Computers and Geotechnics 59 (2014) 54–66

Contents lists available at ScienceDirect

Computers and Geotechnics

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

A unified plasticity model for large post-liquefaction shear

deformation of sand
Rui Wang a, Jian-Min Zhang b,⇑, Gang Wang c
a
School of Civil Engineering, State Key Laboratory of Hydroscience and Engineering, Department of Hydraulic Engineering, Tsinghua University, Beijing, China
b
School of Civil Engineering, State Key Laboratory of Hydroscience and Engineering, Tsinghua University, Beijing 100084, China
c
Ertan Hydropower Development Company Limited, Chengdu 610051, China

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

Article history: Based on previous experimental findings and theoretical developments, this paper presents the formula-
Received 21 November 2013 tion and numerical algorithms of a novel constitutive model for sand with special considerations for
Received in revised form 21 February 2014 cyclic behaviour and accumulation of large post-liquefaction shear deformation. Appropriate formulation
Accepted 25 February 2014
for three volumetric strain components enables the model to accurately predict loading and load reversal
behaviour of sand, fully capturing the features of cyclic mobility. Compliance with the volumetric com-
patibility condition, along with reversible and irreversible dilatancy, allows for physically based simula-
Keywords:
tion of the generation and accumulation of shear strain at zero effective stress after initial liquefaction. A
Constitutive model
Sand
state parameter was incorporated for compatibility with critical state soil mechanics, enabling the unified
Liquefaction simulation of sand at various densities and confining pressures with a same set of parameters. The deter-
Critical state mination methods for the 14 model parameters are outlined in the paper. The model was implemented
Simulation into the open source finite-element framework OpenSees using a cutting-plane stress integration scheme
with substepping. The potentials of the model and its numerical implementation were explored via sim-
ulations of classical drained and undrained triaxial experiments, undrained cyclic torsional experiments,
and a dynamic centrifuge experiment on a single pile in liquefiable soil. The results showed the model’s
great capabilities in simulating small to large deformation in the pre- to post-liquefaction regime of sand.
Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction
Large post-liquefaction deformation is a major cause for seismic
liquefaction induced hazards, and has been a subject of extensive
research since its observations in several well documented earth-
quakes (e.g. [1–4]). In this paper, we focus on the cyclic mobility
of sand and the accumulation of large but limited shear strains
after sand reaches ‘‘initial liquefaction’’ [5] observed in numerous
laboratory experiments (e.g. [6–8]), which is referred to as large
post-liquefaction shear deformation (Fig. 1).
Numerous constitutive models have been developed aiming to
simulate the stress–strain behaviour of saturated sands during
cyclic loading, including generalized plasticity models
(e.g. [9,10]), hypoplasticity models (e.g. [11,12]), multi-surface
models (e.g. [13–18]) and bounding surface plasticity models
(e.g. [19–23]). Pastor et al. [9] suggested predicting cyclic mobility
through applying a ‘‘discrete memory factor’’ to the plastic modulus
in their generalized plasticity model. Wu and Bauer [11] developed
⇑ Corresponding author.
E-mail address: zhangjm@mail.thu.edu.cn (J.-M. Zhang).
a simple hypoplasticity model that accounts for basic cyclic behav-
iour of sand, though the model’s independence from stress history
limits its application under complex stress paths. Wang et al. [19]
proposed a bounding surface hypoplasticity model for sand which
was able to simulate cyclic stress path through reducing plastic
shear modulus with the accumulation of plastic shear strain. Papa-
dimitriou et al. [20] and Dafalias and Manzari [22] developed
bounding surface plasticity models that simulated sand behaviour
under cyclic loading by applying evolving fabric tensors on the plas-
tic modulus and dilatancy rate respectively, enhancing the contrac-
tion upon unloading and thus allowing the stress path to approach
liquefaction during undrained loading. These models all made sig-
nificant contributions to the description of cyclic mobility, but none
are able to reflect the accumulation of shear strain at liquefaction
during each load cycle after initial liquefaction, with stress–strain
relationship following almost the same path each cycle, which con-
tradicts experimental findings.
To reflect the shear strain generated at liquefaction, Boulanger
and Ziotopoulou [23] further modified the model by Dafalias and
Manzari to achieve the accumulation of shear strain after initial
liquefaction by adding fabric history and cumulative fabric terms.

http://dx.doi.org/10.1016/j.compgeo.2014.02.008
0266-352X/Ó 2014 Elsevier Ltd. All rights reserved.
 R. Wang et al. / Computers and Geotechnics 59 (2014) 54–66 55

40 Zhang [32] and Zhang [33] proposed that the volumetric strain of
(a)
sand consisted of two basic components: mean effective stress
20 γ0 change induced evc, and dilatancy induced evd. The dilatancy
induced evd was further decomposed into a reversible and an irre-
τ (kPa)

versible component, namely evd,re and evd,ir, as shown in Fig. 2(c)

0
and (d). Irreversible dilatancy is the shear induced contraction of
sand, which is generally caused by packing and crushing of parti-
-20 Pre-liquefaction cles. Reversible dilatancy refers to shear induced expansion and
Post-liquefaction
the reversal of such expansion normally caused by particle sliding
-40 and reorientation. The decomposition would then be expressed as:
-0.08 -0.04 0 0.04 0.08
γ ev ¼ ev c þ ev d ¼ ev c þ ev d;ir þ ev d;re ð1Þ
(b) 40
Zhang and Wang [31,34] pointed out that since evc is solely
dependent on the change in effective confining pressure, there
20 exists a threshold evc,0 at which zero effective stress is reached.
Once this threshold is reached, the evc would then be determined
τ (kPa)

0 the volumetric compatibility Eq. (1). For sand to leave liquefaction

state when evc < evc,0, sufficient dilatancy would be needed, and
-20 hence sufficient shear strain would be required according to dilat-
ancy relations. Based on the proposed mechanism, Zhang and
Wang [31] formulated a constitutive model within the framework
-40
0 20 40 60 80 100 of bounding surface plasticity suited for two dimensional stress
p' (kPa) space. The model proved capable in simulating the cyclic mobility
and large post-liquefaction shear deformation of sand. However,
Fig. 1. Stress–strain relation and stress path of undrained cyclic torsional test for the model does underestimate contraction during initial loading
Toyoura sand at Dr = 70%. (Data from Zhang et al., 1997).
and may overestimate it during load reversal. And as critical state
behaviour was not considered, it does not comply with critical
state soil mechanic principles and is not able to provide unified
In the multi-surface models by Parra-Colmenares [15], Elgamal
description of sand under different densities and confining pres-
et al. [16,17] and Yang et al. [18], in order to model the accumula-
sures with a same set of parameters.
tion of shear strain near liquefaction state, an additional shear
This paper builds on the work of Zhang and Wang [31] to pres-
strain accumulation was introduced at a ‘‘neutral phase’’ when
ent the formulation of a unique model that (1) achieves the simu-
the effective stress path crossed the phase transformation line at
lation of post-liquefaction shear deformation based on its physics,
relatively low effective confining pressure. These two sets of more
allowing the unified description of pre- and post-liquefaction
recent models have taken a big step forward in the simulation of
behaviour of sand; (2) directly links the cyclic mobility of sand
liquefaction behaviour of sand, however both models lack the
with reversible and irreversible dilatancy, enabling the unified
physical basis for the formulation of post-liquefaction shear
description of monotonic and cyclic loading; (3) introduces critical
strains, causing the shear strain accumulation to occur at some-
state soil mechanics concepts to achieve unified modelling of sand
what high shear stress instead of at liquefaction.
under different states. Modelling of large post-liquefaction shear
Significant progress on clarifying the role of critical state [24,25]
deformation was achieved based on the physics proposed by Zhang
for sand has been made over the past few decades through rigorous
and Wang. The proposed model is able to appropriately describe
work by various researchers, including Been and Jefferies [26], Ishi-
hara [27], Wood et al. [28], Li and Dafalias [29], etc. Thus making the
unified constitutive description for sand of different densities and
0.4
confining pressures possible, which has been an approach adopted (a)
by numerous recent constitutive models (e.g. [20,22,23,28,29]). 0.0
These previous research have provided valuable insights into -0.4
the mechanical behaviour of sands. This paper looks to further 3
(b)
the understanding of sand behaviour by presenting a constitutive 0
model with special attention given to providing physically based -3
modelling of large but limited shear deformation after initial lique- 0.0 (c) Drained torsional test
faction at or near zero effective stress, which has been observed in SaturatedToyoura sand
0.5
undrained cyclic triaxial and torsional shear experiments con- Dr=70%
ducted by various researchers (e.g. [6–8]). 1.0
Through observations from a number of cyclic undrained tor- 1.5
sional shear tests conducted on hollow cylinder specimens of
2.0
Toyoura sand, Zhang [30] noted that after initial liquefaction, while
the stress path of each cycle was very much similar (Fig. 1(b)), 2.5
large but finite shear strain was generated near zero effective -0.8
(d)
stress state during each cycle (Fig. 1(a)), which was referred to as -0.4
large post-liquefaction deformation. The shear strain generated at 0.0
liquefaction state was observed to increase with the number of 0 5 10 15 20
loading cycles, and was defined as a ‘‘fluid-like shear strain’’ c0
Number of cycles, N
[31].
To explain the physics of post-liquefaction deformation, based Fig. 2. The decomposition of reversible and irreversible dilatancy components in
on observations on drained cyclic experiments, Shamoto and drained cyclic torsional test. (Data from Shamoto et al., 1997).
56 R. Wang et al. / Computers and Geotechnics 59 (2014) 54–66

 12
some important features of sand, including dilatancy during load- ð2:973  ein Þ2 p
ing and unloading, and softening of dense sand. The model is first G ¼ Go pa ð4Þ
1 þ ein pa
described in triaxial stress space, and then generalized into multi-
axial stress space with three dimensional mapping rules for the  12
calculation of plasticity and dilatancy. The determination of the 1 þ ein p
K¼ pa ð5Þ
model parameters is described. The cutting-plane stress integra- j pa
tion scheme and the Pegasus procedure for the three dimensional
implementation of the model into the finite-element framework where ein is the initial void ratio, pa is the atmospheric pressure for
OpenSees [35] are presented in detail. The OpenSees platform normalization, Go and j are material constants. The elastic moduli
was chosen for its great capabilities in geotechnical earthquake were defined using the initial void ratio so that elastic volumetric
engineering simulation and most importantly for the model to be strain is strictly dependent only on effective stress, which decouples
openly available to the technical community. Finally, the perfor- the elastic moduli with shear induced volumetric strains. Thus mak-
mance of the model and its three dimensional implementation is ing it consistent with our decomposition of volumetric strains.
evaluated by simulations of classical drained and undrained triax-
ial experiments on Toyoura sand by Verdugo and Ishihara [36] and
2.3. State parameter
undrained cyclic torsional experiments on Toyoura sand. Simula-
tion of a centrifuge shaking table test on a single pile in liquefiable
For the incorporation of critical state behaviour, and for the uni-
ground conducted by the authors was also carried out.
fied description of sand at various densities with a same set of
In this paper, tensorial quantities are denoted by bold sans-serif
parameters, the state parameter W proposed by Been and Jefferies
letters to distinguish them from plain serif letter scalars.
[26] is introduced to consider the dependency of sand behaviour
on the current state.
2. Model formulation in triaxial stress space
W ¼ e  ec ð6Þ
A description of the proposed constitutive model is first
provided in triaxial stress space. The model operates within the with e being the current void ratio and ec the critical void ratio. At
framework of bounding surface plasticity proposed by Dafalias critical state, Roscoe et al. [24] and Schofield and Wroth [25] pro-
and Popov [37] and adopts features of the hypoplasticity model posed that e = ec and qc/pc = M, where M is defined as the critical
developed by Wang et al. [19]. The model uses the two dilatancy stress ratio as shown in Fig. 3. The relationship between the critical
induced volumetric strain components for the successful descrip- void ratio ec and mean effective stress is determined using Li and
tion of the dilatancy behaviours of sand in both monotonic and Wang’s [39] power formulation which have proved effective for var-
cyclic loading, and accounts for the generation of post-liquefaction ious types of sands [39,40]:
shear deformation at zero effective confining stress based on the
physics of post-liquefaction deformation proposed by Zhang and ec ¼ e0  kc ðpc =pat Þn ð7Þ
Wang [31]. Critical state soil mechanics principles were incorpo-
rated into the model to allow unified description of sand by intro- where e0 is the void ratio at pc = 0 and kc and n are constants. By cor-
ducing the state parameter [26]. rect consideration for the variation of plasticity and dilatancy with
In the current model framework, mean effective pressure the state parameter, the model can be made applicable to different
change induced volumetric strain is assumed to be elastic and pressures and densities using a single set of parameters.
shear induced volumetric strain is assumed to be fully plastic,
matching the decomposition of volumetric strain stated previously
in Eq. (1) with traditional elastic and plastic decompositions. 2.4. Plastic loading and load reversal

2.1. Basic equations By adopting Wang’s [19] hypoplasticity approach, plastic load-
ing and load reversal is determined by the sign of:
The incremental stress–strain relations follow the typical equa-
tions for elastic–plasticity: l ¼ g_ ðg  ain Þ ð8Þ

q_ p_ where ain is the stress ratio at the previous load reversal. Plastic
e_ eq ¼ ; e_ ev ¼ ð2Þ
3G K loading is induced when l > 0, and load reversal occurs at l < 0.

g_
e_ pq ¼ ; e_ pv ¼ Dje_ pq j ð3Þ
Maximum stress
H
q ratio line Mm
where the elastic and plastic strains are denoted by superscripts e
and p respectively. G and K are the elastic shear and bulk moduli,
H is the plastic shear modulus and D dilatancy rate. M
The model neglects the plastic volumetric strains during con- Reversible dilatancy
stant stress ratio loading for simplicity, though this proposition η line Md
would cause underestimation of the reconsolidation settlement of r
r
soil. The elastic volumetric strain eev corresponds to the mean effec- 0
tive stress change induced volumetric strain evc, and epv corresponds α in p
to the dilatancy induced volumetric strain evd,ir + evd,re in Eq. (1). Md
Critical state line M
2.2. Elastic moduli
ηm

The elastic shear and bulk moduli are defined as suggested by Fig. 3. Schematic illustration of critical state, maximum stress ratio and reversible
Richart et al. [38]: dilatancy lines with mapping rules in triaxial stress space.
 R. Wang et al. / Computers and Geotechnics 59 (2014) 54–66
2.5. Plastic modulus
The plastic modulus and relevant mapping rules in the model
are modified from the work of Wang et al. [19]. The peak mobilized
stress ratio, which has been shown to be dependent on the state
parameter [26], is here defined using an exponential variation with
W as proposed by Li and Dafalias [29] to be Mp = M exp(nbW). A
maximum stress ratio (Mm) surface is proposed to define the map-
ping rule, as shown in Fig. 3 in triaxial stress space. By using the
peak mobilized and maximum stress ratios, the plastic modulus
can be determined in a way that is related to the bounding surface
plasticity concept. Thus, the plastic modulus is expressed as:
    (Fig. 2(c)). Irreversible dilatancy rate Dir is defined to satisfy these
h M expðnp WÞ q
 features as:
H ¼ G expðnp WÞ 1
p Mm q
where h is a model parameter; np is a model constant; q is the dis-
tance from the current stress ratio g to ain; and q is the distance be-
tween the projection of current stress on the maximum stress ratio
surface and ain (Fig. 2).
The maximum stress ratio Mm surface is defined to expand
according to the maximum stress ratio that has occurred
during loading, until it reaches the peak mobilized stress ratio
M exp(nbW). Once the current stress ratio reaches outside
M exp(nbW), the maximum stress ratio follows the current stress
ratio until it falls on the peak mobilized stress ratio again. Through
this formulation, the plastic modulus H = 0 is satisfied at the critical
state and the softening response where H < 0 is also made possible
when the stress ratio exceeds the peak mobilized stress ratio Mp.
2.6. Dilatancy
The determination of dilatancy is a unique part of the proposed
model. According to the propositions made by Shamoto and Zhang
[32] and Zhang [33], the dilatancy of sand is decomposed into a
reversible and an irreversible component, through which the dilat-
ancy during load reversal and cyclic loading can be properly
reflected. In this model, the dilatancy rate D is determined by
combining the reversible part Dre and irreversible part Dir:
e_ pv e_ v d;re e_ v d;ir
D¼ ¼ Dre þ Dir ¼ p þ p
je_ pq j je_ q j je_ q j
Through experimental observations Zhang and Wang [31]
pointed out that reversible dilatancy remained on the expansion
side, generating and releasing during loading and unloading cycles.
The generation and release rate of reversible dilatancy are here
defined using separate equations. The generation rate of reversible
dilatancy, for which Dre,gen is negative, is defined in a form similar
to Rowe’s dilatancy theory [41]:
Dre;gen ¼ dre;1 ðM d  gÞ
where dre,1 is a reversible dilatancy parameter, and Md = M exp(nd-
W) is the stress ratio at which reversible dilatancy changes from
contraction to expansion, and also follows an exponential variation
with W [29], shown as the reversible dilatancy line in Fig. 3. Revers-
ible dilatancy remains non-positive and is released after load rever-
sal, the release rate is defined as:
2
Dre;rel ¼ ðdre;2 vÞ =p
dre,2 is another dilatancy parameter used to calculate the release of
 
e
reversible dilatancy. v ¼ min dir evprd;re ; 1 is a function controlling
v d;ir
the reversible dilatancy release process, where dir is an irreversible
dilatancy constant and eprv d;ir is the evd,ir at previous load reversal.
This function v guarantees Dre,rel to be zero when evd,re is completely
released, and restricts the release rate from becoming overly large,
57
thus avoiding the overestimation of contraction upon load reversal.
Note that prior to the first load reversal, v is set as 0 and the release
rate Dre,rel is constantly 0.
Reversible dilatancy can thus be expressed together as:
(
e_ v d;re Dre;gen ; jgj P M d;c=e & jgj > 0
Dre ¼ ¼ ð13Þ
je_ pq j Dre;rel ; jgj < M d;c=e or jgj < 0
It has been observed that the irreversible dilatancy induced vol-
umetric strain evd,ir remains contractive, and followed the pattern
of accumulating asymptotically during loading with a decreasing
rate during each monotonic shearing since the last stress reversal
ð9Þ
e_ v d;ir
Dir ¼ ¼ dir expðnd W  aev d;ir Þ
je_ pq j
0 !2 1
cd;r h1  expðnd WÞi
 @hMd  giexpðvÞ þ Að14Þ
cd;r h1  expðnd WÞi þ cmono
Here a is a parameter controlling the decrease rate of irrevers-
ible dilatancy, cmono is the shear strain since the last stress reversal
and cd,r is a reference shear strain. h i are the MacCauley brackets
that yield hxi = x if x > 0 and hxi ¼ 0 if x 6 0. The exp(ndW  aevd,ir)
part of the equation reflects asymptotic accumulation of irrevers-
 2
cd;r h1expðnd WÞi
ible dilatancy, and the part c h1expðn d WÞiþc reflects the
d;r mono
decreasing dilatancy rate during each monotonic loading process.
By introducing hMd  gi into the formulation, the initial contrac-
tion during loading can be appropriately reflected, and exp(v) en-
hances the contraction upon load reversal. The incorporation of
state parameter allows the formulation to take density and effec-
tive pressure into consideration and comply with critical state soil
mechanic principles.
A most significant result of incorporating the state parameter in
the definition of plastic modulus, reversible and irreversible
dilatancy is that the model becomes fully compatible with critical
state soil mechanics requirements and capable of simulating the
behaviour of sand at various densities with the same set of mate-
rial constants. The unique formulation of plastic modulus and
ð10Þ
dilatancy rates allows for the appropriate simulation of both
monotonic and cyclic loading. These features will be shown in
the simulations later in this paper.
2.7. Post-liquefaction shear deformation
As plasticity caused by constant g loading is neglected in this
model, the volumetric strain caused by mean effective stress
change evc proposed by Zhang and Wang [31] in Eq. (1) matches
ð11Þ
eev in Eq. (2). By substituting Eq. (5) into Eq. (2) and integrating
from the current effective stress p to zero effective stress, the
threshold evc at which zero effective stress is reached is expressed
as a function of p:
 12
2j p
ev c;0 ¼ f ðpÞ ¼  ð15Þ
1 þ ein pa
ð12Þ Based on Zhang and Wang’s [31] theory, once the current evc (or
eev ) decreases beyond the threshold value evc,0, sand liquefies and p
_
remains constant at 0, e_ ev ¼ Kp in Eq. (2) becomes invalid and evc is
then determined by the volumetric compatibility equation Eq. (1)
and is able to exceed evc,0.
For sand to exit the state of liquefaction when evc < evc,0, suffi-
cient dilation must occur for evc to become greater than evc,0 again.
According to the dilation equation in Eq. (3), sufficient dilation
would depend on the shear strain epq generated at liquefaction
58 R. Wang et al. / Computers and Geotechnics 59 (2014) 54–66

state, which is the cause of large post-liquefaction shear 1

gðhÞ ¼
deformation: 1 þ M p ð1 þ sin 3h  cos2 3hÞ=6 þ ðM p  M p;o Þ cos2 3h=Mp;o
Z
ð24Þ
epv ¼ ðDir þ Dre Þdjepq j ð16Þ
6 sin /f
Mp ¼ ð25Þ
3  sin /f
3. Multiaxial formulation pffiffiffi
2 3 tan /f
With the model presented fully in triaxial stress space, its mul- Mp;o ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð26Þ
3 þ 4 tan2 /f
tiaxial generalization is then possible. The basic equations for the
multiaxial generalization are:
Mp = M exp(nbW) is the peak mobilized stress ratio at triaxial
p_ s_ compression and /f is the corresponding friction angle, Mp,o is the
e_ ev ¼ ; e_ e ¼ ð17Þ peak mobilized stress ratio under torsional shear after isotropic
K 2G
consolidation.
e_ pv ¼ hLiD; e_ p ¼ hLim ð18Þ Similar to the triaxial space formulation, plastic loading is
determined in three dimensional space by the load index L:
p = tr(r)/3 is the mean effective stress, with r being the effective
stress tensor; s = r  pI is the deviatoric stress, I being the rank L : r_ pr_ : n
L¼ ¼ ð27Þ
two identity tensor; ev = tr(e) is the volumetric strain, e being the H H
strain tensor; e = e  ev/3I is the deviatoric strain tensor. L is the
Here n is a unit deviatoric tensor serving as the loading direc-
plastic loading index and m the deviatoric strain flow direction.
tion in deviatoric stress space in the model, and the loading direc-
The deviatoric stress ratio tensor is here defined as r ¼ ps, and
qffiffiffiffiffiffiffiffiffiffiffi tion L is defined as L ¼ n  13 ðn : rÞI. Plastic loading is induced
q ¼ 32 s : s, g ¼ qp. when L > 0, and load reversal occurs at L < 0.
It is further assumed that the deviatoric strain flow direction m
The total stress–strain relation can be formulated by combining
in Eq. (18) is the same as the loading direction in deviatoric stress
Eqs. (17) and (18) to be:
    space so as:
1 1 1 D pffiffiffiffiffiffiffiffiffi
e_ ¼ pr_ þ rþ I p_ þ m þ I hLi ð19Þ m ¼ n ¼ r= r : r ð28Þ
2G 2G 3K 3
with the elastic moduli G and K still defined by Eqs. (4) and (5). Here r represents the projection of the current stress point on the
The critical, maximum stress ratio and reversible dilatancy maximum stress ratio surface in deviatoric stress space (Fig. 4), the
surfaces shown schematically in Fig. 4 are defined by: mapping rule for this projection is adopted from the work of Wang
et al. [19]. As shown in Fig. 4, the projection of current stress ratio on
fc ðrÞ ¼ g  MgðhÞ ¼ 0 ð20Þ the maximum stress ratio surface r is defined as the intersection be-
tween the extension of the line from the previous load reversal
fm ðrÞ ¼ g  M m gðhÞ ¼ 0 ð21Þ point ain to r and the maximum stress ratio surface:

fd ðrÞ ¼ g  M d gðhÞ ¼ 0 ð22Þ r ¼ ain þ bðr  ain Þ ð29Þ

where h is the lode angle calculated according to: where b can be solved by substituting Eq. (28) into Eq. (20). It needs
  pointing out that although theoretically n should be the unit normal
1 1 s:s:s
h ¼ sin  ð23Þ to the maximum stress ratio surface in deviatoric stress ratio space,
3 6q due to the numerical difficulty in calculating the normal to the
The function g(h) in this model follows Zhang’s [30] proposition surface, a compromise is made for the model to be numerically
which showed excellent agreement with test data, and is defined applicable by setting the deviatoric loading and flow directions to
as: be the same as r (Fig. 4), which is the same approach taken by
Andrianopoulos et al. [42].
When the loading index L is positive, plastic loading occurs.
Once L becomes negative, load reversal takes place and the projec-
tion centre ain is updated to be the current stress ratio.
The plastic modulus H can then be defined based on the
mapping rule and Eq. (9):
   
2 M expðnb WÞ q

H¼ hgðhÞG expðnp WÞ 1 ð30Þ
3 Mm q
where the factor 23 is for the equation to be compatible with that in
triaxial stress space, q
 is the distance between r and ain, and q the
distance between r and ain.
The mapping rule for reversible dilatancy is defined so that the
projection of the current stress ratio on the reversible dilatancy
surface rd is the intersection between r and the reversible dilatancy
surface:

Md M expðnd WÞ
rd ¼ r ¼ r ð31Þ
Mm Mm

Fig. 4. Schematic illustration of critical state, maximum stress ratio and reversible The generation and release of reversible dilatancy can then be
dilatancy surfaces with mapping rules. judged by the angle between rd  r and n:
 R. Wang et al. / Computers and Geotechnics 59 (2014) 54–66 59

(
e_ v d;re Dre;gen ; ðrd  r : n < 0Þ -0.5
Dre ¼ ¼ ð32Þ
c_ p Dre;rel ; ðrd  r : n > 0Þ Reversible dilatancy sign
change point
While the release rate still follows the triaxial formulation Eq.

ε vd,re (%)
-0.3
(12), the generation rate of reversible dilatancy now becomes:

rffiffiffi
2 -0.1
Dre;gen ¼ dre;1 ðrd  rÞ : n ð33Þ
3
qffiffi
2
is introduced for consistency with the formulation in triaxial 0.1
3
stress space.
-0.3 -0.1 0.1 0.3
Irreversible dilatancy rate Dir defined in the triaxial formulation τ /p'
by Eq. (14) is still valid in the multiaxial formulation. The genera-
Fig. 5. Stress ratio and reversible dilatancy component relations after 18 cycles of a
tion of post-liquefaction shear deformation also follows the drained cyclic torsional test for Toyoura sand (data from Shamoto et al., 1997).
description in the triaxial stress space formulation.

4. Determination of model parameters 5. Model implementation

Table 1 lists the 14 parameters used in the model. Some param- The constitutive model has been implemented into the open
eters used in this model have been documented by previous source finite element framework OpenSees [35] using a cutting
researchers, including the elastic modulus constants (G0, j) plane algorithm [43] with substepping for the stress integration
[31,38], plastic modulus parameter (h) [19] and critical state scheme. Solid–fluid coupled elements needed for the undrained
parameters (M, kc , e0, n) [29], readers are advised to follow the cal- and partially drained analysis of sand, which is essential for lique-
ibration methods suggested for these parameters in the respective faction analysis, are already incorporated into OpenSees (e.g. u-p
studies. elements by Yang et al. [44] and SSP u-p elements by McGann
The state parameter constants np and nd can be determined et al. [45]).
through np = ln (M/gp)/Wp and nd = ln (Md/M)/Wd derived from
Eqs. (9) and (11) [29], where gp and Wp are g and W at peak stress 5.1. Numerical treatment for zero effective stress state
ratio in a monotonic drained triaxial test, and Md and Wd are those
at reversible dilatancy sign change points. In the implementation of the model, to avoid numerical difficul-
It is important to note here that the determination method of nd ties at zero effective stress during liquefaction, a pmin was set as the
is different to that suggested by Li and Dafalias [29] due to the way minimum effective confining pressure, thus for p to be:
dilatancy is defined in this paper. Drained cyclic torsional or triax- 
ial tests should be used for the determination of nd here, as Md can
p_ ¼ K e_ ev ; ev c > ev c;0 ð34Þ
only be acquired once irreversible dilatancy is negligible after a
p ¼ pmin ; ev c 6 ev c;0
number of loading cycles. For example, Fig. 5 shows the stress ratio The threshold mean effective stress change induced volumetric
and reversible dilatancy component relations after 18 cycles of a strain then becomes:
drained cyclic torsional test for Toyoura sand. At this stage, irre-
 12  1 !
versible dilatancy becomes negligible, and Md can be determined 2j p pmin 2
from the figure to be 0.222. Using the critical state parameters
ev c;0 ¼  ð35Þ
1þe pa pa
for Toyoura sand given by Li and Wang [39], the state parameter
Wd at this state is -0.232, thus yielding a nd of 7.8 for Toyoura sand. The effectiveness of this approach has been validated by Zhang
The reversible dilatancy parameters dre,1 can be determined and Wang [31].
using the relationship between g and ddecvpd from drained cyclic tests
as suggested by Zhang and Wang [31], and dre,2 should then be cho- 5.2. Stress integration scheme
sen to ensure the release of reversible dilatancy.
For the irreversible dilatancy parameters (dir and a especially, The cutting-plane algorithm was chosen for its simplicity and
cdr can generally be set at 0.05), a trial-and-error process should efficiency, being a semi-explicit integration scheme, relatively
be adopted to simulate the stress strain behaviour of undrained small time increments are needed for stability. To increase stabil-
cyclic torsional/triaxial tests of different initial confining pressure ity, a substepping technique is proposed.
or shear stress amplitude, as was described by Zhang and Wang At the beginning of each step, the strain increment from the glo-
[31]. The parameter dir mainly determines how fast liquefaction bal converged state is used to calculate an elastic prediction of
is reached in undrained cyclic tests, and a controls the decrease stress increment, note once again ev c ¼ eev and p is expressed as a
rate of irreversible dilatancy. function of evc:

Table 1
Model parameters for the simulations of element and centrifuge experiments.

Sand Go j h M dre,1 dre,2 dir a cd,r np nd kc e0 n

Toyouraa 200 0.008 1.8 1.25 0.6 30 1.4 20 0.05 1.1 7.8 0.019 0.934 0.7
Toyourab 200 0.008 1.8 1.35 0.35 30 0.75 20 0.05 1.1 7.8 0.019 0.934 0.7
Fujian 200 0.006 1.7 1.3 0.45 30 0.6 40 0.05 1.1 8.0 0.023 0.837 0.7
a
Toyoura sand used by Verdugo and Ishihara, 1996.
b
Toyoura sand used by Zhang et al., 1997 and Chiaro et al., 2013.
60 R. Wang et al. / Computers and Geotechnics 59 (2014) 54–66

 
ðev c Þtrial
nþ1 ¼ ðev c Þn þ ðDev Þnþ1 ð36Þ ðkþ1Þ ðkþ1Þ
pnþ1 ¼ g ðev c Þnþ1 ;
8   
 12 2 ðkþ1Þ ðkþ1Þ ðkþ1Þ
snþ1 ¼ sn þ 2Gnþ1 Denþ1  ðDep Þnþ1 ð48Þ
  >
< trial trial
trial pa ppina þ 1þe in
; ðe v c nþ1 ðev c Þnþ1 > ev c;0
Þ
ptrial
nþ1 ¼ g ðev c Þnþ1 ¼ 2k
>
: 5. Check residual of consistency condition for convergence.
pmin ; ðev c Þtrial
nþ1 6 ev c;0 ðkþ1Þ ðkþ1Þ ðkþ1Þ ðkþ1Þ ðkþ1Þ ðkþ1Þ
/ðkþ1Þ ¼ Dsnþ1 : nnþ1  Dpnþ1 rn : nnþ1  Lnþ1 Hnþ1 ð49Þ
ð37Þ
ðkþ1Þ
If j/ j > tolerance, convergence is not reached, k = k + 1 and
strial trial go to step 4; else go to step 6.
nþ1 ¼ sn þ 2Gnþ1 Denþ1 ð38Þ
where n is the current step and n + 1 the next step. The elastic pre- 6. Update stress, strain and internal variables.
diction of stress increment along with the shear strain increment is
used to determine the number of sub-steps needed: ðkþ1Þ
pnþ1 ¼ pnþ1 ; snþ1 ¼ snþ1 ;
ðkþ1Þ ðkþ1Þ
ðev c Þnþ1 ¼ ðev c Þnþ1 ð50Þ

nsub ¼ max dDcnþ1 =tolerance1e; dDgnþ1 =tolerance2e ð39Þ
As the cutting-plane algorithm is semi-explicit, the continuum
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi tangent operator is used in the solution of the global finite element
where Dcnþ1 ¼ 23 Denþ1 : Denþ1 , tolerance1 and tolerance2 are the equations:
tolerable shear strain and shear stress invariable increment in each
rnþ1 ¼ rn þ Dep : Denþ1 ð51Þ
sub-step, de is a round up function.
After determining the number of sub-steps, the strain incre-   
ment at each sub-step is acquired through dividing the strain De : m þ Dir þD
3
re
I  n  13 ðr : nÞI : De
Dep ¼ De     ð52Þ
increment by nsub. The cutting-plane algorithm is then used for
H þ n  13 ðr : nÞI : De : m þ Dir þD3
re
I
the stress integration during each sub-step. Fig. 5 provides a graph-
ical illustration of the cutting-plane algorithm. The main concept of where De is the elastic tangent operator.
the cutting plane algorithm is to first make an elastic estimate of
the stress increment and then bring the stress back to the solution 5.3. Determination of projection point on maximum stress ratio
through plastic correction by enforcing the consistency condition surface
/ = 0 using the first order Taylor series expansion of the consis-
tency condition during each iteration (Fig. 5). The main steps of For the three dimensional implementation of the model, the
the algorithm are as follows. determination of the projection of the current stress state on the
maximum stress ratio surface is needed. As analytically solving b
1. Initialize the local iteration number k, plastic strain increment in Eq. (28) is of some difficulty, in the implementation b is solved
and loading index. numerically using the Pegasus procedure developed Dowel and
 p ðkÞ Jarratt [46] guaranteeing fast unconditional convergence, which
ðkÞ
k ¼ 0; e_ v nþ1 ¼ 0; ðe_ p Þnþ1 ¼ 0; L ¼ 0; ð40Þ has been used in the implementation of constitutive models by
Sloan et al. [47] and Andrianopoulos et al. [48]. The main steps of
2. Elastic prediction of stress state at next step.
ðkÞ the procedure are as follows.
ðev c Þnþ1 ¼ ðev c Þn þ ðDev Þnþ1 ð41Þ
  1. Set b0 = 0 and b1 = 1 initially.
ðkÞ ðkÞ ðkÞ ðkÞ
pnþ1 ¼ g ðev c Þnþ1 ; snþ1 ¼ sn þ 2Gnþ1 Denþ1 ð42Þ 2. Calculate.

3. Check consistency condition to determine whether plastic load- rðb0 Þ ¼ ain þ b0 ðr  ain Þ; rðb1 Þ ¼ ain þ b1 ðr  ain Þ ð53Þ
ing or load reversal occurs.
ðkÞ ðkÞ ðkÞ
/ðkÞ ¼ ðsnþ1  sn Þ : nnþ1  ðpnþ1  pn Þrn : nnþ1  LHnþ1
ðkÞ ðkÞ
ð43Þ fm ðb0 Þ ¼ gðb0 Þ  M m gðhðb0 ÞÞ; f m ðb1 Þ ¼ gðb1 Þ  Mm gðhðb1 ÞÞ ð54Þ
3. Judge whether rðb0 Þ and rðb1 Þ are on either side of the
If /(k) > 0, plastic loading is induced, go to step 4; else, load
maximum stress ratio surface.
reversal happens and stress–strain relationship at current
If fm(b0) fm(b1) < 0 and fm(b1) > 0 go to step 4, else if fm(b0) fm
step is assumed to be elastic, the projection centre is updated,
(b1) > 0 and fm(b1) < 0 then set b0 = b1 and b1 = 2b1 and go to step 2.
(ain)n+1 = rn, go to step 6.
4. Calculate
4. Plastic correction through the calculation of loading index
increment, as shown in Fig. 5.
fb ðb1 Þðb1  b0 Þ
,  b ¼ b1  ð55Þ
ðkÞ fb ðb1 Þ  fb ðb0 Þ
ðkÞ ðkÞ @/
DL ¼ /
@L rðbÞ ¼ ain þ bðr  ain Þ ð56Þ
ðkÞ
/
¼   ð44Þ fm ðbÞ ¼ gðbÞ  M m gðhðbÞÞ ð57Þ
ðkÞ ðkÞ ðkÞ ðkÞ ðkÞ ðkÞ
Hnþ1 þ 2Gnþ1  K nþ1 Dnþ1 rnþ1 : nnþ1
If |fm(b)| < tolerance, convergence is reached, else go to step 5.
Update loading index and stress–strain state
Lðkþ1Þ ¼ LðkÞ þ DLðkÞ ð45Þ 5. Update b0 and b1 according to the relative value of fm(b0), fm(b1)
and fm(b).
 ðkþ1Þ ðkÞ ðkþ1Þ ðkÞ
Depv nþ1
¼ Lðkþ1Þ Dnþ1 ; ðDep Þnþ1 ¼ Lðkþ1Þ nnþ1 ð46Þ
If fm(b)fm(b1) < 0, then b1 = b and fm(b1) = fm(b), then go to
  ðkþ1Þ  step 4; else if fm(b)fm(b1) > 0, then fm ðb0 Þ ¼ fmfmðbðb00Þþf
Þfm ðb1 Þ
; b1 = b
ðkþ1Þ
ðe v c Þnþ1 ¼ ðev c Þn þ ðDev Þnþ1  Depv nþ1 ð47Þ m ðb1 Þ

and fm(b1) = fm(b), go to step 4.

 R. Wang et al. / Computers and Geotechnics 59 (2014) 54–66 61

which will be used in the simulations of cyclic torsional tests as

Fig. 6. Schematic illustration of the cutting plane stress integration algorithm.
In the OpenSees implementation of the model, an elastic mate-
rial stage was incorporated into the code to avoid numerical insta-
bility in generating the initial stress state, further details of this
technique can be found in the OpenSees command manual for
the material ‘‘CycLiq’’ (http://opensees.berkeley.edu).
6. Model performance
The performance of the model is here evaluated by conducting
element experiment simulations and simulation of a centrifuge
shaking table test on a single pile in liquefiable soil.
6.1. Undrained and drained triaxial experiment simulation
well. However, as the Toyoura sand (D50 = 0.18 mm, emax = 0.973
and emin = 0.635) used by Zhang et al. varied slightly from that used
by Verdugo and Ishihara, and more importantly, the experiments
by Zhang et al. and those by Verdugo and Ishihara were conducted
using different preparation methods, different dilatancy parame-
ters (dre,1, and dir) were used for the simulation of the two respec-
tive sets of experiment (Table 1). The critical state parameters
(M, kc , e0, n) used in the simulations were adopted from the work
of Li and Wang [39] and Li and Dafalias [29].
Fig. 7 shows the simulation results of the undrained experi-
ments. With the confining pressure ranging from 0.1 to 3 MPa,
and void ratio from 0.907 to 0.735, the model was able to achieve
good agreement with experiment data using a single set of model
parameters. Thus proving the proposed model to be capable in the
simulation of highly dilative and contractive sand behaviour. The
unloading processes were also simulated, exhibiting the effective-
ness of the reversible and irreversible dilatancy formulations in the
model.
Drained triaxial experiment simulations are presented in Fig. 8.
Again, the proposed model showed good agreement with experi-
ment data over a range of initial void ratios and confining pres-
sures during both the loading and unloading stages. These
simulation results are comparable to those of existing models well
known for their simulative capabilities under such stress paths
(e.g. [22], [40]).

Triaxial data of undrained and drained loading from Verdugo
and Ishihara has been widely used in the evaluation and validation
of constitutive models (e.g. [21,22], as they cover a wide range of
confining pressures and initial void ratios. The Toyoura sand used
in these experiments had a mean diameter of D50 = 0.17 mm, max-
imum and minimum void ratios of 0.977 and 0.597 (see Fig. 6).
The model parameters used in the simulation are provided in
Table 1. Most parameters for the simulation of Toyoura sand exper-
iments were calculated via previously stated determination meth-
ods using experimental data from Zhang [30] and Zhang et al. [8]
6.2. Cyclic torsional experiment simulation
Two undrained hollow cylinder cyclic torsional experiments
were simulated using the proposed model to evaluate its capabili-
ties in modelling the cyclic response, and especially the large post-
liquefaction shear deformation of sand. The experiments were on
Toyoura sand of different densities by Zhang et al. [30].
The two tests on Toyoura sand were conducted on sand of 60%
and 48% respectively, with the shear stress amplitude for the tests
being 25 kPa, and the initial consolidation stress were 100 kPa. The

(a) 4
Calculated Experiment
3
q (MPa)

0
0 10 20 30 0 10 20 30
ε 1 (%) ε 1 (%)
(b) 4
Calculated Experiment
3 e = 0.735
e = 0.833
q (MPa)

2
e = 0.907

0
0 1 2 3 0 1 2 3
p' (MPa) p' (MPa)

Fig. 7. Simulation of undrained triaxial tests on Toyoura sand of different void ratio and initial effective confining pressure. (Experiment data from Verdugo and Ishihara,
1996).
62 R. Wang et al. / Computers and Geotechnics 59 (2014) 54–66

(a) 1.6
Calculated Experiment
1.2

q (MPa) 0.8
p0 = 0.5MPa
p0 = 0.1MPa
0.4

0
0 10 20 30 0 10 20 30
ε 1 (%) ε 1 (%)
(b) 1.6
Calculated Experiment
1.2
q (MPa)

0.8

0.4

0
0.78 0.82 0.86 0.9 0.94 0.98 0.78 0.82 0.86 0.9 0.94 0.98
e e

Fig. 8. Simulation of drained triaxial tests on Toyoura sand of different void ratio and initial effective confining pressure. (Experiment data from Verdugo and Ishihara 1996).

(a) 40
20
τ (kPa)

0
-20
-40
0 20 40 60 80 100 0 20 40 60 80 100
p' (kPa) p' (kPa)
(b) 40
20
τ (kPa)

0
-20 Calculated Experiment
-40
-0.1 -0.06 -0.02 0.02 0.06 0.1 -0.1 -0.06 -0.02 0.02 0.06 0.1
γ γ
Fig. 9. Simulation of undrained cyclic torsional test for Toyoura sand at Dr = 60%. (Experiment data from Zhang et al., 1997).

(a) 40
20
τ (kPa)

0
-20
-40
0 20 40 60 80 100 0 20 40 60 80 100
p' (kPa) p' (kPa)
(b) 40
20
τ (kPa)

0
-20 Calculated Experiment
-40
-0.1 -0.06 -0.02 0.02 0.06 0.1 -0.1 -0.06 -0.02 0.02 0.06 0.1
γ γ
Fig. 10. Simulation of undrained cyclic torsional test for Toyoura sand at Dr = 48%. (Experiment data from Zhang et al. 1997).
 R. Wang et al. / Computers and Geotechnics 59 (2014) 54–66 63

(a) 40

20

τ (kPa)
0

-20
0 20 40 60 80 100 0 20 40 60 80 100
p' (kPa) p' (kPa)
(b) 40
τ (kPa)

20

0
Calculated Experiment
-20
-0.2 -0.1 0 0.1 0.2 0.3 0.4 -0.2 -0.1 0 0.1 0.2 0.3 0.4
γ γ
Fig. 11. Simulation of undrained cyclic torsional test with a static shear stress bias for Toyoura sand at Dr = 46.6%. (Experiment data from Chiaro et al. 2013).

model parameters used are provided in Table 1, which were deter-

mined as mentioned previously. Figs. 9 and 10 compares the calcu-
lated stress path and stress–strain relationship with experiment

300
results of Toyoura sand at 60% and 48% relative density, and shows
excellent agreement between them. The model fully captures the
features of cyclic mobility of sand during loading and reverse load-

30
ing. It is worth noting that the generation of shear strain at zero Dr =0.5
effective stress after initial liquefaction and its accumulation with

500
the increasing number of load cycles is very well simulated. This
900

600
feature is a great advantage of this model over most existing mod-
els which either are unable to simulate the accumulation of shear
strain at zero effective stress (e.g. [20], [22]) or artificially generate
the shear strain at non-liquefaction state (e.g. [16], [23]). Again, Dr =0.8

250
for the simulation of cyclic tests on Toyoura sand of different den-
sities, the same set of parameters were used due to appropriate
incorporation of critical state behaviour.
To display the ability of the model in modelling large shear 1500
strains, an undrained cyclic torsional test with a static shear stress
bias for Toyoura sand at Dr = 46.6% conducted by Chiaro et al.
(2013) [49] was simulated. The preparation method for the sample Strain guage
used in the test was through air pluviation, which was the same as Accelerometer
that of Zhang’s experiments, hence the same parameters were Pore pressure transducer
used. Fig. 11 shows the shear strain accumulation reached 32% in
both the experiment and simulation. However, for more accurate (a) Centrifuge model setup (units: cm)
reflection of such or even higher levels of shear strain, geometric
nonlinearity should be taken into consideration for both the consti-
tutive model and the finite element formulation, which is beyond
the scope of the current paper.

6.3. Centrifuge experiment simulation

The capability of the proposed model was further explored

through simulation a centrifuge shaking table experiment on a sin-
gle pile in liquefiable soil conducted at the TH-50 g-ton geotechni-
cal centrifuge facility of Tsinghua University. The centrifuge has an
effective radius of 2 m, with an output of 50 kW and maximum
centrifuge acceleration of 250 g, and has successfully conducted
over 600 experiments [50]. The shaking table is capable of generat-
ing maximum accelerations of 20 g within a frequency range of
10–250 Hz.
The centrifuge shaking table experiment was carried out at 30 g
centrifugal acceleration, for clarity dimensions and measurements (b) Finite element mesh
mentioned here on will only be given in prototype scale. The basic
setup of the model is shown in Fig. 12(a). The level ground Fig. 12. Centrifuge model and finite element mesh of a single pile in liquefiable soil.
64 R. Wang et al. / Computers and Geotechnics 59 (2014) 54–66

Acceleration (m/s2) 5 20
Surface
3

Pressure (kPa)
Horizontal

Excess Pore
15
1
-1 10

-3
5
-5 1.2m

5 0
Acceleration (m/s2)

3
5m 30
Horizontal

Pressure (kPa)
1

Excess Pore
-1 20
Calculated
-3
Experiment
10
-5
2.4m
5
Acceleration (m/s2)

Input 0
3
Horizontal

40
1

Pressure (kPa)
Excess Pore
-1 30

-3 Calculated
20
-5
Experiment
0 5 10 15 20
10
Time (s) 3.3m
Fig. 13. Calculated and measured acceleration time histories at various depths. 0
0 5 10 15 20
Time (s)
consisted of two layers of saturated Fujian sand, a Chinese standard
Fig. 14. Calculated and measured short term excess pore pressure time histories at
sand, with a top medium dense (50%) layer of 5 m thick and a bot- various depths.
tom dense (70%) layer of 2.5 m. The Fujian sand used had specific
gravity of 2.65, maximum and minimum void ratios of 0.879 and
0.555 respectively. The permeability of the top and bottom layer matched experiment measurements reasonably well. Deamplifica-
was measured to be 1.5  104 m/s and 6.0  105 m/s respec- tion of acceleration was notable in both the experimental data and
tively. The aluminium pile used was 6 m long with a square cross simulation results. A comparison between short term calculated
section of 0.3 m  0.3 m and was clamped into a 0.3 m thick and measured excess pore pressures at 1.2 m, 2.4 m and 3.3 m
2.7 m  2.7 m aluminium pile cap. A steel block of 10.8 ton was depths are shown in Fig. 14, which depicts the generation process
supported by a 3 m tall 0.6 m  0.6 m steel column built on top of excess pore pressure during shaking. Good agreement is shown
of the pile cap to represent the structure. A laminar box was used between calculated and measured results, though measured pore
for the experiment. Acceleration, pore pressure in the soil and pressure results tend to show more fluctuation. It can be observed
bending moment of the pile was measured using accelerometers, that the top 3 m of soil had liquefied after about 8.5 s during the
pore pressure transducers and strain gauges (Fig. 12(a)) respec- experiment. Fig. 15 provides the long term excess pore pressure
tively. The input seismic motion had a maximum acceleration of time history at 2.4 m depth, which exhibits the models ability in
4.95 m/s2, as shown in Fig. 14. simulating the post-earthquake dissipation process of excess pore
The finite element mesh for the numerical simulation of the pressure. However, although the settlement of the ground was not
centrifuge experiment is shown in Fig. 12(b); only half of the mod- measured during the experiment, we acknowledge that the
el was needed due to symmetry. The pile, pile cap and structure
were simulated using solid Brick elements and an elastic isotropic
constitutive model. Pile cross section in the simulation consisted of 30
6 elements. The two layers of sand were simulated using Brick UP
elements and the proposed constitutive model. The model param-
Pressure (kPa)
Excess Pore

eters for Fujian sand are given in Table 1 (the critical state param- 20
eters for Fujian sand reported by Yang and Sze [51] were used).
Corresponding nodes on the two sides of the model in the shak-
10
Experiment
ing direction were tied together using the EqualDOF command
while the other two sides were fixed against lateral displacement Calculated
perpendicular to the shaking direction to simulate the effect of
the laminar box. The bottom of the model was fixed in all direc- 0
0 20 40 60 80
tions to follow the input motion, and the top of the model was free
Time (s)
with free drainage condition for soil surface.
Fig. 13 presents the simulated and measured acceleration time Fig. 15. Calculated and measured long term excess pore pressure time histories at
histories at ground surface and 5 m depth. Numerical results 2.4 m depth.

calculated ground surface settlement (5.2 cm) is most likely an
underestimation of the actual value, which is a common problem
among current plasticity models. Different researchers have attrib-
uted this underestimation of liquefaction induced ground settle-
ment to various possible causes: (1) Elgamal et al. (2003)
suggested that a highly nonlinear bulk modulus might be the
difference between constitutive model and experiment [16]; (2)
Boulanger and Ziotopoulou (2012) noted that the sedimentation
of sand after liquefaction cannot be appropriately reflected in cur-
rent constitutive frameworks [23]; (3) Zhang and Wang (2012)
noted that the U-p formulation for coupled elements neglects the
velocity effects of the liquid phase, which has an important role
at high centrifugal acceleration levels in dynamic centrifuge exper-
iments [31]; (4) Shahir et al. (2012) proposed that the variation of
permeability as a major cause for the difference between calcula-
tions and experiments [52]. Also, the proposed model in this paper
neglects the plastic volumetric strains during constant stress ratio
loading, and would also to some extend cause underestimation of
the reconsolidation settlement of sand.
Fig. 16 compares the calculated bending moment history at pile
head with experiment result, showing excellent agreement. The
maximum measured moment at the pile head was 58kNm, while
the maximum calculated moment was 60kNm. The peak bending
moment in the pile was observed at 7s. Fig. 17 presents the peak
bending moment distribution along the pile. The maximum bend-
ing moment was observed at the pile head, while no bending
moment occurred at the pile tip as it was not fixed against rotation.
Overall, the simulations of element and centrifuge experiments
showed great agreement with experimental data, the simulations
40
Moment (kNm)
0
-40
-80
0 5 10
Time (s)
Fig. 16. Calculated and measured pile bending moment time histories at pile top.
Moment (kNm)
-80 -40
0
1
2 China (No. 51079074 and No. 51038007) for funding the work
Depth (m)
3 the paper for their advice and suggestions.
4 References
5 during earthquakes. J Geotech Eng Div 1979;105(2):201–55.
6 Calculated
Experiment
7 related damage to structures during 1991 Telire-Limon, Costa Rica,
Fig. 17. Calculated and measured maximum bending moment distribution along
the pile.
R. Wang et al. / Computers and Geotechnics 59 (2014) 54–66 65
proved the potential of the model and its numerical
implementation.
7. Conclusions
This paper presents the formulation and numerical implemen-
tation of a unified constitutive model for seismic liquefaction
analysis based on the physics of post-liquefaction deformation
established by Zhang and Wang [31]. The model is unique in that
it provides a unified description of sand of different conditions
from pre- to post-liquefaction under monotonic and cyclic loading.
By enforcing the volumetric compatibility equation at liquefac-
tion according to Zhang and Wang’s [31] proposition, the model
provides physically based computation of the generation and
accumulation of shear strain at zero effective stress, which is a
significant advantage over most existing models.
Through the appropriate formulation of two dilatancy compo-
nents, namely reversible and irreversible, the model explicitly links
the phenomenon of cyclic mobility to soil dilatancy, providing
excellent modelling capabilities for both monotonic and cyclic
response of sand.
The state parameter W was incorporated into the model for
compatibility with the critical state soil mechanics concept. The
model was unified to allow the simulation of sand at different
relative densities and confining pressures with a same set of
parameters.
In the multiaxial formulation, the mapping rules for plasticity
and dilatancy were proposed to be suitable in three dimensional
space and accommodate the three dimensional numerical
implementation of the model. Calibration methods for the model
parameters were also provided.
Calculated Using a cutting-plane algorithm with substepping as the stress
Experiment integration scheme, and the Pegasus procedure to locate the
projection of current stress state on the maximum stress ratio sur-
face, the model was implemented in the OpenSees finite element
framework, making it openly available to the technical community.
The constitutive model and its three dimensional numerical
implementation were validated against drained and undrained
triaxial experiments, undrained cyclic torsional experiments and
15 20 a centrifuge experiment on single pile in liquefiable ground, show-
ing the great capabilities of the model in simulating sand response
of a wide range of densities and confining pressure, and highlight-
ing its advantage in simulating large post-liquefaction shear
deformations.
However, as the current model underestimates the reconsolida-
tion settlement of sand, which is something that could be looked
0 into in future improvements.
Acknowledgements
The authors thank the National Natural Science Foundation of
presented in this paper, and would like to thank the reviewers of
[1] Seed HB. Soil liquefaction and cyclic mobility evaluation for level ground
[2] Hamada M. Large ground deformations and their effects on lifelines: 1964
Niigata earthquake. case studies of liquefaction and lifelines performance
during past earthquake. Technical Report NCEER-92-0001, National Centre for
Earthquake Engineering Research, Buffalo; 1992.
[3] Yoshida N, Watanabe H, Yasuda S. Liquefaction-induced ground failure and
earthquake. In: Proceedings from the 4th Japan–U.S. Workshop on
earthquake resistant design of lifeline facilities and countermeasures for soil
liquefaction, Hawaii; 1992. p. 37–52.
66 R. Wang et al. / Computers and Geotechnics 59 (2014) 54–66
[4] Eberhard MO, Baldridge S, Marshal J, Mooney W, Rix GJ. The MW 7.0 Haiti
earthquake of January 12, 2010, USGS/EERI Advance Reconnaissance Team;
2010.
[5] Seed HB, Lee KL. Liquefaction of saturated sands during cyclic loading. J Soil
Mech Found Eng Div ASCE 1966;92(SM6):105–34.
[6] Arulmoli K, Muraleetharan KK, Hossain MM, Fruth LS. VELACS: verification of
liquefaction analysis by centrifuge studies, laboratory testing program, soil
data report. The Earth Technology Corporation, Project No. 90–0562, Irvine,
California; 1992.
[7] Kutter BL, Chen Y, Shen CK. Triaxial and torsional shear test results for sand.
Naval Facilities Engineering Service Center, Contact Report CR 94.003-SHR,
Port Hueneme, California; 1994.
[8] Zhang JM, Shamoto Y, Tokimatsu K. Moving critical and phase-transformation
stress state lines of saturated sand during undrained cyclic shear. Soils Found.
1997;2(37):51–9.
[9] Pastor M, Zienkiewicz OC, Chan A. Generalized plasticity and the modelling of
soil behaviour. Int J Numer Anal Meth Geomech 1990;14(3):151–90.
[10] Zienkiewicz OC, Mroz Z. Generalized plasticity formulation and application to
geomechanics. Mech Eng Mater. In: Desai CS, Gallagher RH, editors, John Wiley
& Sons; 1984. p. 655–79.
[11] Wu W, Bauer E. A simple hypoplastic constitutive model for sand. Int J Numer
Anal Meth Geomech 1994;18(12):833–62.
[12] Wu W, Bauer E, Kolymbas D. Hypoplastic constitutive model with critical state
for granular materials. Mech Mater 1996;23(1):45–69.
[13] Mroz Z, Norris VA, Zienkiewtcz OC. An anisotropic hardening model for soils
and its application to cyclic loading. Int J Numer Anal Meth Geomech
1978;3(2):203–21.
[14] Prevost JH. A simple plasticity theory for frictional cohesionless soils. Int J Soil
Dyn Earthquake Eng 1985;4(1):9–17.
[15] Parra-Colmenares EJ. Numerical modeling of liquefaction and lateral ground
deformation including cyclic mobility and dilation response in soil systems,
Ph.D. dissertation, Dept. of Civil Engineering, Rensselaer Polytechnic Institute,
Troy, New York; 1996.
[16] Elgamal A, Yang ZH, Parra E. Computational modeling of cyclic mobility and
post-liquefaction site response. Soil Dyn Earthquake Eng 2002;22(4):259–71.
[17] Elgamal A, Yang Z, Parra E, Ragheb A. Modeling of cyclic mobility in saturated
cohesionless soils. Int J Plast 2003;19(6):883–905.
[18] Yang Z, Elgamal A, Parra E. Computational model for cyclic mobility and
associated shear deformation. J Geotech Geoenviron Eng
2003;129(12):1119–27.
[19] Wang ZL, Dafalias YF, Shen CK. Bounding surface hypoplasticity model for
sand. J Eng Mech 1990;116(5):983–1001.
[20] Papadimitriou AG, Bouckovalas GD, Dafalias YF. Plasticity model for sand
under small and large cyclic strains. J Geotech Geoenviron Eng
2001;127(11):973–83.
[21] Wang ZL, Dafalias YF. Simulation of post-liquefaction deformation of sand.
Constitut Model Geomater 2003:100–7.
[22] Dafalias YF, Manzari MT. Simple plasticity sand model accounting for fabric
change effects. J Eng Mech 2004;130(6):622–34.
[23] Boulanger RW, Ziotopoulou K. Formulation of a sand plasticity plane-strain
model for earthquake engineering applications. Soil Dyn Earthquake Eng
2013;53:254–67.
[24] Roscoe KH, Schofield AN, Wroth CP. On the yielding of soils. Geotechnique
1958;8(1):22–53.
[25] Schofield AN, Wroth CP. Critical state soil mechanics. London: McGraw-Hill;
1968.
[26] Been K, Jefferies MG. A state parameter for sands. Geotechnique
1985;35(2):99–112.
[27] Ishihara K. Liquefaction and flow failure during earthquakes. Geotechnique
1993;43(3):351–415.
[28] Wood MD, Belkheir K, Liu DF. Strain softening and state parameter for sand
modelling. Geotechnique 1994;50(4):449–60.
[29] Li XS, Dafalias YF. Dilatancy for cohesionless soils. Geotechnique
2000;50(4):449–60.
[30] Zhang JM. Cyclic critical stress state theory of sand with its application to
geotechnical problems, Ph. D thesis, Tokyo Institute of Technology, Tokyo;
1997.
[31] Zhang JM, Wang G. Large post-liquefaction deformation of sand, part I:
physical mechanism, constitutive description and numerical algorithm. Acta
Geotech 2012;7(2):69–113.
[32] Shamoto Y, Zhang JM. Mechanism of large post-liquefaction deformation in
saturated sands. Soils Found 1997;2(37):71–80.
[33] Zhang JM. Reversible and irreversible dilatancy of sand. Chin J Geotech Eng
2000;1(22):12–7.
[34] Zhang JM, Wang G. Mechanism of large post-liquefaction deformation of
saturated sand. Chin J Geotech Eng 2006;28(7):835–40 [in Chinese].
[35] McKenna F, Fenves GL. OpenSees manual, PEER Center; 2001. <http://
OpenSees.berkeley.edu>.
[36] Verdugo R, Ishihara K. The steady state of sandy soils. Soils Found
1996;36(2):81–92.
[37] Dafalias YF, Popov EP. A model of nonlinearly hardening materials for complex
loading. Acta Mech 1975;21(3):173–92.
[38] Richart Jr FE, Hall JR, Woods RD. Vibrations of soils and
foundations. Englewood Cliffs, N.J.: Prentice-Hall Inc.; 1970.
[39] Li XS, Wang Y. Linear representation of steady-state line for sand. J Geotech
Geoenviron Eng 1998;124(12):1215–7.
[40] Ling HI, Yang S. Unified sand model based on the critical state and generalized
plasticity. J Eng Mech 2006;132:1380–91.
[41] Rowe PW. The stress-dilatancy relation for static equilibrium of an assembly of
particles in contact. Proc R Soc Lond A 1962;269(1339):500–27.
[42] Andrianopoulos KI, Papadimitriou AG, Bouckovalas GD. Bounding surface
plasticity model for the seismic liquefaction analysis of geostructures. Soil Dyn
Earthquake Eng 2010;30(10):895–911.
[43] Simo JC, Ortiz M. A unified approach to finite deformation elastoplastic
analysis based on the use of hyperelastic constitutive equations. Comput
Methods Appl Mech Eng 1985;49(2):221–45.
[44] Yang Z, Lu J, Elgamal A. OpenSees soil models and solid-fluid fully coupled
elements user manual. San Diego, California: University of California; 2008.
[45] McGann CR, Arduino P, Mackenzie-Helnwein P. Stabilized single-point 4-node
quadrilateral element for dynamic analysis of fluid saturated porous media.
Acta Geotech 2012:1–15.
[46] Dowell M, Jarratt P. The ‘‘Pegasus’’ method for computing the root of an
equation. Bit Numer Math 1972;12(4):503–8.
[47] Sloan SW, Abbo AJ, Sheng D. Refined explicit integration of elastoplastic
models with automatic error control. Eng Comput 2001;18(1/2):121–54.
[48] Andrianopoulos KI, Papadimitriou AG, Bouckovalas GD. Explicit integration of
bounding surface model for the analysis of earthquake soil liquefaction. Int J
Numer Anal Meth Geomech 2010;34(15):1586–614.
[49] Chiaro G, Kiyota T, Koseki J. Strain localization characteristics of loose
saturated Toyoura sand in undrained cyclic torsional shear tests with initial
static shear. Soils Found 2013;53(1):23–34.
[50] Wang R, Zhang G, Zhang JM. Centrifuge modelling of clay slope with
montmorillonite weak layer under rainfall conditions. Appl Clay Sci
2010;50(3):386–94.
[51] Yang J, Sze HY. Cyclic strength of sand under sustained shear stress. J Geotech
Geoenviron Eng 2011;137:1275–85.
[52] Shahir H, Pak A, Taiebat M, Jeremi C. B. Evaluation of variation of permeability
in liquefiable soil under earthquake loading. Comput Geotech 2012;40:74–88.