HOSTED BY Available online at www.sciencedirect.

com

ScienceDirect
Soils and Foundations xxx (2018) xxx–xxx
www.elsevier.com/locate/sandf

Induced fabric anisotropy of granular materials in biaxial tests

along imposed strain paths
Jingshan Shi, Peijun Guo ⇑
Department of Civil Engineering, McMaster University, Hamilton, Ontario L8S 4L7, Canada

Received 15 February 2017; received in revised form 7 November 2017; accepted 23 November 2017

Abstract

In granular materials, loading on an initially isotopic assembly usually induces particle rearrangement, and this is referred to as the
induced anisotropy of fabric. A series of biaxial tests are conducted along various strain paths using DEM to investigate the evolution of
the induced anisotropy. The evolution of both the overall contact network and the sub-networks (the strong and weak) are examined
separately. Results of DEM simulations indicate that the evolution of the fabric deviator in the overall contact network can be described
as a power function of the stress ratio prior to the peak stress ratio that depends on the imposed dilation rate. A unique fabric-stress
relation is obtained for the strong sub-network, which is independent of the strain path, the initial porosity and the confining pressure.
Moreover, deformation instability is observed only along dilatant strain paths, which can be related to the degradation or even collapse
of a weak sub-network, even though the strong sub-network dominates the strength of the granular assembly.
Ó 2018 Production and hosting by Elsevier B.V. on behalf of The Japanese Geotechnical Society.

Keywords: Granular material; Fabric evolution; Contact network; Imposed strain paths

1. Introduction soils, including dilation (i.e. shear-induced volume change),

The mechanical behavior of a granular material can be
inherently anisotropic as a consequence of the original
manner in which the material was formed or deposited.
Such anisotropy is easily altered when the granular mate-
rial is subjected to shear distortion, because of the rear-
rangement of particles via relative particle movements,
including rolling and sliding at particle contacts. The spa-
tial arrangement and inter-connectivity of discrete particles
subjected to stress change is also known as stress-induced
anisotropy (Oda, 1993). Both experimental and theoretical
studies suggest that induced anisotropy and its evolution
contribute to key aspects of the mechanics of granular
Peer review under responsibility of The Japanese Geotechnical Society.
⇑ Corresponding author.
E-mail addresses: shij9@mcmaster.ca (J. Shi), guop@mcmaster.ca
(P. Guo).
failure and instability during deformation (Fu and
Dafalias, 2011; Muhunthan and Collins, 2003; Oda, 1993;
Radjaı̈ et al., 2012; Wan et al., 2007; Wan and Pouragha,
2015; Wan and Guo, 2004). Research incorporating aniso-
tropy into the constitutive model has been conducted by
Guo and Stolle (2005), Lade (2008), Li and Dafalias
(2011) and Gao et al. (2014) among others. Yet, one of
the obstacles of micromechanically formulated constitutive
laws lies in how to mathematically describe the anisotropy
and its evolution with the deformation history. For granu-
lar materials, geometrical anisotropy can be quantified by a
fabric tensor based on various direction-dependent mea-
sures (Kanatani, 1984). The fabric tensor defined by
Satake (1978) in terms of the contact normal orientation
is expressed as /ij ¼ hni nj i, where ni is the i-th component
of contact normal vector and hAi denotes the average of
quantity A over all N c contacts within the granular
assembly.

https://doi.org/10.1016/j.sandf.2018.02.001
0038-0806/Ó 2018 Production and hosting by Elsevier B.V. on behalf of The Japanese Geotechnical Society.

Please cite this article in press as: Shi, J., Guo, P., Induced fabric anisotropy of granular materials in biaxial tests along imposed strain paths, Soils Found.
(2018), https://doi.org/10.1016/j.sandf.2018.02.001
2 J. Shi, P. Guo / Soils and Foundations xxx (2018) xxx–xxx
Various models have been developed to quantify the
fabric evolution or induced anisotropy characterized by
the evolution law of fabric in terms of applied stresses
(Fu and Dafalias, 2015; Mehrabadi and Nemat-Nasser,
1983; Nakai, 1989; Oda, 1993). Results of biaxial tests on
a stack of photo-elastic rods (Satake, 1987) demonstrate
that the ratio of principal stresses (r1 =r2 ) is approximately
proportional to the square root of the corresponding ratio
of the principal values of the fabric tensor (/1 =/2 ) such as
1=2
ðr1 =r2 Þ  ð/1 =/2 Þ . The same conclusion is also obtained
by Maeda et al. (2006). By defining the stress tensor and
fabric tensor as rmn ¼ N c^l0 hf m nn i and F mn ¼ N c^l0 hnm nn i,
Oda et al. (1982) obtain a relation between their principal
components ri and F i (i ¼ 1; 2) in the form of
ri ¼ a0 F i þ b0 F 2i , where a0 ; b0 are material constants with
f and ^l0 being the contact force vector and the average
length of the branch vector, respectively. This relation is
verified using the results from biaxial compression tests
on assembly of photo-elastic rods (Oda et al., 1982). In
DEM simulations by Ng (2001) for 3D compression tests
of ellipsoidal arrays, the strength indicator defined as
b1 ¼ lnð/1 =/3 Þ is found to be directly proportional to the
principal stress ratio ðr1 =r3 Þ. Under triaxial stress condi-
tions, Wan and Guo (2004) assume that the rate of change
of the fabric tensor components is proportional to change
in the deviatoric stress ratio. Therefore, at a critical state,
all components of the fabric tensor eventually achieve con-
stant values. By conducting 3D drained and undrained
tests using DEM simulations, Zhao and Guo (2013) obtain
a unique relation at critical states between the critical fabric
anisotropy (acr c ) and the mean effective stress (p) as
acr
c ¼ a C  k a lnðp=p a Þ, where aC is the fabric anisotropy
when p ¼ pa and ka depends on the loading paths.
When deformation and shear resistance are concerned,
not all contacts play the same role. According to Radjaı̈
et al. (1996), the overall contact network can be separated
into a sub-network of ‘‘strong contacts” with normal forces
larger than the average normal force hf n i, and a sub-
network of ‘‘weak contacts” with normal forces smaller
than hf n i. The strong sub-network is the ‘‘loading-
bearing” network in which the contacts are non-sliding,
whereas the weak network is the dissipative network in
which dissipation takes place due to sliding at contacts
inside this network (Radjaı̈ et al., 1999). They also observe
that the strong sub-network carries the whole deviatoric
load while the weak network contributes to the hydrostatic
pressure only. Furthermore, based on the results from
biaxial DEM simulations on an assembly of oval particu-
lates, Antony et al. (2004) propose a relation,
1=2
ðq=pÞ  ð1=2Þð/s22 =/s11 Þ , between deviatoric stress ratio
(q=p) and the fabric ratio (/s22 =/s11 ) for the strong sub-
network. Kuhn et al. (2015) note that the strong-contact
measures of fabric more closely follow an increase in the
stress ratio than the measure that includes all contacts. In
addition, the load-bearing strong sub-network carries a
direct geometrical anisotropy induced by shear, but it gives
Please cite this article in press as: Shi, J., Guo, P., Induced fabric anisotropy of granular materials in biaxial tests along imposed strain paths, Soils Found.
(2018), https://doi.org/10.1016/j.sandf.2018.02.001
rise via buckling to an indirect anisotropy inside the dissi-
pative weak network with a preferred direction orthogonal
to the major principal direction of the stress tensor. Intu-
itively, one expects that the strong and weak sub-
networks play different roles in the deformation and failure
process of granular materials.
Tests along proportional strain paths have been consid-
ered more suitable for the investigation of deformation
instability of granular materials since the difference
between the imposed rate of volume change and the inher-
ent potential of dilation determines whether or not the
material succumbs to unstable deformation (Guo and Su,
2007). In addition, during these tests, instability occurs
strictly inside the conventional plastic limit condition and
corresponds to the vanishing of the second-order work
(Prunier et al., 2009). These tests can be found, for exam-
ple, in Darve and Laouafa (2000), Guo and Su (2007),
Nicot et al. (2015, 2013), Prunier et al. (2009) and Wan
et al. (2005). By conducting biaxial tests along proportional
strain paths on assemblies of photo-elastic particles, Wan
et al. (2005) observe that the lateral confinement (the weak
force columns) indicates the strength of central core (strong
force columns) which can locally collapse, and hence leads
to failure at the macroscopic level. They also note that
while the strong force columns in the central core of spec-
imen are critical for the material stability, the evolution of
these strong and weak force columns is hard to trace in lab-
oratory tests. One expects that some in-depth observation
towards the relation between deformation instability and
fabric evolution may be obtained by conducting DEM sim-
ulations along proportional strain paths.
The objective of this paper is to examine the evolution of
fabric and its correlation with the deformation instability
of granular materials sheared along proportional strain
paths. The evolution of fabric for different contact net-
works (i.e., the overall contact network, the strong and
weak sub-networks) is traced in biaxial tests along various
strain paths using DEM. The results reveal that the devia-
tor of the fabric tensor based on the strong sub-network
shows a unique relation with the deviatoric stress ratio,
particularly for specimens along proportional strain paths
corresponding to forced dilation. The maximum value of
the deviator of fabric tensor varies with the applied strain
ratio, or the maximum dilatancy rate in stress-controlled
tests. It is confirmed that the strong sub-network plays
the primary role in shear resistance, while the weak sub-
network and its evolution are significantly affected by the
imposed dilatancy rate. The examination of the second-
order work shows that the instability of deformation is a
consequence of the internal structure collapse of the weak
sub-network, which is associated with a decrease in the fab-
ric anisotropy of the strong sub-network.
It should be noted that no attempt has been made to
reproduce any laboratory test results in this paper. It is
the microscopic mechanism and physical aspects of the
fundamental constitutive features of granular material
observed in the laboratory experimental tests that are of
 J. Shi, P. Guo / Soils and Foundations xxx (2018) xxx–xxx
special interest in this research. Biaxial DEM simulations
are carried out in order to capture these microstructural
aspects and can be extended to 3D simulations for a better
understanding of the catastrophic failures that occur in real
cases.
2. Quantification of microstructure
The measures for fabric anisotropy developed to date
can be classified into the following four categories: the pre-
ferred orientations of the particle body, the particle sur-
face, the contact normal and the void space (Kuhn et al.,
2015). Among these measures, the most apparent and effi-
cient measure associated with the mechanical response of
granular materials is based on the orientation of the con-
tact normal. The fabric tensor quantifying the contact nor-
mal orientation of the overall contact network is defined as
(Satake, 1978)
1 X Nc
/ij ¼ nk nk
N c k¼1 i j
where nki is the i-th component of the unit vector represent-
ing the contact normal at the k-th contact.
Following Radjaı̈ et al. (1996), we evaluate geometrical
anisotropy of a granular assembly by determining the
fabric tensor for the network of all contacts and for sub-
networks that carry normal forces lower (weak sub-
network) or higher (strong sub-network) than the average.
By following the definition of the fabric tensor in Eq. (1),
the fabric tensors for the strong and weak sub-networks
are expressed as
1 X 1 X
Ns
c Nwc
/sij ¼ s nki nkj ; /wij ¼ w nk nk
N c k¼1 N c k¼1 i j
where the superscripts s and w stand for quantities associ-
ated with the strong and weak contact networks, respec-
tively. N sc and N wc are the numbers of contacts in the
strong and weak sub-networks, respectively. The total
number of contacts Nc in the granular assembly is
N c ¼ N sc þ N wc . By introducing the relative proportion of
the strong contacts in the system, a ¼ N sc =N c , the relation
between the fabric tensor /ij of the overall contact network
and those of the strong and weak sub-networks can be
expressed as /ij ¼ ð1  aÞ/wij þ a/sij . For biaxial cases, the
degree of fabric anisotropy for different contact networks
within a granular assembly can be characterized by the cor-
responding fabric deviators, q/ ¼ /1  /2 , qs/ ¼ /s1  /s2
and qw/ ¼ /w1  /w2 , respectively. It should be noted that
/ii ¼ 1 holds true in all contact networks according to
Eqs. (1) and (2).
The stress tensor in a granular medium is a volumetric
averaging of dyadic product of contact force and branch
vector on all contacts within the representative element vol-
ume (REV) (see, e.g., Christoffersen et al., 1981; Cundall
and Strack, 1979; Rothenburg and Selvadurai, 1981):
Please cite this article in press as: Shi, J., Guo, P., Induced fabric anisotropy of granular materials in biaxial tests along imposed strain paths, Soils Found.
(2018), https://doi.org/10.1016/j.sandf.2018.02.001
3
Nc c
r¼ hf  lc i ð3Þ
V
where V is the volume of the assembly. The contact force f c
includes a normal component and a tangential component
as f c ¼ f c;n nc þ f c;t tc . Here, nc is the outward unit normal
vector of the contact c aligned with the contact force com-
ponent of f c;n , tc is the unit tangent vector aligned with the
tangential component of contact force f c;t . The branch vec-
tor is expressed as lc ¼ lc mc in which lc and mc are the
length and direction of the branch vector, respectively.
In the following sections, the convention in soil mechan-
ics is employed to describe the stress and deformation
states. Under biaxial conditions, the following quantities
with i, j = 1, 2 are used in later sections: s ¼ p ¼ rii =2,
t ¼ ðsij sij =2Þ1=2 ¼ ðr1  r2 Þ=2, ev ¼ eii , c ¼ ð2eij eij Þ1=2 ,
sin /m ¼ t=s, sin wm ¼ _epv =_cp . In these expressions, s and
sij are the spherical and deviatoric parts of the Cauchy
stress tensor, and ev and eij are the volumetric strain and
the deviatoric strain tensor. e_ pv and c_ p are the plastic volu-
ð1Þ
metric and shear strain increments, respectively. The angles
um and wm are referred to as the mobilized friction angle
and the mobilized dilation angle, respectively.
3. DEM simulation of biaxial tests along proportional strain
paths
According to Wan et al. (2007) and Guo and Su (2007),
the strain softening and material instability for granular
materials are dominated by the rate of forced volume
change and the intrinsic potential of dilation. From a kine-
matic point of view, the collapse of a specimen can be illus-
trated as a larger imposed dilatancy rate than the intrinsic
ð2Þ dilatancy rate of the granular assembly. Compared to tests
along selected stress paths, controlled proportional strain
path tests can provide in-depth observation towards the
mechanism of strain-softening and deformation instability
of a granular material. As such, DEM simulations for biax-
ial tests along proportional strain paths are carried out to
investigate the connection between deformation instability
and fabric in this study.
It should be noted that the selections of the contact stiff-
ness and the inter-particle friction coefficient l have a
direct influence on the results of DEM simulations. In gen-
eral, both the Young’s modulus and the Poisson’s ratio of
the bulk material are intensively affected by the tangential
to normal stiffness ratio v ¼ k s =k n . The micromechanical
analyses based on different hypotheses of homogenization
reveal that the Poisson’s ratio of a 3D granular assembly
composed of equal sized particles can be related to
v ¼ k s =k n via v ¼ ð1  vÞ=ð4 þ vÞ for the kinematic hypoth-
esis (Bathurst and Rothenburg, 1988; Chang et al., 1995)
and v ¼ ð1  vÞ=ð2 þ 3vÞ for the static hypothesis, respec-
tively. The unrealistic result of v ¼ 0 when v ¼ 1 is likely
owing to the assumption of equal-sized particles. Accord-
ing to a series of DEM simulations, Mohamed and
Gutierrez (2010) find that k s =k n ¼ 1:0 gives a Poisson’s
4 J. Shi, P. Guo / Soils and Foundations xxx (2018) xxx–xxx
ratio of 0.25, which is realistic and in good agreement with
experimental results. Consequently, the assumption
k s =k n ¼ 1:0 has been adopted in much of the research
involving DEM simulations (Antony et al., 2004; Nguyen
et al., 2009; Zhu et al., 2016). Recent studies reveal that
the inter-particle friction could influence the critical state
and the stress-dilatancy relationship. According to Huang
et al. (2014), when l < 0.5, the dilatancy tends to increase
with the increase of l. However, the behaviour of granular
material at the critical state seems to be insensitive to fur-
ther increases in l when l  0.5. Moreover, high friction
coefficient could result in negative Poisson’s ratio, espe-
cially when l ! 1 (Rothenburg and Kruyt, 2004).
l ¼ 0:5 has been widely adopted for DEM simulations;
e.g., by Kruyt (2012), Rothenburg and Kruyt (2004) and
Jiang et al. (2006).
In this study, a rectangle packing of polydisperse disks is
considered as the DEM model, which is constrained by four
frictionless rigid sidewalls. The normal contact stiffness at
the walls is chosen to be sufficiently high, and the tangential
contact stiffness at the walls is set to zero. A linear force-
displacement contact law is employed where the contact
behavior is governed by the normal stiffness kn, tangential
stiffness ks and the friction coefficient l. Here we adopt the
assumptions that k s =k n ¼ 1:0 and l ¼ 0:5. The basic charac-
teristics of the initial assembly are given in Table 1.
The initial packing of the assembly for each test was
generated by filling the box with randomly generated par-
ticles and then the specimen was slowly consolidated up
to an initial hydrostatic confining pressure, as shown in
Fig. 1(a). A total of nineteen numerical tests were con-
ducted. They can be divided into two categories: tests along
proportional strain paths with a constant strain increment
ratio R ¼ _e11 =_e22 and tests with constant horizontal stress
that was achieved by controlling the movement of the ver-
tical and horizontal walls (as illustrated in Fig. 1b).
In this study, the average macroscopic strain is deter-
mined  according to the displacement gradient as
eij ¼ 12 @ui =@xj þ @uj =@xi . The strain rates in the vertical
(_e11 ) and horizontal (_e22 ) directions are kept constant.
Table 1
DEM parameters and material properties.
Parameter Value
Number of particles 26,000
Particle density 2000 kg/m2
Smallest particle diameter Dmin 1.6 cm
Largest particle diameter Dmax 2.0 cm
Average particle diameter D 1.8 cm
Initial porosity/void ratio 0.14–0.22/0.16–0.28
Inter-particle friction coefficient l 0.5
Particle-wall friction coefficient l 0.0
Height (H), width (W) of sample H/D = 212, D/W = 110
Initial axial stress r11 170–400 kPa
Initial confining stress r22 170–400 kPa
Normal contact stiffness k n k n =D ¼ 2800 MPa
Tangential contact stiffness k s k s =D ¼ 2800 MPa
Please cite this article in press as: Shi, J., Guo, P., Induced fabric anisotropy of granular materials in biaxial tests along imposed strain paths, Soils Found.
(2018), https://doi.org/10.1016/j.sandf.2018.02.001
ε11
W
1-1
2-2
ε 22
H
1-2
2-1
(a) (b)
Fig. 1. (a) Schematics of the specimen, and (b) the associated loading
boundaries in a 2D test.
Referring to Suiker and Chang (2004) and Christoffersen
et al. (1981), the uniform strain field eij in a granular assem-
bly is the best fit for describing the actual contact displace-
ments and can be considered as a plausible assumption.
The macroscopic strain rates in the assembly are related
to the relative velocities of the boundary walls as
vt11 ¼ vt12 ¼ 12 e_ 11  H t and vt21 ¼ vt22 ¼ 12 e_ 22  W t with
H t and W t being the height and width of the specimen at
time t. Similar methods can be found in Thornton (2000)
as well as Suiker and Chang (2004). The vertical strain rate
e_ 11 is always positive (compression), while the lateral strain
rate e_ 22 ¼ _e11 =R can be positive or negative depending on
the select R value. Since R R ð1; 0Þ and the boundary
walls are frictionless, the directions of the major principal
stress and strain are always vertical with e_ 11 ¼ e_ 1 and
r11 ¼ r1 . Generally, the major P and minor principal
P strains
can be calculated as e1 ¼ e_ 1 Dt and e2 ¼ e_ 2 Dt respec-
tively. Here Dt is the length of time within a step.
In all DEM simulations, the velocities of the boundary
walls are slow enough (e.g., e_ 1 = 105/s) to maintain a
quasi-static condition throughout the entire loading his-
tory. We introduce the parameter # ¼ e_ v =_c, which is the
ratio of volumetric strain rate (_ev ¼ e_ 1 þ e_ 2 ) to the shear
strain rate (_c ¼ e_ 1  e_ 2 ) and can be related to the strain
ratio R as # ¼ ðR  1Þ=ðR þ 1Þ. Hence, all proportional
strain paths can be categorized into forced dilation with
# < 0 (or 0 < R < 1) and forced contraction # > 0 (or
jRj > 1), respectively. Hereby, a value of # ¼ 0 (or
R ¼ 1) is associated with the isochoric deformation condi-
tion, while the hydrostatic compression and K0-
compression conditions are obtained when # ! 1
(R ¼ 1) and # ¼ 1 (R ! 1). It should be noted that
for initially isotropic specimens, the strain paths with
1 < R < 0 are the same as those with R < 1 that
belongs to a subgroup of tests with jRj > 1.
For the sake of convenience, each individual test along
imposed strain path is described according to the following
conventions: (a) the first letter V stands for vertical com-
pression; (b) the second letter, C or E, refers to the horizon-
tal compression or horizontal extension; and (c) the third
letter, C, I or D, stands for volumetric contraction, iso-
choric deformation or volumetric dilation, respectively.
 J. Shi, P. Guo / Soils and Foundations xxx (2018) xxx–xxx 5

Table 2
Details about tests.
Test type Test label Initial stresses Macroscopic boundary conditions
r2 ¼ constant VP200 r1 ¼ r2 ¼ 200 kPa e_ 1 > 0; r2 ¼ 200 kPa
VP300 r1 ¼ r2 ¼ 300 kPa e_ 1 > 0; r2 ¼ 300 kPa
VP400 r1 ¼ r2 ¼ 400 kPa e_ 1 > 0; r2 ¼ 400 kPa
Controlled strain paths VED r1 ¼ r2 ¼ 170 kPa e_ 1 > 0; e_ 2 < 0; j_e1 j < j_e2 j
VEI e_ 1 > 0; e_ 2 < 0; j_e1 j ¼ j_e2 j
VEC e_ 1 > 0; e_ 2 < 0; j_e1 j > j_e2 j
VCC e_ 1 > 0; e_ 2 > 0; j_e1 j > j_e2 j
Note: The sign convention for strain and strain increment is positive for compression.

response changes from stable (hardening) in VEC and

VCC tests corresponding to forced contraction (# > 0) to
300 unstable (strain softening) in VED tests with # < 0. More
PLS
CSL specifically, for the high dilative strain path (i.e.,
VEC
R ¼ 0:43), shearing induces a continuous decrease in the
mean effective stress. The deviator stress first increases
200 VEI and then decreases until ‘‘flow” failure occurs, as if the
specimen were a saturated loose sand subjected to
t, kPa

VCC undrained shearing. For the moderate dilative strain path

100 VED
0 strain hardening. Nevertheless, the effective stress path
0 200 400 600 800
s, kPa
Fig. 2. Stress path responses for tests along various strain paths.
For example, VEI represents a test in which the specimen
has vertical compression, horizontal extension with the vol-
ume being a constant (R ¼ 1:0). For tests with constant
confining pressure, the second letter P stands for the con-
stant confining pressure. VP200 is used to represent a test
in which the specimen undergoes vertical compression with
r2 ¼ 200 kPa. Table 2 shows the test matrix in this study.
4. Effective stress paths and stress-strain responses
Fig. 2 presents the stress paths in the s-t space along a
variety of imposed strain paths ranging from extreme dila-
tion (# ¼ 0:40; R ¼ 0:43) to extreme contraction (# ¼ 21;
R ¼ 1:1). The initial consolidation pressure was 170 kPa.
The critical state line (CSL) and the failure envelope (PSL)
corresponding to the peak state on the stress-strain curves
in drained compression tests under a constant confining
pressure are also presented in Fig. 2. The peak and critical

state friction angles are determined as up ¼ 23:6 and

ucv ¼ 16 , respectively. The stress-strain responses in differ-
ent tests are presented in Fig. 3.
One immediately observes that the material response is
not bounded by the stress paths in the constant lateral con-
fining pressure (akin to a drained test) and the isochoric
(akin to an undrained test of a saturated specimen) condi-
tions. As the imposed strain ratio changes, the material
with R ¼ 0:76, the mean effective stress first decreases while
the deviator stress increases at the beginning of shearing.
Then the effective stress path changes direction with both
s and t increasing with axial strain, which reflects typical
1000 snaps back shortly after the maximum friction angle at
(t/s)max is mobilized. The effective stress paths along low
dilative strain paths with R ¼ 0:86–2:2 have the same fea-
ture as a saturated dense sand subjected to undrained
shear: the mobilized friction angle (or the t/s ratio) gradu-
ally increases to its peak value and then tends to approach
the critical state, while the mean effective and the deviator
stresses both increase monotonically. For VED tests (# < 0
or 0 < R < 1), the effective stress paths at large shear
strains are bounded by the peak state line (PSL) and the
critical state line (CSL). These observations are consistent
with those of Guo and Su (2007), Wan et al. (2007) and
Wan and Guo (2004) based on laboratory test results and
constitutive simulations. It should be noted that the ulti-
mate case with # ! 1 is the isotropic compression. The
stress-strain responses along different dilative strain paths
are presented in Fig. 3(b).
The deformation states corresponding to the two types
of contractant strain paths, the VEC and VCC tests, are
illustrated in Fig. 4. On the s-t plane, the strain paths in
the VEC and VCC tests are separated by the stress path
of K0-compression at # ¼ 1 and R ! 1, as shown in
Fig. 2. The stress-strain responses, presented in Fig. 3(c)
and (d) for VEC and VCC tests, are very distinctive
depending on the deformation. For VEC tests with # > 0
and R > 1, a peak value of t=s is obtained, with the ulti-
mate deformation state being different from the critical
state. As the value of R increases, the lateral strain tends
to decrease and gradually approaches the (biaxial) K0-
condition with e2 ¼ 0. For strain paths for VCC with
# > 0 and R < 1, the t/s ratio increases monotonically

Please cite this article in press as: Shi, J., Guo, P., Induced fabric anisotropy of granular materials in biaxial tests along imposed strain paths, Soils Found.
(2018), https://doi.org/10.1016/j.sandf.2018.02.001
6 J. Shi, P. Guo / Soils and Foundations xxx (2018) xxx–xxx

0.4 0.5

0.4
0.3
0.3
σ2=400kPa
0.2 σ2=300kPa

t/s
0.2 ℜ=54
t/s
σ2=200kPa
ℜ=10.8
0.1 0.1
e0=0.16 ℜ=1.0
e0=0.183
0 0
0 0.05 0.1 0.15 0.2 0.25 0 0.03 0.06 0.09 0.12 0.15
γ γ
(a) (b)
0.5
0.5
0.4 0.4
0.3 0.3
t/s

t/s
0.2 ℜ=−5.4 0.2 ℜ=1.0
ℜ=−2.2 ℜ=0.76
0.1 ℜ=−1.1 0.1 ℜ=0.43
ℜ=1.0 e0=0.183
e0=0.183
0 0
0 0.05 0.1 0.15 0 0.03 0.06 0.09 0.12 0.15
γ γ
(c) (d)
Fig. 3. Stress-strain responses in (a) compression tests with r2 = constant; (b) VED tests with 0 < R 6 1; (c) VEC tests with R P 1 and (d) VCC tests
with R < 1.

in different contact networks when shearing takes place

along various strain paths.

5.1. Internal structure evolution of overall contact network

5.1.1. Fabric evolution in VED and VEI tests

(a)
Fig. 4. Two types of forced contractant tests: (a) the VEC test, and (b) the
VCC test.
with shear strain, gradually approaching a constant that
depends on the strain ratio R and is smaller than sin ucv .
These observations are generally consistent with the find-
ings of Wan et al. (2007).
5. Fabric evolution
As the mechanical behavior of granular materials is a
macroscopic summation of all particle level activities, it is
affected by the evolution of the spatial arrangement of par-
ticles and the distribution of inter-particle contact orienta-
tions. This section discusses the evolution of fabric tensors
The degree of fabric anisotropy of the overall contact
network is characterized by q/ ¼ /1  /2 , in which /1
and /2 are the principal components of the fabric tensor
in the overall contact network. The directions of /1 and
(b) /2 are coaxial with those of r1 and r2 , respectively. As
shown in Fig. 5(a), the value of q/ increases with the shear
strain initially until it approaches the peak value. There-
after, q/ deceases as shearing continues and approaches a
constant at large strain. This trend is similar to the evolu-
tion of the deviatoric stress. However, both the peak value
and the ultimate value of q/ at large shear strains depend
on the imposed strain ratio R. In particular, the peak
and ultimate values of q/ increase as R decreases, implying
that the contact orientations show most intensive direc-
tional preference to resist deformation in the VED tests
(0 < R < 1 or # < 0), in which the specimen collapses
much easier than in other tests. The evolution of q/ with
the stress ratio (t/s) is plotted in Fig. 5(b), in which the
deformation starts from an initially isotropic stress state.
Prior to the peak stress ratio, q/ increases with the increase
of (t/s) and the overall contact network resists deformation
by rearranging contact orientation with more contacts in

Please cite this article in press as: Shi, J., Guo, P., Induced fabric anisotropy of granular materials in biaxial tests along imposed strain paths, Soils Found.
(2018), https://doi.org/10.1016/j.sandf.2018.02.001
 J. Shi, P. Guo / Soils and Foundations xxx (2018) xxx–xxx 7

0.15 0.15
ℜ=1.0
ℜ=0.92
0.1 0.1 ℜ=0.86
ℜ=0.76
ϕ1 – ϕ2

ϕ1 – ϕ 2
ℜ=0.65
0.05 ℜ=1.0 ℜ=0.92 0.05 ℜ=0.43
ℜ=0.86 ℜ=0.76 e0=0.183
e0=0.18
0 0
0 0.05 0.1 0.15 0 0.1 0.2 0.3 0.4 0.5
γ t/s
(a) (b)
Fig. 5. Evolution of fabric anisotropy in the overall contact network (a) with shear strain, and (b) with stress ratio in the VED and VEI tests.

0.15 0.15
VED ℜ=10.8
0<ℜ<1 ℜ=2.2
VEI ℜ=1.0
0.1 ℜ=1 0.1 ℜ=0.86
ϕ1– ϕ2

ϕ 1 – ϕ2
ℜ=0.76
VEC ℜ=0.65
0.05 ℜ>1 0.05 ℜ=0.43

VCC
ℜ<-1 e0=0.183
0 0
0 0.2 0.4 0.6 0 0.1 0.2 0.3 0.4 0.5
t/s t/s
(a) (b)
Fig. 6. (a) Evolution of the fabric deviator for the overall contact network with t/s; and (b) the best fit of pre-peak data in different tests.

the vertical direction. At the peak stress state, the overall
contact network reaches its maximum bearing capacity
and buckles. After the peak, a reduction of q/ is observed
as t/s decreases and gradually approaches the critical state.
Before the peak stress ratio, the relation between q/ and t/s
is non-linear, which is consistent with the conclusion of
Mehrabadi and Nemat-Nasser (1983). Nevertheless, after
the peak stress ratio, the variation of q/ with t/s ratio can
be described as a linear expression approximately, as
shown in Fig. 5.
5.1.2. Fabric evolution in VEC tests
As shown in Fig. 6(a), significant fabric anisotropy
develops in VED tests since a rapid lateral extension causes
a quick reduction of lateral constraint, which in turn
induces more contacts to align in the vertical direction to
resist deformation. Thereafter, failure of the overall con-
tact network happens after the fabric anisotropy
approaches its maximum. The lower fabric anisotropy in
the VEC tests than that in the VED and VEI tests is due
to the smaller extension rate (_e2 ¼ _e1 =R; R > 1) in the
lateral direction than the vertical compression rate e_ 1 . With
the increase of R, the horizontal extension rate j_e2 j becomes
smaller, which tends to produce relatively stable force
chains in the direction of r2. In other words, with an
increase of R, the force chains in the lateral direction pro-
vide stronger support to the strong force chains in the ver-
tical direction. When the value of R is large enough, the
weak force chains in the lateral direction will always be
stable during deformation and provide sufficient support
to the strong force chains without collapsing. Correspond-
ingly, the stress paths could not approach the peak state
line, for example, in the test with strain path of R ! þ1
in Fig. 2.
5.1.3. Fabric evolution in VCC tests
In the VCC tests, both e_ 1 and e_ 2 are positive (compres-
sion) with e_ 1 > e_ 2 > 0. When compared with other tests,
the enhanced lateral constraint in the VCC tests helps to
build more contacts and increase the contact forces in the
lateral direction. As a result, the directional variation of
the contact distribution decreases, which results in a
reduced magnitude of fabric anisotropy in the overall con-
tact network; as can be observed in Fig. 6(a).
5.1.4. Effect of stress on fabric evolution in overall contact
network
Fig. 6(b) presents the evolution of fabric deviator
q/ ¼ /1  /2 with respect to the stress ratio t/s in the
VED, VEI test and some VEC tests. A regression analysis
is performed to obtain a quantitative description for the
fabric evolution. Prior to the peak stress state, the variation

Please cite this article in press as: Shi, J., Guo, P., Induced fabric anisotropy of granular materials in biaxial tests along imposed strain paths, Soils Found.
(2018), https://doi.org/10.1016/j.sandf.2018.02.001
8 J. Shi, P. Guo / Soils and Foundations xxx (2018) xxx–xxx

of q/ with t/s and the imposed strain ratio can be described
by
n 5.2.1. Fabric evolution in VED and VEI tests
/1  /2 ¼ Aðt=sÞ
where A and n are functions of the imposed strain ratio R
or #. As shown in Fig. 7 (a), A and n can be reasonably
expressed as A ¼ 1:5968  0:9446# and
n ¼ 0:5049  0:7595#, respectively. Eq. (4) plays the role
of the fabric evolution law that is important for a stress-
strain model based on continuum mechanics.
In the post-peak deformation stage, the value of q/
reduces with the decrease of t/s following a linear relation
q/ ¼ qmax
/ þ jp ½t=s  ðt=sÞmax with qmax
/ ¼ Aðt=sÞnmax ; as
shown in Fig. 5(b). jp can be expressed as a linear function
of the imposed strain ratio jp ¼ 0:5054# þ 0:2377; see
Fig. 7(b). The reason that Eq. (4) does not hold true for
the post-peak deformation stage may be attributed to the
different patterns of fabric evolution in the strong and weak
sub-networks as well as the potential deformation instabil-
ity. Owing to the rigid boundary conditions adopted in the
DEM simulations, the post-peak deformation is still
homogeneous without noticeable localized or diffused
deformation patterns, and the deformation instability man-
ifests itself as the collapse or buckling of force chain
segments.
5.2. Evolution of strong sub-network fabric
ð4Þ
Fig. 8 presents the evolution of fabric deviator
qs/ ¼ /s1  /s2 of the strong sub-network under different
conditions. Fig. 8(a) shows the variation of qs/ with the
t/s ratio in the VED and VEI tests with R 2 ½0:43; 1:0 .
One remarkable finding is that a unique relation between
qs/ and t/s can be defined prior to the peak deviator stress
ratio along all strain paths, which is different from the fab-
ric evolution of the overall contact network presented in
Fig. 5(b). The maximum induced anisotropy qs/;peak appears
simultaneously with (t/s)max. The post-peak data are
slightly scattered; however, they localize in a narrow band
that almost overlaps the pre-peak data. Since /s1 þ /s2 ¼ 1,
the following relation can be used to describe the variation
of qs/ with the t/s ratio:
/s1  /s2 t
s ¼ j ð5Þ
/1 þ /2
s
s
where j is constant. The best estimate of j ¼ 1 can be
applied to different strain paths, as shown in Fig. 8(a).
Eq. (5) is verified by tests along imposed strain path
R ¼ 0:7 for different initial void ratios in Fig. 8(b) and

0.3 3
A A = -0.7595ϑ + 0.5049
n
0.2 2
n
κp

0.1 n = -0.9446ϑ + 1.5968

A,

1
κ p = 0.5054ϑ+ 0.2377

0 0
-0.3 -0.2 -0.1 0 0.1 -0.6 -0.3 0 0.3 0.6
ϑ = dεv /dγ -ϑ = -dεv /dγ
(a) (b)
Fig. 7. Best fits of (a) the parameters A, n; and (b) the parameter of jp.

0.5 0.4 0.4

ℜ∈[0.43,1.0] e0=0.16, 0.19, 0.22, 0.25, 0.28 σ2=200kPa, 300kPa,400kPa
0.4 0.3
0.3
ϕs1 – ϕs2

ϕ s1 – ϕ s 2

0.3
ϕ s1 – ϕ s2

0.2 0.2
0.2
0.1 0.1
0.1 e0=0.16
ℜ= 0.7
0 0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4
t/s t/s t/s
(a) (b) (c)

Fig. 8. Evolution of the fabric deviator in the strong sub-network with stress ratio t/s (a) in the VED and VEI tests (# 6 0), (b) at different initial
porosities, and (c) under different confining pressures.

Please cite this article in press as: Shi, J., Guo, P., Induced fabric anisotropy of granular materials in biaxial tests along imposed strain paths, Soils Found.
(2018), https://doi.org/10.1016/j.sandf.2018.02.001
 J. Shi, P. Guo / Soils and Foundations xxx (2018) xxx–xxx 9

by conventional biaxial compression tests with constant r2
(VP200, VP300 and VP400) on specimens with an initial
void ratio of 0.16 in Fig. 8(c). In both cases, Eq. (5) can
be used to describe the relation between t=s and
ðt=sÞ relation. Moreover, the occur-
ð/s1  /s2 Þ, regardless of the initial porosity of the specimens
and the variation of the dilatancy rate dev =dc in the con-
ventional biaxial compression tests. With j ¼ 1 holding
true for all cases, Eq. (5) can be alternatively expressed as
/s1 : /s2 ¼ r1 : r2 .
Eq. (5) can also be used to describe the variation of the
strong sub-network anisotropy in the post-peak deforma-
tion stage, as shown in Fig. 8a for tests along controlled
strain paths and Fig. 8b for tests with constant confining
stresses. The variation of qs/ with the shear strain depends
on the rate of volume change, while a critical state for
t=s relations for
qs/;cr is identified at large shear strain level in Fig. 8. The
decrease of qs/ implies a degradation, even collapse of
strong sub-network, which results in a decrease of shear
resistance and potential deformation instability. The corre-
lation between the evolution of strong sub-network aniso-
tropy and deformation instability will be discussed in later
sections.
5.2.2. Fabric evolution in VEC tests
In VEC tests, shearing takes place along contractant
strain paths, the qs/
ðt=sÞ curves are all in a narrow
band around the line of /s1  /s2 ¼ t=s; as shown in
Fig. 9(a). Similar to the development of anisotropy in
VED tests, a peak and a residual anisotropy are
observed along strain paths with relatively small R val-
ues, for example R ¼ 2:2. Eq. (5) can be used to describe
the prior peak qs/
rence of (t/s)max and qs/;max implies that the maximum
resistance to deformation of the strong sub-network is
mobilized due to intensive reduction of the lateral
constraints.
However, when R > 54, the expansion of the specimen
in the lateral direction is relatively small. As a result, the
strong force chain network can resist deformation without
approaching its maximum capacity due to the strong sup-
port of the horizontal weak force chains. Therefore, qs/
increases monotonically with the stress ratio (t/s) following
Eq. (5). Compared with the scattered q/
the overall contact network in Fig. 6, Fig. 9(a) suggests a
strong influence of t/s on the internal structure of strong
sub-network in tests along contractant strain paths with
R > 1. In addition, the ultimate fabric deviator qs/;ult
depends on the value of R in these cases. qs/;ult is higher
than the critical value qs/;cr in the VED and VEI tests.
5.2.3. Fabric evolution in VCC tests
For imposed contractant strain paths with # > 0 and
R < 1 in the VCC tests, qs/ increases monotonically with
t/s and the shear strain; as shown in Figs. 9(b) and (d). The
ultimate fabric deviator qs/;ult increases as the strain paths
moves toward K0-compression at jRj ! 1. As the jRj

0.5
0.5 ℜ=−1.1
e0=0.183
0.4 ℜ=−2.2
0.4
ℜ=−2.7
0.3 0.3 ℜ=−3.2
ϕ s1 – ϕs 2

ϕs1– ϕs2

ℜ=−5.4
0.2 0.2
ℜ∈[2.2,+∞]
0.1 0.1
ϑ∈[0.38,1.0]
e0=0.183
0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
t/s t/s
(a) (b)
0.5 0.5

0.4 0.4
ℜ=−1.1
ℜ=2.2 0.3 ℜ=−2.2
0.3
ϕ s 1– ϕ s 2

ℜ=10.8 ℜ=−2.7
ϕ s 1 – ϕs 2

0.2 ℜ=21.6 0.2 ℜ=−3.2

ℜ=32.4 ℜ=−5.4
0.1 0.1
ℜ=54.0 e0=0.183
e0=0.183 0
0
0 0.02 0.04 0.06 0.08
0 0.05 0.1 0.15
γ γ
(c) (d)
Fig. 9. Evolution of qs/ ¼ /s1  /s2 with t/s in VEC tests (a), in VCC tests (b), with the shear strain in VEC tests (c) and in VCC tests (d).

Please cite this article in press as: Shi, J., Guo, P., Induced fabric anisotropy of granular materials in biaxial tests along imposed strain paths, Soils Found.
(2018), https://doi.org/10.1016/j.sandf.2018.02.001
10 J. Shi, P. Guo / Soils and Foundations xxx (2018) xxx–xxx

ℜ =1.0 K0 -1
0.6 ψmax

-0.5
0.4
ϕs1– ϕs2
0

(%)
VEC VCC upper limit σ2=400kPa
0.2 σ2=400KPa σ2=300kPa
0.5

εv
σ2=300KPa σ2=200kPa
σ2=200KPa
e0=0.16
0 1
-0.5 0 0.5 1 1.5 2 2.5 3 0 0.05 0.1 0.15 0.2 0.25
dεv/dγ γ
(a) (b)

Fig. 10. (a) The upper limits of qs/ ¼ /s1  /s2 ; and (b) Volumetric strain in biaxial compression tests when r2 = constant.

value increases, the volume contraction becomes dominant
and the effect of shearing decreases. As a special case when
R ¼ 1, the specimen is subjected to hydrostatic compres-
sion. One expects smaller qs/;ult values as R increases, which
is confirmed by the results in Figs. 9(b) and (d). With the
decrease in the R value, the lateral constraint becomes
weaker, and hence severe anisotropy develops in the strong
sub-network.
5.2.4. Maximum anisotropy of strong sub-network
Fig. 10 show the variation in the maximum fabric ani-
sotropy qs/;max in the strong sub-network with # ¼ dev =dc
in controlled strain path tests. The data can be divided
into two sections by the K0-compression (R ! 1)
where the value of qs/;max approaches its maximum value.
The right section of the data points corresponds to the
VCC tests (R < 1) in which the specimen has compres-
sive deformation in the directions of both r1 and r2 . The
qs/;max tends to decrease as # ¼ dev =dc increases in these
tests, which is induced by the increased fraction of strong
contacts in the lateral direction due to the over-
constraint imposed by deformation in the direction of
r2 . However, qs/;max increases with the increase of # in
the left section of the data points in VEC tests, in which
the specimen is allowed to expand in the direction of r2 .
In biaxial tests with constant r2 , the induced anisotropy
is close to that along dilative strain paths with
# ¼  sin wmax in which wmax is the maximum dilatancy
angle defined as sin wmax ¼ dev =dc in conventional biax-
ial tests (see Fig. 10b).
It can be shown that the theoretical maximum value of
fabric deviator in the strong sub-network is 0.5. For contin-
uous distribution of contact normal, /sij defined in Eq. (2)
can be alternatively expressed as
Z 2p
/ij ¼
s
Es ðnÞni nj dh ð6Þ
0
with Es ðnÞ being the probability density function defining
the directional distribution of contact normal in the strong
sub-network. The first-order Fourier expansion of Es ðnÞ
provides an adequate measure of the geometrical aniso-
tropy (Radjaı̈ et al., 1998):
1  
Es ðnÞ  1 þ Asc cos 2ðh  hsc Þ ð7Þ
2p
where Asc and hsc define the amplitude of anisotropy and the
major principal direction of the fabric tensor for the strong
sub-network. It can be shown that /s1  /s2 ¼ Asc =2. The
non-negative requirement of Es ðnÞ requires Asc 6 1, which
yields qs/;max 6 0:5 and is consistent with the results pre-
sented in Fig. 10(a).
5.3. Evolution of weak network fabric
The evolution qw/ ¼ /w1  /w2 of the weak sub-network
with the t/s ratio and shear strain along different strain
paths are presented in Fig. 11. Minor anisotropy in the
weak sub-network develops in VED tests (0 < R < 1), with
the major principal direction of /wij perpendicular to the
direction of r1 and hence the direction of /s1 in the strong
sub-network. Following an initial buildup of anisotropy
upon loading, qw/ reaches a maximum value and then
decreases with continuous deformation. The initial buildup
of qw/ along dilatant strain paths can be described as a lin-
ear function of t/s in the form of:
/w1  /w2 t
¼ jw ð8Þ
/w1 þ /w2 s
in which jw is a coefficient that varies with the value of R.
As shown in Fig. 12(a), jw ranges from 0.3 to 0.45 when the
dilation rate increases form # ¼ 0:4 (R ¼ 0:43) to # ¼ 0
(R ¼ 1:0). After the peak anisotropy, the degree of aniso-
tropy diminishes gradually and the weak sub-network
eventually becomes isotropic at large strain in tests of high
dilation rate (e.g., R ¼ 0:43–0:65); as shown in Fig. 12(a).
For the VEC tests (R > 1) and VCC tests (R < 1), the
induced anisotropy in the weak sub-network tends to
increase significantly as the deviatoric stress ratio t/s
increases. With the increase of shear strain, the value of
qw/ approaches a limit depending on the imposed strain
ratio R; see Fig. 12(a).

Please cite this article in press as: Shi, J., Guo, P., Induced fabric anisotropy of granular materials in biaxial tests along imposed strain paths, Soils Found.
(2018), https://doi.org/10.1016/j.sandf.2018.02.001
 J. Shi, P. Guo / Soils and Foundations xxx (2018) xxx–xxx 11

0.4 0.4
ℜ=-2.2 ℜ=-5.4 ℜ=-2.2
ℜ=-5.4
0.3 ℜ=10.8 0.3 ℜ=10.8
ℜ=2.2
ℜ=2.2 ℜ=1.0
qwϕ

qwϕ
0.2 0.2
ℜ=0.76
ℜ=1.0 ℜ=0.43
0.1 ℜ=-1.1 0.1 ℜ=-1.1
ℜ=0.76
ℜ=0.43
0 0
0 0.05 0.1 0.15 0 0.15 0.3 0.45
γ=ε1-ε2 t/s
(a) (b)
Fig. 11. Evolution of qw/ ¼ /w1  /w2 (a) with shear strain, and (b) with stress ratio along various strain paths.

ℜ=0.43 ℜ=1.0 ℜ=10.8 ℜ=−5.4

0.2
ℜ=1.0
ℜ=0.92
0.15 ℜ=0.86
W
ℜ=0.76
0.1
qwϕ

S
0.05

0 O
0 0.1 0.2 0.3 0.4 0.5
t/s
(a) (b)
Fig. 12. (a) Linear relation between qw/ and t/s prior to the qw/;max in VED and VEI tests; and (b) the histogram of the strong sub-network (S), the weak sub-
network (W) and the overall contact network (O) when c = 3%.

Different trends are observed regarding the evolution of
qw/ with the t/s ratio in the VEC and VCC tests. As shown in
Fig. 12(b), for the VEC tests, the degree of fabric aniso-
tropy within the weak sub-network maintains at a steady
level in the post-peak t/s-ratio stage. The level of aniso-
tropy is much lower than that developed in the strong
sub-network. On the other hand, when compression takes
place in both directions with e_ 1 > e_ 2 > 0 and R < 1
(i.e., VCC tests), the induced anisotropy in the weak sub-
network reaches a much higher level as the t/s increases if
the strain paths are close to K 0 -compression. As a special
case, no significant anisotropy is observed in the weak sub-
network when a specimen is subjected to isotropic com-
pression (R ¼ 1). It should be emphasized that the major
principal direction of the weak sub-network is perpendicu-
lar to that of the strong sub-network.
Fig. 13 compares the evolution of fabric in the strong
and weak sub-networks along contractant strain paths in
which # > 0 and R < 1. For R ¼ 1.1 to 5.7, the evo-
lution of qs/ and qw/ are very close, especially at R ¼ 1:1. It
should be noted that the strain path of R ¼ 1 corre-
sponds to isotropic compression. For an initially isotropic
(or weakly anisotropic) contact network, the induced ani-
sotropy of the overall contact network when R ¼ 1:1
and 2.2 is expected to be minimal, as is demonstrated
in Fig. 6. However, due to the non-uniform distribution
of contact forces, a strong and a weak sub-network can still
be identified with a certain level of anisotropy, which is
stronger for the strong sub-network and weaker for the
weak sub-network in the r1 direction. The minor aniso-
tropy of the overall contact network requires that the fabric
anisotropies in the strong and weak networks are both
small, as demonstrated in Fig. 13.
6. Fabric evolution and deformation instability
Strain softening and instability of geomaterials may lead
to catastrophic deformation, including failure. In mechan-
ical terms, stability is obtained if a small stress increment
yields a small strain increment, while instability is defined
as a behaviour in which large plastic strains develop rapidly
due to the inability of material to sustain a given stress or
load. The theoretical basis for the concept of stability is
Drucker’s postulate of non-negative second-order plastic
work, i.e., d2 Wp ¼ drij depij > 0 (Drucker, 1957). Hill
(1958) extended Drucker’s postulate to the second-order
increment of total work. According to Hill (1958), for
any (drij ; deij ) linked by their constitutive relation, a mate-
rial is stable when d2 W ¼ drij deij 6 0. Hence, a positive
second-order work constitutes a sufficient, but not neces-

Please cite this article in press as: Shi, J., Guo, P., Induced fabric anisotropy of granular materials in biaxial tests along imposed strain paths, Soils Found.
(2018), https://doi.org/10.1016/j.sandf.2018.02.001
12 J. Shi, P. Guo / Soils and Foundations xxx (2018) xxx–xxx
0.6
qwK=-
ϕ ℜ=-5.4
1.1
qsK=1.
ϕ
0.4 0s
qwϕ, qsϕ ℜ=-2.2
0.2
ℜ=-1.1
0 0
0 0.1 0.2
t/s
(a)
Fig. 13. (a) Evolution of qs/ and qw/ with t/s; and (b) Relation between qs/ and qw/ in VCC tests.
sary, condition of stability. When d2 W 6 0, the effective
collapse of a homogeneous sample depends on the loading
direction and the controlled loading variables (Nicot et al.,
2013; Prunier et al., 2009). Alternatively, d2 W 6 0 can be
interpreted as a bifurcation criterion. At a bifurcation
point, the response path can be unique or nonunique and
stable or unstable. For example, in the tests along propor-
tional strain paths, the failure corresponding to the vanish-
ing of the second-order work is thought to be non-effective
(Nicot et al., 2013).
Attempts have been made to investigate material insta-
bility in granular assemblies on the microscopic scale in
terms of the microstructure of the overall contact network;
see, e.g., Wan et al. (2007), Nicot and Darve (2006, 2005)
and Nicot et al. (2015). It has been revealed that the origin
of potential instabilities as a macro-scale phenomenon may
be directly related to the constitutive nature of the local
contact model at particle contacts (Nicot and Darve,
2006). In addition, Nicot et al. (2007) observe that changes
in the fabric of the medium, associated with rearrange-
ments caused by sliding and rolling, cause nonequivalence
between the microscopic and macroscopic second-order
works. The following analyses, however, will focus on the
variation of fabric in different contact networks when the
macroscopic second-order work vanishes.
Following the stability analysis for sand tested along
proportional strain paths undertaken by Wan et al.
(2007), Darve and Laouafa (2000) and Wan et al. (2005),
the second-order work under biaxial stress conditions can
be expressed as
 
dr2  r2 monotonically. These observations indicate that the onset
d2 W ¼ dr1  de1 ¼ d r1  de1 ð9Þ
R R
with R ¼ de1 =de2 as defined previously. With the strain
increment de1 being non-negative, the condition d2 W ¼ 0
is satisfied when dðr1  r2 =RÞ ¼ 0, for which
ðr1  r2 =RÞ ¼ max
The above equation implies that instability starts at the
peak point of the curve obtained by plotting ðr1  r2 =RÞ
versus axial strain e1 .
Please cite this article in press as: Shi, J., Guo, P., Induced fabric anisotropy of granular materials in biaxial tests along imposed strain paths, Soils Found.
(2018), https://doi.org/10.1016/j.sandf.2018.02.001
0.5
0.4
0.3
q sϕ
ℜ=1.0 0.2 ℜ=1.0
ℜ=−1.1
0.1 ℜ=−2.2
ℜ=−5.4
0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
qwϕ
(b)
Fig. 14 presents the variation of ðr1  r2 =RÞ and its
relation with the fabric deviator in the strong and weak
sub-networks. In particular, the variation of ðr1  r2 =RÞ
with axial strain e1 along various strain paths is plotted
in Fig. 14(a), where the peak points corresponding to the
onset of deformation instability according to Eq. (10) can
be easily identified in the VED tests (0 < R < 1). For the
VEC (R > 1) and VCC (R < 1) tests, however, the value
of ðr1  r2 =RÞ tends to increase monotonically with e1 ,
indicating that deformation is always stable. Fig. 14(b) pre-
sents the correlation between qs/ ¼ /s1  /s2 and
ðr1  r2 =RÞ in the VED tests with R ¼ 0:43–1. The thicker
dotted segment on each curve represents states of unstable
deformation. An immediate observation is that the strong
sub-network is responsible for deformation instability since
the peak value of qs/ is reached when d2W = 0. More specif-
ically, for tests of high dilation rate with R ¼ 0:43–0:65, qs/
increases initially with ðr1  r2 =RÞ, both reaching their
maximum simultaneously; see Figs. 14(a) and (b). Refer-
ring to Figs. 2 and 3(b), for high dilative strain paths with
R ¼ 0:43–0:65, d2W = 0 occurs at the ultimate deforma-
tion state when t=s ¼ ðt=sÞmax and the shear resistance
approaches zero. For lower rates of dilation with
R ¼ 0:76–0:86, qs/ and ðr1  r2 =RÞ also reach their maxi-
mum at the same time. However, the degradation of the
strong sub-network with the reduced degree of anisotropy
takes place following the onset of deformation instability;
as shown in Fig. 14(b). In the CEI test with R ¼ 1, the peak
value of qs/ develops during deformation that is always
stable according to Fig. 14(b) since ðr1  r2 =RÞ increases
of deformation instability is accompanied by a degradation
or progressive collapse of the strong sub-network. How-
ever, the degradation of the strong sub-network does not
guarantee unstable deformation. In other words, the degra-
dation of strong sub-network is a necessary but not suffi-
ð10Þ cient condition for deformation instability.
Fig. 14(c) displays the evolution of qw/ ¼ /w1  /w2 with
ðr1  r2 =RÞ. For dilation rates in the range of
R ¼ 0:43–0:86, the highest value of qw/ is observed before
 J. Shi, P. Guo / Soils and Foundations xxx (2018) xxx–xxx 13

ℜ=−1.1 ℜ=10.8
ℜ=2.2
1100
ϑ>0

σ1 – σ2 / ℜ, kPa
600 d2W>0
ℜ=1
ℜ=0.76
100
ϑ <0
ℜ=0.43 ℜ=0.65 d2W=0
-400
0 0.02 0.04 0.06 0.08 0.1
Axial strain

(a)
0.5 0.2
ℜ=0.76 ℜ=1
0.4 ℜ=0.65 ℜ=1
0.15
ℜ=0.86
0.3 ℜ=0.43
ℜ=0.76
q sϕ

ℜ=0.86 0.1 ℜ=0.65

qwϕ
0.2
ℜ=0.43
0.1 0.05

0
0
-400 -200 0 200 400 600
-400 -200 0 200 400 600
(σ1-σ2/ℜ), kPa
(σ1-σ2/ℜ), kPa
(b) (c)
Fig. 14. (a) Variation of r1  r2 =R with axial strain, (b) evolution of fabric deviator for the strong sub-network and (c) for the weak sub-network with
r1  r2 =R in VED tests (the thick dotted line represents the d2W = 0).

0.5 0.5

0.4 0.4
ℜ=2.2 0.3 ℜ=−1.1
0.3
ℜ=10.8 ℜ=−2.2
q sϕ

q sϕ

0.2 ℜ=21.6 0.2 ℜ=−2.7

ℜ=32.4 ℜ=−3.2
0.1 ℜ=54 0.1 ℜ=−5.4

0 0
0 1000 2000 3000 4000 0 2000 4000 6000
(σ1-σ2/ℜ), kPa (σ1-σ2/ℜ), kPa
(a) (b)
0.4 0.4

0.3 0.3
ℜ=−1.1
ℜ=2.2 ℜ=−2.2
ℜ=10.8
qwϕ

0.2 0.2 ℜ=−2.7

ℜ=21.6
qwϕ

ℜ=−3.2
ℜ=32.4
0.1 0.1 ℜ=−5.4
ℜ=54.0

0 0
0 1000 2000 3000 4000 0 2000 4000 6000
(σ1-σ2/ℜ), kPa (σ1-σ2/ℜ), kPa
(c) (d)
Fig. 15. Evolution of qs/ ¼ /s1  /s2 with r1  r2 =R in VEC tests (a) and VCC tests (b), and evolution of qw/ ¼ /w1  /w2 with r1  r2 =R in VEC tests (c)
and VCC tests (d).

Please cite this article in press as: Shi, J., Guo, P., Induced fabric anisotropy of granular materials in biaxial tests along imposed strain paths, Soils Found.
(2018), https://doi.org/10.1016/j.sandf.2018.02.001
14 J. Shi, P. Guo / Soils and Foundations xxx (2018) xxx–xxx
ðr1  r2 =RÞmax is achieved. In all these cases, a significant
decrease in qw/ takes place when deformation becomes
unstable. In other words, the degradation of the weak
sub-network is initiated before deformation becomes
unstable. In the case of the VEI test (R ¼ 1), deformation
is always stable and practically no degradation of qw/
occurs.
Fig. 15 compares the relation between ðr1  r2 =RÞ and
fabric deviators in the weak and strong sub-networks in the
VEC (R > 1) and VCC tests (R < 1). In all cases, the
deformation is always stable since ðr1  r2 =RÞ never
reaches the maximum in these tests. qw/ increases monoton-
ically and approaches a steady state that depends on the
ultimate stress states; as shown Figs. 15(c) and (d). On
the other hand, qs/ only increases monotonically in the
VCC tests; see Fig. 15(b). However, in the VEC tests, a
peak value of qs/ develops at R ¼ 2:2 followed by a
decrease in qs/ ; as shown in Figs. 15(a) and 9(c).
Based on the discussion about the results presented in
Figs. 14 and 15, one may conclude that the degradation
of the strong sub-network (i.e., the decrease of qs/ ) is a nec-
essary but not a sufficient condition of deformation insta-
bility. In other words, deformation instability is
accompanied by strong network degradation but not the
opposite. On the other hand, a decrease in qw/ is the indica-
tor of deformation instability in tests along all imposed
strain paths.
7. Concluding remarks
In this study, the behaviour of granular materials in
biaxial tests along various strain paths was simulated using
the discrete element method. Without attempting to repro-
duce any laboratory test results, this study sheds light on
the capacity of discrete element methods to simulate the
constitutive features observed during laboratory tests, with
the microstructural context revealing information about
both the deformation instability and failure. To character-
ize the microstructure and its evolution in granular assem-
blies, the overall contact network was split into a strong
and a weak sub-network, both described by two indepen-
dent fabric tensors. The correlation between the evolution
of different contact networks and material behaviour
including deformation instability was examined. The fol-
lowing conclusions are obtained:
(1) The fabric evolution of the overall contact network
along an imposed strain path can be related to the
stress ratio (t/s). However, such relation varies with
the imposed dilatancy rate dev =dc .
(2) Along dilatant strain paths, a unique fabric-stress
relation r1 =r2 ¼ /s1 =/s2 , or equivalently
ð/1  /2 Þ=ð/s1 þ /s2 Þ ¼ t=s, is established in the
s s
strong sub-network. The maximum value of qs/ for
all tests occurs in K 0 -compression tests.
Please cite this article in press as: Shi, J., Guo, P., Induced fabric anisotropy of granular materials in biaxial tests along imposed strain paths, Soils Found.
(2018), https://doi.org/10.1016/j.sandf.2018.02.001
(3) The strong sub-network provides resistance to devia-
tor stresses of a granular assembly, while the weak
force chain network and its evolution are significantly
affected by the imposed dilatancy rate.
(4) The instability of a granular assembly is closely
related to the degradation or even collapse of the
weak force chain network, even though the strength
of the assembly depends on the strong sub-network.
Acknowledgement
Funding provided by the Natural Sciences and Engi-
neering Research Council of Canada and the China Schol-
arship Council (CSC) is gratefully acknowledged.
References
Antony, S.J., Momoh, R.O., Kuhn, M.R., 2004. Micromechanical
modelling of oval particulates subjected to bi-axial compression.
Comput. Mater. Sci. 29 (4), 494–498. https://doi.org/10.1016/
j.commatsci.2003.12.007.
Bathurst, R.J., Rothenburg, L., 1988. Micromechanical aspects of
isotropic granular assemblies with linear contact interactions. J. Appl.
Mech. 55 (1), 17–23. https://doi.org/10.1115/1.3173626.
Chang, C.S., Chao, S.J., Chang, Y., 1995. Estimates of elastic moduli for
granular material with anisotropic random packing structure. Int. J.
Solids Struct. 32 (14), 1989–2008. https://doi.org/10.1016/0020-7683
(94)00225-L.
Christoffersen, J., Mehrabadi, M.M., Nemat-Nasser, S., 1981. A microme-
chanical description of granular material behavior. J. Appl. Mech.
Trans. ASME 48 (2), 339–344. https://doi.org/10.1115/1.3157619.
Cundall, P.A., Strack, O.D.L., 1979. A discrete numerical model for
granular assemblies. Géotechnique 29 (1), 47–65. https://doi.org/
10.1680/geot.1979.29.1.47.
Darve, F., Laouafa, F., 2000. Instabilities in granular materials and
application to landslides. Mech. Cohesive-Frictional Mater. 5 (8), 627–
652. https://doi.org/10.1002/1099-1484(200011)5:8<627::AID-
CFM109>3.0.CO;2-F.
Drucker, D.C., 1957. A definition of stable inelastic material. J. Appl.
Mech. 26 (1), 101–106. https://doi.org/10.2307/1146143.
Fu, P., Dafalias, Y.F., 2015. Relationship between void- and contact
normal-based fabric tensors for 2D idealized granular materials. Int. J.
Solids Struct. 63, 68–81. https://doi.org/10.1016/j.ijsolstr.2015.02.041.
Fu, P., Dafalias, Y.F., 2011. Fabric evolution within shear bands of
granular materials and its relation to critical state theory. Int. J.
Numer. Anal. Methods Geomech. 35 (18), 1918–1948. https://doi.org/
10.1002/nag.988.
Gao, Z., Zhao, J., Li, X.S., Dafalias, Y.F., 2014. A critical state sand
plasticity model accounting for fabric evolution. Int. J. Numer. Anal.
Methods Geomech. 38 (4), 370–390. https://doi.org/10.1002/nag.2211.
Guo, P., Su, X., 2007. Effect of dilatancy on instability, pre-instability
strain softening of sand along proportional strain paths. Soils Found.
47 (4), 757–770.
Guo, P.J., Stolle, D.F.E., 2005. On the failure of granular materials with
fabric effects. Soils Found. 45 (4), 1–12.
Hill, R., 1958. A general theory of uniqueness and stability in elastic-
plastic solids. J. Mech. Phys. Solids 6 (3), 236–249. https://doi.org/
10.1016/0022-5096(58)90029-2.
Huang, X., Hanley, K.J., O’Sullivan, C., Kwok, C.Y., 2014. Exploring the
influence of interparticle friction on critical state behaviour using
DEM. Int. J. Numer. Anal. Methods Geomech. 38 (12), 1276–1297.
https://doi.org/10.1002/nag.2259.
 J. Shi, P. Guo / Soils and Foundations xxx (2018) xxx–xxx
Jiang, M., Yu, H., Harris, D., 2006. Kinematic variables bridging discrete
and continuum granular mechanics. Mech. Res. Commun. 33 (5), 651–
666. https://doi.org/10.1016/j.mechrescom.2005.06.013.
Kanatani, K.I., 1984. Stereological determination of structural aniso-
tropy. Int. J. Eng. Sci. 22 (5), 531–546. https://doi.org/10.1016/0020-
7225(84)90055-7.
Kruyt, N.P., 2012. Micromechanical study of fabric evolution in quasi-
static deformation of granular materials. Mech. Mater. 44, 120–129.
https://doi.org/10.1016/j.mechmat.2011.07.008.
Kuhn, M.R., Sun, W.C., Wang, Q., 2015. Stress-induced anisotropy in
granular materials: fabric, stiffness, and permeability. Acta Geotech.
10 (4), 399–419. https://doi.org/10.1007/s11440-015-0397-5.
Lade, P.V., 2008. Failure criterion for cross-anisotropic soils. J. Geotech.
Geoenviron. Eng. 134 (1), 117–124. https://doi.org/10.1061/(ASCE)
1090-0241(2008) 134:1(117).
Li, X.S., Dafalias, Y.F., 2011. Anisotropic critical state theory: role of
fabric. J. Eng. Mech. 138 (3), 263–275. https://doi.org/10.1061/
(ASCE)EM.1943-7889.0000324.
Maeda, K., Hirabayashi, H., Ohmura, A., 2006. Micromechanical influence
of grain properties on deformation-failure behaviors of granular media
by DEM. Geomech. Geotech. Particulate Media, 173–179.
Mehrabadi, M.M., Nemat-Nasser, S., 1983. Stress, dilatancy and fabric in
granular materials. Mech. Mater. 2 (2), 155–161. https://doi.org/
10.1016/0167-6636(83)90034-0.
Mohamed, A., Gutierrez, M., 2010. Comprehensive study of the effects of
rolling resistance on the stress–strain and strain localization behavior
of granular materials. Granul. Matter 12 (5). https://doi.org/10.1007/
s10035-010-0211-x, 3527-3271.
Muhunthan, B., Collins, I.F., 2003. On the relationship between stress–
dilatancy, anisotropy, and plastic dissipation for granular materials.
Géotechnique 53 (7), 611–618. https://doi.org/
10.1680/geot.2003.53.7.611.
Nakai, T., 1989. An isotropic hardening elastoplastic model for sand
considering the stress path dependency in three-dimensional stresses.
Soils Found. 29 (1), 119–137.
Ng, T., 2001. Fabric evolution of ellipsoidal arrays with different particle
shapes. J. Eng. Mech. 127 (10), 994–999. https://doi.org/10.1061/
(ASCE)0733-9399(2001) 127:10(994).
Nguyen, N.S., Magoariec, H., Cambou, B., Danescu, A., 2009. Analysis
of structure and strain at the meso-scale in 2D granular materials. Int.
J. Solids Struct. 46 (17), 3257–3271. https://doi.org/10.1016/j.
ijsolstr.2009.04.019.
Nicot, F., Daouadji, A., Hadda, N., Jrad, M., Darve, F., 2013. Granular
media failure along triaxial proportional strain paths. Eur. J. Environ.
Civ. Eng. 17 (9), 777–790. https://doi.org/10.1080/
19648189.2013.819301.
Nicot, F., Darve, F., 2006. Micro-mechanical investigation of material
instability in granular assemblies. Int. J. Solids Struct. 43 (11–12),
3569–3595. https://doi.org/10.1016/j.ijsolstr.2005.07.008.
Nicot, F., Darve, F., 2005. A multi-scale approach to granular materials.
Mech. Mater. 37, 980–1006. https://doi.org/10.1016/j.
mechmat.2004.11.002.
Nicot, F., Sibille, L., Donze, F., Darve, F., 2007. From microscopic to
macroscopic second-order work in granular assemblies. Mech. Mater.
39 (7), 664–684. https://doi.org/10.1016/j.mechmat.2006.10.003.
Nicot, F., Sibille, L., Hicher, P.Y., 2015. Micro-macro analysis of granular
material behavior along proportional strain paths. Contin. Mech.
Thermodyn. 27 (1–2), 173–193. https://doi.org/10.1007/s00161-014-
0347-8.
Oda, M., 1993. Inherent and induced anisotropy in plasticity theory of
granular soils. Mech. Mater. 16 (1–2), 35–45. https://doi.org/10.1016/
0167-6636(93)90025-M.
Please cite this article in press as: Shi, J., Guo, P., Induced fabric anisotropy of granular materials in biaxial tests along imposed strain paths, Soils Found.
(2018), https://doi.org/10.1016/j.sandf.2018.02.001
15
Oda, M., Konishi, J., Nemat-Nasser, S., 1982. Experimental microme-
chanical evaluation of strength of granular materials: effects of particle
rolling. Mech. Mater. 1 (4), 269–283. https://doi.org/10.1016/0167-
6636(82)90027-8.
Prunier, F., Laouafa, F., Lignon, S., Darve, F., 2009. Bifurcation
modeling in geomaterials: from the second-order work criterion to
spectral analyses. Int. J. Numer. Anal. Methods Geomech. 33 (9),
1169–1202. https://doi.org/10.1002/nag.762.
Radjaı̈, F., Delenne, J.Y., Azéma, E., Roux, S., 2012. Fabric evolution
and accessible geometrical states in granular materials. Granul. Matter
14 (2), 259–264. https://doi.org/10.1007/s10035-012-0321-8.
Radjaı̈, F., Jean, M., Moreau, J.-J., Roux, S., 1996. Force distributions in
dense two-dimensional granular systems. Phys. Rev. Lett. 77 (2), 274–
277. https://doi.org/10.1103/PhysRevLett. 77.274.
Radjaı̈, F., Roux, S., Moreau, J.J., 1999. Contact forces in a granular
packing. Chaos An Interdiscip. J. Nonlinear Sci. 9 (3), 544–550.
https://doi.org/10.1063/1.166428.
Radjaı̈, F., Wolf, D., Jean, M., Moreau, J.-J., 1998. Bimodal character of
stress transmission in granular packings. Phys. Rev. Lett. 80 (1), 61–64.
https://doi.org/10.1103/PhysRevLett. 80.61.
Rothenburg, L., Kruyt, N.P., 2004. Critical state and evolution of
coordination number in simulated granular materials. Int. J. Solids
Struct. 41 (21), 5763–5774. https://doi.org/10.1016/j.
ijsolstr.2004.06.001.
Rothenburg, L., Selvadurai, A.P.S., 1981. A micromechanical definition
of the Cauchy stress tensor for particulate media. Mech. Struct. media,
469–486.
Satake, M., 1987. Graph-theoretical approach to the mechanics of
granular materials. In: Spencer, A.J.M. (Ed.), Continuum Models of
Discrete Systems. Taylor & Francis, Boston, pp. 163–172.
Satake, M., 1978. Constitution of mechanics of granular materials
through the graph theory. Contin. Mech. Stat. approaches Mech.
Granul. Mater. 47–62.
Suiker, A.S.J., Chang, C.S., 2004. Modeling failure and deformation of an
assembly of spheres with frictional contacts. J. Eng. Mech. 130 (3),
283–293.
Thornton, C., 2000. Numerical simulations of deviatoric shear deforma-
tion of granular media. Géotechnique 50 (1), 43–53. https://doi.org/
10.1680/geot.2000.50.1.43.
Wan, R., Al-Mamun, M., Guo, P., 2007. Experimental investigation of
instabilities of granular materials in relation to dilatancy and fabric
issues. Bifurcations, Instabilities, Degradation in Geomechanics.
Springer, Berlin Heidelberg, pp. 71–93.
Wan, R., Pouragha, M., 2015. Fabric and connectivity as field descriptors
for deformations in granular media. Contin. Mech. Thermodyn. 27 (1–
2), 243–259. https://doi.org/10.1007/s00161-014-0370-9.
Wan, R.G., Al-Mamun, M., Guo, P.J., 2005. How do fabric and dilatancy
affect the strength of granular materials, in: Proceedings of the
International Conference on Soil Mechanics and Geotechnical Engi-
neering. pp. 863–868.
Wan, R.G., Guo, P.J., 2004. Stress dilatancy and fabric dependencies on
sand behavior. J. Eng. Mech. 130 (6), 635–645. https://doi.org/
10.1061/(ASCE)0733-9399(2004) 130:6(635).
Zhao, J., Guo, N., 2013. Unique critical state characteristics in granular
media considering fabric anisotropy. Géotechnique 63 (8), 695–704.
https://doi.org/10.1680/geot.12.P.040.
Zhu, H., Nicot, F., Darve, F., 2016. Meso-structure evolution in a 2D
granular material during biaxial loading. Granul. Matter 18 (1), 1–12.
https://doi.org/10.1007/s10035-016-0608-2.