Computers and Geotechnics 68 (2015) 137–146

Contents lists available at ScienceDirect

Computers and Geotechnics

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

Research Paper

Particle mechanics modeling of creep behavior of rockfill materials

under dry and wet conditions
Zhihong Zhao ⇑, Er-xiang Song
Department of Civil Engineering, Tsinghua University, Beijing 100084, China

a r t i c l e i n f o a b s t r a c t

Article history: Rockfill is an important construction material for infrastructure engineering, such as dams, railways and
Received 11 January 2015 airport foundations, which display a long-term post-construction settlement. However, the main mech-
Received in revised form 16 March 2015 anisms for rockfill creep and weathering influence still remain poorly understood. Particle mechanics
Accepted 10 April 2015
method is used to understand the rockfill creep process under dry and wet conditions. Different bond-ag-
ing models and wetting models that represent different degradation and weakening mechanisms are
compared, in order to clarify the principle and secondary mechanisms for rockfill creep and weathering
Keywords:
influence. The results show that rockfill aggregate breakage in terms of angularity abrasion is the main
Rockfill creep
Particle mechanics method
source for rockfill creep under dry state. Wetting can induce additional strain mainly due to the reduction
Water weakening of contact friction coefficient, i.e. lubrication, and the bond strength reduction just plays a secondary role
Wetting–drying cycles in producing additional strain. The earlier the wetting occurs during rockfill creep, the more rapidly the
Particle breakage rockfill becomes stable. The wetting–drying cycles can induce strain evolution in a ‘stepped’ way, which
Abrasion is in agreement with experimental observation. The practical implications from the modeling and the
outstanding issues in this study are also discussed.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction Over the years, many laboratory or in-situ experiments have

Rockfill materials, which are coarse and mainly composed of
quarry rock debris or crushed rock fragments, have been widely
used in many infrastructure projects, such as dams, railways and
airport foundations in mountain area. It is observed that creep set-
tlement at these infrastructures continues for a long period after
their construction [52,43]. The operability and safety of railways
and airports impose tight limits for post-construction settlements
of the supporting embankments. Significant post-construction set-
tlements may affect the infrastructure’s serviceability, or even
induce engineering disaster. In addition to time-dependent strains
under constant external loading, significant rockfill settlement can
also be caused by other environmental factors, such as rainfall or
flooding. There is therefore a pressing need to clearly understand
the physical and chemical mechanisms responsible for rockfill
creep under dry and wet conditions, in order to derive more com-
prehensive constitutive models, interpret safety monitoring
records without misleading conclusions, make more reliable pre-
dictions of creep settlement and minimize maintenance costs dur-
ing the infrastructure’s working life.
⇑ Corresponding author.
E-mail address: zhzhao@tsinghua.edu.cn (Z. Zhao).
been conducted to study the complex settlement behavior of rock-
fill-type materials (Takei et al. [53]; McDowell and Khan [40]; Lim
[26]; Ionescu [18]; Oldecop and Alonso [45]; Huang et al. [13]; Cao
[3]; among others). The results showed that rockfill creep, during
which crushing continuously occurs, is mainly affected by stress
level, initial density, particle characteristics, saturation and other
environmental factors such as temperature and freeze and thaw
actions. Based on these experimental results, a number of constitu-
tive models have been proposed to account for rockfill creep, in
terms of a logarithmic relationship between time and long-term
strain (e.g. [52,41,21]. However, due to the fact that physical tests
are not able to monitor the microscopic rockfill response at an indi-
vidual aggregate level, the nature of creep and the underlying
mechanism of degradation have not been fully understood. In this
way, most of the constitutive models are purely empirical and
lacking in theoretical support, and therefore cannot reflect the nat-
ure of the observed time-dependent rockfill settlement.
The particle mechanics method, as an alternative to laboratory or
in-situ experiments, has been used by many researchers to explore
the micro-mechanism of granular materials. There are mainly three
approaches to represent rockfill aggregates in particle mechanics
models: (1) the rockfill blocks are directly modeled using rigid cir-
cular or polygonal-shaped particles [55,42,30,29,50,56,14], but note

http://dx.doi.org/10.1016/j.compgeo.2015.04.008
0266-352X/Ó 2015 Elsevier Ltd. All rights reserved.
138 Z. Zhao, E.-x. Song / Computers and Geotechnics 68 (2015) 137–146
that the circular elements cannot consider the angularity of rockfill
blocks compared with polygonal elements; (2) using a ‘clump’,
which is a single entity of overlapping particles and is rigid inter-
nally and deformable at the external boundary, rockfill blocks of
irregular shapes can be modeled [6,31–34,17]; (3) one treats rockfill
blocks as clusters of bonded circular particles, which can reflect the
angularity of rockfill blocks to some extent [20,39,4,27,54,51,23,24].
It is known that crushing (i.e., grain fracture process) in a gran-
ular material subjected to compression is one of the main causes
for rockfill settlement. Particle mechanics method has also been
attempted to model rockfill block breakage. The basic elements
in particle mechanics models cannot break, but an alternative of
replacing the original particles that fulfill failure criterion with a
set of smaller particles was proposed to simulate particle breakage
[55,30,29]. However, this replacement cannot conserve mass bal-
ance. Clumps are also unbreakable, and this disadvantage restrains
their use in accurate modeling of mechanical behavior of rockfill
materials. The cluster of bonded circular particles can be regarded
as an cohesive granular aggregate, where micro-cracks can initiate
and propagate depending on the bond strengths and contact forces
and eventually the macro-cracks form [20,4,54,51,57].
Generally, the above studies have demonstrated that particle
mechanics method is able to and also a robust tool to simulate
the rockfill mechanical behavior. However, those studies mainly
focused on the rockfill behavior subjected to static and dynamic
loading, with rare consideration of rockfill creep and other weath-
ering factors like saturation or wetting–drying cycles. Kwok and
Bolton [23,24] studied soil (e.g., sand or clay) creep using particle
mechanics method, considering time-dependent contact friction
coefficient and bond strength. Tran et al. [54] and Silvani et al.
[51] incorporated bonding deterioration models into particles
mechanics method, in order to simulate the rockfill creep, but they
did not consider the time-dependent contact friction coefficient.
In addition, other numerical methods like the finite element
method incorporated with continuum damage mechanics [28]
and the combined finite-discrete element method [37,36] are also
able to simulate particle breakage, but they have not been widely
used to simulate rockfill mechanical behavior.
Overall, the main objective of this study is to further advance
the understanding of the micro-mechanisms of rockfill creep using
particle mechanics method, and the main novelty is to systemati-
cally study the effects of wetting and wetting–drying cycles on
rockfill settlement.
2. Methodology
2.1. Mechanism of rockfill creep
Creep settlement of rockfill ground is mainly due to delayed
grain fracture process [51], and particle breakage mainly occur in
the forms of abrasion and total fragmentation. Grain fracture pro-
cess is a rate-dependent process that can start at a relatively low
stress, and results in gradual changes in rockfill fabric and packing,
which is mainly governed by grain size and shape, the magnitudes
of the applied stresses, and the mineralogy and strengths of indi-
vidual grains.
For a brittle rock that is compressively loaded, two strengths
can be defined: (1) when the applied stress reaches the ‘short-term
strength’ of the rock block, failure occurs immediately; (2) other-
wise, failure progressively occurs when the applied stress reaches
a value defined as the ‘long-term strength’. The ratio of long-term
strength over short-term strength is always less than 1. In addition,
a threshold for the stress is defined above which strength degrada-
tion initiates [25,49]. This strength degradation process can be
understood as a process of crack growth due to stress corrosion,
from a microscopic perspective. A terminology of ‘subcritical crack
growth’ can be used, and crack growth velocity is zero for stress-in-
tensity factor less than its limit values. Serial laboratory tests car-
ried out on different rocks showed that wreathing processes, such
as water, temperature, freeze and thaw actions, can accelerate the
crack propagation velocity and then reduce the rock’s strength and
failure time. The influence of these actions is different for each
material and also depends on the state of the rock studied (rock
alteration, degree of saturation, temperature) [1,44].
2.2. Effects of water on rockfill creep
It was observed that an accelerated rate of compression was
induced by wetting the rockfill specimens [22,46]. Water acts as
lubricant at the interparticle contact points, and results in lower
friction coefficient [12]. On the other hand, water can cause the ini-
tial crushing strength of a single particle to reduce and strength
variability to increase, due to the following two weakening mech-
anisms: (1) physically the pressurized pore water can weaken and
embrittle rocks; (2) chemically the pore water can lower the frac-
ture energy [2]. In other words, by wetting rockfill materials,
mechanical parameters such as short term strength, long term
strength, threshold of strength degradation and contact friction
coefficient may suddenly decrease. Many experiments have found
that the stress–strain behavior of rockfill initially follows the dry
state curve, and is transformed to follow the wet state curve when
subjected to saturation at elevated stress level [11,38,9,10,18].
2.3. Modeling rockfill creep using particle mechanics method
The bond force vector (Fb) can be resolved into normal (Fn) and
shear (Fs) components with respect to the contact plane (Fig. 1a),
and thus the stress state (r) in a parallel bond can be expressed as,
r ¼ rn~
n þ s~
s ð1Þ
where rn and s represent the normal and shear stresses, respec-
tively; ~
n and ~s are the unit vectors that define the contact plane.
If the tensile stress exceeds the tensile strength (rn P rbc) or the
shear stress exceeds the shear strength (s P sbc), then the parallel
bond breaks, and it is removed from the model along with its
accompanying force, moment and stiffness. A few bond-ageing
models, in terms of reducing parallel bond strengths or diameter
(Fig. 1b), have been developed in particle mechanics method
[48,54,51], some of which have been used to model rockfill creep.
Below these bond-ageing models are briefly reviewed and
compared.
2.3.1. Bond-ageing model in Tran et al. [54]
Tran et al. [54] proposed a bond-ageing law for the evolution of
bond strength in the following form,
8
>
> r0b b1 r0b > rn
>
<    
Rt
rtb ¼ r0b 1  b3 t0
exp b2 rrn0  b1 dt r0b  rn  b1 r0b ð2aÞ
>
> b
>
:
0 rn > r0b
8
>
> r0b b1 r0b > rn
>
<    
Rt
stb ¼ s0b 1  b3 t0 exp b2 rrn0  b1 dt r0b  rn  b1 r0b ð2bÞ
>
> b
>
:
0 rn > r 0
b
where r0b and rtb are the short- and long-term normal strength,
respectively; s0b and stb are the short- and long-term shear strength,
respectively; b1, b2 and b3 are three empirical parameters, which
depend on aggregate’s size and shape and the material properties.
 Z. Zhao, E.-x. Song / Computers and Geotechnics 68 (2015) 137–146 139

s resist, |Fs| represents the absolute value of shear force. The bond
damage occurs when g0 P 0, and the bond breaks if gu P 0. The evo-
n
Fb lution law of D is in the following form,
8
>
> D0 gu < g0 < 0
>
< h   ih i
1 mt jF s j=lþF n
D¼ D0 þ m
exp g  1 C0
 ð1  mD0 Þ g u < 0  g 0
d >
>  
>
: 1 jF j= lþF
m
1  s Cu n ! m1 0  gu < g0
ð6Þ
(a) A parallel bond
 
where g is a characteristic time; the term 1
m
1  jF s j=CluþF n is a value
of ultimate damage, denoted as Dc, and is determined according to
gu = 0. Note that D may jump from its ultimate value Dc to its max-
imum value 1/m. For a bond under a condition of g0 P 0 and gu < 0,
bond breakage occurs in finite time, which is determined as follow,
"  1 #  
Initial parallel bond g jF s j=l þ F n g C0
tf ¼ ln 1 þ 1 þ ln 1  ð7Þ
m ð1  mD0 ÞC 0 m Cu

2.3.3. Diameter-ageing model in Potyondy [48]

Based on stress corrosion theory, a corrosion rate law for the
bond diameter, dt, was proposed by Potyondy [48],
8 0
>
>
<
d
 
ra > rn
t 0 Rt
d ¼ d  a1 t0
exp a2 rrna dt r  rn  ra
0
b ð8Þ
>
>
:
0 rn > r 0
b
Bond degradation in terms Bond degradation in terms 0
where d is the initial parallel bond diameter; rn is the normal
of strength reduction of diameter reduction stress acting on the parallel bond; ra is a threshold stress, below
which the stress corrosion reaction ceases; a1 and a2 are two rate
(b) Two approaches considering bond degradation constants. As damage proceeds, the effective bond stiffness
Fig. 1. Conceptual diagram of parallel bond degradation in terms of bond or decreases, which allows a macroscopic load redistribution to occur
diameter reduction. throughout the material. The elapsed time to bond failure can be
determined by,
The parameter b1 defines stress threshold for strength degradation 0 0  
(or called activation stress). If the applied normal stress r exceeds d d rn
tf ¼ exp a2 ð9Þ
r0b , the parallel bond breaks and rtb becomes zero suddenly. When a1 ra
the applied normal stress rn is larger than b1 r0b but smaller than 0
where d is the parallel bond diameter at failure. Note that in this
r0b during the period [t0, t], the parallel bond strength decreases model bond diameter is degraded instead of bond strengths.
with time, and the time corresponding to the breakage (tf = t  t0)
of a given bond under normal stress rn can be determined by, 2.3.4. Comparison
1  rrn0 For the above three damage-rate laws, the elapsed times to bond
tf ¼  b
 ð3Þ failure under normalized normal stresses/forces are compared in
b3 exp b2 rrn0  b1 Fig. 2. If proper values were given to the empirical parameters in
b these bond-ageing models, similar elapsed times could be obtained
from the three models. This means that any one of them can be used
to study the rockfill creep in a generic study, if the empirical param-
2.3.2. Bond-ageing model in Silvani et al. [51]
eters could be calibrated with experiments. In this paper, bond-
A damage bond model that is thermodynamically consistent
ageing model in Tran et al. [54] was used because of its relatively
was proposed by Silvani et al. [51], in which an interface damage
simple mathematical form and clear physical meaning.
term D changing in [0, 1/m] is defined to describe the surface den-
sity of the microdefects occurring at the bond. Note that m > 0 is a
2.4. Modeling water weakening in particle mechanics model
softening parameter, and the bond is broken and replaced by a fric-
tional contact when D = 1/m. Instead of explicitly expressing the
Water weakening can basically be characterized as two mecha-
short- and long-term strengths, two implicit failure criteria are
nisms: reducing fracture energy and friction coefficient, and these
expressed in terms of normal force Fn and shear force Fs on the
two mechanisms are positively correlated, i.e. the operative mech-
bond, which are corresponding to the short- and long-term
anisms are effective on both crack propagation and frictional slid-
strengths, respectively,
ing [2]. In order to consider effects of wetting on bond degradation
g 0 ðF s ; F n ; DÞ ¼ l1 jF s j þ F n  C 0 ð1  mDÞ ð4Þ and rockfill arrangement, both normal and shear bond strengths
and contact friction coefficient are reduced by a specified ratio,
g u ðF s ; F n ; DÞ ¼ l1 jF s j þ F n  C u ð1  mDÞ ð5Þ k1 , after an instantaneous and homogeneous wetting. It is assumed
that after a wetting–drying cycle, both normal and shear bond
where l is the contact friction coefficient; C0 is the initial damage strengths and contact friction coefficient partially convert back to
threshold; and Cu P C0 is the maximum tensile force a bond can their initial values. Compared with the bond-ageing model
140 Z. Zhao, E.-x. Song / Computers and Geotechnics 68 (2015) 137–146

1.0 A rockfill aggregate is represented by an two-dimensional dense

Tran et al. (2009) packing assembly of circular disks of non-uniform sizes, and the
0.9
Silvani et al. (2009) disks are bonded in the normal and shear directions at all contacts
0.8 Potyondy (2007) that possess finite (normal and shear) stiffness and (tensile and
0.7 shear) strengths [47]. The contact points between rockfill aggre-
gates are assigned zero bond strengths, but a nonzero friction coef-
0.6 ficient. In this way, the surface roughness of rockfill aggregates can
tf (s)

0.5 be naturally simulated by the arrangement of disks, but the irreg-

ular shapes of rockfill aggregates cannot be modeled perfectly.
0.4
More details about the assumptions and laws of particle flow can
0.3 be found in the literature (e.g., [19,47]). This section mainly focuses
on the model setup and modeling procedure of rockfill creep dur-
0.2
ing one-dimensional confined compression.
0.1 According to the basic specimen-genesis procedure in Potyondy
0.0
and Cundall [47] and Itasca [19], 5000 disks were packed into a
1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3
square box with a side length of 200 mm to build up rock matrix,
through four steps, i.e., compacting initial assembly, establishing
Normalized normal stress/force
specified isotropic stress, removing ‘floating’ particles and instal-
Fig. 2. Comparison of elapsed time to bond failure between three bond-aging ling parallel bonds. The average disk radius was 1.4 mm, with max-
models. (1) Bond-ageing model in Tran et al. [54] (Eq. (2)): b1 = 0.3, b2 = 4.0, b3 = 0.7, imum and minimum of 1.8 mm and 1.1 mm, respectively. By
rn is normalized by r0b . (2) Bond-ageing model in Silvani et al. [51] (Eq. (10)): projecting a sketch of artificial rockfill specimen including 120
g = 0.5, m = 1.0, D0 = 0, Fs = 0, C0/Cu = 0.3, Fn is normalized by Cu. (3) Diameter-ageing aggregates on the particle mechanics model (Fig. 3a), and removing
model in Potyondy [48] (Eq. (9)): a1 = 0.25 and a2 = 5, rn is normalized by ra.
the disks located in the voids, a numerical rockfill specimen was
generated (Fig. 3b). The average radius of rockfill aggregates was
accounting for wetting in Tran et al. [54], we consider water weak-
about 9.18 mm, with the maximum and minimum of 11.18 and
ening by reducing bond strengths instead of three empirical
6.82 mm, respectively. Other micro-parameters are listed in
parameters (b1, b2 and b3). Compared with the way of accounting
Table 1, the values of which were the same or similar to those in
for wetting in Silvani et al. [51], short-term strength, stress thresh-
the literature pertaining to rock behavior modeled using PFC
old for strength degradation and contact friction coefficient
[57]. Numerical biaxial tests [19] were carried out, and showed
decrease simultaneously in this study, instead of only short-term
that the micro-parameters can reflect the nature of rockfill
strength or long-term strength.
materials.

3. Modeling
3.1. Model setup
Cao [3] conducted large scale compression experiments on
rockfill specimens, under dry and wet conditions, respectively.
The results showed that wetting can speed up creep, and thus
cause a larger creep strain. Wetting–drying cycles can induce creep
strain evolve in a pattern of ‘steps’. In this study, rockfill creep
under dry and wet conditions was modeled using two-dimensional
particle flow code, PFC2D [19], and was compared with experimen-
tal observation in Cao [3].
3.2. Modeling of one-dimensional compressional creep
During one-dimensional compression, the normal stresses on
the upper wall gradually increased to the required values and then
remained constant using the servo control algorithm [19]. Once
bond degradation became active, micro-cracking in rockfill aggre-
gates may be induced and macroscopic settlement of rockfill spec-
imen occurred. The development of micro-cracks and breakage of
rockfill aggregates, normal displacement of the upper wall and
bond force chain in the rockfill specimen were monitored during
the creep test.

(a) Artificial rockfill specimen (b) Numerical rockfill specimen

Fig. 3. Rockfill specimen consisting of 120 aggregates in a squared box of side length 200 mm.
 Z. Zhao, E.-x. Song / Computers and Geotechnics 68 (2015) 137–146 141

During the creep test, mechanical and bond degradation calcula-
tions were run circularly [48]. During mechanical calculation, bond
degradation calculation was hung on, and sufficient time steps are
cycled until a quasi-static state was reached. During bond degrada-
tion calculation, the state variables like forces and displacements
were kept constant, and bond strengths were adjusted based on
the bond-aging law and state variables. This approximation is rea-
sonable because mechanical process is relatively shorter compared
with bond degradation process. Note that the time step in mechan-
ical calculation is a ‘false’ time, whereas the time step in bond
degradation calculation represents the real time. The first false time
can be understood as the ‘numerical time’ that is needed for the
computer to make the system reach the balanced state. The latter
real time is the ‘creep time’ during which the bond strength evolves
but there is no evolution of the whole granular structure [54].
Alternatively, the time step in bond degradation calculation can
be fixed at a constant value or self-adaptively adjusted. Potyondy
[48] proposed a self-adaptive procedure based on the estimated
elapsed time (tf),
Dt b ¼ t f =nc
where Dtb is the time step for bond degradation calculation; tf can
be calculated using Eq. (4) and represents the estimated elapsed
Table 1
Micro-parameters for the rockfill agglomerates.
Parameter (unit) Value
Number of balls (–) 5000
Density (kg/m3) 2230
Contact modulus (GPa) 4.0
Ratio of ball shear to normal stiffness (–) 0.4
Contact friction coefficient (–) 0.3
Parallel bond radius multiplier (–) 1.0
Parallel bond modulus (GPa) 4.0
Ratio of parallel bond shear to normal stiffness (–) 0.4
Parallel bond normal strength, mean (MPa) 10.0
Parallel bond normal strength, standard deviation 3.0
(MPa)
Parallel bond shear strength, mean (MPa) 10.0
Parallel bond shear strength, standard deviation 3.0
(MPa)
Empirical parameters b1, b2 and b3 0.4, 40 and 5  1011
time to the next parallel bond failure; nc is the number of cycles
until the next parallel bond breaks.
3.3. Modeling of wetting and wetting–drying cycles
Since no models that consider the effects of wetting or wetting–
drying cycles on rockfill creep have been widely used, the wetting
model in Tran et al. [54] was used and some new models based on
the water weakening mechanisms in Baud et al. [2] are explored:
(1) Water weakening model I (W1), Tran et al. [54] took into
account the wetting effects by decreasing b2 and increasing b3,
i.e., changing the degradation rate; (2) Water weakening model II
(W2), only bond strengths decrease by k1 due to the lower specific
surface energy in a fluid environment; (3) Water weakening model
III (W3), only contact friction coefficient decreases by k1 ; (4) Water
weakening model IV (W4), both contact friction coefficient and
bond strengths decrease by k1 , which can be regarded as an addi-
tion of W2 and W3. The reduction ratio of 4% was chosen according
to Baud et al. [2], which showed that the ratio of the values of sand-
stone mechanical parameters under wetting state over that under
ð10Þ dry state is about 0.83–0.97. Through comparing the obtained
results by using different models, it is intended to clarify the dom-
inant weakening mechanisms during wetting.
When one dimensional compressional creep started, wetting
occurred instantly at a specified time, which was modeled using
the above four models, respectively. It was assumed that the rock-
fill specimen can be wetted in a sufficiently short time that was
neglected in this study. For each wetting–drying cycle, firstly the
above wetting models were used for wetting, and then the param-
eters were partially converted back to their original values for
modeling drying process. This is supported by the recent experi-
mental studies which showed that rock mechanical properties
gradually decreased with the increasing numbers of wetting–dry-
ing cycles [7,8].
4. Results
4.1. Rockfill creep under one dimensional compression
The creep strain–time curves under different constant normal
stresses of 0.94 MPa and 1.14 MPa are presented in Fig. 4, and both
primary creep and secondary creep phases were observed.
Generally, the higher the normal stress is, the larger are the creep
strain rate during primary creep phase and the eventual creep

0 0
Experiment (Cao, 2011) Experiment (Cao, 2011)
PFC (dt = 100 s) -1 PFC (dt = 100 s)
-1
PFC (adaptive dt) PFC (adaptive dt)
-2
-2
Strain (10-4)

Strain (10-4)

-3
-3
-4

-4
-5

-5 -6

-6 -7
0 20 40 60 80 100 120 0 20 40 60 80 100
Time (minute) Time (minute)
(a) Normal stress of 0.94 MPa (b) Normal stress of 1.14 MPa
Fig. 4. Rockfill creep under dry condition.
142 Z. Zhao, E.-x. Song / Computers and Geotechnics 68 (2015) 137–146

strain. This indicates that normal stress is a critical factor that not
only determines the absolute value of creep strain, but also affects
the rate of creep strain evolution. Compared with experimental
curves, the creep strain curves exhibited a drastic ‘jump’, which
is probably due to rockfill aggregate crushing and rearrangement.
The curves of creep strain that were obtained by employing fixed
or self-adaptive time steps for bond degradation calculation, are
also compared in Fig. 4. Even though both approaches can capture
the main features of rockfill creep, some differences existed during
the primary creep phase. For the modeling in this study, the calcu-
lation time by both approaches were similar. In other words, the
80
60
Micro-crack number
procedure of self-adaptive time steps did not improve the calcula-
tion efficiency significantly. The scheme of fixed time step was
used in modeling rockfill creep under wet condition, and thus
the wetting and drying time can be accurately controlled.
Using the case under normal stress of 1.14 MPa to explain the
microscopic mechanisms responsible for rockfill creep, the axial
strain and the number of micro-cracks are plotted in Fig. 5. The
similarity between the strain and micro-crack curves indicates that
the reduction of parallel-bond strength with time and the forma-
tion of micro-cracks are the main mechanism of producing creep
[48]. Micro-cracks distribution is presented in Fig. 6, which shows
that the micro-cracks formed mainly due to tensile failure. Micro-
cracks mainly occurred in those rockfill aggregates which are on
-7
the force chains (Fig. 6). Rockfill aggregate breakage occurred in
the form of abrasion (angularity/corner breakage), which was also
-6 observed in Takei et al. [53]. However, total fragmentation (particle
splitting) was not found in this study. Indraratna et al. [15] also
showed that most ballast degradation under cyclic loading is pri-

-5
marily the consequence of corner breakage, rather than particle
Strain (10-4)
-4 splitting.
40
-3 4.2. Effects of wetting and wetting–drying cycles on rockfill creep

-2 When a dry rockfill specimen was saturated during one dimen-

20
sional compression creep, an initial sudden settlement was
-1
Micro-crack number observed, especially when the wetting model W4 was used
Creep strain
(Fig. 7a). Using the wetting model W4 predicted the largest strain,
0 0
0 20 40 60 80 100 whereas the smallest strain was obtained when the wetting model
Time (minute) W1 was used. The similar strains were obtained by employing the
wetting models W3 and W4, but the strain increased more rapidly
Fig. 5. The number of micro-crack increase during rockfill creep under normal for W4, immediately after wetting. From the above comparison, it
loading of 1.14 MPa. may be concluded that the initial sudden settlement is primarily

(a) Bond force distribution (b) Micro-crack distribution

(c) Enlarged micro-cracking zones (d) Agglomerate crushing (Takei et al. 2001)

Fig. 6. Rockfill aggregate breakage during one dimensional compression creep under normal loading of 1.14 MPa. Black and red short lines indicate the tensile and shear
micro-cracks, respectively. The angularity (sharp corners) of aggregate leads to stress concentrations and breakage.
 Z. Zhao, E.-x. Song / Computers and Geotechnics 68 (2015) 137–146 143

0
Dry
-1 W1
W2
-2 W3
W4
-3
Strain (10-4)
-4

-5

-6

-7

-8

0 40 80 120 160 200

Time (minute)
(a) Strain-time curve

(b) Bond force distribution (c) Micro-crack distribution

(d) Enlarged micro-cracking zones (left: before wetting; right: after wetting)
Fig. 7. Effects of wetting on rockfill creep using different wetting models.

induced by the reduced contact friction coefficient, which results
into rockfill sliding and rearrangement. Wetting induced weaken-
ing (bond strength reduction) plays a secondary role in speeding
up the creep rate. Due to wetting, some bond force chains changed
and more micro-cracks at rockfill aggregate angularities were
induced (Fig. 7b–d).
The effect of wetting time when the wetting occurred during
creep was also studied, and the corresponding strain evolution
curves modeled using W4 are compared in Fig. 8. A similar final
creep strain was obtained for different wetting times. During the
primary creep phase, micro-cracking and stress redistribution
occurred in a drastic way, wetting could further accelerate the
micro-cracking propagation speed. Therefore, rockfill specimen
that was wetted earlier can reach the secondary creep phase more
quickly. However, if wetting occurs during the secondary creep
phase, the stable compact state of rockfill was broken again.
Additional time was required for the rockfill specimen to reach
another stable state. Wetting-induced strain is strongly dependent
on the stress state and aggregate structure.
The effects of wetting–drying cycles on rockfill creep were also
modeled using W4, and the creep–strain curves are compared with
the experimental results [3] in Fig. 9. Two wetting–drying cycles
were considered in this study. Similar to the experimental observa-
tion, the numerical results also show that the significant strains
were induced during wetting–drying cycles, and the induced strain
in the second wetting–drying cycle was smaller than that in the
first one. During each wetting–drying cycle, both primary creep
and secondary creep phases were observed. Compared with exper-
imental results [3], the wetting induced strain increased more
rapidly in the numerical modeling. The possible reason is that
144 Z. Zhao, E.-x. Song / Computers and Geotechnics 68 (2015) 137–146

the numerical rockfill specimen was assumed to become wet 5. Discussion

instantly, but more time was needed to wet the rockfill specimen
during experiments. 5.1. Micro-mechanism for rockfill creep under dry state

0 Luong [35] summarized that the deformation of granular mate-

rials under loading mainly results from three mechanisms: (1) con-
Wetting time
-1 solidation – the change in volume of particle assemblies; (2)
Dry
20 min distortion – the change in aggregate shape due to sliding and roll-
-2
40 min ing; and (3) attrition – particle crushing and breakage leading to
-3 60 min rearrangement and compaction or dilation. This study shows that
Strain (10-4)

80 min aggregate breakage is the main factor contributing to rockfill creep

-4 (Fig. 10), whereas both consolidation and distortion play a minor
-5
role in rockfill creep. Takei et al. [53] studied the time-dependent
deformation of chalk bar specimen, and their results showed angu-
-6 larity/corner breakage was the dominant failure pattern of chalk
bars (Fig. 6d), which is the same as the modeling results in
-7
Section 4.1. In addition, particle splitting also appeared in a few
-8 chalk bars in Takei et al. [53], but it did not occur in this modeling.
The possible reason is that regular (cubic or hexagonal) packing in
-9 their experiments may induce a pair of relatively larger compres-
0 40 80 120 160 200 sion force in one direction than that in other directions. However,
Time (minute) numerical rockfill aggregates are not perfectly circular and ran-
domly packed in this paper, and thus the contact forces acting on
Fig. 8. Effects of wetting time on rockfill creep evolution.
the single aggregate are in a complicated way. Similarly, both
Takei et al. [53] and this study did not find significant aggregate
rotation, which also plays a negligible role in rockfill creep.
0
Dry
-1
5.2. Proper wetting models in particle mechanics method
-2 Wetting-drying cycles
Experiment
Ciantia et al. [5] classified the bonds into depositional bond,
-3 PFC
which is consisting of mineral powders and can be totally damaged
Strain (10-4)

-4 as water penetrates the rock pores, and diagenetic bond which is

st
1 wetting formed during diagenesis and may be gradually weathered by
-5 water for a long period of time, depending on both the rate of
st

-6
1 drying fluid–mineral reactions and the ionic composition of the bulk fluid.
These mechanisms have been included in wetting models W1, W2
nd
-7 2 wetting and W4 through speeding up the weathering rate or reducing bond
strengths. In addition, wetting models W3 and W4 considers con-
-8 nd
2 drying tact friction coefficient reduction in a fluid environment. The
-9 results show that the additional strain induced by bond strength
0 40 80 120 160 200 240 280 reduction or weathering rate increase is less than that caused by
Time (minute) contact friction coefficient reduction (Fig. 7). In Fig. 9, the strain
evolution during two wetting–drying cycles modeled by W4 is in
Fig. 9. Strain evolution in a ‘stepped’ way induced by wetting–drying cycles. agreement with the laboratory experiments in Cao [3]. The

(a) Initial state (b) Creep under dry condition (c) Creep under wet condition
Fig. 10. Schematic diagram of mechanisms of rockfill aggregate creep under dry and wet conditions. Abrasion (angularity/corner breakage is represented by the gray area) is
the main mechanism for rockfill creep under dry condition. Sliding/rotation due to lubrication and further abrasion due to weakening are the main mechanisms for water
induced strain.
 Z. Zhao, E.-x. Song / Computers and Geotechnics 68 (2015) 137–146
similarity of strain evolution in terms of ‘stepped’ shape indicates
that a reduction in contact friction coefficient and bond strength
should be the main weathering mechanism due to wetting.
Therefore, it is concluded that sliding/rotation due to the reduced
contact friction coefficient and further abrasion due to weakening
are the main mechanisms for water induced strain (Fig. 10). For
practical analysis the wetting model W4 can be more appropriate
to predict the settlement caused by wetting, but careful calibration
is needed to obtain the parameters. Note that these wetting models
are proposed mainly based on Baud et al. [2], which addressed that
the specific surface energy and the frictional coefficient are
reduced in the presence of water. In reality, the reduction extent
of bond strengths and friction coefficient should be carefully cali-
brated with experimental results. In the present modeling of wet-
ting–drying cycles, the values of bond strengths and contact
friction coefficient are assumed to partially recover back to their
original values upon drying. This is just a first approximation,
and further experiments and more accurate models are needed
to understand the changes of bond strengths and contact friction
coefficient during wetting–drying cycles. In nature an acidic solu-
tion may accelerate the dissolution rate of minerals, and also the
reduction rate of bond strengths. This effect is not considered in
the present study as the mechanism is not yet sufficiently clear.
Silvani et al. [51] also studied the effects of wetting time on
rockfill creep behavior using particle mechanics method. Even
though different wetting models were used in Silvani et al. [51]
and this study, respectively, a similar phenomenon found in these
two studies is that an instantaneous strain increment was induced
when the wetting occurs during the primary creep. Silvani et al.
[51] found that the later the water saturation occurs, the lower
the additional creep will be, but this study shows that a similar
final creep strain was obtained for different wetting time. This dif-
ference can be caused by different rockfill aggregate arrangement.
In the future, experiments are encouraged to clarify the different
wetting effects during primary and secondary creep phases.
5.3. Effects of fouling materials
In addition to normal loading and wetting, other factors may
also significantly affect rockfill creep. Effects of fouling materials
are briefly discussed below. Clay fouling can affect rockfill mechan-
ical behavior, depending on its amount. Using large-scale triaxial
tests, Indraratna et al. [16] showed that a relatively small amount
of clay fouling may slightly increase the initial rate of compression,
and this can be attributed to the clay-coated aggregates providing
a lubrication effect at smaller axial strains. With the increase of
fouling, the ballast particles experienced less breakage, attributed
to the cushioning effect of clay, which would reduce the high inter-
nal contact stresses and the associated attrition of rock particles.
Huang et al. [13] and Huang and Tutumluer [14] used experimental
methods and discrete element method to characterize the shear
strength of fouled rail road ballast, and showed that fouling mate-
rials such as coal dust, plastic clayey soil and mineral filler can gen-
erally decrease the shear strength of ballast samples. A similar
trend was also found in Indraratna et al. [17]. However, these stud-
ies did not pay particular attention to the effects of fouling materi-
als on rockfill creep under dry or wet conditions, which will be
studied in the future.
5.4. Implication and limitation
The results show that aggregate crushing in terms of abrasion
is the main cause of rockfill creep under dry state, and this sug-
gests that avoiding aggregates with sharp angularities or corners
in practice can reduce the creep strain to some extent. This is
because the relatively small contact area between sharp corners
145
may easily induce stress concentration and aggregate breakage.
The wetting occurring at the primary phase of creep can result
in rapid settlement and transition from primary creep to sec-
ondary creep, and this suggests that the time gap between the
complement of construction and raining season should be con-
sidered in design.
Note that two dimensional disk models used in this study refer
to a section of rockfill materials, so the width of the model can be
understood as the same magnitude as the rockfill aggregate radius.
This assumption may or may not represent a strict theoretical
plane strain model, but it is helpful to understand the micro-mech-
anisms that cause rockfill creep. The methodology derived in this
study can be easily extended to three dimensional particle
mechanics modeling of rockfill creep. It was assumed that both
wetting and drying processes occurred instantaneously, but they
may need some time for water penetration in practice. This
approximation also needs calibration in the future.
6. Conclusion
Particle mechanics method, incorporating bond-aging models
and wetting models, is used to understand the rockfill creep mech-
anisms under dry or wet conditions. The results show that particle
mechanics method, not only is a useful tool to capture the main
features of rockfill creep, but also can provide insights to the distri-
bution of contact force chains and micro-cracking process, which
cannot be easily monitored in experiments. The main concluding
remarks are summarized below:
(1) Under dry state, micro-cracking due to the reduction of bond
strengths is the main mechanism to generate rockfill creep.
In a rockfill specimen where aggregates are irregularly
packed, breakage mainly occurs at the angularities or cor-
ners of aggregates. Normal stress determines the rates and
values of rockfill creep strain.
(2) A wetting-induced reduction of contact friction coefficient
(lubrication) is the primary cause of a sudden settlement
immediately after wetting, and a wetting-induced reduction
of bond strengths (weakening) plays a secondary role in
rockfill creep, which may speed up the creep strain
evolution.
(3) The earlier the wetting occurs during rockfill creep, the more
rapidly the rockfill specimen becomes stable. The wetting–
drying cycles can induce strain evolution in a ‘stepped’
way, but the magnitude of step decreases with the increas-
ing number of wetting–drying cycles.
Acknowledgements
The work is finically supported by the National Basic Research
Program of China (973 Program) (contract number
2014CB047003). The two anonymous reviewers are acknowledged
for their valuable comments.
References
[1] Alonso E, Oldecop L. A model for rockfill compressibility. Géotechnique
2001;51:127–39.
[2] Baud P, Zhu W, Wong T-F. Failure mode and weakening effect of water on
sandstone. J Geophys Res 2000;105:16371–89.
[3] Cao G. Study on post-construction settlement of high fill foundation in
mountainous airport. Doctoral thesis at Tsinghua University; 2011.
[4] Cheng YP, Nakata Y, Bolton MD. Discrete element simulation of crushable soil.
Géotechnique 2003;53:633–41.
[5] Ciantia MO, Castellanza R, di Prisco C. Experimental study on the water-
induced weakening of calcarenites. Rock Mech Rock Eng 2014;48:441–61.
146 Z. Zhao, E.-x. Song / Computers and Geotechnics 68 (2015) 137–146
[6] Deluzarche R, Cambou B. Discrete numerical modelling of rockfill dams. Int J
Numer Anal Methods Geomech 2006;30:1075–96.
[7] Deng H, Li J, Sun X, Zhu M, Yuan X, Luo Q. Experimental research on fracture
mechanical effect of sandstone under water corrosion. Chin J Rock Mech Eng
2012;31:1342–8 [in Chinese].
[8] Deng H, Li J, Zhu M, Wang K, Wang L, Deng C. Experimental research on
strength deterioration rules of sandstone under ‘‘saturation-air dry’’
circulation function. Rock Soil Mech 2012;33:3306–12 (In Chinese).
[9] Donaghe RT, Cohen MW. Strength and deformation properties of rockfill.
Technical report No. S-78-1. Vicksburg, USA: U.S. Army Waterways
Experimental Station; 1978. p. 98.
[10] Frassoni A, Hegg U, Rossi PP. Large scale laboratory tests for the mechanical
characterization of granular materials for embankment dams. In: Proceeding
of the 14th international congress on large dams, vol. 1; 1982. p. 727–51.
[11] Fumagalli E. Tests on cohesionless materials for rockfill dams. J Soil Mech
Found Div 1969;95:313–30.
[12] Ham TG, Nakata Y, Orense R, Hyodoet M. Influence of water on the
compressive behavior of decomposed granite soil. J Geotech Geoenviron Eng
2010;136:697–705.
[13] Huang H, Tutumluer E, Dombrow W. Laboratory characterization of fouled
railroad ballast behavior. Transport Res Rec: J Transport Res Board
2009;2117:93–101.
[14] Huang H, Tutumluer E. Discrete element modeling for fouled railroad ballast.
Constr Build Mater 2011;25:3306–12.
[15] Indraratna B, Lackenby J, Christie D. Effect of confining pressure on the
degradation of ballast under cyclic loading. Géotechnique 2005;55:325–8.
[16] Indraratna B, Tennakoon N, Nimbalkar S, Rujikiatkamjorn C. Behaviour of clay-
fouled ballast under drained triaxial testing. Géotechnique 2013;63:410–9.
[17] Indraratna B, Ngo N, Rujikiatkamjorn C, Vinod JS. Behaviour of fresh and fouled
railway ballast subjected to direct shear testing – a discrete element
simulation. Int J Geomech 2014;14:34–44.
[18] Ionescu D. Evaluation of the engineering behavior of railway ballast. PhD
thesis at University of Wollongong; 2004.
[19] Itasca Consulting Group Inc. PFC2D user’s guide. Minneapolis, MN; 2008.
[20] Jensen RP, Plesha ME, Edil TB, Bosscher PJ, Ben Kahla N. DEM simulation of
particle damage in granular media-structure interfaces. Int J Geomech
2001;1:21–39.
[21] Justo JL, Durand P. Settlement-time behaviour of granular embankments. J
Numer Anal Methods Geomech 2000;24:281–303.
[22] Kjaernsli B, Sande A. Compressibility of some coarse grained materials. In:
Proceedings of European conference on soil mechanics and foundation
engineering, Wiesbaden; 1963. p. 245–51.
[23] Kwok CY, Bolton MD. DEM simulations of thermally activated creep in soils.
Géotechnique 2010;60:425–33.
[24] Kwok CY, Bolton MD. DEM simulations of soil creep due to particle crushing.
Géotechnique 2013;63:1365–76.
[25] Lajtai EZ, Schmidtke RH, Bielus LP. The effect of water on the time dependent
deformation and fracture of a granite. Int J Rock Mech Min Sci Geomech Abstr
1987;24:247–55.
[26] Lim WL. Mechanics of railway ballast behaviour. PhD Thesis at University of
Nottingham, Nottingham, U.K.; 2004.
[27] Lim WL, McDowell GR. Discrete element modeling of railway ballast. Granul
Matter 2005;7:19–29.
[28] Liu H, Lindqvist P-A, Åkesson A, Kou S, Lindqvist J-E. Characterisation of rock
aggregate breakage properties using realistic texture-based modelling. Int J
Numer Anal Methods Geomech 2012;36:1280–302.
[29] Lobo-Guerrero S, Vallejo EL, Vesga LE. Visualization of crushing evolution in
granular materials under compression using DEM. Int J Geomech
2006;6:195–200.
[30] Lobo-Guerrero S, Vallejo LE. Crushing a weak granular material: experimental
numerical analyses. Géotechnique 2005;55:245–9.
[31] Lu M, McDowell GR. Discrete element modelling of ballast abrasion.
Géotechnique 2006;56:651–5.
[32] Lu M, McDowell GR. The importance of modelling ballast particle shape in
DEM. Granul Matter 2007;9:69–80.
[33] Lu M, McDowell GR. Discrete element modelling of railway ballast under
triaxial conditions. Geomech Geoeng: Int J 2008;3:257–70.
[34] Lu M, McDowell GR. Discrete element modelling of railway ballast under
monotonic and cyclic triaxial loading. Géotechnique 2010;60:459–67.
[35] Luong MP. Stress–strain aspects of cohesion-less soils under cyclic and
transient loading. In: Balkema AA, editor. International symposium on soil
under cyclic and transient loading. Rotterdam; 1982. p. 315–24.
[36] Ma G, Zhou W, Chang XL. Modeling the particle breakage of rockfill materials
with the cohesive crack model. Comput Geotech 2014;41:132–43.
[37] Mahabadi OK, Grasselli G, Munjiza A. Y-GUI: a graphical user interface and
pre-processor for the combined finite-discrete element code, Y2D,
incorporating material heterogeneity. Comput Geosci 2010;36:241–52.
[38] Marsal RJ. Mechanical properties of rockfill, embankment – dam engineering,
Casagrande volume. New York, USA: John Wiley and Sons; 1973. p. 109–200.
[39] McDowell GR, Harireche O. Discrete element modelling of soil particle
fracture. Géotechnique 2002;52:131–5.
[40] McDowell GR, Khan JJ. Creep of granular materials. Granul Matter
2003;5:115–20.
[41] Mesri G, Castro A. Ca/Cc concept and K0 during secondary compression. J
Geotech Eng 1987;113:230–47.
[42] Mirghasemi AA, Rothenburg L, Matyas EL. Influence of particle shape on
engineering assemblies of two-dimensional polygon-shaped particles.
Géotechnique 2002;52:209–17.
[43] Mukai M, Takino Y, Yamazaki H. Long term assessment of super-high naming
with decomposed granite soil. In: Proceedings of 54th annual conference of
JSCE, 3-(A). Hiroshima, Japan: Japan Society of Civil Engineers; 1999 [in
Japanese]. p. 780–1
[44] Oldecop LA, Alonso EE. Suction effects on rockfill compressibility.
Géotechnique 2003;53:289–92.
[45] Oldecop LA, Alonso EE. Theoretical investigation of the time-dependent
behaviour of rockfill. Géotechnique 2007;57:289–301.
[46] Pigeon Y. The compressibility of rockfill. PhD Thesis at University of London,
London, U.K.; 1969.
[47] Potyondy DO, Cundall PA. A bonded-particle model for rock. Int J Rock Mech
Min Sci 2004;41:1329–64.
[48] Potyondy DO. Simulating stress corrosion with a bonded-particle model for
rock. Int J Rock Mech Min Sci 2007;44:677–91.
[49] Sangha CM, Dhir RK. Influence of time on strength, deformation and fracture
properties of lower devonian sandstone. Int J Rock Mech Min Sci 1992;9:343–54.
[50] Seyedi Hosseininia E, Mirghasemi AA. Numerical simulation of breakage of
two-dimensional polygon-shaped particles using discrete element method.
Powder Technol 2006;166:100–12.
[51] Silvani C, Désoyer T, Bonelli S. Discrete modeling of time-dependent rockfill
behavior. Int J Numer Anal Methods Geomech 2009;33:665–85.
[52] Sowers GF, Williams RC, Wallace TS. Compressibility of broken rock and the
settlement of rockfills. In: Proceeding of 6th international conference on soil
mechanics and foundation engineering, Montreal, vol. 2; 1965. p. 561–5.
[53] Takei M, Kusakabe O, Hayashi T. Time-dependent behavior of crushable
materials in one-dimensional compression tests. Soils Found 2001;41:97–121.
[54] Tran T-H, Vénier R, Cambou B. Discrete modelling of rock-ageing in rockfill
dams. Comput Geotech 2009;36:264–75.
[55] Tsoungui O, Vallet D, Charmet J-C. Numerical model of crushing of grains
inside two-dimensional granular materials. Powder Technol 1999;105:190–8.
[56] Zhao D, Nezami EG, Hashash YMA, Ghaboussi J. Three-dimensional discrete
element simulation for granular materials. Eng Comput 2006;23:749–70.
[57] Zhao Z. Gouge particle evolution in a rock fracture undergoing shear: a
microscopic DEM study. Rock Mech Rock Eng 2013;46:1461–79.