Unsaturated behaviour of soils: a refined theoretical model with

complex volumetric mechanisms

J.-M. Pereira & H. Wong

ENTPE, Rue Maurice Audin, Vaulx-en-Velin Cedex, France

Ph. Dubujet
ENISE, Jean Parot, Saint-Etienne Cedex, France

ABSTRACT: This paper proposes a general formulation which allows to adapt existing constitutive models
developed for dry or saturated soils to take into account suction effects. The main advantages of such a conceptu-
alisation is to reuse reliable models presenting good capacities to closely simulate soil behaviour under complex
loading paths. These loading paths may lead to sophisticated behaviour (contractancy followed by dilatancy in
the course of loading etc.) well simulated by such models. The choice of a particular effective stress, combined
with suction, is made. The definition of this effective stress is based on the formulation of an equivalent pore
pressure. Following the approach presented, an existing elastoplastic model (CJS model), is extended to cover
unsaturated states. A classic geotechnical in situ investigation method (pressuremeter) is then simulated. This
simulation allows to determine the effects of suction on the mechanical behavior of the soil.

1 INTRODUCTION 2 GENERAL FORMULATION

In front of an increasing number of problems to which Confronted to the diversity of approaches, an attempt is
engineers are confronted involving unsaturated soils, made to facilitate the adaptation of existing models to
a need of accurate models taking into account partial unsaturated states via a general formulation proposed
saturation of soils appears. These models have to sim- in this section. Only isotropic behavior is considered.
ulate soil responses under complex loading paths. In This formulation can include most of recent unsat-
particular and concerning volumetric behaviour, some urated model proposals. Interested readers will find
soils are characterized by complex mechanisms like more details in (Pereira et al. 2003).
contractancy-dilatancy for instance.
Several propositions have been made to obtain
a realistic description of unsaturated soils behavior. 2.1 Assumptions and definitions
They constitute adaptations to unsaturated states of Some general assumptions are now presented. Suc-
existing reliable elastoplastic models. However, many tion s is defined as the difference between pore gas
of these propositions are based on simple models (like and water pressures: s = pg pl . The existence of an
Modified Cam Clay for instance) and are not able to air-entry suction se is accounted for: it represents the
predict complex behaviors. suction value below which the saturation degree Sl
In this paper, a particular effective stress and suc- remains equal to unity during a drying test.
tion are chosen as state variables. As it will be seen, The following two proposals are inspired by
this option simplifies considerably existing behav- Coussy & Danglas developments in (Coussy and
ior models extension to unsaturated domain. In this Dangla 2002): (i) an equivalent pore fluid pressure
framework, a general isotropic model formulation is is introduced such that, under saturated conditions,
firstly presented. It can be seen as a methodology to it would induce an equivalent behavior as that of
adapt most of existing behaviour models to unsaturated the unsaturated soil at the present state; (ii) the ten-
states. In a second part, this formulation is the basis sor = + 1 plays the role of an effective stress
of a complex behavior model (CJS model) extension according to the current meaning of soil mechan-
to unsaturated states. Finally, numerical simulations ics. Furthermore, the Terzaghis effective stress is
of pressuremeter tests in an unsaturated soil using the assumed to remain valid for suctions below air-entry
proposed model are presented. value. Note that the equivalent pore pressure is

473
Copyright 2005 Taylor & Francis Group plc, London, UK
counted positively for positive suctions contrary to the Functions l1 and l2 allow to take into account the
proposal described in (Coussy and Dangla 2002). fact that a suction increase induces an expansion of
the elastic domain (Eq. 5). In order to stay com-
patible with the validity of the saturated effective
2.2 Generalized isotropic model stress in the 0 s se domain, the elastic boundary
p must remain constant equal to p 0 thus imposing the
2.2.1 Pore pressure and effective stress
following conditions:
According to the assumptions developed above, the
isotropic effective stress is defined by:

Physically, the equivalent pore pressure must repro- Equations 6 and 7 ensure by the way the continuity of
duce an increase of effective stress during a suction the yield surface at air entry suction.
increase (ie (s) is an increasing function of s). The
continuity of at the saturation limit (s = se ) is also 2.2.4 Flow rule
supposed. By considering that the effective stress in unsaturated
In the saturated domain (0 s se ), the validity of states plays an equivalent role to that of saturated
Terzaghis effective stress, uncontested, imposes the states, the flow rule must necessary be written as:
following condition:

2.2.2 Elastic behavior p

where v is the volumetric part of plastic strain tensor
A general form for the elastic part of strains can be increment and g is the plastic potential. The plas-
defined in an incremental way as (where K may depend tic multiplier d is calculated from the consistency
on the stress/strain state): condition: f = 0 and f = 0 under plastic loading.

2.2.5 Hardening law

The hardening law can formally be written in the
following way:
2.2.3 Yield surface
A suction increase (resp. decrease) induces an increase
(resp. decrease) of the effective stress. However, in
order to model the elastic behavior for a suction The definition of the hardening function H depends on
increase above the air-entry value and the collapse phe- the model from which the adaptation as to be carried
nomenon during wetting, the preconsolidation pres- out. It is worthy to note that the hardening process is
sure p must vary more rapidly than pg + when held by the preconsolidation pressure at zero suction
suction increases. In other words, the expansion of p 0 thus simplifying the transition between saturated
the elastic domain must be faster than the increase and unsaturated conditions.
of effective stress. Use of effective stress alone cannot The hardening modulus H can therefore be evalu-
describe these behaviors. This requires that the yield ated. By definition, we have:
surface f also depends on suction as well as effective
stress (see (Loret and Khalili 2000) and (Dangla 2002)
for more details).
The following expression is therefore proposed for
the yield surface:
Use of yield surface definition (Eqs. 4 and 5) as well
as the flow rule (Eq. 9) leads to identify the hardening
modulus:

where p 0 and p are the mean effective preconsolida- In addition to the role of functions l1 and l2 on the
tion stresses respectively at saturation and at a given elastic domain range, Equation 11 shows a more par-
suction (hence a given pore pressure ). The formu- ticular role of l1 : it allows to couple suction effects
lation of p here given is inspired from (Loret and to mechanical hardening thus affecting the expansion
Khalili 2000). rate of the yield surface.

474
Copyright 2005 Taylor & Francis Group plc, London, UK
 This generalized isotropic model sets the bases of As previously mentioned, once the equivalent pore
what can be regarded as a methodology allowing to pressure is defined, the effective stress is given by the
adapt most of the commonly used models of soil relation:
mechanics to unsaturated states. The various func-
tions introduced above only have to be identified in
each case.
The capabilities of this general formulation are now As shown in the methodology presentation, the mod-
illustrated by the adaptation of a complex elastoplastic ification of the model only concerns yield surfaces
model to unsaturated states. definition. The other relations remain unchanged
except that the modified effective stress (as defined
by Eqs. 15, 12 and 13) replaces the classic one.
3 ADAPTATION OF CJS MODEL
3.2.2 Elastic mechanism
3.1 Original CJS model The elastic mechanism is nonlinear. Elastic strains
The purpose of this paper is not to expose extensively are incrementally defined by the hypoelastic-type
the CJS model; interested reader should find more relation:
information in (Cambou and Jafari 1988), (Elamrani
1992) and (Maleki et al. 2000).
The CJS (Cambou-Jaffari-Sidoroff) model was
originally developed for saturated granular materials
(sands and silts). Its formulation includes three strain where I1 is the first invariant of stress tensor and s
mechanisms: one non linear elastic and two plastic is the stress deviator tensor, K e = K0e [I1 /(3pa )]n and
mechanisms. These two plastic mechanisms concern G e = 3K e (1 2)/[2(1 + )] is deduced from Poisson
respectively isotropic and deviatoric loadings and are coefficient .
submitted to strain-hardening (isotropic in the case of Parameters associated to the elastic mechanism are:
the isotropic mechanism and both isotropic and kine- K0e , and n. pa is the atmospheric pressure and is taken
matic in the case of the deviatoric one). In the sake of as a reference datum.
simplicity, kinematic hardening associated to the devi-
atoric plastic mechanism is disabled thus leading to a 3.2.3 Isotropic plastic mechanism
simplified version of CJS model. However, this ver- Concerning the isotropic mechanism, the yield surface
sion is sufficient to illustrate the proposed adaptation is modified in the following way (same modification
methodology. as in the general formulation):

3.2 Adaptation to unsaturated soils

3.2.1 State variables
According to the chosen framework, state variables where Q0 is the maximum mean (saturated) effective
used in the model are a particular effective stress and stress to which the material has been submitted during
suction. The equivalent pore pressure needed for the its history. Q0 plays the role of the hardening parame-
effective stress definition is based on the work of ter (see Eq. 22). This yield surface represents a plane
Dangla & Coussy founded on an energetic approach orthogonal to (O I1 ) axis in the effective stress space
(Coussy and Dangla 2002): (see Figure 1).
l1 (s) and l2 (s) definitions are inspired from Loret
and Khalilis work (Loret and Khalili 2000) and (Loret
and Khalili 2002):

As a first approximation, the sorption curve definition

linking suction and saturation degree is supposed to be
a objective function. The empirical formula proposed
by Brooks and Corey will be used:
An associated model is considered so that the isotropic
plastic potential associated to the isotropic mechanism
is identical to the yield surface:

where 1 is a material parameter.

475
Copyright 2005 Taylor & Francis Group plc, London, UK
Thus, the strain increment associated to the isotropic axis from (O I1 ) axis (see Figure 1). In this version
plastic mechanism is classically defined by: of CJS model, X remains constant (equal to null ten-
sor so that q = s at any time) whereas in the complete
elastoplastic model, it is a hardening variable introduc-
ing kinematic hardening associated to the deviatoric
plastic mechanism.
where dI is the plastic multiplier associated to the Function l3 (s) in Equation 23 is defined in the same
isotropic plastic mechanism. way as l2 (s) that is:
The hardening law associated with the isotropic
mechanism is isotropic and defined by:

It should be noted that this modification leads to a

translation of the yield surface in the direction of ten-
sile stresses. This choice is founded on experimental
p observations (suction does not influence internal fric-
Parameters associated to isotropic mechanism are: K0
and n. Those coming from the model extension to tion angle) tending to show that the cone angle of
unsaturated states are: k1 and k2 . the yield surface must remain constant while suctions
varies.
3.2.4 Deviatoric plastic mechanism Figure 1 represents both isotropic and deviatoric
In order to introduce the possibility to extend (under yield surfaces in 3-dimensional effective stresses
suction increases) the elastic domain delimited by the space. The surface denoted f R corresponds to the rup-
deviatoric yield surface, the deviatoric yield surface is ture surface: it delimits the admissible stress states and
modified in the following way: allows to account for rupture states that may occur
under certain stress paths. Actually, this surface is
implicitly implemented in CJS model via the laws con-
trolling evolution of the hardening variables (R only in
where the version used here) which tends asymptotically to
a limit value.
Figure 2 represents yield surfaces of the adapted
model in (I1 /3, qII h(q ), s) space.
A plastic potential is introduced in order to obtain
a better description of the volumetric strains by means
of the characteristic state concept (allowing dilatant or
q is Lodes angle of q tensor, qII is the second invariant contractant behaviour upon the stress state).
of q tensor and m = 6. This deviatoric plastic potential is implicitly via a
Scalar R represents the mean solid angle of the kinematic condition on the volumetric plastic strain
cone formed by the deviatoric yield surface in the increment induced by the deviatoric mechanism dD :
p
stress space and plays the role of a hardening vari-
able. Tensor X represents the deviation of the cone

fR qII

1 I 1 h (q) fD
s fI

fD

se
fI

3

2
Q0 Q (s) I 1/3

Figure 1. CJS model yield surfaces: 3-dimensional repre- Figure 2. Extended CJS model yield surfaces: representa-
sentation in effective stresses space (s = 0 kPa). tion in (I1 /3, qII /h(q ), s) space at given q .

476
Copyright 2005 Taylor & Francis Group plc, London, UK
This results in a plastic potential defined by its 4 NUMERICAL SIMULATIONS
gradient:
Some results concerning non-homogeneous paths are
now presented. They concern the pressuremeter test
which is a classic geotechnical in situ investigation
method. A computation code using the finite element
method has been developed in our laboratory. It works
with under plane strains and axial symmetry assumptions
so that all quantities only depend upon the radial
space variable. In what follows, the pore gas pressure
pg has been assumed to be in equilibrium with the
atmospheric pressure so that pg = 0.
Pressuremeter tests have been simulated at differ-
ent suctions (0, 20, 100, 200, 300, 400 and 500 kPa)
under undrained condition (corresponding to rapid
loadings). This condition must be interpreted as a
constant specific water volume condition, this latter
having been introduced by Wheeler (Wheeler 1996)
as the volume of water and solids in a volume of soil
p
containing unit volume of solids. Under this consid-
where eD is the plastic strain deviator associated to the eration, a relation appears linking volume changes to
deviatoric plastic mechanism and s is Lodes angle of liquid saturation degree variations:
stress deviator s.
The strain increment associated to the deviatoric
plastic mechanism is then classically defined by:

thus allowing to compute equivalent pore pressure

variations induced by displacement increments (see
(Pereira, Dubujet, and Wong 2003) for more details).
where dD is the plastic multiplier associated to the Materials constants used in these simulations (see
deviatoric plastic mechanism. Table 1) have been identified from triaxial compres-
Only the isotropic hardening is conserved in this sion experimental data (Cui 1993) on an unsaturated
version of the CJS model. It is defined by the following compacted silt from Jossigny, France (this validation is
relation giving the variations of R: presented in a paper submitted for publication). For all
tests, the initial total stress is the same (100 kPa) but
initial void ratios are computed taking into account
suction increases (under isotropic conditions) from

li functions are submitted to the following condi-

Table 1. CJS material constants.
tions in the saturated domain:
Material
constant Unit Value

K0e MPa 8.0

0.125
0.8
thus ensuring continuity of description between satu- A kPa1 0.27
rated and unsaturated states. B kPa1 1.0
Parameters associated to the deviatoric plastic Rc 0.26
mechanism are: , , Rc , A and B. k3 comes from the 0 0.2
p
model extension to unsaturated states. K0 MPa 6.7
The extended model thus counts 9 saturated con- n 0.9
stants, 2 constants linked to the sorption curves (se k1 0.18
and ) and 3 constants coming from its extension k2 1.8
to unsaturated states (ki coefficients) so that a total k3 0.0
se kPa 15.0
of 14 material constants have to be identified from 2.1
experimental data.

477
Copyright 2005 Taylor & Francis Group plc, London, UK
a reference void ratio e = 0.7 at zero suction (see lower at higher suctions). It can also be remarked that
Table 2). once the soil becomes unsaturated, the differences are
This procedure can be seen as the simulation of lower between two distinct suctions: an identical suc-
a soil drainage in order to quantify influence of the tion increment has less impact at higher suctions. It can
latter on soil mechanic properties. Results are pre- also be observed that the limit pressure (characteriz-
sented in Figure 3. They show that a suction increase ing material rupture) increases as suction is higher (see
results in a higher soil stiffness (cell displacements are Table 2 and Figure 4). This increase is however reduced
once the soil is unsaturated. It should be noted that the
limit pressure has been arbitrary chosen here as the
Table 2. Initial void ratios and limit pressures.
cell pressure necessary to obtain a volume increase of
Initial suction s Initial void ratio Limit pressure 20% of the cell volume.
(kPa) () (kPa) It is worthy to note that these pressuremeter tests
do not start from the same effective stress (identical
0 0.700 315 initial total stress but different initial suctions). This
20 0.693 405 can largely explain the smaller displacements observed
100 0.687 484 at higher suctions (due to the larger effective stress,
200 0.683 545 hence stiffness).
300 0.680 590
400 0.678 626
500 0.677 658
5 CONCLUSIONS

A general isotropic model formulation for unsaturated

0.2 soils is presented within the framework of two inde-
0.18 pendent state variables, one of which being an effective
stress. This formulation can be regarded as a method-
0.16
ology allowing existing behaviour models adaptation
0.14 to unsaturated states in a straightforward manner.
0.12 This formulation abilities are illustrated by the
V/V0

0.1
extension to unsaturated soils of a complex behaviour
model (CJS model). This model has then been used
0.08
to numerically simulate pressuremeter tests under
0.06 undrained condition and at different initial suctions.
0.04
s = 0 kPa
s = 20 kPa
The results, although not confronted with experimen-
s = 100 kPa
s = 200 kPa tal data, are encouraging and are in good accordance
0.02 s = 300 kPa
s = 400 kPa
s = 500 kPa
with physical intuition.
0
100 200 300 400 500 600 700
Cell pressure (kPa)
REFERENCES
Figure 3. Pressuremeter tests at different suctions.
Alonso, E. E., A. Gens, and A. Josa (1990). A constitutive
700 model for partially saturated soils. Gotechnique 40(3),
405430.
650 Cambou, B. and K. Jafari (1988). Modle de comportement
des sols non cohrents. Revue Franaise de Gotechnique
600 44, 4355.
Coussy, O. and P. Dangla (2002). Approche nergtique du
Limit pressure (kPa)

550
comportement des sols non saturs. In O. Coussy and
500
J.-M. Fleureau (Eds.), Mcanique des sols non saturs.
Paris: Lavoisier.
450 Cui, Y. J. (1993). Etude du comportement dun limon com-
pact non satur et de sa modlisation dans un cadre
400 lasto-plastique. Thse, Ecole Nationale des Ponts et
Chausses.
350 Dangla, P. (2002). Plasticit et hystrsis. In O. Coussy and
J.-M. Fleureau (Eds.), Mcanique des sols non saturs.
300
0 50 100 150 200 250 300 350 400 450 500 Paris: Lavoisier.
Initial suction (kPa) Elamrani, K. (1992). Contributions la validation du modle
CJS pour les matriaux granulaires.Thse, Ecole Centrale
Figure 4. Limit pressure vs suction. de Lyon.

478
Copyright 2005 Taylor & Francis Group plc, London, UK
Loret, B. and N. Khalili (2000). A three-phase model test. In 16th ASCE Engineering Mechanics Conference,
for unsaturated soils. International Journal for Numer- University of Washington, Seattle, USA.
ical and Analytical Methods in Geomechanics 24(11), Pereira, J.-M., H. Wong, and P. Dubujet (2003). A general
893927. framework for constitutive models adaptation to unsatu-
Loret, B. and N. Khalili (2002). An effective stress elastic- rated states. In H. Di Benedetto, T. Doanh, H. Geoffroy,
plastic model for unsaturated porous media. Mechanics of and C. Sauzat (Eds.), Deformation Characteristics of
Materials 34(2), 97116. Geomaterials, IS-Lyon03, Lyon, France, pp. 12691276.
Maleki, M., B. Cambou, and P. Dubujet (2000). Modlisation Balkema.
hirarchise du comportement des sols. Revue Franaise Wheeler, S. J. (1996). Inclusion of specific water volume
de Gnie Civil 4(78), 895928. within an elasto-plastic model for unsaturated soil. Cana-
Pereira, J.-M., P. Dubujet, and H. Wong (2003). Numer- dian Geotechnical Journal 33, 4257.
ical modeling of unsaturated soils in a pressuremeter

479
Copyright 2005 Taylor & Francis Group plc, London, UK