Computers and Geotechnics 128 (2020) 103787

Contents lists available at ScienceDirect

Computers and Geotechnics

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

Research Paper

DEM analysis of monotonic and cyclic behaviors of sand based on critical T

state soil mechanics framework
Xiaoqiang Gu, Jiachen Zhang, Xin Huang
⁎

Department of Geotechnical Engineering & Key Laboratory of Geotechnical and Underground Engineering of the Ministry of Education, Tongji University, Shanghai, China

ARTICLE INFO ABSTRACT

Keywords: A series of monotonic and cyclic triaxial shearing tests on Toyoura sand are simulated by discrete element
DEM method (DEM) incorporating rolling resistance. The input parameters are calibrated against laboratory triaxial
Rolling resistance test data considering different initial states and monotonic loading paths. The calibrated DEM model successfully
Critical state captures the density and stress dependency of sand behavior under both monotonic and cyclic conditions. It is
Coordination number
shown that when approaching liquefaction, contact yielding alters from sliding-dominant to rolling-dominant.
Liquefaction
The expression of mechanical redundancy index (IR) is derived considering both effects of contact sliding and
rolling. The IR = 1.0 condition is shown to be a unified criterion distinguishing between solid state and liquid
state upon both monotonic and cyclic liquefactions. A unique micro critical state line (CSL) is obtained in MCN-p′
plane. The cyclic resistance ratio is exponentially correlated with both the macro state parameter (ψe0) derived
from the e-p′ CSL and the micro state parameter (ψMCN0) determined from the MCN-p′ CSL. The specimens with
the same ψe0 may have distinct cyclic behaviors, while cyclic behaviors of specimens with the same ψMCN0 are
close to each other, indicating that ψMCN0 is a more plausible state variable for characterizing the behavior of
sand than ψe0.

1. Introduction which it deforms continuously with a constant stress state and void
ratio. Within the CSSM framework, Been and Jefferies (1985) used
Liquefaction is a phenomenon wherein a saturated sand loses shear critical state as a reference state and defined the distance between the
resistance and changes its behavior from solid-like to liquid-like (Castro initial void ratio and the void ratio at critical state with the same ef­
and Poulos, 1977). The liquefaction of saturated sand can lead to slope fective stress as state parameter, which embraces the effects of both
failure, building collapse, ground subsidence, road destruction, etc. void ratio and confining pressure. The critical state approach has also
Therefore, it is of great significance to study the liquefaction char­ been employed in the cyclic liquefaction resistance analysis (Boulanger,
acteristics of sand. Most of the fundamental understandings of sand 2003; Yang and Sze, 2011). However, the shear strain that can be
behaviors were derived from well-controlled laboratory tests (Miura reached is usually limited in laboratory tests. Meanwhile, due to uneven
and Toki, 1982; Vaid and Chern, 1983; Vaid et al., 1985; Tatsuoka sample preparation and rubber membrane effect, the samples are prone
et al., 1986; Toki et al., 1986; Yoshimine et al., 1998; Vaid et al., 2001). to strain localization after large deformation.
When loading is cyclic rather than monotonic, the responses of sand are On the other hand, sand is a type of granular assembly composed of
more complex. The liquefaction resistance essentially depends on a large number of discrete particles. From a microscopic perspective,
packing density and confining stress. However, the experimental ob­ macroscopic behaviors are closely associated with microstructure evo­
servations about how initial state affects the cyclic liquefaction re­ lution. However, it is difficult to obtain the evolution of microstructure
sistance are controversial (Castro and Poulos, 1977; Vaid et al., 2001; in laboratory tests. Discrete element method (DEM) developed by
Seed and Lee, 1966; Sze, 2010). Cundall and Strack (1979) provides an ideal tool to analyze the me­
The critical state soil mechanics (CSSM) has been recognized as a chanical behavior and the microstructure of granular materials sub­
useful unified conceptual framework to model and understand the be­ jected to various loading conditions (Ng and Dobry, 1994; Sitharam,
haviors of soils (Roscoe et al., 1958; Schofield and Wroth, 1968; Been 2003; Dinesh et al., 2004; O'Sullivan et al., 2008; Wei and Wang, 2015;
et al., 1991; Verdugo and Ishihara, 1996; Yang and Li, 2004). Re­ Nguyen et al., 2016; Gong et al., 2019). Characteristics of micro­
gardless of the initial state, soil will reach a unique critical state, at structure at critical state have also been extensively investigated by

⁎
Corresponding author.
E-mail address: [email protected] (X. Huang).

https://doi.org/10.1016/j.compgeo.2020.103787
Received 11 March 2020; Received in revised form 21 June 2020; Accepted 10 August 2020
0266-352X/ © 2020 Elsevier Ltd. All rights reserved.
X. Gu, et al. Computers and Geotechnics 128 (2020) 103787

previous researchers. For example, Gu et al. (2014) found that the co­
ordination number and anisotropies at critical state depend on the ef­
(m)
fective stress. Nguyen et al. (2018) also found that the coordination
number reaches a constant value when approaching critical state.
Huang et al. (2014) indicated that there is a clear linear correlation
between micro-scale state parameter of coordination number and
macro-scale state parameter of void ratio for drained simulations.
DEM has also been used for cyclic liquefaction simulations and the

9.0 mm
evolution of microstructure is investigated pre- and post-liquefaction.
For example, Xu et al. (2015) explored the micromechanics of lique­
faction resistance and its correlation with the shear wave velocity of
granular soils. Wei and Wang (2017) analyzed the influence of initial
fabric on the liquefaction resistance and shear strain evolution of sands.
The limiting coordination numbers that can support a stable non-li­
quefaction state were suggested (Wang et al., 2016; Evans and Zhang,
2019). Despite these efforts, little attention has been paid to interpret
the cyclic liquefaction resistance within the CSSM framework from a
microscopic perspective, which is of importance to the geotechnical
community.
Furthermore, most of the above studies simplified sand particles as
frictional disks or spheres without considering the effect of rolling re­
sistance induced by irregular shape of sand grains. Oda et al. (1982)
pointed out that particle rolling is a major microscopic deformation Fig. 1. Specimen in DEM simulations (walls are omitted).
mechanism, especially when interparticle friction is large. Jiang et al.
(2015) found that the strength increases but the coordination number between two particles is greater than Fs*, sliding occurs. The rolling
decreases as rolling resistance increases, implying that greater rolling resistance moment Mr is accumulated incrementally with respect to the
resistance can improve the robustness of the force network. However, increment of relative rotation between contacting particles Δθ:
many studies only considered contact sliding and neglected rolling
(Gong et al., 2012; Thornton, 2015), while the interplay between Mr : =Mr kr (2)
rolling and sliding at the contacts has been shown to be significant where the rolling stiffness kr is linked to the shear stiffness ks by
(Huang et al., 2014). k r = k s R̄2 , in which R̄ is the effective contact radius given by:
In this study, DEM simulations are conducted to investigate the
1 1 1
state-dependent behaviors of sand under both the monotonic and cyclic = +
R¯ R1 R2 (3)
triaxial loading conditions. A rolling resistance linear contact model is
employed to consider the influence of particle shape on the mechanical where R1 and R2 are the radii of two contacting particles. This accu­
behavior of sand. In order to better capture the real macroscopic re­ mulated rolling resistance moment is limited by its maximum value M*:
sponses of sand, the input contact parameters are calibrated under
¯ n|
|M *| = |µr RF (4)
different drainage conditions. The evolution of mechanical stability is
explored considering both contact sliding and rolling. Both the macro Similar to the tangential contact force, rolling occurs when Mr be­
and micro critical state lines (CSLs) are identified, based on which the tween two particles is greater than the limiting value.
macro and micro state parameters are defined. The mechanical beha­
viors of sand under both monotonic and cyclic loading conditions are 2.2. Specimen generation
linked to the state parameters within the CSSM framework.
The specimen is a 4.5 × 4.5 × 9.0 mm3 cuboid bounded by six rigid
2. DEM model walls, as shown in Fig. 1. Spherical particles coated by different colours
scaled by the radius are used to represent the sand grains. As shown in
2.1. Rolling resistance linear contact model Fig. 2, the particle-size distribution curve of Toyoura sand is simulated,

The DEM simulations are performed using PFC3D (Itasca Consulting 100
Group, 2015). Many studies based on linear contact model used sim­ Experiment
plified particle shapes, such as disks or spheres (Huang et al., 2014;
80 DEM
Sitharam and Vinod, 2009; Hu et al., 2010), which leads to excessive
rolling as compared to real granular materials. Iwashita and Oda (1998)
Percentage (%)

indicated that rolling is the dominant micro deformation mechanism 60

controlling the strength and dilatancy of granular soils. In order to
consider the influence of sand particle shape, a rolling resistance linear 40
contact model is adopted. Compared with the common linear contact
model, this rolling resistance linear contact model incorporates a rolling
20
contact component in addition to the normal and tangential contact
components. The main micro contact parameters in this model involve
a normal stiffness kn, a shear stiffness ks, a friction coefficient μ and a 0
rolling resistance coefficient μr. The limit of tangential contact force Fs* 0 .0 1 0 .1 1
is given by: Particle diameter (mm)
|Fs*| = µ |Fn| (1) Fig. 2. Particle size distributions of Toyoura sand in experiment and DEM si­
where Fn is the normal contact force. When the tangential contact force mulations.

2
X. Gu, et al. Computers and Geotechnics 128 (2020) 103787

Table 1 1.0
Input parameters in DEM simulations. μ=0 μ=0.1 μ=0.2 μ=0.5
Property Value 0.9

Number of particles 44,488

Particle density 2640 kg/m3 0.8
Wall stiffness 1 × 1010 N/m

e
Wall-particle friction coefficient 0
Interparticle friction coefficient (μ) 0.45 0.7
Particle normal stiffness (kn) kn = k0 × r
Particle shear stiffness (ks) ks = α × kn
Local damping 0.7
0.6
Rolling resistance coefficient (μr) 0.5

0.5
Note: k0 = 8 × 108 N/m2, α = 0.15, and r is the radius of the particle.
0.0 0.2 0.4 0.6 0.8 1.0

which has a mean grain size (d50) of 0.16 mm and a uniform coefficient μr
(Cu) of 1.46 (Fukushima and Tatsuoka, 1984). Note that the particles
(a)
with diameter below 0.1 mm are removed for computational efficiency.
Table 1 lists the input parameters in DEM simulations. The specimen is 1.0
generated by radius expansion method. Firstly, 44,488 non-overlapping μr=0
μr=0 μr=0.1
r =0.1 μr=0.2
μr=0.2 μr=0.5
r =0.5
particles at half of their diameters are randomly generated in the space 0.9
enclosed by rigid walls, which are expanded to their target sizes. The
specimen is cycled to equilibrium state. During these stages, the friction
coefficient μ and rolling resistance coefficient μr are set to zero. Then 0.8
the specimen is compressed until an isotropic stress state of 10 kPa is

e
reached. Specimens with different void ratios can be generated under 0.7
different combinations of input contact parameters (Gu et al., 2017).
The void ratio (e) increases as μ and μr increase, and their influences on
0.6
e are interrelated. As shown in Fig. 3a, when μ is small, the influence of
μr on e is not obvious, while μr has a more significant influence on e as μ
further increases. The effect of μ on e is similar to that of μr (Fig. 3b). 0.5
0.0 0.2 0.4 0.6 0.8 1.0
The rolling resistance causes arching structure, which thereby leads to
larger voids (Iwashita and Oda, 1998). Fig. 3 also reveals that both the μ
influences of μ and μr on packing density are limited. Later, μ and μr are
set to the values calibrated for Toyoura sand before shearing. (b)
Fig. 3. The relationships among e, μ and μr (p′ = 10 kPa): (a) the relationships
2.3. Loading method between e and μr at different μ values; (b) the relationships between e and μ at
different μr values.
In this study, both monotonic and cyclic triaxial tests are simulated.
The mean effective stress p′, deviatoric stress q and pore pressure u are As shown in Table 2, different methods have been used to simulate
defined as (Huang et al., 2018): undrained cyclic tests using DEM, which can be classified into three
p = ( xx + yy + zz )/3 (5) groups: 1) stress-controlled method, in which the volume and stress are
controlled simultaneously. Such a servo control scheme is closest to the
q= zz ( xx + yy ) /2 (6) stress-controlled undrained cyclic loading test which is the most
common approach to studying the cyclic soil liquefaction behaviors.
u=p p0 (7) Representative researches adopting such a servo scheme include Xu
et al. (2015) and Huang et al. (2018). However, when approaching li­
where σ′xx, σ′yy, σ′zz are the average effective normal stresses along the
quefaction, the target stress is difficult to maintain due to the disin­
x, y and z directions, and p′0 is the initial mean effective stress (initial
tegration of force transmission network and drastic reduction in contact
confining pressure) after isotropic compression.
forces. Therefore, abrupt failure is usually observed under such a servo
Both drained and undrained monotonic simulations are performed.
control scheme; 2) strain-controlled method which varies the axial
In the drained simulations, the top and bottom walls are moved verti­
strain in a sinusoidal manner. The loading direction is reversed when
cally at the same velocity, while the confining pressure is maintained
the strain amplitude has been reached. The lateral strain rate can be
constant by continuously adjusting the positions of the lateral walls. In
derived from the axial strain rate based on the constant-volume as­
the undrained simulations, the positions of the lateral walls are adjusted
sumption. Liquefaction is considered to occur when the effective stress
continuously in accordance with the movement of top and bottom walls
decreases to zero. Typical examples using such a strain-controlled
so that the sample volume remains constant. This is a widely used ap­
scheme can be referred to Sitharam and Vinod (2008) and Zhang and
proach of simulating undrained behavior of saturated sand using DEM,
Evans (2018). The deformation is strictly controlled under such a
which has been shown by (Shafipour and Soroush, 2008) amongst
loading scheme, which can avoid the abrupt collapse even when the
others to be equivalent to fluid-coupled approach.
stiffness of the system is very low. However, some special liquefaction
The axial strain rate is set to be 1 s−1 for both drained and un­
phenomena, such as cyclic mobility, cannot be observed by this loading
drained monotonic simulations with a timestep of 1 × 10−7. Except
method; 3) Wang et al. (2016) and Wang and Wei (2016) proposed
when approaching liquefaction, the inertia number, I = di g / p
using a hybrid loading scheme, in which a constant strain rate is applied
(where is the strain rate, di is the average particle diameter, ρg is the
initially, and the loading direction is reversed when the deviatoric
particle density and p′ is the mean effective stress) (Da Cruz et al., 2005)
stress amplitude is attained. The loading direction is reserved again
remains smaller than the maximum allowable value 0.003 for quasi-
once the deviatoric stress amplitude in the opposite direction is
static state (Nguyen et al., 2016; Lopera Perez et al., 2016).

3
X. Gu, et al. Computers and Geotechnics 128 (2020) 103787

Table 2
Summary of cyclic loading methods.
Reference Contact model Particle shape Loading method

Da Cruz et al. (2005) Equivalent Hertzian with linear force 3D spherical particles Strain-controlled undrained cyclic triaxial (f = 1.42 Hz)
displacement
Xu et al. (2015) Hertz–Mindlin 3D spherical particles Stress-controlled undrained cyclic triaxial (f = 1000 Hz)
Wang et al. (2016) Linear 2D circular particles Hybridly-controlled undrained cyclic biaxial (low loading rate)
Lopera Perez et al. (2016) Hertz–Mindlin 2D circular particles Hybridly-controlled undrained cyclic simple shear (shear strain
rate = 0.01 s−1)
Sitharam and Vinod (2008) Hertz–Mindlin 3D spherical particles Stress-controlled undrained triaxial (f = 5 Hz)
Wang and Wei (2016) Linear 3D spherical particles Strain-controlled undrained cyclic triaxial and simple shear (f = 2 Hz)
This study Rolling resistance linear 3D spherical particles Hybridly-controlled undrained cyclic triaxial (axial strain
rate = 0.25 s−1)

reached. Such a loading cycle is repeated until liquefaction occurs or expansion method to generate DEM samples, which leads to an initial
the prescribed strain level has been attained. This loading scheme is fabric closer to moist tamping approach used by Verdugo and Ishihara
more stable than the stress-controlled method and could also capture (1996). Nevertheless, since spherical particles are used in the current
the variety of undrained cyclic loading behavior of sands including flow study, the difference in initial anisotropy induced by different sample
deformation, cyclic mobility and plastic strain accumulation. Con­ preparation methods is negligible. Furthermore, the DEM simulations
sidering its merits, we use the hybrid loading method in this study by adopt a height-to-width ratio of 2 when generating the sample, which is
applying a constant axial strain rate to the top and bottom walls. The consistent with the height-to-diameter ratio used in experiments. The
loading direction is reversed when the cyclic deviatoric stress qcyc final calibrated input contact parameters are as follows: k0 = 8 × 108
reaches the cyclic deviatoric stress amplitude qmax, during which the N/m2, α = 0.15, μ = 0.45, μr = 0.5. And emax = 0.839 (when μ = 0.45,
specimen volume remains unchanged. The sensitivity analysis on strain μr = 0.5), emin = 0.555 (when μ = 0, μr = 0). In total, calibration was
rates (0.1–1 s−1) indicates that when the strain rate is smaller than conducted based on 12 simulations considering different relative den­
0.25 s−1, the impact of strain rate on cyclic behaviors can be ignored. sities and confining pressures under both drained and undrained con­
Therefore, a strain rate of 0.25 s−1 is used under cyclic conditions in the ditions. Due to the space limit, we only show 6 comparison cases be­
current study. tween the simulation results and experimental data in Fig. 4. Fig. 4(a)
and (b) compare the evolutions of deviatoric stress and volumetric
2.4. Contact model parameters calibration strain for DEM samples with a relative density of around 40% under
three confining stress levels (100, 200, 400 kPa) with experimental data
In DEM simulations, the input contact parameters determine the reported by Fukushima and Tatsuoka (1984), and Fig. 4(c) and (d)
macroscopic mechanical behaviors of sand. Some parameters have a compare the undrained simulation results on samples with relative
coupled effect on the macroscopic behaviors (Widuliński et al., 2009; densities of around 40% under three confining pressures (100, 1000,
Yan and Dong, 2011; Estrada et al., 2013; Hosn et al., 2017). Therefore, 2000 kPa) with experimental data conducted by Verdugo and Ishihara
it is necessary to calibrate the input contact parameters under different (1996). The calibration data for the remaining 6 simulations are shown
drainage conditions. Monotonic triaxial simulations have been used in the Supplements (see Figs. S1 and S2). It can be seen that the drained
effectively to calibrate DEM parameters against the mechanical re­ simulation results and the undrained simulation results at low packing
sponses of soils (Belheine et al., 2009; Plassiard et al., 2009; Syed et al., density and confining pressures are in good agreement with the ex­
2017). Since the emax and emin in DEM simulations are different from the perimental data of Fukushima and Tatsuoka (1984); whereas the dif­
actual void ratios of Toyoura sand, relative density Dr is used for ferences between undrained simulation results and experimental data
comparison in the parameter calibration. The relative density Dr is of Verdugo and Ishihara (1996) at high packing densities and confining
defined as: pressures are more noticeable. The reasons for these discrepancies can
emax e be multi-fold. Firstly, it is very difficult to prepare the DEM samples
Dr = with exactly the same relative density as that of experimental speci­
emax emin (8)
mens; secondly, the discrepancies may also be caused by the difference
The sample is isotropically compressed until p′ reaches 10 kPa with in initial fabric induced by different sample preparation methods be­
a reference stiffness k0 = 6 × 108 N/m2 (kn = k0 × r) and a stiffness tween DEM simulations and experiments; thirdly, we attributed the
ratio α = 0.25 (ks = α × kn). The emin in DEM is defined as the void more noticeable discrepancies at high stress levels to contact yielding
ratio at this stage exempt from friction and rolling resistances. Since the and particle crushing, both of which are very likely to occur at high
emax in DEM is affected by input contact parameters, i.e., the same emax stress levels but are not considered in the current study. Note that if the
can be obtained under different combinations of input parameters, this current model is used to simulate the behavior of a different type of
study determines emax by trial and error. The simulation results are sand, the calibration process should be reconducted.
compared with the experimental data. The parameters (k0, α, μ, μr) are
adjusted until the final parameters (μ, μr) yield close matches between
the experimental data and simulation results at different relative den­ 3. Simulation results
sities corresponding to the same emax. The physical properties of
Toyoura sand used by Fukushima and Tatsuoka (1984) are The simulation ID sequentially consists of initial confining pressure,
d50 = 0.16 mm, Cu = 1.46, Gs = 2.64, emax = 0.977, emin = 0.605, relative density and loading mode which are separated by dashes. As for
while those are d50 = 0.17 mm, Cu = 1.7, Gs = 2.65, emax = 0.977, loading mode, ‘CDC’ and ‘CUC’ respectively denote isotropically-con­
emin = 0.597 for Toyoura sand used by Verdugo and Ishihara (1996). solidated drained and undrained compression triaxial simulations,
Considering that the differences between the two sets of Toyoura sand while ‘cyc’ indicates isotropically-consolidated undrained cyclic triaxial
are negligible, we adopt the former set of physical properties in our simulations. For cyclic triaxial simulations, an additional indicator is
DEM simulations. Air pluviation method is used in Fukushima and given after ‘cyc’ to present the cyclic stress ratio. For example, one can
Tatsuoka (1984) to generate testing sample, while moist tamping ap­ easily interpret that 100-40%-CDC corresponds to the drained experi­
proach was adopted by Verdugo and Ishihara (1996). We use radius ment/simulation on a sample with a relative density of 40% under a

4
X. Gu, et al. Computers and Geotechnics 128 (2020) 103787

Fig. 4. Calibration results: (a) deviatoric stress (drained); (b) volumetric strain (drained); (c) deviatoric stress (undrained); (d) stress path (undrained).

Table 3 3.1. Monotonic behaviors

Summary of monotonic and cyclic triaxial simulations.
Simulation ID Dr0 p′0 (kPa) e0 MCN0 Loading ψe0 ψMCN0
3.1.1. Macro responses and critical state
mode Fig. 5 shows the critical state lines (CSLs) in both q-p′ and e-p′ planes
determined from drained and undrained triaxial tests. The CSLs do not
50-49%-CDC 49% 50 0.701 4.437 CDC −0.103 0.833 depend on the initial states and drainage conditions. Note that the
100-11%-CDC 11% 100 0.808 4.047 CDC 0.007 0.340
100-50%-CDC 50% 100 0.698 4.431 CDC −0.102 0.838
specimen 100-11%-CUC and 100-20%-CUC show static liquefaction,
200-49%-CDC 49% 200 0.699 4.872 CDC −0.096 1.025 which are not displayed in Fig. 5. As seen in Fig. 5a, although higher
400-48%-CDC 48% 400 0.704 4.943 CDC −0.080 0.904 confining pressure in drained tests and larger relative density in un­
800-50%-CDC 50% 800 0.697 5.115 CDC −0.066 0.814 drained tests cause higher strength, the stress ratio q/p′ tends to con­
100-11%-CUC 11% 100 0.808 4.047 CUC 0.007 0.340
verge to about 1.16 at critical state, which is basically consistent with
100-20%-CUC 20% 100 0.781 4.160 CUC −0.020 0.453
100-30%-CUC 30% 100 0.754 4.355 CUC −0.047 0.647 the experimental value 1.25 (Sze, 2010; Verdugo and Ishihara, 1996).
100-50%-CUC 50% 100 0.698 4.545 CUC −0.102 0.838 As shown in Fig. 5b, almost all the specimens contract initially and they
100-69%-CUC 69% 100 0.643 5.256 CUC −0.158 1.549 finally reach a consistent relation of e and p′ at critical state. The e-p′
800-50%-CUC 50% 800 0.697 5.115 CUC −0.066 0.814 curves in DEM simulations are slightly lower than that in experiments,
20-11%-cyc 11% 20 0.807 3.841 cyc 0.001 0.332
50-10%-cyc 10% 50 0.812 3.920 cyc 0.008 0.316
which indicates that the particle shape effect on packing cannot be fully
50-49%-cyc 49% 50 0.701 4.437 cyc −0.103 0.833 compensated by the rolling resistance model. This may be because the
100-11%-cyc 11% 100 0.808 4.047 cyc 0.007 0.340 particles are essentially spherical. The rolling resistance model is only a
100-30%-cyc 30% 100 0.754 4.355 cyc −0.047 0.647 mathematical and numerical approximation of the constraints to rota­
100-50%-cyc 50% 100 0.698 4.545 cyc −0.102 0.838
tional motion induced by particle’s irregularity, which cannot fully
100-53%-cyc 53% 100 0.687 4.844 cyc −0.113 1.137
100-64%-cyc 64% 100 0.658 5.055 cyc −0.143 1.348 capture the void structure formed by real irregular particles. This may
100-69%-cyc 69% 100 0.643 5.256 cyc −0.158 1.549 be overcome by substituting spheres with clumps composed of adequate
300-11%-cyc 11% 300 0.809 4.158 cyc 0.019 0.205 sub-spheres that accurately capture the surface profile and texture of
300-50%-cyc 50% 300 0.698 4.835 cyc −0.091 0.882 real sand grains. The relationships between e and p′ at the critical state
in Fig. 5b can be represented by a power function as:

confining pressure of 100 kPa and 100-40%-CUC corresponds to the p

0.9

undrained experiment/simulation on a sample with a relative density of e = 0.8077 + 0.0069

pa (9)
40% under a confining pressure of 100 kPa. Simulation details are given
in Table 3. where pa is the atmospheric pressure (101 kPa). Gu et al. (2014) and
Huang et al. (2014) found that the critical void ratio varies linearly with

5
X. Gu, et al. Computers and Geotechnics 128 (2020) 103787

5000 6.0
300
50-49%-CDC 100-11%-CDC
200
4000 5.5 100-50%-CDC 200-49%-CDC
100 400-48%-CDC 800-50%-CDC
q/p' = 1.16
0 5.0
3000 0 100 200 300
q (kPa)

MCN
4.5
2000
50-49%-CDC 100-11%-CDC 4.0
100-50%-CDC 200-49%-CDC
1000 400-48%-CDC 800-50%-CDC
100-30%-CUC 100-50%-CUC 3.5
100-69%-CUC 800-50%-CUC
0
3.0
0 1000 2000 3000 4000 5000
0 10 20 30 40 50
p' (kPa)
εa (%)
(a)
(a)
0.9

e = 0.8077 - 0.0069 (p'/101) 0.9

6.0
R2 = 0.9987
0.8
5.5

0.7 5.0
e

MCN 4.5
50-49%-CDC 100-11%-CDC
0.6 100-50%-CDC 200-49%-CDC
400-48%-CDC 800-50%-CDC 4.0
100-30%-CUC 100-50%-CUC
100-69%-CUC 800-50%-CUC 100-11%-CUC 100-20%-CUC
0.5 3.5 100-30%-CUC 100-50%-CUC
10 100 1000 10000 100-69%-CUC 800-50%-CUC
p' (kPa) 3.0
0 10 20 30 40 50
(b)
εa (%)
Fig. 5. Critical state lines in: (a) q – p′ plane; (b) e – p′ plane (the solid squares
mark the initial state and the hollow squares indicate the critical state). (b)
Fig. 6. Evolutions of mechanical coordination number in: (a) drained tests; (b)
p′ and (p′)0.7. Sze (2010) and Verdugo and Ishihara (1996) found that undrained tests.
the power of stress variable was approximately 0.6 in laboratory tests.
Whereas, a power of 0.9 is used herein as it gives the closest correlation constant values during shearing in drained tests. Although the specimen
with simulation data. 100-11%-CDC is contractive, which means the soil becomes denser
during shearing, MCN decreases consistently and the value at critical
3.1.2. Microscopic interpretation state is smaller than the initial one. Generally, MCN increases with
These above behaviors indicate that our current DEM model can increasing confining pressure. As shown in Fig. 6b, the two loosest
effectively capture the monotonic behaviors of sand. Apart from mac­ specimens 100-11%-CUC and 100-20%-CUC liquefy completely.
roscopic behavior analyses, DEM also allows an examination of the Therefore, MCN drops abruptly to very small values, indicating the
evolutions of microstructure represented by several microscopic quan­ collapse of microstructure. For the remaining medium dense specimens,
tities. Due to the absence of gravity, a number of particles are floating MCN decreases at beginning and reaches the minimum at phase
and have no contribution to the overall force transmission during the transformation state, after which MCN increases again until approxi­
shearing process. These particles are usually referred to as rattlers or mately constant values are reached at critical state.
floaters. Thornton (2000) proposed the mechanical coordination Rothenburg and Kruyt (2004) indicated that shear deformations
number (MCN) which only considers particles with two or more con­ tend to destroy interparticle contacts and create locally unstable con­
tacts: figurations that regain stability by forming new contacts or restoring
old interparticle contacts. The rate of contact loss and that of contact
2Nc Np1 formation are equalized at critical state. There is a unique relationship
MCN =
Np Np0 Np1 (10) between MCN and p′ at critical state irrespective of initial states and
drainage conditions, as shown in Fig. 7. The relationship can also be
where Nc is the total number of contacts, Np is the total number of expressed by a power-law function as:
particles, Np1 is the number of particles with only one contact, and Np0
0.45
is the number of particles with no contact. MCN has been found to p
MCN = 3.32 + 0.39
correlate with many macro-scale quantities, including the small strain pa (11)
stiffness (Gu and Yang, 2013; Gu et al., 2017).
Fig. 6 shows the evolutions of MCN in drained and undrained tests. Similar to the definition of macro state parameter ψe (the distance
The initial MCN increases with increasing initial relative density and between the current void ratio and the critical-state void ratio eCS at the
confining pressure. As shown in Fig. 6a, the MCN values of the speci­ same p′ in e-p′ plane) (Been and Jefferies, 1985); this study defines the
mens with a confining pressure of 100 kPa continuously decrease to micro state parameter ψMCN as the distance between the current

6
X. Gu, et al. Computers and Geotechnics 128 (2020) 103787

7
50-49%-CDC 100-11%-CDC
100-50%-CDC 200-49%-CDC
400-48%-CDC 800-50%-CDC
6 100-30%-CUC 100-50%-CUC
100-69%-CUC 800-50%-CUC
MCN

4
MCN = 3.32 + 0.39 (p'/101) 0.45
R2 = 0.9994
3
10 100 1000 10000
p' (kPa)
Fig. 7. Critical state line of mechanical coordination number (the solid symbols
mark the initial state and the hollow symbols indicate the critical state).

5
MCN0

50kPa
100kPa
200kPa
4
400kPa
500kPa
800kPa
3
0.5 0.6 0.7 0.8 0.9
e0
(a)

2.0

1.5
ψMCN0

1.0

0.5
ψMCN0 = 14.38ψe0 2 - 4.89ψe0 + 0.34
R² = 0.9433
0.0
-0.3 -0.2 -0.1 0.0 0.1
ψe0
Fig. 9. Evolutions of sliding and rolling contact fractions: (a) 100-50%-CDC; (b)
(b) 100-50%-CUC; (c) 100-11%-CUC.

Fig. 8. Relationships between macro and micro state: (a) MCN0 – e0; (b) ψMCN0 –
ψe0. macroscopic and microscopic perspectives, respectively.
From the perspective of structural mechanics, a composite structure
could only maintain its stability when the number of constraints ex­
mechanical coordination number and the critical-state mechanical co­
ceeds the number of degrees of freedom (Thornton, 2015; Huang et al.,
ordination number MCNCS corresponding to the same p′. Fig. 8a reveals
2019; Martin et al., 2019). Considering a 3D system of particles with μ
that there are apparent correlations between e0 and MCN0 after iso­
= ∞ and μr = ∞; i.e., no contact sliding or rolling occurs, the number
tropic consolidation. This may be because MCN also depends on p′.
of degrees of freedom per particle is 6 (three translations and three
However, as seen in Fig. 8b, the relation between ψe0 and ψMCN0 is
rotations) and the number of constraints (unknown contact forces) per
unique, which is consistent with the DEM simulation result by Huang
contact is 5 (the normal force, two tangential force components and two
et al. (2014). These two state parameters can be used to analyze the
rolling resistance components). In this study, the friction and rolling
responses of sand under monotonic and cyclic loading from

7
X. Gu, et al. Computers and Geotechnics 128 (2020) 103787

2.5 calculating IR:

(5 2fs 2fr 4fsr ) Nc

2 IR =
6(Np Np0 ) (12)

1.5 where fs is the sliding contact fraction, fr is the rolling contact fraction,
fsr is the fraction of contacts that both the sliding and rolling limits are
IR

reached. The isostatic condition is IR = 1.0, i.e., the total number of

1 constraints equals to the total number of degrees of freedom.
100-50%-CDC Fig. 9 shows the evolutions of the fractions of sliding contacts (fs),
0.5 100-50%-CUC
rolling contacts (fr) and contacts where both the sliding and rolling
100-11%-CUC
limits are reached (fsr) during the monotonic simulations. Overall,
under both drained and undrained conditions, the number of contacts
0
only reaching the ‘plastic’ sliding limit is higher than that of contacts
0 10 20 30 40 50
only reaching the ‘plastic’ rolling limit. Both fs and fr are larger than fsr
εa (%) under drained condition (Fig. 9a). However, fr is almost the same as fsr
Fig. 10. Evolutions of mechanical redundancy index in monotonic simulations. under undrained condition (Fig. 9b). The very loose specimen 100-
11%-CUC which liquefies exhibits a larger fraction of ‘plastic’ contacts
in comparison to the medium dense specimen 100-50%-CUC. For this
resistance are finite; therefore, when the tangential contact force and
very loose specimen, fs increases rapidly to a value larger than 20% and
the moment between two particles reach the limiting value, contact
both fr and fsr also increases to near 20% at an axial strain of 1.5%
sliding and rolling occur, respectively. At this point, the tangential force
(Fig. 9c). It is noteworthy that fr gets closer to fs when approaching
and the rolling moment will be related to the normal contact force,
liquefaction, indicating that rolling mechanism becomes more notable
which reduces the total number of constraints (unknown contact
in the process of liquefaction.
forces). The redundancy index (IR) defined as the ratio of the number of
Fig. 10 shows the evolutions of mechanical redundancy index. The
constraints to the number of degrees of freedom can effectively reflect
IR of the medium dense specimen under drained (100-50%-CDC) and
the mechanical stability of granular medium (Huang et al., 2018; Kruyt
undrained (100-50%-CUC) loading conditions is always larger than 1.0,
and Rothenburg, 2009). Since the particles with no contacts are un­
which indicates that the system is consistently hyperstatic and stable.
likely to participate in force transmission, they are neglected when
For the very loose specimen 100-11%-CUC, the IR decreases due to the

Fig. 11. Cyclic behaviors of very loose specimen 50-10%-cyc-0.1: (a) axial strain vs N; (b) pore pressure vs N; (c) stress-strain; (d) stress path.

8
X. Gu, et al. Computers and Geotechnics 128 (2020) 103787

Fig. 12. Cyclic behaviors of loose specimen 100-30%-cyc-0.3: (a) axial strain vs N; (b) pore pressure vs N; (c) stress-strain; (d) stress path.

increasing number of ‘plastic’ contacts and becomes smaller than 1.0 specimen shows dilatancy immediately after cyclic instability state, and
when approaching liquefaction and the system changes from hyper­ the effective stress increases after reaching the local minimum that
static to hypostatic. corresponds to the cyclic phase transformation state (PTS) (Fig. 12c and
d). Strain softening ceases due to phase transformation, which is de­
3.2. Cyclic behaviors fined as limited deformation (Sze and Yang, 2013). Corresponding to
the limited deformation, the excess pore pressure also rises rapidly until
3.2.1. Macro stress-strain responses the initial liquefaction occurs. After liquefaction, the axial strain gra­
As seen in Fig. 11, the very loose specimen 50-10%-cyc-0.1 does not dually accumulates on both directions (Fig. 12a).
exhibit notable axial deformation until the eighth loading cycle. The Different from the looser specimens, for the medium dense specimen
triggering of this abrupt deformation (more than 10%) is associated 50-49%-cyc-0.35 (Fig. 13a and b), there is no abrupt change of the axial
with a rapid buildup of excess pore pressure (Fig. 11a and b), after strain and pore pressure with the increase of cycle numbers. The excess
which the sample collapses, showing a typical flow liquefaction beha­ pore pressure of the medium dense specimen gradually increases to the
vior (Hyodo et al., 1994). As shown in Fig. 11b, the first cycle can be initial effective stress until the onset of liquefaction. After liquefaction,
divided into two half cycles, i.e., a compression half-cycle from qcyc = 0 the effective stress recovers instantaneously due to dilatancy. As shown
to qcyc = qpeak and then returning to qcyc = 0, and an extension half- in Fig. 13c, when the deviatoric stress is close to zero, the axial strain
cycle from qcyc = 0 to qcyc = qvalley and then returning to qcyc = 0. accumulates rapidly. As the deviatoric stress increases to the deviatoric
Fig. 11c and d show that the very loose specimen exists cyclic instability stress amplitude qmax, the axial strain reaches the maximum on one
state (IS), which is characterized by a peak mobilized deviatoric stress side, showing the cyclic mobility behavior typical for medium dense
state before strain softening in the last half loading cycle before li­ sand. The stress path is in a butterfly loop and is always reversed at the
quefaction. The peak deviatoric stress in the last half loading cycle is same point (cyclic critical state, CS) on each side in subsequent loading
smaller than the cyclic deviatoric stress amplitude qmax. The cyclic in­ cycles (Fig. 13d). The denser specimens also show cyclic mobility, but
stability state is generally followed by the triggering of flow deforma­ the axial strain and the pore pressure increase more slowly than the
tion. medium dense specimen presented in Fig. 13. Their dilatancy is
Fig. 12 shows the cyclic stress-strain behavior of a loose specimen stronger than the medium dense specimen, and in some cases, even
100-30%-cyc-0.3. At the beginning of cyclic loading, the axial strain of negative pore pressure is present after initial liquefaction. Due to the
the loose specimen is not obvious, and the excess pore pressure accu­ limited space, the stress-strain behaviors of denser specimens are not
mulates gradually. After a certain number of loading cycles, the axial shown.
strain increases abruptly to a relatively large value after cyclic in­ Although the DEM simulations capture the general trend of the
stability state. Different from the very loose specimen, the loose stress-strain behavior of Toyoura sand subjected to undrained cyclic

9
X. Gu, et al. Computers and Geotechnics 128 (2020) 103787

Fig. 13. Cyclic behaviors of medium dense specimen 50-49%-cyc-0.35: (a) axial strain vs N; (b) pore pressure vs N; (c) stress-strain; (d) stress path.

loading condition, the closeness between the DEM data and the ex­ by dilation but increases again in the next loading cycles when ap­
perimental data is not as good as that in the monotonic cases presented proaching liquefaction. This mechanism is repeated in the subsequent
in Fig. 4. Due to this limitation, we only show the numerical data in this loading cycles.
study. Fig. 16 illustrates the evolutions of IR during cyclic loading. For all
specimens, IR is larger than 1.0 before the onset of liquefaction, i.e., the
3.2.2. Microscopic interpretation microstructure is stable. When approaching liquefaction, IR decreases to
Fig. 14 plots the evolutions of MCN and p′ against the number of below 1.0, and the microstructure becomes unstable. This verifies that
cycles in undrained cyclic triaxial tests. The variation of p′ is closely IR can also be used as a micromechanical method to determine when
related to MCN. It can be seen from Fig. 14a that the change of MCN of liquefaction occurs.
the very loose specimen 50-10%-cyc-0.1 is not noticeable until the The relationship between IR and MCN for all cyclic triaxial simula­
eighth cycle, after which MCN drops abruptly to a smaller value due to tions is shown in Fig. 17. When IR > 1.0, the system is hyperstatic,
the loss of valid contacts. Different from the very loose specimen, the while when IR < 1.0, the system is hypostatic. The IR = 1.0 line di­
MCN values of the specimens showing cyclic mobility decrease gradu­ vides the data into a liquefied (p′ → 0) zone and a non-liquefied (p′ > 0)
ally in the cyclic loading process before liquefaction and fluctuate zone. These two sets of data intersect the IR = 1.0 line at one single
around a certain value after liquefaction (Fig. 14b and c). point of around 3.2, which is very close to the critical value predicted
Fig. 15 shows the evolutions of sliding and rolling fractions during by Eq. (11) with p′ = 0. This indicates that regardless of the initial state
cyclic loading for samples with different initial states. Before liquefac­ and loading path there may exist a unique threshold MCN, below which
tion, due to the constraint of surrounding particles, rotation is heavily liquefaction occurs.
inhibited. Therefore, the interaction between contacting particles is
dominated by sliding. There are more sliding contacts than rolling 3.3. Cyclic liquefaction resistance
contacts. Nonetheless, in the process of liquefaction, due to the reduc­
tion of coordination number (Fig. 14) and system stability, particles can Different criteria have been used to determine the cyclic liquefac­
rotate more freely. Thereby, the rolling contact fraction increases and tion resistance: initial liquefaction is deemed to occur either when the
exceeds the sliding contact fraction when approaching liquefaction, i.e., effective mean stress p′ drops to zero or when the axial strain reaches a
rotational motion becomes dominant. For the very loose specimen (50- certain value. For the loose sample, the onset of liquefaction is usually
10%-cyc-0.1), both fs and fr surge up almost vertically, indicating the defined as the instant when flow deformation occurs, while for samples
abrupt disintegration of the force transmission network. For other exhibiting cyclic mobility, initial liquefaction is considered as the in­
specimens showing cyclic mobility, after the onset of liquefaction, the stant when the double-amplitude axial strain reaches 5% (Sze and Yang,
rolling effect drops due to the increase of coordination number induced 2013). Fig. 18 plots the relationships between cyclic stress ratio and the

10
X. Gu, et al. Computers and Geotechnics 128 (2020) 103787

5 1.2 40
fsfs
1.0 frfr
4
30 fsr
fsr
0.8
50-10%-cyc-0.1

f s, f r, f sr (%)
3
MCN

p'/p'0
0.6 20
2
MCN
MCN 0.4

1 p'/p'0
p'/p'0 10
0.2
50-10%-cyc-0.1
0 0.0
0 2 4 6 8 10 0
0 2 4 6 8 10
N
N
(a)
(a)
5 1.2
40
1.0 fsfs
4
frfr
0.8 30 fsr
fsr
3
100-30%-cyc-0.3
MCN

f s, f r, f sr (%)
p'/p'0
0.6
2 20
MCN
MCN 0.4

1 p'/p'0
p'/p'0
0.2
100-30%-cyc-0.3 10
0 0.0
0 1 2 3 4 5
N 0
0 1 2 3 4 5
(b) N
5 1.2
(b)
1 40
4 fsfs
0.8 frfr
3 30 fsr
fsr
MCN

p'/p'0

0.6
50-49%-cyc-0.35
f s, f r, f sr (%)

2
MCN
MCN 0.4 20

1 p'/p'0
p'/p'0
0.2
50-49%-cyc-0.35 10
0 0
0 1 2 3 4 5 6
N 0
(c) 0 1 2 3 4 5 6
N
Fig. 14. Evolutions of MCN and p′ during cyclic loading: (a) 50-10%-cyc-0.1;
(b) 100-30%-cyc-0.3; (c) 50-49%-cyc-0.35.
(c)
Fig. 15. Evolutions of sliding and rolling contact fractions during cyclic
number of cycles to failure obtained from different failure criteria for
loading: (a) 50-10%-cyc-0.1; (b) 100-30%-cyc-0.3; (c) 50-49%-cyc-0.35.
the very loose specimen 100-11%-cyc (flow deformation) and medium
dense specimen 100-50%-cyc (cyclic mobility). There is almost no dif­
ference between these two failure criteria for the looser specimen, and CSR required to cause initial liquefaction after 10 loading cycles as the
the cyclic strength estimated by 5% deformation is a little higher than liquefaction resistance, denoted as cyclic resistance ratio CRR10.
that estimated by the p′ = 0 criterion for the denser specimen. Despite Fig. 19 shows the relationships between CSR and N for specimens
this discrepancy, the CSR-N correlations obtained from different criteria with different relative densities but the same confining pressure of
can be expressed by a power-law function (Idriss and Boulanger, 2006): 100 kPa. The CSR-N curves locates higher with a steeper declining trend
with the increase of relative density. A denser sample has a higher
CSR = aN b (13) CRR10, in accordance with the experimental observations of Sze (2010).
Fig. 20 presents the effect of confining pressure on the CSR-N re­
where CSR is the cyclic stress ratio (qcyc/(2p′0)), N is the number of
cycles to liquefaction, a and b are fitting parameters. Herein, the lationship. The liquefaction resistances of very loose specimens with
Dr = 10% are insensitive to confining pressure (Fig. 20a), which is
number of cycles to initial liquefaction is estimated by using the p′ = 0
criterion. To better analyze the cyclic strength, this study defines the consistent with Vaid et al. (2001). While when Dr reaches 50%, the

11
X. Gu, et al. Computers and Geotechnics 128 (2020) 103787

2.5
50-10%-cyc-0.1
100-30%-cyc-0.3
2 50-49%-cyc-0.35

1.5
IR

0.5
0 0.5 1 1.5 2
N/Np'=0
Fig. 16. Evolutions of mechanical redundancy index during cyclic loading.

2.5

non-liquefied
2.0 (p' > 0)

1.5 hyperstatic
IR

1.0
1.2
1.1
liquefied
0.5 1.0 hypostatic
(p' → 0)
0.9
3.1 3.2 3.3 3.4
0.0
2.5 3.0 3.5 4.0 4.5 5.0 5.5
MCN
20-11%-cyc 50-10%-cyc 50-49%-cyc
100-11%-cyc 100-30%-cyc 100-50%-cyc
100-53%-cyc 100-64%-cyc 100-69%-cyc
300-11%-cyc 300-50%-cyc
Fig. 18. Cyclic stress ratio (CSR) against the number of cycles to failure de­
Fig. 17. Relationship between MCN and IR during cyclic loading (the solid termined from different failure criteria: (a) very loose specimen 100-11%-cyc
symbols are non-liquefied data and the hollow symbols are liquefied data). (flow deformation); (b) medium dense specimen 100-50%-cyc (cyclic mobility).

CSR-N curves slightly deviate from each other, indicating that higher
confining pressure is associated with higher cyclic strength (Fig. 20b).
Since the cyclic liquefaction resistance of sand is affected by both
density and confining pressure, it is reasonable to use the state para­
meter to combine these two effects within the CSSM framework. Fig. 21
shows the relations between liquefaction resistance and state para­
meters. As seen in Fig. 21, CRR10 overall reduces non-linearly with ψe0
following an exponential correlation (Yang and Sze, 2011) (Eq. (14)),
while it also increases non-linearly with ψMCN0 in an exponential form
(Eq. (15)):

CRR10 = 0.105e 11.57 e0

(14)

CRR10 = 0.062e1.685 MCN0 (15)

However, the data in Fig. 20b indicate an increasing trend of CRR10

Fig. 19. Cyclic stress ratio (CSR) against the number of cycles to initial lique­
with ψe0, which is contradictory to the general trend expressed by Eq.
faction for samples with different relative densities.
(14). This indicates that this macroscopic state parameter may not be
sufficient to describe the cyclic behavior of sand.
To verify the above hypothesis, a special method is used to generate After that some rattlers (floaters) are randomly deleted from the
samples with the same void ratio but different mechanical coordination medium dense specimen to match the void ratio of the very loose
numbers or the same mechanical coordination number but different specimen, and at the meantime the confining pressure remains con­
void ratios at the same initial confining pressure. One very loose spe­ stant. Since the rattlers (floaters) have no contribution to the mechan­
cimen (VL1) and one medium dense specimen (MD1) are generated by ical coordination number but are considered when calculating the void
radius expansion method mentioned in Section 2.2, which are then ratio, the specimen after deleting rattlers (floaters) has the same MCN0
isotropically consolidated to the same confining pressure (50 kPa). but with a larger e0 than the original medium dense one. Finally, the

12
X. Gu, et al. Computers and Geotechnics 128 (2020) 103787

Fig. 21. Liquefaction resistance against state parameters derived from: (a) void
ratio; (b) mechanical coordination number.
Fig. 20. Cyclic stress ratio (CSR) against the number of cycles to initial lique­
faction for samples with different confining pressures: (a) Dr = 10%; (b)
Dr = 50%. Table 4
Comparison of specimens generated by different methods.

new specimen is cycled to equilibrium, which is named as VL2. Specimen ID e0 MCN0 ψe0 ψMCN0
The details of the three specimens are listed in Table 4. As shown in
VL1 (50-10%-cyc) 0.812 3.920 0.008 0.316
Fig. 22, although the specimen VL2 has the same ψe0 as VL1, its CSR-N MD1 (50-49%-cyc) 0.701 4.437 −0.103 0.833
curve is closer to MD1 which has the same ψMCN0. This observation VL2* 0.812 4.437 0.008 0.833
suggests that the micro mechanical coordination number is a more ef­
fective state variable reflecting the cyclic strength of sand than the void * This specimen is generated by a different method.
ratio. In order to illustrate this more clearly, CRR10 of VL2 is also
plotted in Fig. 21. It is interesting to note that the data point of VL2
deviates from the massive data described by Eq. (14) in Fig. 21a but
well fits the data described by Eq. (15) in Fig. 21b. This further
strengthens our previous hypothesis that the micro state parameter
ψMCN0 is a more effective state variable that characterizes the initial
state effect on cyclic liquefaction resistance than the macro state
parameter ψe0.

4. Summary and conclusions

In this study, the macroscopic and microscopic mechanical beha­

viors of sand under drained and undrained conditions in the monotonic
and cyclic triaxial tests were investigated using DEM simulations in­
corporating the rolling resistance. The major findings are summarized
as follows:
Fig. 22. Cyclic stress ratio (CSR) against the number of cycles to initial lique­
(1) The rolling resistance linear contact model calibrated against the faction for VL1, MD1 and VL2.
experimental data effectively reflects the influence of irregular
grain shape on both the monotonic and cyclic behaviors of sand.

13
X. Gu, et al. Computers and Geotechnics 128 (2020) 103787

The density-and-stress dependent behaviors of sand are successfully Appendix A. Supplementary data
captured. All the specimens reach the same critical state lines in the
q-p′ and e-p′ planes, regardless of initial states and drainage con­ Supplementary data to this article can be found online at https://
ditions. The density dependent deformation modes under cyclic doi.org/10.1016/j.compgeo.2020.103787.
loading (flow deformation, limited deformation and cyclic mobi­
lity) are also captured by DEM. The cyclic stress ratio varies with References
the number of loading cycles to liquefaction in a power-law
manner. Been, K., Jefferies, M.G., 1985. A state parameter for sands. Géotechnique 35 (2), 99–112.
(2) The mechanical coordination number (MCN) also reaches a unique Been, K., Jefferies, M.G., Hachey, J., 1991. The critical state of sands. Géotechnique 41
(3), 365–381.
critical state, irrespective of the initial MCN and drainage condition. Belheine, N., Plassiard, J.P., Donzé, F.V., et al., 2009. Numerical simulation of drained
The critical-state MCNs increase linearly with (p′/pa)0.45, which is triaxial test using 3D discrete element modeling. Comput. Geotech. 36 (1–2),
defined as the micro critical state line. 320–331.
Boulanger, R.W., 2003. Relating Kα to relative state parameter index. J. Geotech.
(3) When the tangential contact force or moment between two particles Geoenviron. Eng. 129 (8), 770–773.
reaches their limiting value, contact sliding or rolling occurs. There Castro, G., Poulos, S.J., 1977. Factors affecting liquefaction and cyclic mobility. J.
are more sliding contacts than rolling contacts when specimen is Geotech. Geoenviron. Eng. 103 (6), 501–516.
Cundall, P.A., Strack, O.D.L., 1979. A discrete numerical model for granular assemblies.
non-liquefied. The contact yielding mechanism alters from sliding
Géotechnique 29 (1), 47–65.
dominant to rolling dominant when approaching liquefaction. Da Cruz, F., Emam, S., Prochnow, M., et al., 2005. Rheophysics of dense granular ma­
Evolutions of sliding and rolling fractions follow different patterns terials: discrete simulation of plane shear flows. Phys. Rev. E 72 (2), 021309.
Dinesh, S.V., Sitharam, T.G., Vinod, J.S., 2004. Dynamic properties and liquefaction be­
within a loading cycle before and after liquefaction.
havior of granular materials using discrete element method. Curr. Sci. 87 (10),
(4) The mechanical redundancy index (IR), i.e., the ratio of the number 1379–1387.
of constraints to the number of degrees of freedom of the active Estrada, N., Azéma, E., Radjai, F., et al., 2013. Comparison of the effects of rolling re­
particles, is defined considering the sliding and rolling fractions. sistance and angularity in sheared granular media. AIP Conf. Proc. 1542 (1),
891–894.
Liquefaction occurs when IR drops below 1.0 under both monotonic Evans, T.M., Zhang, L., 2019. A numerical study of particle friction and initial state effects
and cyclic loading. A threshold value of MCN distinguishing a flow on the liquefaction of granular assemblies. Soil Dyn. Earthquake Eng. 126, 105773.
state (liquefied) from a solid state is identified corresponding to Fukushima, S., Tatsuoka, F., 1984. Strength and deformation characteristics of saturated
sand at extremely low pressures. Soils Found. 24 (4), 30–48.
IR = 1.0. Gong, J., Nie, Z.H., Zhu, Y.G., et al., 2019. Exploring the effects of particle shape and
(5) The micro state parameter ψMCN0 is defined as the distance between content of fines on the shear behavior of sand-fines mixtures via the DEM. Comput.
the initial mechanical coordination number MCN0 and the critical- Geotech. 106, 161–176.
Gong, G.B., Thornton, C., Chan, A.H.C., 2012. DEM simulations of undrained triaxial
state mechanical coordination number MCNCS with the same ef­ behavior of granular material. J. Eng. Mech. 138 (6), 560–566.
fective stress. It is found that the cyclic resistance ratio CRR10 re­ Gu, X.Q., Huang, M.S., Qian, J.G., 2014. DEM investigation on the evolution of micro­
duces non-linearly as ψe0 increases and increases non-linearly as structure in granular soils under shearing. Granular Matter 16 (1), 91–106.
Gu, X.Q., Hu, J., Huang, M.S., 2017. Anisotropy of elasticity and fabric of granular soils.
ψMCN0 increases, both of which can be well represented by ex­
Granular Matter 19 (2), 1–15.
ponential functions. Using a special method, it is shown that the Gu, X.Q., Lu, L.T., Qian, J.G., 2017. Discrete element modeling of the effect of particle
cyclic behaviors of specimens with the same initial void ratio but size distribution on the small strain stiffness of granular soils. Particuology 32, 21–29.
Gu, X.Q., Yang, J., 2013. A discrete element analysis of elastic properties of granular
different initial MCNs deviate from each other, while the specimens
materials. Granular Matter 15 (2), 139–147.
with different initial void ratios but the same MCN have similar Hosn, R.A., Sibille, L., Benahmed, N., et al., 2017. Discrete numerical modeling of loose
CRR10. This indicates that ψMCN0 is a more effective state variable soil with spherical particles and interparticle rolling friction. Granular Matter 19
than ψe0 in characterizing the cyclic behaviors of sand. (1), 4.
Hu, M.Y., O’Sullivan, C., Jardine, R.R., et al., 2010. Stress-induced anisotropy in sand
under cyclic loading. Granular Matter 12 (5), 469–476.
Note that in this study the undrained condition is approximated by Huang, X., Hanley, K.J., O'Sullivan, C., et al., 2014. Exploring the influence of inter­
imposing a constant-volume geometric constraint on the simulating particle friction on critical state behaviour using DEM. Int. J. Numer. Anal. Meth.
Geomech. 38 (12), 1276–1297.
specimens. The real fluid-particle interaction mechanism underlying Huang, X., O'Sullivan, C., Hanley, K.J., et al., 2014. Discrete-element method analysis of
the liquefaction phenomena requires further exploration by adopting the state parameter. Géotechnique 64 (12), 954–965.
the fluid-coupling scheme. Huang, X., Kwok, C.Y., Hanley, K.J., et al., 2018. DEM analysis of the onset of flow de­
formation of sands: linking monotonic and cyclic undrained behaviours. Acta
Geotech. 13 (5), 1061–1074.
Huang, X., Hanley, K.J., Zhang, Z.X., et al., 2019. Structural degradation of sands during
CRediT authorship contribution statement cyclic liquefaction: Insight from DEM simulations. Comput. Geotech. 114, 103139.
Hyodo, M., Tanimizu, H., Yasufuku, N., et al., 1994. Undrained cyclic and monotonic
triaxial behaviour of saturated loose sand. Soils Found. 34 (1), 19–32.
Xiaoqiang Gu: Conceptualization, Methodology. Jiachen Zhang: Idriss, I.M., Boulanger, R.W., 2006. Semi-empirical procedures for evaluating liquefaction
Investigation, Visualization, Data curation, Writing - original draft. Xin potential during earthquakes. Soil Dyn. Earthquake Eng. 26 (2–4), 115–130.
Huang: Writing - review & editing. Itasca Consulting Group, 2015. Particle Flow Code in three dimensions: user's manual,
version 5.0. Minneapolis, USA.
Iwashita, K., Oda, M., 1998. Rolling resistance at contacts in the simulation of shear band
development by DEM. J. Eng. Mech. 124 (3), 285–292.
Declaration of Competing Interest Jiang, M.J., Shen, Z.F., Wang, J.F., 2015. A novel three-dimensional contact model for
granulates incorporating rolling and twisting resistances. Comput. Geotech. 65,
147–163.
The authors declare that they have no known competing financial Kruyt, N.P., Rothenburg, L., 2009. Plasticity of granular materials: a structural-mechanics
view. AIP Conf. Proc. 1145 (1), 1073–1076.
interests or personal relationships that could have appeared to influ­
Lopera Perez, J.C., Kwok, C.Y., O'Sullivan, C., et al., 2016. Assessing the quasi-static
ence the work reported in this paper. conditions for shearing in granular media within the critical state soil mechanics
framework. Soils Found. 56 (1), 152–159.
Martin, E.L., Thornton, C., Utili, S., 2019. Micromechanical investigation of liquefaction
of granular media by cyclic 3D DEM tests. Géotechnique 1–10.
Acknowledgements Miura, S., Toki, S., 1982. A sample preparation method and its effect on static and cyclic
deformation-strength properties of sand. Soils Found. 22 (1), 61–77.
The work presented in this paper is supported by National Natural Ng, T.T., Dobry, R., 1994. Numerical simulations of monotonic and cyclic loading of
granular soil. J. Geotech. Eng. 120 (2), 388–403.
Science Foundation of China (Grant No. 51822809, 41772283) and the Nguyen, H.B.K., Rahman, M.M., Fourie, A.B., 2016. Undrained behaviour of granular
Fundamental Research Funds for the Central Universities. These sup­ material and the role of fabric in isotropic and K0 consolidations: DEM approach.
ports are gratefully acknowledged. Géotechnique 67 (2), 1–15.

14
X. Gu, et al. Computers and Geotechnics 128 (2020) 103787

Nguyen, H.B.K., Rahman, M.M., Fourie, A.B., 2018. Characteristic behavior of drained Toki, S., Tatsuoka, F., Miura, S., et al., 1986. Cyclic undrained triaxial strength of sand by
and undrained triaxial compression tests: DEM study. J. Geotech. Geoenviron. Eng. a cooperative test program. Soils Found. 26 (3), 117–128.
144 (9), 04018060. Vaid, Y.P., Chern, J.C., 1983. Effect of static shear on resistance to liquefaction. Soils
Oda, M., Konishi, J., Nemat-Nasser, S., 1982. Experimental micromechanical evaluation Found. 23 (1), 47–60.
of strength of granular materials: effects of particle rolling. Mech. Mater. 1 (4), Vaid, Y.P., Chern, J.C., Tumi, H., 1985. Confining pressure, grain angularity, and lique­
269–283. faction. J. Geotech. Eng. 111 (10), 1229–1235.
O'Sullivan, C., Cui, L., O'Neill, S.C., 2008. Discrete element analysis of the response of Vaid, Y.P., Stedman, J.D., Sivathayalan, S., 2001. Confining stress and static shear effects
granular materials during cyclic loading. Soils Found. 48 (4), 511–530. in cyclic liquefaction. Can. Geotech. J. 38 (3), 580–591.
Plassiard, J.P., Belheine, N., Donzé, F.V., 2009. A spherical discrete element model: ca­ Verdugo, R., Ishihara, K., 1996. The steady state of sandy soils. Soils Found. 36 (2),
libration procedure and incremental response. Granular Matter 11 (5), 293–306. 81–91.
Roscoe, K.H., Schofield, A.N., Wroth, C.P., 1958. On the yielding of soils. Géotechnique 8 Wang, R., Fu, P.C., Zhang, J.M., et al., 2016. DEM study of fabric features governing
(1), 22–53. undrained post-liquefaction shear deformation of sand. Acta Geotech. 11 (6),
Rothenburg, L., Kruyt, N.P., 2004. Critical state and evolution of coordination number in 1321–1337.
simulated granular materials. Int. J. Solids Struct. 41 (21), 5763–5774. Wang, G., Wei, J.T., 2016. Microstructure evolution of granular soils in cyclic mobility
Schofield, A.N., Wroth, C.P., 1968. Critical state soil mechanics. McGraw-Hill, New York. and post-liquefaction process. Granular Matter 18 (3), 51.
Seed, B., Lee, K.L., 1966. Liquefaction of saturated sands during cyclic loading. J. Soil Wei, J.T., Wang, G., 2015. Microstructure evolution of granular soils during liquefaction
Mech. Found. Div. 92. process. Geomechanics from micro to macro 251–256.
Shafipour, R., Soroush, A., 2008. Fluid coupled-DEM modelling of undrained behavior of Wei, J.T., Wang, G., 2017. Discrete-element method analysis of initial fabric effects on
granular media. Comput. Geotech. 35, 673–685. pre-and post-liquefaction behavior of sands. Géotechnique Letters 7 (2), 161–166.
Sitharam, T.G., 2003. Discrete element modelling of cyclic behaviour of granular mate­ Widuliński, Ł., Kozicki, J., Tejchman, J., 2009. Numerical simulations of triaxial test with
rials. Geotech. Geol. Eng. 21 (4), 297–329. sand using DEM. Arch. Hydro-Eng. Environ. Mech. 56 (3–4), 149–172.
Sitharam, T.G., Vinod, J., 2008. Numerical simulation of liquefaction and pore pressure Xu, X.M., Ling, D.S., Cheng, Y.P., et al., 2015. Correlation between liquefaction resistance
generation in granular materials using DEM. Int. J. Geotech. Eng. 2 (2), 103–113. and shear wave velocity of granular soils: a micromechanical perspective.
Sitharam, T.G., Vinod, J.S., 2009. Critical state behaviour of granular materials from Géotechnique 65 (5), 337–348.
isotropic and rebounded paths: DEM simulations. Granular Matter 11 (1), 33–42. Yan, W.M., Dong, J.J., 2011. Effect of particle grading on the response of an idealized
Syed, Z., Tekeste, M., White, D., 2017. A coupled sliding and rolling friction model for granular assemblage. Int. J. Geomech. 11 (4), 276–285.
DEM calibration. J. Terramech. 72, 9–20. Yang, J., Li, X.S., 2004. State-dependent strength of sands from the perspective of unified
Sze, H.Y., Yang, J., 2013. Failure modes of sand in undrained cyclic loading: impact of modeling. J. Geotech. Geoenviron. Eng. 130 (2), 186–198.
sample preparation. J. Geotech. Geoenviron. Eng. 140 (1), 152–169. Yang, J., Sze, H.Y., 2011. Cyclic behaviour and resistance of saturated sand under non-
Sze, H.Y., 2010. Initial shear and confining stress effects on cyclic behaviour and lique­ symmetrical loading conditions. Géotechnique 61 (1), 59–73.
faction resistance of sands. The University of Hong Kong. Yang, J., Sze, H.Y., 2011. Cyclic strength of sand under sustained shear stress. J. Geotech.
Tatsuoka, F., Ochi, K., Fujii, S., et al., 1986. Cyclic undrained triaxial and torsional shear Geoenviron. Eng. 137 (12), 1275–1285.
strength of sands for different sample preparation methods. Soils Found. 26 (3), Yoshimine, M., Ishihara, K., Vargas, W., 1998. Effects of principal stress direction and
23–41. intermediate principal stress on undrained shear behavior of sand. Soils Found. 38
Thornton, C., 2000. Numerical simulations of deviatoric shear deformation of granular (3), 179–188.
media. Géotechnique 50 (1), 43–53. Zhang, L., Evans, T.M., 2018. Boundary effects in discrete element method modeling of
Thornton, C., 2015. Granular dynamics, contact mechanics and particle system simula­ undrained cyclic triaxial and simple shear element tests. Granular Matter 20 (4), 60.
tions. Springer.

15