Computers and Geotechnics 122 (2020) 103542

Contents lists available at ScienceDirect

Computers and Geotechnics

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

Research Paper

DEM exploration of the effect of particle shape on particle breakage in T

granular assemblies
⁎
Tao Zhang, Chi Zhang, Jiuqun Zou, Baosheng Wang, Fangnian Song, Weihao Yang
State Key Laboratory for Geomechanics and Deep Underground Engineering, China University of Mining and Technology, Xuzhou 221116, China

A R T I C LE I N FO A B S T R A C T

Keywords: The effect of particle shape on particle breakage is investigated through triaxial tests on sand assemblies mixed
Discrete element method with non-spherical agglomerates using the discrete element method. Mixed assembly is generated in terms of
Sphericity cumulative distribution of sphericity of real sands. Comparable macro-mechanical results are found between
Particle shape DEM and experimental observations, including non-linear stress-strain relationship, peak deviatoric stress, and
Particle breakage
peak friction angle. More broken bonds and agglomerates numbers are generated in mixed assembly than that in
Energy input
spherical assembly. The hyperbolic relationship between particle breakage and energy input is found to have
relevance to particle shape, and the fitting result of mixed assembly is in better accordance with real sands.

1. Introduction breakage. In the last two decades, the discrete element method (DEM)
has proved to be a powerful numerical methodology that allows pre-
Particle breakage frequently occurs in granular materials as the diction of particle flows, including particle breakage, both at the in-
exterior energy acting on soil particle exceeds its strength [1–4]. A great dividual particle level and as a whole process.
deal of experimental studies have shown that the macro-mechanical The earliest model of the effect of particle shape in DEM simulations
behaviors of granular soils, such as compressibility, shear strength, and consisted of 2D elliptically-shaped or polygon-shaped unbreakable
stress-dilatancy, are highly influenced by particle breakage [5–8]. As a particles [13–16]. This was followed by a model of 2D polygon-shaped
non-negligible inherent soil characteristic, particle shape plays a sig- breakable particles which were replaced with smaller inter-connected
nificant role in determining the evolution of particle size during particle sub-particles [17–19]. Although the mechanical behavior of assemblies
breakage. Cavarretta et al. [9] found that the crushing strength of of 2D particles reflects general features of granular materials, magni-
homogeneous particles is roundness dependent rather than size de- tudes of parameters such as porosity and friction coefficient are much
pendent. Wang and Coop [10] revealed that the particles with higher lower than those of real sands. Aided by 3D DEM, many researchers
roundness tend to break in an explosive mode with higher strength, have investigated the macro- and micro-behaviors of breakable gran-
whereas those with lower roundness tend to split with lower strength, ular soils, such as shear strength, volume change, plastic deformation,
which were also found by Todisco et al. [11]. This is due to the local grading evolution, anisotropy, energy, etc. Among these, two basic
stress concentration for particles with different shapes, and sharper methods have been applied to simulate the breakable particles: 1)
contacts result in particle breakage when the failure stress is larger than forming an agglomerate by bonding elementary balls together, the
the particle strength. In order to provide a deeper understanding on the bonds between elementary balls are deformable and breakable. It has
contribution of particle shape to the breakage in granular soils, Karatza been widely used in many previous researches [20–27], and 2) repla-
et al. [12] conducted a series of strain-controlled oedometric com- cing the original particle by a group of smaller ones when a certain
pression tests on well-rounded, highly spherical and uniform zeolite strength is reached [28–30]. The first method is used in this study due
particle assemblies. The results showed that less spherical particles tend to its flexibility to generate breakable agglomerates with arbitrary
to break more. However, as evident from previous studies, experimental shapes within realistic-shaped boundaries. However, these 3D DEM
investigation of the effect of particle shape on particle breakage in simulations have ignored the effect of particle shape on particle
natural sands does not achieve more results due to the limitation of breakage in granular assemblies.
experimental equipment. Especially, it is very difficult to trace the More recently, the development of three-dimensional imaging
evolutions of particle grading and energy dissipated by particle technologies, such as 3D laser scanner (LS) and X-ray microcomputed

⁎
Corresponding author.
E-mail address: [email protected] (W. Yang).

https://doi.org/10.1016/j.compgeo.2020.103542
Received 3 December 2019; Received in revised form 7 February 2020; Accepted 11 March 2020
Available online 20 March 2020
0266-352X/ © 2020 Published by Elsevier Ltd.
T. Zhang, et al. Computers and Geotechnics 122 (2020) 103542

(a) S = 0.684 (b) S = 0.705 (c) S = 0.762

(d) S = 0.817 (e) S = 0.890 (f) S = 0.934

Fig. 1. 3D view of reconstructed sand particles with different sphericity values.

tomography (μCT), are usually used to capture realistic shape of gran- discussed at great length. Possible relationship between particle
ular materials. 3D LS method relies on the spatial coordinates of points breakage and energy input is explored and compared with previous
on the exterior reconstructed surface, which is often composed of tri- experimental results.
angular surface meshes [31–34]. Continued progress of LS technology
leads to higher resolution and less time consuming. Whereas the in- 2. Simulation procedure
formation on the internal properties (e.g. density) of a particle are not
measurable. X-ray μCT scanning is currently believed as the most 2.1. Modelling of sand agglomerates with different shapes
comprehensive method of measuring geometrical properties of parti-
cles. Due to its high resolution, X-ray μCT has been used for the mea- Sand agglomerates in real underground engineering have all dif-
surement and reconstruction of small sand particles (e.g. 1 mm-2 mm) ferent shapes with different surface areas. In this study, three re-
[35–38]. Particularly, in the examination of particle kinematics of presentative agglomerates with different shapes are generated: regular
sands, particles at different shear strains were reconstructed with the tetrahedron, regular octahedron, and sphere. Volume-based sphericity
‘ID track’ approach, although particles have small differences in particle definition, proposed by Wadell [44], is widely used to quantify the
volume and particle surface area during shearing [38]. Meanwhile, the morphology of a 3D non-spherical particle in various research fields
evolution of particle size distribution (PSD) during test can also be [45–47]. Sphericity describes how closely a particle resembles a sphere,
obtained through the image analysis [38,39]. By these means, a more or it provides a measure of the degree to which a particle’s shape ap-
realistic and accurate modeling of granular assemblies, in terms of proaches a sphere, which is a global and comprehensive shape para-
macro- and micro-behaviors, are likely to be obtained, irregular parti- meter compared to 2D parameters (e.g., sphericity, roundness and
cles with different sphericities being much more susceptible to crush roughness in 2D) such that the 2D microscopic images only reflect the
than spherical ones. projected shape of stabilized state and provide limited morphological
However, the most widely adopted modeling approach in DEM is to information [39,48,49]. Particle sphericity S is the ratio of the surface
clump or overlap balls to generate irregular particles, which are con- area of a sphere, of the same volume as the given particle, to the actual
sidered as rigid bodies [40–42]. To simulate particle breakage with surface area of the particle. Aparticle is the surface area of the given
non-spherical agglomerates, Fu, Hu and Zhou [43] generated breakable particle, and Asphere is the surface area of a sphere having the same
agglomerates with realistic particle shapes based on X-ray μCT and volume as the given particle.
spherical harmonic analysis. The compression tests showed that particle
shape has a significant effect on both macro-properties and micro- Asphere
S=
properties of granular assemblies. However, the breakage behavior of Aparticle (1)
non-spherical agglomerates with different sphericities in triaxial tests is
not considered in the previous studies. Fig. 1 presents the 3D view of representative reconstructed Fujian
In this paper, the non-spherical geometries are limited to regular standard sands, which are widely used for geotechnical tests in China,
tetrahedron and regular octahedron. The aim of this work is to perform with different sphericity values. It shows that the shape of particles
the effect of non-spherical agglomerates on particle breakage during approaches a sphere with increasing magnitude of sphericity values.
shearing. As the basis for simulation, uniaxial compression tests on Surface area is the total area of triangular elements on particle surface,
single agglomerate are conducted to ensure the feasibility and validity and particle volume is the total volume of voxels within the boundary of
of the model parameters. Then, a series of triaxial tests of mixed ag- reconstructed surface. Sphericity data from a large number of sand
glomerates and spherical agglomerates are conducted respectively. The particles have been measured to obtain reliable and statistical results.
effect of non-spherical agglomerates on the macro-mechanical beha- The number of sand particles ranging from 0.6 mm to 1.6 mm is more
viors is investigated and compared with experimental results. The than 4000. From the statistical calculation of these sphericity data,
evolutions of particle breakage are also traced in detail, and its effects normal and cumulative distributions are shown in Fig. 2, in which the
on the micro-behaviors of mixed assembly and spherical assembly are corresponding sphericity at the peak frequency count is about 0.852
and the average value of sphericity is about 0.803. The sphericities (ST,

2
T. Zhang, et al. Computers and Geotechnics 122 (2020) 103542

600 100 2.2. Crushing tests for sand agglomerates

Frequency of sphericity
500 Fig. 5 shows the typical simulation results of compression force
Cumulative distribution 80

Cumulative distribution (%)

versus displacement for two non-spherical agglomerates and their
corresponding spherical agglomerate. Fig. 4(c) presents the final frag-
400 ments after peak forces. This breakage mode is very similar to the
60
splitting mode observed by Wang and Coop [10]. For the spherical
Counts

300 agglomerate, an overall steep slope relationship and an obvious peak

force can be observed during compression, and this trend is approxi-
40
200 mately consistent with the experimental result of platen compression
test on single quartz sand by Nakata et al. [51]. For the non-spherical
20 agglomerates, an overall trend of a slight decrease of peak forces can be
100 observed in both tetrahedral and octahedral agglomerates. This is al-
most certainly attributed to the breakage of the local asperities and
0 0 corners [10,43].
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 For the real sand particles, the strength may vary widely due to the
Sphericity complex internal flaws. It has been verified that the fracture strengths
Fig. 2. Frequency distributions of sphericity for sand particles.
of granular soils follow a Weibull distribution [52,53]. Robertson and
Bolton [20] firstly performed crushing tests on single agglomerates with
random bonding flaws. Cheng et al. [22] randomly removed 20% of
elementary balls to provide a statistical variability to the strength of the
agglomerates. In order to provide a statistical variability to the fracture
1.0
strength affected by the agglomerate shape, similar to that of a real
sand particle, 30 agglomerates with the same shape were created by
giving each elementary ball a probability of existence of 80%. For the
0.8
sake of simplicity, the survival probability of a batch of 30 agglomerates
is calculated following the procedures in Cheng et al. [22].
Particle sphericity

Fig. 6 shows the normalized curves of the survival probability Ps and

0.6
Weibull distributions of the spherical and non-spherical agglomerates
comparing with the experimental results from Nakata et al. [51]. The
Weibull modulus m can be calculated by the following equation:
0.4
1 σ
ln ⎡ln ⎛ ⎞ ⎤ = m ln ⎛ ⎞
⎜ ⎟ ⎜ ⎟
⎢ ⎝ Ps ⎠ ⎥ ⎝ σ0 ⎠ (2)
0.2 ⎣ ⎦
where σ0 is the characteristic fracture stress corresponding to 1/e or
37% survival probability, and m is the Weibull modulus.
0.0 It can be seen that the curves agree well with the Weibull dis-
Tetrahedron Octahedron Sphere tribution. The simulation results are comparable with some experi-
Fig. 3. Agglomerates with different sphericity values. mental data [10,51,54–56]. The Weibull moduli are approximately
3.25, 4.12, and 4.69 for tetrahedral, octahedral, and spherical ag-
glomerates, which are also comparable to the results of quartz particles
SO, and SS) for regular tetrahedron, regular octahedron, and sphere are
presented by Wang and Coop [10], McDowell [55], Nakata et al. [51]
0.671, 0.846 and 1.000, as shown in Fig. 3.
and Robertson and Bolton [20].
To generate a dense and uniform agglomerate consisting of nu-
merous bonded elementary balls, a geometry with specified shape (see
Fig. 4(a)) and FISH functions were used to make sure that every ball 2.3. Sample preparation
was completely within the geometry boundaries and cycled to equili-
brium so as to exclude overlaps. Considering the computational time, The numerical samples containing 475 agglomerates were gener-
the radius of the elementary ball was set to 0.12 mm in this study, and ated by random orientations within the initial cylinder containers with
these three type agglomerates consisted of 54, 53, and 55 elementary a diameter of 8 mm and height of 16 mm, and there were no overlap
balls in close packing respectively. Fig. 4(b) shows the final states of between any two agglomerates, as illustrated in Fig. 7(a) and (b). Both
elementary balls for tetrahedron, octahedron and sphere agglomerates. of them followed the same initial particle size distribution (PSD) shown
It is clear that the elementary balls are evenly and closely distributed in Fig. 8. Jamilkowski et al. [57] proposed that the ratio of the sample
within the geometry boundaries. After reaching the final equilibrium size to the maximum particle size should be larger than 5, to eliminate
state, the linear parallel bond model was applied to the contacts be- the effect of the sample size and minimize stress non-uniformities inside
tween elementary balls to form a breakable agglomerate. The ad- a sample. The minimum ratio approaches 6.05 in this study, thus rea-
vantage of the parallel bond model against the contact bond model is sonable conclusions can be obtained from the small samples here. To
that it is capable to provide a rotational resistance and visible physical ensure the validity of the simulation results, experimental results de-
fracture between bonded balls [26,50]. The parameters for the ag- rived from real sands were examined in a parallel study for comparison,
glomerates are summarized in Table 1. The parameters for elementary including the stress–strain relationship, deviatoric stress, and peak
balls are from Cheng et al. [23] and Bolton et al. [24]. The parameters friction angle. The experimental sample was illustrated in Fig. 7(c) and
of parallel bond model used in this study are based on those from Wang (d), and its initial PSD was presented in Fig. 8. During sample pre-
and Yan [50] and Fu et al. [43]. paration, all samples contained approximately 1.18 g Fujian standard
sands, the density of Fujian sand is 2610 kg/m3, and the initial void
ratios of all samples were 0.779.
Based on the cumulative distribution in Fig. 2, the proportion of

3
T. Zhang, et al. Computers and Geotechnics 122 (2020) 103542

Fig. 4. Illustrations of crushing behavior of three agglomerates: (a) geometry boundaries; (b) before crushing; and (c) after crushing.

Table 1 binary mixtures. In this respect, an extended shape measure, is pro-

Parameters of breakable agglomerates. posed as follows to generalized combined sphericity SC for 3D mixtures
Equivalent diameter of each agglomerate (mm) 1.0
SC = ST × 10% + SO × 70% + SS × 20% (3)
Diameter of elementary ball (mm) 0.24
Density of ball (kg/m3) 2610 The SC value of 0.859 for the mixed assembly is slightly larger than
Normal and shear stiffness of ball (N/m) 4 × 106 the average value of sphericity of real sands. The concept of combined
Friction coefficient of ball 0.5
sphericity is reasonable because the spherical and non-spherical ag-
Normal and shear parallel bond strength (N/m2) 5 × 108
Normal and shear parallel bond stiffness (N/m3) 5 × 1014 glomerates have the same gradation and specific gravity in an as-
Normal and shear stiffness of wall (N/m) 2 × 106 sembly. This quantitative method has also been adopted in many pre-
Friction coefficient between walls and balls 0.0 vious studies [59,60].
It is noted that the shear stiffness was set to a low value and friction
coefficient between balls was temporarily reduced to zero to increase
efficiency and reach a dense packing of agglomerates during prepara-
tion stage. The initial void ratios of mixed sample and spherical sample
were 0.784 and 0.798 respectively, which were slightly larger than that
of experiment. The investigated void ratio was calculated following the
method used in Fu et al. [43]

2.4. Isotropic and triaxial compression tests

After a sample was generated, it was isotropically compressed until

a complete equilibrium condition was reached. A servo-controlled al-
gorithm was introduced to achieve and keep the target confining
pressures (2 MPa, 3 MPa, 4 MPa, and 5 MPa), and the difference be-
tween the pressure obtained from the walls was smaller than a toler-
ance of 0.1%. After the isotropic compression, the friction coefficient
between balls was reset to 0.5 for shear.
Subsequently, the sample was subjected to triaxial test by moving
Fig. 5. Compression tests for three type agglomerates comparing with experi- the top and bottom walls towards each other with the constant loading
mental results of Nakata et al. [51]. speed of 0.005 m/s. The servo control in the lateral wall remained
unchanged, and the whole parameters were set to the final values at the
beginning of this stage. A computer workstation with Intel ®Core ®CPU
each agglomerate in a sand assembly is 0.1: 0.7: 0.2 (tetrahedron: oc-
i9-9900 k was used in this simulation, and it took approximately
tahedron: sphere). To obtain the overall shape data for sand assembly,
28 days for each DEM test with a high clock speed of 5.0 GHz.
Yang and Wei [58] proposed a concept of combined sphericity for

4
T. Zhang, et al. Computers and Geotechnics 122 (2020) 103542

Fig. 6. Survival distribution curves of single agglomerate crushing: (a) normalized distribution comparing with the experimental results; (b) Weibull modulus.

Fig. 7. Simulated sand assemblies and natural sand sample: (a) spherical agglomerates; (b) mixed agglomerates; (c) before shearing; (d) after shearing.

5
T. Zhang, et al. Computers and Geotechnics 122 (2020) 103542

stress–strain responses show an initial linear trend followed by a cur-

vature to the peak stress, and higher confining pressures lead to higher
mobilised deviatoric stresses. In addition, comparing the stress–strain
curves at a given confining pressure shows that the deviatoric stress in a
spherical assembly is slightly lower than that of mixed assembly, this is
due to the more serious interlocking effect between non-spherical ag-
glomerates [18,61]. Furthermore, the axial strain at peak state in-
creases with increasing the content of non-spherical agglomerates or
with increasing the confining pressure. These behaviors have also been
observed in experimental results [60] and other DEM simulation results
[62].
Fig. 10 compares the relationships between the volumetric strain
and axial strain for spherical and mixed assemblies. All samples exhibit
an obvious initial contraction, and the contraction becomes more evi-
dent with an increase in confining pressure. When the axial strain
comes to the phase-transformation state where the volumetric strain is
zero, all samples begin to expand, and the volumetric expansion de-
creases as confining pressure increases for a specific assembly. The in-
Fig. 8. Particle size distribution. itial contraction of mixed assembly is more evident than that of sphe-
rical assembly. For example, the maximum contraction at phase-
3. Macro-mechanical results transformation state is 2.108% for mixed assembly but only 1.503% for
spherical assembly at the same confining pressure σc = 5 MPa. At the
3.1. Results of triaxial tests volumetric expansion stage, spherical assembly shown in Fig. 10(a)
exhibits slightly more dilation than that of mixed assembly, this is be-
Fig. 9 shows the relationships of deviatoric stress and axial strain for cause particle breakage in mixed assembly is more pronounced than
spherical and mixed assemblies at different confining pressures (i.e., that in spherical assembly. During breakage at high stress where
σc = 2, 3, 4, and 5 MPa). The DEM results are also compared with that breakage is dominant over dilation, particles reduce the sample’s void
of experiments, the variation trends are basically consistent. The ratio, and facilitates volumetric compression [63].

Fig. 9. Stress–strain relationships for mixed and spherical assemblies comparing with the experimental results: (a) σc = 2 MPa; (b) σc = 3 MPa; (c) σc = 4 MPa; (d)
σc = 5 MPa.

6
T. Zhang, et al. Computers and Geotechnics 122 (2020) 103542

Fig. 10. Volumetric strain versus axial strain: (a) spherical assembly; (b) mixed assembly.

3.2. Peak friction angle noticeable that the number of broken bonds in mixed assembly is
slightly larger than that in spherical assembly at this initial stage, this is
The peak friction angle φp is the maximum value of mobilised due to the angularity breakage caused by a decrease in coordination
friction angle φm, which is defined as follows number or greater contact force between agglomerates in the mixed
σa − σc assembly.
sin ϕm = After this, the broken bonds increase linearly till the end of the tests.
σa + σc (4)
This tendency agrees with the DEM simulation results presented by
where σa and σc are the major and minor effective stresses, respec- Wang and Yan [50] and Liu et al. [26]. The original agglomerates si-
tively. Fig. 11 indicates that the peak friction angle φp depends on multaneously break up into several new agglomerates, which may un-
particle morphology. An increase in combined sphericity SC at a given dergo crushing again after rearrangement. Fig. 12(d) shows the com-
confining pressure leads to a reduction in peak friction angle. In addi- parison of the number of agglomerates for all tests. It can be seen that
tion, an increase in confining pressure also leads to a reduction in peak agglomerate breakage is not significant at the onset of test even with
friction angle for a specific assembly. These trends are consistent with high confining pressure, and the number of agglomerates almost line-
the experiments in Xiao et al. [60] and DEM simulation results in Xu arly increases after axial strain reaches a certain level. In addition, the
et al. [64]. rate of agglomerate breakage of mixed assembly is more pronounced
than that of spherical assembly. For example, the increasing number is
4. Effect of particle shape on particle breakage 357 for mixed assembly but only 223 for spherical assembly at the same
confining pressure (σc = 5 MPa). Therefore, the extent of particle
4.1. Evolution of broken bonds breakage is linked not only with the confining pressure but also with
particle sphericity.
The evolutions of the number of broken bonds and agglomerates are Although the breakable agglomerates are formed by several ele-
examined during shearing. As shown in Fig. 12(a) and (b), the dis- mentary balls, the effect of particle sphericity on the evolutions of both
tributions of broken bonds are even within the rigid walls. Fig. 12(c) the numbers of broken bonds and agglomerates can be reflected. These
shows the evolutions of the number of broken bonds for spherical and findings are basically consistent with other DEM simulation observa-
mixed assemblies at the confining pressure of σc = 5 MPa. There are tions, which greatly make up some limitations in laboratory testing.
few bonds break initially until the axial strain reach nearly 2.5%. It is

4.2. Evolution of particle size distribution

The evolutions of particle size distribution curves are plotted in

Fig. 13 and Fig. 14. As illustrated in Fig. 15, the PSDs are used to
quantify particle breakage by relative breakage (Br) [1]. Br has the re-
markable advantage of representing all changes of the grading com-
pared with other mono-sized computational methods [6,51,65]. It is
clearly that the PSD transforms to a relatively more well-graded dis-
tribution with the increase of confining pressure, axial strain, and
particle roughness. Meanwhile, the size of large-diameter agglomerates
remains nearly constant, and the small-diameter ones undergo sig-
nificant changes. These similar phenomena have also been observed in
experimental tests [2,66] and DEM simulations [50,67]. However, the
effect of non-spherical agglomerates with different sphericities on the
evolution of particle breakage has not been investigated in previous
studies.
Fig. 16 shows the comparison of relative breakage between sphe-
rical and mixed assemblies. Particle breakage in relative breakage in-
Fig. 11. Peak friction angles for mixed and spherical assemblies comparing creases with increasing axial strain, and higher confining pressure leads
with the experimental results. to more significant particle breakage. Furthermore, relative breakage is

7
T. Zhang, et al. Computers and Geotechnics 122 (2020) 103542

Fig. 12. Evolutions of the number of broken bonds when σc = 5 MPa, and comparison of the number of agglomerates for all samples during shearing: (a) spherical
assembly; (b) mixed assembly; (c) accumulated number of broken bonds; (d) number of agglomerates.

found to increase apparently with decreasing combined sphericity SC input of isotropic compression, and ES is the energy input of shearing. A
with other factors unchanged, which is related to the interlocking effect large number of studies have been conducted to investigate the corre-
or greater contact force at the edges of non-spherical agglomerates. For lation between relative breakage and plastic work per unit volume
example, at the end of the test, Br increases from 5.71% to 9.97% as SC [2,6,67–71]. Liu et al. [69] and Liu et al. [70] proposed to replace
decreases from 1 to 0.859 at σc = 2 MPa, and Br increases from 26.51% plastic work with total energy input, because the elastic work is neg-
to 33.86% as SC decreases from 1 to 0.859 at σc = 5 MPa. The differ- ligible compared to the magnitude of plastic work.
ence of Br between spherical assembly and mixed assembly is clearly It is notable that the energy input during isotropic compression is
found to increase with increasing confining pressure and axial strain. also considered even though the evolution of PSD is only tracked during
shearing. From what have been investigated, particle breakage is
principally influenced by particle strength and stress state, especially
5. Correlation between particle breakage and energy input
under higher pressure [2,7]. Lobo-Guerrero and Vallejo [72] found that
more crushing was observed when a deviator stress was imposed rather
As investigated by Lade and Yamamuro [6], the total energy input
than isotropic condition. Yu [2] also found that shearing was much
per unit volume during a conventional triaxial test is the sum of energy
more important in causing particle breakage than isotropic consolida-
of isotropic compression and shearing
tion. This reflect the fact that fewer particle breakage is found during
ET = EC + ES (5) isotropic compression and this does not imply the absence of particle
breakage after isotropic compression.
where ET is the total energy input per unit volume, EC is the energy

8
T. Zhang, et al. Computers and Geotechnics 122 (2020) 103542

100 100

Percent finer by weight (%)

Percent finer by weight (%)

80 80
0.0% 0.0%
60 2.3% 60 2.3%
4.7% 4.7%
7.1% 7.1%
40 9.5% 40 9.5%
11.9% 11.9%
20 14.3% 20 14.3%
16.7% 16.7%
19.1% 19.1%
0 0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8
Particle size (mm) Particle size (mm)
(a) (b)
100 100

Percent finer by weight (%)

Percent finer by weight (%)

80 80
0.0% 0.0%
2.3% 60 2.3%
60
4.7% 4.7%
7.1% 7.1%
40 9.5% 40 9.5%
11.9% 11.9%
14.3% 14.3%
20 20
16.7% 16.7%
19.1% 19.1%
0 0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8
Particle size (mm) Particle size (mm)
(c) (d)
Fig. 13. Evolutions of particle size distribution curves for spherical assemblies: (a) σc = 2 MPa; (b) σc = 3 MPa; (c) σc = 4 MPa; (d) σc = 5 MPa.

Herein the correlations between Br and ET for spherical assembly distribution of sphericity of real sands. The main conclusions are
and mixed assembly under different confining pressures are presented summarized as follows:
in Fig. 17. It is clearly that Br increases gradually with increasing total
energy input, which is consistent with a hyperbolic equation. More 1. The results of single particle compression tests show that spherical
importantly, the extent of particle breakage is approximately related to particle has greater crushing strength and higher Weibull modulus
energy input, independent of confining pressure. The data distribution than those of non-spherical particles. Comparable macro-mechan-
may be fitted well by the following hyperbolic equation ical results are found between DEM and experimental observations,
including non-linear stress–strain relationship, peak deviatoric
ET
Br = stress, and peak friction angle.
a + bET (6)
2. An increase in combined sphericity SC at a given confining pressure
where a and b are the hyperbolic curve-fit parameters. Obviously, results in a reduction in peak friction angle. Higher confining
the values of parameter b are close to zero, which stands for the ap- pressure and larger axial strain are found to have a significant in-
proximately linear relationship between Br and ET. Similar relationship fluence on particle breakage, which simultaneously reflects the
has been observed in experimental test [2] and DEM simulation result variations of broken bonds, agglomerate numbers and particle size
[26]. However, the values of parameter a, which determine the slope of distribution. More broken bonds and agglomerates numbers are
fitting curves, have a significant difference between spherical assembly generated in mixed assembly than that in spherical assembly.
and mixed assembly. Consequently, particle breakage is more likely to 3. The PSDs, which transform to more well-graded distributions with
occur in mixed assembly with the same energy input. Fig. 17(b) shows increasing confining pressure, axial strain, and particle roughness,
that the value of parameter a for mixed assembly is 8.785, which is are used to quantify particle breakage by relative breakage index
closer to real silica sands (a = 7.847) obtained by Yu [2]. This finding (Br), and larger Br is obtained in mixed assembly.
proves that the proposed generation method of mixed assembly is more 4. The hyperbolic relationship between relative breakage and energy
powerful and reasonable to reflect the crushing behaviors than sphe- input is found to have relevance to particle shape, and the fitting
rical assembly. result of mixed assembly is in better accordance with real sands. The
proposed generation method of mixed assembly is more powerful
6. Conclusions and reasonable to reflect the crushing behaviors than spherical as-
sembly.
This paper discussed the effect of particle shape on particle breakage
of granular materials under triaxial tests using 3D discrete element
method (DEM). Mixed assembly is generated in terms of cumulative

9
T. Zhang, et al. Computers and Geotechnics 122 (2020) 103542

100 100

Percent finer by weight (%)

Percent finer by weight (%) 80 80
0.0% 0.0%
60 2.3% 60 2.3%
4.7% 4.7%
7.1% 7.1%
40 9.5% 40 9.5%
11.9% 11.9%
20 14.3% 14.3%
20
16.7% 16.7%
19.1% 19.1%
0 0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8
Particle size (mm) Particle size (mm)
(a) (b)
100 100
Percent finer by weight (%)

Percent finer by weight (%)

80 80
0.0% 0.0%
60 2.3% 60 2.3%
4.7% 4.7%
7.1% 7.1%
40 9.5% 40 9.5%
11.9% 11.9%
20 14.3% 20 14.3%
16.7% 16.7%
19.1% 19.1%
0 0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8
Particle size (mm) Particle size (mm)
(c) (d)
Fig. 14. Evolutions of particle size distribution curves for mixed assemblies: (a) σc = 2 MPa; (b) σc = 3 MPa; (c) σc = 4 MPa; (d) σc = 5 MPa.

100
B A
0.075mm
80
Percent finer by weight (%)

60

40 Initial state
After test
D
20 Total breakage: Bt=Area(ADCA)
Breakage potential: Bp=Area(ABCA)
C Relative breakage: Br = Bt / Bp
0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8
Particle size (mm)
Fig. 16. Relative breakage versus axial strain.
Fig. 15. Computational method of relative breakage, Br.

Declaration of Competing Interest

CRediT authorship contribution statement

The authors declare that they have no known competing financial

Tao Zhang: Conceptualization, Investigation, Writing - original
interests or personal relationships that could have appeared to influ-
draft, Writing - review & editing. Chi Zhang: Resources, Supervision,
ence the work reported in this paper.
Funding acquisition. Jiuqun Zou: Validation, Formal analysis, Data
curation. Baosheng Wang: Data curation, Visualization. Fangnian
Song: Project administration. Weihao Yang: Methodology, Software, Acknowledgements
Funding acquisition.
The authors gratefully acknowledge the support received from

10
T. Zhang, et al. Computers and Geotechnics 122 (2020) 103542

Fig. 17. Correlations between Br and ET: (a) spherical assembly; (b) mixed assembly.

National Key Research and Development Program of China (Grant simulated using DEM. Géotechnique 2004;54(2):131–41.
2016YFC0600904), National Natural Science Foundation of China [24] Bolton MD, Nakata Y, Cheng YP. Micro- and macro-mechanical behaviour of DEM
crushable materials. Géotechnique 2008;58(6):471–80.
(Grant 51804300), and China Postdoctoral Science Foundation (Grant [25] Laufer I. Grain crushing and high-pressure oedometer tests simulated with the
2018M632418). discrete element method. Granul Matter 2015;17(3):389–412.
[26] Liu Y, Liu H, Mao H. DEM investigation of the effect of intermediate principle stress
on particle breakage of granular materials. Comput Geotech 2017;84:58–67.
References [27] de Frias Lopez R, Larsson S, Silfwerbrand J. A discrete element materials model
including particle degradation suitable for rockfill embankments. Comput Geotech
[1] Hardin BO. Crushing of soil particles. J Geotech Eng 1985;111(10):1177–92. 2019;115:103166.
[2] Yu F. Characteristics of particle breakage of sand in triaxial shear. Powder Technol [28] McDowell GR, de Bono JP. On the micro mechanics of one-dimensional normal
2017;320:656–67. compression. Géotechnique 2013;63(11):895–908.
[3] Xiao Y, Liu H. Elastoplastic constitutive model for rockfill materials considering [29] de Bono JP, McDowell GR. DEM of triaxial tests on crushable sand. Granul Matter
particle breakage. Int J Geomech 2017;17(1):04016041. 2014;16(4):551–62.
[4] Xiao Y, Sun Z, Stuedlein AM, Wang C, Wu Z, Zhang Z. Bounding surface plasticity [30] Zhou W, Yang L, Ma G, Chang X, Cheng Y, Li D. Macro–micro responses of crushable
model for stress-strain and grain-crushing behaviors of rockfill materials. Geosci granular materials in simulated true triaxial tests. Granul Matter
Front 2019. https://doi.org/10.1016/j.gsf.2018.1010.1010. 2015;17(4):497–509.
[5] Hagerty MM, Hite DR, Ullrich CR, Hagerty DJ. One-dimensional high-pressure [31] Latham JP, Munjiza A, Garcia X, Xiang JS, Guises R. Three-dimensional particle
compression of granular media. J Geotech Eng 1993;119(1):1–18. shape acquisition and use of shape library for DEM and FEM/DEM simulation.
[6] Lade PV, Yamamuro JA, Bopp PA. Significance of particle crushing in granular Miner Eng 2008;21(11):797–805.
materials. J Geotech Eng 1996;122(4):309–16. [32] Rossi AM, Hirmas DR, Graham RC, Sternberg PD. Bulk density determination by
[7] Luzzani L, Coop MR. On the relationship between particle breakage and the critical automated three-dimensional laser scanning. Soil Sci Soc Am J 2008;72(6):1591–3.
state of sands. Soils Found 2002;42(2):71–82. [33] Asahina D, Taylor MA. Geometry of irregular particles: direct surface measurements
[8] Yu F. Stress-dilatancy behavior of sands incorporating particle breakage. Acta by 3-D laser scanner. Powder Technol 2011;213(1–3):70–8.
Geotech Slov 2017;14(1):55–61. [34] Hirmas DR, Gimnez D, Mome Filho EA, Patterson M, Drager K, Platt BF, et al.
[9] Cavarretta I, O'Sullivan C, Coop MR. The relevance of roundness to the crushing Quantifying soil structure and porosity using three-dimensional laser scanning. In:
strength of granular materials. Geotechnique 2017;67(4):301–12. Hartemink AE, Minasny B, editors. Digital soil morphometrics. Progress in soil
[10] Wang W, Coop MR. An investigation of breakage behaviour of single sand particles science. Springer International; 2016. p. 19–36.
using a high-speed microscope camera. Géotechnique 2016;66(12):984–98. [35] Zhou B, Wang J, Zhao B. Micromorphology characterization and reconstruction of
[11] Todisco MC, Wang W, Coop MR, Senetakis K. Multiple contact compression tests on sand particles using micro X-ray tomography and spherical harmonics. Eng Geol
sand particles. Soils Found 2017;57(1):126–40. 2015;184(14):126–37.
[12] Karatza Z, Andò E, Papanicolopulos SA, Viggiani G, Ooi JY. Effect of particle [36] Zhao B, Wang J. 3D quantitative shape analysis on form, roundness, and com-
morphology and contacts on particle breakage in a granular assembly studied using pactness with μCT. Powder Technol 2016;291:262–75.
X-ray tomography. Granul Matter 2019;21(3):44. [37] Zhou B, Wang J. Generation of a realistic 3D sand assembly using X-ray micro-
[13] Rothenburg L, Bathurst RJ. Micromechanical features of granular assemblies with computed tomography and spherical harmonic-based principal component analysis.
planar elliptical particles. Géotechnique 1992;42:79–95. Int J Numer Anal Meth Geomech 2017;41(1):93–109.
[14] Ng TT. Numerical simulation of granular soil using elliptical particles. Comput [38] Cheng Z, Wang J. A particle-tracking method for experimental investigation of ki-
Geotech 1994;16:153–69. nematics of sand particles under triaxial compression. Powder Technol
[15] Mirghasemi AA, Rothenburg L, Matyas EL. Numerical simulations of assemblies of 2018;328:436–51.
two-dimensional polygon-shaped particles and effects of confining pressure on [39] Fonseca J, O'Sullivan C, Coop MR, Lee PD. Non-invasive characterization of particle
shear strength. Soils Found 1997;37:43–52. morphology of natural sands. Soils Found 2012;52(4):712–22.
[16] Mirghasemi AA, Rothenburg L, Matyas EL. Influence of particle shape on en- [40] Cho NA, Martin C, Sego D. A clumped particle model for rock. Int J Rock Mech Min
gineering properties of assemblies of two-dimensional polygon-shaped particles. Sci 2007;44:997–1010.
Géotechnique 2002;52(3):209–17. [41] Lu M, McDowell GR. The importance of modelling ballast particle shape in the
[17] Hosseininia ES, Mirghasemi AA. Numerical simulation of breakage of two-dimen- discrete element method. Granul Matter 2007;9:69–80.
sional polygon-shaped particles using discrete element method. Powder Technol [42] Manso J, Marcelino J, Caldeira L. Crushing and oedometer compression of rockfill
2006;166:100–12. using DEM. Comput Geotech 2018;101:11–22.
[18] Hosseininia ES, Mirghasemi AA. Effect of particle breakage on the behavior of si- [43] Fu R, Hu X, Zhou B. Discrete element modeling of crushable sands considering
mulated angular particle assemblies. Particuology 2007;5:328–36. realistic particle shape effect. Comput Geotech 2017;91:179–91.
[19] Ueda T, Matsushima T, Yamada Y. DEM simulation on the one-dimensional com- [44] Wadell H. Volume, shape, and roundness of rock particles. J Geol
pression behavior of various shaped crushable granular materials. Granul Matter 1932;40(5):443–51.
2013;15:675–84. [45] Bullard JW, Garboczi EJ. Defining shape measures for 3D star-shaped particles:
[20] Robertson D, Bolton MD. DEM simulations of crushable grains and soils. Proc. sphericity, roundness, and dimensions. Powder Technol 2013;249:241–52.
Powders and Grains 2001;Sendai,:623–6. [46] Trahan J, Graziani A, Goswami DY, Stefanakos E, Jotshi C, Goel N. Evaluation of
[21] Harireche O, McDowell GR. Discrete element modelling of cyclic loading of pressure drop and particle sphericity for an air–rock bed thermal energy storage
crushable aggregates. Granul Matter 2003;5(3):147–51. system. Energy Procedia 2014;57:633–42.
[22] Cheng YP, Nakata Y, Bolton MD. Discrete element simulation of crushable soil. [47] Zheng J, Hryciw RD. Traditional soil particle sphericity, roundness and surface
Géotechnique 2003;53:633–41. roughness by computational geometry. Géotechnique 2015;65(6):494–506.
[23] Cheng YP, Bolton MD, Nakata Y. Crushing and plastic deformation of soils [48] Suh HS, Kim KY, Lee J, Yun TS. Quantification of bulk form and angularity of
particle with correlation of shear strength and packing density in sands. Eng Geol

11
T. Zhang, et al. Computers and Geotechnics 122 (2020) 103542

2017;220:256–65. 2019;145(2):04018105.
[49] Rorato R, Arroyo M, Andò E, Gens A. Sphericity measures of sand grains. Eng Geol [61] Varadarajan A, Sharma KG, Venkatachalam K, Gupta AK. Testing and modeling two
2019;254:43–54. rockfill materials. J Geotech Geoenviron Eng 2003;129(3):206–18.
[50] Wang J, Yan H. On the role of particle breakage in the shear failure behavior of [62] Gong J, Liu J, Cui L. Shear behaviors of granular mixtures of gravel-shaped coarse
granular soils by DEM. Int J Numer Anal Meth Geomech 2013;37(8):832–54. and spherical fine particles investigated via discrete element method. Powder
[51] Nakata Y, Hyde AFL, Hyodo M, Murata H. A probabilistic approach to sand particle Technol 2019;353:178–94.
crushing in the triaxial test. Géotechnique 1999;49(5):567–83. [63] Ghafghazi M, Shuttle DA, DeJong JT. Particle breakage and the critical state of
[52] McDowell GR, Amon A. The application of Weibull statistics to the fracture of soil sand. Soils Found 2014;54(3):451–61.
particles. Soils Found 2000;40(5):133–41. [64] Xu M, Hong J, Song E. DEM study on the effect of particle breakage on the macro-
[53] Lim WL, McDowell GR, Collop AC. The application of Weibull statistics to the and micro-behavior of rockfill sheared along different stress paths. Comput Geotech
strength of railway ballast. Granul Matter 2004;6(4):229–37. 2017;89:113–27.
[54] Xiao Y, Sun Z, Desai CS, Meng M. Strength and surviving probability in grain [65] Lee KL, Farhoomand I. Compressibility and crushing of granular soil. Can Geotech J
crushing under acidic erosion and compression. Int J Geomech 1967;4(1):68–86.
2019;19(11):04019123. [66] Yu F. Particle breakage and the critical state of sands. Géotechnique
[55] McDowell GR. On the yielding and plastic compression of sand. Soils Found 2017;67(8):713–9.
2002;42(1):139–45. [67] Zhou L, Chu X, Zhang X, Xu Y. Numerical investigations on breakage behaviour of
[56] Nakata Y, Kato Y, Hyodo M, Hyde AL, Murata H. One-dimensional compression granular materials under triaxial stresses. Geomech Eng 2016;11(5):639–55.
behaviour of uniformly graded sand related to single particle crushing strength. [68] Miura N, O-hara S. Particle crushing of a decomposed granite soil under shear
Soils Found 2001;41(2):39–51. stresses. Soils Found 1979;19(3):61–76.
[57] Jamiolkowski M, Kongsukprasert L, Lo Presti D. Characterization of gravelly geo- [69] Liu J, Liu H, Zou D, Kong X. Particle breakage and the critical state of sand: By
materials. Proceedings of the fifth international geotechnical conference, Bangkok, Ghafghazi, M., Shuttle, DA, DeJong, JT, 2014. Soils Found 54 (3), 451–461. Soils
Thailand, November. 2. Asian Institute of Technology; 2004. p. 29–56. Found 2015;55(1):220–2.
[58] Yang J, Wei LM. Collapse of loose sand with the addition of fines: the role of particle [70] Liu H, Zou D, Liu J. Constitutive modeling of dense gravelly soils subjected to cyclic
shape. Géotechnique 2012;62(12):1111–25. loading. Int. J Numer Anal Meth Geomech 2014;38(14):1503–18.
[59] Yang J, Luo XD. Exploring the relationship between critical state and particle shape [71] Zhang BY, Jie YX, Kong DZ. Particle size distribution and relative breakage for a
for granular materials. J Mech Phys Solids 2015;84:196–213. cement ellipsoid aggregate. Comput Geotech 2013;53:31–9.
[60] Xiao Y, Long L, Evans TM, Zhou H, Liu H, Stuedlein AW. Effect of particle shape on [72] Lobo-Guerrero S, Vallejo LE. Crushing a weak granular material: experimental
stress-dilatancy responses of medium-dense sands. J Geotech Geoenviron Eng numerical analyses. Géotechnique 2005;55(3):245–9.

12