See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/220170686

Numerical simulation of unsaturated soil behaviour

Article  in  International Journal of Computer Applications in Technology · January 2009

DOI: 10.1504/IJCAT.2009.022697 · Source: DBLP

CITATIONS READS

10 717

2 authors, including:

Ayman Abed
Aalto University
25 PUBLICATIONS   73 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

THEBES - THMC Behaviour of the Swelling Clay Barriers View project

All content following this page was uploaded by Ayman Abed on 30 July 2014.

The user has requested enhancement of the downloaded file.

2 Int. J. Computer Applications in Technology, Vol. 34, No. 1, 2009

Numerical simulation of unsaturated soil behaviour

Ayman A. Abed* and Pieter A. Vermeer

Institute of Geotechnical Engineering,
University of Stuttgart,
Pfaffenwaldring 35,70569 Stuttgart, Germany
E-mail: ayman.abed@igs.uni-stuttgart.de
E-mail: pieter.vermeer@igs.uni-stuttgart.de
*Corresponding author

Abstract: The mechanical behaviour of unsaturated soils is one of the challenging topics in
the field of geotechnical engineering. The use of finite element techniques is considered to be a
promising method to solve settlement and heave problems, which are associated with unsaturated
soil. Nevertheless, the success of the numerical analysis is strongly dependent on the constitutive
model being used. The well-known Barcelona Basic Model is considered to be a robust and
suitable model for unsaturated soils and has, thus, been implemented into the PLAXIS finite
element code (Vermeer and Brinkgreve, 1995). This paper provides the results of numerical
analyses of a shallow foundation resting on an unsaturated soil using the implemented model.
Special attention is given to the effect of suction variation on soil behaviour.

Keywords: unsaturated soil; constitutive modelling; finite element method; shallow foundation.

Reference to this paper should be made as follows: Abed, A.A. and Vermeer, P.A. (2009)
‘Numerical simulation of unsaturated soil behaviour’, Int. J. Computer Applications in
Technology, Vol. 34, No. 1, pp.2–12.

Biographical notes: Ayman A. Abed graduated in Civil Engineering from Al-Baath University
in Syria where he also obtained his Master’s Degree in Geotechnical Engineering in the field
of Unsaturated Soil. Since January 2004, he has been pursuing his PhD in Stuttgart University,
Germany under Professor Pieter A. Vermeer’s supervision. He is currently working on the
constitutive modelling of unsaturated soil, with special focus on the mechanical behaviour of
swelling soft clays.

Pieter A. Vermeer graduated in Civil Engineering from Delft University of Technology in the
Netherlands, where he obtained his PhD in 1980. He taught Geotechnical Engineering in Delft
from 1980 to 1994 with research projects in computational geomechanics, dam construction and
deep excavations. In 1994, he moved to Stuttgart, Germany, to become the Head of the Stuttgart
Institute of Geotechnical Engineering. Since the early 1970s, he has been involved in research
and consulting projects ranging from the foundation of extremely large coastal structures to the
manufacturing of very small industrial powder compacts.

1 Introduction
Unsaturated soil is characterised by the existence of three
different phases, namely, the solid phase, the liquid
phase and the gas phase. An important consequence is the
occurrence of surface tension at the air-water interface
within the pores. These forces increase with continuous
drying of the soil and, vice-versa, surface tension will be
reduced upon wetting of the soil. Surface tension causes
a difference between the air pressure ua and the pore
pressure uw. The difference s = ua – uw is named the matrix
suction (Fredlund and Rahardjo, 1993) In this paper, matrix
suction will simply be referred to as suction.
The relation between suction and water content of a
particular soil is named the soil water characteristic curve.
The water content of a soil sample is defined as its (mass)
amount of pore water with respect to (mass) amount of
solid material. Figure 1 shows typical characteristic curves
for two different soils, namely clayey silt and fine sand.
Such curves play a key role in unsaturated ground water
flow calculations and unsaturated soil deformation analyses.
It can be seen from Figure 1 that suction plays a more
important role in the case of fine-grained soil than in
the case of coarse-grained sand. Indeed, at the same water
content, clay or silt exhibits much more suction than sand.

Copyright © 2009 Inderscience Enterprises Ltd.

 Numerical simulation of unsaturated soil behaviour
For that reason, one may expect more suction related
problems for foundations on clay than on sand.
Figure 1 The soil water characteristic curves for clayey silt and
fine sand
Soil shrinkage is a well recognised problem which
is associated with suction increase. Similarly, soil
swelling as caused by suction decrease is a main problem in
foundation engineering. A decrease of suction due to
wetting is especially dangerous in some very loose soils,
where suction provides the stability of soil particles.
As soon as this suction is reduced the soil fabric may
become unstable and cause the so-called soil fabric collapse.
Both soil shrinkage and soil swelling affect foundations
if no special measures are taken during the design process.
The damage reparation cost level could reach high
numbers e.g., as much as $9 billion per year in the USA
only (Nelson and Miller, 1992).
Many empirical procedures have been proposed in the past
to predict the volumetric changes due to suction variations,
but during the last 15 years, research attention has shifted
to more theoretical models. In combination with the FE
method, robust constitutive models give the designer a tool
to understand the mechanical behaviour of unsaturated soils
and reach better design criteria.
2 Unsaturated soil modelling
While surveying the literature one can classify the
modelling methods into empirical and theoretical
approaches.
2.1 Empirical methods
Empirical methods are mostly based on data from
one-dimensional compression tests with zero lateral strain.
These tests give clear information only about the sample’s
initial conditions and final conditions, but no information
about the suction variation during the saturation process.
A nice review and evaluation of these methods can be
found, for example, in the paper by Djedid et al. (2001).
It is believed that such empirical correlations only give
satisfactory results as long as they are applied to the same
soils which are used to derive them. This reduces their use
to a very narrow group of soils.
3
2.2 Theoretical methods
This category uses the principles of soil mechanics, together
with sophisticated experimental data, for the formulation of
a constitutive stress-strain law. An early attempt was made
by Bishop (1959). He extended the well-known effective
stress principle for fully saturated soils to unsaturated soils.
Bishop proposed the effective stress measure
V c V  ua  F ˜ (ua  uw ) (1)
where
V: total stress
ua: pore air pressure
uw: pore water pressure
F: factor related to degree of saturation.
where F = 0 for dry soil and F = 1 for saturated soil.
According to Bishop, the effective stress always decreases
on wetting under constant total stress. As the effective stress
decreases, an increase in the volume of the soil should
be observed in accordance with the above definition of
effective stress. However, experimental data often show
additional compression on wetting, which is contrary to
the prediction based on Bishop’s definition of effective
stress. Many critiques (Fredlund and Rahardjo, 1993;
Fredlund and Morgenstern, 1977) were expressed regarding
the use of a single effective stress measure for unsaturated
soils and there has been a gradual change towards the
use of two independent stress state variables. Fredlund
Morgenstern (1977) proposed to use the net stress V – ua
and the suction s as two independent stress state variables to
describe the mechanical behaviour of the unsaturated soil.
Considering these two stress measures within the critical
state soil mechanics, an elastoplastic constitutive model for
unsaturated soil has been developed by Alonso et al. (1987),
and later by Alonso et al. (1990). Later, other constitutive
models have been proposed, but all of them remain within
the framework of the Alonso and Gens model, which
became known as Barcelona Basic Model (BB-model).
3 Barcelona Basic Model
The BB-model is based on the Modified Cam Clay model
for saturated soil with extensions to include suction effects
in unsaturated soil (Alonso et al., 1990). This model uses
the net stresses V – ua and the suction s as the independent
stress measures. Many symbols have been used for the net
stresses such as V cc and V * . The latter symbol will be used
here. It is assumed that the soil has different stiffness
parameters for a change of the net stress and for a change of
the suction.
3.1 Isotropic loading
In soil mechanics, the soil porosity n is often measured
by its void ratio e being defined as the volume of the pores
4 A.A. Abed and P.A. Vermeer

over the volume of the solid phase. For unloading-reloading, wf wp p wp p

the rate of change of the void ratio is purely elastic and f *
˜ p *  ˜ s  p ˜ Hvp 0 (9)
wp ws wH v
related to the net stress and the suction
with
p * s
e ee N ˜ Ns ˜ (2)
p* s  patm wf
1 (10)
wp*
where N is the well known swelling index and Ns is
the suction swelling index, patm is the atmospheric pressure wp p O  Of pp
˜ p p ˜ E ˜ ln c (11)
and p* is the mean net stress ws O N p
1 * wp p 1 e
p* (V 1  V 2*  V 3* ). (3) ˜ pp . (12)
3 wH p
O N
v

In terms of volumetric strain, equation (2) yields

It follows from the above equations that
*
e N p N s
Hv Hve  ˜  s ˜ (4) O  Of § pp · O N 1 *
1 e 1  e p* 1  e s  patm Hvp  ˜ E ˜ ln ¨ c ¸ ˜ s  ˜ ˜ p . (13)
1 e ©p ¹ 1 e pp
where compressive strains are considered positive.
For primary loading, both elastic and plastic strains This equation is in full agreement with equation (5), but
develop. The plastic component of volumetric strain is instead of pp0 it involves the stress measures s and p* .
given by Equation (13) shows the so-called soil collapse upon
wetting. Indeed, upon wetting we have s  0 and the above
O0  N p p 0 equation yields an increase of volumetric strain, i.e., Hvp ! 0
Hvp ˜ (5)
1 e p p0 even at constant load, i.e., for p * 0 .

where O0 is the compression index and p p 0 is the
preconsolidation pressure in the saturated state (Alonso
et al., 1990). The above equation is in accordance
with critical state soil mechanics. The difference from
the critical state soil mechanics is the yield function
f p*  p p
3.1 More general states of stress
For the sake of convenience, the elastic strains will not
be formulated for rotating principal axes of stress and
strain. Instead, restriction is made to non-rotating principal
stresses. For such situations equation (4) can be generalised
(6) to become

with V i* Dij ˜  H ej  K s1 ˜ G j ˜ s  for i, j 1, 2, 3 (14)

O0 N *
§ p p 0 · O N where Hie is the principal elastic strain rate, V i is the
c
pp p ˜¨ c ¸ (7) principal net stress, Gj = 1 for j = 1, 2, 3 and
© p ¹
Ns
O Of  (Of  O0 ) ˜ e E ˜s (8) K s1 (15)
3 ˜ (1  e) ˜ ( s  patm )
where O and pp are the compression index and the
ª1 Q Q Q º
suction dependent preconsolidation pressure, respectively. E «
Hence, for full saturation, we have s = 0, O = O0 Dij ˜ Q 1 Q Q »» (16)
(1  2Q ) ˜ (1  Q ) «
and p p p p 0 . The larger the suction, the smaller the ¬« Q Q 1 Q ¼»
compression index O. In the limit for s = f, the
above expression yields O = Of. The index ratio Of/O0 is where Q is the elastic Poisson ratio. Young’s modulus is
typically in the range between 0.2 and 0.8. The constant pc stress dependent
is mostly in the range from 10 to 50 [kPa]. The constant E 1 e
controls the rate of decrease of the compression index E 3 ˜ (1  2 ˜Q ) ˜ K with K ˜ p* . (17)
N
with suction; it is typically in the range between 0.01
and 0.03 [kPa–1]. The monotonic increase of soil stiffness The term K s1 ˜ G j ˜ s in equation (14) represents the
with suction is associated with an increase of the contribution of suction loading-unloading (drying-wetting)
preconsolidation pressure pp according to equation (7). to the elastic strain rates, whereas the other term represents
In order to study equation (5) in more detail, we the net stress loading-unloading contribution.
consider the consistency equation f 0 , as it finally leads For formulating the plastic rate of strain, both the plastic
to equation (5). In terms of partial derivatives, the potential and the yield function have to be considered.
consistency equation yields For the BB-model the yield function reads
 Numerical simulation of unsaturated soil behaviour 5

f q 2  M 2 ( p *  ps ) ˜ ( p p  p * ) (18) 4 Numerical implementation

where M is the slope of the critical state line, as also The numerical implementation is done using an implicit
indicated in Figure 2, and return mapping algorithm according to the well-known
elastic predictor/plastic corrector strategy.
1 The stress correction is performed in the space of stress
q (V 1*  V 2* ) 2  (V 2*  V 3* ) 2  (V 3*  V 1* ) 2 (19)
2 invariants p* and q. On Gauss point level, the question is
to evaluate the values of Vi* and the hardening parameter pp
ps = a . s. (20) corresponding to a given strain increment 'Hi and a given
It can be observed from Figure 2 that ps reflects the suction increment 's assuming that their initial values are
extension of the yield surface in the direction of tension known at time tn.
part due to apparent cohesion. The constant a determines Box 1 shows the basic steps followed during the
the rate of ps increase with suction. return mapping, where [ stands for the deviatoric tensor of
stresses being defined as [ij = Vij* – p*.Gij where Vij* is
Figure 2 Yield surface of Barcelona Basic Model defined as the Cauchy stress tensor and Gij is the Kronecker
delta . The symbol || . || means the Euclidean norm of a
second order tensor and n̂ is the normalised deviatoric
tensor.

Box 1 Return mapping for BB-model

The yield function (18) reduces to the Modified

Cam Clay (MCC) yield function at full saturation with s = 0.
In contrast to the MCC-model, the BB-model has a
non-associated flow rule, which may be written as
wg
Hip /˜ (i 1, 2,3) (21)
wV i*

where Hip stands for the principal rate of plastic strain,

/ is a multiplier and g is the plastic potential function
g D ˜ q 2  M 2 ( p*  ps ) ˜ ( p p  p* ). (22)

The flow rule becomes associated for D , but Alonso

et al. (1990) recommend to use
M ( M  9)( M  3) O0 For solving the multiplier /, one needs to construct
D ˜ . (23) the residual vector r and the unknown vector x defined as
9(6  M ) O0  N
follows
In this way the crest of the plastic potential in p* – q-plane
­ tr wg ½
is increased. Finally it leads to realistic K0-values °qn 1  qn 1  3G ˜ / ˜ wq °
in one-dimensional compression, whereas the associated ° n 1
° ­qn 1 ½
MCC-model has the tendency to overestimate K0-values °° * tr wg °° ° * °
* ° pn 1 °
(Roscoe and Burland, 1968). r ® pn 1  pn 1  K ˜ / ˜ wp* ¾ , x ® ¾ (25)
In combination with equations (14) and (21) the ° n 1
° °/ °
° f ° ° pp °
consistency condition f 0 yields the following ° ° ¯ n1 ¿
Hvp /( O *N *)
expression for the plastic multiplier °¯ p pn1  p pn ˜ e °¿
1 wf T 1 § wf wf T · With G = 3K(1 – 2Q)/(1 + Q) is the elastic shear modulus,
/ ˜ * Dij H j  ˜ ¨  K s1 ˜ * Dij G j ¸ ˜ s
H wV i H © ws wV i ¹ N* = N/(1 + e) and O* = O/(1 + e).
Using the Newton-Raphson iteration technique, one can
with solve the previous system of four nonlinear equations.
It is worth noting that Hvp can be expressed as
wf wg wf T wg Hvp / ˜ wg / wp * in the last residual equation. Box 2, shows
H  p
˜ *  * ˜ Dij ˜ * . (24)
wH v wp wV i wV j solution strategy following the Newton-Raphson method.
6 A.A. Abed and P.A. Vermeer
The adopted scheme is considered as a fully implicit one
in the sense that all the unknowns are updated implicitly
during the iteration process. A complete convergence study
is available for testing the implementation efficiency but it
is beyond the scope of this paper.
By checking Step 2 in Box 1, one can see that a suction
reduction under constant net stress triggers the plastic
correction routine if the stress point was already on the yield
surface. This feature helps in capturing the soil structure
collapse phenomenon. In what follows, the implemented
model is used to solve some boundary value problems to
illustrate the effect of suction on both soil strength and
stiffness.
Box 2 Newton-Raphson algorithm
5 Settlement analysis
Figure 3 shows the geometry, the boundary conditions
and the finite element mesh for the problem of a rough
strip footing resting on partially saturated soil. The material
properties shown in Table 1 are the same as those given by
Compas and Vargas (1991) for a particular collapsible silt.
However, as they did not specify the M-value, we assumed a
critical state friction angle of 31q, which implies M = 1.24.
The ground water table is at a depth of 2 m below the
footing.
The initial pore water pressures are assumed to be
hydrostatic, with tension above the phreatic line. For the
suction, this also implies a linear increase with height above
the phreatic line, as in this zone the pore air pressure ua is
assumed to be atmospheric, i.e., s = ua–uw = –uw. Below the
phreatic line, pore pressures are positive and we set ua = uw,
as also indicated in Figure 3.
For uw < 0 the linear increase of uw implies a decreasing
degree of saturation, as also indicated in Figure 3. In fact,
the degree of saturation is not of direct impact to the present
settlement analysis, as transient suction due to deformation
and changing degrees of saturation are not considered.
The distribution of saturation being shown in Figure 3, was
computed using the van Genuchten model (Van Genuchten,
1980), together with additional data for the silt. For the
sake of convenience however, a constant (mean) value of
17.1 kN/m3 has been used for the soil weight above the
phreatic line. For the initial net stresses the K0-value of 1
has been used. The finite element mesh consists of 6-noded
triangles for the soil and 3-noded beam element for the strip
footing. The flexural rigidity of the beam was taken to be
EI = 10 MN.m2 per metre. This value is representative for a
reinforced concrete plate with a thickness of roughly 20 cm.
Figure 3 Geometry, boundary conditions and finite element
mesh
Table 1 Material and model parameters
e0 1.67 E 0.0215 [kPa–1]
O0 0.22 v 0.2
N 0.006 Ns 0.008
p p0 80 [kPa] a 0.6
M 1.24 pc 18.1 [kPa]
Of 0.123
Computed load-settlement curves are shown in Figure 4
both for the Barcelona Basic model and the Modified Cam
Clay model. For the latter MCC-analysis, suction was fully
neglected. In fact it was set equal to zero above the phreatic
line. On the other hand suction is accounted for in the
BB-analysis, but we simplified the analysis by assuming no
change of suction during loading. In reality, footing loading
will introduce a soil compaction and thus some change of
both the degree of saturation and suction. As yet this has not
been taken into account.
Up to an average footing pressure of 80 kPa, both
analyses yield the same load-displacement curve. This
relates to the adoption of preconsolidation pressure
pp0 = 80 kPa. For pressures beyond 80 kPa, Figure 4 shows
a considerable difference between the results from the
BB-analysis and the MCC-analysis. Indeed, the BB-analysis
yields much smaller settlements than the MCC-model.
Hence, settlements are tremendously overestimated when
suction is not taken into account. The impact of suction is
also reflected in the development of the plastified zone
below the footing. For the BB-analysis the plastic zone with
f = 0 is indicated in Figure 5(a). The MCC-analysis shows
 Numerical simulation of unsaturated soil behaviour 7

a larger plastic zone underneath the footing, as shown in models nearly yield the same final settlement of about
Figure 5(b). 49 mm.

Figure 4 Footing pressure-settlement curves Figure 6 Vertical displacement of soil surface due to wetting

7 Bearing capacity
Figure 5 The plastic zones from BB and MCC model for footing
pressure of 150 kPa (see online version for colours) From Figure 4 it might be seen that the bearing capacity of
the footing is nearly reached, at least for the MCC-analysis
without suction. However, the collapse load is far beyond
the applied footing pressure of 150 kPa, at least for a
Drucker-Prager type generalisation of the Modified Cam
Clay model and a CSL-slope of M = 1.24. The applied
Drucker-Prager generalisation involves circular yield
surfaces in a deviatoric plane of the principal stress
space, which is realistic for small friction angles rather
than large ones. For this reason, we will analyse the
bearing capacity of a strip footing for a relatively low
6 Increase of ground water level
CSL-slope of M = 0.62. Under triaxial compression
Having loaded the footing up to an average pressure of conditions we have M 6 ˜ sin Mcs /(3  sin Mcs ) and we get a
150 kPa, we will now consider the effect of soil wetting friction angle of Mcs = 16.4o. However, we consider the
by increasing the ground water table up to ground surface. plane strain problem of a strip footing. For planar
This implies an increase of pore water pressures and thus
deformation it yields M 3 sin Mcs (Chen and Baladi,
a decrease of effective stresses, being associated with soil
heave. On simulating this rise of the ground water level by 1985), and it follows that Mcs = 21o. Table 2 gives the soil
the MCC-model, both the footing and the adjacent soil parameters. Figure 7 shows the boundary conditions and the
surface are heaving, as plotted in Figure 6. Due to the fact finite element mesh for the bearing capacity problem of a
that we adopted an extremely low swelling index of only shallow footing on unsaturated soil.
0.006 (see Table 1) the heave is relatively small, but for
Figure 7 Finite element mesh and boundary conditions for the
other (expansive) clays it may be five times as large. Similar
bearing capacity problem
to the MCC-analysis, the BB-analysis yields soil heave
as also shown in Figure 6. In contrast to the MCC- analysis,
however, the footing shows additional settlements.
Here it should be realised that Figure 6 shows vertical
displacements due to wetting only, i.e., an extra footing
settlement of about 25 mm.
The BB-analysis yields this considerable settlement of
the footing, as it accounts for the loss of so-called capillary
cohesion as soon as the suction reduces to zero. In text
books (Fredlund and Rahardjo, 1993), this phenomenon is
referred to as soil (structure) collapse. The different
performance of both models is nicely observed in Figure 4.
Here the BB-analysis yields a relatively stiff soil behaviour
when loading the footing up to 150 kPa, followed by
considerable additional settlement upon wetting. In contrast,
the MCC-model yields a relatively soft response upon In this analysis, the soil has been loaded up to failure using
loading and footing heave due to wetting. Finally, both again both the BB-model and the MCC-model.
8 A.A. Abed and P.A. Vermeer

In order to be able to compare the numerical results Figure 9 Incremental shear strain at failure for s = 20 kPa
with theoretical values, we used a uniform distribution for (see online version for colours)
suction in the unsaturated part of s = 20 kPa. The soil is
considered to be weightless and the surcharge soil load
is replaced by a distributed load of 25 kN/m2 per unit
length, which is equal to a foundation depth of about
1.5 m. A value of K0 = 1 is used to generate the initial net
stresses.
The same finite element types as in the previous
problem are used here for the soil and the footing.
According to Prandtl, the bearing capacity is given by
qf c ˜ Nc  q0 ˜ Nq  J ˜ b ˜ NJ (26) Figure 10 Total displacement increments for s = 20 kPa
(see online version for colours)
where c is the soil cohesion, q0 is the surcharge load
at footing level and b is the footing width. The factors Nc, Nq
and NJ are functions of the soil friction angle
1  sin M S ˜tan M
Nq ˜e , Nc ( Nq  1) ˜ cot M. (27)
1  sin M

In the present analysis, J is taken equal to zero and the

corresponding NJ-factor is not needed. For the zero-suction
case, we have c = 0, and the bearing capacity qf is found to Table 2 Soil properties
be 177 kPa.
According to the BB-model, the cohesion c increases e0 0.9 E 0.0215 [kPa–1]
with suction s linearly, according to the formula O0 0.14 v 0.2
c a ˜ s ˜ tan M . (28) N 0.015 Ns 0.01
p p0 60 [kPa] a 1. 24
On using a = 1.24 and s = 20 kPa, we find c = 9.5 kPa.
For this capillary cohesion of 9.5 kPa, the Prandtl equation M 0.625 pc 43 [kPa]
yields qf = 327 kPa. Figure 8 shows the calculated Of 0.036
load-displacement curves using the BB-model and the
MCC-model. The figure shows that an increase of suction
Table 3 Bearing capacity values
value by 20 kPa was enough to double the soil bearing
capacity. Shear bands at failure, as shown in Figure 9, Suction [kPa] 0 20
are typically according to the solution by Prandtl. Theoretical bearing capacities [kPa] 177 327
In Figure 10, the displacement increments show the
Numerical bearing capacities [kPa] 160 315
failure mechanism represented by footing sinking, which
is associated with soil heave at the edges. By comparing Relative error [%] 9.6 3.7
the theoretical bearing capacity values with the computed
ones (Table 3), it is clear that the results are quite It is believed that we can capture better bearing capacity
satisfactory with a relatively small error. values by adopting a more advanced failure criterion
than the Drucker-Prager criterion being used in this
Figure 8 Loading curves for BB- and MCC-analysis analysis. One can use a modified version of the well-known
Mohr-Coulomb failure criterion which accounts for
suction effects, or Matsuoka et al. criterion (Matsuoka et al.,
2002) which offers us a failure surface without singular
boundaries and, as a consequence, a more suitable criterion
for numerical implementation.

8 Groundwater flow
Ground water flow is governed by the ground water head
h = y + uw/Jw, where y is the geodetic head and uw/Jw is the
pressure head, which will be denoted as \ for the sake of
 Numerical simulation of unsaturated soil behaviour 9

simplicity. Jw is the unit pore water weight. In most practical
cases, there will not be a constant ground water head, but a
variation with depth and, consequently, ground water flow.
Indeed, in reality there will be a transient ground water
flow due to varying rainfall and evaporation at the soil
surface. This implies transient suction fields and footing
settlements that may vary with time. For most footing,
settlement variations will be extremely small, but they will
be significant for expansive clays as well as collapsible
subsoil. In order to analyse such problems, we will have
to incorporate ground water flow. Flow in an isotropic soil
is described by the Darcy equation
In order to solve the differential equations (29) and (31),
boundary conditions are required. For studying footing
problems, one would need the water infiltration or the rate
of evaporation q at the soil surface.
The PlaxFlow finite element code (Brinkgreve
et al., 2003) has been used to calculate suction in
unsaturated zone. In the following section, the numerical
results of PlaxFlow are checked by analytical solutions
in the case of one-dimensional unsaturated groundwater
flow.
Figure 11 Gardner permeability model

wh
qi  krel ˜ ksat ˜ (29)
wxi

where qi is a Cartesian component of the specific discharge

water, ksat is the well-known permeability of a saturated
soil and krel is the pore pressure head-dependent relative
permeability. Gardner (1958) proposed a simple exponential
relative permeability function of the form
krel eD ˜\ for \  0
(30)
krel 1 for \ t 0
8.1 Analytical solution for one-dimensional
where D is a fitting parameter. It is worth pointing out that unsaturated stationary groundwater flow
the pressure height \ has a negative value in the unsaturated
For one-dimensional vertical steady-state flow equation (31)
zone. Figure 11 shows a graphical representation of
reduces to equation (34) in terms of pore pressure head \
equation (30) for D = 2 m–1.
In order to do ground water flow calculations, one has w ª § w\ ·º
to supplement Darcy’s equation (29) with a continuity « k sat ˜ krel (\ ) ˜ ¨  1¸ » 0. (34)
wy ¬ © wy ¹¼
equation of the form
wqi wh Gardner (1958) gave an analytical solution for the
C˜ 0 (31) differential equation (34) with infiltration or evaporation
wxi wt
boundary conditions at soil surface, as shown in
where repeated subscripts stand for summation. C is the Figure 12. Using equation (30) for the permeability
effective storage capacity, which is often expressed as function, the solution is given as
(Brinkgreve et al., 2003)
1 ª q q º
\ ˜ ln «(  1) ˜ e D ˜ y  ». (35)
dS r D ¬ k sat k sat ¼
C Csat  n ˜ (32)
ds
Gardner used negative values for q to indicate infiltration
where Csat is the saturated storage capacity and Sr the and positive values to indicate evaporation.
degree of saturation. The latter is a function of pore pressure
head \. In this publication we will use the following simple Figure 12 Boundary conditions used in Gardner solution
equation as in Srivastava and Yeh (1991)
Sr S res  ( S sat  S res ) ˜ eD\ (33)

Sres is the residual degree of saturation, Ssat is the degree

of saturation at full saturation which is usually taken as one.
Strictly speaking, soil deformation implies changing soil
porosity n and pore fluid flow cannot be separated from soil
deformation. For many practical problems, however, the soil
porosity remains approximately constant and flow problems
may be simulated without consideration of coupling terms.
10 A.A. Abed and P.A. Vermeer
8.2 Verification of the numerical results
A silty soil with saturated permeability of ksat = 1 m/day and
D = 2 m–1 has been used to compare PlaxFlow numerical
results with the analytical solution of equation (34).
Figure 13 shows the finite element mesh and boundary
conditions as used in the verification examples. Six noded
triangular elements are used with closed vertical boundaries
to recover 1-D conditions. The ground water table is at
3.0 m depth and a Neumann type boundary condition is
applied at the soil surface.
Figure 13 Finite element mesh and boundary conditions used
in numerical calculations
The problem has been solved for three common
practical situations, namely, hydrostatic conditions with
q = 0, evaporation with q = 0.002 m/day and infiltration
with q = 0.1 m/day. The hydrostatic conditions represent
a pore pressure distribution in unsaturated soil which has
no interaction with surface water, for example a soil directly
underneath a raft foundation. Evaporation represents soil
moisture decrease due to temperature increase during a dry
season and infiltration represents soil moisture increase
due to rainfall for instance.
Figure 14 shows a very good agreement between
analytical and numerical results for this particular kind
of problem. At the same time, it gives a nice idea about
negative pore water pressure profiles in such common cases.
Figure 14 Analytical vs. numerical results
8.3 Analytical solution for one-dimensional
unsaturated transient groundwater flow
The analytical solution of equation (32) in the case of
one-dimensional vertical transient flow i.e., negative pore
water pressure head \ as a function of time and vertical
position is presented by Srivastava and Yeh (1991) for the
particular water characteristic and permeability functions
given in equations (33) and (30), respectively
1
\ ˜ ln( R ) (36)
D
where
q § q · 4 ˜ q ( L  y ) / 2 t* / 4
R ¨  1¸ ˜ e  y  ˜e ˜e ˜
k sat © k sat ¹ ksat
(37)
2 *
f
sin(Zi ˜ y ) ˜ sin(Zi ˜ L) ˜ e Zi ˜t
¦
i 1 1  ( L / 2)  2Zi2 ˜ L
n ˜ D ˜ ksat ˜ t
t* (38)
Ssat  Sres
Zi is the ith root of the characteristic equation (39), L is the
layer thickness and t is the time.
tan(Z ˜ L)  2 ˜ Z 0. (39)
8.4 Verification of the numerical results
The same material properties mentioned in Section 8.2 and
the same geometry and boundary conditions are used
here considering a hydrostatic suction distribution as an
initial condition. The soil is exposed to infiltration rate of
q = 0.1 m/day. The analytical solution is presented by the
solid lines in Figure 15, whereas the numerical results
are represented by dots. This figure shows good agreement
between the numerical and analytical solution. It is also
interesting to notice that Gardner’s solution for the steady
state is obtained after five days of continuous infiltration.
Figure 15 Negative pressure head at different time steps
Since we are confident about the numerical results of the
ground water flow module, we will study the mechanical
behaviour of a shallow foundation during a rainfall event in
the next section.
9 Footing deformation with time
In reality, suction varies with position and time. Up to
this point, we have studied only the case of shallow
foundation which is exposed to a ‘sudden’ suction
reduction by increasing ground water level.
 Numerical simulation of unsaturated soil behaviour 11

In this section, a more realistic simulation is considered Figure 17 Suction and vertical displacements with time
where the deformations of a shallow foundation are
calculated during soil saturation due to rainfall. The
geometry, the boundary conditions and the finite element
mesh are shown in Figure 16. The mechanical properties of
the soil are the same as in Table 2 with a preconsolidation
pressure of pp0 = 30 kPa and pc = 10 kPa. The hydraulic
properties are the same as those used to check the
unsaturated ground water flow finite element code in
Section 8. The finite element mesh consists of 6-noded
triangles for the soil and 3-noded plate elements for the
strip footing. The flexural rigidity of the plate was taken
to be EI = 10 MN.m2 per metre. The initial stresses are
considered to be isotropic, with K0 = 1. A hydrostatic
distribution is adopted as an initial condition for the
transient ground water flow calculations.

Figure 16 Finite element mesh, initial and boundary condition

for transient deformation analysis (see online version
for colours)

Figure 18 Load-settlement curve at the end of infiltration

The footing is loaded upto 50 kPa. Afterwards a transient

water flow calculation is done to simulate a five days
rainfall event with infiltration rate of q = 0.1 m/day.
The calculated suction values at different time steps are
provided to the deformation routine to evaluate the
resulting displacements. Figure 17 shows suction and Figure 19 Net stresses distribution at the end of analysis
vertical displacements underneath the footing with time. (see online version for colours)
In this analysis, we expect a final value of suction equal to
that shown in Figure 15 which is also the value given by
Gardner’s solution for the steady-state situation, however it
is obvious from Figure 17(a) that suction variations are
negligible after two days. This has a direct consequence
on the footing settlements, which take place essentially
during the first two days and then reach a steady situation
as shown in Figure 17(b). The rainfall causes 30 mm of
additional settlement to the 60 mm due to the 50 kPa
of loading. The additional deformation reproduces a
well-known type of unsaturated soil behaviour upon wetting
10 Conclusions
called the partial soil structure collapse, as shown in
Figure 18. The word ‘partial’ means that additional wetting The present study illustrates the possibility of simulating
of the soil will lead to further settlement until the soil the mechanical behaviour of unsaturated soil, using the
reaches a full saturation. A typical distribution of the net finite element method with a suitable constitutive model.
stresses underneath the footing for such type of problems is On incorporating suction, soil behaviour was shown to
shown in Figure 19. be much stiffer than without suction. Moreover, it has
12 A.A. Abed and P.A. Vermeer
been shown that soil collapse was well simulated. This
phenomenon is well-known from laboratory tests, but it also
applies to footings as shown in this study.
In general, shallow foundations will not be built on
collapsible soils, but many footings have been constructed
on swelling clays and this will also be done in the future.
From an engineering point of view, pile foundations may
be preferred, but they are often too costly for low-rise
buildings. Therefore, heave and settlement of shallow
foundations on expansive clays will have to be studied
in full detail. At this point, a one-dimensional transient
flow calculation of an infiltration and evaporation processes
can be very helpful. By applying transient boundary
conditions, one can simulate the variation of a suction
profile with time; typically for two or three years
(Abed, 2007). Depending on the results, the designer can
pick the lowest and the highest suction values in the studied
period. With this information in hand, deformation analyses
for these cases can be done to determine the absolute
foundation deformation variations as well as the differential
settlements with respect to neighbouring footings. Such
movements due to suction variations can introduce quite
high bending moments in the beams, columns and walls of
superstructures if they have not been considered in design.
Another important application of unsaturated soil
mechanics is seen in the field of slope stability. Many
natural slopes have low factors of safety and slope failures
are especially imminent after wetting by rainfall. Hence,
soil collapse computations would seem to be of greater
interest to slopes than to footings, as considered in this
study. Not only natural slopes, but also river embankments,
suffer upon wetting. High river water levels tend to occur
for relatively short periods of time, so that there is partial
wetting. This offers also a challenging topic of transient
ground water flow and deformations in unsaturated ground.
Acknowledgements
We are grateful to Dr. Peter van den Berg of GeoDelft,
the Netherlands, for providing support for this study.
Special thanks are due to Mr. John van Esch of GeoDelft,
to Professor Antonio Gens from the University of Catalunia
and Dr. Klaas Jan Bakker of the Plaxis company for fruitful
discussions on unsaturated soil behaviour.
References
Alonso, E.E., Gens, A. and Hight, D.W. (1987) ‘Special problem
soils’, General Report, Proc. 9th Eur. Conf. Soil Mech.,
Dublin.
Alonso, E.E., Gens, A. and Josa, A. (1990) ‘A constitutive
model for partially saturated soils’, Geotechnique, Vol. 40.
pp.405–430.
View publication stats
Abed, A. (2007) Abed, ‘Numerical simulation of a trial wall on
expansive soil in Sudan’, Plaxis Bulletin, Vol. 21, pp.16–18.
Bishop, A.W. (1959) The Principle of Effective Stress, Teknisk
Ukebladi, p.39.
Brinkgreve, R., Al-Khoury, R. and van Esch, J. (2003)
PLAXFLOW User Manual, Balkema, Rotterdam.
Chen, W.F. and Baladi, G.Y. (1985) Soil Plasticity, Elsevier.
de Compas, T.M. and Vargas, E.A. (1991) ‘Discussion:
a constitutive model for partially saturated soils’,
Geotechnique, Vol. 41, pp.273–275.
Djedid, A., Bekkouche, A. and Mamoune, S.M. (2001)
‘Identification and prediction of the swelling behaviour
of some soils from the Telmcen region of algeria’, Bulletin
des Laboratories des Ponts et Chaussees, July–August, p.233.
Fredlund, D.G. and Morgenstern, N.R. (1977) ‘Stress state
variables for unsaturated soils’, Journal of the Geotechnical
Engineering Division, Proceedings, American Society of
Civil Engineering (GT5), pp.447–466.
Fredlund, D.G. and Rahardjo, H. (1993) Soil Mechanics for
Unsaturated Soils, John Wiley & Sons.
Gardner, W.R. (1958) ‘Some steady-state solutions of the
unsaturated moisture flow equation with applications to
evaporation from a water table’, Soil Science, Vol. 85,
pp.228–232.
Matsuoka, H., Sun, D., Kogane, A., Fukuzawa, N. and
Ichihara, W. (2002) ‘Stress-strain behaviour of unsaturated
soil in true triaxial tests’, Can. Geotech. J., Vol. 39,
pp.608–619.
Nelson, J.D. and Miller, D.J. (1992) Expansive Soils, John
Wiley & Sons.
Roscoe, K.H. and Burland, J.B. (1968) ‘On the generalized
stress-strain behaviour of ‘wet’ clay’, Engineering Plasticity,
Cambridge University Press, pp.535–609.
Srivastava, R. and Yeh, J. (1991) ‘Analytical solutions for
one-dimensional, transient infiltration toward the water table
in homogeneous and layered soils’, Water Resources
Research, Vol. 27, pp.753–762.
Van Genuchten, M.Th. (1980) ‘A closed-form equation for
predicting the hydraulic conductivity of unsaturated soils’,
Soil Sci. Soc. Am. J., Vol. 44, pp.892–898.
Vermeer, P.A. and Brinkgreve, R. (1995) PLAXIS – Finite Element
Code for Soil and Rock Analysis, Balkema, Rotterdam.