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

net/publication/237089710

Parameter identification for elasto-plastic modelling of unsaturated soils

from pressuremeter tests by parallel modified particle swarm optimization

Article  in  Computers and Geotechnics · March 2013

DOI: 10.1016/j.compgeo.2012.08.004

CITATIONS READS

18 181

3 authors:

Youliang Zhang Domenico Gallipoli

Chinese Academy of Sciences Université de Pau et des Pays de l'Adour
25 PUBLICATIONS   117 CITATIONS    129 PUBLICATIONS   1,673 CITATIONS   

SEE PROFILE SEE PROFILE

Charles Augarde
Durham University
154 PUBLICATIONS   1,979 CITATIONS   

SEE PROFILE

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

Tunnels, Excavations, Settlements and Buildings View project

SHIBAM / MECAD View project

All content following this page was uploaded by Domenico Gallipoli on 19 October 2017.

The user has requested enhancement of the downloaded file.

 Paper published in:
Y. Zhang, D. Gallipoli, C.E. Augarde (2013). Parameter identification for elasto-plastic modelling of
unsaturated soils from pressuremeter tests by parallel modified particle swarm optimization.
Computers and Geotechnics, 48: 293–303 - http://dx.doi.org/10.1016/j.compgeo.2012.08.004

Parameter identification for elasto-plastic modelling of unsaturated soils from

pressuremeter tests by parallel modified particle swarm optimization

Youliang Zhang1 Domenico Gallipoli2 Charles Augarde3

1 State Key Laboratory of Geomechanics and Geotechnical Engineering, Institute of Rock and
Soil Mechanics, Chinese Academy of Sciences, Wuhan, P.R. China
ylzhang@whrsm.ac.cn
2 Department of Civil Engineering, University of Glasgow, Glasgow, U.K.

gallipoli@civil.gla.ac.uk
3 School of Engineering & Computing Sciences, Durham University, Durham, U.K.

charles.augarde@dur.ac.uk

Abstract: This paper presents a methodology for the identification of parameter values in the
Barcelona Basic Model (BBM) by inverse analysis of the experimental cavity pressure-cavity strain
curve from pressuremeter tests in unsaturated soils. This methodology involves a high-dimensional
optimization process which is particularly challenging due to the existence of a large number of
local minima caused by the nonlinearity of the BBM. A novel parallel modified particle swarm
optimization algorithm is utilized to minimize the difference between measured and computed
values on the cavity pressure-cavity strain curve. The computed cavity pressure-cavity strain curve
is obtained by using a finite element model of an unsaturated soil whose mechanical behaviour is
described by the BBM. An example is presented to validate the proposed methodology making use
of artificial experimental results that had been calculated by a finite element simulation of
pressuremeter tests. Finally, the application to a real case is presented by showing that the
proposed methodology can safely identify the values of at least six BBM parameters through
inverse analysis of pressuremeter tests at different suction levels.

Keywords: Barcelona Basic Model (BBM); parameter identification; Hybrid moving boundary
particle swarm optimization (hmPSO); pressuremeter test; parallel computing

1
1. INTRODUCTION

Research into the constitutive behaviour of unsaturated soils has traditionally focused on the
interpretation of material behaviour rather than on the selection of parameter values inside specific
constitutive models. Currently there are a number of constitutive models, which are able to capture
the main features of unsaturated soil behaviour (Alonso et al. 1990; Cui and Delage 1996;
Fredlund et al. 1996; Gallipoli et al. 2003; Wheeler and Sivakumar 1995). However, a major
obstacle to the use of these models is the need to determine a relatively large number of soil
parameters, requiring costly and time consuming laboratory tests on small scale undisturbed
samples.

In-situ testing provides an appealing alternative to small scale laboratory experiments because it is
generally less expensive and allows faster material characterization. In particular, pressuremeter
tests have been widely used to obtain in-situ measurements of stiffness and strength in saturated
soil deposits. A pressuremeter test consists of the application of an increasing pressure to the
sidewalls of a section of a cylindrical borehole by an inflatable probe (the pressuremeter),
previously lowered to the required depth. The pressure applied by the pressuremeter is plotted
against the corresponding radial expansion of the borehole to give what is conventionally referred
to as the “cavity pressure – cavity strain” curve (the cavity strain is the change in the borehole
radius divided by the initial radius). Unlike laboratory tests, pressuremeter tests do not require
coring of samples from the field, thus limiting soil disturbance prior to testing. This feature is
especially advantageous for unsaturated soils, which are often characterized by an open soil fabric
with large pores and a metastable structure held together by suction that makes sampling difficult.
Pressuremeter tests are widely used for the determination of the engineering properties of
saturated soils. However, they are still only tentatively employed for the characterization of
unsaturated soil deposits because of the lack of recognized experimental standards and reliable
interpretation methods.

Soil parameters can be obtained from pressuremeter tests by performing an inverse analysis of the
cavity pressure – cavity strain relationship measured in the field. In order to achieve this, the
pressuremeter test is usually simulated as the expansion of an infinitely long cylindrical cavity
inside an unbounded uniform medium by using closed-form analytical solutions or approximate
numerical models, such as finite elements (Carter et al. 1986; Cudmani and Osinov 2001; Fahey
and Carter 1993; Hsieh et al. 2002; Yu and Houlsby 1991; Yu and Houlsby 1995). Closed-form
analytical solutions are normally preferred to numerical models because they reduce computational
costs and offer greater accuracy. However, if the complexity of the soil constitutive model makes it
difficult to obtain a closed-form solution, the use of approximate numerical models becomes the

2
only viable alternative. Moreover, in most elasto-plastic models, a relatively large number of
parameter values needs to be defined, which makes empirical curve fitting impractical so that the
use of optimization algorithms for matching simulations to experiments becomes a necessity.

The goal of the present work is to develop an effective and reliable procedure for the
characterization of the engineering properties of unsaturated soils by interpreting results from
pressuremeter tests. In particular, the paper presents a methodology for the identification of
parameter values in the Barcelona Basic Model (BBM) (Alonso et al. 1990) by inverse analysis of
the experimental cavity pressure-cavity strain curve measured during pressuremeter tests in the
field. BBM is arguably one of the most popular elasto-plastic constitutive models for unsaturated
soils and has been widely used in the analysis of boundary value problems in geotechnical
engineering due to its capability of providing a unified interpretation of plastic soil behaviour under
both changes of load and suction.

In this research, the Nelder-Mead local search method (Nelder and Mead 1965) has been used in
conjunction with a novel evolutionary global optimization algorithm (Zhang et al. 2009a) to match
the cavity pressure - cavity strain curves calculated by finite element models to the field
measurements. The proposed evolutionary global optimization algorithm has been formulated
within the framework of Particle Swarm Optimization (PSO) (Kennedy and Eberhart 1995;
Eberhart and Kennedy 1995; Shi and Eberhart 1998a; Shi and Eberhart 1998b) and has been
implemented in a parallel form to improve computational efficiency.

In Section 2 of this paper, the BBM model is briefly introduced followed by a description of the
proposed methodology for the identification of model parameters. Subsequently, an example is
presented to validate the above methodology making use of artificial experimental results produced
by finite element simulations. Finally, the identification of soil parameter values from real
pressuremeter tests is demonstrated.

2. THE BARCELONA BASIC MODEL FOR UNSATURATED SOILS

The BBM (Alonso et al. 1990) is an elasto–plastic constitutive model that extends the framework of
Modified Cam-clay to the case of unsaturated soils. The BBM is able to reproduce the main
features of the mechanical response of an unsaturated soil, such as the increase of shear strength
and preconsolidation pressure with suction, the development of reversible swelling strains during
wetting at low confining stresses and the occurrence of irreversible collapse strains during wetting
at high confining stresses (Alonso et al. 1990; Vaunat et al. 2000).

3
The BBM is formulated in terms of two independent stress variables, namely the net stress tensor
and the suction scalar. The net stress tensor,  ij is defined as  ij   ij*   ij ua where  ij* is the total

stress tensor, ua is the pore air pressure and  ij is the Kronecker delta. The suction, indicated by

the symbol s is defined as s  ua  uw where u w is the pore water pressure.

The yield locus is defined in terms of stress invariants by the equation:

F  p, q, s   q 2 - M 2  p  ks   p0  s  - p   0 (1)

which describes an ellipse in the  p, q  plane at constant suction s , where p is the net mean

stress and q is the deviator stress. In Equation (1), M is the slope of the critical state line in the

 p, q  plane at constant suction, k is the parameter controlling the increase of apparent cohesion

with suction and p0  s  is the isotropic preconsolidation stress, which depends on suction s

according to:

  0  
 p  0    s  
p0  s   p c  0 c  . (2)
 p 

Equation (2) describes the relationship between the isotropic preconsolidation stress and suction in
the  p, s  plane at q  0 and is often referred to as the Loading-Collapse (LC) yield curve. In

Equation (2),   s  and   0  are the slopes of the normal compression lines at suctions equal to

s and 0 respectively,  is the elastic swelling index with respect to changes of mean net stress,
p0  0  is the isotropic preconsolidation stress under saturated conditions and p c is a reference

pressure such that, when p0  0   p c , the LC yield curve reduces to a straight line with equation

p0  s   p c .

The normal compression lines at constant suction s are defined as:

p
v  N ( s)   ( s) ln (3)
pc

where v is the specific volume. The change of intercept N  s  and slope   s  of the normal

compression lines with suction s is described as:

4
 s  patm
N ( s)  N (0)   s ln
patm

(3bis)

 (s)   (0) (1  r )e  s  r 

where  s is the elastic swelling index with respect to changes of suction, N  0  and   0  are the

intercept and slope of the saturated normal compression line at zero suction, r is a parameter
controlling the range of variation for the slopes of the constant suction normal compression lines,
 is a parameter controlling the rate at which the slopes of normal compression lines change with
suction and patm is the atmospheric pressure.

Isotropic plastic hardening is governed by the variation of the parameter p0  0  , which regulates

the change in size of the yield locus and depends on the plastic volumetric strain  vp according to:

dp0 (0) 1 e
 d  vp (4)
p0 (0)   0   

where e is the void ratio.

A non-associative flow rule relates increments of plastic shear and volumetric strains:

d  sp 2q
 2 (5)
d  v M  2 p  ks  p0 ( s) 
p

where  sp is the plastic shear strain while  is a constant depending on the value of Jaky‟s

coefficient of earth pressure at rest and has been introduced to improve prediction of soil behaviour
during K0 loading.

The elastic variation of volumetric and deviatoric strains, denoted by the symbols  ve and  se

respectively, is calculated as:

 dp s ds
d  ve   (6)
1 e p 1  e s  patm

G
d  se  dq (7)
3
5
where G is the elastic shear modulus.

The five BBM parameters N (0) ,  (0) , r ,  and p c , as well as the initial value of the hardening

parameter p0 (0) , are therefore associated with the plastic behaviour of the soil under isotropic

stress states. The values for such parameters are conventionally obtained from laboratory tests
involving isotropic compression of unsaturated samples at constant suction. In particular, once the
two swelling indices  and  s have been fixed based on loading-unloading and drying-wetting

paths, a minimum of three experimental isotropic virgin loading tests at different suctions are
needed to define the values of five parameters N (0) ,  (0) , r ,  , p c and the initial value of the

hardening parameter p0 (0) (see, for example, the calibration procedure by Gallipoli et al. 2010). It

is also useful to note that, as an alternative to the selection of parameter N (0) , it is possible to
define an initial value of void ratio.

The three parameters G , M and k are associated with the soil behaviour under shearing up to
failure and their values are usually obtained from laboratory tests involving triaxial compression at
constant suction.

In summary, there are ten parameter values and one initial hardening parameter value to be
selected in the BBM, ( G , M , k ,  s ,  , N (0) ,  (0) , r ,  , p c , p0 (0) ) although, as discussed

later, the proposed optimization procedure is rarely used to search for the full set of parameter
values.

3. IDENTIFYING BBM PARAMETER VALUES FROM PRESSUREMETER TESTS

The general framework adopted in this work for the calibration of the BBM from pressuremeter
tests is initially described in this section. Subsequently, the two main components of the proposed
methodology, i.e. the finite element simulation of pressuremeter tests and the optimization
algorithm, are discussed in greater detail.

3.1 Methodology

The identification of parameter values in soil constitutive models can be achieved through inverse
optimization of a boundary value problem for which experimental measurements are available. To
accomplish this, it is first necessary to develop a computational simulator of the chosen boundary
value problem, which must also incorporate the required constitutive model. The values of the
constitutive parameters are then adjusted until the difference between the predictions from the
simulator and the experimental measurements is minimized. The goal is to find an optimum set of

6
parameter values for which the predicted response is closest to measurements. Mathematically,
the problem consists of finding the set of n parameters, arranged in a vector x, which minimizes an
objective function measuring the difference between computed and measured values:

f (x)  uic (x)  uim (8)

where uim is the ith measured value while uic is the corresponding ith computed value, the latter

depending on the n parameters in vector x , and . is a suitable norm. Bounds are set on the

values of the parameters in vector x and these are given by the maximum and minimum vectors
x max and x min , respectively. The computed values uic can be obtained from either closed-form
analytical solutions or numerical models of the chosen boundary value problem.

The flow chart in Figure 1 illustrates the algorithm for inverse analysis of pressuremeter test results.
The flow chart starts from the consideration of two different pieces of information: on one side,
there is a set of experimental pressure-expansion curves measured during pressuremeter tests in
the field while, on the other side, there is a corresponding set of finite element simulations, each
reproducing one of the field tests. The initial set of simulations is performed by assuming some trial
values of the soil constitutive parameters. The objective function is then defined to measure the
difference between the two sets of measured and simulated curves. If this difference is larger than
a given tolerance, new values of the constitutive parameters are selected by the optimization
algorithm and a new set of finite element simulations is performed with these updated values. The
value of the objective function is then recalculated to provide an updated measure of the difference
between measured and simulated curves. This process is repeated until the optimum parameter
values are found, i.e. until the value of the objective function becomes smaller than a set tolerance.
The above generic algorithm has been implemented in this work by using a PSO routine to seek
the optimum set of BBM parameter values. A single piece of software integrates seamlessly the
finite element simulator, implementing the BBM, together with the PSO search engine.

If the objective function is highly non-linear with a multi-modal nature, a large number of
evaluations of the objective function are required before the optimum set of parameter values is
obtained. The computation time for a single finite element simulation can range from seconds to
hours depending on the problem of interest and the size of the model. For the cavity expansion
problem studied here, an average of 3 minutes was required on a standard PC for a single
evaluation of the objective function (note, however, that in the PSO routine a single evaluation of
the objective function requires multiple FE analyses) and more than 50000 evaluations of the
objective function were needed to obtain the optimal values of a sub-set of 8 BBM parameters.

7
This would typically need more than 100 days if using a standard PC. This large computational
cost makes the proposed optimization procedure not suitable for a PC unless the algorithm is
implemented in a parallel form that makes use of multiple processors to speed up computations.

3.2 Finite element simulation of pressuremeter tests

In this work, the computed cavity pressure – cavity strain curves are obtained from the finite
element simulation of an infinitely long cylindrical cavity expanding inside an unsaturated soil
modelled by the BBM. Due to the relative complexity of the constitutive model, there is currently no
closed-form analytical solution for this particular boundary value problem.

The above finite element simulations were carried out using the in-house finite element code
MUSEFEM (Zhang and Gallipoli 2007; Zhang et al. 2007), which was developed and validated
against a number of benchmarking problems during the MUSE project (Gallipoli et al. 2006). In
MUSEFEM, unsaturated soils are modelled as three-phase porous media consisting of solid grains,
pore liquid and pore gas (with the gas phase containing only air and the liquid phase containing
both water and dissolved air) with the mechanical material behaviour described by BBM.
Displacements are interpolated using quadratic shape functions while the air and water pressures
are interpolated using linear shape functions.

The 1D expansion of an infinitely long cylindrical cavity is simulated by using a 2D axisymmetric

finite element mesh (see Figure 2) comprising a single line of eight-noded quadrilateral elements
together with the imposition of a plane strain condition in the direction perpendicular to the axis of
radial symmetry, which is achieved by restraining vertical displacements on the top and bottom
sides of the mesh (see Figure 2). The mesh consists of 35 elements with a total of 178 nodes.
Inspection of Figure 2 also shows that the mesh is finer in the region close to the borehole wall,
where the pressure is applied and where a highly non-linear pattern of displacement is expected.
Cavity expansion is produced by an incremental increase of the pressure applied at the borehole
boundary of the mesh while the outer boundary is free to move horizontally (x-direction). The air
pressure and water pressure are kept constant throughout the test simulating either a drained
unsaturated soil response or an undrained unsaturated soil response that generates no excess
pore pressures.

The outer boundary of the finite element mesh extends to a radius that is 50 times the
pressuremeter radius, here equal to 42mm while the height of the mesh is 10cm. To check that the
chosen position of the outer boundary has no influence on computed results, a suite of preliminary
finite element simulations were performed. Figure 3 shows the variation of displacement in the x-
direction for nodes along the centreline of the mesh due to an increase in total stress at the

8
borehole wall from an initial value of 60kPa to a final value of 160kPa. It is clear that, by using this
mesh size, the effect of the pressure applied at the borehole face is fully attenuated and
deformations become negligible before the outer boundary is reached. Normal stresses in the x-
direction are also shown in Figure 3, which shows that this stress drops quickly to the in-situ value
of 60kPa as the distance from the axis of symmetry is increased, confirming once again that the
outer boundary is sufficiently far from the borehole wall.

As mentioned earlier, a constant suction equal to the initial in-situ value is imposed throughout the
mesh during all finite element simulations of cavity expansion performed in this work. This is
consistent with the experimental observations from Schnaid et al. (2004), who performed pre-bored
pressuremeter tests in a residual unsaturated soil while simultaneously measuring soil suction at
different radial distances around the pressuremeter by means of tensiometers. The results
presented in Schnaid et al. (2004) indicate that, for these particular tests, suction varies negligibly
in the soil region affected by the pressuremeter expansion and it is reasonable to assume that the
initial in-situ value of suction remains unvaried throughout the test.

3.3 Hybrid moving boundary particle swarm optimization (hmPSO)

Due to the complexity of the optimization problem considered in this work, a robust and effective
optimization algorithm is required to obtain the global minimum of the objective function and, hence,
the corresponding set of optimum parameter values. Evolutionary algorithms, such as the genetic
algorithm and PSO, have been applied effectively to optimization problems in geotechnical
engineering (Carrera et al. 2005; Cekerevac et al. 2006; Feng et al. 2006; Levasseur et al. 2007;
Levasseur et al. 2010; Malecot et al. 2004; Schanz et al. 2006). Unlike other optimization methods,
evolutionary algorithms do not require information about the gradient of the objective function,
which makes them particularly suited to inverse analyses where computed values are obtained at
discrete points over the search space by means of numerical (e.g. finite element) simulations of the
chosen boundary value problem.

PSO is inspired by the social behaviour of swarms in which a group of independent particles fly
over the search space trying to achieve a goal by acting in a complex and coordinated way. The
goal is to find the position corresponding to the optimum particle fitness or, in other words, to find
the lowest value of the objective function f(x) in the search space of the vector x whose
components are the model parameters to be selected. The generic ith particle has several
attributes, which are: the current position in the search space xi and the current velocity vi
(measuring the last change of the particle‟s position), the corresponding fitness f(xi), the best

9
position Pi achieved so far by the particle and corresponding fitness f(Pi), the best position Pg
achieved so far across the entire swarm and its corresponding fitness f(Pg).

PSO starts by randomly initializing each particle‟s velocity vi and position xi and calculating the
corresponding fitness f(xi). The initial values of vi and xi are uniformly distributed within their
bounds to improve performance of the algorithm. The bounds of xi are given by the maximum and
minimum vectors x max and x min , as previously explained. Then particles “fly” around the search

space, generation after generation, seeking the optimum fitness. Each generation corresponds to
an iteration of the algorithm where the velocity and position of all particles are updated according to
the following rules:

v ik 1  wk v ik  c1r1 Pi  x i   c2 r2 Pg  x i  (11)

x ik 1  x ik  v ik 1 (12)

where k denotes the generation number , wk is the inertia weight of the current generation, r1 and r2
are two random factors in the interval [0, 1] and c1 and c2 are the „„cognitive” and „„social” weights,
respectively (default values for most problems are c1 = c2 = 2.0, Shi and Eberhart, 1998a). Equation
(11) shows that c1 affects the step size in the direction of the particle best position and c2 affects the
step size in the direction of the global best position. The inertia weight wk controls the momentum
of the particle, which decreases linearly with the generation number according to the following rule
(Shi and Eberhart, 1998b):

wk  wmax  wmin * MaxIter  k  MaxIter  wmin (13)

where MaxIter is the maximum number of generations, taken here equal to 20000 which acts as a
stopping criterion for the algorithm, while wmax and wmin are the maximum and minimum inertia
weights, set in this work equal to 0.75 and 0.4, respectively. The number of particles is set as five
times of the number of unknowns.

For some challenging optimization problems, the classic PSO algorithm suffers from premature
convergence as the algorithm becomes trapped in a position of the search space corresponding to
a local minimum of the objective function, and stops exploring other areas of the search space
where the global minimum may lie. In addition, the classic PSO algorithm needs a large number of
generations to find the optimal solution, which makes it computationally expensive. In order to
overcome these weaknesses, a novel evolutionary algorithm, developed by the authors (Zhang et
al. 2009a, 2009b) and named the “hybrid moving boundary particle swarm optimization” (hmPSO)
algorithm, has been used in this work. The algorithm is the result of hybridization of the classic

10
PSO algorithm (Kennedy and Eberhart 1995) with a Nelder–Mead local search method. The
hmPSO algorithm combines the effectiveness of the classic PSO algorithm for global search with
the speed of the Nelder–Mead method for local search. It also includes two distinct particle swarms,
one flying over the „„global” domain (exploring the entire search space) and the other over the
„„local” sub-domain (centred on the current best particle position). Further details of the hmPSO
algorithm can be found in Zhang et al. (2009a).

In order to speed up computation the hmPSO algorithm is implemented in an asynchronous

parallel form based on the client-server model, which takes advantage of multiple client processors
running in parallel. The server-client model consists of one server and a number of client
processors divided into two distinct groups, i.e. the group of particle clients responsible for
evaluating the objective function at different particle positions and the group of local search clients
responsible for searching in the local sub-domain. Particle clients request information directly from
the server and do not communicate among themselves. On the other hand, local search clients
interact according to a master-slave model. The master is responsible for managing and
coordinating all slaves while the server only communicates with the master of the local search
client group. The server is responsible for storing and managing shared data, listening to queries
by all clients and returning the relevant responses. Further details of the asynchronous parallel
implementation of the hmPSO algorithm can be found in Zhang et al. (2009a). The values of the
hmPSO algorithm parameters used in this work are the same as those employed in the simulation-
based optimization problem presented in Zhang et al. (2009a).

4. VERIFICATION

The performance of the proposed hmPSO algorithm in terms of accuracy, efficiency and
robustness has already been evaluated against some well-defined mathematical optimization
problems and a geotechnical infiltration problem in a previous paper (Zhang et al. 2009a). In this
section, we verify the effectiveness of the parallel hmPSO algorithm by determining the values of
BBM parameters by inverse analysis of artificial pressuremeter tests which have been simulated by
means of a numerical (FE) model using the same constitutive model.

A set of pressuremeter expansion curves at constant suction is first generated from FE models of
infinitely long cylindrical cavities expanding in an unsaturated soil modelled by the BBM using a set
of pre-defined parameter values. The goal is to test the ability of hmPSO algorithm to find the same
set of pre-defined BBM parameter values by inverse analysis of these simulated pressuremeter
expansion curves. The advantage of using simulated “measurements” instead of real ones is that
the former can be exactly matched by the optimization algorithm and the value of the objective

11
function corresponding to the global optimum is therefore zero, which is ideal as a verification
example.

Three verification tests are carried out with different numbers of parameters to be identified and
different search bounds. In particular, the numerical example of a pressuremeter test presented in
Gallipoli (2000) is modified in this work by neglecting the finite length of the pressuremeter and
using instead a simpler one-dimensional model which reproduces the expansion of an infinitely
long cylindrical cavity. The BBM parameters used to produce the simulated “field data” are equal to
those used by Gallipoli (2000) (with the only exception of shear modulus G) and are summarized in
Table 1a. Three cavity pressure - cavity strain curves, at differing suctions, are produced by means
of FE simulations using the pre-defined BBM parameters (Figure 4). Both air and water pressures
are kept constant throughout the simulated cavity expansion with the water pressure taken as 0, -
100kPa and -200kPa (depending on the particular simulation considered) while the air pressure is
assumed to be atmospheric. Other specifications for the example are listed in Table 1b.

For the optimization problem, the objective function is defined as a measure of the cavity strain
difference at the same cavity pressure between computations and “measurements” as follows:

  
N
f x     cm,i
s 2
c ,i (14)
i 1
s
where  c,i are the cavity strains from the finite element simulations performed by the optimization
m
algorithm using different sets of trial parameters and  c,i are the cavity strains from the

“measurements” calculated using the pre-defined BBM parameters. Each sampling point where the
two are compared is indexed i. Note that all curves at different suctions are simultaneously
matched during optimization so N is the total number of points on all three curves of Figure 4.

The first verification test involves simultaneous identification of 6 parameters in the BBM, i.e. x =
{M, k, κ, r, β, pc}T. This reduced set is chosen since, as indicated in Section 2, parameters G, λ(0)
and  s can be determined from laboratory tests with relative ease and may therefore be readily

available in a real case. N (0) is not required given that the alternative assumption of an initial void
ratio, i.e. a void ratio at the start of the test, is made. The hardening parameter p0(0) is excluded
from the set of unknown parameters because the soil is assumed to be normally consolidated in
this case. In particular, after initialization of the unknown BBM parameters, the value of the
hardening parameter p0(0) is calculated assuming that the initial stress state lies on the yield locus
with p0(s) and p0(0) being calculated according to Equations 1 and 2 respectively. The chosen
parameter bounds are listed in Table 2 and a total of 30 particles are used for the search. The
solution, shown in Table 3, is found after 4 local searches and 512757 objective function

12
evaluations by particles. Results show there to be little difficulty in simultaneously finding the 6
unknown BBM parameters with very high accuracy and a corresponding value of the objective
function of 2.1 10 6 .

In the second verification test we attempt to identify simultaneously 8 parameters of the BBM
model, i.e x = {G, M, k, κ, r, λ(0), β, pc}T. The parameter bounds are again listed in Table 2. This
case is much more challenging than the first verification test, which required optimization of only 6
BBM parameters, because a larger number of local minima exist in the search space. In this case,
we find that 7 parameters can be identified with satisfactory accuracy, as shown in Table 3, with a
corresponding value of the objective function equal to 1.5  10 4 . Despite the relative inaccurate
estimate of parameter pc, the use of these parameter values in a FE simulation results in very close
agreement of the simulations to the “measured” behaviour as shown in Figure 5. In a third and final
verification test, smaller parameter bounds (as shown in Table 2) are used keeping everything else
the same as in the second verification test. This is termed a two-stage optimization procedure in
Zhang et al. (2008). All 8 parameters can be identified more accurately than in the second
verification test as shown in Table 3, with a corresponding value of the objective function of
3.4  10 6 .

The results from these three tests serve to verify the efficiency of the hmPSO algorithm in the
determination of BBM parameter values from inverse analysis of observed data. At least 6
parameters can be identified with high accuracy but, if a multi-stage procedure is adopted where
parameter bounds are successively tightened (e.g. by making use of laboratory test results), the
number of parameters that can be accurately identified increases. In the next section we apply the
hmPSO algorithm to the case where the “measured” data do not come from numerical simulations,
as above, but from real tests performed at a given site. In that case, the goal therefore becomes
finding the BBM parameter values than could be used for geotechnical modelling at that site.

5. IDENTIFICATION OF BBM PARAMETERS USING FIELD DATA

Field data are obtained from Suction Monitored Pressuremeter Tests (SMPTs), which employ
tensiometers to measure suction in the proximity of the borehole while pressuremeter expansion
takes place (Schnaid et al. 2004). As mentioned earlier, the finite element simulations of
pressuremeter tests performed in this work assume that suction remains constant over the entire
mesh throughout the pressuremeter expansion. This is because readings from field tensiometers
by Schnaid et al. (2004) gave a clear indication that suction remained constant around the cavity.
While it is acknowledged that the hypothesis of constant suction during pressuremeter expansion
is an approximation, and suction might not remain perfectly constant close to the pressuremeter

13
face, it is this very approximation that allows us to avoid the complication of performing a fully
coupled analysis. A fully coupled analysis would in fact require introduction of a water retention
curve together with a permeability relationship and, consequently, it would need calibration of
additional model parameters.

In addition, experimental evidence from laboratory compression/shearing tests performed on

unsaturated soil samples at constant water content (i.e. undrained) indicate that, when degree of
saturation is low, soil suction tends to remain constant during deformation. This further justifies the
assumption of constant suction during the finite element simulation of pressuremeter tests,
especially in those cases where tests are performed in relatively dry soils (i.e. at a low degree of
saturation) because, in such cases, suction is expected to remain approximately constant even if
the cavity expansion takes place under undrained conditions.

In this section, field data from the pressuremeter tests performed by Schnaid et al. (2004) are used
to investigate the capability of the proposed methodology to identify BBM parameter values
through inverse analysis of the measured cavity pressure-cavity strain curve. The location of the
field tests described in Schnaid et al. (2004) is the city of Porto Alegre, Brazil. Testing was
undertaken in weak erodible yellow granite (YG) residual soil at a depth of 2m. Table 4 lists some
properties of the soils where G, κ and λ(0) were obtained from laboratory tests on saturated
undisturbed samples. Tensiometers were installed at the same depth of the pressuremeter tests at
30 and 60 cm from the centre of the borehole to measure suction during the tests. Measured
suctions remained approximately constant throughout the expansion of the pressuremeter. Three
cavity pressure – cavity strain curves from Schnaid et al. (2004), with different suction levels
(43kPa, 33kPa and 0kPa), are used here (note that the curve labelled as 66kPa in Figure 21 of the
Schnaid et al. (2004) paper should be labeled as 33kPa). The experimental curves were obtained
by digitization of the original figures and subsequent fitting of cubic splines (see Figure 6 for a
comparison of the digitized data and the fitted spline curves). Discrete points on the fitted cubic
splines were retrieved at intervals of cavity pressure of 10kPa starting from 0kPa, with the chosen
sampling interval matching that of the corresponding FE simulation. These interpolated points were
then used when evaluating the objective function in the hmPSO algorithm. In particular, the
number of increments of cavity pressure and the final cavity pressure, which define the points
where the objective function is evaluated, were 55/550kPa, 35/350kPa and 35/350kPa
corresponding to the three suction levels 43kPa, 33kPa and 0kPa, respectively.

Five optimization scenarios were investigated to find different sets of BBM parameter values under
different assumptions (see Table 5 for full details). The initial net stress state is assumed to be
isotropic in all simulations with q = 0 and p = 65kPa. For Cases 1 and 4, where the initial value of
the hardening parameter p0(0) is not part of the fitting process, the value of p0(0) is calculated by

14
imposing an initial soil state defined by net stress and suction lying on the yield locus after
initialization of the unknown BBM parameters. The values of p0(s) and p0(0) are therefore calculated
according to Equations 1 and 2 respectively. For the remaining Cases 2, 3 and 5, where p0(0) is
part of the fitting process, the soil is assumed to be over-consolidated and p0(0) is initialized by the
optimization algorithm. In these three cases, the range of variation of p0(0) is [65kPa, 220kPa] and,
given that the initial isotropic net stress is always equal to 65 kPa (regardless of suction level), the
soil state at the start of the simulation falls always inside the yield locus in (p,q,s) space.

The bounds for all five cases (Table 6) are set large enough to cover a wide range of soils. In Case
1, the values of 8 BBM parameters are sought over a large search space, which makes this a
particularly challenging problem. In Cases 2, 3 and 4 eight or nine parameters are sought but the
minimum and maximum bounds of some parameters are closer compared to Case 1. For example,
the minimum bounds of G, κ, λ(0) are 20% less than the corresponding laboratory-measured values
while the maximum bounds are 20% larger than corresponding laboratory-measured values, thus
making these three cases not as challenging as Case 1. Both Cases 3 and 4 assume a non-
associative flow rule.

The above cases have been chosen to demonstrate the ability of the hmPSO algorithm to deal with
a range of different requirements in terms of number of sought parameters, availability of
information from laboratory tests and material behaviour. The best sets of parameter values
identified for each of the five cases are listed in Table 7. The cavity pressure - cavity strain curves
calculated by the FE model by using these optimal parameter values are shown in Figures 7 to 11.

For Case 1 the best set of parameter values is obtained in correspondence with a minimum value
of the objective function f(x) = 0.0650. This is the smallest value of the objective function in all five
cases and corresponds to a match between computed and field curves that is visually very good
(Figure 7). However, for three of these parameters (M, k, r), the identified values coincide with one
of the specified bounds. Therefore, this solution, while apparently good in optimization terms,
yields parameter values that are not very realistic. For Case 2 the best solution is obtained
corresponding to a minimum value of the objective function f(x) = 0.0720. Five of the identified
parameter values are close to the bounds, though the match between field and simulated curves is
once again visually good (Figure 8). The minimum value of the objective function achieved in Case
3, for which a non-associated flow rule is assumed, is f(x) = 0.0696, i.e located between the
minima obtained for Cases 1 and 2, respectively. In Case 3 we also identify three parameter
values (G, k, λ(0)) which are close to the set bounds. For Case 4, the minimum value of the
objective function f(x) = 2.5993 is much higher than for the preceding three cases and the match
between the pressure-expansion curves measured in the field and those calculated from the finite
element models using the identified BBM parameter values is correspondingly poor (Figure 10). In

15
this case the initial value of the hardening parameter p0(0) is not found by the optimization
algorithm but is prescribed by setting the soil state to be normally consolidated at the start of
pressuremeter expansion, i.e. by imposing that the initial soil state, defined in terms of net stress
and suction, lies on the yield locus. In Case 5 the minimum value of the objective function f(x) =
0.1094 is of the same order of magnitude as in Cases 1, 2 and 3. In this case, G, κ and λ(0) are
assumed known from laboratory tests. Although the minimum value of the objective function
achieved is greater in Case 5 than in Cases 1 to 3, the identified values of the BBM parameters are
more realistic than in the other cases. The match between field and computed curves in Case 5 is
also visually good with parameters that are sensible from an engineering point of view (Figure 11).

While Cases 1 to 3 result in lower minima of the objective function and, hence, better solutions in
terms of optimization than Cases 4 and 5, the identified BBM parameter values appear less
realistic. In this respect, it is important to bear in mind that the ultimate goal of the proposed
optimization procedure is the provision of a set of parameter values for geotechnical design and it
is therefore more important to attain realistic values for the selected parameter values than the
global minimum of the objective function. It is also worth remembering that, as highlighted above, it
is not possible for the objective function to be equal to zero when dealing with field data as these
data are not perfectly consistent with the idealized BBM soil behaviour. A minimum value of the
objective function greater than zero is therefore not necessarily indicating an inadequate set of
parameter values and engineering judgement should always be used in conjunction with the
optimization procedure. For example, in Case 5, where the values of some parameters are
assumed to be known from laboratory tests, the optimization procedure returns values for the
remaining parameters that are both reasonable from the engineering point of view and well inside
the set bounds, despite the corresponding minimum of the objective function not being the lowest.
For Case 5, the top 5 best solutions are listed in Table 8 and these could all be suitably used in
engineering practice.

An important finding from the point of view of computational efficiency is that the size of the chosen
parameter bounds plays a major role in both the quality of the solution and the speed with which
this solution is obtained. A wider search space contains potentially better solutions, however
practical bounds should always be imposed based on the physical meaning of model parameters
and taking into account soil properties.

Given that the optimization problem is ill-posed and the solution is non-unique due to its over-
determination, a sensible choice of the best set of parameter values should always be backed up
with standard knowledge of the geological properties of the field and the physical meaning of
model parameters. In the BBM, for example, laboratory tests can be used to reduce the number of
unknown parameters to set smaller bounds, which can enhance accuracy of the solution.

16
Parameters such as G, λ(0),  s ,  and M can be determined from laboratory tests with relative

ease, while parameters pc, p0(0), which are difficult to obtain from laboratory tests, can be identified
by the optimization algorithm.

6. CONCLUSIONS

In this paper, an optimization algorithm for identifying BBM parameters from field pressuremeter
tests is proposed. The proposed hmPSO algorithm has been first verified against “artificial” data
produced by FE simulations. This has shown the capability of the procedure to accurately identify
parameters, the degree of accuracy depending on the numbers of parameters to be identified and
the bounds of the search space. For realistic field problems, where the input data come from
pressuremeter tests, it has been shown that satisfactory results are obtained providing, where
feasible, the number of parameters to be identified is reduced with the aid of laboratory tests. Six
BBM parameters can be safely identified, and in some cases a larger number too. Attempts to
identify too many parameters at once, can however prove risky due to the non-uniqueness of the
solution found by the algorithm. The combination of laboratory tests, field pressuremeter tests and
the hmPSO algorithm presents a practical way to determine parameter values for geotechnical
modelling in unsaturated soils. Despite the need to use a parallel implementation, the proposed
methodolgy is likely to be a competitive alternative to expensive laboratory tests for the
determination of all parameters. This is particularly the case if the anticipated rise in the use of
GPU-based parallel systems is rolled out.

ACKNOWLEDGEMENTS

The authors acknowledge the support of the UK Engineering and Physical Science Research
Council (EPSRC) under grant no. EP/C526627/1 in undertaking this research. Support from the
“Hundred Talents Programme” of the Chinese Academy of Sciences is also acknowledged in
preparation of this paper. The authors also thank the reviewers for their valuable comments.

REFERENCES

1. Alonso EE, Gens A, Josa A. A constitutive model for partially saturated soils. Géotechnique
1990; 40(3):405-430.
2. Carrera J, Alcolea A, Medina A, Hidalgo J, Slooten L. Inverse problem in hydrogeology.
Hydrogeological Journal 2005; 13(1):206–222.
3. Carter JP, Booker JR, Yeung SK. Cavity expansion in frictional cohesive soils. Géotechnique
1986; 36(3):349-358.

17
4. Cekerevac C, Girardin S, Klubertanz G, Laloui L. Calibration of an elasto-plastic constitutive
model by a constrained optimisation procedure. Computers and Geotechnics 2006; 33(8):432-
443.
5. Cudmani R, Osinov VA. The cavity expansion problem for the interpretation of cone
penetration and pressuremeter tests. Canadian Geotechnical Journal 2001; 38(3):622-638.
6. Cui YJ, Delage P. Yielding and plastic behaviour of an unsaturated compacted silt.
Géotechnique 1996; 46(2):291-311.
7. Eberhart R, Kennedy J. A new optimizer using particle swarm theory. In: Proc. of the Sixth
International Symposium on Micro Machine and Human Science, Nagoya; 1995, p39–43.
8. Fahey M, Carter, JP. A finite element study of the pressuremeter test in sand using a nonlinear
elastic plastic model. Canadian Geotechnical Journal 1993; 30(2):348-362.
9. Feng XT, Chen BR, Yang C, Zhou H, Ding X. Identification of visco-elastic models for rocks
using genetic programming coupled with the modified particle swarm optimization algorithm.
International Journal of Rock Mechanics and Mining Sciences 2006; 43(5):789–801.
10. Fredlund DG, Xing A, Fredlund MD, Barbour SL. Relationship of the unsaturated soil shear
strength to the soil-water characteristic curve. Canadian Geotechnical Journal 1996; 33(
3):440-448.
11. Gallipoli D. Constitutive and numerical modelling of unsaturated soils. PhD thesis, University of
Glasgow, U.K. 2000.
12. Gallipoli D, Gens A, Sharma R, Vaunat J. An elastoplastic model for unsaturated soil
incorporating the effects of suction and degree of saturation on mechanical behaviour.
Geotechnique 2003; 53(1):123-135.
13. Gallipoli D, Toll DG, Augarde CE, Tarantino A, De Gennaro V, Vaunat J, Wheeler S, Mancuso
C. The MUSE Network: sharing research expertise on unsaturated soils across Europe. 4th
International Conference on Unsaturated Soils, Arizona, 2006, p2075-2085.
14. Gallipoli D, D‟Onza F and Wheeler SJ. A sequential method for selecting parameter values in
the Barcelona Basic Model. Canadian Geotechnical Journal 2010; 47(11): 1175-1186.
15. Hsieh YM, Whittle AJ, Yu HS. Interpretation of pressuremeter tests in sand using advanced soil
model. ASCE Journal of Geotechnical and Geoenvironmental Engineering, 2002; 128(3):274-
278.
16. Kennedy J, Eberhart,R. Particle swarm optimization. Proc. IEEE International Conference on
Neural Networks, Piscataway, 1995, p1942-1948.
17. Levasseur S, Malecot Y, Boulon M, Flavigny E. Soil parameter identification using a genetic
algorithm. International Journal for Numerical and Analytical Methods in Geomechanics 2007;
32(2):189–213,.
18. Levasseur S, Malecot Y, Boulon M, Flavigny E. Statistical inverse analysis based on genetic
algorithm and principal component analysis: Applications to excavation problems and
pressuremeter tests. International Journal for Numerical and Analytical Methods in
Geomechanics 2010; 34(5):471–491.
19. Malecot Y, Flavigny E, Boulon, M. Inverse analysis of soil parameters for finite element
simulation of geotechnical structures: pressuremeter test and excavation problem. Proc. of the
Symposium on Geotechnical Innovations, Essen, 2004, p659–675.
20. Nelder J, Mead R. A simplex method for function minimization. Computer Journal 1965;
7(4):308–313.
21. Russell AR, Khalili N. A Unified bounding surface plasticity model for unsaturated soils.
International Journal of Numerical and Analytical Methods in Geomechanics 2004; 30(3):181-
212.

18
22. Schnaid F, Kratz de Oliveira LA, Gehling WYY. Unsaturated constitutive surfaces from
pressuremeter tests. ASCE Journal of Geotechnical and Geoenvironmental Engineering 2004;
130(2):174-185.
23. Schanz T, Zimmerer MM, Datcheva M, Meier J. Identification of constitutive parameters for
numerical models via inverse approach. Felsbau 2006; 24(2):11–21.
24. Shi Y, Eberhart R. Parameter selection in particle swarm optimization. Proc. of the 1998
Annual Conference on Evolutionary Computation, New York, 1998a, p591-600.
25. Shi Y, Eberhart R. A modified particle swarm optimizer. IEEE International Conference on
Evolutionary Computation, Anchorage 1998b, p69-73.
26. Simoni L, Schrefler BA. Parameter identification for a suction dependent plasticity model.
International Journal for Numerical and Analytical Methods in Geomechanics 2001; 25(3):273-
288.
27. Vaunat J, Cante JC, Ledesma A, Gens A. A stress point algorithm for an elastoplastic model in
unsaturated soils. International Journal of Plasticity 2000; 16 (2):121-141.
28. Wheeler SJ, Sivakumar V. An elasto-plastic critical state framework for unsaturated soil.
Géotechnique 1995; 45(1):35-53.
29. Yu HS, Houlsby GT. Finite cavity expansion in dilatant soils: Loading analysis. Géotechnique
1991; 41(2):173-183.
30. Yu HS, Houlsby GT. A large strain analytical solution for cavity contraction in dilatant soils.
International Journal for Numerical and Analytical Methods in Geomechanics. 1995;
19(11):793-811
31. Zhang Y, Augarde CE, Gallipoli D. Identification of hydraulic parameters for unsaturated soils
using particle swarm optimization. Proc. of the 1st European Conference on Unsaturated Soils,
Durham, 2008, p765-771.
32. Zhang Y, Gallipoli D. Development of an object-oriented parallel finite element code for
unsaturated soils. 2nd International Conference on Mechanics of Unsaturated Soils, Weimar,
2007, p177-186.
33. Zhang Y, Gallipoli D, Augarde CE. Parallel computing of unsaturated soils using element-by-
element and domain decomposition methods. 3rd Asian Conference on Unsaturated Soils,
Nanjing, 2007, p425-430.
34. Zhang Y, Gallipoli D, Augarde CE. Simulation-based calibration of geotechnical parameters
using parallel hybrid moving boundary particle swarm optimization. Computers and
Geotechnics 2009a; 36 (4): 604-615.
35. Zhang Y, Gallipoli D, Augarde CE. Parallel Hybrid Particle Swarm Optimization and
Applications in Geotechnical Engineering. In: Cai Z, Li Z, Kang Z, Liu Y. Lecture Notes in
Computer Science: Advances in Computation and Intelligence, Heidelberg: Springer Berlin
5821; 2009b, p.466-475.

19
Figure 1: Flow chart of proposed algorithm for soil parameters identification from pressuremeter
tests.

Figure 2: Finite element mesh for pressuremeter test simulations.

20
Figure 3: Results from preliminary sensitivity tests on FE mesh. Plot of displacement and normal
stress in the radial direction at the end of loading.

Figure 4: “Measured” cavity pressure – cavity strain curves produced by using pre-defined BBM
parameter values.

21
Figure 5: Verification test 2: close agreement between predictions (using identified BBM parameter
values) and “measured” cavity pressure – cavity strain curves.

Figure 6: Curves from field data and fitted cubic splines used for the optimization procedure.

22
Figure 7: Comparison of measurements with optimal solution for Case 1.

Figure 8: Comparison of measurements with optimal solution for Case 2.

23
Figure 9: Comparison of measurements with optimal solution for Case 3.

Figure 10: Comparison of measurements with optimal solution for Case 4.

24
Figure 11: Comparison of measurements with optimal solution for Case 5.

25
Table 1a: BBM parameters used during calculation of the “measured” curves for the verification
tests.

x G M k κ λ(0) r  pc s N(0)
-1
MPa Pa MPa
Value 3.019 0.900 0.500 0.025 0.130 1.500 1.000E-5 2.000 0.020 1.660

Table 1b: Specifications for the verification tests.

Parameter Adopted value

Initial vertical effective stress 60kPa

Initial horizontal effective stress 60kPa

Initial gas pressure atmospheric pressure

Initial water pressure 0

-100kPa
-200kPa
Applied increase of cavity 130kPa over 900 s
pressure

Loading discretization 50 increments over 18 s

26
Table 2: Search bounds for BBM parameters in the verification tests.

Parameter First and second verification tests Third verification test

Minimum value Maximum value Minimum value Maximum value

G (MPa) 1.000E+05 1.000E+07 2.500E+06 3.500E+06

M 0.100 1.400 0.850 0.950

k 0.100 0.700 0.450 0.550

 0.010 0.100 0.015 0.025

λ(0) 0.050 0.250 0.080 0.140

r 1.050 1.800 1.300 1.600

β (Pa-1) 1.000E-06 1.000E-04 8.000E-06 2.000E-05

pc (MPa) 1.000E+04 1.000E+07 1.000E+04 1.000E+07

27
 Table 3: Results for the verification tests.
X G M k κ λ(0) r  pc p0(0) s N(0) f(x)
-1
MPa Pa MPa kPa
Predefined 3.019 0.900 0.500 0.025 0.130 1.50 1.00E-5 2.000 NA 0.02 1.66 NA
value
Identified value NA 0.900 0.500 0.025 NA 1.50 1.00E-5 2.000 NA NA NA 2.1E-06
st
(1 verification
test)
Identified 3.013 0.901 0.498 0.023 0.123 1.46 1.02E-5 1.053 NA NA NA 1.5E-04
nd
value(2
verification test)
Identified 3.020 0.900 0.500 0.025 0.130 1.50 1.00E-5 1.990 NA NA NA 3.4E-06
th
value(3
verification test)

Table 4: Properties of YG soils (Schnaid et al. 2004).

Initial Initial Initial yield

shear horizontal cavity
suction
modulus stress pressure κ λ(0)
(kPa)
G P0M Pf
(MPa) (kPa) (kPa)

43 2.1 62 382 - -

43 2.6 67 272 - -

0 1.8 65 235 0.023 0.258

33 2.6 79 243 - -

28
Table 5: Details of the five cases for the field problem.

Case Unknowns Assumptions

Associated flow rule;

8 parameters (p0(0) calculated by assuming that soil is
1
x={G, M, k, κ, λ(0) ,r, β, pc} initially normally consolidated);
larger search space
Associated flow rule;
9 parameters
2 p0(0) to be identified;
x={G, M, k, κ, λ(0), r, β, pc, p0(0)}
smaller search space

9 parameters Non-associated flow rule;p0(0) to be identified;

3
x={G, M, k, κ, λ(0), r, β, pc, p0(0)} smaller search space;

Non-associated flow rule;

8 parameters (p0(0) calculated by assuming that soil is
4
x={G, M, k, κ, λ(0), r, β, pc} initially normally consolidated);
smaller search space

Non-associated flow rule;

6 parameters
5 p0(0) to be identified;
x={M, k, r, β, pc, p0(0)}
G, κ, λ(0) determined from laboratory tests

29
Table 6: BBM parameter bounds for the field problem.

Parameter names Cases Ranges

Case 1 [1.000E+05, 1.000E+07] Pa

G
Case 2,3,4 [1.840E+06, 2.760E+06] Pa
Case 1 [0.800, 2.000]
M
Case 2,3,4,5 [1.000, 1.800]
k Case 1,2,3,4,5 [0.100, 0.700]
Case 1 [0.010, 0.100]
κ
Case 2,3,4 0.018, 0.028
Case 1 [0.050, 0.350]
λ(0)
Case 2,3,4 [0.200, 0.310]
r Case 1,2,3,4,5 [0.400, 1.800]
β Case 1,2,3 ,4, 5 [1.000E-06, 1.000E-04] Pa-1
pc Pa
Case 1,2,3,4,5 [10.000, 5.000E+08]
p0(0) Pa
Case 2,3,5 [6.500E+04, 2.200E+05]

30
Table 7: Results for the field problem.

Case G M k κ λ(0) r  pc p0(0) f(x)

-1
MPa Pa MPa kPa
1 2.289 2.000 0.100 0.0240 0.256 1.800 6.934E-05 0.1649 0.0650

2 2.751 1.791 0.699 0.0275 0.309 1.386 9.956E-05 1.532 171.90 0.0720

3 2.759 1.657 0.699 0.0187 0.307 0.780 2.789E-05 0.0129 195.80 0.0696

4 2.428 1.603 0.147 0.0290 0.157 1.298 5.836E-05 57.070 2.5993

5 1.456 0.500 1.308 8.581E-06 9.517 84.020 0.1094

31
 Table 8: Best solutions for Case 5, x = {M, k, r, β, pc, p0(0)}

Objective
Solution
Parameter values function
number
minimum value

1 1.456, 0.500, 1.308, 8.581E-06Pa-1, 9.517E+06MPa, 8.402E+01kPa 0.1094

2 1.470, 0.541, 1.415, 9.659E-06Pa-1, 6.458E+06MPa, 7.722E+01kPa 0.1115

3 1.520, 0.358, 1.319, 8.007E-06Pa-1, 3.570E+07MPa, 6.954E+01kPa 0.1148

4 1.520, 0.463, 1.428, 2.986E-05Pa-1, 7.282E+05MPa, 7.888E+01kPa 0.1165

5 1.410, 0.372, 1.497, 1.126E-05Pa-1, 1.318E+06MPa, 8.806E+01kPa 0.1482

32

View publication stats