Computers and Geotechnics 49 (2013) 100–110

Contents lists available at SciVerse ScienceDirect

Computers and Geotechnics

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

Loading path dependence and non-linear stiffness at small strain using

rate-dependent multisurface hyperplasticity model
Dedi Apriadi a, Suched Likitlersuang b,⇑, Thirapong Pipatpongsa c
a
Faculty of Civil and Environmental Engineering, Institut Teknologi Bandung, Bandung, Indonesia
b
Department of Civil Engineering, Faculty of Engineering, Chulalongkorn University, Payathai Road, Pathumwan, Bangkok 10330, Thailand
c
Global Scientific Information and Computing Centre, Tokyo Institute of Technology, Tokyo, Japan

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

Article history: Many interpretations of small-strain experiments indicate a non-linear dependence of soil stiffness on
Received 31 December 2009 pressure. This shows that small-strain stiffness can be expressed as a power function of the mean effec-
Received in revised form 25 September tive stress rather than as a linear function of stress. Many cases in the field show the importance of these
2012
behaviours to load–deformation prediction in a soil-structure interaction problem. This paper presents a
Accepted 15 November 2012
numerical implementation and validation of non-linear pressure-dependent stiffness in a hyperplasticity
model using a strain-driven forward-Euler integration scheme. The kinematic hardening function was
incorporated into a finite number of multiple-yield surfaces of Modified Cam Clay to characterise
Keywords:
Hyperplasticity
small-strain stiffness and accommodate smooth transition of stiffness corresponding to different loading
Small strain conditions and stress histories. Experimental results of current state and loading history dependence in
Rate-dependent overconsolidated clay are compared to the model prediction.
Recent stress history Ó 2012 Elsevier Ltd. All rights reserved.
Non-linear stiffness

1. Introduction transition of stiffness corresponding to different loading conditions

and stress histories, the kinematic hardening function was incor-
Interpretations of experimental data show that initial soil stiff- porated in both energy and yield functions into a finite number
ness or small-strain stiffness is a non-linear function of mean effec- of multiple-yield surfaces. However, numerical implementation
tive stress [36,21,27,37]. Stiffness is also affected by other and calibration of the model were restricted to a linear stress–
variables, such as the void ratio, anisotropic stress state and/or strain relationship with stiffness proportional to isotropic pressure
the preconsolidation pressure [8,14,38,30,33]. Detailed experimen- [23].
tal investigations into the stress–strain response of overconsolidat- Hyperplastic formulations are capable of describing a non-lin-
ed soil have also shown that soil characteristics are dependent on ear response from small to large strain level. Small-strain non-lin-
both non-linearity and the most recent loading paths [19,5,34,12]. ear stiffness can be expressed as a power function of the stress
The effect of current loading history has also been observed using a level to meet the non-linear response of materials observed in
different experimental approach [18,15,32]. It has been observed experiments. The current study extends the previous work on the
that zones exist at small strains, and they can change in both shape linear version to the non-linear version, and the difficulty lies in
and size as the soil is subjected to different loading histories. working out the mathematics and numerical implementation.
The importance of these two behaviours in predicting load Apriadi [2] and Apriadi et al. [3] employed a numerical implemen-
deformation in soil-structure interaction problems has been ob- tation of the non-linear KHMCC model using a strain-driven for-
served in many cases in the field [17,16,7,1]. To include certain ward-Euler integration scheme under triaxial conditions by
physical principles such as conservation of mass, conservation of adopting small-strain stiffness in the form of the power function
energy and the laws of thermodynamics in soil constitutive model- of pressure [11] in an energy function. Their work also presented
ling, Likitlersuang [25] and Likitlersuang and Houlsby [24] an approach to determine the necessary parameters, obtained from
proposed the hyperplasticity kinematic hardening Modified Cam bender element tests, for regulating the small-strain stiffness char-
Clay (KHMCC) model based on thermodynamics principles. To acteristic of the hyperplasticity KHMCC model. Numerical model
characterise small-strain stiffness and accommodate smooth implementation was verified against an analytical solution of ideal
undrained triaxial test on normally consolidated clay [31,28]. The
⇑ Corresponding author. Tel.: +66 2 2186343; fax: +66 22157304. model’s capabilities in characterising the effect of recent stress
E-mail address: Suched.L@eng.chula.ac.th (S. Likitlersuang). history, the dependence of effective stress path on recent stress

0266-352X/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compgeo.2012.11.007
 D. Apriadi et al. / Computers and Geotechnics 49 (2013) 100–110 101

qffiffiffiffiffiffiffiffiffiffiffiffiffi
history during undrained shear, and the smooth and irreversible are ap = aii and aq ¼ 2 0 0
a a , where a0ij ¼ aij  13 aii dij is a deviatoric
3 ij ij
unloading–reloading responses under triaxial stress–strain condi-
tions were highlighted. Some comparisons with experimental data component of the internal variable tensor.
on clay soils have also been carried out, such as the response of
Bangkok clay to the K0-consolidated undrained direct simple shear 2.2. Hyperplasticity formulation
(CK0UDSS) test.
This work performs further numerical implementation of the Under the hyperplasticity framework (for details, see 10), the
non-linear KHMCC model in general stress conditions and validates entire constitutive model is fundamentally governed by two scalar
it with an experiment using small strain levels on clay soils. It im- functions: energy function and yield function or dissipation func-
proves an existing model by incorporating a more realistic depen- tion. In a typical formulation, Gibbs free energy function is sug-
dency of the small-strain stiffness on the mean effective stress. gested to associate stresses, material parameters and material
Owing to the limited data available on both consolidation pressure memories (internal variables) with a unique potential function in
dependent and stress path dependent stiffness, the present work which the reference pressure for zero volumetric strain is defined
focuses on the latter. However, the effect of consolidation pressure at 1 atm. It is noted that the hyperplasticity framework employs
on soil stiffness can be described with a single set of parameters total energy instead of a rate form of energy, which is commonly
[23]. The model demonstrates the non-linear stiffness response to used in classical plasticity theory. Since Gibbs free energy is equiv-
generalised stresses. Therefore, it is not only consolidation pressure alent to complementary energy, the negative elastic stored energy,
but also the stress path direction that affects soil stiffness. negative dissipation energy and positive hardening energy are
This article aims to present the capabilities of multiple-yield sur- combined.
face types of the constitutive model, which is based on hyperplas- A non-linear version of KHMCC is created to allocate non-linear
ticity theory. A slight modification of mathematical expressions elastic moduli at a small strain [11] to the Gibbs free energy, which
from the previous work is made based on the experimental evi- is expressed in the form E ¼ Eðrij ; aij ; a
^ ij Þ:
dence. Further numerical implementation under the hyperplastici- p2n
e p
ty framework of the non-linear dependence of soil stiffness on E¼ þ  rij aij
p1n
r kð1  nÞð2  nÞ kð1  nÞ
pressure, which is shown by many interpretations of the experi- Z !
1 b pa
H ^ 2kk Hb qa
^ 0ij a
^ 0ij
mental data, is developed in this work to present an experimentally þ þ dg; ð1Þ
realisable implementation of the hyperplasticity constitutive 0 2 3
model. qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
An experiment of undrained triaxial test on Speswhite kaolin where pe ¼ p2 þ kð1nÞ 3g
q2 is defined as an equivalent stress variable
[34] was selected to calibrate model parameters. This experiment for convenience.
was conducted to show the dependence of non-linearity, current The variables k,g and n are dimensionless material constants
state and loading history on relative directions of the current and calibrated from elastic stress–strain relations at small strain levels.
previous loading paths in overconsolidated clay. Model validation Atmospheric pressure, 1 atm (approximately 100 kPa), is defined
was performed with the use of the response of tangent stiffness for pr as a reference pressure, while aij is the total plastic strain ten-
against stresses. Further, the experimental result of unloading– sor. Integration of differential hardening energy is evaluated in
reloading at small strains was compared to the model prediction, terms of the internal coordinate g, which is limited to values from
and a study of the effect of the number of yield surfaces on 0 to 1. Here, g = 0 represents the initial hardening stage (the high-
stress–strain response was likewise carried out. est hardening response), while g = 1 represents the final hardening
The model formulation is initially described in terms of general stage (zero hardening response). The variable a ^ ij is the kinematic
stress conditions rather than triaxial stress conditions. Detailed internal variable tensor function of g, which can be integrated to
numerical implementation into three-and two-dimensional obtain aij using Eq. (2). Therefore, aij can be regarded as a definite
stress–strain cases and model parameter determination are then integral area of the functional variable a ^ ij over the domain of g. It is
presented, followed by a brief explanation of the test programme. noted that all variables with ‘‘^’’ (hat) throughout this paper are re-
Finally, remarks are made on the results of the model validation. ferred to as internal variable functions of g.
This study is expected to contribute to the refinement of load– Z 1
deformation prediction in soil-structure interaction problems. aij ¼ a^ ij dg: ð2Þ
0

2. Non-linear kinematic hardening Modified Cam Clay model b p; H

H b q are non-linear kinematic hardening functions corresponding
to isotropic and deviatoric hardening responses, respectively, as ex-
2.1. Notation pressed in the following equations:
1n n
The standard soil mechanics sign convention of taking compres- b p ¼ kp0 pr ð1  gÞ3 ;
H ð3Þ
sive stresses and strains as positive is used throughout this re- 2ða  1Þ
search, and all stresses are effective stresses. The following 3gp n 1n
bq ¼ 0 pr
notation is adopted: rij is the effective Cauchy stress tensor, eij is H ð1  gÞ3 ; ð4Þ
2ða  1Þ
the small-strain tensor and dij is Kronecker’s delta (dij = 1 if i = j,
dij = 0 if i – j, where i, j 2 {1, 2, 3}). The stress invariants are where p0 is a mean pre-consolidation pressure at the end of the con-
qffiffiffiffiffiffiffiffiffiffiffiffiffiffi
p ¼ 13 rii ; q ¼ 32 r0ij r0ij , where r0ij ¼ rij  pdij is a deviatoric compo- solidation stage and a is a material constant.
In the hyperplasticity framework, total strain components are
nent of the effective stress tensor. The corresponding strain invari-
qffiffiffiffiffiffiffiffiffiffiffiffi considered conjugate variables of stresses, which are derived from
ants are ep = eii and eq ¼ 23 e0ij e0ij , where e0ij ¼ eij  13 eii dij is a the Gibbs free energy function. From Eq. (1), the strains eij can be
deviatoric component of the strain tensor. Similarly, generalised obtained via differentiation with stresses rij as shown in the fol-
qffiffiffiffiffiffiffiffiffiffiffiffiffi
lowing equation:
stress invariants are vp ¼ 13 vii ; vq ¼ 32 v0ij v0ij , where
 
v0ij ¼ vij  vp dij is a deviatoric component of the effective general- @E 1 p r0ij
eij ¼  ¼  1 d ij þ þ aij : ð5Þ
ised stress tensor. The corresponding internal variable invariants @ rij 3kð1  nÞ p1n
r pne 2gpr1n pne
102 D. Apriadi et al. / Computers and Geotechnics 49 (2013) 100–110

*rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi +2
However, the Gibbs free energy function expressed in Eq. (1) and  2
the strain components derived in Eq. (5) can produce only a con- v^ 2p þ v^Mq  ^c
^i2
hy
stant modulus (n = 0) or power function of pressure-dependent w¼ ¼ ; ð12Þ
modulus (0 < n < 1). For a linear pressure-dependent modulus 2l 2l
(n = 1), these equations will be replaced by the following equations where l is the artificial viscosity coefficient in units of stress  time
[10]: [6]. It is defined as 1 kPa  second for rate-dependent calculations if
    Z 1 b ^2 b ^2! the SI physical unit of stress in kPa and unit of time in seconds are
p p q2 H p ap H q aq
E¼ ln 1   rij aij þ þ dg; ð6Þ employed. The operator h i is called Macaulay brackets, which de-
k pr 6gp 0 2 2
    0;  < 0
r0ij fine hi ¼ .
@E 1 p q2 dij ;  P 0
eij ¼  ¼ ln  þ þ aij : ð7Þ
@ rij k pr 6gp2 3 2gp However, l can actually be defined as a true viscosity coeffi-
cient to incorporate rate-dependent behaviours of soil, such as
The second derivatives of Gibbs free energy are derived in Box 1.
creep problems.
This form is applicable to 0 6 n 6 1.
According to Eqs. (5) and (7), total strain components are de-
Box 1: Second derivatives of Gibbs free energy (after 10).
rived as functions of rij and aij. Following the flow rule, the rate
of change of total strain components can be written by the follow-
ing equation:
8   9
 n > nq2 dij dkl np > " #
r r Z 1" #
< 1 0 0 =
@ E 2
1 pr k
þ 2
3gpe 9
 6gpe 2 ij d kl þ dij kl
 ¼   @2E @2E
@ rij @ rkl pr pe > fe_ ij g ¼  fr_ kl g þ  ^_ kl gdg:
fa ð13Þ
: þ1 d d  2g
d
ik jl
kl dij
3
 nkð1nÞ
r0 r0 >
4g 2 p2e ij kl
; @ rkl @ rij 0 ^ kl @ rij
@a

By using Eqs. (11) and (12), the rate form of the stress–strain rela-
tionship obtained in Eq. (13) can be expressed.
@2E
 ¼ dij dkl ¼ identity matrix " # Z 1" #

@ aij @ rkl
^ @2E @2E @w
 fr_ kl g ¼ fe_ ij g   dg: ð14Þ
@ rij @ rkl 0 @ rij @ a
^ kl @ v^ kl

The derivative of flow potential w in Eq. (11) is shown in the follow-

In the non-linear KHMCC model, the yield function in terms of ing equation:
generalised stress variables is defined as follows: *rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi +
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  2
 2 v^ 2p þ v^Mq  H b pa
^p g
^¼ v
^ 2p þ
v^ q 2
v^ p dij þ M32 v^ 0ij
y  ^c; ð8Þ @w 3
M ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2ffi : ð15Þ
@v
^ ij l
2 v ^ 2p þ v^Mq
where M is a frictional critical state parameter [31], which is the va-
b pa
lue that the stress ratio q/p0 attains at critical state; ^c ¼ H ^ p g is a
hardening stress which represents the size of the yield surface func-
tion; and v^ ij are generalised stresses. When the stress is inside the 3. Numerical implementation
yield surface ðy ^ < 0Þ, materials behave elastically, but they behave
plastically when the yield surface is active ðy ^ P 0Þ. 3.1. Incremental stress–strain response
The generalised stresses v ^ ij are defined as the changing of the
free energy functional with respect to the internal kinematic vari- The rate form of the stress–strain relationship based on contin-
ables. The Gibbs free energy functional b E can be defined as the uous hyperplasticity is described in the previous section. However,
internal energy function of g. So the integration of b E throughout continuous hyperplasticity is suitable for monotonic loading with
the domain of g, as shown in Eq. (9), is the Gibbs free energy. simple hardening functions. In order to handle complicated loading
Z 1
conditions, multisurface hyperplasticity is employed in numerical
E¼ b
Edg; ð9Þ implementations. Non-linearity is expressed by multiple piece-
0 wise responses (see also [26]). Therefore, multiple internal vari-
p2n b ^ 2p
Hpa b ^ 2q
Hqa ables play a role as discrete memories of materials, and the smooth
where b p
E ¼  p1n kð1nÞð2nÞ þ kð1nÞ
e
 rij a
^ ij þ
2
þ 2 .
r
transition between piece-wise responses depends on the finite
Therefore, v^ ij is referred to in Eq. (10) as the derivatives of b
E
number of internal variables. For linear and non-linear KHMCC
with respect to a ^ ij .
models, continuous yield surface is discretised to a finite number
@b
E 2 of yield surfaces [25,24,3]. The integration operator simply turns
v^ ij ¼  ^ ¼ rij  Hb p a^ p dij  Hb qa
^ 0ij : ð10Þ to the summation operator without losing the general meaning.
@ aij 3
Each yield surface has its own state variables, which are general-
The evolution rule of the kinematic internal variable functions ised and hardening stresses. A finite number of hardening stresses
a^ ij is derived from the derivatives of flow potential w with respect can be considered to be multiple material memories, which are up-
to generalised stress variables [10]: dated when the multiple-yield surfaces are active. An illustration
@w of multiple-yield surfaces in a three-dimensional stress space is
a^_ ij ¼ ^ : ð11Þ shown in Fig. 1.
@ vij
The rate-dependent incremental response is obtained by inte-
One may associate this kind of evolution rule with the flow rule grating the incremental stress–strain relation using the strain-dri-
used in classical plasticity. The flow potential w is defined based ven forward-Euler integration scheme. This indicates that an
on elasto-viscoplastic theory (for details, see 40) in relation to the increment of the variable based on the rate form x_ ¼ f ðxÞ can
^, as shown in Eq. (12).
yield function y typically be written in the manner xi+1  xi = f(xi)Dt, where
 D. Apriadi et al. / Computers and Geotechnics 49 (2013) 100–110 103

2. Determine plastic parameters.

for m 2 {1, 2, . . . , N}
if y(m,t) 6 0 then a_ ijðm;tÞ ¼ 0
@w
else a_ ijðm;tÞ ¼ @ v ðm;tÞ
ijðm;tÞ

3. Compute incremental variables.

" #1 " " # !#

@2E 1X N
@2E
Drkl ¼ e_ ij   a_ klðm;tÞ Dt
@ rij @ rkl N m¼1 @ aklðm;tÞ @ rij

4. Update state variables.

rijðtþ1Þ ¼ rijðtÞ þ Drij

for m 2 {1, 2, . . . , N}

Fig. 1. Non-linear KHMCC yield surfaces in three-dimensional stress space [3].

aijðm;tþ1Þ ¼ aijðm;tÞ þ a_ ijðm;tÞ Dt

Dt = ti+1  ti with i + 1 representing the current step number. 2

Eq. (14) combined with (12) is used to update stress components vijðm;tþ1Þ ¼ rijðtþ1Þ  HpðmÞ akkðm;tþ1Þ dij  HqðmÞ aijðm;tþ1Þ
3
in any increment of strain, and can be used to update the general-
ised stress components using Eq. (10) after the internal variable 5. Go to the next incremental step.
has been updated from the evolution rule given by Eq. (11). The
algorithm of the rate-dependent numerical implementation is ex-
plained in Box 2. 3.2. Two-dimensional stress–strain implementation
The subscripts m and t are positive integers representing the in-
dex of the yield surfaces and the increment of specific variables, For general stress and strain, there are six independent stress
respectively. A subscript index enclosed by parentheses () is used and strain components, i.e., the components of the stress tensor
to distinguish between the tensor expression index. For an arbi- rij, where rij = rji, and the strain tensor eij, where eij = eji. Based
trary variable tensor , the component of the tensor in accordance on Eq. (14), the incremental stress–strain relationships of the
with the yield surface-mth at increment-ith is represented by ij(m,t), rate-dependent strain-driven forward-Euler integration scheme
where N is the number of yield surfaces. All variables with ‘‘^’’ are under general stress can be generally formulated. The components
removed and notified by a subscript index (m), as well as replacing of compliance tensor Cijkl (second derivatives of Gibbs free energy
a term g with a ratio m/N and dg with a ratio 1/N in a numerical with respect to stresses) is calculated from Box 1. The evolution
process of discretisation. rule of kinematic internal variable function tensor aij(m,t) is deter-
Box 2: Strain-driven forward-Euler integration scheme. mined from Eq. (2), using Eqs. (11) and (15).
In two-dimensional cases, usually four stress tensor compo-
nents are involved: three normal stresses r11, r22, r33 and one
For given Gibbs free energy E; e_ ij ; Dt; eijðtÞ ; rijðtÞ ; aijðm;tÞ
shear stress r12. Then, the increment in stress–strain relationships
for m 2 {1, 2, . . . , N}:
of the rate-dependent strain-driven forward-Euler integration
1. Initialise.
scheme can simply be shown by the following equation:
1X N 2 38 9 8 9 8 9
eijðtþ1Þ ¼ eijðtÞ þ e_ ij Dt; aijðtÞ ¼ aijðn;tÞ C 11 C 12 C 13 C 14 > Dr1 > > De1 > > a_ 1ðm;tÞ >
N n¼1 >
> >
> >
> >
> >
> >
>
6C C 24 7 < = < De = 1 X N < _
a2ðm;tÞ =
6 21 C 22 C 23 7 Dr2 2
6 7 ¼  Dt:
4 C 31 C 32 C 33 C 34 5>>
> Dr3 >
>
>
>
>
> De3 >
>
>
N m¼1>
>
> a_ 3ðm;tÞ >
>
>
for m 2 {1, 2, . . . , N} : ; : ; : ;
C 41 C 42 C 43 C 44 Dr4 De4 a_ 4ðm;tÞ
ð16Þ
kp0 pr1n  m3  m3
n
3gpn0 p1n
r
HpðmÞ ¼ 1 ; HqðmÞ ¼ 1
2ða  1Þ N 2ða  1Þ N For computer programming purposes, r1, r2, r3, r4 are defined
as r11, r22, r33, r12 and e1, e2, e3, e4 are defined as e11, e22, e33, c12,
where c12 = e12 + e21 is an engineering shear strain. A similar defi-
2 nition is also applied to v1, v2, v3, v4, which are defined as v11,
vijðm;tÞ ¼ rijðtÞ  HpðmÞ akkðm;tÞ dij  HqðmÞ a0ijðm;tÞ v22, v33, v12, and to a1, a2, a3, a4, which are defined as a11, a22,
3
a33, a12 + a21. The Gibbs free energy and yield function can be
rewritten as shown in Box 3. The component of compliance matrix
m
cðm;tÞ ¼ HpðmÞ akkðm;tÞ ; yðm;tÞ Cij in Eq. (16) can be simply defined in expanded form as shown in
N
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Box 4. The effective generalised stress tensor vi(m,t) is calculated
vkkðm;tÞ 3 v0ijðm;tÞ v0ijðm;tÞ
2
using Eqs. (17) and (18). The kinematic internal variable function
¼ þ  cðm;tÞ tensor ai(m,t) is calculated using Eqs. (19) and (20).
9 2 M2
@E X
3
2
2 viðm;tÞ ¼  ¼ riðtÞ  Hpðm;tÞ akðm;tÞ  Hqðm;tÞ a0iðm;tÞ ;
hyðm;tÞ i @ aiðm;tÞ k¼1
3
wðm;tÞ ¼
2l i ¼ 1; 2; . . . ; 3; ð17Þ
104 D. Apriadi et al. / Computers and Geotechnics 49 (2013) 100–110

@E 1 Yield function:
v4ðm;tÞ ¼  ¼ r4ðtÞ  Hqðm;tÞ a04ðm;tÞ ; ð18Þ
@ a4ðm;tÞ 3
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
@wðm;tÞ  2 v 2 X3
m
a_ iðm;tÞ ¼
@ viðm;tÞ
yðm;tÞ ¼ vpðm;tÞ þ qðm;tÞ  HpðmÞ aiðm;tÞ ;
M i¼1
N
!
X
3
2
vkðm;tÞ þ M32 v0iðm;tÞ where
9
hyðm;tÞ i k¼1
¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ;
l P3 2 P3  02  rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
v
02
2v4ðm;tÞ þ
k¼1
vkðm;tÞ 1X 3
3  02 X3  
2 k¼1 kðm;tÞ
þ 32 vpðm;tÞ ¼ v ; vqðm;tÞ ¼ 2v4ðm;tÞ þ v02iðm;tÞ :
9 M2 3 i¼1 iðm;tÞ 2 i¼1

i ¼ 1; 2; . . . ; 3; ð19Þ

@wðm;tÞ
a_ 4ðm;tÞ ¼
@ v4ðm;tÞ
hyðm;tÞ i M
6
2 v04ðm;tÞ Box 4: Compliance matrix in expanded form for two-dimen-
¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 : ð20Þ
l P3 2 P3  02  sional implementation.
v 2v024ðm;tÞ
þ v kðm;tÞ
k¼1
k¼1 kðm;tÞ
2 9
þ 32 M2

Box 3: Gibbs free energy and yield function for two-dimensional 8   n n  o

> 1 pr nq2 1 npr0
implementation. >
>
> pr pe
1
k
þ 3gp 2 9
 3gp2i þ 3g 1
 nkð1nÞ
4g 2 p2e
r02i ; i ¼ j 2 f1; 2; 3g
>
> 8 e e
  9
>
>   <
>
> n  np r0i þr0j =
>
> 1 pr nq2 1 nkð1nÞ 0 0
Gibbs free energy for n – 1: <p p 1
k
þ 3gp2 9  6gp2  6g  4g2 p2 ri rj ; i – j 2 f1; 2; 3g
1

C ij ¼
r e : e e e ;
>
>  n n o
>
> npr 0

X
4 >
> 1 pr
 3gp24  nkð1nÞ r0i r0j ; i ¼ 4 xor j ¼ 4
p2n p >
> pr pe 2g 2 p2e
eðtÞ
E ¼  p1n kð1nÞð2nÞ ðtÞ
þ kð1nÞ  riðtÞ aiðtÞ >
>   n
e
o
> 1 pr
> n
r
i¼1
:
p p
1
g
 nkð1nÞ
g 2 p2
r024 ; i ¼ j ¼ 4
0   2 P  1
r e e

a0
P3 2 Hqðm;tÞ s2
4ðm;tÞ 3
þ a02
X
N B
a 2 i¼1 iðm;tÞ C note: qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

þ N1 BHpðm;tÞ i¼1 iðm;tÞ
þ C: 2 P3 P3 02 
@ 2 3 A  pe ¼ p2 þ kð1nÞq ; p ¼ 1
r
i¼1 i ; q ¼ 3
2r02 4 þ i¼1 ri
m¼1 3g 3 2
 i = j 2 {1, 2, 3} ) C11, C22, C33
 i – j 2 {1, 2, 3} ) C12, C13, C21, C23, C31, C32
Gibbs free energy for n = 1:
 i = 4 xor j = 4 ) C14, C41, C24, C42, C34, C43
 i = j = 4 ) C44
    q2 X
4
pðtÞ pðtÞ
E¼ k
ln pr
 1  6gpðtÞðtÞ  riðtÞ aiðtÞ
i¼1
0  2 P  1
a0 4. Model parameter determination
P3 2 Hqðm;tÞ 2
4ðm;tÞ
þ
3
a02
XN B
a 2 i¼1 iðm;tÞ C
þ N1 BHpðm;tÞ i¼1 iðm;tÞ
þ C;
@ 2 3 A The non-linear KHMCC model has a small number of parame-
m¼1
ters, namely three dimensionless material constant parameters
(g, k and n), one parameter for critical state (M) and a kinematic
where
hardening parameter (a). These parameters are obtained through
the processes of parameter calibration from the experimental re-
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
sults. The dimensionless material constants g, k and n are related
kð1  nÞq2ðtÞ 1X 3
peðtÞ ¼ p2ðtÞ þ ; pðtÞ ¼ riðtÞ ; to elastic stiffness, which can be directly determined from experi-
3g 3 i¼1 mental measurements of small-strain stiffness such as in bender
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3  02 X3   element tests, or determined from empirical equations for
qðtÞ ¼ 2r4ðtÞ þ r02iðtÞ : preliminary analysis, as summarised in Table 1. The critical state
2 i¼1
frictional parameter M is determined at the stage of failure

Table 1
Small-strain stiffness empirical equation G = AF(e)pn (after [39]).

Soil type A F(e) N Void ratio e Test method References

Reconstituted NC kaolin 3270 ð2:973eÞ2 0.5 0.5–1.5 Resonant column Hardin and Black [9]
1þe
Several 3270 ð2:973eÞ2 0.5 0.5–1.7 Resonant column Hardin and Black [9]
1þe
undisturbed NC clays
Several undisturbed silts and clays 1726 ð2:973eÞ2 0.46–0.61 0.4–1.1 Resonant column Kim and Novak [20]
1þe
Undisturbed 90 ð7:32eÞ2 0.6 1.7–3.8 Cyclic triaxial Kokusho et al. [22]
1þe
NC clays
Undisturbed Italian clays 4400–8100 ex 0.4–0.58 0.6–1.8 Resonant column and bender element Jamiolkowsky et al. [13]
ðx ¼ 1:11  1:43Þ
 D. Apriadi et al. / Computers and Geotechnics 49 (2013) 100–110 105

Fig. 2. Determination of initial stiffness G from specified stress–strain curve.

Fig. 5. Prediction of tangent stiffness Gt against deviatoric stress q of the 3-SKH

model compared to experimental results (after 34).

Table 2
Summary of 3-SKH model parameters of Speswhite kaolin (after 34).

M k⁄ ecs j⁄ G (kPa) T S w
0.89 0.073 1.994 0.005 1964p0:65 R0:2 0.25 0.08 2.5
o

Fig. 3. Determination of parameter a using steps (1)–(3).

Fig. 6. Series of stress points with different stress paths for overconsolidated clay.

G ¼ gpn pr1n ; ð22Þ

n
K ¼ kp p1n
r : ð23Þ
Apriadi et al. [3] showed that Eq. (22), which is derived from the
constitutive model as shown in Fig. 2, is identical to the experi-
mental observation from bender element tests [36,21,27,37]:
Fig. 4. Determination of parameter a using a small parametric study.
G ¼ Cpn ; ð24Þ
where C and n are constants. This equation can be normalised into
conditions. It can be approximated by the relationship in the fol- the form suggested by Houlsby et al. [11], shown in Equation (25),
lowing equation: to obtain the dimensionless material constants g and n. The dimen-
sionless material constant k can be determined using the elastic
6 sin / relationship as presented in Eq. (26). It is assumed that during iso-
M¼ : ð21Þ tropic consolidation, the ratio of K/G is constant regardless of con-
3  sin /
solidation pressure. The value of Poisson’s ratio m is made to
It can be shown that the initial stiffness G and K, after completion of range in a certain limit and can be related to the coefficient of earth
isotropic consolidation, are given by: pressure at rest K0, as shown in Eq. (27).
106 D. Apriadi et al. / Computers and Geotechnics 49 (2013) 100–110

(0 6 n 6 1) in modelling of the non-linear responses of materials;

therefore, the accuracy on the initial stiffness is improved from
the previous version by dropping the limited assumption on spec-
ifying n = 1.
According to Houlsby and Puzrin [10], the kinematic hardening
parameter a is calibrated to fit the stress–strain curve for specific
test data (such as triaxial undrained or drained tests), as shown
in Fig. 3. This parameter characterises the first loading curve (also
referred to by many as backbone curve) of the stress–strain re-
sponse with a hyperbolic curve. The following is the general
step-by-step procedure for determining parameter a:

(1) Normalise undrained or drained stress at the vertical axis by

undrained or drained strength c.
(2) Multiply the strain response at the horizontal axis by the
appropriate initial stiffness and divide by c.
(3) Find a as the inversion slope (1/secant stiffness) of a point at
Fig. 7. Parametric study of kinematic hardening parameter a. 50% of the normalised stress.

However, steps (1)–(3) are applicable if the laboratory stress–

Table 3
strain curve can be approached with a simple hyperbolic stress–
Summary of non-linear KHMCC model parameters of Speswhite kaolin.
strain curve. In many practical scenarios, kinematic hardening
k 1012 Dimensionless material constants parameter a can be directly obtained from a small parametric
g 467
study to match the stress–strain curve of the specific test, as shown
n 0.65
M 0.89 Slope of critical state line in the q–p plane
in Fig. 4.
a 15 Kinematic hardening parameter

5. Model validation

Atkinson et al. [5], Jardine [18,15], Smith et al. [23], Stallebrass

and Taylor [34] and Houlsby [12] observed an additional influence
on the stress–strain behaviour concerning the loading path depen-
dence or immediate stress history of soil, described by the most re-
cent loading. This may take the form of an extended period of rest
or a sudden change in the direction of the stress path. They found
that zones exist at small strains, which can change in both shape
and size as the soil is subjected to different loading histories.
Stallebrass [35] and Stallebrass and Taylor [34] have conducted
an experimental undrained triaxial test on Speswhite kaolin and
simulated the results using the Three-Surface Kinematic Hardening
(3-SKH) model [4]. They plotted the results with tangent shear
stiffness Gt against deviatoric stress q, as shown in Fig. 5. Table 2
shows the 3-SKH soil model parameter in this simulation.
The experiment was simulated with a given series of stress
points (A, B, C, D and E), as shown in Fig. 6. Four cases of consoli-
dated stress paths are given, i.e., ACB, ACAB, ACBDB and ACBEB,
which represent a sudden change in the direction of the stress path
Fig. 8. Position of yield surfaces at initial conditions. before the undrained shearing stage at point B, which has an Ro,
OCR in terms of isotropic pressure, of 2.4 (p0 = 300 kPa and
 n pm = 720 kPa). Also shown in Fig. 6 are all stress histories located
G p
¼g ; ð25Þ within a bounding yield surface. Hereafter, a symbol p labelled to
pr pr
figures and equations represents an effective mean stress.
K k 2ð1 þ mÞ In this simulation, the small-strain stiffness parameters g, k and
¼ ¼ ; ð26Þ
G g 3ð1  2mÞ n are obtained using stiffness relationships normalised to Equation
K0 (25) with 1 atm (100 kPa) as a reference pressure. Eq. (28) shows
m¼ : ð27Þ
1 þ K0 the small-strain stiffness relationship used in this simulation.
As emphasised in Table 1, the non-linear initial stiffness with a  0:65
G p
power pressure dependency of stiffness using material dimension- ¼ 467 : ð28Þ
pr pr
less parameter of n fits the experimental data better than that of a
linear pressure dependency. Therefore, the value of n shown in Eq. The constant k can be determined using Eq. (26), where Pois-
(25) is an input parameter determined from bender element test, son’s ratio m is 0.3. The kinematic hardening parameter a is deter-
resonant column test or cyclic triaxial test, not a fitting parameter mined from a small parametric study of the actual stress–strain
to particular loading conditions and stress histories as illustrated in curve, as shown in Fig. 7. Table 3 summarises the non-linear
this study. According to Apriadi [2], comparisons on the initial stiff- KHMCC model in the simulation. Figs. 8 and 9 show the fields of
ness with a variation of n convinces that linear dependent stiffness yield surfaces at initial conditions and after the four different stress
(n = 1) is essentially inferior to power pressure dependent stiffness paths, respectively. The overlapping of yield surfaces observed in
 D. Apriadi et al. / Computers and Geotechnics 49 (2013) 100–110 107

Fig. 9. Position of yield surfaces after four different stress histories.

Fig. 10. Prediction of tangent stiffness Gt against deviatoric stress q of the non- Fig. 11. Simulation of stress paths of the non-linear KHMCC model after different
linear KHMCC model compared to experimental results. loading histories.

Fig. 9 is permissible. The possibility of overlapping of yield surfaces

has already been discussed based on the laws of thermodynamics paths activate more yield surfaces than do other paths; therefore,
[29]. discretised responses are observed, but can be reduced by increas-
Fig. 10 shows the simulated results reflecting the plot of norma- ing the number of yield surfaces. Though measured data are not
lised tangent shear stiffness Gt against the deviatoric stress. This available, Fig. 11 aims to represent simulations of the behaviours
plot clearly shows that small-strain stiffness is affected by loading of undrained effective stress paths due to stress path rotations
path history. Prediction of the model is also in good agreement after compression and after swelling. The figure shows the result-
with the experimental result. However, slightly non-smooth re- ing four undrained stress paths from the non-linear KHMCC model.
sponses appear in the curve corresponding to the paths after AC- It is obvious that the non-linear KHMCC model can predict loading
BEB as a result of numerical discretisation. Complicated stress path dependence during undrained shear.
108 D. Apriadi et al. / Computers and Geotechnics 49 (2013) 100–110

that stiffness reduction for larger stress changes is in good agree-

ment with the experimental results. The simulations used 100
yield surfaces, except for the unloading–reloading response, which
was carried out using 50 yield surfaces. This simple computation
was run using Intel Pentium M 1.6 GHz.

6. Effect of number of yield surfaces

This additional numerical study aims to determine the effect of

the number of yield surfaces on the response of different stress
paths and tries to find the optimum number of yield surfaces to ob-
tain accurate results. The numerical study is performed using 1, 2,
10 and 100 yield surfaces, and the effect is shown in stress–strain
and stress path responses in Figs. 13 and 14. The figures clearly
show that the stress–strain and stress path responses are highly
influenced by the number of yield surfaces. It can be concluded
that 20 yield surfaces are adequate to obtain accurate results.
Fig. 12. Prediction of unloading–reloading response at small strain levels of the
non-linear KHMCC model compared to experimental results.
7. Conclusions

Finally, the model was validated to simulate un/reloading A numerical implementation of non-linear pressure-dependent
behaviour at small strains compared to the experimental test data, stiffness in the hyperplasticity model under general stress condi-
as shown in Fig. 12. It is noted that the calibration of parameter a in tions using a strain-driven forward-Euler integration scheme was
Fig. 7 is carried out only for the first loading path, but the valida- presented. Further model verification with a small-strain experi-
tion in Fig. 12 also shows un/reloading path forming a hysteresis ment on overconsolidated clay soil was conducted. Verification
loop, which is close to the experimental result. clearly shows that the multisurface hyperplasticity model with a
The model prediction shows that it can model a non-recover- small number of parameters can accurately simulate the key
able strain during an unloading–reloading response. It also shows features of small-strain behaviour using a unified framework. The

Fig. 13. Effect of number of yield surfaces on the stress–strain response.

 D. Apriadi et al. / Computers and Geotechnics 49 (2013) 100–110 109

Fig. 14. Effect of number of yield surfaces on the stress path response.

unloading–reloading response in the model prediction shows a [3] Apriadi D, Likitlersuang S, Pipatpongsa T, Ohta H. On the numerical
implementation of hyperplasticity non-linear kinematic hardening modified
non-recoverable strain and loading path dependence, which are
cam clay model. IES J Part A: Civil Struct Eng 2009;2(3):187–201.
in good agreement with the experimental observation. [4] Atkinson JH, Stallebrass SE. A model for recent stress history and non-linearity
In summary, this study achieved its two main purposes. First, it in the stress–strain behaviour of overconsolidated soil. In: Proc. 7th int. conf.
showed that using the multiple-yield surface could explain the ef- on computer methods and advances in geomechanics, Cairns 1; 1991. p. 555–
60.
fect of stress history (i.e., the different stress paths give the differ- [5] Atkinson JH, Richardson D, Stallebrass SE. Effect of recent stress history on
ent non-linear stiffness responses). The more yield surfaces are stiffness of overconsolidated soil. Geotechnique 1990;40:531–40.
used, the smoother the transition of stiffness that is provided. Sec- [6] Griffiths DV. Finite element analyses of walls, footings and slopes. Ph.D. thesis,
University of Manchester; 1980.
ond, it provides the actual non-linear pressure-dependent stiffness [7] Gunn MJ. The prediction of surface settlement profiles due to tunnelling. In
even inside the first yield surface. This is because when the stress predictive soil mechanics. In: Houlsby GT, Schofield AN, editors. Proc. writh.
state is located inside the first yield surface, the multiple-yield-sur- memorial symp.. Oxford, London: Thomas Telford; 1993. p. 304–16.
[8] Hardin BO. The nature of stress–strain behaviour for soils. Earthquake
face model still offers a purely elastic response. These develop- engineering and soil dynamics. In: Proceedings of the ASCE geotechnical
ments could benefit cases of small strain levels, especially during engineering division specialty conference, Pasadena; 1978. p. 3–90.
cyclic loading such as traffic load, small vibrations or a small [9] Hardin BO, Black WL. Vibration modulus of normally consolidated clays. J Soil
Mech Found, ASCE 1968;94(SM2):353–69.
earthquake. [10] Houlsby GT, Puzrin AM. Principles of hyperplasticity: an approach to plasticity
theory based on thermodynamics principles. London: Springer-Verlag London
Ltd.; 2006. ISBN-10: 184628239X, ISBN-13: 9781846282393.
Acknowledgements [11] Houlsby GT, Amorosi A, Rojas E. Elastic moduli of soils dependent on pressure:
a hyperelastic formulation. Géotechnique 2005;55(5):383–92.
[12] Houlsby GT. A model for the variable stiffness of undrained clay. In: Proc. int.
This work was supported by AUN/SEED-Net JICA. The authors symp. on pre-failure deformation of soils, Torino, 26–29 September, Balkema,
would like to express their sincere appreciation for the grant. vol. 1; 1999. p. 443–50 [ISBN: 9058090760].
[13] Houlsby GT, Wroth CP. The variation of the shear modulus of a clay with
pressure and overconsolidation ratio. Soils Found 1991;31(3):138–43.
[14] Jamiolkowski, M, Lancellotta, R, Lo Presti, DCF. Remarks on the stiffness at
References small strains of six Italian clays. Shibuya S, Mitachi T, Miura S, editors. vol. 2:
A.A. Balkema; 1994. p. 817–836.
[1] Addenbrooke TI, Potts DM, Puzrin AM. The influence of pre-failure soil stiffness [15] Jardine RJ. Some observations on the kinematic nature of soil stiffness. Soils
on the numerical analysis of tunnel construction. Geotechnique Found Jpn Soc Soil Mech Fdn Eng 1992;32(2):111–24.
1997;47(3):693–712. [16] Jardine RJ, St. John HD, Hight DW, Potts DM. Some practical applications of a
[2] Apriadi D. Development and numerical implementation of thermodynamics- non-linear ground model. Proc. 10th Eur. conf. on soil mech. and found. eng.
based soil model. Ph.D. thesis, Chulalongkorn University; 2009. florence 1991;1:223–8.
110 D. Apriadi et al. / Computers and Geotechnics 49 (2013) 100–110

[17] Jardine RJ, Potts DM, Fourie AB, Burland JB. Studies of the influence of non- [31] Roscoe KH, Burland JB. On the generalised stress–strain behaviour of wet clay.
linear stress–strain characteristic in soil–structure interaction. Géotechnique Engineering plasticity. Cambridge University Press; 1968. p. 535–610.
1986;36(2):377–96. [32] Smith PR, Jardine RJ, Hight DW. The yielding of Bothkennar clay. Géotechnique
[18] Jardine RJ. Investigations of pile-soil behaviour with special reference to the 1992;42(2):257–74.
foundations offshore structures. Ph.D. thesis, University of London; 1985. [33] Soga K, Nakagawa K, Mitchell JK. Measurement of stiffness degradation
[19] Jardine RJ, Symes MJ, Burland JB. The measurement of soil stiffness in triaxial characteristics of clays using a torsional shear device. In: First international
apparatus. Géotechnique 1984;34(3):323–40. conference on earthquake geotechnical engineering, Tokyo, November 14–16;
[20] Kim TC, Novak M. Dynamic properties of some cohesive soils of Ontario. 1995. p. 107–12.
Canadian Geotech. J. 1981;18(3):371–89. [34] Stallebrass SE, Taylor RN. The development and evaluation of a constitutive
[21] Kohata Y, Tatsuoka F, Wang, Lo, Jiang GL, Hoques E, et al. Modelling the model for the prediction of ground movements in overconsolidated clay.
non-linear deformation properties of stiff geomaterials. Géotechnique 1997; Géotechnique 1997;47(2):235–54.
47(3):563–80. [35] Stallebrass SE. The effect of recent stress history on the deformation of
[22] Kokusho T, Yoshida Y, Esashi Y. Dynamic properties of soft clay for wide strain overconsolidated soils. Ph.D. thesis, City University; 1990.
range. Soils Found. 1982;22(4):1–18. [36] Tanizawa F, Techavorasinskun S, Yamaguchi J, Sueoka T, Goto S. Measurement
[23] Likitlersuang S, Houlsby GT. Predictions of a continuous hyperplasticity model of shear wave velocity of sand before liquefaction and during cyclic mobility.
for Bangkok clay. Geomech Geoeng 2007;2(3):147–57. In: Shibuya S, Mitachi T, Miura S, editors. Proceedings of the international
[24] Likitlersuang S, Houlsby GT. Development of hyperplasticity model for soil symposium on pre-failure deformation characteristics of geomaterials, vol.
mechanics. Int J Numer Anal Method Geomech 2006;30:229–54. 1. Rotterdam: Balkema; 1994. p. 63–8.
[25] Likitlersuang S. A hyperplasticity model for clay behaviour: an application to [37] Techavorasinskun S, Amornwithayalax T. Elastic shear modulus of Bangkok
Bangkok clay. D.Phil. thesis, Oxford University; 2003. clay during undrained triaxial compression. Géotechnique 2002;52(7):
[26] Mroz Z, Norris VA. Elastoplastic and viscoplastic constitutive models for soils 537–40.
with application to cyclic loading. In: Pande GN, Zienkiewicz OC, editors. Soil [38] Viggiani G. Small-strain stiffness of fine grained soils. Ph.D. thesis, City
mech. cyclic transient loads. NY: John Wiley & Sons; 1982. p. 173–218 [chapter 8]. University, London; 1992.
[27] Pennington DS, Nash DFT, Lings ML. Anisotropy of G0 shear stiffness in Gault [39] Yimsiri S, Soga K. Anisotropy of highly-overconsolidated clay in small-and
clay. Géotechnique 1997;47(3):391–8. intermediate-strain levels. In: Proc. of the 14th Southeast Asian geotechnical
[28] Potts DM, Ganendra D. An evaluation of substepping and implicit stress point conference, 14th SEAGC, Hong Kong; 2001.
algorithm. Comput Methods Appl Mech Eng 1994;119:341–54. [40] Zienkiewicz OC, Cormeau IC. Viscoplasticity, plasticity and creep in elastic
[29] Puzrin AM, Houlsby GT. On the non-intersection dilemma in multi-surface solids – a unified numerical solution approach. Int J Numer Methods Eng
plasticity. Geotechnique 2001;51(4):369–72. 1974;8:821–45.
[30] Rampello S, Silvestri F, Viggiani G. The dependence of G0 on stress state and
history in cohesive soils. In: Proc. 1st int. conf. on pre-failure deformation
characteristics of geomaterials, Sapporo 2; 1994. p. 1155–60.