You are on page 1of 14

Granular Matter (2016) 18:37

DOI 10.1007/s10035-016-0645-x

ORIGINAL PAPER

Shear strength of unsaturated granular soils: three-dimensional


discrete element analyses
Zhifu Shen1,2 · Mingjing Jiang 1,2 · Colin Thornton3

Received: 26 June 2015 / Published online: 7 May 2016


© Springer-Verlag Berlin Heidelberg 2016

Abstract To investigate the shear strength behavior of A new shear strength function for unsaturated soils, consid-
unsaturated granular soils, the three-dimensional discrete ering the anisotropic effects of matric suction, is proposed
element method has been used to model soils in triaxial com- and validated using experimental data in the literature.
pression tests. A simple but effective contact model with
an attractive capillary force was implemented. The effects Keywords Unsaturated granular soils · Shear strength ·
of matric suction and packing density on shear strength of Capillary water · Discrete element method · Fabric ·
unsaturated granular assemblies were examined. The Mohr– Anisotropy
Coulomb strength parameters (apparent cohesion and friction
angle) were fitted for each matric suction examined. The
results show that matric suction can increase the strength 1 Introduction
and modulus of granular soils and lead to increased dilation.
The peak friction angle depends on the packing density but Soil mechanics based on the effective stress concept [1]
seems independent of matric suction. The apparent cohesion has been successfully applied in describing the mechanical
increases with matric suction non-linearly at a decreasing behavior of saturated soils where soil pores are filled with liq-
rate. Similar values of cohesion were observed for both uid, e.g. water. However, unsaturated soils (where soil pores
dense and loose assemblies, which can be explained by the are filled with both gas and liquid phases) are more often
anisotropic distribution of capillary force network. Based on encountered in engineering practice, especially in arid/semi-
the microscopic observations, the stress-induced anisotropy arid areas where the underground water table is deep. The
of contact distributions leads to an anisotropic distribution failures of slopes, shallow foundations and embankments are
of capillary water and, as a consequence, the capillary stress often associated with poorly-predicted strength reduction of
is anisotropic, imposing a shear effect on an assembly in unsaturated soils due to the change of water content [2]. A
parallel with that imposed by the externally applied load- strong need to properly describe the strength of unsaturated
ing. Consequently, the strength of unsaturated granular soil soils therefore arises. However, the mechanical behavior of
is controlled by the combined effects of packing density, and unsaturated soils is not well understood at present due to the
the magnitude and degree of anisotropy of the capillary stress. complex interactions of its three constituent phases, i.e., soil
skeletons, pore fluids and pore gases.
B Mingjing Jiang Large numbers of laboratory experiments have been car-
mingjing.jiang@tongji.edu.cn ried out on unsaturated soils. Based on experimental data,
1 Department of Geotechnical Engineering, College of Civil
the shear strength function was predicted in a single-variable
Engineering, Tongji University, 1239 Siping Road, theory with Bishop’s ‘effective stress’ equation [3] or in a
Shanghai 200092, China double-variable theory [4]. Each theory includes a vague
2 State Key Laboratory of Disaster Reduction in Civil material parameter that changes with matric suction, creating
Engineering, Tongji University, Shanghai 200092, China difficulty in parameter calibration for engineering practice.
3 School of Chemical Engineering, University of Birmingham, More recently, strength functions have been associated with
Edgbaston, Birmingham B15 2TT, UK the soil–water characteristic curve (SWCC) of the soils,

123
37 Page 2 of 13 Z. Shen et al.

with which the strength can be predicted using the sat- normal vectors are investigated. The anisotropic effects of
urated strength parameters. Several such procedures have matric suction are examined and explained in the micro-
been proposed and validated using experimental data [5– scopic view using the stress-induced contact anisotropy. A
10]. Comprehensive comparisons have been made on the new shear strength function for unsaturated soils is proposed
capability of such procedures in predicting the strength of and validated using experimental data in the literature to
unsaturated soils [11–13]. It has been found that no proce- capture the anisotropic effects of matric suction on shear
dure can capture the strength behavior of different types of strength.
soils over a large suction range. Further investigation seems
necessary to establish a rigorous and accurate shear strength
2 DEM modeling of unsaturated granular soils
function.
The SWCC is usually measured in the laboratory with-
2.1 Simplified capillary force model
out applying stresses or only under a nominal stress state,
with the effects of stress history, stress level and microscopic
For unsaturated granular soils in the pendular state, capillary
fabric not considered. However, the stress state and soil struc-
water exists at inter-particle contact as liquid bridge which is
ture have been found to influence the SWCC [12,14–16].
isolated from each other while the air phase is continuous. At
SWCC has also been found to be related to the soil particle
each contact, the capillary water leads to an attractive force
size distribution [17,18]. These findings suggest the need to
acting on the two contacting grains. The capillary force FI
examine the strength behavior of unsaturated soils with more
includes contributions from both the matric suction and the
insight into the microscopic mechanisms. This is because the
surface tension of the contractile skin (air–water interface).
complicated behavior of unsaturated soil is dictated by the
Figure 1 presents a schematic diagram used to derive FI by
interactions of the three phases at the micro-scale level.
examining a cross-section halving the liquid bridge shared by
At the microscopic level, mono-sized regularly packed
two grains (i.e. “gorge method”). The grains are idealized as
sphere assemblies (e.g. simple cubic packing and tetrahe-
spheres for simplicity. The shape of the liquid bridge can be
dral packing) have been used to reveal the main features of
approximated by a toroidal curvature that can be described by
unsaturated granular soils [19,20]. General micro-structural
two radii R1 and R2 in Fig. 1. Note the following geometric
analyses have been carried out by incorporating capillary
relationships,
water effects into the mechanics of granular materials [20–
23]. The discrete element method (DEM), first proposed by
R1 = r (sec θ − 1) (1)
Cundall and Strack [24], has been widely used in geome-
chanics to numerically investigate the mechanical behavior R2 = −r (1 + tan θ − sec θ ) (2)
of soils which are idealized as dynamic polydisperse particle
systems with random packing [25–29]. With the introduc- where r is the sphere radius and θ is the half-filling angle.
tion of interparticle capillary forces in DEM contact models, The wetting angle is assumed zero.
the strength and deformation behavior of unsaturated granu- The matric suction Su in this paper is calculated using the
lar assemblies can be reproduced [18,30–34]. It is revealed theoretical work in [38],
from these microscopic analyses that the effects of capillary
water on the macroscopic behavior of unsaturated soils are Su = T (1/R1 + 1/R2 ) (3)
anisotropic due to its non-uniform distribution within soils
where T is the surface tension, 0.0735 N/m at a temperature
[20–22,34–36]. Such anisotropy results from the anisotropic
of 15 ◦ C.
arrangement of the soil grains and the contacts, which is
difficult to measure in experiments and has not yet been con-
sidered in shear strength functions and constitutive modeling
of unsaturated soils [37]. Instead, the effect of capillary water
is considered as an equivalent isotropic “effective stress”,
which is a quantity scaled from the matric suction. How to
consider the anisotropic effects of matric suction on the shear
strength of unsaturated granular soils constitutes the strong
motivation in this paper.
In the paper, a simplified three-dimensional capillary
water contact model is used in DEM modeling of unsaturated
granular material subjected to triaxial compression tests. The
strength and deformation behavior of the material as well as
the fabric described by the directional distribution of contact- Fig. 1 Liquid bridge and capillary force

123
Shear strength of unsaturated granular soils: three-dimensional discrete element analyses Page 3 of 13 37

Similar to [18], the inter-particle capillary force can be


expressed as:

FI = πR22 Su − 2πR2 T (4)

When a liquid bridge is shared by two spheres with differ-


ent radii r1 and r2 , the following harmonic mean radius can
be defined as:

r = 2r1r2 /(r1 + r2 ) (5)

It can be estimated according to [39] that this approxima-


tion gives an error within 5 %. This is the same level of
error introduced by the toroidal approximation as exam-
ined in [40]. The capillary force between two non-contacting Fig. 2 Mechanical components of the contact model with capillary
water (CW capillary water): a normal direction and b tangential direc-
spheres is not considered. The three strong assumptions (no tion
capillary force between non-contacting spheres, harmonic
mean radius approximation and toroidal approximation) have
already been relaxed in the literature [32,34]. Further refined force acts immediately. When the two spheres separate, the
simulations are indeed needed in this study. In reality, real capillary force disappears. Figure 3a presents the mechanical
grains are rough on the surface and often contact through responses implemented in DEM. The normal force Fn can
flat areas that change the shape and size of the liquid bridge, be expressed as:
leading to a capillary force different from the one determined 
K n u n − FI un ≥ 0
by Eq. (4) [41]. The realistic behavior of capillary water is Fn = (6)
very complex in terms of liquid bridge rupture and recovery 0 un < 0
[42,43]. In addition, there may be coupling effect between
where K n is the normal contact stiffness and u n is the overlap
the capillary force and rolling resistance [18]. Anyway, DEM
of a particle pair with a negative value denoting separation.
simulation can still help identify the most relevant aspects
Figure 2b presents the mechanical components of the
controlling the strength of unsaturated granular material,
contact model in the tangential direction considering the
although the DEM material is quite different from real soil in
effect of capillary force. The components consist of a spring
particle shape and capillary force. Note that the observation
and a dashpot for elastic property and energy dissipation,
in DEM simulation only serves as a qualitative guidance in
respectively. A slider is used so that tangential sliding would
quantitative evaluation of the strength envelope of real unsat-
occur only when the tangential force reaches the thresh-
urated soil in Sect. 4. The DEM defects are tolerated to reduce
old controlled by K n u n (a result of externally applied load
the additional complexity in the analyses and computational
and the capillary force). Figure 3b presents the perfectly
costs as long as the geotechnical engineering implication
elasto-plastic contact behavior considered in the tangential
from the DEM simulation is retained.
direction. The tangential force Fs is incrementally calculated
according to the incremental tangential displacement u s as:
2.2 Simplified contact model with capillary force
Fs ← Fs + K s u s (7)
Granular materials are idealized as sphere assemblies in
the DEM simulations in this study. A simple normal and where K s is the tangential contact stiffness.
tangential spring model with constant stiffness is adopted, The Mohr–Coulomb sliding criterion is adopted as:
regardless of possible rolling resistance at inter-particle con-
tacts. Figure 2a presents the mechanical components of |Fs | ≤ μK n u n (8)
the contact model in the normal direction plus the attrac-
tive capillary force. The mechanical components include a where μ is the inter-particle frictional coefficient.
spring with elastic behavior, a dashpot accounting for vis- Besides, viscous contact damping forces (Fnv in the normal
cous energy dissipation, and a divider preserving no tension direction and Fsv in the tangential direction) can be applied
as two spheres separate. In addition, due to the attractive when u n ≥ 0:
nature of the capillary force, a CW attractor (CW: Capillary

Water) is introduced in parallel with the spring and dash- Fnv = −cn u̇ n
(9)
pot. Once two spheres come into contact again, the capillary Fsv = −cs u̇ s

123
37 Page 4 of 13 Z. Shen et al.

Fig. 3 Mechanical responses of


the implemented contact model
in a normal direction and b
tangential direction

where cn , cs are viscous damping coefficients and u̇ n , u̇ s capillary forces (determined by the prototype particle size)
are relative normal and tangential velocities. Another type were amplified 106 times so that the model-scale behavior is
of damping, local damping, is also available which exerts a equivalent to the prototype-scale behavior because the stress
damping force proportional to unbalanced force on particle does not change with the scale-up. The similarity theorem
center. The two types of damping have no significant effects has been derived in [18].
in quasi-static behavior. Local damping has several advan- A uniform rectangular parallelepiped assembly consist-
tages identified in PFC3D [44] and therefore was used in this ing of 10 000 spheres with a slenderness of 2 (sample
paper with a dimensionless damping coefficient of 0.7. height/width = 2) was generated using the multi-layer under-
compaction method (UCM) [27]. With UCM, an assembly
2.3 DEM simulation procedure was generated layer by layer, five layers in total, in a fric-
tionless rigid wall container. Each layer consisted of an
The above contact model was implemented into a DEM code identical number of spheres. Frictionless wall was used to
PFC3D to carry out numerical simulations of triaxial com- apply principal stress on the boundary and to reduce stress
pression tests. A particle size distribution (PSD) typical of inhomogeneity. To ensure the uniformity of a dense assem-
silty sands was used for which the effects of capillary forces bly with a target void ratio (the ratio of void volume to the
are significant. Figure 4 presents the PSD having 10 types of solid volume) of 0.5 (packing fraction = 0.67), for exam-
diameter ranging from 6.6 to 10.5 μm with a mean diameter ple, the first layer of spheres were generated randomly and
d50 = 7.75 μm and a uniformity coefficient d60 /d10 = 1.2. one-dimensionally compressed to a void ratio of 0.528. Like-
If the prototype size (6.6 to 10.5 μm) is used, both the particle wise, the sphere assembly was compressed to a void ratio of
mass and contact force/capillary force are very small. Com- 0.527, 0.521, 0.512 and 0.500 respectively after the second,
putation error may arise when one small value is divided by third, fourth and fifth layer of spheres were generated. Sim-
another small value. This can be avoided by amplifying par- ilarly, the sequence of void ratios for a loose assembly with
ticle size. The particle sizes were amplified 103 times and the a target void ratio of 0.82 (packing fraction = 0.55) was
0.930, 0.914, 0.895, 0.870 and 0.820. In preparation of a
dense (loose) assembly, the inter-particle friction was set as
100
Percent smaller than by weight %)

0.0 (1.0) to allow the assembly to reach a dense (loose) state.


After UCM, inter-particle friction was set as 0.5 and was
80
assumed independent of matric suction. The way to generate
uniform assembly by changing friction was developed by trial
60
and error. Then, an assembly was isotropically compressed
Mean grain diameter to confining pressures of 50, 100, and 200 kPa. The gravity
40
d 50 = 7.75 m is not present in the simulation. Figure 5 presents a typical
Uniformity coefficient
model-scale (amplified) dense assembly before shear load-
20
Cu = d60 /d 10= 1.2 ing under a confining pressure of 50 kPa. Then, the capillary
force was implemented at each contact and the assembly was
0
5 6 7 8 9 10 11 allowed to reach equilibrium. Matric suction of 0 kPa (sat-
Sphere diameter ( m) urated or dry state), 50, 100, 200 and 400 kPa was applied,
respectively. This ensures that the assemblies with different
Fig. 4 Prototype particle size distribution used in the DEM simulations
matrix suctions have the same initial void ratio and particle

123
Shear strength of unsaturated granular soils: three-dimensional discrete element analyses Page 5 of 13 37

O are frictionless, the principal stresses of the net stress tensor


x y σi j (σ1 , σ2 , σ3 ) are normal to the wall faces, i.e. in the z-, x-,

257 mm
and y-axis directions, as illustrated in Fig. 5.
z Imagine that an effective contact force network in a dry
sample is first in equilibrium with boundary loads (“external”
relative to the soil material). Then, a capillary force network
124
mm mm (being “internal” because it arises from inter-particle capil-
124 lary water) is applied and the effective contact force network
has to adjust itself to reach a new equilibrium state. There-
Fig. 5 Typical dense assembly generated by UCM with amplified par-
fore, the effective contact force in an unsaturated sample is
ticle size (confining pressure σ3 = 50 kPa, amplified mean particle size
d50 = 7.75 mm, matric suction = 0 kPa, inter-particle friction coef- the result of both the externally applied loads and the capillary
ficient = 0.5, normal contact stiffness = 1.5 × 106 N/m, tangential forces. The effective contact force can be used to calculate a
contact stiffness = 1.0 × 106 N/m) stress tensor σij in DEM simulations over a Representative
Element Volume (REV) using forces acting on its boundary
S as:
arrangement. Then, any difference in mechanical behavior
can be attributed to matrix suction. The capillary forces were
1  β β
not considered at the contacts between spheres and bound- σij = xi F j (11)
ary walls. A total of 30 DEM numerical tests were carried Vm
β∈S
out on dense and loose assemblies. The normal and tangen-
tial contact stiffnesses were 1.5 × 106 and 1.0 × 106 N/m, β β
where Vm is the volume of REV, and xi and F j are the coor-
respectively. Quasi-static axial strain was applied at a rate of
dinate and the effective contact force vector at contact β on
5 %/min while keeping the confining pressure and matric suc-
the boundary S, i.e. Fβ = K n u n n+Fs where n is the contact
tion constant. The time step used in the DEM simulations was
normal and Fs is the tangential contact force vector. The ten-
6 × 10−6 s. Note that within unsaturated granular media, the
sor σi j is usually termed as effective stress from the general
existence of only pair-wise liquid bridges is possible for low
view of the mechanics of granular material, if the material
saturation degrees. With the increase of saturation degree, the
is dry or fully saturated where capillary force disappears,
radius of the menisci separating air and water increases. For a
since it uniquely controls both the strength and deformation
medium saturation degree, liquid cluster between three grains
of the material in these two cases. However, from the spe-
is more stable than three liquid bridges and the fusion of the
cific view of geomechanics, σi j in Eq. (11) cannot be used to
menisci occurs. Further increase of saturation degree would
uniquely define the constitutive behavior of soil regardless of
result in the voids abruptly flooding with water according
the saturation. For example, when a sample becomes unsat-
to Fredlund’s study [2], which can be numerically simulated
urated from a fully saturated state, the mean effective stress
by 2D DEM [18] but is extremely difficult to simulate by
obtained from Eq. (11) will increase and, according to fun-
3D DEM. This is why we did not carry out DEM tests on
damental knowledge in Soil Mechanics, the sample should
the assembly with matric suction between 0 and 50 kPa in
be more contractive, but numerical and experimental results
this paper. However, such simplicity does not affect the main
reveal that it behaves more dilatively. For this reason, this
conclusions in this study.
σi j was alternatively termed as generalized effective stress in
[18] and the term “effective stress” is left only for a stress
2.4 Stress variables in unsaturated granular assembly
tensor that can uniquely define the mechanical constitutive
law of unsaturated soil. Due to the geomechanics background
A stress tensor σi j due to loading on the boundary can be
of this study, the terminology in [18] is still followed here.
readily measured in laboratories or calculated in DEM sim-
It has been checked that using all contacts mapped into
ulations using boundary data. The conventional way widely
the REV actually yielded very close generalized effective
used to measure σi j is:
stresses to those calculated by Eq. (11). Two measurement
  spheres with a diameter of 2/5h (h: the height of an assem-
σi j = f i /s j (10) bly), each containing about 1,500 particles, are centered at
1/3h and 2/3h as REVs in order to measure σij during loading.
where f i is the load acting in the i direction on the boundary Since the generalized effective stress in an unsaturated sam-
area of s j with the normal direction as j. The stress tensor σi j ple is the result of both the total stress and the capillary stress
is equivalent to the net stress tensor in unsaturated soils since (stress due to capillary water effect), it is straightforward to
the pore air pressure u a is only taken as a reference pressure, extract the total stress σi j from the generalized effective stress
being zero in the DEM simulations. Since the boundary walls σij to calculate the capillary stress σiwj :

123
37 Page 6 of 13 Z. Shen et al.

σiwj = σij − σi j (12) slightly denser before shearing and show more significant
dilation at large axial strain than saturated assemblies. For
loose assemblies in the saturated state, as shown in Fig. 7,
strain-hardening and contractive behavior is observed. On
3 DEM simulation results the contrary, unsaturated loose assemblies are characterized
by dilation which is more apparent under lower confining
3.1 Stress–strain relationships pressure. Unsaturated loose assemblies demonstrate higher
moduli and deviator stresses than saturated assemblies under
Figures 6 and 7 present the stress–strain relationships of the same confining pressure.
dense and loose assemblies obtained from DEM simulations Generally, for the range of matric suction investigated,
in the saturated state (Su = 0 kPa) and unsaturated state, capillary water can increase the strength and initial modulus
respectively. For conciseness, only the simulation results of granular assemblies and induce more dilation regardless
from Su = 200 kPa are shown for the unsaturated state. of the density and confining pressure.
The stresses in Figs. 6 and 7 are net stresses calculated by
Eq. (10) and the strains were calculated according to the
motion of the boundary walls. 3.2 Strength parameters
Dense assemblies in Fig. 6a demonstrate strain-softening
behavior with the initial moduli and peak deviator stresses The shear strength under each confining pressure means the
increasing with confining pressures. The modulus and peak peak deviator stress for dense assemblies in Fig. 6 (at an axial
deviator stress are higher of the unsaturated assembly than strain around 4 %) or the deviator stress at an axial strain of
of the saturated counterpart under each confining pressure 15 % for loose assemblies in Fig. 7. The strength parameters
examined. Dense assemblies in Fig. 6b show apparent dila- (friction angle and apparent cohesion) are obtained by fitting
tion after the initial contraction. Unsaturated assemblies are the test data (described by net stresses) against the Mohr–

500 0.75
(kPa)

Su=200 kPa
Su= 0 kPa
400 0.70
=200 kPa
3

3
- 1

Void ratio

300
Deviator stress

0.65
200 3
=100 kPa
Su=200 kPa Su=0 kPa
0.60 3
= 50 kPa
100
3
=100 kPa

3
=50 kPa 3
=200 kPa
0 0.55
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
Axial strain (%) Axial strain (%)
(a) (b)
Fig. 6 Mechanical behavior of dense assemblies described by net stresses (under confining pressure of σ3 = 50, 100 and 200 kPa and matric
suction of Su = 0 and 200 kPa). a Deviator stress versus axial strain. b Void ratio versus axial strain

300 0.85
(kPa)

Su=200 kPa Su=200 kPa Su=0 kPa


250 Su= 0 kPa 3
= 50 kPa
3

3
=100 kPa
-

200 0.80
1

Void ratio

3
=200 kPa
Deviator stress

3
=200 kPa
150
3
=100 kPa
100 0.75

50
3
=50 kPa
0 0.70
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
Axial strain (%) Axial strain (%)
(a) (b)
Fig. 7 Mechanical behavior of loose assemblies described by net stresses. a Deviator stress versus axial strain. b Void ratio versus axial strain

123
Shear strength of unsaturated granular soils: three-dimensional discrete element analyses Page 7 of 13 37

the principal stresses of the generalized effective stress σij .

Apparent cohesion (kPa)


30 8
Friction angle ( ) 28 Figure 9 shows that there seems to be a unique peak strength
26 6
envelope for granular assemblies even with different matric
24
22
4 suctions provided that the generalized effective stresses are
20 used. According to Mohr–Coulomb strength theory, sin ϕ 
2
18 Friction angle
Dense
Cohesion
Dense (ϕ  is the friction angle) equals to the slope of the strength
Loose Loose
16 0 envelop in Fig. 9. The friction angle deduced from Fig. 9
are 28.1◦ (20.0◦ ) for dense (loose) assemblies, which is very
0 100 200 300 400
Matric suction (kPa)
close to the average friction angle of 28.6◦ (19.9◦ ) deduced
Fig. 8 Effect of matric suction on strength parameters described by from Fig. 8. It is hence recognized that the generalized effec-
net stresses tive stresses, or definitely the microscopic effective contact
forces used to define the generalized effective stresses, deter-
mine the strength of granular materials, irrespective of the
Coulomb strength envelope. Figure 8 presents the effect of saturation state (i.e. dry, saturated or unsaturated).
matric suction on the strength parameters of granular mate-
rials. Figure 8 shows that dense assemblies exhibit greater
3.3 Critical microscopic mechanisms
peak friction angle than loose assemblies but seemingly have
the same cohesion as loose assemblies subjected to the same
Figure 10 presents the effects of confining pressure and
matric suction. Figure 8 also shows that the capillary force
matric suction on coordination number Z of loose and dense
leads to a non-linear increase in cohesion with matric suc-
assemblies. Rattlers with less than 2 contacts were excluded
tion regardless of the density. However, the capillary force
in calculating Z . Figure 10a, b shows that upon the imple-
has little effect on the friction angle, which is because the
mentation of capillary forces at contacts before shearing,
inter-particle friction coefficient is assumed independent of
Z increases since capillary forces tend to clump particles
matric suction. This agrees with experiment results on one
together to slightly change the micro-structure to a denser
type of soils (e.g. in [45–47]). However, experiment results
state, as demonstrated by the slight decrease of void ratio
on the other type of soils indicate that the peak friction angle
in Figs. 6b and 7b. Z increases obviously with the confin-
does change with matric suction (e.g. in [16,48]), implying
ing pressure and increase slightly with matric suction within
that the frictional coefficient may be dependent on matric
the examined range. Figure 10c shows that the coordination
suction. Therefore, the conclusions in this study should be
number of the dense assembly under each confining pres-
applied to soils of which the friction angle does not vary
sure is larger than that of the loose counterpart and they
greatly with matric suction. However, the major topic in this
finally merge. Three critical microscopic mechanisms behind
study, i.e. the anisotropic effects of capillary water as will be
the observed macroscopic behavior of unsaturated granular
discussed below, is important for both types of soils.
materials can be identified.
Figure 9 presents the stress paths and peak strength
envelopes described by the generalized effective stresses
under
  different
 matric suctions,
  where
 q  = σ1 /2 − (1) The higher coordination number in an unsaturated
σ2 + σ3 /4, p = σ1 /2 + σ2 + σ3 /4 and σ1 , σ2 , σ3 are
    assembly than in a saturated one induces higher shear

250 150
sin ' = 0.471 sin ' = 0.342
o
' = 28.1 o ' = 20.0
Deviator q' (kPa)

200
Deviator q' (kPa)

Su=0 kPa 100 Su=0 kPa


150 Su=100 kPa Su=100 kPa
Su=200 kPa Su=200 kPa
Su=400 kPa
100 Su=400 kPa
50
50
=200 kPa
3
3
=200 kPa
0 0
0 100 200 300 400 500 0 100 200 300 400
Mean stress p' (kPa) Mean stress p' (kPa)
(a) (b)

Fig. 9 Stress paths and peak strength envelopes described by the generalized effective stresses under different matric suctions. a Dense assemblies.
b Loose assemblies

123
37 Page 8 of 13 Z. Shen et al.

3 50kPa 3 100kPa 3 200kPa 50kPa 100kPa 200kPa


3 3 3
6.4 1 =0% 5.0 1 =0%
4.8
6.3 4.6

Z
6.2 4.4
4.2
6.1 4.0
0 100 200 300 400 0 100 200 300 400
5.1 =4% 5.0
1=15%
1

4.8
4.8
Z

Z
4.6
4.5 4.4
0 100 200 300 400 0 100 200 300 400
Matric suction (kPa) Matric uction (kPa)
(a) (b)

6.5
Su=200 kPa Dense sample
6.0 Loose sample
Coordinate number Z

5.5 =100 kPa


3
3
=200 kPa
5.0

4.5
3
=50 kPa
4.0
0 2 4 6 8 10 12 14 16
Axial strain (%)
(c)
Fig. 10 Effects of confining pressure and matric suction on coordination number. a Dense sample at axial strains of ε1 = 0 and 4 %. b Loose
sample at axial strains of ε1 = 0 and 15 %. c Evolution of coordination number with axial strain

modulus and greater shear strength described by the net 3.4 Anisotropic effects of matric suction
stresses, and enhances dilation macroscopically.
(2) The capillary force can increase the effective normal Figure 11 presents the eigenvalues of the capillary stress σiwj
force and hence increase the sliding resistance at a con- with σ1w in the z-axis direction and σ2w (σ3w ) lying on the
tact. According to Eq. (6), u n increases when FI enters to x–y plane (σiwj is found to be coaxial with σi j and σij ). Fig-
play if Fn counterbalancing the externally applied loads ure 11 shows that σiwj is not isotropic because σ1w is greater
remains fixed. In this way, according to Eq. (8), FI con- than σ2w and σ3w and the difference increases with axial
tributes to the tangential sliding resistance at a contact by strain. The values of σ2w and σ3w are very close, indicating
increasing the effective contact force K n u n . Therefore, an axis-symmetric state. For dense assemblies, σ1w fluctu-
due to FI , an increased strength described by net stresses ates significantly while σ2w and σ3w decrease at a decreasing
is expected since the macroscopic strength is essentially rate with the increase in axial strain. For loose assemblies, σ1w
controlled by the sliding resistance at contacts. first increases slightly while σ2w and σ3w fluctuate over a small
(3) A higher coordination number and therefore more liq- range with the increase in axial strain. All σiw (i = 1, 2, 3)
uid bridges in a dense assembly are expected to result tend to be steady at an axial strain of 15 %.
in a higher apparent cohesion than in a loose assembly Considering the anisotropy and axis-symmetry, the stress
if described by the net stresses. However, the apparent tensor σiwj can be rewritten as:
cohesion under each matric suction examined in Fig. 8
appears to be independent of the density of granular
⎡ ⎤
assemblies. This can be explained by the stress-induced σ1w 0 0
anisotropy of the directional contact distribution and, σiwj = ⎣0 σ2w 0 ⎦
therefore, of the capillary force network within an assem- 0 0 σ3w
bly, which in turn leads to the anisotropic effect of matric ⎡ ⎤
1 + 2w 0 0
suction. This phenomenon will be discussed in detail in
= σmw ⎣ 0 1 − w 0 ⎦ (13)
the following sub-section.
0 0 1 − w

123
Shear strength of unsaturated granular soils: three-dimensional discrete element analyses Page 9 of 13 37

40 40 40 w
1
35 w
35 w 35
1 1

(kPa)

i (kPa)

(kPa)
30 30 30
25 25 25

w
w

i
i
w w
20 20 20 2 3
w w w w
15 2 3 15 2 3 15
=50 kPa =100 kPa 3
=200 kPa
3 3 10
10 10 0 2 4 6 8 10 12 14 16
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
Axial strain (%) Axial strain (%) Axial strain (%)
(a)

40 40 40
35 35 35
i (kPa)

(kPa)
30 w 30 30

(kPa)
w w
1
1 1
25 25 25
w

w
i
20 20 20

i
w w w w
w w 2 3 2 3
15 2 3 15 15
3
=50 kPa 3
=100 kPa 3
=200 kPa
10 10 10
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
Axial strain (%) Axial strain (%) Axial strain (%)
(b)
Fig. 11 Evolution of the eigenvalues of the capillary stress (Su = 200 kPa). a Dense assemblies, b loose assemblies

w
40 0.30
Su=200 kPa Dense Loose Capillary water anisotropy index Su=200 kPa Dense Loose
35
50 kPa 0.25 50 kPa
3
=100 kPa 3
=100 kPa
200 kPa 0.20 200 kPa
m (kPa)

30
0.15
w

25
0.10
20 0.05

15 0.00
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
Axial strain (%) Axial strain (%)

Fig. 12 Evolution of σmw with axial strain (Su = 200kPa) Fig. 13 Evolution of capillary water anisotropy index w with axial
strain (Su = 200 kPa)

 
where σmw = σ1w + σ2w +σ3w /3  is the
 mean value and rate with axial strain for both dense and loose assemblies
w = σ1w − σ2w + σ3w /2 / 3σmw is an anisotropy under each confining pressure. A higher value of w is
index. observed in dense assemblies and under lower confining pres-
Figure 12 presents the evolution of σmw with axial strain sures.
under a matric suction of Su = 200 kPa. Figure 12 shows Note that σiwj essentially depends on the contact distri-
that σmw in a dense assembly decreases under each confining bution that determines the distributions of capillary water
pressure with axial strain and then fluctuates about a mean and capillary force network in a granular assembly. Loading
value after the axial strain exceeds 4 %. For loose assemblies, can induce a higher proportion of contacts along the loading
σmw increases slightly before the axial strain exceeds 1 % and direction compared to the proportion in the x–y plane. This
then fluctuates in a small range. As expected, a higher σwm stress-induced contact anisotropy leads to the anisotropy of
is observed under a higher confining pressure or in a dense σiwj as revealed in Fig. 11. To quantitatively examine the con-
assembly due to the greater coordination number and there- tact anisotropy, a second-order contact normal fabric tensor
fore more liquid bridges. Fc can be defined to represent the statistical arrangement of
Figure 13 presents the evolution of the anisotropy index inter-particle contact directions in an assembly. According to
w with axial strain under a matric suction of Su = 200 kPa. the work in [25], the tensor component Fc,i j can be calculated
Figure 13 shows that w generally increases at a decreasing as

123
37 Page 10 of 13 Z. Shen et al.

1 c k k
2N matric suction will be discussed in the next section, in which
Fc,i j = ni n j (14) the similar cohesion of dense and loose assemblies observed
2Nc
k=1 in Fig. 8 can be explained.

where Nc is the total contact number and n ik (i = 1, 2, 3) are


the direction cosines of n, which is a unit vector normal to
4 Shear strength function considering anisotropic
the tangential plane at a contact.
effects of matric suction
Let Fci denote the eigenvalues of Fc with i = 1, 2, 3
and Fc1 + Fc2 + Fc3 = 1. The tensor Fc exhibits an axis-
The above DEM simulations provide evidence of the anisotro-
symmetric state with the direction of F1c very close to the
pic effect of matric suction which definitely produces an
z-axis direction and F2c being almost equal to F3c on the
internal shear effect in a granular assembly that can move
x–y plane. An index measure of the contact anisotropy can
the assembly towards failure even under externally applied
be defined as c = Fc1 − (Fc2 + Fc3 ) /2, so that Fc can be
isotropic loading. Under externally applied deviator load-
written as
ing, this anisotropic effect of matric suction can bring the
⎡ ⎤ ⎡ ⎤ assembly to failure. Consequently, the strength parameters of
Fc1 0 0 1 + 2c 0 0 unsaturated soils usually described by the net stresses (cor-
1
Fc = ⎣ 0 Fc2 0 ⎦ = ⎣ 0 1 − c 0 ⎦
responding to the externally applied loading) should depend
0 0 Fc3 3 0 0 1 − c
on the anisotropic distributions of the capillary forces. This
(15) issue will be clarified by examining an assembly loaded in
an axis-symmetric way. Based on the observed uniqueness of
Figure 14 presents the evolution of c with axial strain. strength envelope when described by the generalized effec-
The observed initial anisotropy in Fig. 14 is attributed to tive stress in Fig. 9, the failure condition can be expressed
the one-dimensional compression during sample prepara- as
tion. Figure 14 shows that c increases with axial strain at    
   σ1 + σ1w f − σ3 + σ3w f
a decreasing rate and then decreases slightly after peak for σ1 − σ3
a dense assembly or approaches an asymptotic value for a =    
σ1 + σ3 f σ1 + σ1w f + σ3 + σ3w f
loose assembly. Moreover, dense assemblies exhibit greater  
anisotropy than loose ones and assemblies under lower con- (σ1 − σ3 ) f + σ1w − σ3w f
=   = sin ϕ  (16)
fining pressures are more prone to become anisotropic. (σ1 + σ3 ) f + σ1w + σ3w f
The evolutions of w in Fig. 13 and c in Fig. 14 show
adequate similarity, confirming the aforementioned mecha- where ϕ  is the effective friction angle, assumed independent
nism leading to the anisotropy of σiwj . Since stress-induced of matric suction.
anisotropy is an intrinsic feature of granular materials and Substituting Eq. (13) into (16) and rearranging leads to
some inherent anisotropy can be observed as well before
deviator loading, the accompanied anisotropic effects of (σ1 − σ3 ) f = (σ1 + σ3 ) f sin ϕ 
matric suction need to be considered in advanced modeling of   ∗
unsaturated granular materials. As a preliminary application, + 2 sin ϕ  − 3 − sin ϕ  ∗w σmw (17)
a shear strength function considering anisotropic effects of ∗
where ∗w and σmw are the values of w and σmw at failure,
respectively.
0.14
From Eq. (17), the apparent cohesion due to matric suc-
c

0.12 tion, denoted by cs , can be derived as


Contact anisotropy index

0.10    
3 − sin ϕ  ∗ ∗
0.08 cs = 1 − 
w σmw tan ϕ  (18)
2 sin ϕ
0.06
Dense Loose Equation (18) implies that cs depends on the effective fric-
0.04
50 kPa
Su=200 kPa =100 kPa tion angle ϕ  and the two quantities characterizing the effect
0.02 3 ∗
200 kPa of capillary water (σmw and ∗w ). Generally, a dense assem-

0.00
0 2 4 6 8 10 12 14 16 bly exhibits greater ϕ  , σmw and ∗w than a loose one, which
Axial strain (%) however cannot mathematically guarantee a greater apparent
cohesion for dense assembly due to the combined effects of

Fig. 14 Evolution of contact anisotropy index c ϕ  , σmw and ∗w .

123
Shear strength of unsaturated granular soils: three-dimensional discrete element analyses Page 11 of 13 37

0.20

w
0.16

w
Su=200 kPa Su=200 kPa

Capillary water anisotropy index

Capillary water anisotropy index


0.16 50 kPa 50 kPa
3
=100 kPa 0.12 =100 kPa
3
200 kPa 200 kPa
0.12
0.08
0.08
0.04
0.04

0.00 0.00
0.00 0.04 0.08 0.12 0.16 0.00 0.04 0.08 0.12
Contact anisotropy index c Contact anisotropy index c
(a) (b)
Fig. 15 Relationship between contact anisotropy index c and capillary water anisotropy index w . a Dense assemblies. b Loose assemblies

The term ∗w in Eq. (18) represents the degree of With Eqs. (19) and (20), the shear strength function in the
anisotropy of σiwj in unsaturated soils at failure. Figure 15 τ − σ plane can be written in a complete form as
presents the relationship between w and c obtained in
DEM simulations, which shows that w generally increases τ = c + (σn − u a ) tan ϕ 
with c in a non-linear way. The DEM results imply that ∗w    
3 − sin ϕ 
can be correlated to the degree of contact anisotropy ∗c (the + 1− (1 − ξ )b ∗c Su ξ d tan ϕ  (21)
value of contact anisotropy index c at failure). Note that c 2 sin ϕ 
is used to describe the anisotropy of solid contacts while w
principally should describe the anisotropy of capillary forces where c is the effective cohesion of soil measured in the
due to liquid bridges existing at solid contacts as well as at saturated state and (σn − u a ) is the net normal stress with σn
inter-particle gaps. The effect of capillary forces at inter- and u a being the applied normal force on the failure plane
particle gaps, which exist in reality but were neglected for and the pore air pressure, respectively.
simplicity in our DEM simulations, would lower the correla- Since water migration is less hindered by the water-
tion between ∗w and ∗c . With a decrease in water saturation, particle surface interactions of coarse particles than fine
the proportion of capillary water at inter-particle gaps will particles, the anisotropic effect of matric suction on strength
decrease due to liquid bridge rupture, leading to an increased of granular soils can be more apparent than that for clayey
proportion of capillary water at solid contacts. That is, w soils. For this reason, the triaxial compression tests in [16]
should depend more on c for lower water saturation. With on Korean weathered granite (a kind of non-plastic silty sand
this view, a power law is assumed to associate ∗w with ∗c with c = 19.3 kPa and average ϕ  = 42.6◦ ) are used to pro-
through a normalized water content variable ξ as vide a preliminarily validation of the proposed shear strength
function given by Eq. (21). A slight variation of 2◦ in the
friction angle ϕ  was reported for this weathered granite in
∗w = (1 − ξ )b ∗c (19)
the range of matric suction examined, which is neglected
for practical purpose by using its average value. Figure 16
where b is a material parameter. The normalized water con- presents the SWCC of the weathered granite at zero net
tent variable ξ can be the degree of saturation, the normalized stress and the measured/predicted shear strength values. The
volumetric water content, the normalized gravimetric water weathered granite was tested under confining pressures of
content, etc. 0–300 kPa with a matric suction range of 0–300 kPa. The
Many functions have been proposed to associate σmw∗ in SWCC was used to obtain the value of ξ (degree of satura-
Eq. (18) with some measurable variables in laboratory exper- tion) for each matric suction Su in Eq. (21). The parameters
iments, e.g. matric suction and water saturation. The function b, ∗c and d in Eq. (21) were fitted with b and d stress-level
proposed in [8] was reported to give a good fit to a wide range independent, making Eq. (21) easy to use in practice. A good
of experimental results, will be used in this paper, i.e. agreement is observed between the predicted and measured
shear strengths in Fig. 16b. Note that the fitting parameter
∗c in Fig. 16b decreases with the increase of net normal
σmw∗ =Su ξ d (20)
stress, implying less anisotropic contact distribution under
higher stress-level, which is consistent with the observations
where d is a material parameter. in DEM simulations shown in Fig. 14.

123
37 Page 12 of 13 Z. Shen et al.

100 500
weathered granite b=2.41 d=1.03

Degree of saturation (%)

Shear strength (kPa)


80 400 ( n ua)=300 kPa, *
c = 0.090

60 300 *
( n ua)=200 kPa, c = 0.259

40 200 ( n ua)=100 kPa, *


c = 0.400

*
20 100 ( n ua)=0 kPa, c = 0.556

0 -1 0 1 2 3
0
10 10 10 10 10 0 100 200 300 400
Matric suction S u(kPa) Matric suction Su (kPa)
(a) (b)
Fig. 16 Unsaturated behavior of Korean weathered granite (experimental data modified after [16]): a SWCC at zero net stress and b measured
(points) and predicted (curves) shear strength values

The above interpretation of shear strength considering an anisotropic capillary stress. This anisotropy imposes a
anisotropic effect of matric suction is preliminary and ten- shear effect on the assembly and consequently the appar-
tative. Similar observations about the anisotropic effects of ent cohesion of unsaturated granular soils turns out to be
matric suction may be the same with other cohesive contact a function of the packing density, and the magnitude and
models. Experiments on soils under the same initial stress degree of anisotropy of the capillary stress.
state but applying different stress paths to lead to different (3) A new shear strength function for unsaturated granular
fabric evolutions are needed to further validate the proposed soils has been proposed incorporating the measurable
function. However, although Eq. (21) is proposed for soils matric suction and water saturation using three fitting
with the effective friction angle ϕ  independent of matric suc- parameters. An example of good agreement with exper-
tion, it can be applied to the other type of soils by considering imental data is provided.
ϕ  as a function of matric suction. Moreover, the work here
demonstrates that although Eq. (4) is generally valid only for Acknowledgments The research is funded by China National Funds
for Distinguished Young Scientists with Grant No. 51025932, Program
low saturation degrees, the DEM results can help to establish
of Shanghai Academic Chief Scientist with Grant No. 11XD1405200,
Eq. (21) that appears to be suitable for soils at a wide range and the National Natural Science Foundation of China with Grant No.
of saturation degree. 51179128, all of which are sincerely appreciated. The authors would
also thank the second author’s former Ph.D. student Haijun Hu for his
assistance in numerical simulations.

5 Conclusions
References
This paper presents numerical results of three-dimensional
(3D) DEM modeling of unsaturated granular soils in triaxial 1. Terzaghi, K.: The shear resistance of saturated soils. In: Pro-
compression tests. The anisotropic effects of matric suction ceedings of 1st International Conference on Soil Mechanics and
on shear strength are clarified. Three major conclusions are Foundation Engineering, vol. 1, pp. 54–56 (1936)
2. Fredlund, D.G., Rahardjo, H.: Soil Mechanics for Unsaturated
reached.
Soils. Wiley, New York (1993)
3. Bishop, A.W., Blight, G.E.: Some aspects of effective stress in
(1) The simplified capillary force contact model can repro- saturated and partly saturated soils. Géotechnique 13(3), 177–197
(1963)
duce the typical mechanical responses of unsaturated/sa- 4. Fredlund, D.G., Morgenstem, N.R., Widger, R.A.: The shear
turated dense and loose assemblies in triaxial compres- strength of unsaturated soils. Can. Geotech. J. 15, 313–321 (1978)
sion tests. Matric suction can apparently increase the 5. Fredlund, D.G., Xing, A.: Equations for the soil–water character-
strength and shear modulus of granular soils and enhance istic curve. Can. Geotech. J. 31, 517–532 (1994)
6. Rohm, S.A., Vila, O.M.: Shear strength of an unsaturated sandy
volumetric dilation. The apparent cohesion, obtained by soil. In: Proceedings of 1st International Conference on Unsatu-
fitting the Mohr–Coulomb strength envelope to the data, rated Soils, vol. 1, p. 189 (1995)
increases non-linearly with matric suction at a decreas- 7. Fredlund, D.G., Xing, A., Fredlund, M.D., Barbour, S.L.: The rela-
ing rate. tionship of the unsaturated soil shear strength to the soil–water
characteristic curve. Can. Geotech. J. 33, 440–448 (1996)
(2) The stress-induced anisotropy of contact distributions 8. Vanapalli, S.K., Fredlund, D.G., Pufahl, D.E., Clifton, A.W.: Model
leads to an anisotropic distribution of the capillary force for the prediction of shear strength with respect to soil suction. Can.
network within the examined assembly, which results in Geotech. J. 33, 379–392 (1996)

123
Shear strength of unsaturated granular soils: three-dimensional discrete element analyses Page 13 of 13 37

9. Oberg, A., Sallfors, G.: Determination of shear strength parameters 29. Jiang, M.J., Shen, Z.F., Thornton, C.: Microscopic contact model of
of unsaturated silts and sands based on the water retention curve. lunar regolith for high efficiency discrete element analyses. Com-
Geotech. Test. J. 20(1), 40–48 (1997) put. Geotech. 54, 104–116 (2013)
10. Khatlili, N., Khabbaz, M.H.: A unique relationship for the deter- 30. Gili, J.A., Alonso, E.E.: Microstructural deformation mechanisms
mination of the shear strength of unsaturated soils. Géotechnique of unsaturated granular soils. Int. J. Numer. Anal. Met. Geomech.
48(5), 681–687 (1998) 26, 433–468 (2002)
11. Vanapalli, S.K., Fredlund, D.G.: Comparison of different proce- 31. Liu, S.H., Sun, D.A.: Simulating the collapse of unsaturated soil by
dures to predict unsaturated soil shear strength. In: Shackleford, DEM. Int. J. Numer. Anal. Met. Geomech. 26(6), 633–646 (2002)
C., Houston, S.L., Chang, N.-Y. (eds.) Advances in unsaturated 32. Richefeu, V., El Youssoufi, M.S., Peyroux, R., Radja, F.: A model
geotechnics, Geotechnical Special Publication 99, pp. 195–209. of capillary cohesion for numerical simulations of 3D polydisperse
ASCE (2000). doi:10.1061/40510(287)13 granular media. Int. J. Numer. Anal. Met. Geomech. 32(11), 1365–
12. Kim, W.S., Borden, R.H.: Influence of soil type and stress state on 1383 (2008)
predicting shear strength of unsaturated soils using the soil–water 33. Shamy, UEl, Gröger, T.: Micromechanical aspects of the shear
characteristic curve. Can. Geotech. J. 48(12), 1886–1900 (2011) strength of wet granular soils. Int. J. Numer. Anal. Met. Geomech.
13. Sheng, D.C., Zhou, A., Fredlund, D.G.: Shear strength criteria for 32(14), 1763–1790 (2008)
unsaturated soils. Geotech. Geol. Eng. 29(2), 145–159 (2011) 34. Scholtès, L., Hicher, P.Y., Nicot, F., Chareyre, B., Darve, F.: On
14. Nimmo, J.R.: Modeling structural influences on soil water reten- the capillary stress tensor in wet granular materials. Int. J. Numer.
tion. Soil Sci. Soc. Am. J. 61, 712–719 (1997) Anal. Met. Geomech. 33(10), 1289–1313 (2009)
15. Vanapalli, S.K., Fredlund, D.G., Pufahl, D.E.: The influence of 35. Radjai, F., Richefeu, V.: Bond anisotropy and cohesion of wet gran-
soil structure and stress history on the soil–water characteristics of ular materials. Philos. T. R. Soc. A 367(1909), 5123–5138 (2009)
a compacted till. Géotechnique 49, 143–159 (1999) 36. Zhang, W., Zhao, C.: Micromechanics analysis for unsaturated
16. Lee, I.M., Sung, S.G., Cho, G.C.: Effect of stress state on the unsat- granular soils. Acta Mech. Solids Sin. 24(3), 273–281 (2011)
urated shear strength of a weathered granite. Can. Geotech. J. 42(2), 37. Sheng, D.C.: Review of fundamental principles in modelling unsat-
624–631 (2005) urated soil behaviour. Comput. Geotech. 38(6), 757–776 (2011)
17. Fredlund, M.D., Wilson, G.W., Fredlund, D.G.: Use of the grain- 38. Fisher, R.A.: On the capillary forces in an ideal soil. J. Agric. Sci.
size distribution for estimation of the soil–water characteristic 16, 492–505 (1926)
curve. Can. Geotech. J. 39(5), 1103–1117 (2002) 39. Harireche, O., Faramarzi, A., Alani, A.M.: A toroidal approxima-
18. Jiang, M.J., Leroueil, S., Konrad, J.M.: Insight into shear strength tion of capillary forces in polydisperse granular assemblies. Granul.
functions of unsaturated granulates by DEM analyses. Comput. Matter. 15(5), 573–581 (2013)
Geotech. 31(6), 473–489 (2004) 40. Lechman, J., Lu, N.: Capillary force and water retention between
19. Likos, W.J., Lu, N.: Hysteresis of capillary stress in unsaturated two uneven-sized particles. J. Eng. Mech. 134(5), 374–384 (2008)
granular soil. J. Eng. Mech. 130(6), 646–655 (2004) 41. Cho, G.C., Santamarina, J.C.: Unsaturated particulate materials—
20. Wan, R., Khosravani, S., Nicot, F.: Micromechanical analysis of particle-level studies. J. Geotech. Geoenviron. Eng. (ASCE)
stress in an unsaturated granular medium. In: The 2nd Interna- 127(1), 84–96 (2001)
tional Symposium on Computational Geomechanics (ComGeo II), 42. Mason, G., Clark, W.C.: Liquid bridges between spheres. Chem.
Cavtat-Dubrovnik (2011) Eng. Sci. 20, 859–866 (1965)
21. Li, X.S.: Effective stress in unsaturated soil: a microstructural 43. Lian, G., Thornton, C., Adams, M.J.: A theoretical study of the
analysis. Géotechnique 53, 273–277 (2003) liquid bridge forces between two rigid spherical bodies. J. Colloid
22. Molenkamp, F., Nazemi, A.H.: Micromechanical considerations Interf. Sci. 161(1), 138–147 (1993)
of unsaturated pyramidal packing. Géotechnique 53(2), 195–206 44. Itasca Consulting Group Inc. PFC3D (particle flow code in three
(2003) dimensions). User’s Guide. (2008)
23. Hicher, P.Y., Chang, C.S.: A microstructural elastoplastic model for 45. Krahn, J., Fredlund, D.G., Klassen, M.J.: Effect of soil suction
unsaturated granular materials. Int. J. Solids Struct. 44, 2304–2323 on slope stability at Notch Hill. Can. Geotech. J. 26(2), 269–278
(2007) (1989)
24. Cundall, P.A., Strack, O.D.L.: A discrete numerical model for gran- 46. Rassam, D.W., Williams, D.J.: A relationship describing the shear
ular assemblies. Géotechnique 29(1), 47–65 (1979) strength of unsaturated soils. Can. Geotech. J. 36(2), 363–368
25. Oda, M., Iwashita, K.: Mechanics of Granular Materials: An Intro- (1999)
duction, pp. 27–35. A. A. Balkema Publishers, Rotterdam (1999) 47. Schnellmann, R., Rahardjo, H., Schneider, H.R.: Unsaturated shear
26. Thornton, C.: Numerical simulations of deviatoric shear deforma- strength of a silty sand. Eng. Geol. 162, 88–96 (2013)
tion of granular media. Géotechnique 50(1), 43–53 (2000) 48. Escario, V., Saez, J.: The shear-strength of partly saturated soils.
27. Jiang, M.J., Konrad, J.M., Leroueil, S.: An efficient technique for Géotechnique 36(3), 453–456 (1986)
generating homogeneous specimens for DEM studies. Comput.
Geotech. 30(7), 579–597 (2003)
28. Jiang, M.J., Shen, Z.F., Zhu, F.Y.: Numerical analyses of braced
excavation in granular grounds: continuum and discrete element
approaches. Granul. Matter. 15(2), 195–208 (2013)

123
Reproduced with permission of the copyright owner. Further reproduction prohibited without
permission.

You might also like