-
Trigging Instabilities in Geomaterials within the Finite Element
Method Framework 1. INTRODUCTION
Instability manifests itself through various defor-mation modes
in geomechanics. For example, a natural slope may undergo large
movements as a result of deformations either localizing into shear
bands or developing in a diffuse manner through-out the entire
mass. In both cases, the problem underlies a material instability
phenomenon that originates in the small scale due to
microstructural features of the geomaterial. In the case of
localiza-tion, one finds distinctive forms of concentrated
deformations such as shear bands, compaction bands and dilation
bands [1-2]. These bands lead to an unstable response that is
associated to a bi-furcation phenomenon at the material point
level. That type of instability usually takes place near limit
failure conditions. Paradoxically, another type of instability may
occur well within the plas-tic limit surface [3]. For instance,
when loose sand is sheared in undrained conditions, it may collapse
at stress levels far from the plastic limit surface. In this form
of instability, normally coined as diffuse instability, a rather
generalized failure takes place. The implication is that stress
states deemed to be safe with respect to a limit condition or
localiza-tion can be still vulnerable to another type of
in-stability such as of the diffuse type.
One of the main objectives of this study is to establish
plausible models to numerically capture both localized and diffuse
instabilities and explore the non-uniqueness associated to the
underlying field equations in a boundary value problem. As
anticipated, non-uniqueness can be triggered by introducing small
perturbations in either the con-straints of the boundary value
problem or the ini-tial conditions. It is shown in plane strain
condi-tions that these small perturbations impact di-rectly on the
position of the shear band that is formed, and hence on the overall
response of the structure. Next, diffuse type of instability is
exam-ined under both axisymmetric boundary and load-ing conditions.
One aspect that is investigated is the role of the initial void
ratio distribution in triggering instability. Both homogeneous and
het-erogeneous void ratio distributions (uniform and Gaussian
types) are considered. A density-stress-fabric dependent
elasto-plastic model [4] is used to describe the constitutive
behaviour of the material. All simulations are carried within the
finite ele-ment framework.
2. DENSITY-STRESS-FABRIC DEPENDENT ELASTOPLASTIC MODEL
The ability of a constitutive model to describe soil behaviour
within a high degree of fidelity is crucial in any failure
analysis. A constitutive model for geomaterials must be able to
capture essential be-havioural features such as non-linearity,
irreversi-bility and dependencies of the mechanical behav-iour on
stress level, density and fabric. The model herein used has many of
the attributes described above. It was developed in [4] to describe
the me-chanical behaviour of granular materials, especially sands.
The core framework of this model, referred to as WG-model, is
recalled here.
The WG-model is a two-surface elastoplastic model that accounts
for both deviatoric and com-paction behaviours of granular
materials. It is founded on Rowe’s stress-dilatancy theory [5] for
predicting volumetric changes under deviatoric loading. Rowe’s
original theory has been further enriched to address density,
stress level and fabric dependencies as well as cyclic loading
regime con-ditions. The two surfaces of the WG-model are the
shear-yield surface f(s) and the cap-yield surface f(c), as shown
in Figure 1.
1!
2!
3!
(a) (b)
p0
q
c.s.l.
c.s.l.
TCM1
1"M
#p
$ %sf$ %cf
cp
Figure 1: Generalized WG-model: (a) Trace of yield surfaces in
the meridional plane. (b) Three-dimensional view in principal
stress space. The acronym c.s.l. stands for critical state line (or
surface).
The first yield surface treats deviatoric loading
governed by dilatancy whereas the second surface accounts solely
for isotropic loading producing plastic volumetric compressive
strains. A non-associated flow rule derived from the enriched
stress-dilatancy theory is used to calculate the in-crement of
plastic shear deformations. The update of both shear- and cap-yield
surfaces is controlled
-
2 by two distinct hardening-softening laws. For the shear-yield
surface, the mobilized friction angle acts as plastic hardening
variable, whereas for the cap-yield surface, the pre-consolidation
pressure, pc, governs plastic hardening. Table 1 summarizes the
equations used in the model.
Table 1: Main equations in WG-model for loading in-volving the
shear-yield surface
Yield sur-face
(s)f : q M p"& ' ; m
TCm
6 sinM :
3 sin
"&' "
TC2
M : M(1 ) (1 )t"
(&
) ( ' '(
Potential function
(s)g : q M p*
& ' ; m f
m f
sin sinM :
1 sin sin*" ' "
&' " "
fnp
Ff csp
cs0
esin : sin
e
+ ,- ) . /0 /0" & "/0 /0 /- ) . 1 2
Hardening law
mnp
m cspcs0
esin sin
e
'+ ,. /0 /0" & "/0 /0 /- ) . 1 2
Other evo-lution law
csn
cs cs0 cs0
pe : e exp h
p
3 4+ ,5 6/0 /& ' 05 6/0 //05 61 27 8
where p: mean effective stress; q: deviatoric stress; t = sin39
(9: Lode’s angle); "m, "f, "cs: friction angles mo-bilized, at
failure, and at critical state, respectively; e, ecs: void ratio at
current and at critical states, respec-tively; .p: deviatoric
plastic strain; ecs0, !, -F, -0, nf, nm, ncs and hcs are material
parameters; p0 = 1 kPa.
3. LOCALIZED AND DIFFUSE INSTABILITIES
Stability in its simplest form evokes the idea that a small
perturbation input or load results into a small-bounded output or
response [6]. In this re-gard, the well-known phenomenon of plastic
failure can be viewed as a problem of instability whereby a small
or even null stress input leads to large de-formations. Thus,
according to Lyapunov’s con-cept, a material would exhibit unstable
behaviour whenever its stress state reaches a certain limit state
defined by a failure criterion such as the Mohr-Coulomb. However,
experimental tests have shown that a material may be unstable even
before any failure criterion is violated [3, 7]. The next
sub-sections define briefly two such types of insta-bility in
geomechanics, namely, localized instabil-ity and diffuse
instability. 3.1 Localized Instability The localized stress-strain
response of geomaterials essentially results from their underlying
micro- and macro-mechanical properties. Similarly, boundary
conditions, imposed stress path, relative density and fabric are
among the most important factors that would trigger localization. A
great amount of experimental and numerical effort has been
ex-pended in an attempt to understand and weigh those factors [1,
8]. Mostly, from a theoretical standpoint, localized instability is
viewed as a bi-furcation of the underlying continuum field
equa-tions. In other words, an alternate mode of non-homogeneous
deformation is possible apart from the homogeneous one due to loss
of uniqueness of the governing incremental equilibrium equations.
The non-homogeneous mode involves a surface with a kinematic
discontinuity upon which interfa-cial slip occurs. By imposing
continuity of trac-tions across the discontinuity surface and the
same constitutive relationship throughout the body, Rudnicki &
Rice’s criterion [1] for strain localiza-tion emerges as:
det( : : ) 0&n D n (1)
where D is the tangent constitutive matrix in the sense that d :
d& D! " , and n is the normal to the discontinuity surface or
shear band.
3.2 Diffuse Instability In contrast to localized deformations,
diffuse in-stability is mathematically described using the
en-ergy-based Hill’s stability criterion which is de-fined by the
sign of the second-order work [9]. Whenever the second-order work
given by the product of incremental stress d! and strain d" becomes
negative at a material point during a loading increment,
instability springs. If the loss of positiveness of the
second-order work becomes pervasive within the structure, collapse
will even-tually occur. Locally, the second-order work crite-rion
is given by:
2W : d : d& ! " (2)
This notion of second-order work which in fact underlies a
bifurcation problem can be extended to highlight the directional
character of the loading-response behaviour of a material and its
relation-ship to instability. For instance, the domain of
stress-strain states for which at least one loading-response
direction exists such that W2 becomes non-positive gives rise to a
so-called bifurcation domain. This concept, though, has to be
brought hand-in-hand with that of controllability.
Control-lability, according to Nova [10], highlights the fact that
instability is also a function of the loading programme: force,
displacement, or mixed test control. To illustrate the raised
point, in force con-trolled tests, failure will not occur before
the plas-tic limit condition is reached.
-
3 4. FEM STUDY OF LOCALIZED AND DIFFUSE
INSTABILITY
In the following sub-sections, the occurrence of various
localized and diffuse instability modes is numerically studied in a
boundary value setting via finite element computations. The
density-stress-fabric dependent elasto-plastic model de-scribed in
Section 2 of this paper was implemented into the commercial finite
element code Abaqus [11] through the user material sub-routine
facility called UMAT. As such, full advantage of the ca-pabilities
of Abaqus can be taken in terms of searching algorithm for locating
limit or bifurca-tion points. However, for ensuring robustness in
the plasticity calculations, an implicit stress return and
consistent tangent operator algorithm together with a spectral
decomposition of the stress tensor were developed and implemented
in UMAT.
4.1 Plane Strain Conditions The initial boundary value problem
(BVP) illus-trated in Figure 2 is examined. It represents a
dis-placement controlled biaxial test on a hypothetical 10 x 22 x 1
cm3 sand specimen isotropically com-pressed to 100 kPa and
submitted to a 3 cm verti-cal displacement on its top surface.
Deformations are only allowed to occur along components y and z.
The strain component in the direction x is con-strained to zero so
as to reproduce plane strain conditions. Material parameters and
initial condi-tions are identical for all elements; therefore the
specimen is perfectly homogeneous. A total of 220 three-dimensional
C3D8 elements with eight nodes and full integration were used in
these simulations.
x
z
y
100 kPa
100 kPa
Constraints:
3D element8 nodes8 Gauss points
3 cm
10 cm
22 c
m
xu 0 (front & back)&
zu 0 (bottom)&yu 0 (along x-axis)&
†
Position varies†
Element: C3D8
Figure 2. Geometry, mesh, boundary conditions and element type
used in the finite element simulations.
In view of evaluating the effects of boundary
conditions on the loss of homogeneity in deforma-tions under the
presence of a perfectly homogene-
ous field during loading history, the following sub-tle
displacement constraints that block the rigid-body lateral motion
were examined as shown in Figure 3. Among all configurations, the
case (g) with both top and bottom middle nodes restrained laterally
serves to further illustrate the effect of small perturbation on
the final deformation pat-tern. As a matter of comparison, the case
(h) in Figure 3 illustrates the fundamental solution which refers
to the homogeneous deformation mode. This mode was obtained by
rerunning the whole bound-ary value problem as a single element
problem so that shear band formation would be overcome.
y
z
x
p.
0.100.160.210.270.320.380.43
(a) (b) (c) (g)
(d) (e) (f) (h)
Figure 3. Plastic deviatoric strain: (a)-(g) Deformed meshes
showing different localized deformation re-sponses for the distinct
initial boundary conditions. (h) Theoretical homogeneous response
with no shear band.
Figure 3 also shows the numerical results in
terms of deformed configuration and values of plastic deviatoric
strains for a homogeneous and initially dense specimen (initial
void ratio, 0.60). Although homogeneous conditions are imposed,
localized deformation in the form of an inclined shear band appears
in all cases (a)-(g). From a theoretical viewpoint, the above
numerical results clearly confirm that, under the presence of
mate-rial instability, there is a loss of uniqueness in the
solution of the underlying field equations govern-ing the
structural response of the biaxial specimen. This loss of
uniqueness allows a shear band to emerge with its position being
dictated by the boundary constraint imposed.
Figure 4a shows the effective stress path fol-lowed by elements
on the top boundary for all cases (a)-(h). Given the various
positions and con-figurations of the shear band obtained in the
finite element computations, the compelling consequence on the load
versus displacement curves is that
-
4 various post peak branches are obtained with slightly
different bifurcation points, as shown in Figure 4b. Note that in
the pre-peak regime, specimens respond identically for all cases
despite of small subtleties in the boundary constraint. It is worth
noting that the above numerical results seem to point to the fact
that small imperfections give rise to distinct post-peak response
as soon as localization is implicated. The practical signifi-cance
of this is that no two samples tested in the laboratory can result
into the same response curve as a consequence of initial
imperfections and boundary effects. The open question here is
whether there exists a unique solution for the post peak behaviour
irrespective of the nature of the initial imperfections. In other
numerical experi-ments under plane-strain conditions (not presented
here) we show, though, that material perturbation in which void
ratio is assumed to scatter around a specific mean value in space
is a more natural method whereby the bifurcated solution is more
objectively found irrespective of the imperfections in boundary
conditions.
0
300
600
900
0 200 400 600
p (kPa)
q (k
Pa)
Top-LTop-CTop-RBottom-LBottom-CBottom-R
Bottom-Top-CNo Shear Band
(a)
(b)
0
300
600
900
0 1 2 3vertical displ. (cm)
q (k
Pa)
650
750
850
2.14 2.3 2.46
c.s.l.
Figure 4. Average response for elements located on the top
boundary.
4.2 Axisymmetric Conditions In this section we essentially
examine the same prototype problem as in sub-section 4.1, except
that the geometry and loading conditions are now axisymmetric just
like in a so-called triaxial test in soil mechanics. The specimen’s
size is the same as previously, i.e. 22 cm tall and 10 cm of
diameter.
The initial imposed constraints are: uz = 0 on all nodes of the
bottom surface and ux = uy = 0 on one node of the bottom surface
located at the axis of geometrical symmetry of the problem. A total
of 550 three-dimensional C3D8 elements with eight nodes and full
integration were used.
In order to erase the influence of initial imper-fections as
explored in the previous sub-section, material perturbation is
introduced through a small fluctuation in material properties in
the form of a random distribution of initial void ratio throughout
the specimen. To specify a certain void ratio variation throughout
the specimens, both a uniform and a Gaussian (normal) stochastic
void ratio distribution are used as illustrated in Figure 5. In the
uniform distribution case, the void ratio varies between 0.55 and
0.65, corresponding to a dense sand. By contrast, in the Gaussian
distribu-tion, the void ratio is made to spread about a mean value
of 0.60 with a standard deviation of 0.025. In either case, the
distribution inside the specimen is random and non-spatially
correlated.
0
void ratio
200
400
600
800Fr
eque
ncy
0.50
0.52
0.54
0.56
0.58
0.60
0.62
0.64
0.66
0.68
0.70
UniformGaussian
Figure 5. Clustered frequency plots of void ratio for Uniform
and Gaussian distributions for dense sand.
Figure 6 shows the deformed shapes of the cy-
lindrical sand specimens after applying 3 cm of vertical
displacement for three cases: (a) initial homogeneous condition (e0
= 0.60), (b) initial uni-form and (c) initial Gaussian void ratio
distribu-tions. As expected, the unperturbed case (a) gives the
fundamental homogeneous response. However, the other two cases give
a diffuse deformation mode. Note that depending on the distribution
type there is no clear sight of strain localization. Either bulging
at the top or at the bottom parts of the specimen is obtained.
At last, Figure 7 depicts the evolution of the second-order work
throughout the middle section of the sample for the case of
Gaussian distribution. It is noticeable how the zone of negative
second-order work coincides with the bulged region of the specimen
where diffuse failure manifests more in-tensely after pseudo-time t
= 0.866. At the plastic limit condition, for any strain
localization that would occur, the negative second-order work
crite-
-
5 rion is violated. This brings about the fact that the
localization condition is contained within the bifurcation domain
derived from the second-order work criterion.
e
0.590.610.620.640.660.670.69
x y
z
(a) (b) (c) Figure 6. Undeformed-deformed configurations for:
(a) Homogeneous, (b) Uniform, and (c) Gaussian void ratio
distributions for dense sand
t = 0.226
t = 0.591 t = 0.666 t = 0.866 t = 1.000
t = 0.456 t = 0.507t = 0.351
W > 02 W : 02 Figure 7. Temporal evolution of second-order
work for dense sand specimen with Gaussian void ratio
distribu-tion
5. CONCLUSIONS
The question of failure in its various forms such as diffuse and
localized deformation have been dis-cussed within the context of
bifurcation and loss of uniqueness in the presence of material
instability. These notions are illustrated through finite ele-ment
computations of initial boundary value prob-lems. It is shown that
various post-peak responses can be computed as a function of small
fluctua-tions in the initial boundary conditions. Interest-ingly,
it is possible to capture the bifurcation of a homogeneous
deformation field into a heterogene-ous one with shear localization
without the need to introduce any artificial non-homogeneity in
the
calculations. This is due to enriched features such as strong
dilatancy with pyknotropy, barotropy and anisotropy level
sensitivities of the elastoplas-tic constitutive model that are
sufficient to capture bifurcated failure modes. Under axisymmetric
stress-strain conditions, it was demonstrated that, according to
Hill’s stability criterion, diffuse insta-bility prevails over
shear localization along certain loading directions in drained
testing conditions. Finally, it was shown that a correlation exists
be-tween the spatial distribution of second-order work and the zone
of diffuse deformations in the sample.
REFERENCES
[1.] Rudnicki, J.W. & Rice, J.R. 1975. Conditions for the
localization of deformation in pressure sensitive dilatant
material. J. Mech. Phys. Solids 23: 371-394.
[2.] Vardoulakis, I. 1980. Shear band inclination and shear
modulus of sand in biaxial tests. Intl. J. Num. Anal. Methods in
Geomechanics 4: 103-119.
[3.] Darve, F. 1994. Stability and uniqueness in geoma-terials
constitutive modeling. In Chambon, Des-rues, Vardoulakis (eds),
Localisation and Bifurca-tion Theory for Soils and Rocks: 73-88,
Rotterdam: Balkema.
[4.] Wan, R.G. & Guo, P.J. 2004. Stress dilatancy and fabric
dependencies on sand behavior. J. Eng. Mech. 130(6): 635-645.
[5.] Rowe, P.W. 1962. The stress-dilatancy relation for static
equilibrium of an assembly of particles in contact. Proc. Royal
Soc. London. Series A, Math. Phys. Sciences, 269(1339):
500-527.
[6.] Lyapunov, A.M. 1907. Problème général de la sta-bilité des
mouvements. Annals of the Faculty of Sciences in Toulouse, 9:
203-274.
[7.] Lade, P.V. 2002. Instability, shear banding, and failure in
granular materials. Intl. J. Solids Struc-tures 39: 3337–3357.
[8.] Borja, R.I. 2002. Bifurcation of elastoplastic solids to
shear band mode at finite strain. Comp. Me-thods Appl. Mech. Engrg.
191: 5287-5314.
[9.] Hill, R. 1958. A general theory of uniqueness and stability
in elastic-plastic solids. J. Mech. Phys. Solids, 6: 236-249.
[10.] Nova R. 1994. Controllability of the incremental response
of soil specimens subjected to arbitrary loading programs. J. Mech.
Behavior of Materials, 5(2): 193-201.
[11.] Abaqus, 2006. ABAQUS/CAE: User’s Manual. Version 6.6,
ABAQUS Inc, USA.