You are on page 1of 4

Krabbenhoft, K. and Lyamin, A. V. (2015) Géotechnique Letters 5, 250–253, http://dx.doi.org/10.1680/jgele.15.

00110

Strength reduction finite-element limit analysis


K. KRABBENHOFT* and A. V. LYAMIN*

A procedure for strength reduction analysis using finite-element limit analysis is presented. The
scheme is completely general and does not require decision making regarding the loads needed to
drive the system to failure. Rather, the scheme is based on the ability of modern interior-point methods
to detect infeasibility in a controlled and reliable manner. The new scheme is illustrated by an example
involving a strip footing on top of a slope.

KEYWORDS: limit state design/analysis

ICE Publishing: all rights reserved

NOTATION While classical limit analysis encompasses both these ques-


BT discrete equilibrium operator tions, most work to date has revolved around the first
b body forces acting in the volume V question. In practice, however, the latter question is often
c cohesion the more relevant one. Indeed, for slopes and retaining
cred reduced cohesion implying state of incipient collapse systems not necessarily subjected to any loads of a structural
F yield function
nσ total number of stress points in domain
origin, the task of determining the strength reduction factor
p, p0 vectors of forces leading to a state of incipient collapse is often the only
q load to which footing is subjected relevant one.
Sσ part of surface From a technical computational point of view, the load
t tractions factor problem is significantly easier than the strength
u displacement reduction problem. Conceptually, it is a variant of the
V volume fundamental problem of determining the response of a
α load multiplier system subjected to external loading. Virtually all compu-
ε strain tational schemes, the finite-element method for example,
λ plastic multiplier
have been conceived with this problem in mind and the
σ stresses
σ̂ j nodal stresses extension required to accommodate the load factor problem
ϕ friction angle is conceptually trivial. Moreover, from a mathematical point
ϕred reduced friction angle implying state of incipient collapse of view, the load factor problem does not add any real
7 strain-displacement operator complexity, whereas the strength reduction problem more
7T equilibrium operator often than not gives rise to problems that are non-convex and
therefore are problematic to deal with. This issue is
particularly relevant in computational limit analysis, which
INTRODUCTION almost always relies on optimisation techniques that assume
Ultimate limit state (ULS) design is concerned with deter- or require convexity of the problem solved.
mining the necessary structural dimensions and/or strengths So far, the very limited work on the strength reduction
that, subject to a certain level of safety, will imply a state of problem in the context of computational limit analysis
incipient plastic collapse. When using numerical procedures makes use of a variant of the load factor problem to
for ULS design, an iterative scheme where the structural determine the strength reduction factor iteratively (e.g. as
dimensions (e.g. footing width or retaining wall depth) are reported by Sloan (2011)). This scheme has been applied
assumed fixed is usually employed. The integrity of the soil– only to slope stability where it arguably performs well.
structure system can then be checked and the dimensions Considering a slope with cohesion c and friction angle ϕ and
adjusted until a satisfactory design is obtained. The struc- not subjected to any external loads other than those due to
tural integrity check can usually be formulated in terms of selfweight, the procedure is as follows. For fixed c and ϕ,
one of the following questions. determine the factor α by which the selfweight should be
magnified to induce a state of incipient collapse. If α > 1, the
(a) Given the material strengths, by what factor should
slope is stable and c and tan ϕ are reduced, whereas, if α < 1,
the external loads be magnified to induce a state of
the slope is unstable and c and tan ϕ are increased. In both
incipient collapse?
cases, a new load factor is computed using the adjusted c and
(b) Given the external loads, by what factor should
ϕ and the process is repeated until α is sufficiently close to
the material strengths be reduced to induce a state of
1. The resulting strength reduction factor is usually defined
incipient collapse?
as the factor of safety

c tan ϕ
FS ¼ ¼
Manuscript received 22 July 2015; first decision 7 September cred tan ϕred
2015; accepted 18 September 2015.
Published online at www.geotechniqueletters.com on 29 where cred and ϕred are the reduced parameters implying a
October 2015. state of incipient collapse. Note that the above definition of
*Centre for Geotechnical Science and Engineering, University of FS relies on c and tan ϕ being reduced by the same factor in
Newcastle, NSW, Australia each iteration.

250

Downloaded by [ TUFTS UNIVERSITY] on [25/09/16]. Copyright © ICE Publishing, all rights reserved.
Strength reduction finite-element limit analysis 251
The main drawback of the above scheme is a lack of This problem reproduces the governing equations (2)–(5).
generality. For foundation problems, for example, the self- That is, solving the optimisation problem is equivalent to
weight would need to be minimised (rather than maximised solving the governing equations.
as for slopes) in order to induce a state of incipient collapse. In a finite-element context, the governing equations may
In practice, therefore, the sense of the optimisation (maxi- be discretised by introducing appropriate approximations for
mise or minimise) would need to be adjusted according to the variables involved. The end result is a discrete optimis-
the particular problem. It could of course be argued that, in ation problem of the type (Anderheggen & Knöpfel, 1972;
the case of foundations, the load acting on the foundation, Bottero et al., 1980; Zouain et al., 1993)
rather than the soil unit weight, should be used to drive the
system to collapse. In any case, the principle of driving the maximise α
system to collapse by adjusting the magnitude of the loads subject to B T σ̂ ¼ αp þ p0 ð7Þ
and/or selfweight requires decision making with respect to 
F σ̂ j  0; j ¼ 1; . . . ; nσ
the problem at hand. Moreover, for some problems the appro-
priate decision is not obvious. For example, for a foundation where σ̂ j are the nodal stresses (with the total nodal stress
on top of a slope, both the external load and the soil vector being σ̂ ¼ ðσ̂; . . . ; σ̂ nσ ÞT ), B T is a discrete equilibrium
selfweight (or a combination of the two) could be used to type operator, p and p0 are vectors of forces, and nσ is the
drive the system to collapse. total number of stress points in the domain. Depending
In general, therefore, the approach of using one or on the particular type of discretisation, the solution to
more loads to drive the system to failure and adjust the equation (7) may or may not bound the exact solution from
strength reduction factor accordingly is not suitable. In the above or below in a rigorous manner (Anderheggen &
following, a general scheme that does not require any a priori Knöpfel, 1972; Bottero et al., 1980; Zouain et al., 1993). It is
decision making is outlined. The scheme makes use of the noted that the constitutive model used, rigid perfect plas-
ability of modern optimisation algorithms, in particular ticity with associated flow, is mathematically well posed
interior-point algorithms for conic programming, to detect and does not require any particular considerations of the
infeasibility in a controlled fashion. The new scheme is finite-element mesh in relation to strain localisation.
illustrated with an example involving a strip footing on top
of a slope.

Strength reduction limit analysis


Consider now the case where a system is subjected to a set of
LIMIT ANALYSIS loads p0 and the task is to determine whether or not a sol-
The basic physical assumptions of limit analysis are that ution corresponding to this situation exists. That can be done
the material behaves in a rigid/perfectly plastic manner by solving, or attempting to solve, the following optimisation
and that the deformations at incipient collapse are of a problem
magnitude that justifies the use of conventional small-strain
theory. Under these assumptions, the aim is to determine a maximise 0
solution that satisfies the equilibrium and static boundary subject to B T σ̂ ¼ p0 ð8Þ
conditions 
F σ̂ j  0; j ¼ 1; . . . ; nσ
7T σ þ b ¼ 0 in V
ð2Þ where the choice of a constant (0) as objective function
N σ¼t
T
on Sσ simply implies that the objective is not to optimise a function
the yield condition but rather to determine a feasible solution; that is, a solution
that satisfies the discrete equilibrium and yield constraints.
F ðσ Þ  0 ð3Þ If such a solution can be found, the problem is stable for
the material strengths assumed. If not, the opposite con-
the associated flow rule assuming infinitesimal strains clusion may be drawn. Since any feasible solution is optimal,
ε ¼ 7u ¼ λrF ðσ Þ ð4Þ the task is to determine whether the problem is feasible
(physically stable) or infeasible (physically unstable). This is
and the complementary conditions a fairly routine analysis and in fact an integral part of most
λF ðσ Þ ¼ 0; λ  0 ð5Þ modern interior-point methods for convex programming,
for example, the homogeneous interior-point algorithm for
In the above equations, σ are the stresses, 7T is the second-order cone programming described by Andersen
equilibrium operator (7 being the strain-displacement et al. (2003). It is important to note here that infeasibility is
operator), b are the body forces acting in the volume V, detected in an entirely controlled manner without relying on
and t are the tractions acting on part of the surface Sσ. The various measures of non-convergence. In fact, for infeasi-
yield function is given by F and the strain and displacements bility to be declared, the algorithm is required to converge
are denoted by ε and u, respectively. and the final state allows one to gauge whether the problem
The governing equations can be cast in terms of a number is infeasible or whether the solution is optimal (and thereby
of equivalent variational principles. Consider for example feasible). Moreover, infeasibility is often detected at a sig-
the case where the maximum magnitude of the tractions, nificantly lower cost than that of finding an optimal solution
t, is to be determined. This problem may be cast in terms to a feasible problem. Relatively strict convergence criteria,
of the following variational principle (or optimisation along the lines of Andersen et al. (2003), are used to ensure
problem) that infeasible problems are not falsely declared feasible and
vice versa.
maximise α With this possibility of controlled feasibility detection,
subject to 7T σ þ b ¼ 0 in V an algorithm for strength reduction limit analysis is straight-
ð6Þ forward to construct. One possibility is as follows. First
N T σ ¼ αt on Sσ
assume a minimum factor of safety FSmin=0 and a maximum
F ðσ Þ  0 factor of safety FSmax equal to some appropriately large

Downloaded by [ TUFTS UNIVERSITY] on [25/09/16]. Copyright © ICE Publishing, all rights reserved.
252 Krabbenhoft and Lyamin
Table 1. Algorithm for strength reduction limit analysis 1·8

FSmin : = 0, FSmax : = Inf, FS : = 1


while (FSmax-FSmin)/FS > TOL
1·6
cred = c/FS, arctan(tan( phi)/FS)
FeasibilityCheck(cred, phired)

Factor of safety, FS
if feasible
FSmin : = FS 1·4 Lower Upper
FS : = 1/[(1/FSmin + 1/FSmax)/2] bound bound
else
FSmax : = FS 1·2
FS : = (FSmin + FSmax)/2
end
end 1·0

0·8
0 100 200 300 400 500
q
Footing load, q: kN/m2
6m
1m
1m Fig. 2. Factors of safety (upper and lower bounds) as function of
footing load
2m 6m

30° TOL = 0 · 01. More generally, the algorithm converges


linearly with a rate of convergence of 0·5.
Fig. 1. Strip footing on slope

EXAMPLE
number within the range of machine precision. Then, In the following, an example demonstrating the capa-
starting from an initial factor of safety FS = 1, calculate bilities of the scheme is briefly discussed. The procedures
reduced parameters described in the previous section have been implemented in
the general-purpose geotechnical finite-element package
cred ¼ c=FS; ϕred ¼ arctan ½ðtan ϕÞ=FS ð9Þ OptumG2 (OptumCE, 2015) which is used to resolve the
example problem.
For these parameters, perform a feasibility check. If the The example involves a strip footing on top of a slope, as
problem is feasible, set FSmin = FS and calculate a new factor shown in Fig. 1. The soil is of the Mohr–Coulomb type with
of safety by cohesion c = 10 kPa, friction angle ϕ = 22° and unit weight
 1 γ = 20 kN/m2. The footing is perfectly rigid and has a unit
1  weight of 23 kN/m2. The footing is subjected to a load, q,
FS ¼ FS1 1
min þ FSmax ð10Þ
2 which in the following will be varied.
Conducting upper and lower bound strength reduction
Otherwise, if the problem is infeasible, set FSmax = FS and analysis as outlined in the previous section gives the factor of
calculate a new factor of safety by safety plotted against footing load diagrams shown in Fig. 2.
It can be seen that the factor of safety is unaffected by the
1 footing load up to about q ’ 75 kN/m2. This is reflected in
FS ¼ ðFSmin þ FSmax Þ ð11Þ
2 the corresponding collapse mechanism, which for suffi-
ciently small footing loads does not involve the footing at all,
The use of the harmonic (equation (10)) and arithmetic whereas, for larger loads, the failure mechanism extends
(equation (11)) means in the two cases is somewhat from the footing to the toe of the slope (see Fig. 3). Finally,
arbitrary and these have been chosen mainly because from the results shown in Fig. 2 it can be inferred that
they generalise to all possible values of FSmin and FSmax, the bearing capacity of the footing (corresponding to FS = 1)
including FSmax = ∞ in equation (10) and FSmin = 0 in is approximately 400 kN/m2. This is some 80% of the
equation (11). bearing capacity for a footing at an infinite distance from
This iterative process continues until convergence as the slope.
defined by

FSmax  FSmin
, TOL ð12Þ CONCLUSIONS
FS
A procedure for strength reduction analysis using finite-
where a tolerance TOL ’ 0·01 is appropriate for most element limit analysis has been presented. The scheme is
practical applications. completely general and does not require decision making
The full algorithm is shown in Table 1. It should be noted regarding the loads needed to drive the system to failure.
that the algorithm exits with two strength reduction factors, Rather, the scheme is based on the ability of modern interior-
FSmax and FSmin. If the problem is discretised to produce point methods to detect infeasibility in a controlled and
lower bounds, FSmin represents a rigorous lower bound on reliable manner. The new scheme is illustrated by an example
the factor of safety. Conversely, if the problem is discretised involving a strip footing on top of a slope. The extension of
to produce rigorous upper bounds, FSmax represents a the scheme to consider strength reduction with respect to
rigorous upper bound on the factor of safety. For typical structural members (e.g. sheet pile walls, anchors and so on)
ranges of factors of safety (FS ’ 0·2–5) the algorithm rather than the soil with which they interact is straightfor-
converges in about ten iterations for a tolerance of ward. Moreover, by relevant definition of the factor of safety,

Downloaded by [ TUFTS UNIVERSITY] on [25/09/16]. Copyright © ICE Publishing, all rights reserved.
Strength reduction finite-element limit analysis 253

(a)

(b)

Fig. 3. Collapse mechanisms (lower bound) for (a) q = 50 kN/m2 and (b) q = 200 kN/m2

the scheme is applicable to failure criteria other than Bottero, A., Negre, R., Pastor, J. & Turgeman, S. (1980). Finite
Mohr–Coulomb, for example, Hoek–Brown. element method and limit analysis theory for soil mechanics
problems. Comput. Methods Appl. Mech. Engng 22, No. 1,
131–149.
REFERENCES OptumCE (2015). See www.optumce.com. Copenhagen, Denmark:
Anderheggen, E. & Knöpfel, H. (1972). Finite element limit OptumCE (accessed 24/09/2015).
analysis using linear programming. Int. J. Solids Structs 8, Sloan, S. W. (2011). Geotechnical stability analysis. Géotechnique
No. 12, 1413–1431. 63, No. 7, 531–571.
Andersen, E. D., Roos, C. & Terlay, T. (2003). On implementing a Zouain, N., Herskovits, J., Borges, L. A. & Feijoo, R. A. (1993). An
primal-dual interior-point method for conic quadratic optimiz- iterative algorithm for limit analysis with nonlinear yield
ation. Mathl Programming 95, No. 2, 249–277. functions. Int. J. Solids Structs 30, No. 10, 1397–1417.

WHAT DO YOU THINK?


To discuss this paper, please email up to 500 words to the
editor at journals@ice.org.uk. Your contribution will be
forwarded to the author(s) for a reply and, if considered
appropriate by the editorial panel, will be published as a
discussion.

Downloaded by [ TUFTS UNIVERSITY] on [25/09/16]. Copyright © ICE Publishing, all rights reserved.

You might also like