 0´75
Janbu 2 Slices Prandtl 13´0 71 0´92
Prandtl Plasticity Prandtl 18´4 100 1´00
Present Limit analysis Velocity ®eld 19´7 107 1´03
Brinch Hansen Equilibrium Circle 22´5 122 1´07
Rendulic Extreme Spiral 22´6 123 1´07
Janbu 1 Slices Circle 27´9 152 1´21
Bishop Slices Circle 31´1 169 1´19

C
L

Fellenius
Hansen
Bishop, Janbu 1
Prandtl, Janbu 2

Fig 15. Failure velocities, Prandtl mechanism, and slip surfaces by limit equilibrium methods
870 JIANG AND MAGNAN

Fig 16. Principal strain rates and Prandtl mechanism

relatively simple geometry and strength properties can be encountered in the stability analysis of
of soil (homogeneous, for example). shallow tunnels, where the least tunnel pressure is
Although the ®rst de®nition has been used ex- calculated to prevent collapse, or in seismic earth
tensively in geotechnical practice, it is sometimes pressure, where the resultant seismic coef®cient is
considered ambiguous in physical meaning for ef- to be assessed.
fective stress calculations (Jardine & Hight, 1987). All the results of calculation are summarized in
Indeed, Example 4 shows that the ®rst factor of Table 2 and are compared with those obtained by
safety is less good than the second de®nition be- other methods. In the de®nition of the limit load in
cause, for instance, a factor of safety of 1´19 (the limit analysis is admitted to be exact for stability
Bishop method) by the ®rst de®nition corresponds analysis, it can be concluded that present analyses
in fact to 1´69 times the exact collapse load, so the in most cases give numerical results as good as or
®rst de®nition is a poor indicator of bearing capa- better than those based on limit equilibrium analy-
city. Despite these considerations, it is not to be sis, since the upper bounds (F 2s ) are often less than
concluded here that one de®nition prevails over or close to the values (F 1s ) obtained by other meth-
another in the stability analysis of slopes and ods. From this point of view, it can be said, at least
embankments. If possible, all these de®nitions for the examples discussed, that the limit equili-
should be considered in practice. However, the brium method often overestimates the factor of
second de®nition is more useful in cases where a safety and gives more dangerous results than those
limit load factor is to be determined. Such cases predicted by accurate limit analysis. Nevertheless,

Table 2. Summary of the results of limit analysis and comparison with the results of limit equilibrium analysis
Case Method of limit Author F 1s from F 2s from F 1s ÿ F 2s
equilibrium limit limit F 2s
equilibrium analysis
Example 1 Semi-analytical Low 1´45 1´48 ÿ2´0%
Fellenius and Bishop Low 1´44 ÿ2´7%
Example 2, with ö ˆ 158 Semi-analytical Low 1´38 10´4%
embankment Fellenius Low 1´36 1´25 8´8%
Bishop Low 1´14 ÿ8´8%
Example 2, with Fellenius and Bishop Low 1´31 1´25 4´8%
ö ˆ 08 embankment
Example 3, with 6´1 m berm Slip circle Skempton 1´0 10´5%
Morgenstern & Price Chen 0´864 0´905 ÿ4´5%
Bishop Chen 0´899 ÿ0´7%
Example 3, with 12´2 m berm Slip circle Skempton 1´2 19´2%
Morgenstern & Price Chen 0´974 1´007 ÿ3´3%
Bishop Chen 1´068 6´1%
STABILITY ANALYSIS OF EMBANKMENTS 871
further comparisons between these two classes of by Z. Y. Chen, who used the Morgenstern & Price
methods may be necessary if more con®dence is to method combined with an advanced optimization
be placed on limit analysis for practical use in technique. For the calculation of bearing capacity,
geotechnical engineering. ®nite element limit analysis led to a result much
In hand calculation of the stability by limit closer to the analytical solution than all factors of
analysis, it is generally dif®cult to deal with soil safety presented by Brinch Hansen using limit
masses containing simultaneously ö ˆ 0 and c ÿ ö equilibrium methods.
soils. By the present ®nite element method, such The kinematic limit analysis used in this paper
problems can be treated as easily as problems of gives an upper bound for the limit load. The
homogeneous material. Another advantage of ®nite results obtained are higher than or equal to the
element limit analysis is that curved yield criteria actual limit loads. The difference between the
can also be incorporated in the numerical proce- numerical and actual results generally does not
dure. It is useful to take into account such failure exceed 6% of the actual limit loads in our experi-
surfaces because curvature of the envelopes for ence. This error is produced essentially by two
soils has been observed by many investigators (for factors. The ®rst one is that the kinematically
example, Lade, 1977). Many materials, such as admissible velocities used in the numerical scheme
cohesionless soils and rocks, have a non-linear are kinematically admissible velocities of C 0 con-
yield surface. The use of the upper bound method tinuity, that is, the discretized velocity function is
to evaluate the stability of slopes with a non-linear continuous but the strains are discontinuous. Dis-
yield surface was introduced by Baker & Frydman continuous failure velocities and velocities other
(1983) and was extended by Zhang & Chen than C 0 continuity are excluded. The second factor
(1987). For practical problems with curved yield is due to the error of the numerical approximation
surfaces, limit analysis by the ®nite element meth- to the real in®nite minimization of functionals. No
od would be more advantageous than other meth- matter how much the difference between the nu-
ods. Stresses in the failure zones can also be merical and actual results, it is sure that the present
determined as a by-product of the upper bound results give upper bounds for the limit loads.
calculation and the associated ¯ow rule if a strictly Another advantage of limit analysis over methods
convex curved yield criterion (without any ¯at part of slices is that the ®nite element solution of
of the yield surface) is used. kinematic limit analysis provides failure mechan-
isms that may not be appropriately simulated by
methods of slices. In fact, limit analysis permits
CONCLUSIONS calculation of a variety of realistic critical failure
Limit analysis is a universal method for correct mechanisms which are often composed of plane,
and accurate solution of the stability problem. An circular and=or logarithmic rupture bands and fail-
ef®cient and accurate numerical technique is vital ed soil mass.
to make limit analysis applicable to complicated
geotechnical and structural problems of stability.
To this end, a new ®nite element method of limit ACKNOWLEDGEMENT
analysis has been developed to determine the limit The authors are grateful to Dr Z. Y. Chen of the
load or the factor of safety. This method does China Institute of Water Conservancy and Hydro-
not need the engineer to assume the failure mech- electric Power Research for carrying out the calcu-
anisms because it can automatically ®nd the criti- lation of minimum factors of safety in Example 3
cal failure mechanism which gives the least re- using methods of slices.
sistance of the structure to the applied loads.
Various examples are presented to compare the
factors of safety obtained by limit analysis and REFERENCES
methods of slices. It is demonstrated that limit Anderheggen, E. & KnoÈpfel, H. (1972). Finite element
analysis can be perfectly well applied to calculate limit analysis using linear programming. Int. J. Solids
the stability of embankments and the results are and Structures 8, No. 12, 1413±1431.
often better than or as good as those given by Baker, R. & Frydman, S. (1983). Upper bound limit
methods of slices. analysis of soil with non-linear criterion. Soils Fdns
In two examples, limit analysis gave smaller 23, No. 4, 34±42.
factors of safety than previous results based on Bishop, A. W. (1955). The use of the slip circle in the
stability analysis of slopes. GeÂotechnique 5, No. 1, 7±
limit equilibrium analyses, and the factor of safety
17.
obtained in the ®rst example is only 2´8% higher Brinch Hansen, J. (1966). Comparison of methods for
than that given by Low using the Fellenius method stability analysis. Bulletin No. 21 Danish Geotechni-
and the Bishop's simpli®ed method. However, in cal Institute.
the third example, ®nite element solutions gave Chen, W. F. (1975). Limit analysis and soil plasticity.
factors of safety slightly higher than those obtained Amsterdam: Elsevier.
872 JIANG AND MAGNAN

Chen, W. F. & Snitbhan, N. (1975). On slip surface and of kinematic limit analysis. Int. J. Num. Meth. Engng
slope stability analysis. Soils Fdns 15, No. 3, 41±49. 38, No. 16, 2775±2807.
Chen, Z. Y. & Morgenstern, N. R. (1983). Extensions to Lade, P. V. (1977). Elasto-plastic stress±strain theory for
the generalized method of slices for stability analysis. cohesionless soil with curved yield surfaces. Int. J.
Can. Geotech, J. 20, No. 1, 104±119. Solids Struct. 13, 1019±1035.
Chen, Z. Y. & Shao, C. M. (1988). Evaluation of Low, B. K. (1989). Stability analysis of embankments on
minimum factor of safety in slope stability analysis. soft ground. J. Geotech. Engng, ASCE 115, No. 2,
Can. Geotech, J. 25, No. 4, 735±748. 211±227.
Chen, Z. Y. (1992). Random trials used in determining Morgenstern, N. R. & Price, V. E. (1965). The analysis of
global minimum factor of safety. Can. Geotech J. 29, the stability of general slip surface. GeÂotechnique 15,
No. 1, 225±233. No. 1, 79±93.
Fredlund, D. G. & Krahn, J. (1977). Comparison of slope Nash, D. (1987). A comparative review of limit equili-
stability methods of analysis. Can. Geotech. J. 14, brium methods of stability analysis. In Slope stability
No. 3, 429±439. (eds M. G. Anderson and K. S. Richards). New York:
Henry, F. D. C. (1986). The design and construction of Wiley.
engineering foundations 2nd edn. London: Chapman SalencËon, J. (1983). Calcul aÁ la rupture et analyse limite.
& Hall. Paris, Presses ENPC.
Janbu, N. (1957). Earth pressures and bearing capacity Skempton, A. W. (1946). Earth pressure and the stabil-
calculations by generalized procedure of slices. Proc. ity of slopes. In The principles and application
4th Int. Conf. Soil. Mech. Fdn Engng, Vol. 2, pp. of soil mechanics. London: Institution of Civil
207±212. London: Butterworths. Engineers.
Jardine, R. J. & Hight, D. W. (1987). The behaviour and Spencer, E. (1967). A method of analysis of the stability
analysis of embankments on soft clay. In Embank- of embankments assuming parallel inter-slice forces.
ments of soft clay, Bulletin of the Public Works GeÂotechnique 17, No. 1, 11±26.
Research Centre, Athens, pp. 159±244. Taylor, D. W. (1937). Stability of earth slopes. J. Boston
Jiang, G. L. (1992). Application de l'analyse limite aÁ Soc. Civ. Engrs 24, No. 3, 197±246.
l'eÂtude de la stabilite des massifs de sol. PhD thesis, Whitman, R. V. & Bailey, W. A. (1967). Use of
Ecole Nationale des Ponts et ChausseÂes (ENPC), computers for slope stability analysis. J. Soil Mech,
Paris. Fdns Div., ASCE 93, No. 4, 475±498.
Jiang, G. L. (1994). Regularized method in limit analysis. Zhang, X. J. & Chen, W. F. (1987). Stability analysis of
J. Engng Mech., ASCE 120, No. 6, 1179±1197. slopes with general nonlinear failure criterion. Int. J.
Jiang, G. L. (1995). Non-linear ®nite element formulation Num. Anal. Meth. Geomech, 11, No. 1, 33±55.

Vous aimerez peut-être aussi