Novel Approach for Generating Homogeneous Samples

for Discrete-Element-Method Studies

Tiantian Hu1; Yao Tang2; Daosheng Ling3; Xiukai Wang4; and Bo Huang5
Downloaded from ascelibrary.org by "Indian Institute Of Technology, Madras" on 07/31/24. Copyright ASCE. For personal use only; all rights reserved.

Abstract: A novel approach––the modified multilayer method––is proposed for generating homogeneous specimens for discrete-element-
method (DEM) studies. The “particle boundary” is introduced at the interface between the layers, and the general servocontrolled technique is
extended to the particle boundary to apply the stress field. Each layer of the specimen is compacted by moving the top and bottom boundaries
simultaneously to generate the desired DEM specimen. Unlike existing methods, the proposed approach generates highly homogeneous
specimens. The mechanism underpinning the new approach was determined by investigating the effect of the boundary movement on the
final spatial distribution of the particles. A comparison with experimental direct-shear results shows that homogeneous samples generated
using the proposed approach can accurately model the deformation and strength of granular material, whereas other methods lead to
deviations due to sample inhomogeneity. DOI: 10.1061/(ASCE)GM.1943-5622.0002273. © 2021 American Society of Civil Engineers.
Author keywords: Discrete-element method; Sample preparation; Particle boundary; Servocontrolled technique.

Introduction method, proposed by Cundall (Cundall and Strack 1979), was one
of the earliest. It is applied by moving the boundaries inward to
The discrete-element method (DEM) is a numerical modeling ap- condense a “cloud” of noncontacting particles in the desired density
proach for simulating the behavior of soils and other granular ma- or stress state. Some researchers (Masson and Martinez 2001;
terials based on Newton’s second law and the contact model Thornton 2000) have modified the IC in order to generate loose
between particles (Cundall and Strack 1979). Unlike the finite- specimens by setting a larger interparticle coefficient. However,
element method (FEM), which is based on continuum mechanics, large pores appear in the center of the prepared samples due to the
the DEM explicitly considers individual particles and can obtain arch effect (Chen et al. 2008; Dai et al. 2016; Rui et al. 2016). As
microscopic information on the particle rearrangement, contact for the radius-expansion (RE) method (Belheine et al. 2009; Gu
force, and structure (Furukawa et al. 2017; Guo and Zhao 2013; et al. 2014; Rothenburg and Bathurst 1992), particles of the reduced
Jiang et al. 2011), thus revealing the underlying mechanism spe- size are generated in a domain that is very close to the actual desired
cific to granular materials, such as the shear band, erosion, and liq- specimen size. The particles are gradually restored to the anticipated
uefaction (Jiang et al. 2016; Liu et al. 2020; Mollon et al. 2020; Xu radii, while the contact forces formed drive the particles to become
et al. 2015; Zeng et al. 2018; Zhao and Guo 2015). rearranged. Because the interparticle friction coefficient is low dur-
The first and most essential step in a DEM simulation involves ing particle placement, the RE method also produces dense speci-
creating the initial numerical setup and, in many cases, achieving a mens (Jiang et al. 2003). The gravitational deposit method (Cheng
specified stress level via a preliminary simulation. Many different et al. 2000; Dai 2010; Dai et al. 2015; Siiriä and Yliruusi 2007) rep-
methods have been developed and adopted for generating numerical licates how the soil comes into being in nature by generating parti-
specimens, and each method has its applications (Dai 2010; Lozano cles at some height above the final analysis domain and then
et al. 2016; OuYang et al. 2017). The isotropic-compression (IC) allowing them to fall downwards under a vertical body force. The
prepared numerical specimen has a stress field with a gradient, and
1
Ph.D. Student, MOE Key Laboratory of Soft Soils and Geoenviron- the top part is looser than the bottom (Dai et al. 2016; Fu and
mental Engineering, Center for Hypergravity Experimental and Interdisci- Dafalias 2011; Masson and Martinez 2001). Jiang et al. (2003) intro-
plinary Research, Zhejiang Univ., Hangzhou, China 310058. duced an undercompaction method (UCM) criterion to the traditional
2
Associate Professor, MOE Key Laboratory of Soft Soils and Geoenvir- multilayer (ML) method, as compaction potentially induces energy
onmental Engineering, Center for Hypergravity Experimental and Interdisci- transfer from the upper layers to the underlying layers (Ladd
plinary Research, Zhejiang Univ., Hangzhou, China 310058 (corresponding
author). Email: [email protected]
1978). However, the UCM cannot guarantee homogeneity in an
3
Professor, MOE Key Laboratory of Soft Soils and Geoenvironmental individual layer (Duan and Cheng 2016).
Engineering, Center for Hypergravity Experimental and Interdisciplinary The abovementioned methods include periods where particles
Research, Zhejiang Univ., Hangzhou, China 310058. move dynamically, and so they need massive calculation cycles.
4
Ph.D. Student, MOE Key Laboratory of Soft Soils and Geoenviron- A series of constructive methods have also been proposed, such
mental Engineering, Center for Hypergravity Experimental and Interdisci- as the advancing-front approach and the triangulation approach
plinary Research, Zhejiang Univ., Hangzhou, China 310058. (Bagi 2005; Cui and O’Sullivan 2003; Feng et al. 2003; Kuhn
5
Professor, MOE Key Laboratory of Soft Soils and Geoenvironmental and Bagi 2009; Lozano et al. 2016; OuYang et al. 2017; Valera
Engineering, Center for Hypergravity Experimental and Interdisciplinary
et al. 2015). While these model specimens are created without
Research, Zhejiang Univ., Hangzhou, China 310058.
Note. This manuscript was submitted on May 30, 2021; approved on
the need for DEM calculation cycles, they do require well designed
October 12, 2021; published online on December 9, 2021. Discussion pe- algorithms. In the fixed-point method (Li et al. 2017; Nadimi and
riod open until May 9, 2022; separate discussions must be submitted for in- Fonseca 2018; O’Sullivan et al. 2004), the position and radius of
dividual papers. This paper is part of the International Journal of each particle is determined by the regular packing or is replicated
Geomechanics, © ASCE, ISSN 1532-3641. from visual observations of laboratory tests. This method can be

© ASCE 04021278-1 Int. J. Geomech.

Int. J. Geomech., 2022, 22(2): 04021278

 used to verify analytical expressions or to compare simulated DEM
results with experimental observations. To improve the homogene-
ity of DEM packing, some researchers have proposed a hierarchi-
cal, multiscale modeling approach by applying deformation of
the packing using periodic boundaries via a gradient from a homog-
enous velocity field (Wu et al. 2019a, 2020).
The mechanical response of a granular assembly is significantly
affected by the sample’s homogeneity (Arthur and Phillips 1975;
Chakrabortty and Popescu 2012; Hu et al. 2019; Lu and Peng
2015; Vaid et al. 1999). Thus, in laboratory tests, especially ele-
ment tests, experimentalists have developed and adopted several
Downloaded from ascelibrary.org by "Indian Institute Of Technology, Madras" on 07/31/24. Copyright ASCE. For personal use only; all rights reserved.

methods for preparing homogeneous samples (Al-Shibli et al. (a)

1996; Bendahmane et al. 2008; Li et al. 2020; Tabaroei et al.
2017). As for numerical simulations, existing methods, such as
the IC method and RE method, can effectively generate uniform-
density specimens. However, when modified to generate loose
samples, the homogeneity may not be satisfactory (Jiang et al.
2003; Lai et al. 2014). Moreover, it is known that the behavior of
a sample is closely related to the stress level. However, the applied
stress fields of the samples generated by these methods are not suf-
ficiently homogeneous because only rigid-wall boundaries are used
to impose the confining stress when a servocontrolled mechanism
is used (Li et al. 2013; Thornton and Zhang 2001).
Thus, the generation of homogeneous DEM specimens needs to (b)
be carefully considered in order to be able to improve the accuracy
Fig. 1. Schematic of the proposed method for generating a homoge-
of modeling granular materials. Here, we propose a new approach
neous specimen: (a) the target specimen; and (b) the two different
for generating homogeneous DEM specimens with uniform
boundaries used in the proposed method. Particles A, C, and D, and
void-ratio and stress distributions. In the proposed modified ML
Wall B are used to describe the extension of the servocontrolled tech-
method, the model specimen is prepared in layers, and the particles
nique from the rigid-wall boundary to the particle boundary. Particle E
at the interfaces between the layers are considered to be the new
is an additional particle, used if necessary.
boundaries. The general servocontrolled technique was extended
to the particle boundary in order to apply the stress field. To eval-
uate the homogeneity, DEM samples, generated by different meth- Yang 2017; Jiang et al. 2018a; Kuhn et al. 2014). The application
ods, were compared. Finally, simulations of direct-shear tests were of confining pressure in the DEM is considered to have been com-
conducted on the samples generated using different methods so as pleted when the reaction force on the servocontrolled boundary
to investigate the effect of sample homogeneity on the simulation reaches the desired value, and the quasi-static state is also reached.
accuracy. However, the rigid-wall boundaries do not significantly affect the
interiors of large samples, and the boundary stress condition is
not equivalent to the stress state of every part of the specimen (Li
Proposed Approach to Generating a Homogeneous et al. 2013; Thornton and Zhang 2001). In loose samples generated
DEM Sample by the IC method, the stress in the center is much lower than that
adjacent to the boundary (Jiang et al. 2003).
A “particle boundary” is introduced to improve the limited influ-
Establishing the Proposed Method ence of the rigid walls. As shown in Fig. 1(b), the sample is split
Fig. 1(a) shows the final state of a target specimen with uniform into several layers. The particles near the interface between Layer
particle distribution, an average void ratio of
e = e0 , and a stress n and Layer n−1 are selected as the new additional boundary,
state of σ1 = σy, σ3 = σx. Gravity is not introduced in the generation which is controlled in order to compact Layer n to the target void
of a homogeneous representative elementary volume (REV) sam- ratio and stress state. All the layers are assembled so as to obtain
ple. This is different from natural soil, in which the stress state a homogeneous specimen. When samples are prepared in several
varies with depth due to gravity. However, the proposed MML layers in the geotechnical laboratory, each layer’s upper surface
method is applicable for producing the stress field in natural soils is scarified in order to increase the roughness before placement
if gravity is introduced. As a two-dimensional (2D) case is used of the subsequent layer (Thomson and Wong 2008). This ensures
as an example herein, the void ratio actually refers to the planar close contact between the layers. In the top right of Fig. 1(b), the
void ratio, and the same below. Due to boundary strain, particles particles are aligned with the wall boundary, resulting in spaces be-
overlap, creating forces at the contact points that produce a stress tween the layers after assembly. By contrast, in the proposed
field in the specimen. In a homogeneous specimen, the local void method, the rough surface of the particle boundary locks in the
ratio and stress state for any element, i, of the specimen should neighboring particles, ensuring a close layer-to-layer contact. It is
be close to the average value of the specimen, indicating that ei = necessary to ensure that the particles at the boundary are in contact
e0, σ i1 = σ y , and σ i3 = σ x . with each other or that the gap is small enough to prevent other
In most methods, the sample preparation is terminated when the particles from passing through. If the gap is too large, it is sug-
average void ratio of the sample achieves the target value. The cal- gested that an additional particle is added into the boundary, such
culation continues to ensure the quasistatic state until the ratio be- as Particle E in Fig. 1(b). This particle boundary can ensure the gen-
tween the unbalanced force and the average contact force decreases eration of Layers n or n−1 in Fig. 1, providing the contact force for
to less than 0.01%, as recommended by other researchers (Dai and the inner particles in the servocontrolled stage.

© ASCE 04021278-2 Int. J. Geomech.

Int. J. Geomech., 2022, 22(2): 04021278

 Implementation of the Proposed Method derived using the equation:
The implementation of the proposed approach for preparing a 2D

DEM sample is shown in Fig. 2(a). For the first layer of the speci- fni
σc = (1)
men in Fig. 2(b), the particles are randomly generated in the speci- Aw
fied domain, enclosed by the rigid walls. The top and bottom walls
move toward each other at a constant velocity until the target void
where fni = normal contact force at the ith wall–particle contact point;
ratio, et, is reached. Vertical pressure is applied by moving the top
and Aw = wall boundary area. In Fig. 3(a), as the wall boundary
and bottom walls using the servocontrolled mechanism (as de-
moves, the change in the overlap between the particle and the wall
scribed in the following section). In the prepared Layer 1, the par-
boundary is vΔt, and the normal contact force increases, vΔtkn. By
ticles adjacent to the top wall are defined as the “particle
updating the velocity of the wall boundary, v, during the iterations,
Downloaded from ascelibrary.org by "Indian Institute Of Technology, Madras" on 07/31/24. Copyright ASCE. For personal use only; all rights reserved.

boundary,” which serves as the basal boundary for the generation

it gradually converges on the target stress, σt, and v can be deter-
of Layer 2. As shown in Fig. 2(d), the particle boundary is raised mined using Eq. (2):
to form a new domain for generating the particles of Layer
2. The top wall and bottom particle boundary are moved to achieve
the target void ratio. Vertical stress consolidation is achieved by αAw
v= (σ c − σ t ) (2)
using the servocontrolled mechanism. The particles at the particle kn Nw Δt
boundary are fixed without rotation or horizontal motion so as to
maintain the initial configuration. The procedures are repeated where kn = normal contact stiffness; Nw = number of contacts be-
until the specimen is generated. As shown in Fig. 2(e), after the tween the wall boundary and particles; and Δt = timestep. A relax-
generation of Layer 2, all the particles in Layer 2 are shifted down- ation factor, α, is incorporated in order to improve the stability of
ward by δ so that the gap can be eliminated. In this process, the the servocontrolled procedure; its value is typically 0.5.
shape of the particle boundary also remains unchanged. Since the Unlike the wall boundary, the particle boundary consists of a series
particle boundary remains unchanged and Layer 1 is fixed by the of particles. There is an angle between the normal contact force and
generation of Layer 2, the contacts can be easily restored. Subse- the direction of boundary movement. Thus, while the particle boun-
quently, the layers are assembled, and the restriction on the particle dary moves vΔt in the vertical direction, the relative normal displace-
boundaries is released. Lateral confinement can also be applied by ment increment is not simply vΔt, as shown in the wall boundary in
dividing the specimen into several vertical strips and applying a Eq. (2). The servocontrolled algorithm of the particle boundary is de-
confining stress using the servocontrolled mechanism. rived as follows.
As shown in Fig. 3(b), Particle D is a particle in the particle
boundary, while Particle C is an interior particle. These two parti-
Servocontrolled Algorithm for the Particle Boundary
cles are in contact, with the contact location being marked in
As shown in Fig. 3(a), the top wall boundary moves downward to Fig. 1(b). Particle D moves from D to D′ during timestep Δt, and
compress Particle A [the positions of A and B are marked in the contact point changes from point E to E′ . The increment of
Fig. 1(b)]. The confining stress, σc, on the wall-boundary can be the contact force between D and C is Δ fc, which can be

(b) (c)

(a) (d) (e)

Fig. 2. Implementation of the MML method for DEM sample generation: (a) flowchart for the MML method; (b) and (c) generation of Layer 1; and
(d) and (e) generation of Layer 2.

© ASCE 04021278-3 Int. J. Geomech.

Int. J. Geomech., 2022, 22(2): 04021278

Downloaded from ascelibrary.org by "Indian Institute Of Technology, Madras" on 07/31/24. Copyright ASCE. For personal use only; all rights reserved.

(a) (b)

Fig. 3. Servocontrolled technique for different boundaries: (a) wall boundary; and (b) particle boundary.

decomposed into normal and tangential components, thusly: Table 1. Parameters for DEM sample preparation

Δf c = Δfn nc + Δfs t c (3) Parameter Value

Width of sample (m) 0.4
where nc and tc = unit normal and tangential vectors of the contact Target confining pressure, σt (kPa) 100
between Particles C and D, respectively. In the linear contact Number of particles 8,400
model, Δfn = knΔδn and Δfs = ksΔδs, where Δfn and Δfs = magnitudes Radius of particles (mm) 2–3
of the normal and tangential contact-force increments, respectively; Density of particles, ρ (kg/m3) 2,650
kn and ks = normal and tangential stiffness values; and Δδn and Interparticle friction coefficient, μg 0.5
Δδs = relative displacements in the normal and tangential direc- Normal stiffness, kn (N/m) 1 × 109
tions, respectively. In a small timestep, the displacement of the par- Tangential stiffness, ks (N/m) 1 × 109
ticle boundary, DD′ , is much smaller than the magnitude of the Local damping coefficient, αd 0.7
branch vector, DC. Therefore, we can obtain the relative displace-
ment in the normal direction based on the geometrical relationship: contact force is always consistent with the direction of the wall’s
Δδn = (|d + Δvey | − |d|) sin θ = Δtv sin θ (4) movement.

where ey = unit vector in the y-direction; and d = branch vector Evaluation of Specimen Homogeneity Using the
joining the particle centers. Since the overlap between particles is Proposed Approach
much smaller than the particle radii, the displacement in the tangen-
tial direction, Δδs, can be determined as follows: Numerical Setup and Model Parameters
rb rb The homogeneity of the samples generated by the MML method was
Δδs = (|d + Δvey | − |d|) cos θ = Δtv cos θ (5) evaluated by comparing it with samples produced by other methods,
rb + rw rb + rw
including the IC method, the RE method, the ML method and the
In timestep Δt, the increment of the vertical force between the parti- ML method with undercompaction (i.e., the UCM). Four DEM sam-
cle boundary and the interior particles can be determined by summation: ples with various densities (dense, medium dense, loose, and ex-
′
tremely loose) were generated. To eliminate errors, each sample

Nw
type was generated five times with different random numbers. The
ΔFcy = Δf ci · ey (6) input parameters of all the samples are listed in Table 1. The con-
i=1
trolled initial void ratios for the different samples, prior to consolida-
where N ′w = number of contacts between the particle boundary and tion, are listed in Table 2.
the interior particles. The particle boundary moves at a velocity that In the MML methods, the specimen is generated layer by layer.
depends on the difference between σc and σt. Therefore, the specimen homogeneity is affected by the layer number,
nl, or the ratio between the layer thickness and the mean particle ra-
ΔFcy dius, T /
r. If the layer number is small, the homogeneity is low. If
= α(σ c − σ t ) (7)
Aw the sample has too many layers, the efficiency is also low. Specimens
created using the DEM, with different numbers of layers, were gener-
Thus, the velocity of the particle boundary can be determined as
ated, and the homogeneity of the specimens was evaluated to deter-
follows:
mine the optimum nl.
αAw (σ c − σ t ) The DEM sample prepared with different values of nl was divided
v=   (8) into ng = 10 × 10 cells. The local void ratio of each cell, ei, was calcu-

rb
Δt kn sin θi +
2
ks cos θi
2
lated. The variance, Se, was used to evaluate the sample homogeneity:
rb + rw

 ng  i 2
This improved method can be applied through a general servo-  1  e
controlled technique, which means Eq. (8) can be reduced to Se = −1 (9)
ng − 1 i=1
e
Eq. (2) for the wall boundary if θi = 90° because the normal

© ASCE 04021278-4 Int. J. Geomech.

Int. J. Geomech., 2022, 22(2): 04021278

 Table 2. Intial void ratio of the samples
Soil sample Initial void ratio
Dense 0.195
Medium dense 0.225
Loose 0.250
Extremely loose 0.280
Downloaded from ascelibrary.org by "Indian Institute Of Technology, Madras" on 07/31/24. Copyright ASCE. For personal use only; all rights reserved.

(a)

Fig. 4. Variance of the void ratio of samples with different numbers of

layers obtained from the MML and IC methods. (b)

where
e = mean void ratio of the specimen. A smaller Se = better

homogeneity. As illustrated in Fig. 4, the specimen with one layer
had a lower Se with the MML method than the IC method, indicat-
ing that the MML method provides a significant improvement in
the sample uniformity. The Se of the MML method decreased
with an increase in the number of layers. However, when nl ≥ 4
(T /
r ≤ 22), the Se converged to a small value (of about 0.072),
and the degree of uniformity did not continue to increase. There-
fore, nl is 4 in the subsequent analysis, corresponding to T /
r = 22.

Comparison of Specimen Homogeneity of Different

Methods
Samples with densities ranging from dense to very loose were gen- (c)
erated by the three methods, followed by application of the stress
fields. In addition to Se, which characterizes the homogeneity of Fig. 5. Frequency distribution of the void ratio and stress of the loose
sample porosity, the stress variances Sx and Sy were proposed to specimens (μg = 0.5): (a) void ratio; (b) stress σx; and (c) stress σy.
evaluate the homogeneity of the stress levels of the samples:
1  where σ i = equivalent stress tensor of the ith cell; σ ix and σ iy = the
σi = fc ·d (10) respective x- and y-components of σ i; and σ
x and σ
y = mean values
Vgi
of σ ix and σ iy , respectively.
Fig. 5 shows the frequency distribution of the local void ratio and

 ng  i 2 stress values of the loose specimens generated by different methods.
 1  σx It was found that these variables had Gaussian distributions. The
Sx = −1 (11)
ng − 1 i=1 σ
x more uniform the sample, the greater the peak value of the curve.
Most of the e i values in the specimen from the MML method
clustered around
e, indicating high sample homogeneity. As for
 the other methods, the curves for the void ratio distributions were

 1  ng
σ iy
2
much flatter, indicating that the prepared specimens were less ho-
Sy = −1 (12) mogeneous. The Se of the MML method was only about 65%
ng − 1 i=1 σ
y
that from the IC method. In the ML method, the particles of the

© ASCE 04021278-5 Int. J. Geomech.

Int. J. Geomech., 2022, 22(2): 04021278

 underlying layer were temporarily fixed when preparing the subse- dense to very loose. Note that the IC specimens exhibit a slightly
quent layer, in case excessive compaction energy transferred from lower
e than those generated by the other approaches. The likely rea-
the upper layer. This step is easy to realize in the DEM and has been son is that when the confining pressure was applied, the soil structure
widely adopted by other researchers (Dai et al. 2016; Lai et al. of the IC specimens collapsed, reducing the void ratio. As the histo-
2014; Ding et al. 2019). Its effect on the void ratio distribution is gram in Fig. 6(a) indicates, for different sample densities, the MML
very close to that using the UCM (Se = 0.112 and 0.125, respec- specimens constantly have the smallest Se, which equates to the high-
tively. However, the homogeneity of an individual layer cannot est sample homogeneity. As for the IC and RE method, they are ef-
be ensured (Duan and Cheng 2016). For the stress distribution fective in generating homogenous dense samples but not
[Figs. 5(b and c)], the curve of the MML samples had the highest homogenous looser samples. The inhomogeneity of the looser IC
peak, indicating the best homogeneity, whereas the IC specimens samples results from the arch effect, which is discussed further in
exhibited the worst homogeneity, represented by a flatter curve the following section. When generating looser samples by the RE
Downloaded from ascelibrary.org by "Indian Institute Of Technology, Madras" on 07/31/24. Copyright ASCE. For personal use only; all rights reserved.

and the largest Sx and Sy values. methods, a larger μ is used, indicating a higher interparticle friction
In Fig. 6(a), the dotted lines represent the mean values of the void resistance. After the initial particles are generated randomly, particles
ratios of the specimens with different densities, obtained from differ- will gradually resort to their usual size. During the growth process,
ent methods, and the histogram represents the corresponding varian- particle rearrangement may be inhibited by friction resistance, to
ces, Se, determined using Eq. (9). As Fig. 6(a) shows, the methods some extent. The sample homogeneity from using the RE methods
can be used to generate samples with different densities, from can probably be improved by an algorithm that distributes the initial
particles in a more uniform way; further study of this is needed.
The stress results are given in Figs. 6(b and c). As the dashed lines
indicate, regardless of method or density, the values of σ
x and σ
y are
consistent with the boundary stresses (100 kPa). As the histograms
show, when using different methods to prepare dense samples, Sx
and Sy are very low, indicating that the contact forces are uniformly
spread in the dense samples. However, in generating looser samples,
Sx and Sy increase, with the uniformity decreasing significantly, espe-
cially when using the IC method. By contrast, the Sx and Sy from the
MML specimens are constantly very low, even for the loose samples.
The Sx of the loose sample generated by the MML method is 0.12,
which is about 40% and 50% that of the IC and ML methods,
respectively.
Thus, it can be concluded that the MML method provides more
(a) homogeneous samples for DEM studies than the other methods, es-
pecially where loose samples are required.
In the MML method, the stress state and void ratio of each layer
can be controlled individually. The homogeneous samples can be
further prepared to produce a specimen with the desired stress field
(e.g., a gravitational gradient similar to that operating in natural
soil covered by an overlying load). The servocontrolled techniques
employed in the MML method can ensure a monotonous current
boundary stress approaching the target stress. The stress state in
the sample generation can be controlled under the desired stress
path. The stress history during the generation procedure is essential
in affecting soil behavior, and is well worth an in-depth study.

(b) Homogeneity Mechanism of Specimens Generated Using

the MML method
The effect on sample homogeneity of the boundary moving during
sample generation is discussed below. In implementing the IC
method, four rigid-wall boundaries are controlled, causing a signif-
icant arching effect during sample generation. In the ML method,
the upper wall boundary is controlled to generate the sample
layer by layer, while the UCM follows the same generation techni-
que but uses a different targeted void ratio for each layer. There-
fore, in terms of the boundary effect during sample generation,
the UCM is similar to the ML method. In the RE method, there
is no wall movement during the sample generation stage. Thus,
this method is not discussed further in this section. In the MML
(c) method, a particle boundary is introduced to generate the sample
layer by layer, and the upper and lower boundaries can move simul-
Fig. 6. Variances in the void ratio and stress for different initial densi-
taneously during the generation stage. Therefore, the IC and ML
ties (histograms = variances; and curves = mean values): (a) void ratio;
methods were selected for comparison with the proposed MML
(b) stress σx; and (c) stress σy.
method.

© ASCE 04021278-6 Int. J. Geomech.

Int. J. Geomech., 2022, 22(2): 04021278

Downloaded from ascelibrary.org by "Indian Institute Of Technology, Madras" on 07/31/24. Copyright ASCE. For personal use only; all rights reserved.

(a) (b) (c)

(d) (e) (f )

Fig. 7. Distributions of the contact forces of the DEM specimens using different generation approaches: (a–c) IC method; and (d–f) MML method.

Inhomogeneity and Anisotropy in the Specimens Bathurst 1989) (the dash circle in Fig. 7):
As shown in Fig. 7(a), the center of the IC sample was looser than 1
E(θ) = [1 + ac cos 2(θ − θc )] (13)
the area near the boundary (the local void ratios being 0.249 and 2π
0.240, respectively). These findings are consistent with those of
where ac = degree of anisotropy; the strong force chains are orien-
Jiang et al. (2003), with large pores being present in the center
tated in the principal direction; and θc, which is shown as the dashed
of the loose sample generated by the IC method. This inhomoge-
lines in Fig. 7. Although the entire IC specimen exhibits isotropy
neity can be attributed to the formation of an arch structure during with a low overall value (ac = 0.01), as shown in Fig. 7(b), some
specimen generation. In the IC method, the specimen was gener- areas exhibit a highly preferable orientation of the contact forces.
ated by moving the boundaries inward until the target void ratio For example, in Cell A [Fig. 7(c)], the local ac = 0.20, and most of
was reached. During this process, particles tend to accumulate the contact forces are oriented in the vertical direction, with θc = 88°.
at the corners of the walls. Therefore, the particles in the corners It has been found that arch formation is related to the geometry
rapidly form a strong and continuous force chain, whereas the in- of the boundary during specimen generation (Li et al. 2013). By
terior particles remain loosely distributed and may even float, as contrast, the force chain distribution and rose diagrams of the speci-
indicated in Fig. 7(a), suggesting few contact forces in operation men generated by the MML method show no significant arches at
at the center of the sample. The reason for this is that the dense the corners of the wall boundaries. As shown in Fig. 7(d), the void
structure formed by the force chain can support most of the exter- ratios of the overall specimen and Cell B are almost identical. Also,
nal load, making it difficult to transmit the contact forces to the the anisotropy of the overall specimen and cells is not significant, as
interior particles (Rui et al. 2016). This finding suggests that indicated in Figs. 7(e and f).
sometimes the boundary response may not accurately reflect the
stress state in the center of the specimen (Thornton and Zhang
2001), and it is not feasible to apply a homogeneous stress field Particle Behavior at the Layer Interface of the Specimens
to the whole specimen if only the outside wall boundaries are The specimen generated by the ML method is relatively inhomoge-
being used. neous in the vertical direction, especially at the interface between
In addition to this inhomogeneity, specimens generated by the the layers. The areas between the layers are always looser than
IC method also exhibit high anisotropy in localized areas. The the overall specimen and cannot be improved, even by introducing
polar distribution of the normal contact forces in different direc- undercompression criteria (Jiang et al. 2003). The reason for this
tions is illustrated by the rose diagrams in Fig. 7. The shape of inhomogeneity remains to be determined.
the rose diagram was fitted by a second-order Fourier function in When the tangential contact force is fs < μg|fn|, the contacting
order to evaluate the anisotropy quantitatively (Rothenburg and particles stay in a state of static friction with no energy dissipation.

© ASCE 04021278-7 Int. J. Geomech.

Int. J. Geomech., 2022, 22(2): 04021278

 When fs increases and exceeds μg|fn|, sliding friction occurs and fs pushed by the top wall to compress the middle sublayer. Therefore,
adjusts to μg|fn|. Unlike static friction, when sliding friction occurs, the Sc of the top sublayer decreases, and the Sc of the middle sub-
the particles are rearranged, and the energy is dissipated through the layer increases to the peak value, which is smaller than that of the
friction (Wu et al. 2019b). Therefore, only the sliding contacts are top sublayer because the skeleton of the top sublayer acts as a
discussed below. In an initially loose specimen, sliding causes pore buffer between the top wall and the middle sublayers. The bottom
shrinkage, resulting in the particles becoming rearranged into a sublayer is the farthest from the top wall; thus, its peak slide frac-
denser state (Kong and Lannutti 2000). Below, we explain the in- tion occurs last (during t3) and is the lowest. In Stage 2, the wall
homogeneity of the ML samples by investigating the influence of stops moving, and the residual kinetic energy is dissipated by slid-
wall movement on particle sliding. ing friction. The top sublayer exhibits the highest slide fraction due
As shown in the insert in Figs. 8(a and b), each layer of the speci- to the maximum accumulated energy. The consolidation of Stage 3
men, consisting of 2,100 particles, is distributed randomly and divided is implemented by the servocontrolled top wall. The assembly of
Downloaded from ascelibrary.org by "Indian Institute Of Technology, Madras" on 07/31/24. Copyright ASCE. For personal use only; all rights reserved.

into three sublayers, based on the distance from the top wall. In the the top layer is closest to, and most affected by, the top wall, lead-
ML method, the top wall moves down at 0.01 m/s to simulate the ing to the highest slide fraction among all the layers and indicating
boundary movement. In the MML method, the top and bottom that the efficiency of the servocontrolled wall might be limited as
walls move simultaneously. When the target void ratio is reached, the sample size increases. Therefore, it is necessary to use the inte-
the wall is fixed temporarily, but the iteration continues in order to rior particles as a supplementary boundary in order to apply a uni-
achieve a quasistatic equilibrium. Ultimately, a vertical stress of form stress field to the entire sample.
100 kPa is applied, according to Eq. (2). During this process, particle In each single layer, the effect on particle sliding of the distance
sliding occurs. The evolution of the slide fraction, Sc, is plotted in to the moving top wall results in higher density at the top and lower
Fig. 8; it is defined as the ratio of the slipping contact number, ns, density at the bottom. Therefore, the entire specimen obtained by
to the particle number, np, in the sublayer. the ML method is inhomogeneous in the vertical direction, and
ns the interfaces between the layers are looser than the rest of the
Sc = (14) specimen.
np
In the MML method, the top and bottom wall boundaries can
As shown in Fig. 8(a), in Stage 1, during t1, as the top wall move simultaneously. The evolution of the slide fraction is
moves downward, the noncontacting particles in the top sublayer shown in Fig. 8(b). During t1, driven by the top and bottom
are the first to move and collide; thus, the slip fraction increases. walls, the Sc of the top and bottom sublayers increase to a peak,
During t2, the basic skeleton is formed in the top sublayer and is the value of which is close to that of the top sublayer, shown in
Fig. 8(a). The slide fraction of the middle sublayer rapidly rises
to a lower peak. Although the evolution of the slide fraction in
the middle sublayer lags slightly behind the others, this is compen-
sated for by the relatively higher postpeak slide fraction during t3,
as the middle assembly is compressed by the top and bottom sub-
layers. In the subsequent stages, the slide fraction of the three sub-
layers develops in a similar locus. Therefore, it can be assumed that
the simultaneous movement of the top and bottom walls in the
MML method improves the uniformity of the specimen in the ver-
tical direction. Furthermore, since both the top and bottom walls are
used, fewer cycles are needed than shown in Fig. 8(b) than shown
in Fig. 8(a), indicating that the MML method is less time-
consuming than the other methods.

(a) Application of the Proposed Method in Modeling

Direct-Shear Tests

In the element test and the simulations, the modeled element was a
REV, which needed to be homogeneous. Various approaches have
been developed to prepare homogeneous samples through labora-
tory testing (Bendahmane et al. 2008; Li et al. 2020). To investigate
the effect of sample homogeneity on the mechanical behavior of
granular materials, DEM specimens generated using different
methods were used to simulate the direct-shear test conducted by
Liu (1999). The new method proposed herein can also be used to
generate 2D specimens for biaxial compression or 3D specimens
for triaxial compression. Under these two loading conditions, slip
occurs in the assembly (Rowe 1962), so the simulation of flexible
boundaries is necessary in order to produce more realistic behaviors
(Cheung and O’Sullivan 2008). By contrast, the direct-shear test
(b) only involves rigid boundaries, so it is here considered to avoid ex-
cessive complications. The experimental sample consisted of 3,259
Fig. 8. Evolution of the slide fraction induced by moving the wall
aluminum rods with two different diameters––5 and 9 mm––and a
boundaries of the DEM samples using different generation methods:
mixing ratio of 3:2 by weight. The width of the box was 400 mm.
(a) ML method; and (b) MML method.
To change the void ratio, the height of the sample could be adjusted

© ASCE 04021278-8 Int. J. Geomech.

Int. J. Geomech., 2022, 22(2): 04021278

Downloaded from ascelibrary.org by "Indian Institute Of Technology, Madras" on 07/31/24. Copyright ASCE. For personal use only; all rights reserved.

Fig. 9. Numerical setup of the direct-shear test.

(a)
Table 3. Input parameters for the direct-shear simulation
Parameter Value
Normal stiffness of walls, k ′n (N/m2) 9 × 109
Tangential stiffness of walls, k ′s (N/m2) 3 × 108
Normal stiffness of particles, kn (N/m2) 1 × 109
Tangential stiffness of particles, ks(N/m2) 1 × 108
Interparticle friction angle, ϕμ (°) 16
Density of particles, ρ (kg/m3) 2,700
Local damping coefficient, αd 0.7

using the top plate (266 mm for e0 = 0.201). According to images

of the setup, as provided by Liu (1999), the distribution of the
void ratio could be calculated and the sample was relatively uni-
form. The shearing velocity was 1 mm/s, while the normal stress
was 49 kPa. The test ended when the relative displacement between
the upper and lower boxes exceeded 8 mm. (b)
The 3D test was carried out using long aluminum rods with a
length of 50 mm, which is much greater than the dimensions of Fig. 10. Stress ratio and normal displacement in the direct-shear tests:
the rod cross section (5–9 mm). Thus, a direct-shear test using (a) comparison with the experimental result (e0 = 0.201); and (b) sand
these rods could be simplified to a plane strain issue in the simula- with different initial void ratios (e0 = 0.195, 0.217, and 0.235).
tion (Li et al. 2017). The DEM program PFC 2D by Itasca was used
to conduct the numerical simulation, and the disk element was used
to model the assembly of the rods, as shown in Fig. 9. The model
normal displacement between the top and bottom boundaries. As
specimens with different initial void ratios were generated using the
shown in Fig. 10(a), the peak strength and volume change in the
IC and MML methods for comparison. The parameter selection
mainly refers to the experiment (Liu 1999; Liu et al. 2005), as listed MML sample agree well with the experimental results, indicating
in Table 3. In the experiment, the friction coefficient was directly that the DEM simulation using the MML specimen accurately
obtained from frictional tests on aluminum rods. The contact stiff- captured the dilative response of the dense specimen. By contrast,
ness and local damp coefficient were from the back-calculation. the IC specimen also dilated due to the high-density state;
Yamamoto (1995) and Liu et al. (2005) used the same parameters however, the strength and volume expansion were slightly
to simulate a biaxial compression test and a simple shear test on an underestimated because the localized shear band crossed over
assembly of aluminum rods, respectively. Using these calibrated the center, where the void ratio was higher than that of the entire
parameters, the simulated results––particularly the mechanical be- specimen. In the experiment reported by Liu (1999), the ratio of
havior––were in good agreement with the experiments. A measure- Ds to the sample width was only 2%, without reaching the critical
ment sphere with a diameter of 150 mm was placed at the center of state.
the specimen in order to monitor changes in the local void ratio dur- Another set of direct-shear testing was simulated on the samples
ing the shearing process (the dashed circle in Fig. 9), which was to (e0 = 0.195, 0.217, and 0.235, and the height was 264, 270, and
be compared with the overall void ratio. In this study, we followed 274 mm, respectively). The results are given in Fig. 10(b). The
the suggestions of other researchers, using a measurement circle final Ds was extended to 40 mm, and all the samples eventually
that was 20 times the mean particle size, which could capture rep- reached the stable state, where the stress ratio converged at 0.35,
resentative parameter measurements (Li et al. 2014; Jiang et al. regardless of density and method used, which is consistent with
2016; Lai et al. 2016). the findings of existing studies (Jiang et al. 2018b; Syed et al.
The numerical and experimental results from the dense samples 2017; Wang et al. 2007). At this stage, only the shear zone reached
with void ratios of 0.201 are given in Fig. 10(a). The shear strength the critical state, with a local void ratio of 0.220, while the whole
is represented by the ratio of the shear stress to the normal stress sample did not. However, in the early stages (Ds < 8 mm), the re-
applied to the shear plane. The width of the shear box was fixed sponses of the samples prepared by the two methods were quite dif-
at 0.4 m; thus, the volume change could be represented by the ferent. The stress ratios of the IC sample were lower than those of

© ASCE 04021278-9 Int. J. Geomech.

Int. J. Geomech., 2022, 22(2): 04021278

 (a) (b)
Downloaded from ascelibrary.org by "Indian Institute Of Technology, Madras" on 07/31/24. Copyright ASCE. For personal use only; all rights reserved.

Fig. 11. Force chain network during shearing of samples generated by different methods with a shear displacement of 3 mm (the thickness and gray
scale of the lines denote the magnitude of the normal contact force): (a) MML method; and (b) IC method.

(a) (b)

Fig. 12. Contours of the particle velocity vectors with a shear displacement of 3 mm (dashed arrows = internal movement of the particles): (a) MML
method; and (b) IC method.

the MML samples, leading to an underestimated peak shearing deformation outside the band. The width of the band was about 12
strength. Also, the void ratios of the IC samples became lower times the average particle size, which is consistent with the results
than those of the MML samples, even though their initial values presented in Jiang et al. (2010), Mühlhaus and Vardoulakis
were the same. For e0 = 0.217, in the early stages, the MML speci- (1987), and Rechenmacher et al. (2010). The arrows of the particles
mens dilated, whereas the IC specimens contracted, similarly to a near the top and bottom boundaries point slightly outward in Fig. 12,
looser specimen. It can be assumed that the IC samples were indicating the occurrence of dilation in the entire specimen.
more contractive due to the shear plane cutting across the loose cen- As shown in Fig. 12(b), the shear region in the IC specimen was
ter. The macroscopic differences can be explained from the micro- not as easy to recognize as that in the MML specimen. However, the
scopic perspective as follows. velocity was highly disordered in Regions A and B, which represent
Fig. 11 shows the contact-force network of the samples of the intersections of the predetermined shear plane and the arch struc-
e0 = 0.217, generated by the MML and IC methods when Ds = ture [Fig. 11(b)]. This result indicates that most of the shear forces
3 mm. In the MML specimen, the direction of the primary forces from the boundaries were transmitted to the arch structure. As the
rotated in a clockwise direction (opposite to the shear direction), shearing process developed, the arch structure gradually collapsed,
indicating that the normal contact-force anisotropy had increased and the particles moved inward into the loose center [see the dashed
significantly along the white diagonal line in Fig. 11(a); these arrows pointing inward in Fig. 12(b)]. In other words, the loose cen-
orientated contact forces have also been observed in granular ter absorbed the shear dilation from the dense area; thus, the arrows
shearing experiments (Allersma 2005; Gao et al. 2017; Jewell near the top and bottom boundaries point slightly inward. This find-
and Wroth 1987). By contrast, the forces of the IC specimen ing was confirmed by the change in local void ratio in the measure-
also rotated in a clockwise direction when subjected to shear ment sphere in the center. The central void ratio decreased from
loading; however, most of the contact forces were transmitted 0.224 to 0.220, whereas the overall average void ratio of the sample
through a path around the center [see the white oval in increased slightly from 0.217 to 0.218.
Fig. 11(b)]. In the center of the IC specimen, the initial local Although the overall void ratio of the two samples is similar, in-
void ratio was 0.224, which is too loose to form sufficient con- dicating that both are highly isotropic, the samples have different
tacts between particles to support force chains. Thus, the sur- mechanical properties due to their different homogeneities. The
rounding dense region bore most of the shear load. specimen obtained by the proposed new method had better uni-
Fig. 12 shows the distribution of the particle velocity when Ds = formity than the IC specimen, and the simulation showed a slight
3 mm. The arrows represent the velocity direction, and the grayscale dilation of the medium-dense specimen, where the shear band
indicates the magnitude. The particle velocity of the MML specimen was concentrated in the middle. In the IC specimen, the particles
was disordered in the middle, indicating localized shear strain in this moved from the dense regions near the boundaries to the loose cen-
narrow band (Iwashita and Oda 2000). Outside this region, the veloc- ter; thus, the medium-dense specimen behaved similarly to the
ity field was quite uniform, indicating relatively low shear loose specimen.

© ASCE 04021278-10 Int. J. Geomech.

Int. J. Geomech., 2022, 22(2): 04021278

 Conclusions ey = unit vector in the y-direction;
Fcy = vertical force on the boundary;
A new method for preparing a DEM sample with homogeneous fni = normal contact force at the ith boundary-particle contact
distributions of the void ratio and stress field was proposed, involv- point;
ing the introduction of “particle boundaries” at the interfaces be- fsi = tangential contact force at the ith boundary-particle
tween sample layers. The general servocontrolled algorithm of contact point;
the wall boundary was extended to that of the particle boundary. kn = normal stiffness;
The sample homogeneity was evaluated and verified using DEM ks = tangential stiffness;
simulations and comparison with other sample-generation ap- nc = unit normal vectors of the contact;
proaches. The following conclusions were reached: ng = number of grids;
(1) The sample homogeneity was significantly improved by the nl = layer number;
Downloaded from ascelibrary.org by "Indian Institute Of Technology, Madras" on 07/31/24. Copyright ASCE. For personal use only; all rights reserved.

proposed new approach. The variance in the void ratio and np = total particle number;
stress level of the sample was reduced by more than 50% com- ns = slipping contact number;
pared with the ML, indicating a high degree of sample homo-
r = average particle radius;
geneity. The layer thickness in the proposed approach is rb = radius of interior particle;
suggested to be about 22 times the mean particle size in rw = radius of boundary particle;
order to maintain high homogeneity. Sc = slide fraction;
(2) The evolution of the contact characteristics of the DEM sample Se = variance of the void ratio;
generated by the ML method, with only one moving boundary, Sx = variance of the horizontal stress;
indicated that particle sliding developed progressively from the Sy = variance of the vertical stress;
moving boundary to the fixed boundary. By contrast, using the T = layer thickness;
proposed method, the two boundaries move in opposite direc- v = velocity of the boundary;
tions, compressing the DEM particles, causing the simultane- α = relaxation factor;
ous development of particle sliding in the compression αd = local damping coefficient;
direction, and resulting in high sample homogeneity. δ = gap between layers;
(3) The DEM simulations of the direct-shear tests, using samples Δδn = relative displacements in the normal direction;
generated by different approaches, showed that inhomogeneity Δδs = relative displacements in the tangential direction;
led to deviations. In the IC method, the particles tended to ac- μg = interparticle friction coefficient;
cumulate in the corners, forming a loose core inside the sam- ϕμ = interparticle friction angle;
ple. Under direct-shear loading, the shear band passed ρ = particle density;
through this loose core, leading to an underestimation of the σ i1 = vertical stress of the ith element;
strength and stiffness of the granular samples. By contrast, σ i3 = horizontal stress of the ith element;
the homogenous samples generated using the proposed method σc = current confining pressure;
provided results that were in good agreement with the experi- σt = target confining pressure;
mental results. Δt = timestep increment;
tc = unit tangential vectors of the contact;
θ = angle between the normal contact and the x-direction; and
Data Availability Statement θc = principal direction of the strong contact forces.

All data, models, and code that support the findings of this study are References
available from the corresponding author upon reasonable request.
Allersma, H. 2005. “Optical analysis of stress and strain in shear zones.” In
Proc., Int. Conf. on Powders and Grains, 187–191. Delft, Netherlands:
Acknowledgments Delft University of Technology, Stuttgart.
Al-Shibli, K., E. Macari, and S. Sture. 1996. “Digital imaging techniques
This work was supported by the National Natural Science Founda- for assessment of homogeneity of granular materials.” Transp. Res.
tion of China (Grant Nos. 51988101, 51908493, and Rec. 1526: 121–128. https://doi.org/10.1177/0361198196152600115.
41961144018), the Natural Science Foundation of Zhejiang Prov- Arthur, J. R. F., and A. B. Phillips. 1975. “Homogeneous and layered sand
ince (Grant No. LCZ19E080002), and the Fundamental Research in triaxial compression.” Géotechnique 25 (4): 799–815. https://doi.org
/10.1680/geot.1975.25.4.799.
Funds for the Central Universities (Grant Nos. 2019FZA4016
Bagi, K. 2005. “An algorithm to generate random dense arrangements for
and 2019QNA4035). This support is gratefully acknowledged. discrete element simulations of granular assemblies.” Granular Matter
7 (1): 31–43. https://doi.org/10.1007/s10035-004-0187-5.
Belheine, N., J.-P. Plassiard, F.-V. Donzé, F. Darve, and A. Seridi. 2009.
Notation “Numerical simulation of drained triaxial test using 3D discrete element
modeling.” Comput. Geotech. 36 (1–2): 320–331. https://doi.org/10
The following symbols are used in this paper: .1016/j.compgeo.2008.02.003.
Aw = wall boundary area; Bendahmane, F., D. Marot, and A. Alexis. 2008. “Experimental parametric
ac = degree of anisotropy; study of suffusion and backward erosion.” J. Geotech. Geoenviron.
Eng. 134 (1): 57–67. https://doi.org/10.1061/(ASCE)1090-0241(2008)
Ds = shear displacement;
134:1(57).
d = branch vector joining the particle centers; Chakrabortty, P., and R. Popescu. 2012. “Numerical simulation of centri-

e = average void ratio; fuge tests on homogeneous and heterogeneous soil models.” Comput.
e0 = initial void ratio; Geotech. 41: 95–105. https://doi.org/10.1016/j.compgeo.2011.11.008.
ei = void ratio of the ith element; Chen, Y. M., W. P. Cao, and R. P. Chen. 2008. “An experimental investi-
et = target void ratio; gation of soil arching within basal reinforced and unreinforced piled

© ASCE 04021278-11 Int. J. Geomech.

Int. J. Geomech., 2022, 22(2): 04021278

 embankments.” Geotext. Geomembr. 26 (2): 164–174. https://doi.org Jiang, M. D., Z. X. Yang, D. Barreto, and Y. H. Xie. 2018a. “The
/10.1016/j.geotexmem.2007.05.004. influence of particle-size distribution on critical state behavior of spher-
Cheng, Y. F., S. J. Guo, and H. Y. Lai. 2000. “Dynamic simulation of ran- ical and non-spherical particle assemblies.” Granular Matter 20 (4): 80.
dom packing of spherical particles.” Powder Technol. 107 (1–2): 123– https://doi.org/10.1007/s10035-018-0850-x.
130. https://doi.org/10.1016/S0032-5910(99)00178-3. Jiang, M. J., J. M. Konrad, and S. Leroueil. 2003. “An efficient technique
Cheung, G., and C. O’Sullivan. 2008. “Effective simulation of flexible lateral for generating homogeneous specimens for DEM studies.” Comput.
boundaries in two- and three-dimensional DEM simulations.” Geotech. 30 (7): 579–597. https://doi.org/10.1016/S0266-352X(03)
Particuology 6 (6): 483–500. https://doi.org/10.1016/j.partic.2008.07.018. 00064-8.
Cui, L., and C. O’Sullivan. 2003. “Analysis of a triangulation based ap- Jiang, M. J., J. Liu, Z. F. Shen, and B. L. Xi. 2018b. “Exploring the critical
proach for specimen generation for discrete element simulations.” state properties and major principal stress rotation of sand in direct shear
Granular Matter 5 (3): 135–145. https://doi.org/10.1007/s10035-003 test using the distinct element method.” Granular Matter 20 (2): 25.
-0145-7. https://doi.org/10.1007/s10035-018-0796-z.
Downloaded from ascelibrary.org by "Indian Institute Of Technology, Madras" on 07/31/24. Copyright ASCE. For personal use only; all rights reserved.

Cundall, P. A., and O. D. Strack. 1979. “A discrete numerical model for Jiang, M. J., H. B. Yan, H. H. Zhu, and S. Utili. 2011. “Modeling shear be-
granular assemblies.” Géotechnique 29 (1): 47–65. https://doi.org/10 havior and strain localization in cemented sands by two-dimensional
.1680/geot.1979.29.1.47. distinct element method analyses.” Comput. Geotech. 38 (1): 14–29.
Dai, B. 2010. “Micromechanical investigation of the behavior of granular https://doi.org/10.1016/j.compgeo.2010.09.001.
materials.” Doctor Ph.D. thesis, Dept. of Civil Engineering, Hong Jiang, M. J., Z. Y. Yin, and Z. F. Shen. 2016. “Shear band formation in
Kong Univ. lunar regolith by discrete element analyses.” Granular Matter 18 (2):
Dai, B.-B., and J. Yang. 2017. “Shear strength of assemblies of frictionless 32. https://doi.org/10.1007/s10035-016-0635-z.
particles.” Int. J. Geomech. 17: 11. Kong, C. M., and J. J. Lannutti. 2000. “Effect of agglomerate size distribu-
Dai, B. B., J. Yang, and X. D. Luo. 2015. “A numerical analysis of the tion on loose packing fraction.” J. Am. Ceram. Soc. 83 (9): 2183–2188.
shear behavior of granular soil with fines.” Particuology 21: 160– https://doi.org/10.1111/j.1151-2916.2000.tb01533.x.
172. https://doi.org/10.1016/j.partic.2014.08.010. Kuhn, M. R., and K. Bagi. 2009. “Specimen size effect in discrete element
Dai, B. B., J. Yang, C. Y. Zhou, and X. D. Luo. 2016. “DEM investigation simulations of granular assemblies.” J. Eng. Mech. 135 (6): 485–492.
on the effect of sample preparation on the shear behavior of granular https://doi.org/10.1061/(ASCE)0733-9399(2009)135:6(485).
soil.” Particuology 25: 111–121. https://doi.org/10.1016/j.partic.2015 Kuhn, M. R., H. E. Renken, A. D. Mixsell, and S. L. Kramer. 2014.
.03.010. “Investigation of cyclic liquefaction with discrete element simulations.”
Ding, X. H., T. Ma, and X. M. Huang. 2019. “Discrete-element contour- J. Geotech. Geoenviron. Eng. 140 (12): 04014075. https://doi.org/10
filling modeling method for micromechanical and macromechanical .1061/(ASCE)GT.1943-5606.0001181.
analysis of aggregate skeleton of asphalt mixture.” J. Transp. Eng. Ladd, R. 1978. “Preparing test specimens using undercompaction.”
Part B. Pavements 145 (1): 04018056. https://doi.org/10.1061 Geotech. Test. J. 1 (1): 16–23. https://doi.org/10.1520/GTJ10364J.
/JPEODX.0000083. Lai, H. J., J. J. Zheng, J. Zhang, R. J. Zhang, and L. Cui. 2014. “DEM anal-
Duan, N., and Y. P. Cheng. 2016. “A modified method of generating speci- ysis of “soil”-arching within geogrid-reinforced and unreinforced pile-
mens for a 2D DEM centrifuge model.” In Geo-Chicago 2016: supported embankments.” Comput. Geotech. 61: 13–23. https://doi.org
Sustainable Materials and Resource Conservation, Geotechnical /10.1016/j.compgeo.2014.04.007.
Special Publication 272, edited by K. R. Reddy, N. Yesiller, D. Lai, H., J. Zheng, R. Zhang, and M. Cui. 2016. “Visualization of the forma-
Zekkos, A. Farid, and A. De, 610–620. Reston, VA: ASCE. tion and features of soil arching within a piled embankment by discrete
Feng, Y. T., K. Han, and D. Owen. 2003. “Filling domains with disks: an element method simulation.” J. Zhejiang University-SCIENCE A 17
advancing front approach.” Int. J. Numer. Methods Eng. 56 (5): 699– (10): 803–817. http://doi.org/10.1631/jzus.A1500302.
713. https://doi.org/10.1002/nme.583. Li, B., L. Guo, and F. Zhang. 2014. “Macro-micro investigation of granular
Fu, P. C., and Y. F. Dafalias. 2011. “Study of anisotropic shear strength of materials in torsional shear test.” J. Cent. South Univ. 21 (7): 2950–
granular materials using DEM simulation.” Int. J. Numer. Anal. 2961. http://doi.org/10.1007/s11771-014-2262-3.
Methods Geomech. 35 (10): 1098–1126. https://doi.org/10.1002/nag.945. Li, S., A. R. Russell, and D. M. Wood. 2020. “Influence of particle-size dis-
Furukawa, R., K. Kadota, T. Noguchi, A. Shimosaka, and Y. Shirakawa. tribution homogeneity on shearing of soils subjected to internal ero-
2017. “DEM modelling of granule rearrangement and fracture behav- sion.” Can. Geotech. J. 57 (11): 1684–1694. https://doi.org/10.1139
iours during a closed-die compaction.” AAPS PharmSciTech 18 (6): /cgj-2019-0273.
2368–2377. https://doi.org/10.1208/s12249-017-0719-z. Li, X., H. S. Yu, and X. S. Li. 2013. “A virtual experiment technique on the
Gao, Z. G., Z. C. Li, and M. Alam. 2017. “Study on the microscopic struc- elementary behaviour of granular materials with discrete element
ture and the bearing capacity of the elliptical granular system in the ran- method.” Int. J. Numer. Anal. Methods Geomech. 37 (1): 75–96.
dom particle size.” In Proc., Int. Conf. on Discrete Element Methods, https://doi.org/10.1002/nag.1086.
edited by X. Li, Y. Feng, and G. Mustoe, 161–168. Singapore: Li, Z. F., Y. H. Wang, X. Li, and Q. Yuan. 2017. “Validation of discrete
Springer. element method by simulating a 2D assembly of randomly packed ellip-
Gu, X. Q., M. S. Huang, and J. G. Qian. 2014. “DEM investigation on the tical rods.” Acta Geotech. 12 (3): 541–557. https://doi.org/10.1007
evolution of microstructure in granular soils under shearing.” Granular /s11440-017-0542-4.
Matter 16 (1): 91–106. https://doi.org/10.1007/s10035-013-0467-z. Liu, J. Y., A. Wautier, S. Bonelli, F. Nicot, and F. Darve. 2020.
Guo, N., and J. D. Zhao. 2013. “The signature of shear-induced anisotropy “Macroscopic softening in granular materials from a mesoscale per-
in granular media.” Comput. Geotech. 47: 1–15. https://doi.org/10.1016 spective.” Int. J. Solids Struct. 193–194: 222–238. https://doi.org/10
/j.compgeo.2012.07.002. .1016/j.ijsolstr.2020.02.022.
Hu, Z., D. Wang, X. M. Tong, L. H. Li, and R. P. Behringer. 2019. Liu, S. H. 1999. “Development of a new direct shear test and its application
“Granular scale responses in the shear band region.” Granular Matter to the problems of slope stability and bearing capacity.” Doctor Ph.D.
21 (4): 1–6. https://doi.org/10.1007/s10035-019-0958-7. thesis, Dept. of Civil Engineering, Nagoya Institute of Technology.
Iwashita, K., and M. Oda. 2000. “Micro-deformation mechanism of shear Liu, S. H., D. Sun, and H. Matsuoka. 2005. “On the interface friction in di-
banding process based on modified distinct element method.” Powder rect shear test.” Comput. Geotech. 32 (5): 317–325. https://doi.org/10
Technol. 109 (1–3): 192–205. https://doi.org/10.1016/S0032-5910(99) .1016/j.compgeo.2005.05.002.
00236-3. Lozano, E., D. Roehl, W. Celes, and M. Gattass. 2016. “An efficient algorithm
Jewell, R., and C. Wroth. 1987. “Direct shear tests on reinforced sand.” to generate random sphere packs in arbitrary domains.” Comput. Math.
Géotechnique 37 (1): 53–68. https://doi.org/10.1680/geot.1987.37.1.53. Appl. 71 (8): 1586–1601. https://doi.org/10.1016/j.camwa.2016.02.032.
Jiang, M., F.-Z. Wang, and H. Zhu. 2010. “Shear band formation in ideal Lu, L. Q., and W. J. Peng. 2015. “Test research on inhomogenetiy of triax-
dense sand in direct shear test by discrete element analysis.” [In ial test samples prepared by on-way hierarchical compaction method.”
Chinese.] Rock Soil Mech. 31 (1): 253–257. [In Chinese.] Technol. Highway Transp. 2: 5–8.

© ASCE 04021278-12 Int. J. Geomech.

Int. J. Geomech., 2022, 22(2): 04021278

 Masson, S., and J. Martinez. 2001. “Micromechanical analysis of the shear Thomson, P. R., and R. C. K. Wong. 2008. “Specimen nonuniformities in
behavior of a granular material.” J. Eng. Mech. 127 (10): 1007–1016. water-pluviated and moist-tamped sands under undrained triaxial com-
https://doi.org/10.1061/(ASCE)0733-9399(2001)127:10(1007). pression and extension.” Can. Geotech. J. 45 (7): 939–956. https://doi
Mollon, G., A. Quacquarelli, E. Andò, and G. Viggiani. 2020. “Can friction .org/10.1139/T08-023.
replace roughness in the numerical simulation of granular materials?” Thornton, C. 2000. “Numerical simulations of deviatoric shear deformation
Granular Matter 22 (2): 42. https://doi.org/10.1007/s10035-020-1004-5. of granular media.” Géotechnique 50 (1): 43–53. https://doi.org/10
Mühlhaus, H.-B., and I. Vardoulakis. 1987. “The thickness of shear bands .1680/geot.2000.50.1.43.
in granular materials.” Géotechnique 37 (3): 271–283. https://doi.org/10 Thornton, C., and L. Zhang. 2001. “A DEM comparison of different shear
.1680/geot.1987.37.3.271. testing devices.” In Powders and grains, edited by Y. Kishino, 183–
Nadimi, S., and J. Fonseca. 2018. “A micro finite-element model for soil 190. London: CRC Press.
behaviour: numerical validation.” Géotechnique 68 (4): 364–369. Vaid, Y. P., S. Sivathayalan, and D. Stedman. 1999. “Influence of
https://doi.org/10.1680/jgeot.16.P.163. specimen-reconstituting method on the undrained response of sand.”
Downloaded from ascelibrary.org by "Indian Institute Of Technology, Madras" on 07/31/24. Copyright ASCE. For personal use only; all rights reserved.

O’Sullivan, C., J. D. Bray, and M. Riemer. 2004. “Examination of the re- Geotech. Test. J. 22 (3): 187–195. https://doi.org/10.1520/GTJ11110J.
sponse of regularly packed specimens of spherical particles using phys- Valera, R. R., I. P. Morales, S. Vanmaercke, C. R. Morfa, L. A. Cortés, and
ical tests and discrete element simulations.” J. Eng. Mech. 130 (10): H. D.-G. Casañas. 2015. “Modified algorithm for generating high vol-
1140–1150. https://doi.org/10.1061/(ASCE)0733-9399(2004)130: ume fraction sphere packings.” Comput. Part. Mech. 2 (2): 161–172.
10(1140). https://doi.org/10.1007/s40571-015-0045-8.
OuYang, Y. P., Q. Yang, and L. Yu. 2017. “An efficient dense and stable Wang, J., J. E. Dove, and M. S. Gutierrez. 2007. “Discrete-continuum anal-
particular elements generation method based on geometry.” ysis of shear banding in the direct shear test.” Géotechnique 57 (6):
Int. J. Numer. Methods Eng. 110 (11): 1003–1020. https://doi.org/10 513–526. https://doi.org/10.1680/geot.2007.57.6.513.
.1002/nme.5433. Wu, H., A. Papazoglou, G. Viggiani, C. Dano, and J. Zhao. 2020.
Rechenmacher, A., S. Abedi, and O. Chupin. 2010. “Evolution of force “Compaction bands in Tuffeau de Maastricht: insights from X-ray to-
chains in shear bands in sands.” Géotechnique 60 (5): 343–351. mography and multiscale modeling.” Acta Geotech. 15 (1): 39–55.
https://doi.org/10.1680/geot.2010.60.5.343. https://doi.org/10.1007/s11440-019-00904-9.
Rothenburg, L., and R. J. Bathurst. 1989. “Analytical study of induced an- Wu, H., J. Zhao, and N. Guo. 2019a. “Multiscale modeling of compaction
isotropy in idealized granular materials.” Géotechnique 39 (4): 601– bands in saturated high-porosity sandstones.” Eng. Geol. 261: 105282.
614. https://doi.org/10.1680/geot.1989.39.4.601. https://doi.org/10.1016/j.enggeo.2019.105282.
Rothenburg, L., and R. J. Bathurst. 1992. “Micromechanical features of Wu, Q. X., Z. X. Yang, and X. Li. 2019b. “Numerical simulations of gran-
granular assemblies with planar elliptical particles.” Géotechnique ular material behavior under rotation of principal stresses: microme-
42 (1): 79–95. https://doi.org/10.1680/geot.1992.42.1.79. chanical observation and energy consideration.” Meccanica 54 (4–5):
Rowe, P. W. 1962. “The stress-dilatancy relation for static equilibrium of 723–740. https://doi.org/10.1007/s11012-018-00939-4.
an assembly of particles in contact.” Proc. R. Soc. London, Ser. A Xu, X. M., D. S. Ling, Y. P. Cheng, and Y. M. Chen. 2015. “Correlation
269 (1339): 500–527. between liquefaction resistance and shear wave velocity of granular
Rui, R., F. van Tol, X. L. Xia, S. van Eekelen, G. Hu, and Y. Y. Xia. 2016. soils: a micromechanical perspective.” Géotechnique 65 (5): 337–
“Evolution of soil arching; 2D DEM simulations.” Comput. Geotech. 348. https://doi.org/10.1680/geot.SIP.15.P.022.
73: 199–209. https://doi.org/10.1016/j.compgeo.2015.12.006. Yamamoto, S. 1995. “Fundamental study on mechanical behavior of gran-
Siiriä, S., and J. Yliruusi. 2007. “Particle packing simulations based on ular materials by DEM.” Dr. Eng. thesis, Dept. of Civil Engineering,
Newtonian mechanics.” Powder Technol. 174 (3): 82–92. https://doi Nagoya Institute of Technology.
.org/10.1016/j.powtec.2007.01.001. Zeng, D. Z., E. B. Zhang, Y. Y. Ding, Y. G. Yi, Q. B. Xian, G. J. Yao, H. J.
Syed, Z., M. Tekeste, and D. White. 2017. “A coupled sliding and rolling Zhu, and T. H. Shi. 2018. “Investigation of erosion behaviors of
friction model for DEM calibration.” J. Terramech. 72: 9–20. https://doi sulfur-particle-laden gas flow in an elbow via a CFD-DEM coupling
.org/10.1016/j.jterra.2017.03.003. method.” Powder Technol. 329: 115–128. https://doi.org/10.1016/j
Tabaroei, A., S. Abrishami, and E. S. Hosseininia. 2017. “Comparison be- .powtec.2018.01.056.
tween two different pluviation setups of sand specimens.” J. Mater. Civ. Zhao, J. D., and N. Guo. 2015. “The interplay between anisotropy and strain
Eng. 29 (10): 04017157. https://doi.org/10.1061/(ASCE)MT.1943 localisation in granular soils: a multiscale insight.” Géotechnique 65 (8):
-5533.0001985. 642–656. https://doi.org/10.1680/geot.14.P.184.

© ASCE 04021278-13 Int. J. Geomech.

Int. J. Geomech., 2022, 22(2): 04021278