06 Effect of Heterogeneity of Particle Properties On Variability of Laboratory Sandy Soil Properties A Random Discrete Element Perspective

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

Computers and Geotechnics 141 (2022) 104497

Contents lists available at ScienceDirect

Computers and Geotechnics


journal homepage: www.elsevier.com/locate/compgeo

Effect of heterogeneity of particle properties on variability of laboratory


sandy soil properties: A random discrete element perspective
Jia-Yan Nie a, Yi-Fei Cui a, Zhi-Yong Yang b, *, Yan-Zhou Yin c, Zi-Jun Cao d, Dian-Qing Li d
a
State Key Laboratory of Hydroscience and Engineering, Tsinghua University, Beijing 100084, PR China
b
School of Civil Engineering, Sun Yat-Sen University, Zhuhai 519082, PR China
c
Key Laboratory of Mountain Hazards and Earth Surface Processes, Institute of Mountain Hazards and Environment, Chinese Academy of Sciences, Chengdu, PR China
d
State Key Laboratory of Water Resources and Hydropower Engineering Science, Institute of Engineering Risk and Disaster Prevention, Wuhan University, Wuhan
430072, PR China

A R T I C L E I N F O A B S T R A C T

Keywords: Many experimental results have presented that soil deformation and strength properties for the same type of soil
Random field modeling samples are with a certain variability at the same density and stress states, and even at the identical laboratory
Discrete element method conditions. Micro mechanical interpretations of this phenomenon are the random spatial arrangement of con­
Heterogeneity of particle properties
stituent particles and heterogeneity of particle geometry and mechanical properties. This paper mainly focuses
Drained triaxial compression
on the influence of the latter, which is still an open issue for the community of granular materials, by the random
Variability of soil properties
discrete element method. The random field model is firstly incorporated into the discrete element method to
characterize the heterogeneity of equivalent particle properties for a simple linear rolling resistance contact
model. Two RFM parameters named coefficient of variation and scale of fluctuation are adopted to represent the
variability and spatial correlation pattern of particle properties, respectively. Monte Carlo simulations are then
performed by repeatedly conducting DEM simulations of drained triaxial compression tests on a series of dense
heterogeneous specimens. The macroscopic soil deformation and strength parameters are analyzed statistically.
The contact network characteristics of three representative heterogeneous specimens are discussed and
compared with those of the homogenous specimen to explore the underlying microscopic mechanisms. More­
over, the variability of soil properties due to the random spatial arrangement of constituent particles is compared
with that due to the heterogeneity of particle properties. The study helps to provide more insights into the cross-
scale understanding of the variability of sandy soil properties in the laboratory.

1. Introduction stress states, and even at the similar laboratory conditions (Schwiteilo
and Herle, 2016; Schneider-Muntau et al., 2021). This variability would
Granular soils like sands are ubiquitous on the earth surface, and be more pronounced for in-situ soils because of the complexity of the
have been widely employed as the common construction materials of experienced geological process and stress paths. The variability of soil
geotechnical or hydraulic engineering structures. The serviceability and properties has promoted the probabilistic design or evaluation of the
ultimate limit states of these structures under external loadings are engineering structures (Phoon and Kulhawy, 1999; Fenton and Griffiths,
closely related to the physical and mechanical properties of soils. 2008; Qi and Li, 2018; Gao et al. 2019; Yang et al., 2021). However, the
Traditional limit state analysis methods take the soil deformation and underlying physical mechanisms of the variability of soil properties have
strength parameters, which are generally determined from the labora­ been not yet explored from the perspective of granular mechanics, at
tory or field tests, as inputs for the design or evaluation of the relevant least partially, even at the soil sample scale.
structures (Craig, 2004). Many experimental results have shown that For sandy soils, the complexity of their macro mechanical behaviors
macroscopic soil deformation and strength parameters for the same sort mainly originates from the packing patterns and interactions between
of soil samples may exhibit a certain variability at the same density and constituent particles, which is further related to the particle geometry

* Corresponding author.
E-mail addresses: njiayan@mail.tsinghua.edu.cn (J.-Y. Nie), yifeicui@mail.tsinghua.edu.cn (Y.-F. Cui), yangzhy85@mail.sysu.edu.cn (Z.-Y. Yang), yinyanzhou@
imde.ac.cn (Y.-Z. Yin), zijuncao@whu.edu.cn (Z.-J. Cao), dianqing@whu.edu.cn (D.-Q. Li).

https://doi.org/10.1016/j.compgeo.2021.104497
Received 8 April 2021; Received in revised form 30 September 2021; Accepted 3 October 2021
Available online 13 October 2021
0266-352X/© 2021 Elsevier Ltd. All rights reserved.
J.-Y. Nie et al. Computers and Geotechnics 141 (2022) 104497

properties (e.g., particle size and shape) (Mollon and Zhao, 2013; Vangla sample. Therefore, in this study, the RFM is still adopted to characterize
and Latha, 2015; Yang and Luo, 2015; Gong et al., 2019a) and me­ the heterogeneity of equivalent particle properties (e.g., effective
chanical properties (e.g., particle Young’s modulus and sliding friction) modulus, contact normal to tangential stiffness ratio, sliding friction and
(Antony et al., 2006; Coetzee and Els 2009; Barreto and O’Sullivan, rolling friction coefficients) for a simple linear rolling resistance contact
2012; Huang et al., 2014; Gong et al., 2019b), other than the effect of model in DEM simulations of drained triaxial compression, and further
pore fluid (Ma et al., 2021). Hence, it can be anticipated that there are at explore its effect on the variability of soil properties. And then, two COV
least two factors leading to the variability of soil properties for the same and three δ are combined in pairs to prepare different types of hetero­
type of soil samples at the same density and stress states, namely, the geneous soils. After that, Monte Carlo simulations are conducted by
random spatial arrangement of constituent particles and heterogeneity repeatedly performing DEM simulations on a series of dense specimens
of particle geometry and mechanical properties. Although the effects of for each type of heterogeneous soil. The corresponding macroscopic soil
the polydispersity of particle shape and size on the behaviors of sandy deformation and strength parameters are statistically analyzed to
soils have been experimentally or numerically investigated by many explore the effects of COV and δ of particle properties on the variability
researchers (Minh and Cheng, 2013; Nguyen et al., 2015; Jiang et al., of soil properties. The contact network characteristics of three repre­
2018; Rorato et al., 2018; Yang and Luo, 2018; Nie et al., 2021), the sentative heterogeneous specimens are analyzed and compared with
heterogeneity of particle mechanical properties has received little those of the homogenous specimen to explore the underlying micro­
attention. In fact, this heterogeneity has been proved by many re­ scopic mechanisms. Moreover, the variability of soil properties due to
searchers. For example, a series of micro mechanical tests on different the random spatial arrangement of constituent particles is also
types of natural sand grains (e.g., Leighton Buzzard sand, limestone, compared with that due to the heterogeneity of particle properties.
completely decomposed granite grains) have shown that the Young’s The remaining part of this paper is organized as follows. Firstly, the
modulus and inter-particle friction of the same class of sand grains vary basic theories of the DEM and RFM are briefly introduced, and then the
within a certain range, and the variability differs for different materials coupled RFM and DEM model (RDEM) for drained triaxial compression
(Senetakis et al., 2013; Sandeep and Senetakis, 2018a, 2018b). The is presented in detail. In the following, the macro and micro mechanical
variability of particle Young’s modulus and inter-particle friction mainly behaviors of heterogeneous and homogeneous samples are analyzed,
results from the variable particle mineralogy and surface roughness respectively. Finally, some primary conclusions and discussions are
characteristics. In addition, Ren et al. (2021) further found that the summarized.
uneven distribution of the coating and the morphology of the grains
would introduce more dispersity into these particle properties. And 2. Random discrete element model of drained triaxial
hence a probabilistic selection of input properties might be more compression
appropriate in the discrete element method (DEM) simulations for the
materials showing high discrepancy in their properties (Ren et al., 2.1. Discrete element method
2021). Almost at the same time, Rorato et al. (2021) proposed a novel
method to simplify the calibration of rolling resistance by relating the Discrete element method (DEM) was originally proposed by Cundall
rolling friction and particle true sphericity, which enabled that the (Cundall, 1971) to deal with rock mechanics problems using block ele­
variability of grain shape could be directly mapped into the DEM. These ments. It was further extended to reproduce the granular soils’ behaviors
novel works naturally necessitate the studies with regard to the effect of by adopting ideal disks or spheres to represent real soil particles (Cun­
the heterogeneity of particle properties on the variability of soil prop­ dall and Strack, 1979). The basic principle of the DEM is that at the given
erties using the DEM. Recently, some relevant DEM studies have force or displacement boundary conditions, the translational and rota­
emerged within the geomechanics community. For instance, Yang et al. tional kinematics of constituent spherical particles are governed by the
(2017) adopted the random field model (RFM) to characterize the het­ following Newton’s equation and Euler’s equation, respectively:
erogeneity of particle sliding friction, and investigated its influence on
dvi
the drained biaxial shearing behaviors of dense sands. Rorato et al. Fi = m (1a)
dt
(2018) discussed the effect of particle shape distribution on the drained
triaxial shearing responses of sands, through assigning varying rolling dωi
friction coefficients on the constituent particles in the DEM based on Mi = Ii (1b)
dt
different probability distributions. Zhao and Liu (2020) used the RFM to
simulate the heterogeneity of particle Young’s modulus, bonding where Fi and Mi are the resultant force and resultant torque acting on or
strength and sliding friction, and studied the fragmentation mechanisms around the mass centroid of the spherical particle, respectively, and Fi
of rock samples under uniaxial compression. Nevertheless, the effect of consists of contact force and local damping force fd while gravity force is
the heterogeneity of particle properties on the variability of sandy soil ignored in this study; m and Ii are the mass and principal moment of
properties is still an open issue. inertia of the spherical particle, respectively; vi and ωi are the trans­
The advantage of RFM to characterize the heterogeneity of particle lational and angular velocity of the spherical particle, respectively. The
properties is that two RFM parameters named coefficient of variation standard and extending leapfrog algorithms are adopted to solve the
COV and scale of fluctuation δ can consider the variability of particle above Newton’s and Euler’s equations, respectively. The local damping
property for each particle and the spatial correlation pattern of particle force fd is artificially applied on each particle in order to facilitate the
properties between constituent particles, respectively. Although some dissipation of kinetic energy of the whole granular particles system, and
studies have shown that the particle shape is spatially independent for it is determined as follow:
air-pluviated sand specimens (e.g., Rorato et al., 2020), this does not
fd = − α|fu |vp /|vp | (2)
indicate that the other particle properties or the shape of other sand
specimens especially of the undisturbed or less disturbed soil specimens where fu and vp are the resultant contact force and translational velocity
are not possible to be spatially correlated. The spatial correlation of the particle, respectively; α is the local damping ratio and set as 0.3
assumption for particle properties between constituent particles is based reference to our previous studies (Nie et al., 2020; 2021). The minus sign
on the fact that particles at adjacent locations within undisturbed or less means that the local damping force fd acts along the opposite direction of
disturbed soils may experience the similar formation history. Moreover, the particle translational motion. Note that there is no damping torque
the RFM is able to model the spatially independent particle properties, applied on the particles in this study.
similar with that adopted in Rorato et al. (2018), when the scale of As for the contact force fc , the following two contact models at the
fluctuation is smaller than the minimum particle size within the soil

2
J.-Y. Nie et al. Computers and Geotechnics 141 (2022) 104497

particle scale are employed in the DEM to determine the normal contact simulations (Kawamoto et al., 2016; Wang et al., 2021). Similar with
force fcn and tangential contact force fct , respectively: Iwashita and Oda (1999), a simple rolling resistance model is adopted in
this study and the corresponding rolling resistant momentum Mrc can be
fcn = − kcn δn n (3a)
determined as:
/
fct = − min(|fc’t − kct Δut |, |μs fcn |)t (3b) Mcr = − min(|Mc’r − kcr Δθr , |μr Rfcn |)θr |θr | (6)

where n and t are the unit contact normal and tangential vector,
where Δθr is the incremental relative rotation angle between two con­
respectively, as shown in Fig. 1, whose directions are consistent with
those of relative normal and tangential translation of the given particle tacting particles at the current time step; M’rc is the rolling resistant
(e.g., particle p2) against the contacting particle (e.g., particle p1), and momentum at the previous time step; R is the contact effective radius
hence the minus signs in both equations represent that the directions of and defined through R1 = R1P1 + R1P2 ; krc is the contact rolling stiffness and
fcn and fct are opposite with those of n and t, respectively; δn is the normal can be determined by the contact tangential stiffness ktc through krc =
penetration depth between two contacting particles; Δut is the incre­ 2
ktc R ; μr is the contact rolling friction determining the upper-bond of the
mental relative tangential displacement of the contacting particles at the rolling resistant momentum, and once the magnitude of Mrc reaches to
current time step; fc’t is the tangential contact force at the previous time |μr Rfcn |, the relative rolling occurs between two contacting particles. The
step; μs is the contact sliding friction defining the maximum magnitude minus sign in Eq. (6) also indicates that the direction of Mrc is opposite
of tangential contact force, and once the magnitude of fct reaches to with that of the relative rolling motion of the particle. Similar with μs ,
|μs f nc |, the relative tangential sliding occurs between two contacting μr is also derived from the particle rolling friction of two contacting
particles; knc and ktc are contact normal and tangential stiffness, respec­ particles as follow:
tively. In this study, a simple linear contact model is adopted to derive ( )
the contact normal and tangential stiffness directly from the particle μr = min μrp1 , μrp2 (7)
properties of two contacting particles as follows:
* * where μrp1 and μrp2 are the particle rolling friction of two contacting
2(Ep1 Rp1 )(Ep2 Rp2 )
kcn = * *
(4a) particles, respectively. The above introduced simple linear rolling
Ep1 Rp1 + Ep2 Rp2
resistance contact model in PFC3D has also been adopted by many re­
* * * * searchers to efficiently study the sandy soil behaviors considering the
2(Ep1 Rp1 kp1 )(Ep2 Rp2 kp2 )
kct = * * * *
(4b) effect of particle shape (e.g., Gu et al., 2020; Rorato et al., 2021).
Ep1 Rp1 kp1 + Ep2 Rp2 kp2

2.2. Random field modeling


where E*p1 and E*p2 are the effective modulus of the particle p1 and p2,
respectively, which increase with the particle Young’s modulus; k*p1 and The macroscopic soil deformation and strength parameters exhibit
k*p2 are the contact normal to tangential stiffness ratio of two contacting significant spatial variability because of the complicated geological
particles which increase up to a finite positive value as the particle processes such as weathering, transportation, and sedimentation. Van­
Poisson’s ratio increases; Rp1 and Rp2 are the radius of two contacting marcke (1977) found that the spatial variability of soil properties in the
particles, respectively (Itasca Consulting Group Inc., 2014). In addition, field can be well described by the random field model (RFM). Since then,
the contact sliding friction μs is determined from the particle sliding the RFM has received considerable attention in the probabilistic design
friction of both contacting particles as follow: or evaluation of geotechnical engineering structures. This study adopts a
( ) similar philosophy but uses the RFM at a rather small scale (e.g., particle
μs = min μp1 , μp2 (5) scale) to characterize the heterogeneity of particle properties within a
soil specimen, reference to the previous studies (Yang et al., 2017; Zhao
where μp1 and μp2 are the particle sliding friction of two contacting and Liu, 2020).
particles, respectively. In this study, the RFM considers four basic particle properties of the
Considering that using ideal spheres in the DEM simulation to aforementioned linear rolling resistance model as random variables,
reproduce the soil behavior would acquire a smaller shear strength than
including particle effective modulus E*p , contact normal to tangential
common experimental results, the rolling resistance model are often
introduced into the contacting particles to enhance the soil strength stiffness ratio k*p , sliding friction μp and rolling friction μrp . The four
(Iwashita and Oda, 1999; Jiang et al., 2005; Zhou et al., 2013). This particle properties are assumed to follow the lognormal distributions
alternative way helps to significantly reduce the computation burden of with given mean values (μ) and coefficient of variations (COV) in
random DEM simulations compared with the realistic shape DEM consideration of their non-negativity characteristics. Each of the particle
properties is spatially auto-correlated by a specific autocorrelation

Fig. 1. 2D schematic of the adopted linear rolling resistance contact model in this study (particles rendered with different colors representing particles of varying
particle properties) (modified from Liang and Zhao, 2019).

3
J.-Y. Nie et al. Computers and Geotechnics 141 (2022) 104497

function for different particles within the soil sample. The autocorrela­ and then the corresponding particle property values are set as the mean
tion function is a function of the scale of fluctuation (δ) and spatial values of random particle properties of the heterogeneous samples. To
distance between particles. A smaller δ implies a weaker spatial corre­ this end, a robust numerical model configuration consisting of 5000
lation of the particle property between constituent particles and hence spherical particles of uniform grading is adopted reference to our pre­
the sample is more heterogeneous, and vice versa. Specially, if δ is less vious studies (Nie et al., 2020; 2021), as shown in Fig. 2. The numerical
than the minimum particle size, the particle property of constituent sample is prepared through the isotropic compression of a cloud of
particles is spatially uncorrelated and would be varying according to the particles with non-overlap under a numerical stress servo control
assumed probability distribution, and hence the sample is very hetero­ scheme with the target confining pressure of Pc = 100 kPa. During the
geneous; While if δ tends to be infinitely larger, the particle property of isotropic compression process, the particle sliding and rolling friction
constituent particles within the soil sample is strongly correlated and are assigned with zero, while the other two particle properties are set as
would be kept the same, and therefore the sample becomes their trial mean values. When the sample reaches an initial mechanically
homogenous. equilibrium state, the particle sliding and rolling friction are also reset as
Suppose that four lognormal random fields for E*p , k*p , μp and μrp are to their trial mean values, and a second numerical stress servo control
be constructed. Take the lognormal random field k(x, y, z) of the particle procedure is conducted to acquire the final well-“consolidated” sample
property k as an example, where k denotes E*p , k*p , μp μrp , respectively, before shearing. In the following, a constant loading rate γ̇ of 0.01/s is
adopted to compress the specimen through the top and bottom walls
and x, y and z are spatial coordinate components of the particle mass
under a well-controlled confining stress of 100 kPa, and the compression
centroid. The autocorrelation matrix ρ for the random field k(x, y, z) is
process stops until the axial strain reaches 40% where the soil critical
firstly estimated by the assumed Gaussian autocorrelation function as
state is obviously observed. The density scaling method is adopted to
follow (Fenton and Griffiths, 2008):
acquire a relatively acceptable critical time step Δtcr (O’Sullivan 2011)
[ (( )2 ) ]
( ) Δx,ij
)2 (
Δy,ij
)2 (
Δy,ij to ensure the computation efficiency. The computational time step Δt is
ρij Δx,ij , Δy,ij , Δz,ij = exp − π + + (8) set to 10− 5 s for all simulations. The inertia number Iinertia is less than
δx δy δz
10− 3 during the whole compression process, indicating that the shearing
where ρij is the coefficient of autocorrelation of the particle property k is quasi-static (Midi, 2004). The final mean values of the four particle
between the i-th and j-th particle; Δx,ij, Δy,ij and Δz,ij are the spatial property parameters are determined as: E∗p =1000 MPa, k∗p =0.25, μp =
distances between the mass centroids of the i-th and j-th particles along 0.4 and μpr = 0.4, to make the stress–strain curves and critical internal
the x-, y- and z- directions, respectively; δx, δy, and δz are the scale of friction angle of the homogeneous sample can be able to represent the
fluctuation of the particle property in the x-, y- and z- directions, laboratory results of the common dense sub-rounded quartz sands. The
respectively, which are assumed to be equal to the value of δ in this length of the final densified homogeneous sample before shearing is
study. The Cholesky decomposition is then adopted to decompose the around 33 mm and about 11 times larger than the maximum particle
autocorrelation matrix ρ to a lower-triangular matrix L according to the diameter, which satisfies the basic requirement for the size of numerical
following expression (Yang and Ching, 2021): sample in the DEM so as to relieve the rigid boundary effect (Jamiol­
kowski et al., 2004).
ρ = LLT (9) In terms of the preparation of heterogeneous samples with varying
Based on the lower-triangular matrix L, a correlated standard normal particle properties, three scale of fluctuation (e.g., δ = 0.001 m, 0.018 m,
random field N(x, y, z) is then generated through the following equation 0.033 m) and two coefficients of variation (e.g., COV = 0.03, 0.10) are
(Nie et al., 2020): combined to construct different random field models for four random
particle properties. Different combinations of the coefficient of variation
N(x, y, z) = LU (10) and scale of fluctuation are adopted to prepare different types of het­
erogeneous soils, which would be illustrated in the following. As
where U is an independent standard normal column vector with n ele­ aforementioned, four random particle properties are assumed to be in­
ments; n is the number of constituent particles within the soil sample. dependent and lognormally distributed. Table 1 summarizes the statis­
Finally, the lognormal random field k(x, y, z) of the particle property k tical characteristics of random field models for the four random particle
can be given by (Fenton and Griffiths, 2008): properties. Once the random field model for each type of particle
k(x, y, z) = exp[μlnk + σ lnk ⋅N(x, y, z) ] (11) property are constructed based on the section of 2.2, the particle prop­
erty values for each random field realization are mapped into constitu­
where μlnk and σlnk are the mean and standard deviation of the par­ ent particles within the already generated homogeneous sample based
ticle property k in the normal space, which can be estimated by the on their mass centroid positions to acquire the heterogeneous sample.
followings (Fenton and Griffiths, 2008): Note that the current heterogeneous sample would not be mechanically
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )̅ stable due to the change of particle properties, and a new numerical
σ lnk = ln 1 + COV 2k (12a)
stress servo equilibrium process should be further conducted. A very
slight disturbance is observed during this process and the average value
1 2 of the normalized distances between mass centroids of each particle by
μlnk = lnμk − σ (12b)
2 lnk their corresponding sizes within the homogeneous and heterogeneous
where μk and COVk are the mean and COV of the particle property k. It samples is only about 0.01. Hence, it can be assumed that the initial
should be noted that the cross-correlation of different particle properties particle spatial arrangement and density are consistent with each other
is not considered in this study, and hence there are totally four inde­ for homogeneous and heterogeneous samples, and the effect of particle
pendent random field models mapped into the heterogeneous samples to random spatial distribution on the variability of soil properties can be
consider the heterogeneity of particle properties. neglected in this study. After the preparation of heterogeneous samples,
similar drained triaxial shearing procedure is conducted as that on the
homogeneous samples.
2.3. RDEM of drained triaxial compression Taking the random field model of particle sliding friction μp as an
example, Fig. 3 shows six typical realizations of μp random field with
In this section, the RFM is incorporated into the DEM to conduct different combinations of COV and δ. It can be seen that the random field
drained triaxial compression simulations. Firstly, a reasonable homo­ model can effectively simulate the heterogeneity of particle properties of
geneous sample with constant particle properties needs to be prepared,

4
J.-Y. Nie et al. Computers and Geotechnics 141 (2022) 104497

Fig. 2. (a) Numerical model configuration of the drained triaxial compression simulation and (b) particle size grading curve.

Table 1
Statistical characteristics of random field models for four equivalent particle properties and their range of variation, and some other statistical values of 500 random
simulations for each type of heterogeneous soil sample.
Particle Probability Mean Coefficient of Scale of Range of Minimum and Range of Range of contact
properties Distribution μ Variation (COV) fluctuation (δ), variation maximum average dimensionless friction angle, φs (◦ )
m values stiffness, k

E*p (MPa) Lognormal 1000 0.03 0.001 [863.7,1161.4] 998.6,1001.2 [4318.5,5807.0]


0.018 [850.2,1134.0] 965.8,1033.8 [4251.0,5670.0]
0.033 [875.3,1132.2] 947.8,1051.7 [4376.5,5661.0]
0.10 0.001 [612.1,1638.9] 995.4,1003.8 [3060.5,8194.5]
0.018 [580.8,1514.1] 891.2,1117.5 [2904.0,7570.5]
0.033 [639.9,1505.9] 837.1,1182.2 [3199.5,7529.5]
k*p Lognormal 0.25 0.03 0.001 [0.215,0.290] 0.250,0.250
0.018 [0.220,0.289] 0.244,0.257
0.033 [0.220,0.286] 0.238,0.266
0.10 0.001 [0.150,0.408] 0.249,0.251
0.018 [0.162,0.403] 0.232,0.274
0.033 [0.163,0.388] 0.213,0.308
μp Lognormal 0.4 0.03 0.001 [0.337,0.463] 0.400,0.401 [18.8,24.7]
0.018 [0.351,0.459] 0.390,0.412 [19.3,24.7]
0.033 [0.349,0.457] 0.378,0.419 [19.3,24.7]
0.10 0.001 [0.225,0.649] 0.398,0.402 [12.4,33.0]
0.018 [0.257,0.631] 0.367,0.441 [14.6,32.2]
0.033 [0.253,0.623] 0.331,0.466 [14.0,31.8]
μpr Lognormal 0.4 0.03 0.001 [0.340,0.462] 0.400,0.401
0.018 [0.348,0.457] 0.388,0.411
0.033 [0.354,0.454] 0.379,0.422
0.10 0.001 [0.233,0.645] 0.398,0.402
0.018 [0.251,0.622] 0.360,0.436
0.033 [0.265,0.606] 0.333,0.479

Note that the values of “Rang of variation” are determined from the maximum and minimum particle property parameters of 500 random simulations; “Minimum and
maximum average values” denotes that the minimum and maximum average values of particle property parameters for each sample of 500 random simulations;
“Range of dimensionless stiffness, k” and “Range of contact friction angle, φs (◦ )” are calculated based on the values of “Rang of variation” reference to Aboul et al.
(2017).

granular soils in the DEM model to some extent. For a given COV, the particles and their maximum and minimum average values of 500
sample would be more heterogeneous when δ is smaller. Specially, the random simulations are also summarized in Table 1 for each type of
particle property is spatially uncorrelated for the sample with δ = 0.001 heterogeneous soil. According to the previous literatures (Roux and
m, which is less than the minimum particle size, while it would be kept Combe, 2010; Aboul et al., 2017; Sibille et al., 2019), the dimensionless
the same and the sample becomes homogenous when δ is infinitely large. stiffness k defined as k = E*p/2Pc for each case (shown in Table 1) is
For a given δ, the variability of particle property for each particle in­ significantly larger than the threshold (e.g., 1875) proposed by Aboul
creases with the COV. In this study, 500 number of random field re­ et al. (2017), and the contact rolling stiffness is also higher than the
alizations are independently generated based on the Monte Carlo corresponding extreme, indicating that the adopted random contact
scheme reference to Li et al. (2020) for each random particle property stiffness in this study is large enough to avoid the interplay of local
under a given COV and δ. The acquired 500 samples for each type of elastic parameters and contact friction parameters on the macroscopic
heterogeneous specimen are further used to perform drained triaxial plastic properties. In addition, the contact friction angles for all cases in
compression. Table 1 are basically larger than 15◦ , suggesting that the coupled effects
The variation range of four particle properties of constituent of contact sliding friction and rolling friction on the critical state shear

5
J.-Y. Nie et al. Computers and Geotechnics 141 (2022) 104497

Fig. 3. Six typical random field realizations for particle sliding friction μp with different COV and δ.

strength can be safely ignored. These priors help to uncouple the com­ 3.1.1. Stress–strain relationships
bined effects of the heterogeneity of local elasticity parameters and Fig. 4 and Fig. 5 present the evolutions of the deviatoric stress q and
contact friction parameters on the variability of soil deformation and the void ratio e with the axial strain εa for 500 heterogeneous samples
strength properties. under different combinations of COV and δ, respectively. The corre­
sponding curves of the homogeneous sample are also shown in Fig. 4 and
3. Effect of heterogeneity of particle properties on soil behaviors Fig. 5. It can be seen from Fig. 4 and Fig. 5 that all specimens experience
strain softening after reaching the peak shear strength and finally reach
3.1. Macroscopic behaviors the critical state when the axial strain is larger than 0.30. Meanwhile, all
specimens show shear dilation instantly after compression because of
In this section, the stress–strain relationships and soil deformation the very dense characteristic at the initial state. And the stress–strain
and strength parameters (e.g., Shear modulus G, Poisson’s ratio v, peak curves of heterogeneous specimens appear as a curve cluster with a
internal friction angle φp and critical internal friction angle φc ) of 500 certain width especially at the large deformation stage. The width of the
random samples for each type of heterogeneous specimen are statisti­ curve cluster increases with the increasing of δ for a given COV. The
cally analyzed compared with those of the homogeneous sample. Similar differences of stress–strain curves of heterogeneous specimens result
analysis is also conducted on the additional 500 random simulations of from the distinct spatial correlation pattern and varying average values
homogenous samples with different particle spatial arrangement but the of particle properties, which will be further explained in the following
same density. Based on these statistical results, effects of two random section. In addition, it is interesting to note that the average stress–strain
field model parameters (e.g., COV and δ) and the random spatial curves of 500 heterogeneous specimens are closer to those of the ho­
arrangement of constituent particles on the variability of soil properties mogeneous specimen with the increase of δ for a given COV.
are systematically discussed and compared. On the other hand, for δ = 0.001, the stress–strain curve cluster of

Fig. 4. Stress–strain curves for homogeneous and heterogeneous specimens with different δ and COV = 0.10 at the same initial density and stress states.

6
J.-Y. Nie et al. Computers and Geotechnics 141 (2022) 104497

Fig. 5. Stress–strain curves for homogeneous and heterogeneous specimens with different δ and COV = 0.03 at the same initial density and stress states.

random heterogeneous samples does not fully cover the curve of the between homogeneous and heterogeneous samples more pronounced.
homogeneous specimen especially at the peak shear strength state, Comparing the results of Fig. 4 and Fig. 5, it can be seen that for a given
although this phenomenon is not so obvious for COV = 0.03. This might δ, the discrepancies of stress–strain curves of heterogeneous specimens
be because when δ = 0.001, particle properties are spatially uncorre­ increase with COV. That is, the width of the curve cluster increases with
lated and hence their average values over constituent particles at each COV, and this phenomenon is much clearer at the larger scale of
random field realization are close to their corresponding mean values of fluctuation.
RFM, which are also the particle property values of the homogeneous In order to distinguish and compare the variability of soil properties
sample, as shown in Table 1. While weak particles (particles with due to the random spatial arrangement of constituent particles and that
smaller sliding and rolling friction than the mean values of RFM) within due to the heterogeneity of particle properties, 500 additional random
heterogeneous samples collapse earlier compared with the homoge­ simulations of homogeneous samples are conducted, and the initial void
neous sample, and then induce a smaller peak shear strength. Similar ratio for all simulations is kept the same as the specific homogeneous
phenomenon has also been observed in Rorato et al. (2018). In other sample used to prepare heterogeneous samples. Fig. 6 (a) and (b) present
words, when two particles touch with each other, only the more their stress–strain curves, similar strain softening and shearing dilata­
spherical and smooth particle transmits its rolling and sliding friction to tion behaviors are observed. And it is anticipated that the curves of the
the contact (Rorato et al., 2018). For larger COV (e.g., 0.10) under δ = specific homogeneous sample used for the preparation of heterogeneous
0.001, it means that particle properties of heterogeneous samples are of samples are totally within the curve cluster of other 500 random ho­
greater variability, and more weaker particles within samples reduce the mogeneous samples, which is different with the random heterogeneous
peak shear strength, making the difference of peak shear strength samples with δ = 0.001. Moreover, a relatively large width of the curve

Fig. 6. Stress–strain curves for random homogeneous specimens with different spatial arrangement of constituent particles at the same initial density and stress states
and the corresponding statistical results of soil deformation and strength properties.

7
J.-Y. Nie et al. Computers and Geotechnics 141 (2022) 104497

cluster has been acquired under this case, compared with those of the (PDFs) of both two deformation parameters for random heterogeneous
random heterogeneous samples shown in Figs. 4 and 5, suggesting that samples can be well-fitted by the Gaussian distribution. This conclusion
the variability of peak shear strength due to the random spatial holds for the random heterogeneous samples with other combinations of
arrangement of constituent particles can be greater than that due to the COV and δ, and also the random homogeneous samples (shown in Fig. 6
heterogeneity of particle properties, which will be further analyzed in (c-f)). Fig. 8 further shows the evolutions of the statistical characteristics
the next section. (e.g., the mean and COV) of G and v with δ for different COVs of particle
properties. It can be seen that the COV and δ of E∗p and k∗p have little
3.1.2. Soil deformation and strength parameters influence on the mean values of the shear modulus G and Poisson’s ratio
In order to quantify the effect of heterogeneity of particle properties v; while they have a relatively large effect on the COV of G and v. Spe­
on the variability of soil deformation and strength properties, the shear cifically, the increase of COV or δ of E∗p and k∗p would increase the COV of
modulus G, Poisson’s ratio, v, peak internal friction angle φp , and critical G and v. Moreover, the COV of G and v are always less than the COV of E∗p
internal friction angle φc are determined for all DEM simulations.
and k∗p because of the average effect of particle properties in obtaining
Among which, the shear modulus G and Poisson’s ratio v are calculated
the bulk values.
based on the stress–strain curves within the axial strain of 0.2%, during
Similarly, Fig. 9 shows the frequency distribution histograms of the
which the soil specimen is assumed to be linear elastic, and the corre­
peak and critical internal friction angle, φp and φc of 500 random het­
sponding soil deformation parameters are determined by the following
erogeneous samples for COV = 0.10 and δ = 0.033 m. It can be seen that
two expressions:
PDFs of two soil strength parameters can also be well-fitted by the
G=
q
(13a) Gaussian distribution, and the fluctuation ranges of φp and φc can be up
to 4 and 3 , respectively for the heterogeneous samples with large COV
◦ ◦
3εs
and δ. Fig. 10 gives the evolutions of the mean and COV of φp and φc with
(εx + εy )/2
v= (13b) δ for different COVs of particle properties. Different from the soil
εz deformation parameters G and v, the COV and δ of μpr and μp have a
certain influence on the mean values of φp and φc. Specifically, the
where q is the deviatoric stress and εs is the deviatoric shear strain
decrease of COV or increase of δ of μpr and μp can increase the mean
defined as εs = 2(εz − (εx + εy )/2)/3, and εx , εy and εz are the principal
values of φp and φc. This phenomenon may be because φp and φc are
strain along the direction of x, y and z axis, respectively (Wang et al.,
mainly affected by the average values of μpr and μp within each het­
2020). Considering that the heterogeneity of particle properties, the
erogeneous sample. Reference to Fenton and Griffiths (2008), if the
strains in the x and y direction for heterogeneous samples might be
assumed lognormally distributed random fields of particle properties are
somewhat deviated during compression, and hence the average value of
used, the mean values of the average values of particle properties of 500
εx and εy is set as the radial strain to determine the shear modulus G and
heterogeneous specimens decrease with COV of particle properties and
Poisson’ ratio v.
this decrease is more significant for a smaller δ. The slight dependences
The peak and critical internal friction angles, φp and φc are deter­
of the mean values of G and v on the COV and δ of E∗p and k∗p may be
mined as:
( ) because the changed average values of E∗p and k∗p has a limited influence
φp = sin− 1
σ 1p − σ3
(14a) on the soil deformation parameters within the studied range of this
σ 1p + σ3 study. The COV and δ of μpr and μp have a large influence on the COV of
( ) φp and φc. The COV of φp and φc generally increase with the COV of the
φc = sin− 1 σ1c − σ3
(14b) μpr and μp. They also increase with δ except for φc when COV of μpr and μp
σ1c + σ3 equals to 0.03. The possible reason of this exception is that φc mainly
depends on μpr of particles within the strain localized zone, and hence
where σ1p, σ 1c are the axial stress at the peak and critical state, respec­
the variation of φc would be saturated with δ when the COV of particle
tively. Axial stress within the axial strain from 30% to 40% are averaged
properties is relatively small.
as the critical state axial stress σ1c in order to alleviate the influence of
The mean and COV of four soil property parameters for 500 random
stress fluctuation during the critical state.
homogeneous samples shown in Fig. 6(c-f) have also been plot on Fig. 8
Fig. 7 gives the frequency distribution histograms of the shear
and Fig. 10 for comparing the variability of soil properties due to the
modulus G and Poisson’s ratio v of 500 random heterogeneous samples
particle spatial arrangement and that due to the heterogeneity of par­
for COV = 0.10 and δ = 0.033 m. The shear modulus G and Poisson’s
ticle properties. It is seen that the variability of soil deformation and
ratio v of the specific homogeneous specimen are also shown in Fig. 7
strength parameters of random homogeneous samples are larger than
with blue dot lines. It is found that the probability distribution functions
those of heterogeneous samples with COV of particle properties being

Fig. 7. Frequency distribution histograms of the shear modulus G and Poisson’s ratio v of 500 random heterogeneous samples with COV = 0.10 and δ = 0.033 m.

8
J.-Y. Nie et al. Computers and Geotechnics 141 (2022) 104497

Fig. 8. Evolutions of the mean and COV of the shear modulus G, Poisson’s ratio v with δ for different COVs of particle properties.

Fig. 9. Frequency distribution histograms of the peak friction angle φp and critical friction angle φc of 500 random heterogeneous samples with COV = 0.10 and δ =
0.033 m.

equal to 0.03 or 0.10 if δ is relatively small (e.g., δ = 0.001 m). This of soil properties due to the heterogeneity of particle properties exceeds
result suggests that in this case the contribution of particle spatial the former, and hence this source of variability of soil properties cannot
arrangement is dominant for the variability of soil properties. However, be ignored in such a condition. Moreover, the heterogeneity of particle
when the COV and δ of particle properties become larger, the variability properties exists inherently for natural soils, and it should be taken into

9
J.-Y. Nie et al. Computers and Geotechnics 141 (2022) 104497

Fig. 10. Evolutions of the mean and COV of the peak internal friction angle φp , critical internal friction angle φc with δ for different COVs of particle properties.

account reasonably when conducting DEM studies especially for soils combination of COV and δ, the average values of particle prop­
with highly varying particle properties. The adopted RDEM is readily to erties for constituent particles within 500 random samples are
consider both effects of the random spatial arrangement of constituent almost equal to particle properties of the HOS sample, that is, the
particles and heterogeneity of particle properties. This study consider mean values of four random particle properties, as shown in
them separately in order to explore their own effects more clearly. Table 1. Hence such a selection aims to explore the effect of
correlation pattern of particle properties on soil behaviors. The
particle properties of HOS sample are strongly spatially corre­
3.2. Mechanisms analysis lated while those of HES-A sample are spatially uncorrelated.
(2) The HES-B and HES-C samples are chosen from two random
3.2.1. Representative specimen behaviors simulations for COV = 0.10 and δ = 0.033 m. The stress–strain
In this section, three representative specimens for different types of curves for HES-B and HES-C samples are denoted as black and
heterogeneous samples (e.g., HES-A, HES-B and HES-C) and the specific green lines in Fig. 4c, respectively. The HES-B and HES-C samples
homogeneous sample (HOS) used to prepare heterogeneous specimens are selected representatively as they give the minimum and
(as shown in Fig. 4) are chosen to analyze their micro mechanical be­ maximum critical internal friction angles of 500 random simu­
haviors so as to give some insights into the observed variability of soil lations. Moreover, HES-B and HES-C have similar spatial corre­
properties because of the different heterogeneity characteristics of par­ lation patterns of particle properties but different average values
ticle properties (e.g., different spatial correlation patterns and average of particle properties of constituent particles. Hence the effect of
values of particle properties of constituent particles). Three represen­ average values of particle properties on soil properties can be
tative heterogeneous specimens are selected based on the following discussed.
principles:
The stress–strain curves for HES-A, HES-B, HES-C and HOS have been
(1) The HES-A is selected from one random simulation for COV = redrawn in Fig. 11 for clearer showcase. The corresponding soil defor­
0.10 and δ = 0.001 m. The stress–strain curves for HES-A is mation and strength parameters (i.e., G, ν, φp, φc) and average values of
denoted as pink line in Fig. 4a. It should be noted that for this

10
J.-Y. Nie et al. Computers and Geotechnics 141 (2022) 104497

Fig. 11. Stress–strain curves of the representative heterogeneous and homogeneous specimens.

* *
particle properties of constituent particles (i.e., Ep , kp , μp , μrp ) are also Fig. 12 gives the evolutions of the mean coordination number (MCN)
and mean normal contact force <fn> of four representative samples. It is
attached in the legends of Fig. 11(a) and (b), respectively. It can be easily
interesting that HES-A and HES-B have almost the same MCN and <fn>
seen from Fig. 11 that HOS sample has a larger shear strength than HES-
during shearing, which, to some extent, explains the similar evolutions
A sample at both peak and critical state, suggesting that at the given
of stress–strain curves of these two specimens shown in Fig. 11. In
average values of particle sliding and rolling friction, the shear strength
comparison with the MCN and <fn> of HES-A and HES-B samples, HES-
would increase for samples with more correlated particle friction pa­
C sample has the smallest MCN and largest <fn> while HOS sample has
rameters. While for a given correlation pattern, the shear strength of
intermediate ones. These observations suggest that both the average
HES-C sample is larger than that of HES-B due to the larger average
values and spatial correlation patterns of particle properties affect the
values of particle sliding and rolling friction. Moreover, it is interesting
magnitudes of MCN and <fn> and further affect the bulk behaviors of
to see that stress–strain curves of both HES-A and HES-B samples are
granular soils. Larger average values of μpr and μp would induce a smaller
very consistent, which indicates that the weakening effect of uncorre­
MCN and hence a greater <fn>, while stronger spatial correlations of
lated particle properties on the soil shear strength of HES-A sample has
particle friction parameters would decrease the MCN and increase <fn> .
been counteracted by its larger average values of particle friction pa­
Fig. 13 presents the evolutions of the percentage of sliding contacts ρs
rameters. In addition, it is also seen from Fig. 11 that the macroscopic
and percentage of rolling contacts ρr of four representative samples. It
deformation parameters are mainly determined by the average values of
* can be seen from Fig. 13(a) that four samples have almost the same ρs
local elasticity parameters, and shear modulus G increases with Ep while except a slightly larger one for HES-A owing to more weak constituent
*
Poisson’s ratio v decreases with kp , which are consistent with the results particles. Nevertheless, Fig. 13(b) shows that HES-C and HES-B have the
of Fleischmann et al. (2013). smallest and largest ρr , respectively, which results in the corresponding
largest and smallest shear strength. A small ρr or ρs implies that the
3.2.2. Contact network characteristics specimen has a large resistance to the contact rolling or sliding and
Contact network characteristics (e.g., mean coordination number, further gives a great overall shear strength. As a result, the HES-B and
mean normal contact force, sliding, rolling and weak contacts) are HES-A samples have a final similar shear strength because of the mutual
analyzed in this section in order to explain the obvious shear strength offset effects of the contributions of sliding and rolling contacts to the
differences for four representative specimens from the micromechanics shear strength within two samples.
perspectives. Fig. 14 shows the evolutions of the percentage of weak contacts of
four representative specimens. Herein, the weak contacts are defined as

Fig. 12. Evolutions of the mean coordination number and mean normal contact force of the representative heterogeneous and homogeneous specimens.

11
J.-Y. Nie et al. Computers and Geotechnics 141 (2022) 104497

Fig. 13. Evolutions of the percentages of sliding and rolling contacts of the representative heterogeneous and homogeneous specimens.

arrangement. Effects of two random field parameters on the variability


of soil deformation and strength parameters are analyzed. The vari­
ability of soil properties due to the random particle spatial arrangement
is also compared. The contact network characteristics of four represen­
tative samples are explored to explain the observed shear strength dif­
ferences owing to the heterogeneity of particle properties. Some primary
conclusions from this study are summarized as follows:

(1) The stress–strain curves of random heterogeneous samples with


almost the same particle spatial arrangement but variable particle
properties under drained triaxial compression appear as a curve
cluster with a certain width, which indicates that the heteroge­
neity of particle properties is an important source of the vari­
ability of soil properties. The contribution of particle spatial
arrangement to the variability of soil particles is greater than that
due to heterogeneity of particle properties when their coefficient
of variation is smaller (e.g., 0.03) or larger (e.g., 0.10) under
smaller scale of fluctuation. However, for the larger coefficient of
variation and scale of fluctuation, the contribution of the het­
erogeneity of particle properties is dominant.
(2) The shear modulus, Poisson’s ratio, peak and critical internal
Fig. 14. Evolutions of the percentage of weak contacts of the representative
heterogeneous and homogeneous specimens. friction angles of random heterogeneous specimens follow
Gaussian distributions if lognormally distributed random fields
are used for describing the heterogeneity of uncorrelated particle
the contacts whose contact forces are less than the mean normal contact
properties. The mean values of shear modulus and Poisson’s ratio
force <fn> (Guo and Zhao, 2013). It is interesting to observe that four
of random heterogeneous samples are basically not affected by
specimens have almost the same percentage of weak and strong contacts
COV and δ of particle properties, while the mean values of peak
within the contact networks during shearing. It indicates that the
and critical internal friction angles are. The COV of soil property
average values and spatial correlation pattern of particle properties have
parameters for random heterogeneous specimens generally in­
minor effects on the percentages of weak and strong contacts, and there
creases with the COV and δ of particle properties. Two random
are always more than 60% weak contacts within the soil samples
field model parameters affect the soil strength parameters
regardless of the heterogeneity of particle properties. This phenomenon
through the average values and spatial correlation patterns of
is consistent with the previous DEM studies of homogeneous samples
particle properties of the heterogeneous samples.
(Guo and Zhao, 2013; Nie et al., 2021).
(3) The average values and spatial correlation patterns of particle
properties jointly affect the soil macroscopic and microscopic
4. Conclusions and Discussions
mechanical quantities. The shear strength increases with the
average values of particle sliding or rolling friction and also the
This study incorporates the RFM into DEM to explore the effect of
spatial correlation degrees of these particle properties. And hence
heterogeneity of particle properties on the variability of sandy soil
heterogeneous samples with different average values and spatial
properties. Four equivalent particle properties including particle effec­
correlation degrees of particle properties might have similar
tive modulus, contact normal to tangential stiffness ratio, sliding and
stress–strain curves when subjected to a mutual offset between
rolling friction for a simple linear rolling resistance contact model in the
both effects. The mean coordinate number and percentages of
DEM are assumed as lognormally distributed random variables. Based
sliding or rolling contacts decrease while mean normal contact
on the coupled RFM and DEM model, random DEM simulations of
force increases with the average values of particle sliding or
drained triaxial compression in tandem with MCS are conducted on
rolling friction and the correlation degrees of these particle
heterogeneous samples with almost the same particle spatial

12
J.-Y. Nie et al. Computers and Geotechnics 141 (2022) 104497

properties, which explains the shear strength differences of het­ Gao, G.H., Li, D.Q., Cao, Z.J., Wang, Y., Zhang, L., 2019. Full probabilistic design of earth
retaining structures using generalized subset simulation. Comput. Geotech. 112,
erogeneous and homogeneous specimens micromechanically.
159–172.
Gong, J., Nie, Z., Zhu, Y., Liang, Z., Wang, X., 2019a. Exploring the effects of particle
This study makes a novel attempt to couple the RFM with DEM to shape and content of fines on the shear behavior of sand-fines mixtures via the DEM.
explore the effect of heterogeneity of equivalent particle properties on Comput. Geotech. 106, 161–176.
Gong, J., Zou, J., Zhao, L., Li, L., Nie, Z., 2019b. New insights into the effect of inter-
variability of soil properties. The statistical characteristics of four in­ particle friction on the critical state friction angle of granular materials. Comput.
dependent random field models for spatially varying particle properties Geotech. 113, 103105. https://doi.org/10.1016/j.compgeo.2019.103105.
are chosen based on empirical values to conduct a parametric study. The Gu, X., Zhang, J., Huang, X., 2020. DEM analysis of monotonic and cyclic behaviors of
sand based on critical state soil mechanics framework. Comput. Geotech. 128,
relevant data might be obtained by the emerging non-intrusive soil tests 103787. https://doi.org/10.1016/j.compgeo.2020.103787.
aided with X-ray μCT technique, which has been adopted by many re­ Guo, N., Zhao, J., 2013. The signature of shear-induced anisotropy in granular media.
searchers to image particle size or shape characteristics and even iden­ Comput. Geotech. 47, 1–15.
Huang, X., Hanley, K.J., O’Sullivan, C., Kwok, C.Y., 2014. Exploring the influence of
tify particle minerals (Fonseca et al., 2012; Hurley et al., 2017; Rorato interparticle friction on critical state behaviour using DEM. Int. J. Numer. Anal.
et al., 2019; Nadimi et al., 2020; Cano et al., 2021). The more physical Meth. Geomech. 38 (12), 1276–1297.
RDEM studies will be pursued in our future efforts. Itasca Consulting Group Inc., 2014. PFC — Particle Flow Code, Ver. 5.0. Minneapolis:
Itasca. Available from https://www.itascacg.com/software-faq.
Hurley, R.C., Hall, S.A., Wright, J.P., 2017. Multi-scale mechanics of granular solids from
CRediT authorship contribution statement grain-resolved X-ray measurements. Proc. R. Soc. A: Math. Phys. Eng. Sci. 473
(2207), 20170491.
Iwashita, K., Oda, M., 1999. Mechanics of granular materials: an introduction. CRC Press.
Jia-Yan Nie: Conceptualization, Methodology, Validation, Writing – Jamiolkowski, M., Kongsukprasert, L., & Lo Presti, D. C. (2004). Characterization of
original draft. Yi-Fei Cui: Writing – review & editing, Funding acqui­ gravelly geomaterials. Proceedings of the fifth international geotechnical
sition. Zhi-Yong Yang: Conceptualization, Methodology, Validation, conference. Bangkok 22-26 November 2004 (Vol. 2, pp. 29-56).
Jiang, M.D., Yang, Z.X., Barreto, D., Xie, Y.H., 2018. The influence of particle-size
Writing–review & editing. Yan-Zhou Yin: Writing – review & editing. distribution on critical state behavior of spherical and non-spherical particle
Zi-Jun Cao: Writing – review & editing. Dian-Qing Li: Writing – review assemblies. Granular Matter 20 (4), 1–15.
& editing. Jiang, M.J., Yu, H.-S., Harris, D., 2005. A novel discrete model for granular material
incorporating rolling resistance. Comput. Geotech. 32 (5), 340–357.
Kawamoto, R., Andò, E., Viggiani, G., Andrade, J.E., 2016. Level set discrete element
method for three-dimensional computations with triaxial case study. J. Mech. Phys.
Declaration of Competing Interest Solids 91, 1–13.
Li, K.Q., Li, D.Q., Liu, Y., 2020. Meso-scale investigations on the effective thermal
conductivity of multi-phase materials using the finite element method. Int. J. Heat
The authors declare that they have no known competing financial Mass Transf. 151, 119383.
interests or personal relationships that could have appeared to influence Liang, W., Zhao, J., 2019. Multiscale modeling of large deformation in geomechanics.
Int. J. Numer. Anal. Meth. Geomech. 43 (5), 1080–1114.
the work reported in this paper. Ma, T., Zhang, K., Shen, W., Guo, C., Xu, H., 2021. Discontinuous and continuous
Galerkin methods for compressible single-phase and two-phase flow in fractured
porous media. Adv. Water Resour. 156, 104039.
Acknowledgement
MiDi, G.D.R., 2004. On dense granular flows. Eur. Phys. J. E 14 (4), 341–365.
MINH, N.H., CHENG, Y.P., 2013. A DEM investigation of the effect of particle-size
The authors would like to thank three anonymous reviewers for their distribution on one-dimensional compression. Géotechnique 63 (1), 44–53.
Mollon, G., Zhao, J., 2013. Generating realistic 3D sand particles using Fourier
insightful and constructive comments that help to greatly improve the
descriptors. Granular Matter 15 (1), 95–108.
quality of the manuscript. This study was funded by the National Natural Nadimi, S., Fonseca, J., Andò, E., Viggiani, G., 2020. A micro finite-element model for
Science Foundation of China (42077238, 41941019, 52109144), the soil behaviour: experimental evaluation for sand under triaxial compression.
Open Innovation Fund of Changjiang Institute of Survey, Planning, Géotechnique 70 (10), 931–936.
Nguyen, D.H., Azéma, E., Sornay, P., Radjai, F., 2015. Effects of shape and size
Design and Research (CX2020K07) and the Second Tibetan Plateau polydispersity on strength properties of granular materials. Phys. Rev. E 91 (3),
Scientific Expedition and Research Program (2019QZKK0903). The 032203.
financial support is gratefully acknowledged. Nie, J.Y., Cao, Z.J., Li, D.Q., Cui, Y.F., 2021. 3D DEM insights into the effect of particle
overall regularity on macro and micro mechanical behaviors of dense sands.
Comput. Geotech. 132 (6), 103965.
References Nie, J.Y., Li, D.Q., Cao, Z.J., Zhou, B., Zhang, A.J., 2020. Probabilistic characterization
and simulation of realistic particle shape based on sphere harmonic representation
and Nataf transformation. Powder Technol. 360, 209–220.
Aboul, H.R., Sibille, L., Benahmed, N., Chareyre, B., 2017. Discrete numerical modeling
O’Sullivan, C., 2011. Particulate discrete element modelling: a geomechanics
of loose soil with spherical particles and inter-particle rolling friction. Granular
perspective. CRC Press.
Matter 19 (1), 1–12.
Phoon, K.K., Kulhawy, F.H., 1999. Evaluation of geotechnical property variability. Can.
Antony, S.J., Moreno-Atanasio, R., Hassanpour, A., 2006. Influence of contact stiffnesses
Geotech. J. 36 (4), 625–639.
on the micromechanical characteristics of dense particulate systems subjected to
Qi, X.H., Li, D.Q., 2018. Effect of spatial variability of shear strength parameters on
shearing. Appl. Phys. Lett. 89 (21), 214103. https://doi.org/10.1063/1.2396894.
critical slip surfaces of slopes. Eng. Geol. 239, 41–49.
Barreto, D., O’Sullivan, C., 2012. The influence of inter-particle friction and the
Ren, J., Li, S., He, H., Senetakis, K., 2021. The tribological behavior of iron tailing sand
intermediate stress ratio on soil response under generalized stress conditions.
grain contacts in dry, water and biopolymer immersed states. Granul. Matter 23 (1),
Granular Matter 14 (4), 505–521.
1–23.
Cano, A.C., Van Stappen, J.F., Wolterbeek, T.K., Hangx, S.J., 2021. Uniaxial compaction
Rorato, R., Arroyo, M., Andò, E., & Gens, A., 2019. Sphericity measures of sand grains.
of sand using 4D X-ray tomography: the effect of mineralogy on grain-scale
Eng. Geol., 254, 43-53.
compaction mechanisms. Mater. Today Commun. 26, 101881.
Rorato, R., Arroyo, M., Andò, E., Gens, A., Viggiani, G., 2020. Linking shape and rotation
Coetzee, C.J., Els, D.N.J., 2009. Calibration of discrete element parameters and the
of grains during triaxial compression of sand. Granul. Matter 22 (4), 1–21.
modelling of silo discharge and bucket filling. Comput. Electron. Agric. 65 (2),
Rorato, R., Arroyo, M., Gens, A., Andò, E., & Viggiani, G. (2018). “Particle shape
198–212.
distribution effects on the triaxial response of sands: a DEM study.” In Micro to
Craig, R.F., 2004. Craig’s soil mechanics, seventh edition. CRC Press.
Macro Mathematical Modelling in Soil Mechanics, pp. 277–286. Birkhäuser, Cham,
Cundall, P.A., 1971. A computer model for simulating progressive, large-scale movement
2018.
in blocky rock system. In Proceedings of the International Symposium on Rock
Rorato, R., Arroyo, M., Gens, A., Andò, E., Viggiani, G., 2021. Image-based calibration of
Mechanics, 1971.
rolling resistance in discrete element models of sand. Comput. Geotech. 131,
Cundall, P.A., Strack, O.D.L., 1979. A discrete numerical model for granular assemblies.
103929. https://doi.org/10.1016/j.compgeo.2020.103929.
Geotechnique 29 (1), 47–65.
Roux, J. & Combe, G., 2010. How granular materials deform in quasistatic conditions, in:
Fenton, G.A., Griffiths, D.V., 2008. Risk Assessment in Geotechnical Engineering. John
J.D. Goddard, J.T. Jenkins, P.G. (Ed.), IUTAM-ISIMM Symposium on Mathematical
Wiley & Sons.
Modeling and Physical Instances of Granular Flow, AIP Conference Proceedings (vol.
Fleischmann, J.A., Plesha, M.E., Drugan, W.J., 2013. Quantitative comparison of two-
1227). p. 260.
dimensional and three-dimensional discrete-element simulations of nominally two-
Sandeep, S., Senetakis, K., 2018a. Effect of young’s modulus and surface roughness on
dimensional shear flow. Int. J. Geomech. 13 (3), 205–212.
the inter-particle friction of granular materials. Materials 11 (2), 217.
Fonseca, J., O’sullivan, C., Coop, M.R., Lee, P.D., 2012. Non-invasive characterization of
particle morphology of natural sands. Soils Found. 52 (4), 712–722.

13
J.-Y. Nie et al. Computers and Geotechnics 141 (2022) 104497

Sandeep, C.S., Senetakis, K., 2018b. Grain-scale mechanics of quartz sand under normal Wang, X., Yin, Z.Y., Xiong, H., Dong, S., Feng, Y.T., 2021. A spherical-harmonic-based
and tangential loading. Tribol. Int. 117, 261–271. approach to discrete element modeling of 3D irregular particles. Int. J. Numer. Meth.
Schneider-Muntau B., Medicus G., Desrues J., Andò E., & Viggiani G., 2021. Investigation Eng. https://doi.org/10.1002/nme.6766.
of Uncertainty in Strength Parameter Identification. In: Barla M., Di Donna A., Sterpi Yang, J., Luo, X.D., 2015. Exploring the relationship between critical state and particle
D. (eds) Challenges and Innovations in Geomechanics. IACMAG 2021. shape for granular materials. J. Mech. Phys. Solids 84, 196–213.
Schwiteilo, E., Herle, I., 2016. Vergleichsstudie zur Kompressibilität und zu den Yang, J., Luo, X.D., 2018. The critical state friction angle of granular materials: does it
Scherparametern von Ton aus Ödometer- und Rahmenscherversuchen. Geotechnik depend on grading? Acta Geotech. 13 (3), 535–547.
40, 204–217. Yang, Z., Ching, J., 2021. Simulation of three-dimensional random field conditioning on
Senetakis, K., Coop, M.R., Todisco, M.C., 2013. The inter-particle coefficient of friction at incomplete site data. Eng. Geol. 281, 105987. https://doi.org/10.1016/j.
the contacts of leighton buzzard sand quartz minerals. Soils Found. 53 (5), 746–755. enggeo.2020.105987.
Sibille, L., Villard, P., Darve, F., Aboul Hosn, R., 2019. Quantitative prediction of discrete Yang, Z.Y., Cao, Z.J., Li, D.Q., Phoon, K.K., 2017. Effect of spatially variable friction
element models on complex loading paths. Int. J. Numer. Anal. Meth. Geomech. 43 coefficient of granular materials on its macro-mechanical behaviors using biaxial
(5), 858–887. compression numerical simulation. Eng. Mech. 34 (5), 235–246 (in Chinese).
Vangla, P., Latha, G.M., 2015. Influence of particle size on the friction and interfacial Yang, Z., Nie, J., Peng, X., Tang, D., Li, X., 2021. Effect of random field element size on
shear strength of sands of similar morphology. Int. J. Geosynthet. Ground Eng. 1 (1), reliability and risk assessment of soil slopes. Bull. Eng. Geol. Environ. 80 (10),
6. 7423–7439. https://doi.org/10.1007/s10064-021-02422-z.
Vanmarcke, E.H., 1977. Probabilistic modeling of soil profiles. J. Geotech. Eng. Divis. Zhao, T., Liu, Y., 2020. A novel random discrete element analysis of rock fragmentation.
103 (11), 1227–1246. Int. J. Numer. Anal. Meth. Geomech. 44 (10), 1386–1395.
Wang, R., Dafalias, Y.F., Fu, P., Zhang, J.M., 2020. Fabric evolution and dilatancy within Zhou, B., Huang, R., Wang, H., Wang, J., 2013. DEM investigation of particle anti-
anisotropic critical state theory guided and validated by DEM. Int. J. Solids Struct. rotation effects on the micromechanical response of granular materials. Granular
188-189, 210–222. Matter 15 (3), 315–326.

14

You might also like