INSTALLATION EFFECTS IN GEOTECHNICAL ENGINEERING

PROCEEDINGS OF THE INTERNATIONAL CONFERENCE ON INSTALLATION EFFECTS IN

GEOTECHNICAL ENGINEERING, ROTTERDAM, THE NETHERLANDS, 24–27 MARCH 2013

Installation Effects in Geotechnical

Engineering

Editors
Michael A. Hicks
Section of Geo-Engineering, Delft University of Technology, Delft, The Netherlands

Jelke Dijkstra
Section of Geo-Engineering, Delft University of Technology, Delft, The Netherlands

Marti Lloret-Cabot
Centre for Geotechnical and Materials Modelling, University of Newcastle, Newcastle,
Australia & Section of Geo-Engineering, Delft University of Technology, Delft,
The Netherlands

Minna Karstunen
Department of Civil and Environmental Engineering, Chalmers University of
Technology, Gothenburg, Sweden & University of Strathclyde, Glasgow, UK
CRC Press/Balkema is an imprint of the Taylor & Francis Group, an informa business

© 2013 Taylor & Francis Group, London, UK

Typeset by V Publishing Solutions Pvt Ltd., Chennai, India

Printed and bound in Great Britain by CPI Group (UK) Ltd, Croydon, CR0 4YY.

All rights reserved. No part of this publication or the information contained herein may be reproduced,
stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, by pho-
tocopying, recording or otherwise, without written prior permission from the publisher.

Although all care is taken to ensure integrity and the quality of this publication and the information
herein, no responsibility is assumed by the publishers nor the author for any damage to the property or
persons as a result of operation or use of this publication and/or the information contained herein.

Published by: CRC Press/Balkema

P.O. Box 11320, 2301 EH Leiden, The Netherlands
e-mail: Pub.NL@taylorandfrancis.com
www.crcpress.com – www.taylorandfrancis.com

ISBN: 978-1-138-00041-4 (Hbk + CD-ROM)

ISBN: 978-0-203-74654-7 (eBook)
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Table of contents

Preface ix
Committees xi
Keynote speakers xiii

Computational methods
Large deformation analysis of cone penetration testing in undrained clay 1
L. Beuth & P.A. Vermeer
Adaptive Mesh Refinement for strain-softening materials in geomechanics 8
C.E. Heaney, R.B.J. Brinkgreve, P.G. Bonnier & M.A. Hicks
A dynamic material point method for geomechanics 15
I. Jassim, C. Coetzee & P.A. Vermeer
Pile penetration simulation with Material Point Method 24
L.J. Lim, A. Andreykiv & R.B.J. Brinkgreve
Coupling triangular plate and volume elements in analysis of geotechnical problems 31
S. Tan, M.A. Hicks & A. Rohe
Fracture growth in heterogeneous geomaterials 38
P.J. Vardon & J.D. Nuttall

Constitutive modelling
Modelling rate-dependent behaviour of structured clays 43
M. Karstunen, N. Sivasithamparam, R.B.J. Brinkgreve & P.G. Bonnier
FEM simulation of large vertical deformations caused by land subsidence and verification
of the results by using radar interferometry techniques 51
C. Loupasakis, D. Rozos, F. Raspini & S. Moretti
Numerical modeling of fracturing in soil mix material 58
G. Van Lysebetten, A. Vervoort, N. Denies, N. Huybrechts, J. Maertens,
F. De Cock & B. Lameire
A 3D practical constitutive model for predicting seismic liquefaction in sands 65
A. Petalas, V. Galavi & R.B.J. Brinkgreve
Comparison of two anisotropic creep models at element level 72
N. Sivasithamparam, M. Karstunen, R.B.J. Brinkgreve & P.G. Bonnier

Installation effects
Displacement pile installation effects in sand 79
A. Beijer Lundberg, J. Dijkstra & A.F. van Tol
Cyclic jacking of piles in silt and sand 86
F. Burali d’Arezzo, S.K. Haigh & Y. Ishihara
Influence of installation procedures on the response of capacitance water content sensors 92
M. Caruso, F. Avanzi & C. Jommi

v
The load capacity of driven cast in-situ piles derived from installation parameters 99
D. Egan
On the numerical modelling and incorporation of installation effects of jacked piles:
A practical approach 104
H.K. Engin, R.B.J. Brinkgreve & A.F. van Tol
Analytical and laboratory study of soil disturbance caused by mandrel driven prefabricated
vertical drains 111
A. Ghandeharioon
CEL: Simulations for soil plugging, screwed pile installation and deep vibration compaction 118
J. Grabe, S. Henke, T. Pucker & T. Hamann
Towards a framework for the prediction of installation rate effects 128
S. Robinson & M.J. Brown
Rate dependent shear strength of silt at low stresses 135
S. te Slaa & J. Dijkstra
Control of excess pore pressure development during pile installations in soft sensitive clay 140
T. Tefera, G. Tvedt & F. Oset
Seabed pipelines: The influence of installation effects 146
D.J. White

Offshore construction and foundations

LDFE analysis of installation effects for offshore anchors and foundations 162
L. Andresen & H.D.V. Khoa
Investigation into the effect of pile installation on cyclic lateral capacity of monopiles 169
T. de Blaeij & J. Dijkstra
Development of a coupled FEM-MPM approach to model a 3D membrane with an application
of releasing geocontainer from barge 176
F. Hamad, C. Moormann & P.A. Vermeer
Leg penetration assessments for self-elevating tubular leg units in sand over clay conditions 184
D.A. Kort, S. Raymackers, H. Hofstede & V. Meyer
Investigating the scales of fluctuation of an artificial sand island 192
M. Lloret-Cabot, M.A. Hicks & J.D. Nuttall

Soil improvement
Volume averaging technique in numerical modelling of floating deep mixed columns in soft soils 198
P. Becker & M. Karstunen
Comparison between theoretical procedures and field test results for the evaluation
of installation effects of vibro-stone columns 205
E. Carvajal, G. Vukotić, J. Castro & W. Wehr
Numerical analyses of stone column installation in Bothkennar clay 212
J. Castro, M. Karstunen, N. Sivasithamparam & C. Sagaseta
Execution of Springsol® deep mixed columns: Field trials 219
S. Melentijevic, F. Martin & L. Prieto
A method of modelling stone column installation for use in conjunction with unit cell analyses 226
B.G. Sexton & B.A. McCabe
Cement grout filtration in non-cohesive soils 233
X.A.L. Stodieck & T. Benz
The undrained cohesion of the soil as a criterion for column installation with a depth vibrator 241
J. Wehr

vi
Soil-structure interaction
Modeling of rock fall impact using Discrete Element Method (DEM) 245
G. Grimstad, O. Melhus, S. Degago & R. Ebeltoft
Investigation into the factors affecting the shaft resistance of driven piles in sands 252
D. Igoe, K. Gavin & L. Kirwan
Monitoring and risk assessment in EPB TBM’s in urban environments: High speed railway
tunnel Sants-Sagrera running next to Sagrada Familia Basilica (World Heritage) 258
J.E. Paris Fernández & J. Gómez Cabrera
Effect of roughness on keying of plate anchors 264
D. Wang, C. Han & C. Gaudin

Author index 271

vii
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Preface

The partners of the European project GEO-INSTALL extend a warm welcome to all participants of the
International Conference on Installation Effects in Geotechnical Engineering (ICIEGE). This is the closing
conference of GEO-INSTALL (FP7/2007-2013, PIAG-GA-2009-230638), an Industry-Academia Path-
ways and Partnerships project funded by the European Community from the 7th Framework Programme.
Infrastructure construction involves the installation of structural elements, such as piles and various
ground improvement techniques for soils and rocks. The installation process itself can be quasi-static (for
example jacked piles) or dynamic (vibratory methods, such as stone columns and driven piles), and generally
involves very large deformations and changes in pore pressure. The fact that natural soils are complex geo-
materials, exhibiting structure and rate-dependent behaviour, makes analysis of such problems yet more chal-
lenging. In particular, the influence of installation on key design parameters, such as mobilised strength at the
soil-structure interface and soil stiffness, is difficult to quantify and, as yet, impossible to model. Numerical
analyses using the standard Finite Element Method (FEM) are unable to produce accurate descriptions of
large deformation problems due to excessive mesh distortions and novel techniques need to be developed.
The aim of the conference is to provide an international forum for presenting the latest developments
in monitoring, analysing and managing installation effects in geotechnical engineering. Active discussion
on important topics will be facilitated through invited keynote lectures, which set the scene for the main
theme of the conference. In addition, the partners of GEO-INSTALL will present selected highlights of
their joint research programme, which has been achieved through intense collaboration between industry
and academia.
The peer review papers contained in these proceedings were accepted for presentation at ICIEGE, held
in Rotterdam, The Netherlands 24–27 March 2013. They have been authored by academics, research-
ers and practitioners from many countries worldwide and cover numerous important aspects related to
installation effects in geotechnical engineering, ranging from large deformation modelling to real field
applications. The main topics are:
• Computational methods
• Constitutive modelling
• Installation effects
• Offshore construction and foundations
• Soil improvement
• Soil-structure interaction
The submitted abstracts were reviewed and the authors of those abstracts that fell within the scope
of the conference were invited to submit full papers for peer review. The editors would like to thank the
Scientific Committee who provided assistance in the review process. They would also like to thank the
keynote speakers, authors, participants and members of the Organising Committee. The editors are grate-
ful for the support of the European Community and the partner organisations of the GEO-INSTALL
project: University of Strathclyde, United Kingdom (Project Coordinator); Delft University of Technol-
ogy, The Netherlands; Deltares, The Netherlands; Keller Limited, United Kingdom; Norwegian Geotech-
nical Institute, Norway; Plaxis BV, The Netherlands; Stellenbosch University, South Africa; University
of Stuttgart, Germany.
On behalf of the partners of GEO-INSTALL, we welcome you to The Netherlands and hope that you
find the conference both enjoyable and inspiring.

Michael Hicks
Jelke Dijkstra
Marti Lloret-Cabot
Minna Karstunen
January 2013

ix
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Committees

ORGANISING COMMITTEE

Prof. Michael Hicks, Delft University of Technology, The Netherlands (Chair)

Dr. Jelke Dijkstra, Delft University of Technology, The Netherlands (Co-Chair)
Dr. Marti Lloret-Cabot, University of Newcastle, Australia & Delft University of Technology,
The Netherlands
Ms. Bahar Akbarian, Delft University of Technology, The Netherlands
Mr. Marius Ottolini, Delft University of Technology, The Netherlands
Mr. Remon Romp, Delft University of Technology, The Netherlands

SCIENTIFIC COMMITTEE

Dr. Lars Andresen, Norwegian Geotechnical Institute, Norway

Dr. Patrick Becker, University of Strathclyde, UK
Prof. Alan Bell, Keller, UK
Prof. Thomas Benz, Norwegian University of Science and Technology, Norway
Dr. Ronald Brinkgreve, Plaxis, The Netherlands
Dr. Mike Brown, Dundee University, UK
Dr. Jorge Castro, University of Cantabria, Spain
Dr. Corne Coetzee, Stellenbosch University, South Africa
Dr. Jelke Dijkstra, Delft University of Technology, The Netherlands
Dr. Derek Egan, Keller, UK
Dr. Gustav Grimstad, Oslo and Akershus University College of Applied Sciences, Norway
Dr. Claire Heaney, Plaxis, The Netherlands
Prof. Michael Hicks, Delft University of Technology, The Netherlands
Prof. Minna Karstunen, Chalmers University of Technology, Sweden & Strathclyde University, UK
(GEO-INSTALL Coordinator)
Dr. Marti Lloret-Cabot, University of Newcastle, Australia & Delft University of Technology,
The Netherlands
Mr. Dirk Luger, Deltares, The Netherlands
Prof. Cesar Sagaseta, University of Cantabria, Spain
Dr. Nallathamby Sivasithamparam, Plaxis, The Netherlands
Prof. Pieter Vermeer, Deltares, The Netherlands & University of Stuttgart, Germany
Dr. Jimmy Wehr, Keller, Germany
Prof. Zdizslaw Wieckowski, Technical University of Lodz, Poland
Prof. Zhenyu Yin, Shanghai Jiao Tong University, China

xi
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Keynote speakers

Dr. Lars Andresen, Norwegian Geotechnical Institute, Norway

Prof. Minna Karstunen, Chalmers University of Technology, Sweden & University of Strathclyde, UK
Mr. Alain Puech, Fugro, France
Prof. Pieter Vermeer, Deltares, The Netherlands
Dr. Jimmy Wehr, Keller, Germany
Prof. David White, University of Western Australia, Australia

xiii
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Large deformation analysis of cone penetration testing

in undrained clay

L. Beuth
Deltares, Delft, The Netherlands

P.A. Vermeer
Deltares, Delft, The Netherlands
University of Stuttgart, Germany

ABSTRACT: Cone penetration testing is a widely-used in-situ test for soil profiling as well as estimating soil
properties of strength and stiffness. In this paper, the relationship between the undrained shear strength of clay
and the measured cone tip resistance is investigated through numerical analysis. Such analyses serve to refine
and establish correlations between cone penetration testing measurements and soil properties; thus enabling
more reliable predictions of soil properties. The presented analyses are performed by means of a Material
Point Method that has been developed specifically for the analysis of quasi-static geotechnical problems
involving large deformations of soil. Both, the load-type dependency of the shear strength of undrained clay
as well as the influence of the anisotropic fabric of natural clay on the undrained shear strength are taken
into account through a new material model, the Anisotropic Undrained Clay model. Results indicate that
the deformation mechanism relevant for cone penetration in undrained normally-consolidated clay differs
significantly from predictions based on the Tresca model, but resulting cone factors appear to be useful.

1 INTRODUCTION deformation processes imposed by the cone on

The undrained shear strength of soil, su, represents
no unique soil parameter, but largely depends on
the type of loading. For normally-consolidated
clays, the undrained shear strength found for tri-
axial compression is for instance much larger than
the strength found for triaxial extension. Simple
shear tests render an undrained shear strength that
lies in between the strength values obtained for
triaxial compression and extension. This has to be
taken into consideration when using this param-
eter in geotechnical analyses. When deriving an
undrained shear strength from cone penetration
measurements of tip resistance, qc, it is important
to know which undrained shear strengths domi-
nate the failure mechanism found during cone
penetration testing.
In this paper, results of numerical analyses of
cone penetration testing (CPT) in normally-con-
solidated undrained clay are presented to provide
new insight into the mechanical processes that
occur during a cone penetration test.
An accurate computation of the stress field
that evolves in the vicinity of the penetrating cone
requires one to take into account the complex large
the surrounding soil. Because the cone is pushed
into the ground with a constant rate of penetra-
tion of 2 cm/s, inertia and damping effects can
be neglected. Thus, the considered problem full-
fills the requirements of quasi-static analysis. The
method used in this study, the quasi-static Material
Point Method (MPM), has been developed specifi-
cally for the analysis of such problems.
Furthermore, the non-linear stress-strain rela-
tionship of undrained clay must be considered.
When using the well-known elastic-plastic Tresca
material model for such analyses, su is specified as
the cohesion parameter of the model. Such com-
putations assume an undrained shear strength that
is independent of the loading path. Generally, an
undrained shear strength obtained from simple
shear tests is used as a kind of average for such
simplifed analyses.
Although the widely-used Mohr-Coulomb
model does predict a higher undrained shear
strength for triaxial compression than for triaxial
extension, it overpredicts the magnitudes of both
undrained shear strengths, at least for normally-
consolidated clays. In order to reproduce the mag-
nitudes of undrained shear strengths for triaxial

1
compression and extension more accurately, a
model based on critical state soil mechanics such
as the well-known Modified Cam Clay model
might be used. This model takes into account
the shear-induced volume change of drained clay
(Schofield & Wroth 1968). Indeed, for normally-
consolidated clay as considered in this paper, this
model correctly predicts lower undrained shear
strengths for triaxial compression and extension
than the Mohr-Coulomb model.
Due to the deposition process of clay, natural
clay generally shows a different strength in hori-
zontal directions than in the vertical direction.
In order to further increase the accuracy of CPT
analyses, this anisotropic strength of clay must also
be taken into consideration. Therefore, the Aniso-
tropic Undrained Clay (AUC) model is used in the
presented analyses.
It implements the theory of critical state soil
mechanics and also considers the strength anisot-
ropy of natural clay (Vermeer et al. 2010; Beuth
2012). It is largely based on the S-CLAY1 model
developed by (Wheeler et al. 2003) but consid-
ers neither density and rotational hardening nor
softening.
To the authors knowledge, usage of a quasi-
static MPM in combination with the AUC model
for CPT analysis exceeds the accuracy of numerical
studies reported so far in literature, such as (Van
den Berg 1994) and (Lu et al. 2004).
In the following section, the used numeri-
cal method will be introduced briefly. A detailed
description of it would exceed the scope of this
paper. For further information, the reader is there-
fore referred to (Beuth et al. 2007; Vermeer et al.
2009; Beuth et al. 2011) and (Beuth 2012). The con-
stitutive modelling of undrained clay is treated in
Section 3. Subsequently, the geometry and discreti-
sation of the performed numerical analyses are pre-
sented in Section 4. In Section 5, results obtained
for the analysis with the AUC model are compared
to results for the Tresca model. The paper ends
with an outlook on possible future work.
2 QUASI-STATIC MATERIAL POINT
METHOD
The quasi-static MPM can be considered as an
extension of the classical Updated Lagrangian
Finite Element Method (UL-FEM). With the UL-
FEM, a solid body is discretised by finite elements
that follow the deformations of the solid body.
In case of large deformations of the solid body,
the finite element grid might eventually experi-
ence severe distortions which lead to numerical
inaccuracies and can even render the calculation
impossible.
2
The Material Point Method discretises a solid
body by means of a cloud of material points that
move through a fixed finite element grid. Thereby,
the material points capture the arbitrary large
deformations of the solid body without the occur-
rence of severe deformations of the finite elements.
Material and state parameters of the solid body as
well as applied loads are stored in material points
whereas the mesh does not store any permanent
information.
With the MPM, the underlying finite element
grid is used as with the UL-FEM to solve the sys-
tem of equilibrium equations for an applied load
increment on the basis of those finite elements that
contain material points at the considered loading
step. Once displacement increments are mapped
from nodes to material points, once strain incre-
ments are computed at the locations of material
points, the mesh is usually reset into its original
state. It might be changed arbitrarily.
Obviously, the finite element grid used with the
MPM must cover not only the solid in its initial con-
figuration as with the UL-FEM but the entire region
of space into which the solid is expected to move.
It should be emphasised, that in contrast to
mesh-less Lagrangian methods such as the Dis-
crete Element Method, the material points repre-
sent subregions of a solid body and not individual
particles such as sand grains. The mass and volume
of subregions is memorised with material points,
but changes in the shape are not traced.
With one exception (Guilkey & Weiss 2003),
existing implementations of the MPM are dynamic
codes that employ an explicit time integration
scheme (Sulsky et al. 1994; Wi eckowski et al. 1999;
Coetzee et al. 2005). Using these codes for the
analysis of quasi-static problems is computation-
ally inefficient as explicit integration requires very
small time steps. The quasi-static MPM makes use
of an implicit integration scheme and thus circum-
vents the limitation on step size of dynamic codes.
In recent years, the quasi-static MPM has been
validated with numerous geotechnical benchmark
problems. Furthermore, it has been extended by a
contact formulation for modelling reduced frictional
or adhesive contact between structure and soil based
on interface elements (Vermeer et al. 2009).
3 CONSTITUTIVE MODELLING
3.1 Modelling of undrained elasticity
The total mean stress rate of the undrained soil is
′ , and the
split into the effective mean stress rate, p′
change of excess pore pressures, pw , through con-
sideration of strain compatibility between the two
materials
 K water
pw = εvol and
d p′ K′ εvol (1)
n

where Kwater is the bulk modulus of water, n is the

porosity and K′ is the bulk modulus of the soil
skeleton. The term Kwater/n can be written as

K water 3( vu v′ )
= K′ (2)
n ( − vu ) ( + v′ )

where vu is an undrained Poisson ratio and v′

Figure 1. Yield locus of the AUC model for triaxial
the effective Poisson ratio of the soil skeleton.
compression and extension.
Incompressibility of the soil implies vu being
close to 0.5. In this paper, it is taken to be 0.49 for
the computation with the Tresca model and 0.495
for the computation with the AUC model. The giving the well-known Modified Cam Clay yield
effective Poisson ratio is assumed to be v′ = 0.25 condition for α0 = 0, i.e. for isotropic consolida-
and the Young’s modulus is E′ = 6 MPa. This tion with K0 = 1. In this particular case, the yield
yields K′ = 4 MPa and Kwater/n = 110 MPa for condition is represented by an ’isotropic’ ellipse in
vu = 0.49 and 235 MPa for vu = 0.495. It should p′-q-space.
be noted that elastic behaviour (inside the yield For non-isotropic consolidation, the yield locus
surface) is assumed to be isotropic according to is rotated away from the p′-axis, as illustrated in
Hooke’s law. Figure 1. In the special case of 1-dimensional
consolidation, the rotation parameter, α0, follows
from
3.2 Modelling of undrained plasticity
η02 + 3η0 − M 2 1 K0
1−
The AUC model requires 4 input parameters: the α0 = with η0 = 3 (6)
elastic parameters, E′ and v′, the friction angle 3 1 2K0
for the critical state, φcs, and the overconsolida-
tion ratio, OCR, defined as the preconsolidation where K0 is the coefficient of earth pressure at rest.
stress, σp0, divided by the vertical effective stress. In this paper, normally-consolidated clay is consid-
In this paper, only normally-consolidated clay with ered with K0 = 1 − sin φcs.
OCR = 1 is considered, φcs is taken to be 30°. Only associated plasticity is considered, so that
The yield function of the AUC model is defined as a separate specification of a plastic potential func-
tion is not needed. For OCR = 1, as considered in
T the following, the isotropic preconsolidation stress,
3⎛ α ⎞ ⎛ α ⎞
⎜ s p ′ 0 s0 : s − p ′ 0 s0 ⎟ p′po, can be initialised for the initial stress state
2⎝ q0 ⎠ ⎝ q0 ⎠ from the condition f = 0.
f p′ + − p′p 0
(M 2 2
0 ) p′
(3) 4 GEOMETRY AND DISCRETISATION
where s is the deviatoric stress tensor, and s0 is the A 20° slice of the axisymmetric cone penetration
initial in-situ one. Furthermore, it yields problem is considered as shown in Figure 2. The
mesh extends outwards by 14 cone diameters D,
6 sin φcs
M
3 − sin φcs
d q0 =
3 T
2
(
s0 : s0 ) (4)
D being for the standardised cone penetrometer
3.57 cm. In the initial configuration, the cone tip
is located 4.9 cm below the soil surface which ini-
For special stress loading paths with s propor- tially coincides with the cone base, see Figure 3 (b).
tional to s0, implying q proportional to q0, the yield The cone is slightly rounded in order to circum-
function can be written in the simpler form vent numerical problems induced by a discontinu-
ous edge at the base of the cone. Apart from this
modification, the dimensions of the penetrometer
f p′ +
(q p )2 − p ′p 0 (5) correspond to those of a standard penetrometer:
(M 2 2
0 ) p′ the apex angle is 60° and the horizontal base area
is 10 cm2.

3

Figure 2. (a) Material point discretisation for the initial;
(b) Final configuration at a penetration depth of 18 D.
Figure 3. (a) FE discretisation with local mesh refine-
ment along the cone tip; (b) Geometry of the cone tip.
4
The height of the discretised space has been
chosen so as to accomodate a penetration depth
of up to 18 D measured from the base of the cone.
The lower mesh boundary is located 18 D below
the tip of the cone. Considering a rigid penetrom-
eter, there is no need to discretise the penetrometer
itself which simplifies the application of boundary
conditions along the penetrometer surface.
Rather than incrementally moving the pen-
etrometer downwards into the ground, the soil is
pushed upwards against the fixed penetrometer
surface. The upward displacement of the material
point discretisation is realised by prescribed dis-
placements applied on the bottom boundary of the
mesh. Material points are continuously fed along
the lower boundary into the mesh from a cloud
of material points that is initially located outside
the discretised region (feeder). Figure 2 shows the
material point discretisation including the feeder in
its initial and final configuration.
Along the sides of the mesh, displacements are
fixed normal to the vertical planes (roller bound-
aries). Only smooth contact between soil and
penetrometer is considered by prescribing roller
boundaries fixed normal to the penetrometer sur-
face. Reduced adhesive or frictional contact along
the penetrometer surface might be considered by
placing interface elements along the fixed bound-
ary of the penetrometer but this is not considered
here.
Soil weight is not considered with the presented
computations since the gradient of vertical stress
is not relevant at the greater depths of cone pen-
etration that are reached in the performed analy-
ses. Because the computation with the AUC model
requires initial stresses for initialisation of the extent
and orientation of the yield surface in principal
stress space, a surface traction of 10 kPa is applied.
K0 is set to 0.5. Thus, in an initial loading stage, a
vertical stress of 10 kPa and horizontal stresses of
5 kPa are realised. For the AUC model, the iso-
tropic preconsolidation pressure is then −7.1 kPa.
The computations were performed with
4-noded tetrahedral elements that were extended
by a strain-smoothing algorithm to prevent lock-
ing. A mesh with 15,678 degrees of freedom has
been used. Mesh re-finement has been applied in
the region around the penetrometer cone as shown
in Figure 3 (a) where large stress gradients occur.
Furthermore, a mesh re-finement is required on
the narrow edge along the longitudinal axis of the
discretised slice.
Initially, the soil surface coincides with element
boundaries adjacent to the cone base, see Figure 3
(b). An equal number of 10 material points is
initially placed inside each finite element. In total,
286,410 material points are used for the performed
analyses.
5 RESULTS
FEM simulations of triaxial compression and
extension tests were performed with the AUC model
for anisotropically consolidated soil under K0 = 0.5
up to σp0 = 10 kPa. Similarly, a direct shear test was
computed for K0-consolidation up to σp0 = 10 kPa in
order to obtain the respective values of undrained
shear strength relevant for the performed CPT
analyses. The resulting undrained shear strengths
are for triaxial compresssion su,comp = 3 kPa, for
triaxial extension su,ext = 1.2 kPa and for simple
shear su,ss = 2.15 kPa (see also Figure 1). For the
computation with the Tresca model, an undrained
shear strength of 4 kPa is specified. An overview
of the undrained shear strengths relevant for the
two computations is provided in Table 1.
In Figure 4 the obtained relative tip stresses σc/su,ss
for the two computations are plotted over the rela-
tive cone penetration. Once the influence of the soil
surface on the movement of soil particles around the
penetrating structure subsides, a steady-state defor-
mation process evolves, as can be seen in Figure 4.
The tip stress σc for the fully developed steady-state
penetration process corresponds to the tip resistance
qc measured during cone penetration testing.
The relative tip resistance qc/su corresponds
to the cone factor Nc which is commonly used to
correlate the undrained shear strength of soil to
measurements of cone tip resistance. Equating
su to su,ss results in approximately the same cone
Table 1. Undrained shear strengths [kPa].
su,comp su,ext su,ss
Tresca 4.0 4.0 4.0
AUC 3.0 1.2 2.15
Figure 4. Relative tip stress plotted over the relative
cone displacement for the Tresca and AUC model.
5
factor for the computations with the Tresca and
AUC model. For the Tresca model, a cone factor
of 12.5 is obtained, for the AUC model a cone fac-
tor of 11.6.
The Tresca cone factor of 12.5 agrees well with
results of previous numerical studies based on
this model. Indeed, (Van den Berg 1994) reports
for a smooth cone and a rigidity index Ir = G/
su = 600 a cone factor of approximately 13. (Lu
et al. 2004), who additionally take into considera-
tion the initial stress state, report a matching cone
factor of 12.5. Total stress analyses were consid-
ered in these studies whereas in the computations
of this paper distinction is made between effective
stresses and excess pore pressures as described in
Subsection 3. Supplementary numerical analyses
show that the choice of analyses has little impact
on the obtained cone factors, at least for und-
rained Poisson ratios above a value of 0.49 (Beuth
2012).
Figure 5 (a) shows the loading types found within
the soil body at a penetration depth of 10 D; on the
left half for the AUC model, on the mirrored right
half for the Tresca model. Points colored dark grey
represent regions of triaxial compression states-
of-stress. Light grey points represent regions that
cannot be associated with a specific loading type.
For the Tresca model, triaxial compression states-
of-stress dominate in the vicinity of the cone. The
AUC model predicts triaxial compression states-
of-stress to occur only in a small wedge below the
advancing cone.
Figure 5 (b) shows the effective vertical stresses
for the AUC model and the Tresca model at a pen-
etration depth of 10 D. Compressive stresses are
negative. Low vertical effective stresses occur next
to the cone for both material models forming wing-
shaped stress bulbs. This can be explained by the
reduction of vertical compaction once soil leaves
the ‘shadow’ of the penetrating cone. Within the
shadow of the cone, soil is pushed to the sides and
downward, outside it, soil moves to the sides in an
upward direction as noted by Lu et al. Some of
the vertical compression previously applied on soil
particles is consequently released.
For the AUC model, as a consequence, mean
effective stresses decrease within the region of the
wing-shaped stress bulb. The stress paths obtained
for the AUC model correspond to soil yielding in
a critical state: stress states in this region follow
the critical state line. They do not correspond to
triaxial compression nor triaxial extension which
corresponds to the results shown in Figure 5 (a).
Soil yields in the vicinity of the cone with effec-
tive mean stresses decreasing down to −2.5 kPa
and deviatoric stresses q as low as 3 kPa which
explains the lower tip resistance found with the
AUC model.
Figure 5. (a) Regions of triaxial compression states-of-stress colored dark gray; (b) Comparison of effective vertical
stresses σ ′yy [kPa].
6 CONCLUSIONS
For a normally-consolidated clay, the relationship
between cone tip resistance and the strength of
undrained clay has been investigated. Penetration
of a smooth cone has been simulated. Large soil
deformations are taken into consideration through
use of a quasi-static MPM. The load-type depend-
ency of the undrained shear strength and the
strength anisotropy of natural clay are considered
through use of the new AUC model.
The cone factor obtained for a computation with
the Tresca model agrees well with results from the
AUC model. Hence, the Tresca model seems well
suited for determining cone factors for normally-
consolidated undrained clays, provided the cone
factors are related to an undrained shear strength
for simple shear.
Apart from the Tresca model, the Von Mises
yield criterion might be used for such CPT analy-
ses. However, when fitting the Von Mises to the
Tresca yield surface so that the yield stress coin-
cides for triaxial compression and extension, the
Von Mises model will give larger cone factors.
For a proper cone factor, the Von Mises criterion
should be matched to the Tresca yield criterion in
the case of simple shear. In order to obtain this, the
Von Mises tensile strength, σtension, should be equal
to 3 su,Tresca.
It would be important to also investigate cone
penetration in overconsolidated clays since the
overcon-solidation ratio has a considerable influ-
ence on the undrained shear strength of clay. It
would seem that the AUC model is not only suited
for normally-consolidated but also for overconsoli-
dated clays. This work would allow one to identify,
6
for given soil parameters of strength and stiffness,
the relationship between cone tip resistance and
undrained shear strength as a function of OCR as
suggested by Vermeer (2010). Formulating such a
relation would simplify the analyses of geotechni-
cal problems involving undrained clay and increase
their reliability.
ACKNOWLEDGEMENTS
The research leading to these results has received
funding from the European Community’s Seventh
Framework Programme FP7/2007–2013 under
grant agreement n° PIAG-GA-2009-230638 (Geo-
Install).
The authors would like to express their gratitude
to Deltares, especially to Peter van den Berg and
Hans Teunissen, for their support of the numerical
study on which this publication is based and of the
development of the AUC model.
The authors would like thank Plaxis B.V. and
Deltares for their funding of the development of
the quasi-static MPM. Here, the authors would
especially like to thank Paul Bonnier from Plaxis
B.V for kindly providing his expert advise in this
work and Issam Jassim from the University of
Stuttgart for his contributions to the research
efforts leading to this publication.
REFERENCES
Beuth, L. 2012. Formulation and application of a quasi-
static material point method. Ph. D. thesis, University
of Stuttgart, Holzgartenstr. 16, 70174 Stuttgart.
Beuth, L., Benz, T., Vermeer, P., Coetzee, C., Bonnier, P., &
Van Den Berg, P. 2007. Formulation and validation
of a quasi-static Material Point Method. In Proceed-
ings of the 10th International Symposium on Numerical
Methods in Geomechanics, Volume 10, pp. 189–195.
Taylor & Francis Group.
Beuth, L., Więckowski, Z., & Vermeer, P. 2011. Solution
of quasi-static large-strain problems by the mate-
rial point method. International Journal for Numeri-
cal and Analytical Methods in Geomechanics 35(13),
1451–1465.
Coetzee, C., Vermeer, P., & Basson, A. 2005. The mod-
elling of anchors using the material point method.
International Journal for Numerical and Analytical
Methods in Geomechanics 29(9), 879–895.
Guilkey, J., & Weiss, J. 2003. Implicit time integration
for the material point method: Quantitative and algo-
rithmic comparisons with the finite element method.
International Journal for Numerical Methods in Engi-
neering 57(9), 1323–1338.
Lu, Q., Randolph, M., Hu, Y., & Bugarski, I. 2004.
A numerical study of cone penetration in clay. Geo-
technique 54(4), 257–267.
7
Schofield, A., & Wroth, P. 1968. Critical state soil
mechanics. McGraw–Hill New York.
Sulsky, D., Chen, Z., & Schreyer, H. 1994. A particle
method for history–dependent materials. Compu-
ter Methods in Applied Mechanics and Engineering
118(1–2), 179–196.
Van den Berg, P. 1994. Analysis of soil penetration.
Ph. D. thesis, Delft University of Technology, The
Netherlands.
Vermeer, P., Jassim, I., & Hamad, F. 2010. Need and per-
formance of a new undrained clay model.
Vermeer, P., Yuan, Y., Beuth, L., & Bonnier, P. 2009.
Application of interface elements with the Material
Point Method. In Proceedings of the 18th Interna-
tional Conference on Computer Methods in Mechanics,
Volume 18, pp. 477–478. Polish Academy of Sciences.
Wheeler, S., Näätänen, A., Karstunen, M., & Lojander,
M. 2003. An anisotropic elastoplastic model for soft
clays. Canadian Geotechnical Journal 40(2), 403–418.
Więckowski, Z., Youn, S., & Yeon, J. 1999. A particle–in–
cell solution to the silo discharging problem. Interna-
tional Journal for Numerical Methods in Engineering
45(9), 1203–1225.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Adaptive Mesh Refinement for strain-softening materials

in geomechanics

C.E. Heaney & R.B.J. Brinkgreve

Delft University of Technology, Delft, The Netherlands
Plaxis BV, Delft, The Netherlands

P.G. Bonnier
Plaxis BV, Delft, The Netherlands

M.A. Hicks
Geo-Engineering Section, Faculty of CITG, Delft University of Technology, Delft, The Netherlands

ABSTRACT: This paper describes the implementation of Adaptive Mesh Refinement (AMR) within
the geotechnical software package Plaxis 2D. The algorithm is recovery-based and aims to reduce the
discretisation error estimated as a measure of the incremental deviatoric strain. Once the global discre-
tisation error exceeds a user-defined tolerance, certain elements in the mesh are marked for refinement.
The refinement procedure is based on a combination of regular subdivision and longest-edge bisection.
Mapping from the old mesh to the new mesh is accomplished by using the recovered solutions at the
nodes and the shape functions. The AMR algorithm is demonstrated for a vertical cut problem for a
softening Drucker-Prager material. Regularisation prevents the mesh-dependency which would otherwise
be seen for such softening constitutive models.

1 INTRODUCTION integration points at the nodes; (ii) error

Localisation is commonly observed in soils, often as
a precursor to failure. It manifests itself in the forma-
tion of shear bands, across which develop very high
gradients of displacement and strain. This poses a
considerable numerical challenge. Ideally the high
gradients in the shear band would be modelled with
a fine mesh, whereas, in other areas of the domain
where the spatial gradients are much lower, a coarser
mesh would be used. However, it is not known in
advance where the shear band will form, so the mesh
cannot be pre-refined. Uniform refinement of the
mesh would be wasteful as the shear band occupies
typically only a small part of the domain.
Adaptive Mesh Refinement (AMR) can be used
to tackle such problems. This method monitors
the discretisation error in the finite element (FE)
solution, and, when the global error norm exceeds
a specified tolerance, certain elements are marked
for refinement. Once the mesh has been refined
and the solution has been mapped to the new
mesh, the loading is resumed. This allows the mesh
to be refined in regions where the error is high (due
to the large gradients) resulting in a more accurate
description of the shear band.
The AMR algorithm implemented here has four
steps: (i) recovery—recovering fields defined at the
estimation—estimating the error in the finite ele-
ment solution; (iii) refinement—refining elements
which have an error larger than a pre-defined
tolerance; and (iv) mapping—the FE solution is
mapped from the old mesh to the new mesh so that
the calculation can continue.
Although widely used in some areas, AMR is not
often fully exploited in geotechnics. Some noteable
exceptions are Perić et al. (1996), Hu & Randolph
(1998), Hicks (2000) and Kardani et al. (2012).
2 FORMULATION
The AMR algorithm has been implemented within
Plaxis’s 2D displacement-based FE solver (see
Brinkgreve et al. (2011)). The meshes are unstruc-
tured and composed of 6-noded triangular elements
resulting in a quadratic approximation of the dis-
placements. For elastoplastic materials the load or
displacement boundary conditions are applied in
increments. Once an increment has converged, the
AMR algorithm recovers the nodal fields and calcu-
lates the error. If the error exceeds the user-defined
tolerance, the refinement and mapping algorithms
are called. The FE solver then applies the next
increment. The four stages of the algorithm are

8
now briefly discussed. A more detailed description
can be found in Heaney et al. (2013).
2.1 Recovery
Variables defined at integration points are recov-
ered at the nodes in order to estimate the error
and to facilitate the mapping process. For this pur-
pose, Superconvergent Patch Recovery (SPR) is
used (Zienkiewicz & Zhu 1992a). It is an efficient,
local method involving the inversion of a relatively
small matrix (small when compared to the size of
the global stiffness matrix).
The implementation of SPR involves two stages:
first, defining the patches (groups of neighbouring
elements) and second, fitting least squares sur-
faces to the integration point values in each patch.
Node-based patches are implemented here, which
are formed around each interior vertex node (often
referred to as an “assembly node”). The patch is
made up of all those elements which contain the
assembly node. To recover the solution at nodes
on or near the boundary, the standard node-based
patches containing elements in contact with the
boundary can be extended to include the boundary
nodes (as suggested in Zienkiewicz & Zhu (1995)).
In each patch, a least squares fit is carried out
to the data at integration points. In SPR the least
squares fit is of the same degree as the displacement
shape functions, which are quadratic in this case.
Each patch is mapped onto the domain [–1,1] ×
[–1,1] to avoid ill-conditioning of the matrix which
is inverted to obtain the least squares coefficients.
2.2 Error estimation
Zienkiewicz & Zhu (1992b) prove that their
error estimator (described in Zienkiewicz & Zhu
(1987)) is asymptotically exact for linear elastic
problems. For elastoplastic problems Boroomand &
Zienkiewicz (1999) devised an error estimator based
on incremental energy. More recently Hicks (2000)
employed an error estimator based upon a meas-
ure of incremental deviatoric strain. He defines the
incremental shear strain invariant, Δγ, as
Δγ Δε ijdev Δε ijdev , (1)
where Δε ijdev is the ijth component of the incremen-
tal deviatoric strain tensor. The error in this quan-
tity with respect to the L2 norm is given by
2
|| e ||
Δγ ∫Ω
( γ* γ) 2
, (2)
where Δγ* is the recovered field based on the recov-
ered incremental strains. This estimator should be
9
especially suitable for geomechanical problems
which exhibit large changes in strain.
The error estimator for element iEl is given by
|| e ||iE1
Δγ ∫iiE1( γ* γ )2 , (3)
and from this, the global error norm can then be
calculated:
nE 1
(|| ||Δγ )2 (|| ||Δi γ1 )2 . (4)
iE1 1
Either of the above error measures can be made
relative by dividing by
|| Δγ * |||| ∫Ω ( Δγ **) d Ω .
2
(5)
Both local and global errors can be utilized in
adaptivity. The global estimate is compared
against a user-defined tolerance, and, if it exceeds
the tolerance, mesh refinement is triggered. The
marking strategy (a method of determining which
elements to refine) is now outlined. The approach
followed here is to link the user-defined global tol-
erance with the local, element scale. This method
relies on the concept of an optimal mesh where the
error is equally distributed over the elements (for
example Mar & Hicks 1996)). Given a user-defined
global tolerance η (which is a relative measure), the
condition for triggering refinement is
|| e || Δ γ
≥ η. (6)
|| Δγ * ||
Introducing the assumption that the local errors
are equally distributed over the mesh gives
η || Δγ * ||
Δγ ≥
|| e ||local , (7)
nE1
where nE1 is the total number of elements. So, once
the global error exceeds the user-defined (global)
tolerance, then all elements whose error exceeds
η || Δγ *||
nE1
will be marked for refinement.
2.3 Mesh refinement
In h-adaptivity, the mesh can be refined in two
ways: either by regeneration or by subdivision.
Regeneration is often used and produces elegant-
looking meshes directly indicative of the (numerical
approximation to the) underlying physical mecha-
nism. However, one consequence of regeneration
is that the solution over the entire domain must be
transferred from the old to the new mesh. In order
to limit the numerical diffusion which may occur
as a result of transferring the solution between
meshes, and to be more efficient, subdivision is
used.
Regular bisection refers to the splitting of an
element into four elements by joining the mid-
points of the triangle’s edges (Bank et al., 1983).
Between a regularly-refined element and a non-
refined element will be a non-conforming edge
with hanging nodes. A straightforward solution
to this is adopted here based on Rivara’s longest-
edge bisection method (Rivara 1984). One advan-
tage of using longest-edge bisection is that the
minimum internal angle of subsequent meshes
is never less than half of the smallest interior
angle of the original mesh (Rosenberg & Stenger
1975).
2.4 Mapping
New elements require values of displacement and
stress, and also any state variables which are asso-
ciated with the constitutive model. Displacements
can be mapped using the shape functions and the
nodal values of displacement. In order to map the
stresses, they are first recovered at the nodes using
SPR, then they can be mapped to the new mesh
just as the displacements were. Any integration-
point values can be mapped using this method.
After mapping, the stress field will no longer be
in equilibrium with the applied force. A zero-load
increment can be applied in order to resolve this,
if necessary.
3 REGULARISATION
The stress invariants p (the mean hydrostatic stress)
and q (a measure of deviatoric stress) are used to
define the conical Drucker-Prager yield surface:
f = q + α (p − c cot φ) = 0, (8)
where α is related to the friction angle, φ, and c is
the cohesion of the material. In order to represent
softening, c depends linearly on a scalar measure
of plastic strain, κ, the constant of proportionality
is the softening modulus, H:
c = max (c0 − Hκ, cmin). (9)
Once an integration point starts behaving elasto-
plastically, the cohesion reduces at a rate governed
by the softening modulus from the initial value (c0)
to the minimum value (cmin). The evolution of κ is
governed by
10
dk = 2 / 3 || d p
|| . (10)
The regularisation employed here follows the
non-local approach whereby strains are averaged
in the region around the integration point of
interest. A weighted combination of the non-local
and local strains is used in the calculation of dκ.
During the averaging process a parameter is intro-
duced which effectively controls the minimum
width of the shear band. The interested reader is
referred to Brinkgreve (1994) for further details.
This approach has already shown good results for
standard FE calculations (Rolshoven & Jirásek
(2003), Rezania et al. (2012)).
4 RESULTS
A vertical cut is modelled, the geometry of which can
be seen in Figure 1. The (effective) material param-
eters are shear modulus G = 3355 kNm−2, Poisson’s
ratio υ = 0.49, initial cohesion c0 = 5 kNm−2, mini-
mum cohesion cmin = 1 kNm−2 and softening modu-
lus H = 100 kNm−2. The problem is undrained so the
friction angle is zero and the yield surface becomes
a cylinder. Displacement is applied in increments
along half of the topsurface up to a total displace-
ment of 0.03 m. Plane strain is assumed and self-
weight is not considered.
A number of meshes were used comprising
6-noded triangular elements: the mesh densities
are given in Table 1. First the problem was run for
each mesh with the standard FE method. Then
the problem was run again for each mesh with
regularisation. Finally mesh 2 was used as a start-
Figure 1. A vertical cut. The s-axis labelled here is used
later.
Table 1. Numbers of elements and degrees of freedom
in each mesh used in the standard and the regularised FE
simulations.

Number Number of degrees

of elements of freedom

Mesh 1 147 660

Mesh 2 288 1250
Mesh 3 603 2552
Mesh 4 1150 4794
Mesh 5 2314 9530
Mesh 6 4606 18810
Mesh 7 8011 32552

Table 2. Numbers of elements and degrees of freedom

in the AMR simulation.

Number Number of degrees

of elements of freedom

Initial mesh 288 1250

First refinement 357 1536
Second refinement 490 2078
Third refinement 884 3662
Fourth refinement 1528 6254
Fifth refinement 2310 9046

ing mesh for the AMR algorithm used in combina-

tion with the regularisation.
The AMR algorithm triggered refinement at
the end of the 4th, 5th, 6th, 7th and 8th displace-
ment increments. The mesh densities are given in
Table 2.
Figure 2 shows how the displacement varies
across the shear band along the s-axis (shown in
Figure 1) for the three sets of results: (a) stand-
ard FE results; (b) FE results with regularisation
and (c) AMR results with regularisation. For the
standard FE results it can be seen that, on increas-
ing the number of elements, the gradient tends to
infinity and the shear band width tends to zero.
For the FE results with regularisation, the gra-
dient and shear band width remain finite as the
mesh density increases. This is also true of the
AMR results seen in Figure 4 where there is excel-
lent agreement between results using the finest
mesh with FE and regularisation, and the AMR
results.
Figure 3 shows the load displacement curves for
the three sets of results (the standard FE results;
FE with regularisation; and AMR with regulari-
sation). Figure 4 shows that on increasing mesh Figure 2. The variation of the magnitude of the dis-
density, the FE results are clearly mesh dependent. placement vector along the s-axis (see Figure 1). (The dis-
Figure 4 shows how the regularisation alleviates placement has been normalised by its maximum value.)

11
Figure 3. The total reaction force (kNm-1)along the top surface against the applied displacement (m) for the vertical cut.

12

Figure 4. Contours of Δγ for mesh 7 without regu-
larisation (top), mesh 7 with regularisation (centre), and
AMR with regularisation (bottom).
13
Figure 5. The deformed mesh at the end of the
AMR simulation. Displacements have been exaggerated
by a factor of 10.
mesh dependence. Once the mesh is fine enough
to resolve the shear band, on further increasing
the mesh density, the load displacement curves
converge. Finally, Figure 4 demonstrates that the
AMR algorithm when combined with regularisa-
tion also does not suffer from mesh dependence.
The AMR results are again in excellent agreement
with the finest mesh used in the regularised FE
simulations.
Figure 4 shows contours of Δγ, the variable on
which the error is based. The FE results show that
the orientation of the mesh has an influence of
the results and the shear band is very thin when
compared with the regularised results (with and
without AMR). Figure 4 shows the deformed,
final mesh for the AMR simulation.
5 CONCLUSIONS
An AMR algorithm has been described and imple-
mented in Plaxis 2D. For softening materials, mesh
dependency is a known weakness when modelling
problems with the FE method. As an AMR algo-
rithm will aim to reduce the element-size in the
shear band, a softening material will prove a stern
test of the regularisation. In this paper, non-local
regularisation is used to provide the length scale
required to obtain mesh-independent results. With
fewer elements, the AMR results show excellent
agreement with the fixed-mesh FE results (with
regularisation applied in both cases).
ACKNOWLEDGEMENTS
The authors gratefully acknowledge the support of
GEO-INSTALL: a project funded by the European
Community through the Marie Curie IAPP pro-
gramme (Contract No. PIAP-GA-2009-230638).
REFERENCES
Bank, R., Sherman, A.H. & Weisser, A. 1983. Refine-
ment algorithms and data structures for regular local
mesh refinement. In Scientific computing (IMACS
Transactions), North Holland, pp. 3–17.
Boroomand, B. & Zienkiewicz, O.C. 1999. Recovery
procedures in error estimation and adaptivity. Comput.
Methods Appl. Mech. Engrg 176, 127–146.
Brinkgreve, R.B.J. 1994. Geomaterial models and numeri-
cal analysis of softening. Ph.D. thesis, Delft University
of Technology.
Brinkgreve, R.B.J., Swolfs, W.M. & Engin, E. 2011.
Plaxis 2D 2011. Plaxis BV.
Heaney, C.E., Bonnier, P.G., Brinkgreve, R.B.J. &
Hicks, M.A. 2013. Adaptive mesh refinement with
application to geomaterials. In preparation for submis-
sion to Comput. Geotech.
Hicks, M.A. 2000. Coupled computations for an elastic-
perfectly plastic soil using adaptive mesh refinement.
Int. J. Numer. Anal. Meth. Geomech. 24, 453–476.
Hu, Y. & Randolph, M.F. 1998. H-adaptive FE analy-
sis of elasto-plastic non-homogeneous soil with large
deformation. Comput. Geotech. 23, 61–83.
Kardani, M., Nazem, M., Abbo, A.J., Sheng, D. &
Sloan, S.W. 2012. Refined h-adaptive FE procedure
for large deformation geotechnical problems. Comput.
Mech. 49, 21–33.
14
Mar, A. & Hicks, M.A. 1996. A benchmark computa-
tional study of finite element error estimation. Int.
J. Numer. Meth. Engng 39(23), 3969–3983.
Perić, D., Hochard, C., Dutko, M. & Owen, D.R.J. 1996.
Transfer operators for evolving meshes in small strain
elasto-plasticity. Comput. Methods Appl. Mech. Engrg
137, 331–344.
Rezania, M., Bonnier, P.G., Brinkgreve, R.B.J. &
Karstunen, M. 2012. Non-local regularisation of
Drucker-Prager softening model. In Z. Yang (Ed.),
Proceedings of the 20th UK National Conference of
ACME, Manchester (UK), pp. 275–278.
Rivara, M.C. 1984. Design and data structures of a fully
adaptive multigrid finite element software. ACM
Transactions on Mathematical Software 10, 242–264.
Rolshoven, S. & Jirásek, M. 2003. Numerical aspects of
nonlocal plasticity with strain softening. In Compu-
tational Modelling of Concrete Structures, Austria,
pp. 305–314.
Rosenberg, I.G. & Stenger, F. 1975. A lower bound on the
angles of triangles constructed by bisecting the longest
side. Mathematics of Computation 29, 390–395.
Zienkiewicz, O.C. & Zhu, J.Z. 1987. A simple error esti-
mator and adaptive procedure for practical engineering
analysis. Int. J. Numer. Meth. Engng 24, 337–357.
Zienkiewicz, O.C. & Zhu, J.Z. 1992a. The superconver-
gent patch recovery and a posteriori error estimates.
Int. J. Numer. Meth. Engng 33, 1331–1364.
Zienkiewicz, O.C. & Zhu, J.Z. 1992b. The superconver-
gent patch recovery (SPR) and adaptive finite element
refinement. Comput. Methods Appl. Mech. Engrg 101,
207–224.
Zienkiewicz, O.C. & Zhu, J.Z. 1995. Superconvergence
and the superconvergent patch recovery. Finite Elem.
Anal. Des. 19, 11–23.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

A dynamic material point method for geomechanics

I. Jassim
Institut für Geotechnik, Universität Stuttgart, Germany

C. Coetzee
Department of Mechanical Engineering, University of Stellenbosch, South Africa

P.A. Vermeer
Deltares, Delft, The Netherlands
University of Stuttgart, Germany

ABSTRACT: A dynamic Material Point Method for use in Geomechanics is presented. Soil and structural
bodies are represented by (material) particles, which move inside an unstructured mesh of four-noded 3-D
tetrahedral elements. As such low-order elements tend to show locking for fully developed plastic flow,
a strain-enhancement remedy is described. As a first example, the penetration of a drop anchor into a
Mohr-Coulomb soil is considered. As both a soil body and a metal anchor are considered, an algorithm for
dynamic contact is used and described. An improved type of absorbing boundaries to avoid the reflection
of stress waves is also described. The second example consists of dynamic cone penetration. Finally, the
example of a collapsing tunnel is considered.

1 INTRODUCTION The early beginnings of MPM can be traced back

Over the last decades the Finite Element Method
(FEM) has become the standard tool of analysis
in the field of solid mechanics. However, due to its
reliance on a mesh, the FEM is not well suited for
the treatment of extremely large deformations. To
overcome the mesh dependency of the FEM, mesh-
free methods have been developed, for example the
Element-Free Galerkin Method (Belytschko et al.
1994) and the Material Point Method (MPM). The
latter might be classified as a meshfree method, a
Particle-in-Cell method or an Arbitrary Lagrang-
ian-Eulerian method (Więckowski 2004).
MPM uses two discretizations of the material,
one based on a computational mesh and the other
based on a collection of material points or “par-
ticles”. Within the standard MPM large deforma-
tions are modeled by particles moving through a
fixed mesh. The particles carry all the properties
of the continuum (material properties and state of
stress and strain) as well as external loads, whereas
the mesh and its Gaussian integration points carry
no permanent information at all. The computa-
tional grid is used to determine incremental dis-
placements of particles by solving the governing
equations. Through this approach, MPM combines
the advantages of both Eulerian and Lagrangian
formulations.
to the work of Harlow (1964), who studied fluid flow
by material points moving through a fixed grid. Sul-
sky et al. (1995) extended the method to the mod-
eling of solids. It was then called the Material Point
Method (Sulsky & Schreyer 1996). Bardenhagen
et al. (2000) further extended the method by includ-
ing frictional contact between deformable bodies.
The potential of MPM for simulating granular
flow, e.g. in silo discharge, was first recognized by
Więckowski et al. (1999). Coetzee et al. (2005) used
the MPM for studying the large deformation prob-
lem of anchor pull-out. All previous developments
of MPM are based on dynamics, Beuth et al. (2010)
were the first to develop a quasi-static MPM.
In Chapter 2 of this paper, the weak formulation
and space integration of the momentum equation is
presented. Chapter 3 describes integration in time.
Here, explicit forward marching is used. A numeri-
cal example of a drop anchor is provided in Chap-
ter 4. The formulation of a contact algorithm used
to model the interaction between different bodies
is explained in Chapter 5. In Chapter 6, mesh lock-
ing and strain smoothening are discussed. Absorb-
ing boundaries and wave reflection are discussed in
Chapter 7. A second numerical example is given in
Chapter 8. In this example, dynamic cone penetra-
tion is investigated. Chapter 9 contains the exam-
ple of a collapsing tunnel.

15
2 WEAK FORMULATION AND SPACE and storage advantages, is used instead of the con-
INTEGRATION OF EQUILIBRIUM sistent-mass matrix defined in Eq. (5). On denot-
ing the entries of the lumped-mass matrix, as mi
The Cauchy form of conservation of linear it yields
momentum is given by the equations
⎡ m1 0 ... 0 ⎤
ρ  ∇ + ρ g and t σ n (1) ⎢0 m2 ... 0 ⎥
ML = ⎢ (6)
   ⎥
⎢0 0 ... mn ⎥⎦
where ρ is the material density, ü is the displace- ⎣
ment, a superposed dot denotes differentiation
with time, σ denotes the Cauchy stress tensor and where n denotes the number of degrees of freedom.
g is the gravitational acceleration vector. The sur- In order to conserve the mass of the continum,
face traction acting on the external boundary is particle-based integration is used i.e.
denoted by t and n is the outward unit normal of
the boundary. Applying the virtual work principle np
on a domain of volume V surrounded by boundary
S yields mi ∑ m p Nip and
p =1
np

∫δ ρ u dV = − ∫ δεε T σ dV + ∫ δ ρ g dV ∑ BTp σ p Vp
T T
F int = ∫ B σ dV ≈
T
(7)
V V V V p =1
+ ∫ δ uT t dS (2)
S
where np denotes the number of particles, mp is the
mass of particle p, Nip is the shape function evalu-
where δ implies a virtual quantity, ε is the strain ated at particle p and Vp is the volume associated
tensor and the script T denotes the transpose. with particle p. The drawback of using a lumped-
For space discretisation, the displacement field mass matrix is a slight numerical dissipation of the
u is approximated in terms of interpolation func- kinetic energy (Burgess et al. 1992).
tions N and nodal displacements a by u = Na. The
strain tensor is now written in vector notation as
3 TIME INTEGRATION
ε = Ba with B = LN and
ε=( )
T
(3) On applying Euler-forward time integration with
lumped-mass matrix, Eq. (4) yields
where B is the usual finite element strain-displace-
Δt −1
ment matrix, as computed from the linear dif- a t = a t + Δt  at = ⎡⎣ MtL ⎤⎦
at ,  Ft (8)
ferential operator L and the shape functions N.
Substituting Eq. (3) into Eq. (2) gives
where Δt is the current time increment, a t and a t Δt
δ T
δa T
(F ext
F int
) or M 
a= F are the nodal velocities at time t, t + Δt respectively.
The incremental nodal displacement is obtained by
with F F ext − F int (4) integrating the nodal velocity by the Euler-back-
ward rule and the position of the particles are sub-
in which sequently updated, i.e.
Δ Δt
Δat Δ at , xtp Δ xtp + N p Δat Δt
(9)
M ∫ N N F = ∫ ρ NT g dV + ∫ NT t dS
V V S
where xtp and xtp Δt are the particle positions at
and F int ∫ B σ dV
T
( )
V
time t and t + Δt respectively. Strains and stresses
at particles are updated using the same algorithms
as for Gaussian integration points within the
Equation (4) is identically used within FEM and standard FEM. In updated Lagrangian FEM, one
MPM. However, in the Material Point Method M would use Δa to update the finite element mesh,
can also change in size when particles move into but within the MPM only particles positions are
empty elements. In other words, the total number updated. Particles eventually cross element bound-
of degrees-of-freedom of the system can vary. A aries, which entails that the new element of a cross-
lumped-mass matrix, which offers computational ing particle has to be detected.

16
Figure 1. A torpedo-shaped drop anchor.

4 ANALYSIS OF DROP ANCHOR

The analysis of a drop anchor as shown in Figure 1

is presented in this chapter. These torpedo-shaped
anchors are used in the offshore oil and gas indus-
try as a cost effective anchoring solution in clays.
The anchor is dropped from a particular height and
penetrates the seafloor by the kinetic energy gained
during its free fall (Fig. 1). In the present paper, a
fully dynamic penetration process is simulated. As
the problem is axisymmetric, only a sector of 20o is
discretised. The drop anchor resembles a foundation Figure 2. Initial and final configuration of the problem.
pile, where soil deformations are intense around the
tip. Accurate computations require a relatively dense
mesh around the tip, as also shown in Figure 2a. standard MPM by using a mesh which is fixed to the
Within the standard MPM, however, the anchor (par- anchor so that it moves into the soil. Thereby, the fine
ticles) would move through the mesh and one would part of the mesh will always remain around the tip of
need mesh refinement over the complete penetration the anchor. The occurrence of elements containing
depth. In order to avoid this, we deviated from the particles of different material is also prevented as the

17
Figure 3. Penetration depth for different values of the
friction coefficient. Figure 4. Illustration of two bodies in contact for a
regular mesh.

boundary of the anchor coincides throughout the

analysis with element boundaries.
The material properties of the anchor are: La =
A A
F A , M BL a B FB , M A
L
B
a A+ B = F A
 B (10)
Young’s modulus of 50000 kPa, Poisson’s ratio
of 0 and unit weight of 78 kN/m3. The anchor is From these solutions, predictor velocities a A, a B
modeled as linear elastic. The soil is modeled by the and a A B are computed. Contact at a considered
Mohr-Coulomb model with a Young’s modulus of node is detected by comparing the velocity of a
5000 kPa, Poisson’s ratio of 0.3, cohesion of 5 kPa, single body to the velocity of the combined bodies,
friction angle of 30o and unit weight of 18 kN/m3. as illustrated in the chart of Figure 5, with n being
Simulations were done with different friction coef- the outward unit normal at a considered node.
ficients μ of the contact surface between the anchor When these velocities differ, the considered node is
and the soil. Hence, we do not consider penetration a contact node. Now we detect whether or not the
in undrained clay, but in a drained sand. This is sim- contact at that node is broken (by separation) or
ply done because frictional sand is computationally continued (by approaching). For an approaching
more challenging than (undrained) cohesive clay. contact node, we check for sliding as explained in
Figure 3 shows the computed penetration depth the next section.
for different μ-values plotted over time. For fully
rough contact (μ = 1), the final penetration depth
is found to be 5D, where D is the diameter of the 5.1 Check for sliding and subsequent correction
anchor. In the very beginning the anchor resist- for a contact node
ance is linear with penetration depth, but later the In the following, only body A is considered. The
penetration is slowing down because the anchor relative normal and tangential velocities at a con-
looses its kinetic energy. As the anchor is penetrat- tact node are
ing deeper into the soil, the resistance is getting
higher because the contact between anchor and
soil is increasing and because the shear resistance a n
⎣ ( ⎦)
⎡ a A − a A+ B n ⎤ n and
of the soil increased with depth. The ultimate pen-
etration depth obviously increases with smaller
a t ( )
a A − a A+ B a n (11)
friction coefficients. The largest penetration depth
obviously corresponds to the case of smooth con- The normal and tangential components of the
tact. Figure 2b shows the particle distribution after interaction force at a contact node can then be
a penetration of 6.5D for the case of smooth con- computed from
tact. It also shows soil heave around the anchor.
mA mA
FnA a n and FtA
and a t (12)
5 CONTACT ALGORITHM Δt Δt

Following Bardenhagen et al. (2000), Eq. (8) is where mA is the mass of a contact node computed
solved for the combined bodies A and B (see Fig. 4) only from body A as in Eq. (7). In frictional con-
as well as for each body separately, i.e. tact, the tangential force is limited by

18
 6 MESH LOCKING AND STRAIN
SMOOTHENING

Difficulties arise when determining the displace-

ment field for a solid that is nearly incompressible.
For such a material, the bulk modulus is very large
and small errors in strain will yield large errors in
stress. Furthermore, when dealing with low-order
elements the mesh may lock when incompress-
ibility constraints from neighboring elements are
imposed. For high-order elements, it is common
to prevent locking by reduced integration (Bathe
1982), but for low-order elements a kind of strain
smoothening can be applied, being referred to as
nodal-mixed-discretisation by Detournay and Dzik
(2006). This technique involves first of all the com-
putation of the strain rates for each element in the
usual manner subsequently they are decomposed
into a volumetric strain rate, εv, and a deviatoric
strain rate, ε d . An averaged volumetric strain-rate
εv for an element is now defined as

1 4 
∑ (εv )k
εv ∑ εvi with εvn = k
(15)
4 i =1 ∑ ( )k
k

where the sum is over all elements k attached to the

node n. Ω is the volume of an element. The work-
ing assumption is that deviatoric strain rates need
not be enhanced, only the volumetric components.
As a result, the final strain rate within an element
is redefined as

1 
ε = ε d + ε v I with I = (1, 1, 1, 0, 0, 0 )T (16)
3

Figure 5. Flow chart of the contact algorithm applied This approach is applied in all examples of the
on body A. present paper.

mA ⎡ A
FtA max μ FnA = μ
Δt ⎣
a ( a A+B ⋅ n ⎤
⎦) (13)
7 ABSORBING BOUNDARIES

In numerical simulations of wave propagation,

where μ is the coefficient of friction. Sliding the use of finite boundaries leads to reflection of
between the two bodies will only occur when the waves upon reaching the boundary. In Geo-
FtA FtA,max . Thus, only then a correction of the mechanics rigid boundaries are mostly numeri-
cal antifacts and reflecting waves are not physical
nodal velocity is required. It can be derived that the and they will affect the solution considerably. This
corrected velocity of a sliding contact node is problem might be overcome by choosing the finite
a A
new ⎣(
a A − ⎡ a A )
a A+ B ⋅ n ⎤ ( n + μ t )
⎦
(14)
boundary of the mesh far enough so that no reflec-
tion occurs. But this is not always practical solu-
tion as it makes the mesh unnecessarily large. On
where t is the direction vector of the tangential top of that, the computational effort increases con-
velocity. siderably. A partial solution to this problem was

19
introduced by Lysmer and Kuhlemeyer (1969).
They proposed a solution in which the boundary is
supported on a dashpot. On denoting the normal
stress at a boundary node as tn and the shear stress
as ts, it yields

tn n ρVp an a d ts = α s ρVs as (17)

where an and as are the normal and tangential

velocities at a boundary node respectively, ρ is the
mass density of the material, αn and αs are dimen-
sionless parameters, Vp and Vs are the p-wave
speed and the s-wave speed of the material respec-
tively. It yields

Vp Ec ρ d Vs G ρ (18) Figure 6. Displacement of the bottom of the mesh for

dashpot and spring in parallel.
where Ec is the constrained modulus and G is the
shear modulus. They relate to Young’s modulus
and Poisson’s ratio by the equations On the other hand, the dashpot with spring in
parallel will limit the creep of the boundary as

Ec =
( )E and G =
E shown in Figure 6. This displacement corresponds
(19) to δ = 0.5 m and αn = 2.5. Sensitivity study showed
( ))(( + ) 2( + ) that those are the best values for the problem con-
sidered. The final displacement of the bottom of the
The drawback of supporting the boundary by mesh can be calculated as an = (σ/Ec)δ. For σ = 1 kPa
dashpots is that the boundary will continuously and Ec = 100 kPa, this displacement is found to be
creep as long as the dashpot will receive stresses 0.05 m. Points 1,2 and 3 of Figure 6 indicate that
from the soil body. In order to limit the creep of the wave is just reflected after reaching the bottom
the boundary, a spring is added parallel to the of the mesh. Stress fronts correspond to point 1, 2
dashpot to obtain a Kelvin-Voigt type of bound- and 3 are (1.28, 1.86 and 1.14) kPa respectively.
ary response. Hence Eq. (17) is rewritten as

Ec 8 DYNAMIC PENETRATION TEST

tn n ρVp an + an and
δ
G The analysis of a cone (with a diameter of 3.57 cm)
ts s ρVs as + as (20)
δ being hammered into the soil is considered in this
where δ is a virtual thickness being used to limit the
creep of the boundary.
As a numerical example a soil layer with a thick-
ness of 1 m is considered here. The layer has a
Young’s modulus of 100 kPa, a Poisson’s ratio of 0
and unit weight of 18 kN/m3. A uniformly distrib-
uted load of 1 kPa is applied instantaneously at
the surface. Three different boundary conditions
were considered at the bottom: fully fixed, dashpot
and dashpot with spring in parallel. With the fully
fixed boundary, the entire energy is reflected when
the wave reached the bottom. Hence, the stress is
doubled after reflection and oscillates later con-
tinuously. When replacing the fixities by dashpots,
only a small portion of the energy is reflected,
but (as explained previously), the drawback of
using only dashpots is a continuous creep of the
boundary.
chapter. Block is successively dropped from a cer-
tain height as shown in Figure 7. The maximum
impulsive load, Fmax, corresponding to the drop
weight, can be calculated from the conservation of
momentum i.e.
t pulse
⎛ πt ⎞
∫ Fmax sin ⎜ ⎟ dt = η m 2 gh
⎝ t pulse ⎠
(21)
0
where η is the hammer efficiency (fraction of
energy transferred to the cone). It was chosen as
64% (Borja 1988). The initial configuration of the
problem is shown in Figure 8a. The same moving
mesh technique used to simulate the drop anchor
problem of chapter 5 is used here. The material
properties of the elastic cone are: Young’s modulus
of 50000 kPa, Poisson’s ratio of 0 and unit weight
of 78 kN/m3. Adopting the elastic-plastic Mohr-

20
Figure 7. Applied load of dynamic penetration test.

Coulomb model, the soil properties are: Young’s

modulus of 5000 kPa, Poisson’s ratio of 0.3, cohe-
sion of 5 kPa, friction angle of 30o and unit weight
of 18 kN/m3. The dropped weight has a mass
m = 10 kg, the drop height h = 50 cm, tpulse = 0.02 sec
and tperiod = 0.1 sec. The case of fully rough contact,
μ = 1, between the cone and the soil is considered.
Figure 9 shows the penetration verses the number
of blows. Figure 8b shows the principal stresses at
the end of the penetration.

9 DYNAMIC COLLAPSE OF TUNNEL FACE

Tunnel collapse can pose a danger to life and prop-

erty. Both the life of the workers standing in front
of the tunnel face and individuals at the ground
surface might be in danger. An example of tunnel
collapse in Munich is shown in Figure 10. The aim
of this analysis is to estimate the real collapse time
of a tunnel and the crashing of the ground. The
pattern of the tunnel face collapse is also inves-
tigated. The dimensions of the tunnel mesh are
shown in Figure 11. The boundary conditions are:
the upper surface is free to move, the side surfaces
are roller supported, and the base is fixed.
Figure 8. Initial configuration and final state of stress.
In the first stage of the analysis, initial stresses
are generated based for a Ko-value of 0.5, where
Ko is the co-efficient of lateral earth pressure at
rest. Adopting the elastic-plastic Mohr-Coulomb
model, the soil properties are: Young’s modulus
of 10000 kPa, Poisson’s ratio of 0.3, cohesion of

21
Figure 9. Penetration as a function of number of blows.
Figure 10. Collapse of tunnel face Munich under-
ground in September 1994.
Figure 11. Description of tunnel geometry.
1 kPa, friction angle of 25o and unit weight of
16 kN/m3. A relatively fine mesh is used to discre-
tise the material around the opening of the tun-
nel where the material is expected to flow. The face
support pressure is removed in single step and the
calculation is carried out by applying time steps of
Δt = 0.001 second. For time increment, the parti-
cles move due to the unbalanced forces in the sys-
22
Figure 12. Pattern of the tunnel face collapse.
tem until the kinetic energy is dissipated and static
equilibrium is reached. The total displacements for
the final static equilibrium are shown in Figure 12a.
The collapse time for the tunnel face is found to be
7 seconds. The settlement of the ground surface is
shown in Figure 12b.
10 CONCLUSIONS
Existing dynamic MPM codes are based on a regu-
lar grid. For complex structures, being represented
by a cloud of material particles, this would require
a special CAD type preprocessor. On the other
hand existing user-friendly preprocessors can be
used in combination with non-regular and non-
structured meshes. This is the main reason for the
use of the non-structured meshes. Another advan-
tage of the present non-structured mesh approach
is that it allows for mesh-refinements. In this paper
this has been applied for reason of achieving good
accuracy in zones of intense soil deformations. For
a consequent application of mesh refinement, the
concept of moving mesh has been introduced.
In dynamic soil analyses one usually introduces
absorbing boundaries to prevent the reflection of
stress and strain waves at the more or less arbitrary
bottom of the mesh. Hence, one usually employs
so-called dashpots that will continually creep under
load. In order to limit such non-physical displace-
ments, the dashpot is combined with a spring to
obtain a Kelvin-Voigt type of boundary response.
At present the main limitations of the code is the
lack of pore pressures and a soil model for cyclic
loading. Meanwhile, dynamic generation and dis-
sipation of pore pressures is nearly finished and
remains to be reported. The implementation of a
cyclic loading model for genuine simulations of
pile driving has been planned.
ACKNOWLEDGEMENT
This research was carried out as a part of the ‘’GEO-
INSTALL‘’ project (Modelling Installation Effects
in Geotechnical Engineering). It has received fund-
ing from the European Community through the
program (Marie Curie Industry-Academia Part-
nerships and Pathways) under grant agreement
no PIAP-GA-2009-230638. The PhD study of the
first author at Stuttgart University is funded by the
DAAD (German Academic Exchange Service).
REFERENCES
Bardenhagen, S.G., Brackbill, J.U. & Sulsky, D. 2000.
“The material-point method for granular materials’’.
Computer Methods in Applied Mechanics and Engi-
neering, Vol. (187), 529–541.
Bathe, K.J. 1982, Finite Elements Procedures in Engi-
neering Analysis, Prentice-Hall, Inc., New Jersey.
Belytschko, T. Lu, Y.Y. & Gu, L. 1994, “Element-free
Galerkin methods’’. International Journal of Numeri-
cal Methods in Engineering, Vol. (37), 229–256.
Beuth, L., Więckowski, Z. & Vermeer, P. 2010, “Solu-
tion of quasi-static large-strain problems with the
23
material point method’’. International Journal of
Numerical and Analytical Methods in Geomechanics,
Wiley Online Library (wileyonlinelibrary.com). DOI:
10.1002/nag.965.
Borja, R.I. 1988, “Dynamics of pile driving by the finite
element method’’. Computers and Geotechnics,
Vol. (5), 39–49.
Burgess, D. Sulsky, D. & Brackbill, J.U. 1992, “Mass
matrix formulation of the FLIP particle-in-cell
method’’. Journal of Computational Physics,
Vol. (103), 1–15.
Coetzee, C.J. Vermeer, P.A. & Basson, A.H. 2005,
“The modelling of anchors using the material point
method’’. International Journal for Numerical and
Analytical Methods in Geomechanics, Vol. (29),
879–895.
Detournay, C. & Dzik, E. 2006, “Nodal mixed discretiza-
tion for tetrahedral elements’’, Proceedinh of ‘4 inter-
national FLAC symposium on numerical modeling in
geomechanics, Itasca Consulting Group.
Harlow, F.H. 1964, “The particle-in-cell computing
method for fluid dynamics’’. Methods for Computa-
tional Physics, Vol. (3), 319–343.
Lysmer, J. & Kuhlmeyer, R.L. 1969, “Finite dynamic
model for infinite media’’. Journal of the Engineering
Mechanics Division, Vol. (95), 859–877.
Sulsky, D. Zhou, S.J. & Schreyer, H.L. 1995, “Application
of a particle-in-cell method to solid mechanics’’. Com-
puter Physics Communications, Vol. (87), 236–252.
Sulsky, D. & Schreyer, H.L. 1996, “Axisymmetric form
of the material point method with applications to
upsetting and Taylor impact problems’’. Computer
Methods in Applied Mechanics and Engineering,
Vol. (139), 409–429.
Więckowski, Z. Youn, S.K. & Yeon, J.H. 1999, “A parti-
cle-in-cell solution to the silo discharging problem’’.
International Journal for Numerical Methods in Engi-
neering, Vol. (45), 1203–1225.
Więckowski, Z. 2004, “The material point method
in large strain engineering problems’’. Computer
Methods in Applied Mechanics and Engineering,
Vol. (193), 4417–4438.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Pile penetration simulation with Material Point Method

L.J. Lim & A. Andreykiv

Plaxis BV, Delft, The Netherlands

R.B.J. Brinkgreve
Delft University of Technology, Delft, The Netherlands
Plaxis BV, Delft, The Netherlands

ABSTRACT: Conventional finite element method (FEM) faces mesh distortion and mesh tangling
when it is used in the simulation of extreme deformation in pile penetration. To avoid the shortcoming
of FEM, material point method (MPM) is used owing its abilityto analyse engineering problems involv-
ing extreme deformation. However, MPM generates numerical noise in the calculation of stresses when
material points cross element boundary due to discontinuity of gradient of shape functions. Dual domain
material point method (DDMP), introduced earlier within an explicit framework, provides a continuity
of the gradient of shape functions which helps to eliminate the numerical noise in stress and strain fields.
In this paper we further extended the application of DDMP within an implicit scheme by formulating a
consistent tangent system. Additionally, we have presented a method to couple MPM and FEM analyses
in order to limit the application of MPM to the areas with extreme deformation, which allows to increase
computational efficiency. Numerical analysis results for a pile penetration problem have been presented
and compared with analytical solution for validation.

1 INTRODUCTION solve this problem as well, especially in large defor-

Large deformation in pile penetration simulation
causes finite element method (FEM) to suffer from
mesh distortion and mesh tangling. The material
point method (MPM), a type of “meshless” meth-
ods, can be used to eliminate these issues. MPM
was first introduced by Sulsky (Sulsky et al. 1995),
where it evolved from a fluid dynamics method
called particle-in-cell (PIC) method, introduced by
Harlow (Harlow 1964). MPM can also be compared
to Arbitrary Lagrangian Eulerian finite element
formulations in the way it separates the motion of
the computational and physical domains.
MPM, however, introduces numerical artificial
noise when the material points “cross” over the
boundary of computation grid (Bardenhagen &
Kober 2004). Non-physical numerical noise in MPM
is caused by the discontinuity of the gradient of
the shape functions across the mesh. This artificial
noise can seriously degrade the accuracy of the solu-
tion, as well as introduce instability especially when
explicit time integration method is used (Zhang et al.
2011). The Generalized Interpolation Material Point
Method (GIMP) has been introduced (Bardenhagen
& Kober 2004) to reduce the numerical noise. Later,
Convected Particle Domain Interpolation (CPDI)
(Sadeghirad et al. 2011) and Dual Domain Material
Point (DDMP) (Zhang et al. 2011) were proposed to
mation problem which GIMP is unable to solve. The
advantage of DDMP over the other two methods is
that it does not require additional particle domain
tracking. In Zhang et al. 2011, DDMP is imple-
mented in explicit integration method. We extended
the method by formulating it with the implicit time
integration method. The implicit time integration is
unconditionally stable, thus larger time step can be
employed to analyse dynamic problems, addition-
ally, it allows solution of static problems.
Meanwhile, the surrounding soil continuum
which is not severely deformed by pile penetration
can be analysed by conventional FEM, while the
MPM will only be used on soil region with large
deformation. This will reduce the computation
load of overall analysis because the MPM analysis
by default is more computationally expensive than
FEM. We have proposed a mesh relaxation method
to couple the analysis of MPM and FEM.
The description of MPM and the implementation
of implicit time integration scheme are written in sec-
tion 2. Meanwhile, section 3 will deal with the formu-
lation of tangent stiffness matrix for implicit DDMP.
In section 4, the MPM-FEM coupling technique with
mesh relaxation method is briefly explained. In sec-
tion 5, the pile penetration modelling in 2D and 3D
will be presented. Next, section 6 discusses the results
of the simulation. Finally, conclusions are drawn.

24
2 IMPLICIT FORMULATION OF MPM As mentioned in Zhang et al. 2011, the numeri-
AND DDMP cal noise is caused by the discontinuity of gradient
function of MPM in Eq. 5 when the internal force
2.1 Governing continuum equations is calculated.
Consider a continuous body Ω ⊂ Rn with a bound-
ary Γ = ∂Ω, the conservation equations for mass 2.3 Implicit time integration scheme
and linear momentum can be defined as,
During a load step, the computation grid will be
dρ solved using conventional FEM. The linearized
+ ρ∇⋅v = 0 (1) equation of motion during the Newton iteration
dt κ for current time step can be defined as (Wieck-
owski 2004),
ρa = ∇.σ + ρb (2)
k 1
where ρ is the mass density, v is the velocity, a is the ij du j ( fiext ) − (f
( fiint )k 1
m ij ⋅ a kj 1
Qkj (6)
acceleration, σ is the Cauchy stress tensor, and b is
the specific body force. where K is the stiffness matrix, mij =
∑ p =1 ρυυ pSi ( p ) S j ( p ) is the mass matrix, duj is
Np

2.2 Discretization of continuum equations the incremental displacement of node j of com-

To solve the continuum equations, the strong form
of the equations are discretized to weak form by
using standard FEM procedure (Wells 2009). After
discretization with FEM, linear momentum equa-
tion becomes,
putation grid, Q is the residual vector,and
k is the iteration step. Eq. 6 is solved itera-
tively, until the residual of the system is less
than a defined iteration convergence criteria
(|Q| < ∈). The displacement update is given as,

N u kj = u kj −1 + duu j (7)
∑ ∫Ωc ρ Si S j d c j ∫Ωc ∇Si σ d Ωc
j =1
Meanwhile, the acceleration term can be calcu-
+ ∫ ρ Si d c ∫Γt tSi d Γc (3) lated by discretizing the time derivative with trap-
Ωc
ezoidal rule. The discretized acceleration term is
where N is the total number of nodes in a compu- given as,
tation domain Ωc, i, j are the nodal indices, aj is the
acceleration at node j, Si is the shape function of 4 k −1 4 0
a kj uj − vj a 0j (8)
node i, t is the surface traction, and Γt is the sur- Δt 2 Δt
face boundary of the computation domain. The
first term of the right hand side of Eq. 3 is defined where v 0j and a 0j terms are the nodal velocity and
as the internal force of the system, fiint. The sum of acceleration at the start of time step.
the second and the third terms of the right hand
side of Eq. 3 can also be defined as the external
force of the system, fiext . Comparable to conven- 2.4 Numerical implementation of implicit
tional FEM, the numerical integration of MPM MPM
over the computation domain Ωc is approximated At the start of a time step, the material points carry
by following, current state variables of acceleration a, velocity v,
Np mass m, volume V, and deformation gradient F.
∫Ωc F d Ωc ≈ ∑ υ p F ( p) (4) Meanwhile, the position of a material point can
p =1
be directly takenfrom the spatial position of mate-
rial points in the computation domain. These state
where F is an arbitrary function to be integrated variables are then interpolated to the computation
over the computation domain, xp is the location of grid using the standard shape function interpola-
material point p and υp is the volume of the mate- tion. The nodal velocity (and nodal acceleration)
rial point p. The internal force vector fiint can be can be interpolated by using conservation of
approximated by, momentum,

Np Np
fiint ∫Ωc ∇Si σ d ∑ p∇Si p (x p )
(5) mi i ∑ ρν pSi ( p )v p (9)
p =1 p =1

25
 In the Updated Langrangian formulation, stiff- gradient term, ∇Si (x) with ∇S i ( ), the internal
ness matrix is defined by, force vector Eq. 5 can be defined as,
Np
K ijk −1 = ∫ ∇Si C T
∇S j d Ω 
Ωc f i int = − ∑ v pσ p : ∇ S i (x p ) (13)
+ ∫ ∇Si σ ∇S j d Ω (10) p
Ωc

where Ωc is the current configuration of computa- To ensure conservation of energy, all the gra-
 i ( ) consistently.
dient quantities have to use ∇S
tion domain, Cσ T is the Truesdell rate of elastic tan-
gent modulus and σ is the Cauchy stress tensor. The The gradient of displacement at location x can be
second identity of Eq. 10 also shows that stiffness calculated by,
matrix includes the term of material nonlinearity N
( ) = u ∇
(first term) and geometrical nonlinearity (second
term). The tangent modulus tensor depends on the
∇ ∑ i S i ( x ) (14)
i =1
type of constitutive model of the material and will
not be elaborated in here. Eq. 6 is solved to obtain The term ∇S  i ( ) has enlarged influence to
incremental displacement du. Computation grid is neighbouring elements, where ∇  i ( ) ≠ 0 if com-
then deformed with du, and kinematics of the sys- putation grid element of node i and computation
tem is then updated before next iteration begins. grid element of location x are neighbouringele-
The update of velocity term is given by, ments. As a result, conventional FEM assembly
method cannot be used. In order to include the
2 k influence of neighbouring elements, the calcula-
v ik = v ik −1 + duu i v i0 (11)
Δt tion of gradient of displacement is done by using
two separate assembly procedures. First, the nodal
while the nodal acceleration a ik is updated by using gradient of displacement is calculated by mapping
Eq. 8. If the Newton procedure is converged with gradient of displacement to the nodes with,
defined convergence criteria, a convective stage of N
) = 1
current time step will be performed. In this stage,
the solution from the computation grids is inter-
∇ j
Vj
∑ ui ∫Ωc S j ∇Si d Ω (15)
i =1
polated back to material points by using shape
functions. By the end of the convective stage, the Subsequently, the gradient of displacement at x
computation grid will be disposed because all the is readily interpolated from the nodal gradient of
required state variables are carried by material displacements by using the shape function,
points to the subsequent time steps.
N
 ) S (x)
∇ ( ) = ∑ (∇ (16)
j j
j =1
3 FORMULATION OF THE STIFFNESS
IN DDMP
Eq. 16 can be computed by using the conven-
3.1 Continuous gradient of DDMP tional FEM local assembly, while Eq. 15 is com-
puted using global tensor multiplication. Because
DDMP calculates the value involving gradient of the integral in Eq. 15 is calculated independently,
the shape function by doing two-step mapping we have redefined it as a tensor,
procedure (Zhang et al. 2011). The method intro-
duces a continuous gradient of the shape function, 1
which is defined as,
Aij =
Vj ∫Ωc S j ∇Si d Ω (17)

N S ( )
i ( ) =
∑ V ∫Ωc S j ∇Si d Ω Meanwhile, the assembly of stiffness matrix also
j
∇ (12)
j =1 j requires the calculation of gradient quantities. By
using the same analogy of mapping material point
stress to the node of computation grid in Eq. 15,
where Vj, is the nodal volume of computation grid
( j ∫  ( x ) is continuous over the
)).. ∇
tangent modulus tensor of material stiffness at
Ωc j material point is mapped to the nodes of computa-
boundary of computation grid because it is calcu- tion grid by,
lated by interpolation of the shape function. This
continuous gradient is able to solve the problem of p N
1
numerical noise caused by discontinuity of gradi- ( στ
)mn ∑
VmVn p
p m(
στ
p )C ( p )Sn ( p) (18)
ent quantities. By substituting the conventional

26
 With the nodal tangent modulus, (Cστ)mn, quan-
tity at location x can be approximated by using
shape function,
1 N N
C T (x) = ∑ ∑ Sm ( x )(Cστ )mn Sn ( x )
VmVn m =1 n =1
Substituting Eq. 19 into the first term of
Eq. 10,
N
1 N N p
Kij = ∑ ∑ ∑ ...
VmVn m =1 n =1 p
× pSm ( p ∇Si ( p)(Cστ )mn n ( xp)∇Sj ( p)
N N
= ∑ ∑ A im ( )mn ATnj
m =1 n =1
The assembly of stiffness matrix is now simply
a product of nodal tangent modulus tensor (Cστ)mn
and A tensors. Eq. 21 is calculated by first assem-
bling the global tensor (Cστ)ij using Eq. 18. This ten-
sor is a container of fourth order tensor described
on nodes i and j. A tensor is also assembled locally
to form a global tensor. Because the size of both
terms is large but the elements are mostly zero val-
ued, sparse matrix is used.
The direct use of continuous gradient may cause
numerical instability because of zero-energy mode,
thus Zhang et al. 2011 introduced a modified gra-
dient of shape function that combines both con-
tinuous and conventional gradient quantities. This
hybrid approach was implemented in this work as
well.
4 COUPLING OF MPM WITH FEM
The analysis of MPM requires more computation
effort than FEM. To couple the calculation process
between MPM and FEM, a mesh relaxation tech-
nique is introduced. In this technique, both MPM
and FEM are sharing the same computation mesh.
A special mesh relaxation procedure is carried out
only on MPM region at the end of the load step
calculation to recover the deformed mesh back to
the initial undeformed configuration. To perform
relaxation procedure, the deformed mesh is applied
with same boundary conditions but the external
force contribution is taken out during the relaxa-
tion process so that the mesh will be deforming
back to initial configuration. Addionally, the stiff-
ness of the FEM region is temporarily increased to
prevent the relaxation procedure from “relaxing”
the mesh on the FEM region. Thus, the relaxation
procedure will only recover the deformed mesh in
the MPM region. To prevent the material points
from travelling out of the computation domain,
a buffer zone is provided on the free surface of the
MPM region.
(19)
5 MODELLING OF PILE PENETRATION
SIMULATION
5.1 Two dimensional model
Fig. 1(Left) shows a symmetrical half of a sheet
pile penetrated into a soil medium. The mesh was
generated using Gmsh (Geuzaine & Remacle 2009).
The model was fixed at the sides horizontally, and
(20) fully fixed at the bottom. The analysis was carried
out by coupling MPM analysis and FEM analysis.
(21) MPM region was located exactly below the pile,
where extreme deformation during pile penetration
process. The mesh was progressively finer towards
the penetration region to simulate more accurate
result. The mesh region with less deformation
was defined as FEM region. Conventional FEM
method was used to calculate this region during
the analysis, and relatively coarse mesh was used to
save calculation cost. In the MPM region and the
pile region, the deformed mesh was relaxed back
to its previous undeformed stage at the end of each
load step to prevent excessive deformation in the
mesh. A buffer zone of meshes was defined on
top of the soil to contain the material points from
translating beyond the computation grid. Inthis
simulation, the contact between the pile and the
soil was simulated by the inherent no-slip contact
of MPM. A very fine layer of mesh was predefined
on the predicted soil and pile contact region to
increase the contact accuracy. The Mohr-Coulomb
failure criterion was used as the constitutive model
of the soil. We chose three soil models with differ-
ent cohesion values of 0.25 kN/m2, 0.50 kN/m2 and
Figure 1. (Left)2D plane-strain model of the pile
penetration simulation. (Right) 3D model of the pile
penetration simulation.
27
1.00 kN/m2 to simulate this pile penetration proc-
ess. The soil has elastic stiffness, Es, of 100 kN/m2
and Poisson ratio of v = 0.33. Meanwhile, the pile
stiffness, Ep, was set to 20,000 kN/m2 and the Pois-
son ratio, v was chosen to be zero. The pile was
defined as a hyperelastic material. Both the soil
and the pile were set to be weightless (γsoil = 0,
γpile = 0).The pile penetration was carried out by
applying prescribed displacement of 0.05 m per
step at the top of the pile until a maximum of
2.5 m penetration depth. Linear triangular ele-
ments were used for the meshes, and 12 material
points per element were prescribed on the Gauss
integration points initially. Separately a compari-
son of mesh fineness at contact region was carried
out with quadrilateral elements to investigate the
influence of rigid contact to the penetration force.
Both MPM and DDMP analyses were carried out
in these simulations.
5.2 Three dimensional model
Fig. 1(Right) shows the 3D model of the simula-
tion. The pile was penetrated into the soil by 0.02 m
per step until a maximum of 2.0 m, or about two
third of the actual pile length. Purely cohesive soil
with cohesion value of c = 1.0 kN/m2 was used
in this simulation. The analysis used linear tetra-
hedron elements, where the finest element size in
the contact area was set to 0.10 m, while the size
was progressively increasing in size to 0.50 m at the
boundary between the MPM region and the FEM
region. A total of 40 material points per element
initially was used.
contact. The maximum shear stresslimit is exactly
the cohesion value of the soil. The shear stress
concentration starts at the corner of pile tip, and
spreads out beneath the soil at about 45°. The soil
is also yielded along the contact region between
the side of the pile and the soil. The average ver-
tical stress on top of the pile has been measured
to determine the pressing force during penetration
process. Fig. 3 shows the stress-penetration depth
plot of the simulation for soil samples with dif-
ferent cohesion values. All the soil samples show
distinct elastic zone and hardening zone. In the
hardening zone, the vertical pile stress increases lin-
early with the penetration depth. The contact gap
between the side of the pile and the soil is directly
influenced by the element sizes in the contact zone.
Three different meshes with varying fineness are
used to examine the effect of mesh refinement on
the analysis. Fig. 4 shows the plots of the vertical
stresses of the pile against the penetration depth
for the three samples of mesh refinements for the
first 1.0 m of penetration. The simulation with

6 RESULTS AND DISCUSSIONS

6.1 2D Simulation Result Figure 3. Vertical stress values of the pile with increas-
ing penetration depth for all three soil samples with
Pile penetration for selected load steps is presented
different cohesion values.
in Fig. 2 for the simulation of soil with cohesion
of c = 0.5 kN/m2. High shear stress concentration
has been located at the corner of the pile and soil

Figure 4. Vertical stress values of the pile with increas-

Figure 2. Shear stress (σxy) for the soil at penetration ing penetration depth for different mesh fineness at the
depth of 1.25 m and 2.50 m for soil with c = 0.5 kN/m2. contact region.

28
coarser mesh has given stiffer response, where
higher transitional stress has been reached before
the soil starts to have hardening response. To vali-
date the stress-displacement response, an analytical
solution is established by calculating the pile load
capacity from the contribution of soil end bearing
capacity and the rigid skin friction. The Terzaghi’s
equation of calculating the bearing capacity for
shallow foundation is used (Das 2007). Meanwhile,
the skin friction of the pile is calculated by taking
the rigid friction resistance of the soil, which is the
cohesion of the soil. The skin friction resistance
can be readily approximated by taking the product
of soil cohesion and total contact area. Thus the
analytical solution plotted as the dashed line in Figure 5. Vertical stress of the soil, σyy, along the sec-
tion beneath the pile, for MPM and DDMP at selected
Fig. 4 is given as, pile penetration depths.
2csoil d
qt = qb + , (22)
b

where, qt is the maximum allowable vertical stress

of the pile, qb is the stress calculated from allowable
bearing capacity proposed by Terzaghi, csoil is the
cohesion of the soil, d is the depth of the penetra-
tion, and b is the width of the pile. The stress value
at the transition zone between elastic response and
hardening response is approximately equal to the
end bearing capacity of the pile. As the mesh refines
further, the stress-displacement response of the pile
is expected to converge to the analytical solution.
Because there is a contact gap between the soil and Figure 6. Shear stress distribution (σxy) in three dimen-
the pile, the pile width is overestimated during the sional pile penetration simulation at penetration depth of
simulation, which causes the simulated bearing 1.00 m, for the MPM analysis and DDMP analysis.
capacity of coarser meshes to have higher values.

6.1.1 Comparison of DDMP with MPM
In this case of comparison between DDMP and
MPM analyses, we have used the results obtained
from the model with quadrilateral mesh with finest
mesh size of 0.0625 m at the contact region. The
initial number of material points per element is 16.
The soil is defined with cohesion, c = 0.5 kN/m2.
MPM analysis results show that vertical stress, σyy,
along the soil below the pile for selected penetration
depths has much higher numerical noise, which is
reduced by DDMP analysis. This is depicted in
Fig. 5. These numerical noisesin MPM has caused
the line search method in Newton iterative proce-
dure to have slower convergence rate than DDMP.
6.2 3D Simulation result
Fig. 6 shows the shear stress colour maps of σxy at
penetration depth of 1.00 m. The maximum shear
stress distribution, which is also the cohesion value
of the soil, are spreading from the tip corner of the
pile. Maximum shear stress of the soil is activated
along the rigid contact next to side of the pile. In
the figure, element discretization effect is clearly
seen in the stress distribution in the MPM result,
where clear discontinuity of stress values can be
seen. This discontinuity is reduced by the DDMP
analysis.
7 CONCLUSIONS
The extreme deformation during pile penetration
has been successfully simulated using MPM and
DDMP, which allow to overcome mesh tangling
problems of conventional FEM. In the simulation
with purely cohesive soil samples, the pile resist-
ance of the simulation is compared with the ana-
lytical solution. The comparison has shown that
the simulation accurately predicted the pile stress,
which is contributed by the end bearing of the pile
and rigid contact resistance between the soil and
the pile. The mesh refinement analysis meanwhile
shows that coarser mesh has produced stiffer pile
resistance. Meanwhile, the DDMP analysis has
shown less numerical noise in the stress results. We

29
have also successfully coupled the MPM and the
FEM analyses using the proposed mesh relaxa-
tion method. The coupled system has shown an
improvement in computation efficiency.
REFERENCES
Bardenhagen, S.G. & Kober, E.M. 2004. The generalized
interpolation material point method. Computer
Modeling in Engineering and Sciences 5(3), 477–495.
Das, B.M. 2007. Principles of Foundation Engineering.
Cengage Publisher.
Geuzaine, C. & Remacle, J.F. 2009. Gmsh: a three-
dimensional finite element mesh generator with built-in
pre- and post-processing facilities. International Jour-
nal for Numerical Methods in Engineering 79(79),
1309–1331.
Harlow, F.H. 1964. The particle-in-cell computing
method for fluid dynamics. Methods in Computational
Physics 3, 319.
Sadeghirad, A., Brannon, R.M. & Burghardt, J. 2011.
A convected particle domain interpolation technique
30
to extend applicability of the material point method
for problems involving massive deformations.
International Journal for Numerical Methods in
Engineering 86, 1435–1456.
Sulsky, D., Zhou, S. & Schreyer, H. 1995. Application of
a particle-in-cell method to solid mechanics. Computer
Physics Communications 87, 236.
Wells, G.N. 2009. The Finite Element Method: An
Introduction. University of Cambridge and Delft
University of Technology.
Wieckowski, Z. 2004. The material point method in large
strain engineering problems. Computer Methods in
Applied Mechanics and Engineering 193, 4417–4438.
Zhang, D.Z., Ma, X. & Giguere, P.T. 2011. Material
point method enhanced by modified gradient of
shape function. Journal of Computational Physics 230,
6379–6398.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Coupling triangular plate and volume elements in analysis

of geotechnical problems

S. Tan
Geo-Engineering Section, Faculty of CITG, Delft University of Technology, Delft, The Netherlands

M.A. Hicks
Geo-Engineering Section, Faculty of CITG, Delft University of Technology, Delft, The Netherlands
Deltares, Delft, The Netherlands

A. Rohe
Deltares, Delft, The Netherlands

ABSTRACT: In geotechnics, it is common to have a thin layered material with high stiffness on top of
soil, to prevent damage from external loading or erosion. To model this numerically a very fine mesh is
often needed, which decreases the critical time step in explicit time integration algorithms, severely affect-
ing simulation performance. The use of 2D plate elements connected to 3D elements is investigated to
overcome this problem. The presented three-noded plate element is based on Kirchhoff thin plate theory,
and uses non-conforming polynomial shape functions. The lumped mass matrix, with both the transla-
tional and rotational degrees of freedom considered, is implemented with an explicit time integration
scheme. This plate element is coupled with volume elements and the implementation is tested for several
cases in which analytical or numerical solutions are available. All simulations show that the plate element
with the lumped mass matrix is working properly in geotechnical problems.

1 INTRODUCTION to construct the lumped mass matrix by neglect-

The plate bending problem was one of the first
problems to which the Finite Element Method was
applied in the early 1960s. A three-noded triangular
plate bending element with nine degrees of freedom
was first derived by Bazeley et al. (1965), based on
Kirchhoff thin plate theory. Although it satisfies
the constant strain criterion, it unfortunately does
not pass the test for arbitrary mesh configurations
(Batoz et al., 1980). However, Specht (1988) pro-
posed a successful modification procedure, which
passes the test perfectly by using three fourth-order
terms in the plate polynomial function.
In order to analyze dynamic problems, the
explicit time integration procedure is often used,
which has been extensively developed for dynamic
analyses to meet the increasing demand of geo-
technical applications. Moreover, the consistent
mass matrix is well established, but it requires
a considerable computational cost and storage
effort. The use of a lumped mass matrix simpli-
fies the program coding and significantly reduces
the computational requirements. However, for
plate elements with both translational and rota-
tional degrees of freedom, it has been customary
ing the inertia effects of the rotational degrees of
freedom (Surana, 1978; Batoz et al., 1980). Unfor-
tunately, this can lead to low accuracy results,
especially when solving geotechnical problems
with the combination of plate elements and vol-
ume elements.
Therefore, a lumped mass matrix, considering
both the translational and rotational degrees of
freedom, is introduced for the three-noded trian-
gular plate element. The modified non-conform-
ing shape functions based on area coordinates are
explained in detail, and the lumped mass matrix for
the plate element is derived without neglecting the
rotational degrees of freedom. Based on the plate
theory, the explicit time integration procedure for
dynamic analysis is presented. By assembling the
plate and volume elements in a 3D code, the plate
elements can be used to model a thin layer material
instead of volume elements.
For validation purposes, the implementation is
tested for several cases in which analytical solutions
or numerical results are available. 2D problems of
a square cantilever plate and a circular plate are
considered, followed by a plane strain problem of
a plate resting on an elastic foundation and a cir-

31
cular concrete plate resting on a homogeneous half
space. Finally, the conclusions are provided.

2 METHODOLOGY

Kirchhoff plate theory is an extension of the Eul-

er-Bernoulli beam theory for thin plates. It uses the
mid-surface plane to represent the three dimen-
sional plate in a two dimensional form.
The fundamental assumptions are that straight
lines normal to the mid-surface are infinitely rigid
and that, after deformation, they remain straight
and normal to the mid-surface (Reddy, 2007; Figure 1. Area coordinates of triangular element.
Bauchau & Craig, 2009). Experimental measure-
ments show that these assumptions are valid for thin
plates made of homogeneous, isotropic materials. where bi = yj − yk, ci = xk − xj; x and y are the coor-
All the equations are based on these assumptions. dinates of the nodes (Figure 1); i, j, k are cyclic
permutations of 1, 2, 3; and P is the polynomial
2.1 Shape functions with area coordinate system expression used to define the shape functions in
terms of nine parameters,
For the three-noded triangular plate element, three
degrees of freedom (one deflection and two rota-
tions) at each vertex are used as nodal variables. P = [ 1, L2 , L3 , L1L2 , L2 L3 , L3L1,
The nodal displacement vector at node i can be 1
defined as ai. The element displacement will there- L12 L2 + L1L2 L3{ (1− μ3 ) 1 (1 3 μ3 ) 2 (1 3 μ3 ) 3 },
2
fore be given by a listing of nine nodal displace-
1
ments, i.e. {3(1 μ1 )
L22 L3 + L1L2 L3{3 2 (1 3 μ1 ) 3 (1 3 μ1 ) 1},
2
1
⎧a i ⎫ ⎧wi ⎫ {3(1 μ2 )
L23 L1 + L1L2 L3{3 (1 3 μ2 ) (1 3 μ2 ) 2 }]
⎪ ⎪ ⎪ ⎪ 2
3 1
a e = ⎨a j ⎬ , a i = ⎨θ xxi ⎬ (1)
⎪⎩a k ⎪⎭ ⎪⎩θ yyi ⎪⎭ (5)
where μi is defined as,
in which wi is the deflection of the node; and θxi and
θyi are the rotations relative to the x- and y-axes of lk2 l 2j
the plate, respectively, and are defined as, μi = (6)
li2
∂w ∂w
θ xi , θ yi = − (2) and li is the length of the triangle side opposite to
∂y i ∂x i node i (Figure 2).
L1, L2 and L3 are the area coordinates, which are
The continuous displacement variable, in this defined by the following linear relationship with
case the deflection w, is approximated in terms of the Cartesian system,
discrete nodal values as follows,
x L1x1 L2 x2 L3x3
w ≈ Na e = [ Ni N j N k ][a iT a Tj a Tk ]T (3) y L1y1 L2 y2 L3 y3
(7)
1 = L1 L2 + L3
where N is the matrix of shape functions derived
by Specht (1988), given by,
The coordinate L1 of a point P is defined as the
ratio of the area of the shaded triangle to that of
⎧Pi Pi + 3 + Pk 3 2( Pi + 6 − Pk + 6 ) ⎫ the entire triangle, i.e.
⎪⎪ ⎪⎪
NiT = ⎨−bb j Pk +6 + 6 − Pk 3 bk Pi + 6 ⎬
⎪ ⎪ (4) area P 23
L1 = (8)
⎪⎩−c j ( Pk + 6 − Pk +3
+ ) − ck Pi + 6 ⎪⎭ area 123

32
 where Ai indicates the area related to node i in an
element, as shown in Figure 2.

2.3 Equations of motion in dynamic analysis

Considering a dynamic process using an explicit
time integration scheme, the discrete equation of
motion can be written as,

M L w = F ext Fint (11)

where w is the displacement vector, which includes

Figure 2. Mass lumping for 3-node triangular plate deflections and rotations; ML is the lumped mass
element.
matrix of the plate element; Fext is the nodal exter-
nal force vector, including the external traction
2.2 Lumped mass matrix and body forces; and Fint is the nodal internal force
vector resulting from the bending and torsion
The lumped mass matrix is built up by mapping moments. The internal force vector is given by,
the mass of the surrounding elements to the respec-
tive nodes of the elements. Considering the three
Fint ∫S B
T
degrees of freedom of each node, the construction MdS (12)
of the mass matrix consists of two different types
of terms. For the translational degree of freedom, where S is the area of the element, and M is the ele-
it is equivalent to the normal mass matrix. The ment bending moment vector, which contains Mxx,
mass of the element is divided into three parts by Myy and Mxy. It is calculated as,
the central point of the element (Figure 2). Each
node in the element carries 1/3 of the total mass
(m) and it goes to the term for the translational M DBa e (13)
degree of freedom.
In classical mechanics, the moment of inertia in which B is the matrix of the second derivative of
is the property of a mass in space that measures the shape functions, namely,
its resistance to rotational acceleration around an
axis. It plays the same role in rotational motion as T
⎡ ∂2 ∂2 ∂2 ⎤
mass does in translational motion, describing the B ⎢ 2 2 ⎥ ⎡ Ni N j N k ⎤⎦ (14)
relationship between the moment of momentum ⎣ dx dy 2
dxdy ⎦ ⎣
and angular velocity, torque and angular accelera-
tion. Therefore, the moment of inertia (J) is used and D is called the “flexural rigidity” which is anal-
for the terms of the rotational degree of freedom. ogous to the bending stiffness EI of a beam, i.e.
It is calculated in the local coordinate system of
each node, as shown in Figure 2. Therefore, the ⎡ ⎤
diagonal terms of the mass matrix are, ⎢ D vD 0 ⎥
Et 3
D = ⎢vD D 0 ⎥, D = (15)
⎢ ⎥ 12( − 2 )
⎡m
ML = ⎢ J x1 J y1
m
Jx 2 J y2
m ⎤
J x 3 J y3 ⎥ (9) ⎢0 0 (1 − v )D ⎥
⎣ 3 3 3 ⎦ ⎣ 2 ⎦

where the moments of inertia can be defined as, where t is the thickness of the plate and ν is Pois-
son’s ratio.
In order to reach an the equilibrium state for the
J x1 ∫∫ y12dx
d 1dy
d1 J y1 ∫∫ x12dx1dy1 dynamic problem, the local damping is introduced
A1 A1 as defined in FLAC (1998) by adding an artificial
damping term into equation (11), i.e.
∫∫ ∫∫ ρx2 dx2dy2
2 2
Jx 2 y2 dx2dy2 J y2 (10)
A2 A2
M L w = F ext Fint − F damping (16)
∫∫ ∫∫
2 2
Jx3 y3 dx
d 3dy
d 3 , J y3 x3 dx
d 3dy3
A3 A3 in which Fdamping is the damping force vector. Its
components are defined as,

33

Fi damping α Fi ext − Fiint sig (wi ), i=1,2,
where α is the damping factor and Ndof is the total
number of degrees of freedom.
3 NUMERICAL SIMULATIONS
OF 2D PROBLEMS
The plate element presented above has been imple-
mented in a 3D code and coupled with tetrahedral
elements for representing the underlying soil. The
implementation has been tested for both 2D and
3D cases in which analytical or numerical solutions
are available for validation.
3.2 Circular plate with simply supported edge
Ndof
Figure 4 shows a simply supported circular plate
(17) with a point load applied at its centre. The plate is
modelled as linear elastic, with a Young’s modu-
lus E = 200 MPa, and Poisson’s ratio ν = 0. The
radius of the plate is a = 0.45 m and the thickness is
t = 0.1 m. The applied point load is q = 10 kN.
The analysis has been done using a local damp-
ing factor of 0.5 to reach the equilibrium state.
Figure 5 shows the deflection of the loaded point
as a function of time, with the final deflection of
7.24 mm being in close agreement to the analytical
solution of 7.25 mm. Figure 6 shows the contour
plot of the deflection of the plate.

3.1 Cantilever square plate

The plate element was first tested for a 2D problem;
this being the modelling a square plate, clamped
at one end and subjected to a vertical line load at
the free end, as shown in Figure 3. The plate has
a thickness of 0.01 m. The material is consid-
ered to be linear elastic, with a Young’s modulus
E = 180GPa, Poisson’s ratio ν = 0, and density
ρ = 1000 kg/m2. The magnitude of the lateral dis-
tributed load is assumed to be 2 kN/m. Figure 4. Geometry of circular plate: a. elevation of the
Three finite element meshes have been consid- mesh; b. side view.
ered and the deflection of the corner point B for
each mesh is listed in Table 1, along with the ana-
lytical result. As can be seen in Table 1, the calcula-
tion error reduces to almost zero with increasing
number of elements, indicating that the plate ele-
ment performs well.

Figure 5. Plot of deflection of the loaded point.

Figure 3. Cantilever plate subjected to a uniform end

load.

Table 1. Comparison of results for deflection of Point B.

Deflection Analytical result

(m) (m) Error

4 elements 0.0433 2.47%

8 elements 0.0437 0.0444 1.58%
12 elements 0.0442 0.45% Figure 6. Contour plot of deflection of the circular
plate.

34
4 NUMERICAL SIMULATIONS
OF 3D PROBLEMS
In order to solve the problem of a thin layered
material resting on soil, the 2D plate element has
been implemented in a 3D finite element code
which uses tetrahedral elements for modelling the
soil volume. In this case, the assembling nodes
share information of both the plate element and
tetrahedral element, with five degrees of freedom,
three for translation and two for rotation. Because
of the different shape functions of the plate and
volume elements, the continuity equation is only
applied to the vertical displacements of the nodes,
which belong to both types of element.
4.1 Plate resting on elastic foundation
A plate resting on an elastic foundation has been
considered by using the 3D code to model what
is a plane strain problem, as shown in Figure 7.
The plate has a virtual thickness of 0.05 m, and
a uniform load of p = 12 kN/m2 has been applied
to the middle area of the plate over a width of
0.1 m. Both the plate and soil are modelled as
linear elastic, with the Young’s modulus for soil
being 2000 kN/m2 and for the plate 20000 kN/m2,
and with both having a Poisson’s ratio of 0.0. The
dimensions of the problem are shown in Figure 7.
As expected, the highest stresses and displace-
ments are concentrated just below the loaded area,
as indicated in Figures 8 and 9, respectively. For
Figure 7. Geometry of centrally loaded plate resting on
elastic foundation.
35
Figure 8. Contour plots of vertical stress in the soil
below loaded plate: a. results using implemented plate
element; b. results using Plaxis.
Figure 9. Contour plots of vertical displacement in
the soil below loaded plate: a. results using implemented
plate element; b. results using Plaxis.
comparison, the same problem has been analyzed
by the Plaxis 2D code using Mindlin plate ele-
ments. Similar results are obtained for both the
stresses (Figure 8b) and displacements (Figure 9b),
although the values from Plaxis are slightly higher
than those from the 3D code. However, although
the two meshes are similar, Plaxis 2D uses 6-noded
high order elements, while the 3D code uses low
order elements, which is likely to be the reason for
the slight differences in the results.
4.2 Concrete plate resting on elastic soil
In this subsection, a circular plate has been used to
model a concrete layer resting on top of a homoge-
neous elastic soil layer. The plate is subjected to a
load q = 12 kN/m2 applied centrally over a circular
area of 10 cm diameter. As shown in Figure 10, the
discretized region has a diameter of 90 cm, and a
depth of 80 cm. The plate has a virtual thickness
of 5 cm.
The soil is idealised as linear elastic, with a
Young’s modulus of E2 = 2000 kN/m2 and Pois-
son’s ratio of ν = 0.0. The plate is modelled as lin-
ear elastic, with a Poisson’s ratio of ν = 0.0. In order
to study the interaction between the soil and the
plate, different values of Young’s modulus are used
for the concrete plate, namely, E1 = 10000 kN/m2,
20000 kN/m2 and 40000 kN/m2.
 Figure 11. Vertical stresses in a two layer system
(Burmister, 1958).

Figure 10. Geometry of loaded plate resting on half

space.

This problem can be considered as a two-layer

problem. The concrete plate significantly reduces
the stresses and deflections in the soil, especially
when the modular ratio of the plate and soil (E1/E2)
is large. Solutions for the two-layer problem have
been obtained by Burmister (1945) using strain
continuity equations, which indicate the depend-
ency on E1/E2. Figure 11 shows the vertical stress Figure 12. Contour plots of vertical stress: a. 3D view;
b. cross section through problem (E2 = 20000 kN/m2).
ratio below the centre of the circular plate for the
two-layer system (Burmister, 1958; Yoder & Witc-
zak, 1975), as a function of E1/E2.
After applying the load to the circular area,
stresses and deflections are generated in the soil.
The soil just below the loaded area experiences
the highest stresses, as shown in Figure 12, for the
plate with a Young’s modulus of E2 = 20000 kN/m2.
The vertical stresses in the soil decrease with depth,
but the maximum value occurring at the interface
between the plate and soil is equal to 3.33 kPa,
which is 27.8% of the applied load. According to
Figure 11, for a modular ratio of 10, the vertical Figure 13. Contour plots of settlement: a. top view;
stress at the interface of the two layers is approxi- b. cross section through problem (E2 = 20000 kN/m2).
mately 30% of the applied load. As the maximum
value obtained in the numerical analysis is com- 4.62 kPa and 2.33 kPa, respectively. Once again, there
puted at the stress point of the volume element is a good agreement with the theory (Figure 11).
just below the loaded area, which means that the
parameter z/a in Figure 11 is slightly bigger than
1, the vertical stress is consequently less than 30%. 5 CONCLUSIONS
Therefore, the obtained stresses in the numerical
analysis are reasonable compared with the analyti- Based on Kirchhoff thin plate theory, the lumped
cal results. Figure 13 shows the contour plots of mass matrix of the three-noded triangular plate
the settlements after loading. element has been introduced, considering both the
The calculations were also done with plate Young’s translational and rotational degrees of freedom.
moduli of 10000 kN/m2 and 40000 kN/m2. The In combination with high order non-conforming
maximum vertical stresses occur for both cases at polynomial shape functions, the explicit time inte-
the same stress point as before, with magnitudes of gration scheme was presented for dynamic plate

36
analysis. The static problems considered in this
paper were solved using a transient dynamic pro-
cedure with the dynamic relaxation method. Dif-
ferent cases were studied in 2D and 3D, and the
results were validated using analytical or numerical
solutions. All simulations showed that the lumped
mass matrix works properly with the plate ele-
ment. The plate element can be applied to layered
geotechnical problems, to simulate a thin layered
material instead of using volumetric elements.
ACKNOWLEDGEMENT
This research is funded by the CSC (China Scholar-
ship Council), TU Delft and Deltares. The authors
appreciate the contributions of Professor Vermeer
from Deltares, and Professor Wieckowski from
Lodz University of Technology and Deltares, who
gave valuable advice while doing this research.
REFERENCES
Batoz, J.L., Bathe, K.J. & Ho, L.W. 1980. A study of
three-node triangular plate bending elements. Interna-
tional Journal for Numerical Methods in Engineering,
15: 1771–1812.
Bauchau, O.A. & Craig, J.I. 2009. Structural analysis
with applications to aerospace structures. Springer,
The Netherlands.
37
Bazeley, G.P., Cheung, Y.K., Irons, B.M. & Zienkie-
wicz, O.C. 1965. Triangular elements in plate bending
conforming and non-conforming solutions. Proceed-
ings Conference on Matrix Methods in Structural
Mechanics, Wright Patterson A.F.B., Ohio. 547–576.
Burmister, D.M. 1945. The general theory of stresses and
displacements in layered systems. International Jour-
nal of Applied Physics, 16: 89–94.
Burmister, D.M. 1958. Evaluation of pavement systems
of the WASHO road test by layered systems methods.
Highway Research Board Bulletin 177.
FLAC 1998. Fast lagrangian analysis of continua: theory
and background. Itasca Consultin Group, Inc., Min-
nesota, USA.
Reddy, J.N. 2007. Theory and analysis of elastic plates and
shells. CRC, Taylor and Francis.
Specht, B. 1988. Modified shape functions for the three-
node plate bending element passing the patch test.
International Journal for Numerical Methods in Engi-
neering, 26: 705–715.
Surana, K.S. 1978. Lumped mass matrices with non-zero
inertia for general shell and axisymmetric shell ele-
ments. International Journal for Numerical Methods in
Engineering, 12: 1635–1650.
Yoder, E.J. & Witczak, M.W. 1975. Principles of pave-
ment design (2nd Edition). John Wiley & Sons.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Fracture growth in heterogeneous geomaterials

P.J. Vardon & J.D. Nuttall

Department of Geoscience and Engineering, Delft University of Technology, Delft, The Netherlands

ABSTRACT: A method to simulate fracture growth in heterogeneous geomaterials is presented. The

method links a statistical description of spatial material properties, random fields, with the extended
finite element method. A probabilistic description of the growing fracture is then generated which can be
incorporated into risk and reliability design based methods. Initial model development, verification and
output is presented.

1 INTRODUCTION XFEM provides non-continuity representa-

Geomaterials are by their very nature heterogene-
ous (e.g. Hicks 2007). This affects material behav-
iour and causes uncertainty in analysis and design.
Non-continuous behaviour, such as fracturing and
fracture flow, occurs often in installation of geo-
technical structures and is highly influenced by
heterogeneity in geomaterials in situations where
the process is not highly geometrically dominated.
In fact, in geomaterials, an accurate deterministic
description of material properties and their spatial
variation is difficult, therefore realistic determinis-
tic descriptions of fracturing processes are equally
difficult to describe. Consequently adequate predic-
tion is almost impossible. For continuum (finite ele-
ment) modelling a recent advance is the inclusion
of statistical descriptions of material properties
and a probabilistic description of material response
(Hicks & Spencer 2010). This paper provides the
initial steps towards extending this approach to the
simulation of fracture growth behaviour.
2 EXTENDED FINITE ELEMENT
METHOD
The eXtended Finite Element Method (XFEM)
allows the representation of non-continuous and
non-mesh dependent properties and processes to
be incorporated into the Finite Element Method
(FEM) (e.g. Belytschko & Black 1999; Bordas
et al. 2007). In particular the advantage is that non-
continuities, such as fractures, can be represented
easily where they do not coincide with the mesh
topology. Therefore moving non-continuities, e.g.
fracture growth, can be represented without the
need for explicit remeshing.
tion via locally enriched shape functions, e.g.
Belytschko & Black (1999):
n ⎡ ne ⎤
uh ( ) ∑ Ni ⎢ui + v( )ai ⎥
∑ v( (1)
i =1 ⎢⎣ j 1
j= ⎥
⎦
where uh is the variable numerical approximation,
x = ( , z )T , Ni is the traditional shape function, n is
the number of nodes, ne is the number of enriched
nodes, v is the enrichment function and a is the
enriched degree of freedom.
3 MATERIAL HETEROGENEITY
3.1 Geomaterials
Materials such as those found in the subsurface,
called in this paper geomaterials, are often inher-
ently variable due to the processes in which they are
created and subsequent material history. Therefore
the materials are not randomly variable, but have
variability correlated to location and formation
(e.g. Hicks & Spencer 2010).
3.2 Random fields
Material heterogeneity was initially proposed to be
statistically quantified via random fields by Fenton &
Vanmarcke (1990). The spatial variation of materi-
als is described via a set of statistics e.g. the amount
of variability, the rate of change and the distance
between maximum variation. Mapping this descrip-
tion onto the geometric description of the models
(the domain) allows a variable material to be included
in simulation. A single geometric representation
achieved in this manner is known as a ‘realisation’.

38
3.3 Random finite element method
To incorporate the effect of heterogeneous materials
into the finite element method, the Random Finite
Element Method (RFEM) has been developed,
where random spatial perturbations in material
properties are included in analyses via the incorpo-
ration of random fields e.g. Hicks and Samy 2002;
Hicks and Spencer 2010 and as described in sec-
tion 3.2. In this approach, the material properties
are represented by a mean and standard deviation,
and by a definition of the spatial variation and cor-
relation. This leads to a stochastic analysis, where
multiple realisations (100 s or 1000s) are analysed
and compiled. The results of the analyses may be
quantified in terms of risk and reliability, in con-
trast to traditional deterministic analyses which
generally result in a single characterisation e.g. a
single factor of safety.
the method of local average sub-division (Fen-
ton & Vanmarcke 1990) to maintain global mean
properties.
The material property that is varied can be
expressed as:
u( ) = μ( z ) + σ( z )Z ( ) (3)
where u is the material properties to be varied, μ(z)
is the mean, σ(z) is the standard deviation and Z(x)
is the local average. The local average is based upon
an Gaussian distribution and a spatial correlation
function, giving a covariance function for an iso-
tropic random field, after Hicks & Samy (2002)
and Hicks & Spencer (2010), of:
β τ ) = σ 2 exp ⎛ − τ ⎞
2
(4)
⎝ θ ⎠

where β is the local covariance, τ is the lag and θ is

4 RANDOM-XFEM
4.1 Approach
The approach a deterministic XFEM implemen-
tation and executes multiple realisations, with
material properties randomised spatially in each
realisation using the random field technique. A
Monte Carlo analysis is then carried out to estab-
lish the range of behaviour.
4.2 XFEM
An existing XFEM implementation was utilised
(Pais 2010), offering structured grids and shape
function enrichment for both strong and weak
discontinuities. In this paper only fracture meth-
ods are considered. The Heaviside step function is
used as the enrichment function to represent the
discontinuity in elements that are fully split by a
fracture and linear elastic crack tip displacement
fields are used in elements that contain a fracture
tip. Crack growth angle is modelled utilising the
maximum circumferential stress criterion (Shih &
Asaro 1988) utilised by Moes et al. (1999):
the spatial scale of fluctuation.
Figure 1 shows an example of a random field
with a log-normal distribution for Young’s modu-
lus, E, with a spatial variation, θ = 1.0 m, mean,
Eμ = 1 × 109 kN/m2, and standard deviation
Eσ = 2.5 × 109 kN/m2, with domain dimensions of
6 m × 4 m.
4.4 Deterministic analyses
Deterministic analyses are able to be simulated and
the results are taken as a single possible outcome for
a domain defined by the material statistics. Utilising
a standard deviation of zero yields the same results
as a continuum model including no material hetero-
geneity. This can be used as the reference case.

⎡ ⎛ 2 ⎞⎤
⎛ K ⎞
θc = 2 tan−1 ⎢⎢ ⎜ I − sign ⎥
1 K
i ( K II ) ⎜ I ⎟ + 8 ⎟ ⎥
4 ⎜ K II ⎝ K II ⎠ ⎟
⎢⎣ ⎝ ⎠ ⎥⎦
(2)
where KI and KII are stress intensity factors.

4.3 Random field realisations

A random field generator is used following the Figure 1. Typical example of a Young’s modulus,
approach of Hicks & Spencer (2010) utilising E (kN/m2) random field.

39
4.5 Monte Carlo analysis
Monte Carlo analyses can then be undertaken uti-
lising a series of possible realisations, each realisa-
tion valid based upon the material statistics, and
the results compiled. Each realisation consists of
the random field properties being mapped to the
finite elements and then a deterministic XFEM
analysis. The resulting analysis provides a possi-
ble outcome, therefore the compilation of results
provides the range of possible outcomes. Suffi-
cient realisations must be undertaken so that the
required response of the system, in terms of prob-
ability, has converged. In terms of crack propaga-
tion the required response is likely to be in terms of
geometry, i.e. length, direction, angle, etc.

4.6 Justification
Bordas et al. (2007) showed that the inclusion of soft
and hard inclusions, with contrasting Young’s mod-
uli, created deviations within the crack propagation
as shown in Figure 2, with the analyses repeated by
the Authors’ here for verification purposes.
The presence of the inclusions in Figure 2, illus- Figure 3. Plots of predicted crack propagation pathway
trate that the hard inclusion (Figure 2(a)) repels in the presence of a soft inclusion of varying radii (solid
crack propagation while that of the softer inclu- line propagation path and dashed line inclusion).
sion (Figure 2(b)) attracts the path. This numerical
experiment shows and justifies the need to investi-
gate further the affects of spatial variability within
material properties that vary continuously within a
material when analysing crack propagation.

4.6.1 Inclusion size and placement

The Authors undertook further numerical experi-
ments to try to understand the influence of the

Figure 4. Illustration of predicted crack propaga-

tion in the presence of varying material properties.
(Blue, E = 107 kN/m2, Green, E = 108 kN/m2 and Red,
E = 109 kN/m2).

material variability. This included varying the

Figure 2. Illustration of predicted crack propagation in
the presence of (a) hard and (b) soft inclusions. (Blue,
E = 107 kN/m2 and Red, E = 108 kN/m2).
size of the inclusion in the Bordas et al. (2007)
experiment.
Figure 3 shows a plot of the crack propagation
for the soft inclusion of varying radii, all centred at
the same location.
The results indicate that the size and proximity of
the inclusion to the crack effects the path of propa-
gation. Furthermore, the proposed inclusion of a
random field to model material properties will have

40
a analogous effect, with the propagation paths vary- It can be seen that the two paths are signifi-
ing with the spatially varying Young’s modulus. cantly different and show that the spatially vary-
Figure 4 illustrates the predicted crack propaga- ing Young’s modulus has a significant effect on
tion in the presence of regularly placed regions of the direction and path of the crack propagation.
contrasting Young’s moduli. Figure 7 shows the propagated crack paths for 100
realisations of the methodology.
The predicted paths are shown to be relatively
5 RESULTS AND DISCUSSION smooth, as would be expected due to the local aver-
aging within the random field. This is a character-
5.1 Initial model istic that may change as the random fields become
more abruptly changing, with larger gradients
A model of uni-axial tensile stress was modelled
within the material properties, changes in the mate-
with a central edge crack within a 6 m × 4 m
rial behaviour models, or discrete changes in mate-
domain, as shown in Figure 5.
rial properties. In fact, the model illustrated here,
The boundary conditions on the base of the
shows a material with slowly changing properties,
domain restrict vertical displacement and horizon-
or as the results of material averaging, considering
tal displacement is restricted on the left corner. At
a rather large representative elementary volume
the top of the domain a uni-axial vertical load (of
(or at least significantly larger than grain size).
unity) is enforced. No explicit boundary condi-
Where grain level behaviour, or other local behav-
tions are enforced on the fracture surface allowing
iours, causes much stronger discontinuities these
displacement to occur, but not to support load.
effects can also be incorporated into this model,
The material variation is then model via the
either by altering properties on a element basis or
generation of a random field for Young’s modu-
via XFEM enrichment itself.
lus, E, with a log-normal spatial variation where
θ = 1.0 m, mean, Eμ = 1 × 109 kN/m2, and standard
deviation Eσ = 2.5 × 109 kN/m2. The variation is
then mapped to the domain, during each realisa-
tion of the analysis.
The crack is then allowed to propagate through
the domain using 40 iterative steps of 0.1 m with
the direction of the fracture determined via the cir-
cumferential stress as shown in equation (2).

5.2 Example results

Figure 6 illustrates the path of fracture propa-
gation through the modelled materials for two
example random fields with the same statistical
properties.

Figure 6. Two example realisations of the predicted

crack propagation, including the two random fields of
Figure 5. 6 m × 4 m domain with initial 0.5 m edge crack Young’s modulus generated using identical problem
(blue horizontal line half way down left hand side). statistics.

41
Figure 7. Predicted crack propagation paths of 100
realisations.
This simple initial numerical experiment shows
that by spatially varying the Young’s modulus, by
means of random fields, results in deviations in the
propagation paths of the cracks. There is likely to
be an infinite number of statistically relevant path-
way solutions due to the infinite number of possi-
ble random field variations. These results indicate
that utilising a single deterministic analysis could
cause significant design oversights. Therefore a
probabilistic means evaluating problems of this
nature would be advantageous.
Probabilistically the methodology can used to
provide predictions of pathway or network likeli-
hood as illustrated in Figure 8. This may provide
significant design advantage, for example where
fracture paths tend towards formation bounda-
ries or are distributed (non-) uniformly around
structures. However this methodology can require
several hundreds if not thousands of realisations
and therefore significant computational resources.
However, with more recent computational devel-
opments, where many cores can be situated on a
single chip, these sort of analyses can become more
commonplace. Furthermore many other probabil-
istic performance measures can be easily incorpo-
rated into this methodology and therefore it can
be introduced into reliability and risk based design
and project management tools.
5.2.1 Future work
This work is an initial investigation into the feasi-
bility, reliability and usefulness of the methodology
presented. It is anticipated to be extended into a
coupled hydro-mechanical formulation to increase
validity to further geomechanical processes.
The robustness of the methodology will be fur-
ther tested and compared with experimental results
from the laboratory testing.
The Authors are also keen to incorporate crack
initiation into the model, not only as this aspect
is of practical importance, but also as the meth-
odology should be of particular use when predict-
42
Figure 8. Percentage probability of a fracture occurring
at any location using 100 realisations.
ing the positions of likely crack initiation, and the
probability of its occurrence.
6 CONCLUSIONS
A method to simulate the fracturing process in het-
erogeneous geomaterials is presented. Initial results
demonstrate that coupling random fields into the
eXtended Finite Element Method provides a range
of equally valid solutions. Further analysis can
yield a probabilistic depiction of the fracture and
fracture growth process.
REFERENCES
Belytschko, T. & Black, T. 1999. Elastic crack growth in
finite elements with minimal remeshing. Int. J., Numer.
Meth. Engng. 45: 601–620.
Bordas, S., Nguyen, P.V., Dunant, C., Guidoum, A. &
Nguyen-Dang, H. 2007. An extended finite element
library. Int. J. Num. Meth. Engng. 71 (6): 703–732.
Fenton, G.A. & Vanmarcke, E.H. 1990. Simulation of
random fields via Local Average Subdivision. ASCE J.
Eng. Mech. 116(8): 1733–49.
Hicks, M.A. editor. 2007. Risk and variability in geotech-
nical engineering. London: Thomas Telford.
Hicks, M.A. & Samy, K. 2002. Influence of heterogeneity
on undrained clay slope stability. Quart. J. Eng. Geol.
Hydrogeol. 35(1):41–9.
Hicks, M.A. & Spencer, W.A. 2010. Influence of het-
erogeneity on the reliability and failure of a long 3D
slope. Computers and Geotechnics 37 (7–8): 948–955.
Moes, N., Dolbow, J. & Belytschko, T. 1999. A finite ele-
ment method for crack growth without remeshing.
Int. J. Numer. Meth. Engng. 46: 131–150.
Pais, M.J. 2010. MATLAB eXtended Finite Element
Method (MXFEM): User guide. Gainesville, Florida:
University of Florida.
Shih, C. & Asaro, R. 1988. Elastic-plastic analysis of
cracks on biomaterial interfaces: part I—small scale
yielding. J. App. Mech. 55: 299–316.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Modelling rate-dependent behaviour of structured clays

M. Karstunen
Chalmers University of Technology, Gothenburg, Sweden
University of Strathclyde, Glasgow, Scotland, UK

N. Sivasithamparam
Plaxis BV, Delft, The Netherlands
University of Strathclyde, Glasgow, Scotland, UK

R.B.J. Brinkgreve
Plaxis BV, Delft, The Netherlands
Delft University of Technology, Delft, The Netherlands

P.G. Bonnier
Plaxis BV, Delft, The Netherlands

ABSTRACT: Due to the desire of reducing the embedded CO2 in construction and the pressure in public
finances to get more value for money in big infrastructure projects, the demands for the accuracy of deforma-
tion predictions increase. Instead of piling, alternative environmentally friendly and cost effective solutions,
such as preloading via surcharge, vertical drains and column methods, such as deep-mixing, are becoming
increasingly attractive. Installation of piles and ground improvement into the ground will modify the state of
the soil. This is sometimes beneficial, and sometimes detrimental, and so far this effect has been rarely taken
into account. One reason for this is that the numerical techniques and the constitutive soil models have not
been able to represent the changes in soil structure and state in a satisfactory manner. The aim of the Europe-
an GEO-INSTALL project has been to develop numerical techniques that can be used to model installation
effects in geotechnical engineering. A key part of this has been constitutive model development, and their ro-
bust implementation. The aim of this paper is to discuss some recently developed rate-dependent constitutive
models for structured clays, which formed the basis for new developments, resulting in a new rate-dependent
model able to represent the complex rate-dependent stress-strain behaviour of soft structured clays. The im-
portance of modelling key features of soil behaviour in the context of rate-dependency are discussed in the
light of experimental evidence, and demonstrated through a series of numerical benchmark simulations.

1 INTRODUCTION drains, deep mixing and stone columns, are becoming

Infrastructure construction of soft soils relies on rep-
resentative predictions on long-term deformations,
as often the serviceability considerations are con-
trolling the design solution. This is particularly true
for construction of embankments on soft soils. For
example, in Sweden the earthwork and foundation
costs are typically over 20% of the total construction
cost (SGI 1995), and can form even a higher propor-
tion than that in major infrastructure projects, due
to the soft ground conditions. With the global drive
to reduce the embedded CO2 in construction and
the increasing challenges in both public and private
finance, there is increasing need to consider design
options that can be used as alternatives to rather
CO2-heavy and relatively expensive solutions, such as
piling. Hence, methods such as preloading, vertical
increasingly attractive. This puts additional demands
on the accuracy of deformation predictions. In addi-
tion, more and more construction occurs in densely
populated urban areas on poor ground conditions
and the effects on structures nearby is important.
This creates new demands for soil modelling and
numerical analyses, as due to the complexity of the
materials involved, and the complexity of the actual
problems, no analytical solutions exist.
Most natural soils, perhaps with the exception
of extremely fibrous peats, can be considered as
particulate materials, which when fully saturated
consist of an assembly of soil particles surrounded
by pore water. The strength and stiffness of the
soils depends on the arrangement and packing of
the soil particles, in particular the assembly of the
intergranular contacts, as well as the presence of

43
any apparent bonding between the particles. The
latter results from natural bio-chemical processes
such as the precipitation of calcites by bacteria in
the soil and/or some complex geo-chemical pos-
sesses associated with sedimentation environment
and subsequent geological and flow history.
Particularly fascinating is this aspect are the so-
called sensitive clays, which in extreme cases are
referred to as quick clays. Highly sensitive clays can
be found in large areas of Northern Europe and
North America (Rankka 2003), and they tend to
change from rather stiff consistency to a flowing
liquid gel when disturbed, forming a potential geo-
hazard. Sensitive clays were formed when clay parti-
cles and other fine fractions were sedimented in cold
and salty marine environment after the Pleistocene
epoch. Due to the bi-polar electro-chemical charges
associated with the clay minerals, in salty sedimen-
tation environments the clay particles flocculate and
form very open card-house structures (see e.g. Pusch
1970). Since their formation the clays have been con-
solidating and creeping under their self-weight. Due
to leaching, the salt ions at the surface of the clay
particles have dissolved. The possible leaching proc-
esses involve percolating water through the deposit,
water seeping upwards through the deposit due to
artesian pressure, and diffusion of salts towards
zones with lower ion concentrations (Rankka et al.
2004). This way the initially stable structure of the
clays has become meta-stable and sensitive.
When piles or any column-type of ground improve-
ment are installed in the ground, there are large defor-
mations in the soils, involving both shearing and the
lateral expansion of the soil. Consequently, there are
changes in the vertical and horizontal effective stresses,
and the parameters associated with the state of the soil,
such as void ratio, preconsolidation pressure, density,
K0 (coefficient of earth pressure at rest) etc. are modi-
fied. As demonstrated by Dijkstra et al. (2010, 2011)
using photoelastic techniques, installation causes
significant density changes and rotations of princi-
pal stresses. The former are of course beneficial, but
effects of the latter on soil state are usually ignored.
In granular media, pile driving may also cause particle
crushing Lobo-Guerrero & Vallejo (2005) and break-
age of the apparent bonding in the soil, which in turn
may have a detrimental effect on the pile capacity.
In this paper, the focus is on structured natural
clays, and modelling the changes in their stress-
strain response due to loading and possible instal-
lation effects.
2 CONSTITUTIVE MODELLING
OF STRUCTURED CLAYS
The response of clays is dominated by their micro-
structure, which has been studied with the help of
44
reconstituted clays samples given the difficulties
in micromechanical testing of intact natural clays.
During irrecoverable straining for kaolin clay there
is evidence on re-orientation of particles, and
changes in particle contacts, at microstructural
level (see e.g. Bai & Smart 1996, Hicher et al. 2000).
This supports the macro-level evidence that the
anisotropy of clays, when undergoing large strains,
can evolve.
Based on experimental studies on natural
Otaniemi clay, Wheeler et al. (2003) proposed an
elastoplastic model, called S-CLAY1, with a rota-
tional hardening law describing the changes in the
inclination of the yield surface due to irrecover-
able straining. The calibration of the parameters
for the S-CLAY1 model is rather straightforward
and the model has been thoroughly validated
experimentally by Karstunen and her co-workers
(Karstunen & Koskinen, 2004, 2008).
As shown by i.e., Burland (1990) & Leroueil &
Vaughan (1990), when natural clays are undergoing
large deformations, the initial apparent bonding is
progressively lost, and at large strains, soil starts to
behave as a fully destructured material. The degra-
dation of bonding due to irrecoverable straining is
referred to as destructuration (Leroueil et al. 1979).
Most constitutive models that attempt to account
for the effect of bonding and destructuration, uti-
lize the conceptual ideas by Gens and Nova (1993).
The same applies to the S-CLAY1S model (Kars-
tunen et al. 2005), in which the sudden collapse of
the sensitive clay structure upon yielding is rep-
resented using the concept of intrinsic yield sur-
face (Gens and Nova 1993) in combination with
the rotational hardening of the S-CLAY1 model
(Wheeler et al. 2003). This was the first constitutive
clay model that was able to represent the changes
in the state of the structured clays in a simple com-
prehensive manner. The next stage was to incorpo-
rate rate-effects into the model.
Given it was thought that anisotropy and its evo-
lution has a major effect on the soft clay response,
the S-CLAY1 model was extended to account for
creep and rate effects by Leoni et al. (2008) using the
creep formulation by Vermeer and his co-workers
(Vermeer et al. 1998, Vermeer & Neher 1999). The
resulting model, called ACM (Anisotropic Creep
model), has the advantage that the model param-
eters are easy to derive. Most importantly, the
concept of reference time or time shift (see Leoni
et al. 2008) enables, for the first time a systematic
way to take into account the strain-rate in the tests
that are used for defining the model input param-
eters. This is a very powerful feature of the model,
which gives it a major advantage over the so-called
overstress models, based on Perzyna’s (1963) over-
stress theory, proposed by e.g. Hinchberger & Gu
(2009) and Karstunen & Yin (2010).
 As discussed the in following, and pointed out
by Grimstad et al. (2010) and Karstunen & Yin
(2010), some of the key assumptions in the ACM
model are, however, both inconsistent with experi-
mental evidence and fundamentally wrong (see
also the companion paper by Sivasithamparam
et al. 2013). As a consequence, in practical context,
the ACM model often ends up grossly overpre-
dicting deformations in structured soft soils (see
Karstunen et al. in press), when the values of soil
constants are objectively selected. This affects the
predictive ability of the model and prevents its use
in practical context on structured soil deposits.
The work by Hinchberger & Gu (2009), Kars-
tunen & Yin (2010) and Yin et al. (2011), demon-
strates that for predicting certain features of natural
clay behavior, such as tertiary creep and creep rup-
ture, or indeed the phenomenon of progressive
failure, it is also necessary to account for the effects
of bonding and destructuration. In the following
some features of the ACM model are highlighted,
which necessitated the development of a new rate- Figure 1. Current state surface (CSS) and normal con-
dependent model Creep-SCLAY1 (for details see solidation surfaces (NCC) of the ACM model and the
the companion paper Sivasithamparam et al. direction of viscoplastic strains (triaxial stress space).
2013). The importance of modelling key features
of soil behaviour in the context of rate-dependency
are discussed and demonstrated through a series of In the ACM model, there is a separate failure sur-
numerical benchmark simulations. face on which zero dilatancy condition is imposed
in order to comply with the condition of zero
volumetric strain rate at critical state. Hence, the
3 ACM AND CREEP-SCLAY1 MODELS there is no “dry side” of the critical state. Because
an associated flow is assumed on the CSS, there
The creep formulation in the ACM model (Leoni is actually a “jump” in the volumetric creep rate
et al. 2008) is based on the idea of a Normal Con- when approaching critical state (see Figure 1): the
solidation Surface (NCS), see Figure 1, which is constant volumetric creep rate at the CSS surface
treated as the contour of constant volumetric creep suddenly reduces to zero (results of simulations
strain rate. For simplicity, the model is plotted in showing this can be found in Sivasithamparam
Figure 1 in triaxial stress space, in terms of mean et al. 2013).
effective stress p’ and deviator stress q. The scalar Most importantly, because the volumetric creep
α represents the current degree of anisotropy, M strain rate is assumed to be constant, large volu-
is the stress ratio at critical state and ψ is the dila- metric creep strains and very large deviatoric creep
tancy angle. strains are predicted in the stress space between
The volumetric creep strain rate is given by a the K0 line and critical state line. As shown in the
simple power law as follows: simulations in this paper, this results in severe
overprediction of deformations at boundary value
problems. Because of these assumptions, quite
β
μ ⎛ peq
′ ⎞ λ κ* significant apparent strain-softening is predicted
εvc = ⎜ ⎟ with β = (1) in undrained simulations of shearing of normally
τ ⎝ p′p ⎠ μ* consolidated or lightly overconsolidated samples.
Furthermore, as shown by Sivasithamparam et al.
where μ* is the modified creep index, λ* in the (2013), the ACM model cannot model the isotach
modified compression index, κ * is the modified behaviour that is observed in soft clays caused by a
swelling index and τ is the reference time (see Leoni step-change in strain rate. This is not satisfactory,
et al. 2008 for details). The ratio in the brackets is and these insufficiencies inspired further model
an inverse of a generalised yield stress ratio. Hence, developments.
creep is occurring even within the overconsolidated In the Creep-SCLAY1 model (Figure 2), instead
region, when the current stress surface (CSS) in of assuming constant volumetric creep strain rate,
Figure 1, is smaller than the NCS. similarly to Grimstad et al. (2010) and Yin et al.

45
Figure 2. Current state surface (CSS) and normal con-
solidation surfaces (NCC) of the Creep-SCLAY1 model
and the direction of viscoplastic strains (triaxial stress
space).
(2011), it is assumed that the viscoplastic multi-
plier is constant along the CSS. The viscoplastic
multiplier is defined in such a way that the model
gives identical volumetric creep strain rate to Eq.
(1) under oedometric loading (see Grimstad et al.
(2010) for details) as:
β
μ ⎛ peq
′ ⎞ ⎛ M 2 − α 2 NC ⎞
K0
εvc = ⎜ ⎟ ⎜ 2 ⎟ (2)
τ ⎝ p′p ⎠ ⎜⎝ M − ηK2 NC ⎟⎠
0
Where η = q / p′ is the stress ratio, and subscript
K0NC refers to the normally consolidated K0 stress
path. The corresponding strain rate vectors are
plotted in Figure 2. Due to the evolution of ani-
sotropy, the Creep-SCLAY1 model predicts some
minor apparent strain softening in undrained
shearing in normally consolidated region. Because
there is now a “dry side”, with increasing strain
rate it is also possible for the stress path to pass
over the critical state line, which is in agreement
with experimental evidence (see simulations in
Sivasithamparam et al. 2013).
For finite element analyses, the model needs to
be generalised. Instead of a scalar α, the anisot-
ropy is described with a fabric tensor, and invari-
ants can no longer be used (see Wheeler et al. 2003
for details). In order to account for Lode angle
dependency, instead of the Drucker-Prager model
that assumes constant M, the formulation of Sheng
et al. (2000) has been adapted, in order to have a
smooth variation of the Lode angle of the fabric in
46
Figure 3. The Creep-SCLAY1 model (general stress
space).
the general stress space as the CSS and NCS rotate
(see Figure 3). In order to incorporate the effect
of bonding and destructuration in some of the
simulations, an intrinsic surface has been added
together with a destructuration law that is analo-
gous with the S-CLAY1S model (Karstunen et al.
2005). In the following, this version of the model
is referred to as the Creep-SCLAY1S model. It is
basically a rate-dependent model that accounts for
the changes in fabric arrangement and bonding,
enabling a rather complete representation of the
stress-strain behaviour of structured clays.
The usability of a constitutive model in a practi-
cal context relies on its robust implementation into
the finite elements. The Creep-SCLAY1 model
is implemented into the PLAXIS (Brinkgreve
et al. 2010) finite element code as a user-defined
soil model (UDSM) using a Modified Newton-
Raphson scheme (Sivasithamparam 2012), whilst
the ACM model is using a fully implicit Newton
iteration (Leoni et al. 2008). In the following the
model is applied to a typical benchmark problem,
a simple embankment.
4 BENCHMARK SIMULATIONS
4.1 FE model and model input
The performance the ACM model and the Creep-
SCLAY1 model are demonstrated in a benchmark
problem using the 2D PLAXIS finite element
code. An embankment constructed on soft soil is
assumed to be 2 m high, with a width at the top
of 10 m and the side slopes with a gradient of 1:2.
The soft soil is assumed to have the properties of
soft Bothkennar clay (Symposium 1992), which
extends to a depth of 30 m. The groundwater table
assumed to be located at 1 m below the ground sur-
face. At the surface there is a 1 m depth over-con-
solidated dry crust. The problem has been analysed
as a small strain simulation. The values for the
model parameters for the soft clay layer are shown
in Table 1. The embankment, assumed to be made
of granular material, was modelled with a simple
Mohr Coulomb model and the same applies to
the dry crust layer (see Table 2 for material param-
eters). This problem is expected to be dominated
by the soft soil response and is not sensitive to the
embankment and crust parameters.
The most difficult model parameter to deter-
mine is constant ω representing the evolution of
anisotropy. Hence, for comparison the simulation
is also run with a version of Creep-SCLAY1 model
where the anisotropy has been fixed, i.e. only initial
anisotropy is considered, with no evolution of ani-
sotropy. This assumption was made e.g. by Bodas
Freitas et al. (2011) in their rate-dependent model.
Such a model would of course not be able to repro-
duce the stress-strain response in cases where there
are major changes in the stress path directions, and
in principle would not be able to reproduce the ele-
ment level tests done by Karstunen & Koskinen
(2008).
4.2 Results
The predicted vertical displacements by ACM
and Creep-SCLAY 1 as function of time are plot-
ted in Figure 4 and the corresponding settlement
through at the end of construction and at the end
of consolidation have been plotted in Figure 5.
The settlement predicted by both models are huge
(unrealistically large considering that the embank-
ment is only 2 m high. The results clearly demon-
strate that the problem is not a one-dimensional
problem, as for K0 consolidation the predictions by
the two models would have been the same.
The introduction of the constant viscoplastic
multiplier, is clearly improving the results, and this
is even more apparent in Figure 6, where the lateral

Table 1. Model parameters for the soft clay.

Parameter e0 OCR K0 α0 χ0

Soft clay 2.0 1.50 0.50 0.59 8

γ
Parameter [kN/m3] κ* ν′ λ* M

Soft clay 16.5 6.67E-3 0.20 0.10 1.5

Parameter λ*i ωd ω ζ ζd
Figure 4. Predicted vertical displacements as a function
Soft clay 0.06 1.0 50 9 0.2 of time by ACM and Creep-SCLAY1.

τ
Layer μ*i μ* [days]

Soft clay 2.0E-3 5.07E-3 1

Table 2. Model parameters for the embankment and

the dry crust.

γ Ε ϕ′ c′ ψ
Layer [kN/m3] [kPa] [°] [kPa] ν′ [°]

Embankment 20 40000 40 2 0.35 0

Dry crust 19 3000 30 6 0.20 0 Figure 5. Predicted vertical settlement through by
ACM and Creep-SCLAY1.

47
 Figure 7. Predicted vertical displacements as a func-
tion of time by Creep-SCLAY1, Creep-SCLAY1 with no
Figure 6. Predicted lateral displacements by ACM and rotation and Creep-SCLAY1S.
Creep-SCLAY1 at the end of consolidation.

deformations under the toe of the embankment

predicted by the two models have been plotted.
The assumption in ACM model about CSS being
the contour of constant volumetric creep rate and
combining that with associated flow rule simply
results in much higher shear strain prediction, and
hence also larger horizontal deformations than
predicted by Creep-SCLAY1. Both models predict
notable deformations outside the loaded area (see
Figure 5), which of course is not satisfactory.
In order to investigate what the effect of the
evolution of anisotropy is, the benchmark problem Figure 8. Predicted vertical settlement through by
has been re-analysed with Creep-SCLAY1 model Creep-SCLAY1, Creep-SCLAY1 with no rotation and
with the evolution of anisotropy switched off Creep-SCLAY1S.
(ω = 0). As a consequence, no rotation of the CSS
and NCS is allowed. The predictions of the vertical
and horizontal deformations by this version of the
model are high (Figures 7, 8 and 9), which demon-
strates that the evolution of anisotropy dissipates
plastic energy, so any predictions that ignore that
are essentially over-conservative. Again, as would
be expected, far too large deformations are trig-
gered by the in situ stresses only.
When considering installation effects, account-
ing for changes in anisotropy is extremely impor-
tant. Castro & Karstunen (2010) and Castro et al.
(2013) simulate stone column installation in Both-
kennar clay by modelling the problem as a cavity
expansion with S-CLAY1S model. Their results
show that due to the installation for typical centre
to centre spacings the soil fabric flips totally over.
When the actual structure is then constructed, Figure 9. Predicted lateral displacements by Creep-
the fabric needs to rearrange once again. Prelimi- SCLAY1, Creep-SCLAY1 with no rotation and Creep-
nary analyses suggest that the resulting deforma- SCLAY1S at the end of consolidation.
tions might actually decrease up to 25%, all due
to this dissipation of energy due to evolution of turation is that both the compression index and
anisotropy. the modified creep index are assumed to be con-
One problem with creep models that do not stant. The experimental data on structured clays
account for the effect of bonding and destruc- however clearly demonstrates that neither of

48
these are constants for structured natural clays.
Due to the gradual degradation of bonds, both
the apparent compression index and the appar-
ent creep index have maximum values just after
the onset of yield (see e.g. data by Graham et al.
1983). Indeed, even their ratio (and consequently
exponent β in Eq. 1) is not constant. As shown by
Karstunen & Yin (2010) this type of behaviour
can be represented with a rate-dependent model
that accounts for the effect of bonding and its
gradual destruction.
In order to explore the effect of bonding and
destructuration at boundary value level, the bench-
mark was simulated with Creep-SCLAY1S model,
i.e. an extension of the Creep-SCLAY1 model that
accounts additionally for initial bonding and the
subsequent degradation of bonding. The results
have been included in Figures 7, 8 and 9, labelled
Creep-SCLAY1S. The predicted vertical and hori-
zontal deformations have reduced by nearly 45%
and 55% respectively, and most importantly the
unrealistic large deformations due to the in situ
stresses only have virtually vanished.
CONCLUSIONS
Installation of piles and ground improvement into
the ground will modify the state of the soil. This
sometimes beneficial, and sometimes detrimental,
and so far this effect has been rarely taken into
account. One reason for this is that the numerical
techniques and the constitutive soil models have
not been able to represent the changes in soil struc-
ture and state in a satisfactory manner. The paper
dis cussed some recently developed rate-dependent
constitutive models for structured clays, culminat-
ing in a new rate-dependent model. The numerical
simulations show that the Creep-SCLAY1 model
(Sivasithamparam et al. 2013), when extended to
account for the effects of apparent bonding and
destructuration, appears to have features that
enable realistic predictions of rate-dependent
stress-strain behaviour of soft structured clays.
Furthermore, the results and the experimental
evidence so far suggest that when modelling rate-
dependent behaviour of structured clay, it is essen-
tial to account for both evolution of anisotropy
and the effect of bonding and destructuration, in
order to have a realistic rate-dependent constitu-
tive model.
Creep and rate effects can be considered as
microstructural reaction to stress increment (Pusch
2012). Although the micromechanics of structured
clays is complex, and our sampling and experimen-
tal techniques have not yet developed to explore
this fully, the simulations in this paper suggest
that it is possible to account for these microstruc-
49
tural effects by advanced macroscopic constitutive
models. These models have also the advantage that
they enable us to improve our understanding of
what might be happening, when piles and ground
improvement elements are installed into soft struc-
tured clays.
ACKNOWLEDGEMENTS
The work presented was carried out as part of a
Marie Curie Industry-Academia Partnerships and
Pathways project GEO-INSTALL (PIAP-GA-
2009-230638). The first author is also benefitting
from the financial support by Marie Curie Indus-
try-Academia Partnerships and Pathways project
CREEP (PIAP-GA-2011-286397). The experimen-
tal work facilitating the model developments was
done at Aalto University sponsored by Academy
of Finland (Grant 128459).
REFERENCES
Bai, X. & Smart P. 1996. Change in microstructure of
kaolin in consolidation and undrained shear. Géo-
technique 47(5): 1009–1017.
Bodas Freitas, T.M., Potts, D.M. & Zdravkovic L.
2011. A time-dependent constitutive model for soils
with isotach viscosity. Computers and Geotechnics
38(6):809–820.
Brinkgreve, R.B.J., Swolfs, W.M. & Engin, E. 2010. Plaxis
2D 2010 Manual.
Burland, J.B. 1990. On the compressibility and shear
strength of natural clays. Géotechnique 40 (3):
329–378.
Castro, J. & Karstunen, M. 2010. Numerical simulations
of stone column installation. Canadian Geotechnical
Journal 47:1127–1138.
Castro, J., Karstunen, M., Sivasithamparam, N. & Sagas-
eta C. 2013. Numerical analyses of stone column
installation in Bothkennar clay. Proc. International
Conference on Installation Effects in Geotechnical
Engineering, Rotterdam, NL, 23–27 March 2013.
Dijkstra, J., Alderliests, E.A. & Broere, W. 2011. Pho-
toelastic investigation into plugging of open ended
piles. Frontiers in Offshore Geotechnics II: 483–488.
CRC Press. DOI: 10.1201/b10132–61
Dijkstra, J. & Broere, W. 2010. New full-field stress meas-
urement method using photoelasticity. Geotechnical
Testing Journal 33(3). DOI: 10.1520/GTJ102672
Gens, A. & Nova, R. 1993. Conceptual bases for a
constitutive model for bonded soils and weak rocks.
In Geotechnical Engineering of Hard Soils—Soft
Rocks, Athens, Greece, Anagnostopoulos et al. (eds.).
Balkema, Rotterdam: 485–494.
Graham, J., Crooks., J.H.A. & Bell A.L. 1983. Time
effects on the stress strain behaviour of natural soft
clays. Géotechnique 33(3): 327–340.
Grimstad, G., Abate, S., Nordal, S. & Karstunen, M.
2010. Modelling creep and rate effects in structured
anisotropic soft clays. Acta Geotechnica 5: 69–81.
Hicher, P.Y., Wahyudi, H. & Tessier, D. 2000. Microstruc-
tural analysis of inherent and induced anisotropy in
clay. Mechanics of Cohesive-Frictional Materials 5(5):
341–371.
Hinchberger, S.D. & Qu, G. 2009. Viscoplastic constitu-
tive approach for rate-sensitive structured clays. Cana-
dian Geotech. J. 46(6): 609–626.
Karstunen, M. & Koskinen, M. 2004. Anisotropy and
destructuration of Murro clay.” Proc., Advances
in Geotechn. Eng. Skempton Conf., London, UK,
1: 476–487.
Karstunen, M. & Koskinen M. 2008. Plastic anisotropy
of soft reconstituted clays. Canadian Geotechnical
Journal 45: 314–328.
Karstunen, M., Krenn, H., Wheeler, S.J. Koskinen, M. &
Zentar, R. 2005. Effect of anisotropy and destructu-
ration on the behaviour of Murro test embankment.
ASCE International Journal of Geomechanics 5(2):
87–97.
Karstunen, M., Rezania, M, Sivasithamparam, S. &
Yin, Z.-Y. (in press). Comparison of anisotropic
rate-dependent model for modelling consolidation
of soft clays. International Journal of Geomechanics
doi:10.1061/(ASCE)GM.1943–5622.0000267.
Karstunen, M. & Yin, Z.-Y. 2010. Modelling time-de-
pendent behaviour of Murro test embankment. Géo-
technique 60(10): 735–749.
Leoni, M., Karstunen, M. & Vermeer, P.A. 2008. Aniso-
tropic creep model for soft soils. Géotechnique, 58(3):
215–226.
Leroueil S., Tavenas F., Brucy F., La Rochelle P. &
Roy M. 1979. Behaviour of destructured natural clays.
ASCE Journal of Geotechnical Engineering, 105(6):
759–778.
Leroueil. S. & Vaughan, P.R. 1990. The general and con-
gruent effects of structure in natural soils and weak
rocks. Géotechnique, 40(3): 467–488.
Lobo-Guerrero, S. & Vallejo, L.E. 2005. DEM analysis
of crushing around driven piles in granular materials.
Géotechnique 55(8): 617–623.
Perzyna, P. 1963. The constitutive equations for work-
hardening and rate sensitive plastic materials. Proc.
Vibration Problems Warsaw 3, 281–290.
50
Pusch, R. 1970. Clay microstructure. A study of the
microstructure of soft clays with special reference to
their physical properties. Swedish Geotechnical Insti-
tute, Proceedings No 24. Stockholm.
Pusch, R. Personal communication, 2012.
Rankka, K. 2003. Kviklera—billdning och egenskaper. SGI
Varia 526. Swedish Geotechnical Institute, Linköping.
Rankka, K., Andersson-Sköld, Y. Hulten C., Larsson, R.
Lerocex, V. & Dahlin, T. 2004. Quick clays in Sweden SGI
Report 65. Swedish Geotechnical Institute, Linköping.
SGI. 1995. Geotekniska skadekostnader och behov av
ökad geoteknisk kunskap. Internal Report K94/1825/3.
In Swedish, not publicly available.
Sheng, D., Sloan, S. & Yu, H. 2000. Aspects of finite ele-
ment implementation of critical state models. Compu-
tational Mechanics 26: 185–196.
Sivasithamparam, N. 2012. Modelling creep behaviour
of soft soils. Internal report. Plaxis B.V. & University
of Strathclyde, not publicly available.
Sivasithamparam, N., Karstunen, M., Brinkgreve, R.B.J., &
Bonnier P.G. 2013. Comparison of two anisotropic
rate dependent models at element level. Proc. Interna-
tional Conference on Installation Effects in Geotechni-
cal Engineering, Rotterdam, NL, 23–27 March 2013.
Symposium. 1992. Bothkennar soft clay test site: Charac-
terization and lessons learned (Géotechnique sympo-
sium in print). Géotechnique 42(2): 161–380.
Vermeer P.A. & Neher H.P. 1999. A soft soil model that
accounts for creep. Proc. Int. Symp. Beyond 2000 in
Computational Geotechnics. Amsterdam. Balkema,
Rotterdam, 249–261.
Vermeer P.A., Stolle D.F.E. & Bonnier P.G. 1998. From
the classical theory of secondary compression to
modern creep analysis. Proc. Computer Methods and
Advances in Geomechanics. Balkema, Rotterdam.
Wheeler, S.J., Näätänen A., Karstunen M. & Lojander M.
2003. An anisotropic elasto-plastic model for natural
soft clays. Canadian Geotechnical Journal 40(2):
403–418.
Yin, Z.-Y., Karstunen, M., Chang, C.S., Koskinen, M. &
Lojander, M. 2011. Modelling time-dependent behav-
iour of soft sensitive clay. J. Geotech. Geoenviron.
Eng., 137(11): 1103–1113.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

FEM simulation of large vertical deformations caused by land

subsidence and verification of the results by using radar
interferometry techniques

C. Loupasakis & D. Rozos

Laboratory of Engineering Geology and Hydrogeology, School of Mining and Metallurgical Engineering,
National Technical University of Athens, Athens, Greece

F. Raspini & S. Moretti

Department of Earth Sciences, University of Firenze, Firenze, Italy

ABSTRACT: Land subsidence induced by the over-exploitation of aquifers is a very common phenom-
enon affecting extensive areas worldwide. Knowledge about the range and the rate of the deformations is
necessary for the installation and the protection of constructions. A detailed study was conducted aim-
ing to validate the efficiency of the Mohr–Coulomb and the Hardening soil models, introduced in the
PLAXIS 2D finite-element code, for the simulation of land subsidence. The data used for this case study
came from the Kalochori region on the west side of Thessaloniki, Northern Greece. The rapid develop-
ment of the area and the extensive need of water led to the development of surface subsidence, reaching
maximum values of 3–4 m, in several parts of the study area. The validation of the simulation results was
conducted by using all available subsidence indications as well as a land motion mapping produced by PSI
(Persistent Scatterer Interferometry) analysis.

1 INTRODUCTION

The simulation of land subsidence phenomena

induced by intense water pumping encounters sev-
eral difficulties. The fact that these phenomena have
taken place for several decades, sometimes without
being noticed at the beginning, makes the estima-
tion of the total vertical displacements, and as a
result the conduction of back analysis, impossible.
Furthermore, the great extent of the affected areas,
the interaction of the engineering, geological and
hydro-geological settings, along with the impact of
human activities, makes their study complicated
and requiring numerous data.
Several regions in Greece have experienced land Figure 1. Location map of the study area. Background
subsidence induced by intense water pumping optical images from Google Earth and Visual Earth.
during recent decades. Kalochori village (Fig. 1)
on the west side of Thessaloniki is one of the
most well known and widely investigated areas finite element simulation model. Furthermore, the
(Hatzinakos et al., 1990; Andronopoulos et al., availability of PSI (Persistent Scatterer Interfer-
1991; Rozos & Hatzinakos, 1993; Stiros, 2001; ometry) data allowed cross-checking of the results
Stiros et al., 2001; Rozos et al, 2004; Psimoulis et al., and validation of the simulation procedure. Practi-
2007; Loupasakis & Rozos, 2009; Doukas et al., cally, this case study was considered as the perfect
2004; Psimoulis et al., 2007; Raucoules et al., 2008; chance of applying the FE method for the analysis
Mouratidis et al., 2010). of water pumping induced subsidence phenomena
The numerous studies conducted for the and verifies the results by the use of space meas-
Kalochori region provided plenty of data, covering urement of past displacements derived from PSI
all the requirements for setting up a polyparametric analyses.

51
2 THE STUDY AREA

The Thessaloniki plain, located in the northern

part of the Thermaikos Gulf, is the largest del-
taic area of Greece with an area of more than
1500 km2. From a geological point of view the
plain corresponds to a complex delta formed in the
last thousand years in a 600 m deep Quaternary
graben.
At the bordering mountains Mesozoic lime-
stones, marbles and other metamorphic rocks
outcrop, while Neogene and Pleistocene sediments
occur in the hilly areas. Within the plain the rocky
basement is covered by Neogene and Quaternary
deposits. The Neogene deposits outcrop in the
north and north-eastern parts of the study area.
Figure 2. Equal subsidence contours, in meters, for
Quaternary formations cover the Neogene depos- the period 1955–1980 based on data coming from the
its throughout the rest of the area acquiring a sig- comparison of 1:5000 topographic diagrams, (Stiros,
nificant thickness towards the coast. 2001). The outlines present the land use distribution in
The evaluation of the data coming from geo- Kalochori: A: the residential area of Kalochori village,
technical boreholes drilled in the Kalochori region B: the lagoon—partly flooded area, C: the residential
showed that the quaternary deposits consist of clays, waste dumping site, D: areas containing the productive
silty clays, sands and silty sands including organic wells of the Water Company of Thessaloniki, E: the
material. (Andronopoulos, 1979; Andronopoulos industrial area of Kalochori until the 1980’s. Finally, the
location of the embankments is also presented.
et al., 1990). Particularly, the top horizons, from
the surface to a depth of 4.5 to 15 m, are occu-
pied by yellow-brown fine to medium grain sands
and silty sands. Under the sands and down to a
varying depth of 26.5 to 35.3 m extends a black-
gray silty sands horizon including organic materi-
als (Loupasakis & Rozos, 2009). The geophysical
investigation results and data coming from water
well profiles showed that the above mentioned
alteration routine continues down to the Neocene
basement (Nikolaou & Nikolaidis, 1987).
In agreement with the above described geologi-
cal setting, the formations of the wide study area
contain an unconfined shallow aquifer extending
in the overlaying sands and a system of successive
confined artesian aquifers developing below the
impermeable black-gray silty sand layer.
Figure 3. Emerged well pipes due to the subsidence
phenomenon. Deep productive well located NW of
3 LAND SUBSIDENCE EVOLUTION Kalochori. The pipes have been extracted more than 1 m
above the ground surface.
3.1 Field observations—historical background
The land subsidence phenomena were first noticed Since then several damages and extensions of the
in 1965, in the form of a progressive marine embankment were recorded, but the main con-
invasion. In 1969, during a period of intensive struction managed to resist.
rainfall, the seawater reached the southern houses Through the years several signs referring to the
of the village. Since then several attempts to total amount of the subsidence were reported. For
reclaim the lost land were made by constructing instance, the main square of the village was recon-
embankments. The continuous subsidence phe- structed about 1.2 m higher to prevent it from
nomena combined with the structural defects of flooding, in many wells the pipes were raised above
the embankments led repetitively to their collapse. the ground surface by about 1 m (Fig. 3) and sev-
Finally, a new larger dam was constructed in eral areas around the village are still submerged.
1980 providing security to Kalochori (Fig. 2). Note that in contrast to many other areas in Greece

52
affected by subsidence phenomena, no surface
ruptures were located throughout the entire study
area. This is clearly related to the absence of tec-
tonic lines as well as the big thickness and the rela-
tive uniformity of the Quaternary deposits.
The total amount of subsidence deformations
was estimated to exceed 3 m in several locations.
This aspect was supported by Stiros (2001), who
compared the 1:5000 topographic diagrams for the
period 1955–1980. This comparison proved that,
until the 1980’s, the maximum subsidence regions
(subsidence > 3 m) were located within the areas
containing the productive wells of the Water Com-
pany of Thessaloniki (Fig. 2) as well as the south-
western section of the industrial area containing
the most water-consuming industries. Currently,
along the embankments the inland is at least 4 m
below the sea level.
3.2 Interferometric techniques
3.2.1 Basic principles
Assessment of ground motion is historically based
on conventional geodetic methods (GPS, level-
ling network above everything else), or on timely
updated surveying. These methods represent some
of the most reliable geodetic measurements tools,
but are time consuming and resource intensive. In
addition to conventional geodetic monitoring sys-
tems, Earth Observation (EO) techniques have suc-
cessfully demonstrated, in recent years, to be highly
valuable in measuring land motion in a wide range
of application fields (Tralli et al., 2005).
The study of land subsidence in the wider
Thessaloniki Plain exploited the PSI technique,
a remote sensing technology based on multi-
temporal satellite SAR (Synthetic Aperture
Radar) imagery.
A single SAR image is of no practical use, since it
is not possible to distinguish the different phase con-
tributions related to atmosphere, topography and
noise. Differential Interferometry SAR (DInSAR),
a technique that exploits the phase variations (or
interference) of two SAR images gathered at differ-
ent times on the same target area by the same satel-
lite using the same acquisition mode and properties
(beam, orbit, off-nadir angle, etc), is a far more suit-
able approach (Zebker et al., 1986; Massonnet &
Feigl, 1998, Rosen et al., 2000). Since the 90s sig-
nificant results have been obtained in detecting,
mapping and quantifying surface deformation with
millimeter precision over large, contiguous areas
along the sensor’s Line-Of-Sight (LOS) direction.
Conventional DInSAR is limited by temporal and
geometrical decorrelation (Zebker and Villasenor,
1992). Moreover, phase distortions introduced by
atmospheric effects, can compromise the accuracy
of the results (Massonnet & Feigl, 1995).
53
The main limitations encountered in single-pair
interferograms can be overcome through PSI (Per-
sistent Scatterer Interferometry) techniques, a family
of different approaches based on processing several
multi-temporal space-borne SAR imagery (at least
15–20, or more) of the same target area. The PSI
approach uses long stacks of radar dataset (as those
in the ESA ERS1/2 archive) and analyzes the signals
backscattered from the observed scene to estimate
and remove atmospheric artifacts. The electromag-
netic returns of electromagnetically stable, highly
reflective point-wise targets (the so-called Perma-
nent Scatterers, (PS) is statistically processed and
analyzed to retrieve estimates of the displacements
occurred between different acquisitions (Ferretti et
al., 2000; 2001). PS usually correspond to man-made
structures (i.e. buildings, roads, bridge, monuments,
pylons), as well as natural reflectors, such as out-
cropping rocks. Thanks to this multi-interferometric
approach, at each Permanent Scatterer, the relevant
LOS velocity can be estimated with unprecedented
accuracy, sometimes even better than 0.1 mm/year.
Displacement time series can be retrieved, acquisi-
tion per acquisition, with accuracy on single meas-
urements usually ranging from 1 to 3 mm (Colesanti
et al., 2003). Each measurement is temporally and
spatially referred to a unique reference image and to
a stable reference point, respectively.
Since the pioneering studies of Massonnet et al.
(1993; 1994; 1995; 1996) and Singhroy et al. (1998),
satellite SAR data have been fully exploited to
retrieve ground deformations related to geohazards.
3.3 SAR data used
Within the framework of the ESA GMES Ter-
rafirma project (GMES—Global Monitoring for
Environment and Security), 9 satellite image frames
were processed using a special semi-automated proc-
essor to produce a PSI ground motion map covering
a 65,000 km2 area of Greece. This WAP (Wide Area
Product) over Greece is based on stripmap ERS1/2
images, obtained from the European Space Agency
(ESA). Among these 9 satellite image frames, track
7_3, made of a dataset of 63 SAR imagery in C
band (5.6 cm wavelength), acquired along descend-
ing orbits in 02/05/1992–30/12/2003, was employed.
The reference points of the stack—to which the
displacement estimates are referred—was selected
in the southernmost part of Thessaloniki’s urban
fabric, in a sector unaffected by ground motions.
The master image for Track 7_3 dataset has been
chosen on December 31st, 1996.
3.4 Evaluation of the PSI data
The analysis of the mean annual displacement
velocities, retrieved by PSI processing, provide
essential information. As expected, the density of
measurement points is higher for the urban area
of Thessaloniki (maximum density up to about
300 PS/km2), and over the urban fabric of the
Sindos and Kalochori villages (up to hundred of
PS/km2), but lower for agricultural and vegetated
terrains (few point/km2). Nevertheless, more than
1.000 point-wise measurements targets have been
yielded for the 15 km2 wide-Kalochori village and
its industrial area. An overview of the WAP results
for ERS1/2 datasets is presented in Figure 4,
as measured by the SAR sensor along the satellite’s
line of sight and expressed in mm/y. The black dots
refer to stable areas. The triangles pointing south
represent increasing deformation rates, moving
away from the sensor (subsidence).
The PSI results reveal that the city of Thessalo-
niki shows very low LOS deformation rates, ranging
between −1.5 and 1.5 mm/yr (close to the PSI tech-
nique detection sensitivity), indicating relatively sta-
ble ground conditions since the 90s. Considering that
Figure 4. The LOS deformation rates in Kalochori—Sindos between 1995 and 2001, obtained through the PSI
analysis. Time series for selected PS A, B, C, D and E are shown in Figure 5.
54
a large part of the urban fabric is built over Neogene
formations the above observation is reasonable. On
the contrary the Sindos-Kalochori area presents sig-
nificant deformations. Particularly, the PSI results
show that subsidence was widespread throughout
the western suburbs of the Thessaloniki metropoli-
tan area in the time period from 1995 to 2001. Land
subsidence can be clearly identified in the coastal
area of Kalochori, where the highest LOS velocities
range between 15 and 25 mm/yr in 1995–2001, with
peaks of 40 mm/yr recorded in the SW of the vil-
lage. The observed subsidence rate in Kalochori is
extended to another subsidence bowl affecting the
village of Sindos, on the right side of the Gallikos
River. In this area the LOS deformation rate ranges
between 5 and 15 mm/yr, with peaks of 46 mm/yr.
These two well defined areas of land deformation
are abruptly truncated along a NW-SE alignment:
northeast of Kalochori and Sindos.
The potential of repeat-pass space-borne SAR
interferometry can be exploited not only to map

Figure 5. LOS displacement time series and yearly
velocities of PS A, B, C, D and E, located in Kalochori
village and indicated in Figure 4. Time series are rescaled
to the first available acquisition (i.e. 10 April 1995).
the extension of affected areas but also to evaluate
their deformation history. The displacement time
series, available for each PS in the area of inter-
est, are ideally suited for large-scale, spatially con-
tinuous monitoring of geohazard-related ground
motion (Fig. 5).
The time series of Figure 5, referring to five Per-
manent Scatterers located in areas with different
magnitudes of deformation, reflecting the differ-
ent behavior of the various sectors of the ground,
allow the evaluation of the temporal evolution
of each measurement point. The time series are
rescaled to the first acquisition date (i.e. 10 April
1995) for visualization purposes only.
4 MODELLING ACTIVITY
This simulation of the land subsidence mechanism
was conducted by using the PLAXIS (V 9.02) two-
dimensional finite element code (Brinkgreve et al.,
2002). The application of a two-dimensional code
could be considered sufficient because, considering
the geological setting of the site, the present study
deals with the simulation of a simple cross section
with no horizontal transitions.
The simulation procedure was applied in two char-
acteristic profiles of 100 and 150 m. The thickness
of the profiles was selected by considering the thick-
ness of the loose Quaternary deposits (200–100 m),
the mean depth of the productive drills (100 m) and
the water level variations (25 to 34 m). The altera-
tion and the thickness of the geological formations
are clearly presented in Figure 6 and it is identical in
both profiles. The only difference is that in the 100 m
deep profile the bottom layers are missing.
The simulations were conducted by using
geotechnical data coming from the reevalua-
tion of all available former studies (Table 1).
55
Figure 6. The simulation profile. The soil layers are
indicated in respect to their reference at Table 1.
Both Mohr–Coulomb and Hardening soil models
were used in order to check their performance on
simulating land subsidence phenomena.
The piezometric level in the majority of the deep
water wells constructed in the Kalochori region in
1955 was over the surface. Therefore, for the con-
figuration of the models’ initial pore pressure the
confined aquifers’ piezometric surface was set at
5 m above the ground surface. On the contrary,
the water table depth of the unconfined shallow
aquifer was set 5 m below the ground surface and
it was kept stable during the entire calculation
procedure.
The simulation procedure was divided into four
calculation phases. During the first three phases
the piezometric level of the confined aquifers was
reduced from +5 m to −27 m and during the last
phase the level was reduced for an extra −1 to
−3 m, simulating the variations occurring between
1995–2001 (Soulios, 1999). In order to isolate the
deformations caused by the extra drawdown and
compare them with the PSI data, the displace-
ments were “reset to zero” at the beginning of the
last calculation phase.
The calculations were carried out using the
“consolidation analysis” procedure. The gradual
reduction of the piezometric surface was applied
by means of the “stage construction” routine. Prior
to the four calculation phases mentioned above, the
“Ko procedure” was applied in order to configure
the initial stress fields.
The simulation results (Table 2) indicate that
the displacements calculated by applying the
Mohr–Coulomb model approximate the space
measurements (PSI data) as well as the in situ
records perfectly. The simulations indicated defor-
mations ranging from 10.36 to 40.29 cm, and the
space measurement deformations up to 42.31 cm
(Fig. 4), for the same time period. Considering the
cumulative vertical displacements since the ‘50 s,
the simulation results presented a maximum value
of 4.16 m, coinciding perfectly with the maximum
in situ records along the embankments.
On the contrary, the displacements calculated by
applying the Hardening soil model decline widely
from the measured values. For instance the maxi-
Table 1. Soil data sets parameters coming from the reevaluation of the available laboratory and in situ tests.
Drained
[1] YB [2] BG Silty [3] B Sand [4] BG Silty [5] B Sand [6] B Sand
Draining conditions Sand sand Silty Sand Sand II Silty Sand II Silty Sand III

γunsat [kN/m3] 19.00 17.40 19.50 17.50 18.50 19.50

γsat [kN/m3] 20.00 18.00 22.00 19.80 20.60 22.00
kx, ky [m/day] 8.928 0.638 8.928 0.638 8.928 8.928
[MC]* Eoed [kN/m2] 3950 3100 11600 8800 15000 20000
[HS]** Cc [−] 0.148 0.08 0.08 0.08 0.067 0.423
[HS] Cs [−] 0.0311 0.018 0.0156 0.0168 0.021 0.086
[HS] einit [−] 0.622 0.387 0.387 0.387 0.623 1.243
cref [kN/m2] 10 60 35 80 10 40
ϕ [°] 38.5 7 40 15 38.5 40
ψ [°] 3 0 5 0 3 5
Einc [kN/m2/m] – – – 360 – –
yref [m] – – – 40 – –

*[MC]: Mohr–Coulomb model, **[HS]: Hardening Soil model.

Table 2. FEM vertical displacements. Furthermore the current research presents the
applicability of space-borne interferometry on the
150 m deep 100 m deep study of land subsidence phenomena.
profile profile

Vert. disp. (in cm) MC HS MC

ACKNOWLEDGEMENTS
+5 to −27 m drawdown 376 718 283
1 m extra drawdown 15.6 24.2 10.4 The Terrafirma Extension project has funded the
2 m extra drawdown 27.4 44.3 20.5 SAR imagery processing as well as the interpreta-
3 m extra drawdown 40.3 64.7 30.7 tion presented in this paper. The project is one of ten
services being supported by the Global Monitoring
for Environment and Security (GMES) Service Ele-
ment Programme, promoted and financed by the
mum cumulative vertical displacements indicated European Space Agency (ESA). The authors grate-
by the simulation are more than 7.8 m, exceeding fully acknowledge the German Aerospace Centre
severally all available measured values. (DLR) for having processed SAR data.

5 CONCLUSIONS REFERENCES

Concluding the application of the Mohr–
Coulomb soil model for the direct simulation of
the land subsidence mechanism can be consid-
ered accurate. The required values of the stiffness
parameters can be obtained from oedometer tests
and particularly from the loading stages refer-
ring to the active stresses applied after the ground
water drawdown.
Considering that most of the times land sub-
sidence phenomena cannot be back calculated,
because of the lack of accurate displacement
records, the implementation of a verified simula-
tion procedure cannot be established. In those
cases, the knowledge that the Mohr–Coulomb
soil model, a simple model with low requirements
on mechanical parameters, can provide accurate
results can be proved valuable.
Andronopoulos, V. 1979. Geological and geotechnical
study in the Kalochori (Thessaloniki) area. Institute
of Geology and Mineral Exploration Report, Athens,
p. 90.
Andronopoulos, V., Rozos, D. & Hatzinakos I. 1990.
Geotechnical study of ground settlement in the Kalo-
chori area, Thessaloniki District. Institute of Geology
and Mineral Exploration Report, Athens, p. 45.
Andronopoulos, V., Rozos, D., Hatzinakos, I. 1991.
Subsidence phenomena in the industrial area of
Thessaloniki, Greece. In: Johnson, A. (Ed.). Land
Subsidence, pp. 59–69.
Brinkgreve, R.B.J., Al-Khoury, R., Bakker, K.J. 2002.
Plaxis, Fine Element Code for Soil and Rock Analysis,
2D–Version 8. Balkema, Rotterdam.
Colesanti, C., Ferretti, A., Prati, C. & Rocca, F., 2003.
Monitoring landslides and tectonic motions with the
Permanent Scatterers Technique. Engineering Geol-
ogy (68) pp. 3–14.

56
Ferretti, A., Prati, C. & Rocca, F., 2000. Nonlinear sub-
sidence rate estimation using Permanent Scatterers in
differential SAR interferometry. IEEE Trans. Geosci.
Remote Sens. (38) pp. 2202–2212.
Ferretti, A., Prati, C. & Rocca, F., 2001. Permanent Scat-
terers in SAR interferometry. IEEE Trans. Geosci.
Remote Sens. (39) pp. 8–20.
Hatzinakos, I., Rozos, D., Apostolidis, E. (1990). Engi-
neering geological mapping and related geotechnical
problems in the wider industrial area of Thessalo-
niki, Greece. In: Price, D. (Ed.), Proceedings of Sixth
International IAEG Congress, Amsterdam, Balkema,
pp. 127–134.
Loupasakis, C. & Rozos, D., (2009) Land Subsidence
Induced by Water Pumping in Kalochori Village
(North Greece)—Simulation of the Phenomenon by
Means of the Finite Element Method, Quarterly Jour-
nal of Engineering Geology and Hydrogeology, Geo-
logical Society of London, (42) pp. 369–382.
Massonnet, D., Rossi, M., Carmona, C., Adragna, F.,
Peltzer, G., Feigl, K. & Rabaute, T., 1993. The dis-
placement field of the Landers earthquake mapped by
radar interferometry. Nature (364).
Massonnet, D., Feigl, K.L., Rossi, M. & Adragna, F.
1994. Radar interferometric mapping of deforma-
tion in the year after the Landers earthquake. Nature
(369).
Massonnet, D. & Feigl, K.L., 1995. Discrimination of
geophysical phenomena in satellite radar interfero-
grams. Geophys. Res. Lett., (22).
Massonnet, D., Briole, P. & Arnaud, A. 1995. Defla-
tion of Mount Etna monitored by Spaceborne Radar
Interferometry. Nature, (375).
Massonnet, D., Thatcher, W. & Vadon, H. 1996. Detec-
tion of postseismic fault zone collapse following the
Landers earthquake. Nature (382).
Massonnet, D. & Feigl, K.L., 1998. Radar interferometry
and its application to changes in the Earth’s surface.
Rev. Geophys., (36).
Mouratidis A., Briole P., Astaras A., Pavlidis S.,
Tsakiri M., Ilieva M., Rolandone F. & Katsambalos, K.
2010. Contribution of InSAR and kinematic Gps data
to subsidence and geohazard monitoring in Central
Macedonia (N. Greece), Scientific Annals, School of
Geology, Aristotle University of Thessaloniki, Pro-
ceedings of the XIX CBGA Congress, Thessaloniki,
Greece (100) pp. 535–545.
57
Nicolau, S. & Nicolaidis, M. 1987. Geoelectric study in
Kalochori village of Thessaloniki. Report, Institute of
Geology and Mineral Exploration, Athens, 10.
Psimoulis P., Ghilardi M., Fouache E. & Stiros S., 2007.
Subsidence and evolution of the Thessaloniki plain,
Greece, based on historical leveling and GPS data.
Engineering Geology (90) pp. 55–70.
Rosen, P.A., Hensley, S., Joughin, I.R., Li, F.K., Madsen, S.N.,
Rodriguez, E. & Goldstein, R.M., 2000. Synthetic aper-
ture radar interferometry. Proc. I.E.E.E. (88).
Rozos, D. & Hadzinakos, I. 1993. Geological conditions
and geomechanical behaviour of the neogene sediments
in the area west of Thessaloniki (Greece). Proc. Int.
Symp. on Geotechnical Engineering of Hard Soils—
Soft Rocks, Greece, Anagnostopoulos et al (Eds), A.A.
Balkema, Vol.1, pp. 269–274, 1993 Rotterdam.
Rozos, D. Apostolidis, E. & Hadzinakos, I. 2004. Engi-
neering geological map of the wider Thessaloniki area,
Greece. Bulletin of Intern. Assoc. of Eng. Geol. & the
Environment, Springer—Verlag, Vol. 63, pp.103–108,
2004. Rotterdam.
Singhroy, V., Mattar, K.E. & Gray, A.L. 1998. Landslide
characterisation in Canada using interferometric SAR
and combined SAR and TM images. Advances in
Space Research (21) pp. 465–476.
Soulios, G. 1999. Research for the development of the
aquifers in the low lands on the west of Thessaloniki
for the interests of the Water Company of Thessalo-
niki. unpublished technical report, Aristotle Univer-
sity of Thessaloniki, p. 99.
Stiros, S.C. 2001. Subsidence of the Thessaloniki (north-
ern Greece) coastal plain, 1960–1999. Engineering
Geology, (61) pp. 243–256.
Tralli, D.M, Blom, R.G., Zlotnicki, V., Donnellan, A.,
Evans, D.L. 2005. Satellite remote sensing of earth-
quake, volcano, flood, landslide and coastal inunda-
tion hazards. Journal of Photogrammetry and Remote
Sensing (59) pp. 185–198.
Zebker, H.A. & Goldstein, R.M. 1986. Topographic
Mapping From Interferometric Synthetic Aperture
Radar Observations. Journal of Geophysical Research
(91), pp. 4993–4999.
Zebker, H.A. & Villasenor, J. 1992. Decorrelation in
interferometric radar echoes. IEEE Trans. Geosci.
Remote Sens. (30).
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Numerical modeling of fracturing in soil mix material

G. Van Lysebetten & A. Vervoort

KU Leuven, Leuven, Belgium

N. Denies & N. Huybrechts

Belgian Building Research Institute, Geotechnical division, Limelette, Belgium

J. Maertens
Jan Maertens BVBA, Belgium
KU Leuven, Leuven, Belgium

F. De Cock
Geotechnical Expert Office GEO.BE, Lennik, Belgium

B. Lameire
Belgian Association of Foundation Contractors ABEF, Brussels, Belgium

ABSTRACT: The deep soil mixing technique consists of an in situ mechanical mixing of the soil with
an injected binder (e.g. cement). However, the presence of soil inclusions (poorly or even unmixed soil)
in the artificial material is unavoidable. This heterogeneous character of soil mix material makes it differ-
ent from traditional building materials. The presented research investigates the influence of the volume
percentage of inclusions on the strength, stiffness, stress-strain behaviour and fracture pattern of soil mix
material. 2D numerical simulations are conducted using a Discrete Element Program (UDEC) and the
results are compared with experimental data. It is observed that the reduction of the strength and stiffness
of a sample is significantly larger than the weighted average of the UCS and Young’s modulus, taking into
account the volumes of the well mixed material and the softer inclusions. However, the strength is remark-
ably more affected by the volume percentage of inclusions than the stiffness. Moreover, other parameters
than the percentage of weak material are also important (e.g. shape, size and relative position of the inclu-
sions) and result in wide ranges of resulting strength and stiffness.

1 INTRODUCTION However, since a natural material (i.e. soil) is

being mixed, it is to be expected that the entire wall
1.1 Deep soil mixing technique is not perfectly mixed and homogeneous. The vol-
ume percentage of inclusions depends on the soil
The application of the soil mix technology in
type and mixing technique (e.g. 0 to 3.5% in sandy
Belgium is sharply increasing (Denies et al. 2012a).
soils up to 35% and more in stiff clays (Ganne et al.
Next to soil improvement applications, soil mix
2011). This heterogeneous character makes the
walls are extensively used for excavation support
material different from traditional building materi-
because of its economic and environmental advan-
als. In order to formulate alternative design rules,
tages compared with classical techniques such as
a Flemish regional research project (IWT 080736)
concrete secant pile walls, diaphragm walls and
in collaboration with the BBRI, ABEF and KU
king post walls (i.e. soldier pile walls). The soil mix
Leuven was initiated. This paper deals with one
technique is based on an in situ mechanical mixing
particular aspect of the project which is the influ-
of the soil with an injected binder (e.g. cement). By
ence of the volume percentage of inclusions on the
executing overlapping rectangular panels or cylin-
strength, stiffness, stress-strain behaviour and frac-
drical columns a continuous wall is obtained. As
ture pattern of soil mix material.
soon as the panel or column has been mixed, steel
H or I profiles are inserted into the fresh soil mix
1.2 Aim of the study
material to increase the shear and bending resist-
ance of the wall. Depths up to 20 meters are cur- Apart from an experimental part consisting of lab-
rently reached. oratory experiments on small scale samples as well

58
as large blocks, complementary numerical simula-
tions (2D) are conducted in UDEC, a numerical
program based on the discrete element method.
The followed approach allows simulating frac-
ture initiation and growth in and around the soft
soil inclusions. This allows a correct distinction
between shear and tensile fractures, but helps also
to understand and quantify the effect of heteroge-
neities in the studied material. A large number of
simulations have been executed to study the influ-
ence of the volume percentage, shape, number and
relative position of the inclusions.
1.3 Advantage of discrete simulations
During the past decades, several numerical
approaches have been elaborated to realize the sim-
ulation of crack initiation and propagation. Some
of these codes are based on finite elements, on
finite differences or on displacement discontinuity
boundary elements. Though, it is found that the
Distinct Element Method (DEM) is a more than
valuable alternative (Debecker et al. 2006). Several
DEM codes have been developed. In the Parti-
cle Flow Code (PFC) materials are modeled as a
dense packing of rigid spherical elements, bonded
together at their contact points (Potyondy &
Cundall 2004). For this study, the Universal Dis-
tinct Element Code (UDEC) is chosen. This code
is originally developed for the simulation of frac-
tured rock mass behavior, e.g. slope stability and
rock fall which depend on the activation of existing
fractures (Itasca 2004). However, in the past it has
been successfully used for the numerical modeling
of fracture initiation and growth in rock (Debecker
2009, Tempone & Lavrov 2008).
2 NUMERICAL SIMULATIONS
2.1 Concept and global approach
UDEC is a 2D numerical program that is based
on the discrete element method (Cundall 1971).
The discrete element model consists of discrete
blocks that are mutually connected by contacts.
For these contacts tensile and shear failure crite-
ria are defined, allowing them to open and deform
upon activation. The UDEC solution scheme is
based on a (dynamic) explicit finite-difference
method which is also used in continuum analysis
(Cundall & Board 1988).
The philosophy followed in this study is that by
dividing a medium in multiple discrete small blocks
(tightly bounded together), their boundaries can
act as potential fracture paths when an external
load is applied (Debecker 2009, Vervoort et al.
2012). Or in other words, a contact between two
59
adjacent blocks does not represent a physical crack
as long as it is not activated. An example of a mesh
of triangles for a rectangular medium is shown in
Figure 1a. Since fracture growth is limited to the
available contacts, the distribution of the contacts
should be as uniformly as possible so that there is
no bias by the mesh. Moreover, on a larger scale a
given fracture with a certain overall direction tends
to develop as a combination of activated individ-
ual contacts with sometimes very different orienta-
tions, as illustrated in Figures 1b and 1c for a 60°
inclined crack.
In this study, the blocks only deform elastically.
The strength parameters of the material are thus
not explicitly implemented in the blocks. However,
the contacts are modelled with a Mohr-Coulomb
failure criterion with tension cut-off. After activa-
tion, the strength parameters are set to residual
values. In addition to this, UDEC models the con-
tacts with stiffness, in shear and normal direction
(respectively ks and kn). This stiffness allows defor-
mation of contacts prior to activation. The defor-
mation due to this contact stiffness has an effect on
the stress distribution within the sample (Debecker
et al. 2006).
2.2 Calibration of model parameters
On the one hand, material properties as obtained
from laboratory tests are assigned to the triangu-
lar blocks (Table 1). On the other hand, the stiff-
ness and strength parameters of the contacts are
not physically measurable and their values are
deduced from a calibration process. First, the con-
tact properties for a homogeneous model of well
mixed material are calibrated based on the stress-
strain behavior and typical fracture patterns dur-
ing uniaxial compression tests. These properties
are all well-known from laboratory tests (e.g. peak
strength, Young’s modulus, post-peak behavior,
Figure 1. (a) Simplified example of adjacent triangular
blocks that form the sample. A contact only represents
a physical crack when it is activated. (b) Fracture along
activated contacts with a global dip of 60° (straight line).
(c) Apart from the activated contacts the sample is still
intact.
Table 1. Material properties of the blocks correspond-
ing to the matrix material and the soft inclusions (Van
Lysebetten 2011).
Material properties Matrix Inclusions
Density ρ [kg/m3] 1820 1820
Young’s modulus E [GPa] 11.6 0.165
Poisson’s ratio υ [−] 0.3 0.4
Bulk modulus K [GPa] 9.67 0.275
Shear modulus G [GPa] 4.46 0.059
Figure 2. (a) 2D sample (120 × 240 mm) from a real soil
mix column cross-section. (b) Mesh of the basic model.
failure by 60 to 70° inclined shear fractures,
etc.). This time-consuming process finally leads
to the matrix—matrix contact strength and stiff-
ness parameters. Next, the strength properties of
matrix—inclusion and inclusion—inclusion con-
tacts are easily deduced from the matrix—matrix
contact strength by applying the strength ratios of
mixed and unmixed material.
2.3 Basic model
The starting point for the soil mix samples is a sec-
tion through a real soil mix column, of which an
area of 120 × 240 mm is considered (Fig. 2a). This
2D sample contains 11 inclusions which account
for 11 vol%. Based on this sample, the basic model
is assembled. It consists of 9 of the 11 inclusions
or 10 vol% of inclusions (Fig. 2b). Note that the
term vol% is justified, since the 2D model in fact
has a unit thickness. Moreover, using this termi-
nology the link with the measurements on real soil
mix material is more easily made.
2.4 Mesh generation algorithm
As illustrated by the basic model in Figure 2b, the
mesh of a soil mix sample is built up by a com-
bination of a predefined structure and a certain
degree of randomness. Or in other words, the
60
predefined shape and location of the inclusions
are incorporated in a (random) mesh of triangles.
Since UDEC does not provide a mesh generator
that can meet these requirements, a mesh generator
was implemented in Matlab. The algorithm divides
the sample in multiple small rectangular sectors in
which grid nodes are randomly generated around
the predefined nodes of the inclusions. Finally,
Delaunay triangulation is performed on the entire
collection of nodes (i.e. nodes that define the
boundary of the inclusions as well as the randomly
generated nodes). The triangulation algorithm is
based on the Quickhull algorithm for constructing
convex hulls (Barber et al. 1996). To keep calcula-
tion time in UDEC reasonable, the number of tri-
angles is limited to approximately 8000. Of course,
this limits the minimal representative length of
the individual contacts of the discrete model and
consequently the minimum fracture length that
can be simulated (Debecker & Vervoort 2006). The
quality of the generated mesh is largely dependent
on the choice of the input parameters, such as the
minimum distance between grid nodes in relation
to the dimensions of the sectors.
An appropriate mesh for fracture simulation
must meet several geometrical requirements. First,
the boundary of the inclusions must be respected
by the edges of the generated triangles (the edges
are called ‘elements’ from now on). Second, the
distribution of the element orientation should be
as uniformly as possible, as discussed in paragraph
2.1. Third, adjacent triangles should be of approx-
imately the same size and sharp triangles should
be avoided, both to guarantee reasonable solution
accuracy in UDEC. In order to check the gener-
ated triangles, quality parameters have been intro-
duced (Debecker & Vervoort 2006):
• ‘Minimum triangle angle’: the smallest angle of
the triangle and a degree to evaluate the skinni-
ness of the triangles. Its aimed interval is [20°,
60°].
• ‘Triangle aspect ratio’: a measure for the maxi-
mum edge size to the minimum altitude of the
triangle. Its ideal interval is [1, 2.5].
• ‘Triangle quality’ Q: the ratio of the smallest
angle to the largest angle of the triangle. The
ideal interval of Q is [0.4, 0.9].
The present algorithm provides an overview of
all quality parameters after the mesh has been gen-
erated. Note that all quality parameters are only
indicative and not stringent.
2.5 Results of the simulations
In order to study the influence of several inclusion
properties (e.g. vol%, shape, size, relative position),
about 200 uniaxial compression tests on soil mix
models (120 × 240 mm) with different inclusion
geometries have been simulated.
2.5.1 Influence of volume percentage, shape
and size of the inclusions
On the basis of the real 2D section and the basic
model (Figs. 2a and 2b), 68 additional models are
generated with 1, 5, 10 and 20 vol% inclusions. For
these models, the relative position of the inclu-
sions is varied as well as the shape (e.g. explic-
itly rounded or sharp-ended) and the number of
inclusions (e.g. 1 large rounded inclusion against
3 smaller rounded inclusions).
The evolution of the fracture pattern of the basic
model (Fig. 2b) is presented in Figures 3a, b and c
for 3 different loading steps (just before, at and just
after peak strength). The basic model has a UCS
value of 4.4 MPa against a UCS value of 11.7 MPa
for the pure soil mix sample without inclusions. The
9 inclusions (a total of 10 vol%) reduce the strength
thus with more than 60%. At relatively low vertical
loads (less than 50% of the peak strength) a part of
the material inside the inclusions starts to fail. This
is logical since the inclusions have lower strength
properties. However, failure is clearly limited to
specific parts of the inclusions. At 2.6 MPa vertical
load several vertical splitting type of fractures are
induced in the matrix at the top and at the bottom
of the grouped inclusions (Fig. 3a). At higher ver-
tical loads, especially after the UCS value has been
Figure 3. (a, b, c) Evolution of the fracture pattern of
the basic model. (d) Examples of real fracture patterns of
uniaxial loaded soil mix samples.
61
reached (Fig. 3b), also shear fractures are induced.
Finally, this results in a shear zone that extends
from the upper left corner to the lower right corner
of the sample (Fig. 3c).
Figure 3d presents three detailed fracture pat-
terns of tested samples, cored from a real soil mix
wall with several soil inclusions. The inclusions
are clearly visible, as well as the induced fractures.
Moreover, the observed fractures are comparable
to the simulated fracture patterns around inclu-
sions. For example in the left picture, two fractures
are induced at both extremities of the bottom part
of the inclusion, which is also observed around
several inclusions in the simulations. In the central
and right pictures, fractures are observed some-
where along the middle of the inclusion or at the
most top or bottom point of the inclusion. Again,
these fractures are also visible in the simulations.
Figure 4 compares laboratory data of in situ
cored samples from 12 different Belgian construc-
tion sites with the results of the numerical simula-
tions (69 models). Both sets of results show a similar
downward trend in function of the volume percent-
age of inclusions. The results of the simulations are
clearly situated within the (wider) range of labora-
tory results. The fact that the experimental data have
a wider range is logical, as the soil type at the 12 con-
structions sites is different. Clayey, as well as loamy
and sandy soils were encountered, and of course the
mechanical characteristics of the different soil mix
materials are different (Denies et al. 2012b). For the
numerical simulations one single value is considered
for the model without any inclusions, i.e. a strength
value of 11.7 MPa for the homogeneous sample.
Other reasons for the wider range are that in reality a
real soil mix sample is never perfectly homogeneous
Figure 4. Comparison of the variation of strength
(UCS) in function of the volume percentage of unmixed
material for the numerical simulations (results based on
Van Lysebetten (2011)) and for the results of laboratory
experiments.
and that two samples even without any inclusions
(or with exactly the same amount of inclusions) still
have different strength values because of variations
in e.g. cement content.
Therefore, the numerical simulations are more
suitable to study the effect of the volume percent-
age of unmixed material on the strength of a sam-
ple. The numerical results of Figure 4 are based
on the basic model and the 68 additional models
(diamond-shaped markers). First, the decrease
of strength is clearly larger than the percentage
of inclusions in the sample. For a mere 1% of
unmixed material strength is reduced by 13% on
average. For 10% of inclusions even half of the
strength disappears on average (Van Lysebetten
2011). The same can be concluded for the stiffness
(Vervoort et al. 2012), though it is less influenced
(for 1 and 10 vol% of inclusions the stiffness is
reduced on average with respectively 3 and 32%).
Second, there is an overlapping zone between the
strengths of 5 and 10% and of 10 and 20%. This
overlap is caused by the relatively large range for
each volume percentage of inclusions. For the stiff-
ness, there is just no overlap between successive
volume percentages.
Strength and stiffness of a sample are clearly
not only determined by the volume percentage.
This is further analyzed for additional models with
10% of inclusions. It is observed that strength and
stiffness are reduced more by sharp-ended inclu-
sions than by rounded inclusions. Apart from this,
the strength of a sample also drastically decreases
when less inclusions (but with the same shape and
the same total percentage of inclusions) are present
(Vervoort et al. 2012).
2.5.2 Influence of the relative position
of the inclusions
Apart from the volume percentage, size and number
of inclusions, the relative position has also an influ-
ence on strength and stiffness. 123 additional mod-
els are generated to study this effect. All additional
models represent 10% of inclusions and are com-
posed of two sets of respectively 3 and 5 inclusions.
The main difference between all models is the relative
position of these inclusions. In some models, they
are concentrated, in others they are spread as wide
as possible, or they are nicely situated along a vertical
line or along a diagonal, etc. The ranges of strength
(3.8–8.1 MPa) and stiffness (5–8.7 GPa) increase in
comparison with the results of the 69 models for
the study of the influence of the volume percent-
age, shape and number of inclusions (respectively
3.1–9.2 MPa and 6.3–7.9 GPa). This is logical since
the more simulations, the greater the probability that
new minimum and maximum values are found.
The simulations clearly show the influence of the
relative position of the inclusions on the strength
62
and stiffness of a sample (Van Lysebetten et al.
2012). For example, it is observed that strength
and stiffness decrease if the inclusions are located
along the diagonal of the sample. This configu-
ration probably facilitates a global shear failure.
Figure 5a shows that there is a clear negative trend
between strength and stiffness and wi/t: the ratio
of the total width of inclusions, i, and the sample
width, t (Fig. 5b). However, a significant overlap is
also observed, e.g. the largest UCS value for a width
ratio of 60% is still larger than the smallest value for
20% ratio. For the stiffness, the overlap is probably
even larger. In these additional simulations, it was
also observed that the strength and stiffness also
decrease if the vertical distance between the inclu-
sions decreases, since inclusions tend to act as one
larger inclusion. However, this effect is less impor-
tant than the influence of the ratio wi/t.
2.5.3 Link with simulated fracture patterns
Figs 6a, b, c and d show the evolution of the frac-
ture pattern of 4 models for successive loading
steps. At relative low stress levels (less than 50% of
the strength) vertical extension type of fractures
initiate at the lower and upper parts of the inclu-
sions (see also Figs 3a, b and c for the evolution
of the fracture pattern of the basic model). Only at
stress levels near the UCS value of the sample, shear
fractures are observed at the sides of the inclusions
after which the sample finally fails due to a combi-
nation of shear and extension type of fractures.
Most of the described effects of inclusions can
be related to and explained by the observed fracture
patterns. For example, stress peaks near sharp-
ended inclusions are higher than close to rounded
inclusions, causing fracture initiation at lower
Figure 5. (a) Relation between the strength (left)
and stiffness (right) and wi/t. (b) Definition of the ratio
wi/t = i/t.
Figure 6. Illustration of the fracture pattern at successive loading steps of 4 samples of the original set each with
10% of inclusions. Shape and number of inclusions are different. The stiffness of the models is respectively 7.2 GPa,
8.0 GPa, 6.7 GPa and 7.4 GPa.

stress levels. Finally, this results in lower strength the sample is damaged. Again, this results in lower
and stiffness (compare Figs 6a and c). Fractur- strength and stiffness.
ing also initiates at lower stress levels when 1 large
inclusion is present instead of 3 smaller inclusions,
3 CONCLUSIONS
at least if these 3 inclusions are vertically aligned
(compare Figs 6a, b and d). Again, this is caused
Although calibration of the numerical discrete
by higher stress peaks near the largest inclusions
model is very time-consuming, the results show
(i.e. because a higher horizontal distance has to be
that numerical simulations are very interesting
bridged).
to perform relatively easily sensitivity analyses.
However, the fact that the 3 inclusions are verti-
The influence of soil inclusions on the strength
cally aligned is very important. Figure 6d shows the
and stiffness of soil mix material has been inves-
fracture pattern of a sample with the same 3 inclu-
tigated by a total of 192 simulations. It has been
sions positioned diagonally. This sample results in
observed that soft inclusions have a large impact
about the same strength and stiffness as the sample
on the resulting strength and stiffness of a sam-
with 1 large rounded inclusion. The large reduc-
ple. However, not only the volume percentage is
tion of strength in comparison with the model
important, but also the relative position of the
with 3 vertically aligned inclusions is caused by
inclusions, their shape and number, etc. Moreover,
the diagonal orientation of the inclusions result-
UDEC has shown to be suitable for the simula-
ing in a larger wi/t ratio. Moreover, the stress peaks
tion of fracture initiation and growth in soil mix
around the stress relieved zones interfere and prob-
material with soft inclusions. Similar trends of
ably amplify each other.
the influence of the volume percentage of inclu-
Finally, if the vertical spacing between the inclu-
sions on the strength are observed for the results
sions of Figure 6b is decreased, the inclusions
of laboratory data of in situ cored samples from
tend to form one larger inclusion. Overlapping
different Belgian construction sites and the results
zones with increased stresses expand again and the
of numerical simulations. The simulated fracture
amplifying effect grows. This causes earlier fracture
patterns are also comparable with these observed
initiation (at lower vertical loads), but the vertical
in real soil mix samples.
extension type fractures around the central inclu-
sion also reach faster the upper and lower inclu-
sions. The travelled distance through the stronger ACKNOWLEDGEMENTS
material before reaching other weak inclusions is
thus lower. Because propagation through these This project has been made possible by the finan-
inclusions goes much faster, a much larger part of cial support of the Flemish government agency for

63
Innovation by Science and Technology (IWT). It
is the result of cooperation between the Belgian
Building Research Institute (BBRI), the Belgian
Association of Foundation Contractors (ABEF)
and KU Leuven.
REFERENCES
Barber, C.B., Dobkin, D.P, Huhdanpaa, H. 1996. The
Quickhull algorithm for convex hulls. ACM Trans-
actions on Mathematical Software, Vol. 22 (4),
pp. 469–483.
Cundall P.A. 1971. A computer model for simulating
progressive large scale movements in block rock sys-
tems. Proceedings of the International Society for Rock
Mechanics Symposium, paper II-8.
Cundall P.A. & Board M. 1988. A microcomputer pro-
gram for modeling large-strain plasticity problems.
Proceedings of the 6th International Conference on
Numerical Methods in Geomechanics. Rotterdam
(Netherlands), pp. 2101–2108.
Debecker B. 2009. Influence of planar heterogeneities on
the fracture behavior of rock, Ph.D. Dissertation, KU
Leuven, Leuven (Belgium).
Debecker B., Vervoort A., Napier J.A.L. 2006. Fractur-
ing in and around a natural discontinuity in rock: a
comparison between boundary and discrete element
models. Proceedings of the 5th International Confer-
ence on Engineering Computational Technology, Las
Palmas de Gran Canaria (Spain), paper 168.
Debecker B. & Vervoort A. 2006. A 2D triangular Delau-
nay grid generator for the simulation of rock features.
Proceedings of the 5th International Conference on
Engineering Computational Technology, Las Palmas
de Gran Canaria (Spain), paper 220.
Denies N., Huybrechts N., De Cock F., Lameire B.,
Maertens J., Vervoort A. 2012a. Soil mix walls as
retaining structures, Belgian practice. Proceed-
ings of TC211 International Symposium on Ground
Improvement, Brussels (Belgium), Vol. 3, pp 83–97.
64
Denies N., Huybrechts N., De Cock F., Lameire B.,
Vervoort A., Maertens J. 2012b. Soil mix walls as retain-
ing structures, mechanical characterization. Proceedings
of the TC 211 International Symposium on Ground
Improvement, Brussels (Belgium), Vol. 3, pp. 99–115.
Ganne P., Denies N., Huybrechts N., Vervoort A.,
Tavallali A., Maertens J., Lameire B., De Cock F. 2011.
Soil Mix: influence of soil inclusions on the structural
behavior. Proceedings of the 15th European conference
on soil mechanics and geotechnical engineering, Athens
(Greece), pp. 977–982.
Itasca 2004. UDEC v4.0 manual. Itasca Consulting
Group, Inc., Minnesota, USA.
Potyondy D.O. & Cundall P.A. 2004. A Bonded-
Particle Model for Rock. International Journal for
Rock Mechanics & Mineral Sciences, Vol. 41(8), pp.
1329–1364.
Tempone P. & Lavrov A. 2008. DEM modeling of mud-
losses into single fractures and fracture networks.
Proceedings of the 12th international conference
of the international association for computer meth-
ods and advances in geomechanics, Goa (India),
pp. 2475–2482.
Van Lysebetten G., Vervoort A., Maertens J.,
Huybrechts N. 2012. Discrete modelling for the study
of the effect of soft inclusions on the behaviour of soil
mix material. In preparation.
Van Lysebetten G. 2011. Soil Mix for Construction pur-
poses: Quality control. M.Sc. Thesis (Geotechnical
and Mining Engineering) KU Leuven, Leuven.
Vervoort A., Van Lysebetten G., Tavallali A. 2012.
Numerical modeling of fracturing around soft
inclusions. Proceedings of the Southern Hemisphere
International Rock Mechanics Symposium, Sun City
(South-Africa), pp. 33–46.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

A 3D practical constitutive model for predicting seismic liquefaction

in sands

A. Petalas & V. Galavi

Plaxis BV, Delft, The Netherlands

R.B.J. Brinkgreve
Plaxis BV, Delft, The Netherlands
Delft University of Technology, Delft, The Netherlands

ABSTRACT: This paper presents a three dimensional formulation of a simple and practical constitutive
model developed for evaluating seismic liquefaction in sands. The model is an extension of the two dimen-
sional UBC-SAND model developed at University of British Colombia (Beaty & Byrne 1998) which
utilises isotropic and simplified kinematic hardening rules for primary and secondary yield surfaces, in
order to take into account the effect of soil densification and predict a smooth transition into the liquefied
state during undrained cyclic loading. By means of a simplified Rowe stress-dilatancy theory the model is
capable of modelling cyclic liquefaction for different stress paths. To show the capability of the model to
predict cyclic liquefaction in soils, the mechanical behaviour of some sands is numerically studied under
direct simple shear conditions and compared with experimental data. The effect of densification and
parameters selection on the results is discussed. Finally, the model is used for simulating an experimental
dynamic centrifuge test and the numerical results are compared with the real measurements.

1 INTRODUCTION function based on Drucker-Prager criterion is used

The use of numerical methods in earthquake
engineering practice for liquefaction prediction
has many pitfalls, due to the very complicated
soil behaviour under undrained dynamic loading.
This behaviour is investigated by researchers and
field engineers for decades but still there is not
enough confidence from both sides in order to use
these kind of tools in the design of structures and
superstructures.
The UBC3D-PLM model which is presented in
this paper is an attempt of using a simple but pow-
erful plasticity framework to analyse and predict
the onset of dynamic induced liquefaction. The
UBC3D-PLM is a 3-D extension of the UBC-
SAND model which was introduced by Puebla
et al (1997). The 3D formulation was initially devel-
oped and implemented as a User Defined Model
in PLAXIS by Tsegaye (2010). The formulation
of the model described in this paper is a continua-
tion of the work presented by Petalas et al. (2012)
and new elements are included in order to achieve
higher accuracy and to model more key features of
soil behaviour under dynamic and cyclic loading.
The UBC3D-PLM model uses the MohrCoulomb
yield condition in a 3-D principal stress space. More-
over, a modified non-associated plastic potential
in order to maintain the assumption of stress-strain
coaxiality in the deviatoric plane for a stress path
beginning from the isotropic line (Tsegaye 2010).
In the proposed formulation of the UCB3D-PLM
model a soil densification rule is added to predict
more realistic evolution of excess pore pressures
during cyclic loading. A correct procedure of count-
ing the cycles during dynamic and cyclic loading is
implemented in order to achieve higher accuracy
in the stress paths which do not start from the iso-
tropic line. The densification rule allows the increase
of thepore pressures with a decreasing rate during
shearing which is observed during experimental
studies by different researchers (Martin et al. 1975).
Moreover, a rule which governs the predicted
behaviour during post-liquefaction is introduced
in order to model the degradation of the stiffness
during the liquefied state. This also allows the pre-
diction of cyclic mobility in dense sands and solves
the issue of volumetric locking which was encoun-
tered in previous formulations. Finally, a rule for
taking into account the non-symmetry in cycles is
implemented in order to correct the evolution of
excess pore pressure during loading.
The main characteristics of the model are pre-
sented in the next section after which a validation
of the model is presented.

65
2 KEY FEATURES OF THE UBC3D-PLM 2
⎛ p′ ⎞
np
⎧⎪ ⎛ sin φ ′ ⎞ ⎫⎪
G *
kGp ⎜ ⎨1 − ⎜
mob
m
⎟ RF ⎬ (5)
2.1 Yield surfaces ⎝ PA ⎟⎠ ⎪⎩ ⎝ sin φ p′ ⎠ ⎪⎭
The UBC3D-PLM model uses the well-known
Mohr-Coulomb yield function generalized in 3-D where kGp is the plastic shear modulus number; dλ
principal stress space (Tsegaye 2010). The formula- stands for the plastic multiplier; np is the plastic
tion of the surface is given bellow: shear modulus exponent; φmob is the mobilized fric-
tion angle; φpeak is the peak friction angle; and Rf is
σ ma
′ x −σm
′ in
fm = (1) the failure ratio qf /nult based on the Duncan-Chang
2 model, ranging from 0.5 to 1.0, where qf is the devi-
⎛σ ′ + σm ′ in ⎞ atoric stress ratio at failure and qult is the ultimate
− ⎜ max c ′ cot φ p′ ⎟ i φm
′
⎝ 2 ⎠ mob
stress ratio from the best fit hyperbola. The hard-
ening rule relates the amound of plastic strain with
where σ ma′ x is the maximum principal effective
the mobilized friction angle which is computed
stress and σ mi
′ n is the minimum principal effective
at each step. A schematization of the hyperbolic
stress, c′ is the cohesion of the soil, φ p′ is the
shear hardening rule is depicted in Figure 1.
peak effective friction angle of the soil and φm′ is
the mobilized friction angle during hardening.
The intermediate stress does not influence the yield 2.4 Plastic potential function and flow rule
surface in three dimensional stress space.
A non-associated flow rule based on the Drucker-
Prager plastic potential function is formulated in
2.2 Elastic behaviour 3-D space as follows:
The elastic behaviour which occurs within the yield
6 sinψ m
surface is governed by a stress dependent non-lin- g q− ( p + c cot φ p ) (6)
ear rule defined by Puebla et al. (1997). The bulk 3 − sinψ m
modulus K and the elastic shear modulus G are
stress dependent and are given by the following where, g is a conical surface passing through the tri-
equations: axial compression point and is independent of the
Lode angle θL. The q and p terms are the deviatoric
me
⎛ p⎞ and the isotropic parts of the stress tensor. The influ-
Ke kBe PA ⎜ ⎟ (2) ence of the Lode angle in different stress paths was
⎝ PA ⎠ investigated with a Lode angle dependent formula-
ne tion of the plastic potential function and proved
⎛ p⎞ not to play a key role during modelling the onset of
Ge kGe PA ⎜ ⎟ (3)
⎝ PA ⎠ liquefaction.
In the UBC3D-PLM model the mobilized dila-
where kBe and kGe are the bulk and the shear moduli tancy angle is computed based on the flow rule
numbers respectively, at a reference stress level. The
factors ne and me are parameters to define the rate
of stress dependency of stiffness. In the literature,
the reference stress level Pref is commonly taken
as the atmospheric pressure (PA = 100 kPa). Pure
elastic behaviour is predicted by the model during
unloading.

2.3 Elasto-plastic behaviour

In the latest version of the UBC3D-PLM two yield
surfaces (primary and secondary) are introduced
(Petalas et al. 2012) and its formulation is based
on mixed isotropic and kinematic hardening. The
hardening rule as introduced by Puebla et al. (1997)
for both yield surfaces is described as follows:

⎛ pA ⎞ ∗ Figure 1. Schematization of the hyperbolic shear hard-

d sin φmob . G dλ (4) ening rule for both the primary and secondary yield sur-
⎝ pm ⎠ face (Puebla 1997).

66
 simplified kimematic hardening rule is used for the
secondary surface.
The plastic shear modulus number kGp during
primary loading is identical with the one entered
as input parameter. However, during secondary
loading it increases as a function of the number of
cycles in order to capture the effect of soil densifi-
cation as follows:

⎛ n ⎞
KGp KGp 4 + rev ⎟ × hard f achard (11)
⎝ 2 ⎠

where, nrev is the number of stress reversals from

Figure 2. Schematization of the linear flow rule. loading to unloading or vise versa. Four reversals
in any direction counts for one full cycle. The term
hard is a factor to correct the densification rule for
used in the original UBCSAND model by Puebla loose non cohesive soils and f achard is a multiplier
et al. (1997). It is derived by the stress-dilatancy which is a user input parameter in order to cali-
theory developed by Rowe (1962), linearised and brate the densification rule.
simplified according to energy considerations. The This rule is the result of calibrating a number of
relationship is given as follows: direct simple shear tests. Thus, the calibration factor
plays a key role when the user wants to model dif-
d p
υ si ψ mobdγ p
(7) ferent stress paths (cyclic triaxial tests etc.) and the
final value is a matter of judgement according to
sinψ mob i φm′ ob − sin φcv
sin ′ (8) the most critical stress path for a specific problem.
It finally leads to an increase of the excess pore
where, d∈υp is the plastic volumetric strain incre- pressure during undrained cyclic loading until the
ment and φcυ is the constant volume friction angle. liquefied state is approached. The rate of genera-
The schematization of the flow rule is given in 2. tion of excess pore pressure decreases by increasing
number of cycles which is proven via experiments.
2.5 Stress reversal and densification
2.6 Post liquefaction behaviour and cyclic
Based on the mobilized friction angle an unload- mobility
ing-reloading criterion is defined in the model as
follows: One important issue during the modelling of cyclic
(Unloading; elastic behaviour) liquefaction in sands is the volumetric locking. The
evolution of the volumetric strains, after the stress
sin φmo
e
φmo
0
(9) path reaching the yield surface defined by the peak
b b
friction angle, becomes constant due to the formu-
lation of the flow rule (in Equation 8 φmo ′ b becomes
(Loading or reloading; elasto-plastic behaviour) φ p′ and remains constant while φcv ′ is also constant).
Due to this issue the stiffness degradation of the
sin φmo
e
b φmo
0
b (10) soil due to the post-liquefaction behaviour of loose
non-cohesive soils or due to the cyclic mobility of
where, φmo0
b is the initial and φ mob is the current
e
dense non-cohesive sands, which is observed in the
mobilized friction angle. During unloading, pure experimental studies, cannot be modelled. This lim-
elastic behaviour is predicted until the stress point itation is solved in the formulation of the UBC3D-
reaches the p′ axis. PLM with the implementation of an equation
A soil densification rule is introduced in order which gradually decreases the plastic shear modu-
to have higher accuracy in the predicted evolution lus kGp as a function of the generated plastic devia-
of the excess pore pressures following Beaty & toric strain during dilation of the soil element.
Byrne (1998). A secondary yield surface was also The stiffness degradation is formulated based on
introduced in the model for the secondary load- the plastic deviatoric strain related with the dila-
ing in order to ensure a smooth transition into tion of the soil element, due to the deconstruction
the liquefied state of the soil. The secondary yield of the soil skeleton which occurs during dilative
surface generates less plastic strains compared to behaviour (Beaty & Byrne 2011). This leads to the
the primary yield surface. An isotropic hardening decreasedsoil stiffness during contraction which
rule is used for the primary yield surface, while a follows after the unloading phase. This behaviour

67
 Table 1. Input parameters for the validation of the
UBC3D-PLM in modelling monotonic element tests,
cyclic element tests and a dynamic centrifuge test.

Cyclic and
Monotonic tests dynamic tests

Input Monotonic Monotonic Cyclic Dynamic

parameters T×C DSS DSS centrifuge

φp(deg) 33.7 33.7 33.3 31.2

φcυ(deg) 33 33 33 34.6
KBe (−) 300 300 848 720
KGe (−) 300 300 594 1031
Figure 3. Undrained cyclic shear stress path reproduced KGp (−) 310 98.3 243 700
with UBC3D-PLM for dense sand. Cyclic mobility, stiff- me and ne(−) 0.5 0.5 0.5 0.5
ness degradation and soil densification are mentioned on np(−) 0.5 0.5 0.4 0.4
the graph. Rf (−) 0.95 0.95 0.81 0.74
PA(kPa) 100 100 100 100
is presented in Figure 3 picturing the process of σt(kPa) 0 0 0 0
cyclic mobility of a dense sand. The stiffness deg- f achard (−) 1 1 0.45 0.45
radation is computed as follows: N160(−) 10 10 8 13
facpost(−) 1 1 0.01 0.01
× 110
KGp KGp × e E ; Edil × 110 ≤ fac post (12)

where Edil is accumulation of the plastic deviatoric

strain which is generated during dilation of the soil
element. With the input parameter facpost the mini-
mum value of the exponential multiplier term in
the above mentioned equation is set.

3 VALIDATION OF THE UBC3D-PLM

IN ELEMENT TESTS

3.1 Validation of the UBC3D-PLM

in monotonic loading
The validation of the UBC3D-PLM in monotonic
loading is presented in this section. The behaviour Figure 4. Numerical modeling of loose Syncrude sand
of loose Syncrude sand is modelled with UBC3D- under undrained triaxial compression. Comparisson of
the UBC3D-PLM with the original 2D UBCSAND and
PLM and the numerical results are compared with experimental data.
experimental data as well as numerical results
(modelled with the original 2D UBCSAND model)
both published by Puebla et al. (1997). The input
parameters for modelling the monotonic triaxial
compression test (TxC) and the monotonic direct
simple shear test (DSS) on loose Syncrude sand
are given in Table 1. The results are presented in
Figures 4 and 5. The results of the UBC3D-PLM
are in a good agreement with the experimental and
numerical results of the original UBCSAND as
published by Puebla et al. (1997).
One of the limitations of the current implemen-
tation is that the effect of the rotation of the princi-
pal stresses and its influence in the stiffness cannot
be modelled. Thus, the user has to enter a differ-
ent input parameter for the plastic shear modulus Figure 5. Numerical modeling of loose Syncrude sand
number kGp in order to model properly the triaxial under undrained simple shearing.

68
and the direct simple shear loading path. This issue
has to be taken into account during the calibration
of the model according to the most critical stress
path expected in a specific case.
3.2 Validation of the UBC3D-PLM in cyclic
loading
The behaviour of loose Fraser sand under cyclic
direct simple shear is modelled and the numerical
results are compared with experimental data as
published by Sriskandakumar (2004). The relative
density RD of the tested sand is 40%. Three
different shear stress ratios are used (CSR = 0.08,
0.1, 0.12) with the same set of parameters. The ver-
tical applied stress is 100 kPa in all cases. The K0
factor in the numerical calculations is assumed to
be 0.46 computed with the well known Jaky for-
mula. Therefore the initial stresses after consolida-
tion in the two horizontal directions equal 46 kPa.
In Figures 6, 7 and 8 the evolution of excess
pore pressure in three stress controlled DSS tests
are presented. The input parameters are presented
Figure 6. Evolution of excess pore pressures during
simple shearing on Fraser sand (RD = 40%). CSR = 0.08.
συ = 100 kPa.
Figure 7. Evolution of excess pore pressures during
simple shearing on Fraser sand (RD = 40%). CSR = 0.1.
συ = 100 kPa.
69
Figure 8. Evolution of excess pore pressures during
simple shearing on Fraser sand (RD = 40%). CSR = 0.12.
συ = 100 kPa.
in Table 1. Even though modelling the onset of liq-
uefaction in the framework of classical plasticity
is very complicated, the UBC3D-PLM constitutive
model can reproduce the evolution of excess pore
pressures towards cyclic liquefaction with adequate
accuracy for three different shear stress ratios using
the same set of parameters. The updated formula-
tion of the densification rule is helping the model
not to predict very steep evolution of the excess
pore pressures in the case of the anisotropic con-
solidated soil which was reported by Petalas et al.
(2012) for previous formulations.
One of the main limitations of the model is pre-
sented in the case of the higher shear stress ratio
(CSR = 0.12). The formulation of the UBC3D-
PLM cannot take into account the anisotropic
consolidation effects during the primary loading
which causes higherevolution of the excess pore
pressures during the first full cycle. Due to this
issue the UBC3D-PLM predicts a slower evolution
for this CSR. The slope of the curve in the experi-
mental results is much steeper during the first two
half cycles as can be seen in Figure 8. The ability of
the UBC3D-PLM of reproducing with high accu-
racy the cyclic stress paths which are started from
the isotropic line was presented by Petalas et al.
(2012). The influence of induced anisotropy dur-
ing the process of anisotropic consolidation is an
aspect under research and a new formulation for
the framework of classical plasticity is needed.
In Figure 9 the influence of the post liquefac-
tion formulation can be seen for the case of the
lower shear stress ratio (CSR 0.08). The predicted
total shear strains are in a good agreement with
the experimental results and prove the liquefied
state of the soil. The same good agreement in the
predicted relationship between the shear stress
and the shear strain is observed for the other two
shear stress ratios as well, still with the same set of
parameters.
Figure 9. Evolution of shear strains during undrained
simple shearing. Fraser sand (RD = 40%). CSR = 0.1.
συ = 100 kPa.
4 VALIDATION OF THE UBC3D-PLM
IN A FINITE ELEMENT SCHEME
A dynamic centrifuge test is modelled with PLAXIS
2D dynamics in order to validate the UBC3D-
PLM in a boundary value problem. The numerical
results are compared with the experimental results
published by Byrne et al. The input parameters are
presented in Table 1.
The depth of the sand in the model on proto-
type scale is 38.1 m. The width does not have any
influence in numerical modelling and a 1D soil
response is predicted. The total time of the input
acceleration is 33 seconds. 50 loading cycles are
modelled with constant amplitude of acceleration
equal to 1.96 m/s2.
The results of the predicted evolution of excess
pore pressure are presented in Figure 10. At 13.1
meters depth the numerical predictions is in a close
agreement with the experimental result. The model
can predict the onset of liquefaction with adequate
accuracy for this depth. However, in the two higher
depths, i.e., at 24.8 and 30.8 meters, the model
shows inaccuracy during the first loading cycles. It
predicts a much steeper evolution of excess pore
pressure which lead earlier to the liquefied state
compared with the experiment. This issue gets
more crucial in the deeper layer.
The influence of the higher relative density in the
deeper layer is critical in that case. The UBC3D-
PLM cannot take into account the specific feature
and the layer is treated as homogeneous. Moreover,
the densification of the soil in the two cases is notin
close agreement with the experimental results and
reveals the complexity of the soil behaviour under
undrained cyclic loading. It is concluded from this
research that the absence of a stress densification
feature in the model leads to a rapid evolution of
the pore pressures in the first loading cycles. The
effective confining pressure does not influence the
behaviour of the current model.
70
Figure 10. Comparison of the predicted evolution of
excess pore pressure by the UBC3D with the experimen-
tal results published by Byrne et al. (2004).
5 CONCLUSIONS
The formulation of the UBC3D-PLM as well as
the new features implemented by the authors is
presented in this paper. The model is simplified
but powerful in modelling the onset of liquefac-
tion in sands. The model can predict with high
accuracy the monotonic behaviour of sands and
the comparison with real element test presented
by the authors. One of the main limitations is the
inability of the model to reproduced the behaviour
of the sand under triaxial and direct simple shear
loading conditions with one set of parameters. The
difference in the predicted stiffness due to the rota-
tion of the principal stresses in the direct simple
shear test has to be taken into account when the
input parameters are derived.
It is concluded that after the improvements pre-
sented in this paper the model can predict with
adequate accuracy the onset of liquefaction in var-
ious stress paths as proven when compared with
real element tests. The main limitations arise dur-
ing cycling loading after anisotropic consolidation
of the soil element. The model can not reproduce
the influence of the induced anisotropy after the
consolidation stage which causes steeper evolution
of the excess pore pressures during the first load-
ing cycle.
Finally, when the model is used in a boundary
value problem the aspect of stress densification
plays a key role when modelling the onset of lique-
faction in the deep soil layers. The effect of stress
densification as well as the accuracy of the den-
sification rule should be further investigated for a
future reformulation.
REFERENCES
Beaty, M. & Byrne, P. 1998. An effective stress model
for predicting liquefaction behaviour of sand. Geo-
technical Earthquake Engineering and Soil Dynamics
III ASCE Geotechnical Special Publication No.75. 1,
766–777.
Beaty, M. & Byrne, P. 2011. Ubcsand constitutive model
version 904ar. Itasca UDM Web Site, 69.
Byrne, P.M., Park, S.S., Beaty, M., Sharp, M.,
Gonzales, L. & Abdoun, T. 2004. Numerical model-
ling of dynamic centrifuge tests. 13th World Confer-
ence in Earthquake Engineering.
Martin, G., Finn, W. & Seed, H. 1975. Fundamentals of
liquefaction under cyclic loading. Journal of the Geo-
technical Engineering Division, ASCE 101.
Petalas, A., Galavi, V. & Bringkreve, R. 2012. Validation
and verification of a practical constitutive model for
71
predicting liquefaction in sands. Proceedings of the
22nd european young geotechnical engineers conference,
Gothenburg, Sweden., 167–172.
Puebla, H., Byrne, M. & Phillips, P. 1997. Analysis of
can-lex liquefaction embankments prototype and
centrifuge models. Canadian Geotechnical Journal 34,
641–657.
Rowe, P.W. 1962. The stress-dilatancy relation for static
equilibrium of an assembly of particles in contact.
Proceedings of the Royal Society of London. Series A,
Mathematical and Physical Sciences 269A, 500–527.
Sriskandakumar, S. 2004. Cyclic loading response of
fraser sand for validation of numerical models simu-
lating centrifuge tests. Master’s thesis, The University
of British Columbia, Department of Civil Engineering.
Tsegaye, A. 2010. Plaxis liqueafaction model. report
no. 1. PLAXIS knowledge base.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Comparison of two anisotropic creep models at element level

N. Sivasithamparam
Plaxis BV, Delft, The Netherlands
University of Strathclyde, Glasgow, Scotland, UK

M. Karstunen
Chalmers University of Technology, Gothenburg, Sweden
University of Strathclyde, Glasgow, Scotland, UK

R.B.J. Brinkgreve
Plaxis BV, Delft, The Netherlands
Delft University of Technology, Delft, The Netherlands

P.G. Bonnier
Plaxis BV, Delft, The Netherlands

ABSTRACT: This paper presents a comparison of two anisotropic creep models, ACM and Creep-
SCLAY1, which differ in their formulation of creep strain rate. Creep is formulated in ACM using
the concept of contours of constant volumetric creep strain rate, whereas the newly developed Creep-
SCLAY1 model uses the concept of a constant rate of visco-plastic multiplier. The two models are identical
in the way the initial anisotropy and the evolution of anisotropy are simulated. A key assumption of both
models is that there is no purely elastic domain. The models are compared at element level. The numerical
simulations show that the Creep-SCLAY1 model is able to give a better representation of natural clay
behaviour at element level.

1 INTRODUCTION enhance the confidence of practicing geotechnical

Natural soils behave in a highly anisotropic man-
ner due to the deposition process and subsequent
loading history, and show time-dependent (creep)
behaviour. An accurate description of anisotropy
and rate-dependent behaviour of soft soils is nec-
essary for safe and economic design of structures
on soft soils deposits. To obtain realistic solutions
for geostructures on natural clays, it is essential
to use a constitutive model that accounts for ani-
sotropy and time dependency. Many constitutive
models for time-dependency and anisotropy have
been proposed in the literature. Time-dependent
models that represent only inherent anisotropy
have been proposed (e.g Sekiguchi & Ohta (1977)
and Zhou et al. (2006)) as well as time-dependent
models accounting for both inherent and plastic
strain induced anisotropy (e.g. Leoni et al. (2008)
and Karstunen & Yin (2010)). The constitutive
models should be relatively simple and easy to
understand. Ideally, it should be possible to deter-
mine the values of the model parameters from
standard laboratory tests. This would namely
engineers for adopting the models for numerical
analysis.
The ACM (Leoni et al. 2008) is an extension of
the Soft Soil Creep model (Vermeer et al. 1998)
with rotated ellipses (similar to the S-CLAY1S
model (Karstunen et al. 2005)) used as contours of
volumetric creep strain rates. The formulation for
the Creep-SCLAY1 model was proposed recently
by Siv-asithamparam (2012). Anisotropy in both
models is described by introducing a fabric tensor
to represent the rotation of the constitutive ellipses
in the p′ – q plane, similar to the S-CLAY1S model
(Karstunen et al. 2005). Moreover, a rotational
hardening law describes the evolution of anisot-
ropy due to volumetric and deviatoric creep strain
rates. However, the Creep-SCLAY1 model differs
considerably from the ACM in the formulation of
creep strain rates. Creep is formulated in Creep-
SCLAY1 using the concept of rate of the visco-
plastic multiplier (Grimstad et al. 2010). Unlike
Grimstad et al. (2010) model that used Janbu’s time-
resistance concept, the present model uses the more
familiar creep coefficient, modified creep index μ*

72
which can be easily derived from standard labora-
tory tests. This paper shows a direct comparison
of both models and their prediction capability at
element level.
The first part of this paper gives a short descrip-
tion of the ACM and Creep-SCLAY1 models in
triaxial stress space. In further sections the single
element simulations results obtained using the
SoilTest facility in the finite element code PLAXIS
(Brinkgreve et al. 2012) are presented, followed by
brief conclusions.

2 ANISOTROPIC CREEP MODELS

The elastic and creep parts in both models are

combined with an additive law expressing the total
strain rate as a combination of elastic and creep
component as in classical elasto-plasticity.

c
∈= ∈e + ∈ (1)

where ∈ is strain, a dot over a symbol implies rate

(differentiation with respect to time) and super-
scripts e and c refer to the elastic and creep compo-
nents respectively.
For the sake of simplicity, the mathematical
formulation of both models is presented in triaxial
stress space, which can be used only to model the
testing of cross-anisotropic samples (cut vertically
from the soil deposit) in oedometer or triaxial
apparatus in the laboratory.

2.1 ACM
Leoni et al. (2008) proposed the Anisotropic Creep
Model (ACM) extending from a previously devel-
oped isotropic creep model (Vermeer et al. 1998;
Vermeer & Neher 1999) which is based on ellipses
of Modified Cam Clay (Roscoe & Burland 1968).
An extract of the mathematical formulation from
Leoni et al. (2008) is presented below. The outer
rotated ellipse defines the normal consolidation
surface (NCS) and the size of this ellipse evolves
with volumetric creep strains according to the Figure 1. Current state surface and normal consolida-
hardening law tion surface in triaxial stress space.

⎛ ∈c ⎞
p′p = p′p 0 exp ⎜ * v * ⎟ (2) (CSS). The intersection of the CSS with the hori-
⎝ λ −κ ⎠
zontal axis is called the equivalent mean stress p′eq,
and it is defined as
where λ* and κ* are the modified compression
index and modified swelling index respectively. The
( q − p ′ )2
intersection of the vertical tangent to the ellipse ′ = p′ +
peq (3)
with p′ axis is the isotropic preconsolidation pres- ( M 2 − 2 ) p′
sure p′p (see Figure 2.1).
The inner ellipse passes through the current state where M is the stress ratio at critical state and a
of effective stress called the current stress surface scalar quantity α is used to describe the orientation

73
of the normal consolidation surface and current
stress surface.
The volumetric creep strain rate is given by a
power law as follows:

β
μ * ⎛ peq
′ ⎞
 vc =
∈ ⎜ ⎟ (4)
τ ⎝ p′p ⎠

μ* is referred to as the modified creep index, τ is

called the reference time and is set to 1 day if
the NCS is found by performing a standard 24 h
oedometer test, and β is defined as:

λ* κ *
β= (5)
μ*

The ACM cannot predict swelling on the ‘dry’

side of critical state line as it does not allow the
stress state to cross the failure line represented by
Mohr-Coulomb criterion i.e. allowing d ∈vc ≤ 0.
Because of this, the ACM is limited to the ‘wet’
side of the critical state line only (see Figure 2.1).
In addition, the ACM cannot give a satisfactory
response for strain rate changes in undrained tests
of normally consolidated clays (Grimstad 2009) as
discussed later. For further details of the anisot-
ropy and creep formulation, the interested reader
is referred to Leoni et al. (2008), Wheeler et al.
(2003) and Karstunen et al. (2005).

2.2 Creep-SCLAY1
In Creep-SCLAY1, Eq. (4) is modified to an
expression that gives the rate of the visco-plastic
multiplier as follows:

β
′ ⎞ ⎛ M − α K 0NC ⎞
2 2
 μ * ⎛ peq
Λ= ⎜ ⎟ ⎜ 2 ⎟ (6)
τ ⎝ p′p ⎠ ⎜⎝ M − ηK2 NC ⎟⎠
0

where η = q/p′ and the additional term

Figure 2. Normal Consolidation Surface (NCS) in gen-
( 2 2
K0
(M 2 − ηK2 NC ) is added to ensure that
) /(
/(M eral stress space.
0
under oedometer conditions, the resulting creep
strain corresponds to Eq. (4), where subscript K 0NC
refers to the normally consolidated K0 stress path. of the mathematical formulation of the model, the
Grimstad (2009) suggested that creep expressed reader is referred to Siv-asithamparam (2012).
directly on the rate of plastic multiplier gives the
“proper” response.
3 MODEL PARAMETERS
The current stress surface (CSS) and normal
consolidation surface (NCS) are defined similar
Both models require the same parameters described
to ACM. However, Creep-SCLAY1 predicts swell-
below.
ing on the ‘dry’ side of the critical state line, unlike
ACM (see Figure 2.1). Figure 2 compares the nor- Parameters which are similar to the Modified
mal consolidation surface of ACM and Creep- Cam-clay model include soils constants v′
SCLAY1S in general stress space. For further details (Poisson’s ratio), M (stress ratio at critical state),

74
 λ* (modified compression index) and κ* (modified
swelling index). Furthermore, the initial value for
a state variable p′p (initial size of the ellipse) is
required. In the context of finite element analy-
ses, the initial value of p′p is calculated based on
the OCR (vertical overconsolidation ratio) or
POP (pre-overburden pressure), normally con-
solidated K 0NC value (estimated by Jaky’s formula)
and the initial vertical effective stress.
Parameters describing initial anisotropy (α0)
and its evolution, include soil constants ω (rate
of rotation of the surfaces) and ωd (relative rate
of surface rotation). The scalar value α0 and ωd
can be theoretically derived based on M values
(see Wheeler et al. (2003) for details) as follows:
η02 + 3η0 − M 2
α0 = (7)
3 rained strength predicted by ACM is lower than
3 4M 2 − 4η02 3η0
ωd = (8)
8 η02 M + 2η0
where η0 = 3(
3 1−
1 NC
0 ) /(1 2 K 0NC ) .
The parameter ω can be estimated based on
initial anisotropy (α0), modified compression
index (λ*), M and ωd (see Leoni et al. (2008) for
details) as follows:
1 10M 2 − 2α 0ω d
ω= ln (9)
λ* M 2 − 2α 0ω d
In derivation of Eq. (9), a number of assump-
tions has been made (see Leoni et al. (2008)).
Consequently, with certain parameter combina-
tions Eq. (9) might result with a negative value,
which makes no physical sense. As an alternative,
an empirical formula suggested by Zentar et al.
(2002) to estimate the ω value can be used:
10 20
≤ω ≤ (10)
λ λ
μ* (modified creep index) can be obtained by
measuring the volumetric strain on the long
term and plotting it against the logarithmic
time. τ (the reference time, which is linked to the
definition of vertical preconsolidation stress)
can usually be taken to equal one day (see
Brinkgreve et al. (2012) for details).
4 NUMERICAL SIMULATION
This section discusses the performance of both
models in a single element simulation. Both models
are implemented into the finite element code
75
PLAXIS as user-defined soil models. The Creep-
SCLAY1 model has been implemented by the
first author and the ACM has been implemented
by Leoni et al. (2008). Single element simulations
were done using the PLAXIS SoilTest facility to
highlight the similarities and the differences in the
model predictions. Parameters used for these simu-
lations corresponding to Both-kennar clay parame-
ters (Symposium 1992) are summarized in Table 1.
Firstly, Creep-SCLAY1 and ACM were com-
pared in undrained compression simulations with
two strain rates (10% per day and 100% per day).
Initial effective stress σ 3′ = 100 kPa and K0 = 0.5
were assumed and 10% maximum strain was
applied. Figures 4 and 4 show the stress paths and
deviatoric stress versus axial strain predicted by
the two models. Though both models are able to
predict dependence on strain rate, the peak und-
that predicted by the Creep-SCLAY1 model. In
contrast to Creep-SCLAY1, ACM predicts stress
path approaches the CSL with reducing p′ and q,
converging towards the stress origin due to the
assumption of constant volumetric creep strains.
In the ACM simulations, jumbs were observed
as highlighted in Figure 3 due to the transition
between current state surface to Mohr-Coulomb
failure surface. Furthermore, the ACM cannot
reach to a critical state condition with shearing at
constant volume and effective stresses.
Several publications (e.g. Graham et al. (1983),
Tatsuoka et al. (2002), Leroueil & Marques (1996),
Tavenas et al. (1978) and Vaid & Campanella
(1977)) showed the influence of step changes in
strain rate on the stress-strain behaviour of soft soil
in undrained triaxial compression. Immediately
after an increase in strain rate the stress-strain path
is seen to jump upwards and show an initial stiff
response. If the strain rate is reduced back to the
original strain rate then a downwards stress jump
is observed after which the path rejoins the original
curve defined by the lower strain rate. The paths
in stress-strain curves are indicated to be uniquely
Table 1. Bothkennar clay parameters.
Parameters value
λ* 0.1
v′ 0.2
κ* 0.00667
M 1.5
OCR 1.5
α0 (initial anisotropy coefficient) 0.59
ω (anisotropy coefficient) 50.0
ωd (anisotropy coefficient) 1.0
μ* (viscosity coefficient) 5.07 × 10−3
τ (viscosity coefficient) 1.0 day
Figure 3. Simulation of undrained triaxial compression with varying strain rate.

Figure 4. Simulation of undrained triaxial compression with varying strain rate.

76
defined by the strain rate and the effects of strain
rate changes are observed to be persistent, which
is a characteristic of isotach behaviour, i.e., there
is a unique stress-strain strain-rate relation for
a given soil. Most soft clays in both undisturbed
and reconstituted states, undisturbed natural stiff
clays and cases of soft rock all show isotach vis-
cous behaviour. Figure 4 shows a stepwise change
in strain rate undrained compression simulations
using Creep-SCLAY1 and ACM to verify the
capability of both models to predict the isotach
behaviour. Figure 4 clearly demonstrates that
ACM cannot properly simulate the isotach behav-
iour observed in natural soft clays under a stepwise
change in strain rate. Furthermore, the stress path
simulated by ACM cannot overpass the critical
state as shown in Figure 4. This too is not in agree-
ment with experimental observations for slightly
structured or reconstituted clays (Yin et al. 2010).
There is a mathematical difference between the
two models to calculate the creep strain compo-
nents in general stress space. In ACM and Creep-
SCLAY1, creep strains are calculated as follows:
ACM:
∈ vc ∂peq
′ not be in agreement with experimental observa-
 cij =
∈ ∂peq′
∂σ iij
∂p ′
(11)
Creep-SCLAY1:
 vc
∈ ∂peq
′ Further work will involve comparing the per-
 cij =
∈
( )
(12)
∂peq
′ ∂σ iij
∂p ′
K 0NC
The value of ∂p′eq/∂p′ in the ACM reaches to
infinity when η/M becomes to 1, i.e., the stress
condition reaches to critical state (see Figure 5).
This causes numerical problems.
Figure 5. dp′eq /dp′ versus η/M plot.
77
5 CONCLUSIONS
This paper studies the performance of two aniso-
tropic creep constitutive models at element level.
In ACM, the creep strain rate is formulated using
contours of volumetric creep strain rates whereas in
the newly developed Creep-SCLAY1 model, creep
strain rate is formulated using the concept of rate
of visco-plastic multiplier. The model simulations
demonstrate that the new formulation results in
a better prediction of natural soft soil behaviour.
The following observations are made from the
comparison:
Though both models are able to predict rate
effect dependence in undrained compression simu-
lation, in contrast to the Creep-SCLAY1 model,
the ACM predicts stress paths which approach the
CSL with reducing p′ and q, converging towards
the stress origin. ACM cannot reach a critical state
condition with shearing at constant volume and
effective stresses. Undrained compression using
stepwise change in strain rate simulations demon-
strate that the ACM cannot reproduce the isotach
behaviour observed in natural soft soils. Further-
more, ACM cannot overpass the CSL; this may
tions for slightly structured or reconstituted clays.
There is a mathematical difficulty in ACM. When
calculating creep strain rates, the value of ∂p′eq/∂p′
can reach infinity when η/M becomes to 1, i.e., the
stress condition reaches a critical state.
formance of the models against experimental data
and instrumented test structures.
ACKNOWLEDGMENTS
The research was sponsored by FP7 programmes
GEO-INSTALL (PIAP-GA-2009-230638) and
CREEP (PIAP-GA-2011-286397).
REFERENCES
Brinkgreve, R., Engin, E. & Swolfs, W.M. 2012. PLAXIS
Finite Element Code for Soil and Rock Analyses.
The Netherlands: 2D-Version 2011.
Graham, J., Crooks, J. & Bell, A. 1983. Time effects
on the stress-strain behaviour of natural soft clays.
Geotechnique 33(3), 327–340.
Grimstad, G. 2009. Development of effective stress
based anisotropic models for soft clays. Ph. D. thesis,
Norwegian University of Science and Technology
(NTNU), Norway.
Grimstad, G., Abate, S., Nordal, S. & Karstunen, M.
2010. Modeling creep and rate effects in structured
anisotropic soft clays. Acta Geotechnica 5, 69–81.
Karstunen, M., Krenn, H., Wheeler, S., Koskinen, M. &
Zentar, R. 2005. The effect of anisotropy and
 destructuration on the behaviour of Murro test
embankment. Int. J. of Geomechanics (ASCE) 5(2),
87–97.
Karstunen, M. & Yin, Z.Y. 2010. Modelling time-de-
pendent behaviour of Murro test embankment. Geo-
technique 29, 1–34.
Leoni, M., Karstunen, M. & Vermeer, P. 2008. Aniso-
tropic creep model for soft soils. Gotechnique 58 (3),
215–226.
Leroueil, S. & Marques, M. 1996. State of art: Impor-
tance of strain rate and temperature effects in geo-
technical engineering. measuring and modelling time
dependent behaviour of soils. ASCE, Geotechnical
Special Publication 61, 1–60.
Roscoe, K. & Burland, J. 1968. On the generalised stress-
strain behaviour of wet clay. Engineering Plasticirv,
535–609.
Sekiguchi, H. & Ohta, H. 1977. Induced anisotropy and
time dependency in clays. 9th ICSMFE, Tokyo, Con-
stitutive equations of Soils 17, 229–238.
Sivasithamparam, N. 2012. Modelling creep behaviour
of soft soils. Internal report Plaxis B.V. Symposium
1992. Bothkennar soft clay test site: Characteriza-
tion and lessons learned (Géotechnique symposium in
print). Géotechnique 42(2), 161–380.
Tatsuoka, F., Ishihara, M., Di Benedetto, H. & Kuwano,
R. 2002. Time-dependent shear deformation charac-
teristics of geomaterials and their simulation. Soils &
Foundations 42(2), 103–138.
78
Tavenas, F., Leroueil, S., La Rochelle, P. & Roy, M. 1978.
Creep behaviour of an undisturbed lightly overcon-
soli-dated clay. Can. Geot. J. 15(3), 402–423.
Vaid, Y. & Campanella, R. 1977. Time-dependent behav-
iour of undisturbed clay. ASCE J Geotech Eng Div
103(7), 693–709.
Vermeer, P.A. & Neher, H. 1999. A soft soil model that
accounts for creep. Beyond 2000 in Computational Geo-
technics, R.B.J. Brinkgreve(eds), Rotter-dam. 4, 249–261.
Vermeer, P.A., Stolle, D.F.E. & Bonnier, P.G. 1998. From
classical theory of secondary compression to modern
creep analysis. Proc. 9th Int. Conf. Comp. Meth. and
Adv. Geomech., Yuan(eds) 4, 2469–2478.
Wheeler, S., Näätänen, A., Karstunen, M. & Lojander,
M. ¨ 2003. An anisotropic elasto-plastic model for soft
clays. Can. Geot. J. 40, 403–418.
Yin, Z.Y., Chang, C.S., Karstunen, M. & Hicher, P.Y.
2010. An anisotropic elastic-viscoplastic model for
soft clays. Int. J. of Solids and Structures 47, 665–677.
Zentar, R., Karstunen, M., Wiltafafsky, C., Schweiger, H.F. &
Koskinen, M. 2002. Comparison of two approaches for
modelling anisotropy of soft clays. Proc. 8th Int. Symp.
on Numerical Models in Geomech. (NU-MOG VIII),
115–121.
Zhou, C., Yin, J.-H., Zhu, J.-G. & Cheng, C.-M. 2006.
Elastic anisotropic viscoplastic modeling of the strain-
rate dependent stress–strain behaviour of K0-consoli-
dated natural marine clays in triaxial shear test. Int. J.
Geomech 5(3), 218–232.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Displacement pile installation effects in sand

A. Beijer Lundberg, J. Dijkstra & A.F. van Tol

Delft University of Technology, Delft, The Netherlands

ABSTRACT: Installation of jacked or driven displacement piles imposes large deformations in the soil.
These installation effects will influence the subsequent load-deformation response of the installed piles
and should be taken into account in the description of the soil-structure interaction. A series of model
pile tests were carried out in the geotechnical centrifuge at TU Delft. The tests combined horizontal con-
tact stress measurements on an instrumented model pile and visual observation of the soil deformation
adjacent to the pile. The model scaling, experimental set-up and results of pile installation in samples with
various initial relative densities are discussed. The results indicated good consistency and reproduced the
friction fatigue effect, the lateral loads at rest acting on the pile show surprisingly small influences from
the pile installation.

1 INTRODUCTION White & Lehane, 2004). This phenomenon has

been called friction fatigue or fluffing (White &
1.1 Pile installation Lehane 2004, Kraft, 1991). Several different mod-
els for the rate of shaft friction reduction have been
Pre-fabricated driven and jacked piles are com-
proposed (Randolph et al, 1994).
monly installed by forcing the pile into the ground
Field tests on a pile which is instrumented at
with a pile hammer or a hydraulic jack. In the
several levels suggests that the reduction in normal
installation phase the pile displaces the soil which
stress was proportional to D/L, where D is the pile
causes large soil deformations and high stress levels
diameter and L is the distance from the pile tip
in the ground. These installation effects influence
(Lehane et al, 1993). A possible explanation for the
the subsequent load-deformation response of the
friction fatigue is the effect of cyclic interface load-
piles and can make bearing capacity prediction of
ing between the pile and the soil (White & Lehane,
displacement piles relatively precarious (Klotz &
2004). The gradual reduction of shaft friction
Coop, 2001).
has been introduced in conceptual models of pile
The pile installation results in a pre-stressed soil
installation effects. Cyclic interface loading during
mass around the pile base which results in a stiffer
every installation cycle results in step-wise shearing
pile base response compared to bored piles (Li et al,
and an initial stress build-up at the interface. With
2012). The pile base penetration mechanism is sim-
on-going load cycles the soil near the interface will
ilar to that of a cone penetration test and the ulti-
contract, which subsequently reduces the contact
mate base capacity has shown to be proportional
stress level. This mechanism is reasonably approxi-
to the cone resistance (Lee & Salgado, 1999). An
mated in constant normal stiffness shearbox tests
ultimate base capacity of around 60% of the cone
(DeJong et al., 2006). The effect is most noticeable
resistance is used in most design codes (e.g. Lehane
for sand with high cyclic compressibility, such as
et al, 2005), however the employed normalization
calcareous sands (White, 2005).
and averaging techniques of the measured cone
Most field and model tests have been carried out
resistance complicate direct comparisons between
in variable relative densities or at different stress
the different design codes (Eslami & Fellenius
levels (Lehane et al, 1993, White & Lehane, 2004,
1997).
Klotz & Coop, 2001). The governing mechanisms
The governing mechanisms of the pile shaft
of shaft friction and friction fatigue have therefore
friction are not directly related to the cone resist-
not been systematically studied for soils of different
ance which makes direct correlation between these
relative densities and stress levels in a large number
quantities more cumbersome. Furthermore, both
of tests. Also, the combination of the measurement
field tests and model tests show a reduction of
of contact loads on the pile and deformations in
normal stress and shaft friction between the soil-
the soil has not yet been attempted. These mecha-
pile interface at any given depth as the pile tip
nisms may change with the soil state (Axelsson,
moves further into the ground (Lehane et al, 1993,
1998).

79
 A series of geotechnical centrifuge tests in sand
samples with different initial relative densities have
therefore been performed to bring some further
clarification of this phenomenon.
Stress measurements on the pile-soil contact
and soil deformation measurements adjacent to
the pile have been made. These tests have been per-
formed with a small instrumented model pile. The
influence of the scaling conditions on the effect
of friction fatigue and the difference between the
model and prototype scale is also discussed.
1.2 Scaling of model tests
Previous studies have mainly been concerned with
the influence of friction fatigue for stress meas-
urement of an unloaded pile (White & Lehane,
2004, Lehane et al, 1993). At full shaft friction
mobilization an additional normal stress compo-
nent by dilation at the pile surface will be present
(Boulon & Foray, 1986). This has been called con-
strained dilatancy because of the high level of kin-
ematic constraint around the pile (Axelsson, 1998,
Houlsby, 1991). The additional normal stress Δσ n′
is often approximated, using elastic cavity expan-
sion theory, as:
4G Δh
Δσ n′ = (1)
D
where G is the shear modulus, Δh is the width of
the dilating interface and D is the pile diameter.
The size of the dilating interface depends on the
grain size and is assumed to be relatively constant.
Hence, the latter will not scale in the centrifuge. An
interface width of 0.02 mm has been proposed in
design procedures and has been seen to correlate
well with field measurements (Lehane et al, 2005,
Axelsson, 1998).
Smaller model piles will have a significantly
higher additional normal stress at full mobilization.
Figure 1 shows the additional normal stress for
different width of the dilating interface for a 10 mm
diameter pile with 50 MPa shear modulus.
The much higher interface stress during the
shearing load cycles may result in a different
magnitude of the friction fatigue effect in the
model tests. The length of the frictional interface
is smaller in the centrifuge tests. Ring shear tests
show very large mobilization length is needed to
reach a constant interface friction angle (Klotz &
Coop, 2001, Jardine et al, 2012). The ultimate shaft
capacity will therefore be much larger in the model
test compared to the prototype pile, which has
been observed in centrifuge tests (White & Lehane,
2004). Also, the pile diameter will slightly increase
during pile installation because of the Poisson’s
ratio effect. However, for small stiff piles which are
80
Figure 1. Additional normal stress relative to the size of
the dilative interface according to equation 1.
statically loaded this is negligible. However, even
in model tests this could be significant if dynamic
pile installation and open ended piles are used
(Di Nicola & Randolph, 1993).
Nevertheless, the centrifuge offers necessary
modelling conditions at continuum scale. At
the moment only measurements when the pile is
unloaded and at rest will be discussed here.
2 TESTING PROCEDURE
The pile installation was modelled in the TU Delft
geotechnical centrifuge (Allersma, 1994). The tests
were carried out at 50 g. A 380 mm × 150 mm ×
144 mm strongbox was used for the test. The sam-
ple container was equipped with two Plexiglas
windows where visual observations were made.
The Plexiglas edges were marked at regular inter-
val for correction of lens distortion. The test pro-
cedure consisted of stress measurements and visual
measurements. These were carried out pairwise in
the same sample. The stress measurements were
performed in the center of the sample container
with a 10 mm diameter instrumented model pile
connected to a data acquisition system. The soil
deformation measurements were performed in
half-space at one side of the Plexiglas borders with
a 5 × 10 mm non-instrumented pile. The experi-
mental set-up is shown in Figure 2.
A machine vision camera (VC 4012Nano) was
mounted on the centrifuge for the visual meas-
urements. The sample container was illuminated
with portable LEDs attached to the centrifuge
swing.
Stress measurements were conducted on an
instrumented model pile. The model pile had
two levels of normal contact stress sensors situ-
ated at 2.5 D and 7.5 D distance from the pile tip.
Each consisted of two separate sets of full-bridge
strain gauges embedded inside the pile in sensing
 3 TEST SERIES

In all test the pile was installed approximately

Figure 2. Experimental set-up showing the sample con-
tainer in the centrifuge swing.
100 mm into the soil. Two different installation
methods were used, continuous installation and
cyclic installation. Continuous installation con-
sisted of jacked installation at a continuous rate of
approximately 0.25 mm/s. The cyclic installation
consisted of a downward stroke of 10 mm, which
was followed by an upward stroke which was ter-
minated at zero pile head load. Therefore, this
cyclic installation simulates the load reversal dur-
ing installation, but excludes the stress wave effects.
After the installation the pile was unloaded, and a
load test to 0.2 D (20% of the pile diameter) was
performed. Subsequently, a tension test was exe-
cuted, by completely retracting the pile out of the
soil with the same rate as during installation, whilst
measuring all the loads acting on the pile.
The tests were performed in dry air pluviated
sand. Drained conditions were assumed to be gen-
erally applicable for the prototype. Dry sand only
displays small rate effects of 10%/decade small
variations in the installation rate would therefore
not significantly influence the test results.
Sand samples were prepared at three different
initial densities. Uniform sub-rounded silica sand
with a mean grain size of 0.235 mm was used in
the tests. The sand samples were prepared by air
pluviation and additional vibration was used to
prepare the dense (80% relative density) sand sam-
ples. Due to the modest size of the strongbox, the
whole box was vibrated.
The test series are described in Table 1. The
instrumented model pile was installed in the cen-
tre of the sample, whilst the half width pile wasn’t
instrumented and installed near the transparent
window.

Table 1. Test series with initial relative density of the

Figure 3. Half part of the instrumented pile showing soil samples and the installation method.
sensing membranes equipped with strain gauges and
cables. Unit. rel. Cont. Cycl Stress Visual nr.
dens. inst inst. meas. meas. tests.

membrane in perpendicular positions. Dynamic 40% x x 3

strain gauge bridge excitation was used to reduce 40% x x 2
thermal disturbances. The sensors were mounted 40% x x 3
on two identical 5 × 10 steel bodies, shown in 40% x x 2
Figure 3, connected by Bison epoxy glow. The 60% x x 3
strain gauges were calibrated in a water-pressured 60% x x 3
calibration pressure vessel (Beijer Lundberg et al, 60% x x 3
2012). 60% x x 3
The model pile was also equipped with a base
80% x x 2
stress sensor consisting of strain gauge pairs ori- 80% x x 2
ented in the axial direction of the pile. The pile 80% x x 2
head load was measured with a load cell (Burster 80% x x 2
8431-6005), shown in Figure 2.

81
4 RESULTS

The results consisted of stress measurements and

measurements of soil deformations (from the
image data). Here pile head load measurements
of the continuous installation tests and measure-
ments of normal contact stress for the unloaded
pile during cyclic installation are shown in addition
to displacement path measurements of soil at dif-
ferent initial relative density.

4.1 Stress measurements

The pile head load was measured during the tests.
Figure 4 shows the pile head load measurements
for the continuous installation tests. Figure 5 shows
the pile base stress measurements. Figure 5. Pile base load (MPa) for continuous
The horizontal contact stress was measured installation.
after pile head unloading for the tests with
cyclic installation. Figure 6 shows the horizon-
tal normal stress for the lower stress center with
2.5D distance to the pile tip. The measurements
at both sides were combined to give an average
measurement.
The higher stress sensor at 7.5D distance from
the pile tip was used to measure the effect of
friction fatigue during the measurements in the
same test. Figure 7 shows the measurements in
loose sand. Figure 8 shows the measurements in
medium dense sand. Figure 9 shows the measure-
ments in dense sand. The average normal stress
is either normalized by the initial effective ver-
tical stress before installation (Fig. 10) or either
normalized by the average pile base stress qb
(Fig. 11). All Figures plot the model against the
load. Figure 6. Average horizontal normal stress (kPa) as
function of depth (mm) after pile head unloading dur-
ing cyclic installation with 10 mm increment for the lower
normal stress sensor.

Figure 7. Normal stress measurements for the higher

Figure 4. Pile head load for continuous installation. and lower stress sensor in loose sand.

82
Figure 8. Normal stress measurements for the higher
and lower stress sensor in medium dense sand.
Figure 9. Normal stress measurements for the higher
and lower stress sensor in dense sand.
Figure 10. Average normal stress normalized by the
initial effective vertical stress.
83
Figure 11. Average normal stress normalized by the pile
base stress qb.
Figure 12. Displacement paths around the pile for
medium dense sand after displacement increment extrac-
tion with JPIV.
4.2 Soil deformation measurements
Soil deformation measurements were performed
with the machine vision camera. These measure-
ments were transformed to displacement increment
with the program JPIV (Vennemann et al., 2007).
The displacement increments were collected to
form the full displacement paths. These are shown
for medium dense sand in Figure 12.
The displacement paths for loose, medium
dense and dense are compared in Figure 13. The
displacement paths start at 3.5 D depth and the
horizontal distance of 0 D, 0.5 D, 1 D. 1.5 D
and 2 D.
Figure 13. Displacement paths for loose, medium dense and dense initial density.
5 DISCUSSION & CONCLUSIONS
The results in Figures 4 and 5 show good similarity
with other centrifuge measurements (Klotz & Coop
2001, White & Lehane, 2004), hence supporting the
quality of the setup. The small amount of meas-
urements points for the horizontal measurements
in Figure 6 create somewhat non-continuous stress
distribution, but a clear trend with increasing depth
and dependence on higher initial density is shown.
A friction fatigue phenomenon is also observed in
the current tests (Figures 7–9). Interestingly, the
magnitude of the stress reduction scales well with
the initial stress as registered during the first pas-
sage of the pile, where the initial stress is density
dependent. Also, the normal stress measurements
in Figure 10 show a surprisingly good fit to the
preferred design values for K0 in Kraft, 1991, espe-
cially, when taking into account the complex load-
ing path experienced by the soil near the interface
during installation. Currently, it is unclear whether
this is a test specific result or more generally appli-
cable. A more detailed interpretation of the test
data is required to verify the general applicability.
Also, the unloaded horizontal stress have been
used for this analysis.
The normal stress measurements are normalized
by the base stress in Figure 1. This normalization,
which can be assumed to be proportional to the
cone resistance, shows that there is a clear connec-
tion between the normal stress and the base stress.
The displacement path measurements in Figure 13
show the larger horizontal displacement for dense
sand, which could influence the final normal stress
around the pile tip.
The measurements show a relatively continuous
change in governing mechanism between loose to
dense sand. This implies and that the mechanisms
84
of normal stress distribution along the shaft and
friction fatigue are similar at all densities. If the
different magnitudes of the material behavior can
be taken into account, the same conceptual model
would therefore be applicable for pile design at any
densities. The results also show that the cone resist-
ance is a relatively good tool for normalization of
the measurements, at least when pile installation in
silica sand is modeled.
REFERENCES
Allersma, H.G.B., 1994, The university of Delft Geo-
technical centrifuge, Proc. Int. Conf 1994.
Axelsson, Gary, 1998, Long-term increase in shaft capac-
ity of driven piles in sand, Proc. 4th Int. Conf. on Case
Histories in Geotech. Engng., St. Louis, Missouri.
Beijer Lundberg, A., J. Djikstra, and K. van Beek, 2012,
Measurements of soil contact stress in a harsh envi-
ronment, Instrumentation and Measurement Technol-
ogy Conference (I2MTC), 2012 IEEE International.
IEEE, 2012.
Boulon, M. & Foray, P., 1986, Physical and numeri-
cal simulation of lateral shaft friction along offshore
piles in sand, In numerical methods in offshore pileing,
3rd international conference, Nantes, 1986.
DeJong, J.T., White, D.J. and Randolph, M.F. (2006).
Microscale observation and modeling of soil-structure
interface behavior using particle image velocimetry,
Soils and Foundations, 46(1): 15–28.
De Nicola, A. & Randolph, M.F., 1993, Tensile and com-
pressive shaft capacity of piles in sand, Journal of geo-
technical engineering 119(12), 1952–1973.
Eslami, A. and Fellenius, B.H., 1997, Pile capacity by direct
CPT and CPTu methods applied to 102 case histories.
Canadian Geotechnical Journal. 34(6): 886–904.
Houlsby, G.T., 1991, How the dilatancy of soils affects
their behavior, Proceedings of the 10th European
Conference on Soil Mechanics and Foundation
Engineering.
Huy, N.Q., Dijkstra, J, and Tol, A.F. van, 2005, Influence
of loading rate on the bearing capacity of piles in
sand. Proceedings of the 16th International Confer-
ence on Soil Mechanics and Geotechnical Engineering,
pp. 2125–2128.
Jardine, R.J., Zhu, B.T., Foray, P. & Yang,. Z.X., 2012,
Measurements of stresses around closed ended dis-
placement piles in sand, Geotechnique, 56(9), 1–17.
Klotz, E.U. & Coop, M.R., 2001, An investigation of the
effect of soil state on the capacity of driven piles in
sands, Geotechnique, 51(9), 733–751.
Kraft, L.M., 1991, Performance of axially loaded pipe
piles in sand, J. of Geotech. Engrg, 117(2), 272–296.
Lee, J.H., and R. Salgado, 1999, Determination of pile
base resistance in sands, Journal of Geotechnical and
Geoenvironmental Engineering, 125(8), 673–683.
Lehane, B.M., Jardine, R., Bond, A. & Frank, R., 1993,
Mechanisms of Shaft friction in Sand from instru-
mented pile tests, J. of Geotech. Engrg. 119(1), 1–19.
85
Lehane, B.M., J.A. Schneider, and X. Xu, 2005, The
UWA-05 method for prediction of axial capacity of
driven piles in sand, Proc., 1st Int. Symp. on Frontiers
in Offshore Geotechnics. Perth, Australia: Balkema.
Randolph, M.F., Dolwyn, J. & Beck, R., 1994, Design of
driven piles in sand, Geotechnique, 44(3), 427–448.
Vennemann, P., Lindken R. & Westerweel J., 2007, In
vivo whole-field blood velocity measurement tech-
niques, Experiments in fluids, 42(4), 495–511.
White, D.J., 2005, A general framework for shaft resist-
ance on displacement piles in sand, Proceedings of the
1st International Symposium on Frontiers in Offshore
Geotechnics, ISFOG 2005. Taylor and Francis.
White, D.J., and B.M. Lehane, 2004, Friction fatigue
on displacement piles in sand, Géotechnique, 54(10),
645–658.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Cyclic jacking of piles in silt and sand

F. Burali d’Arezzo & S.K. Haigh

University of Cambridge, Cambridge, UK

Y. Ishihara
Giken Seisakusho Co. Ltd, Kochi, Japan

ABSTRACT: Jacked piles are becoming a valuable installation method due to the low noise and
vibration involved in the installation procedure. Cyclic jacking may be used in an attempt to decrease the
required installation force. Small scale models of jacked piles were tested in sand and silt in a 10 m beam
centrifuge. Two different piles were tested: smooth and rough. Piles were driven in two ways with monot-
onic and cyclically jacked installations. The cyclically jacked installation involves displacement reversal
at certain depth for a fixed number of cycles. The depth of reversal and amplitude of the cycle vary for
different tests. Data show that the base resistance increases during cyclic jacking due to soil compaction
at the pile toe. On the other hand, shaft load decreases with the number of cycles applied due to densifica-
tion of soil next to the pile shaft. Cyclic jacking may be used in unplugged tubular piles to decrease the
required installation load.

1 INTRODUCTION Even though “Press-in” is an appealing installation

Pile jacking is a common installation method for dis-
placement piles due to the low noise and vibration
involved in the installation procedure. A method
called “Press-in” was developed in Japan by Giken
Seisakusho Co. Ltd. Pre-formed tubular piles are
pressed into the ground with a static load gener-
ated by hydraulic rams. The static load is obtained
with a push-pull system by gripping the previously
installed piles (see Figure 1). The jacking machine is
able to ‘walk’ forward after the installation of each
pile while standing on the advancing pile wall. The
centre-to-centre spacing is usually one pile diameter.
Figure 1. Giken piler for tubular piles.
method for piles there are still numerous problems
related to the prediction of strength and stiffness of
jacked piles. Some of them are:
1. The installation procedure involve large dis-
placements for which the standard Finite Ele-
ment Method (FEM) is unsuitable.
2. “Press-in” piles are positioned at lower spacing
when compared to pile groups. At the moment
there are only few information on performance
of contiguous piles to vertical loads.
3. Tubular piles cannot be easily tested in the centri-
fuge due to the difficulties in placing strain gauges
along the pile shaft. Furthermore during the instal-
lation of tubular piles the shaft friction develops in
a localised region along the pile shaft and scaling
effects can occur in centrifuge (Balachowski 2006).
White (2005) studied the behaviour of jacked
piles during installation. The author described the
behaviour of soil elements next to the pile while
the pile is being installed. The key observation was
that while the pile is being installed along the pile
shaft an interface layer of crushed grains forms.
Grain crushing leads to volume reduction in the
interface layer and hence to a decrease in the shear
stress. White & Bolton (2004) called the occur-
rence “friction fatigue”. Field data and model tests
link friction fatigue to the displacement amplitude
applied at the interface, the mode of shearing and
the number of shearing cycle applied.

86
 Cyclically jacked installation can be used to
promote friction fatigue and hence reduce the jack-
ing load by applying displacement reversal at a cer-
tain depth for a fixed number of cycles. The paper
presents centrifuge tests on jacked piles installed in
flight with cyclic jacking. The depth of reversal and
amplitude of the cycles vary for different tests. The
paper examines the effect of stroke amplitude and
installation rate on friction fatigue. Pile stiffness and
strength after the installation are also assessed.
2 CENTRIFUGE MODELLING
Centrifuge modelling is widely used in soil mechan-
ics due to the possibility of reproducing at model
scale the correct stress level. The geotechnical cen-
trifuge increases the gravitational acceleration in
the model by spinning it at high centrifugal accel-
eration. The increase in acceleration compensates
the reduction in the model size. Scaling laws are
derived for correspondence between the model and
the prototype (Taylor 1995). A detailed description
of the geotechnical centrifuge at Cambridge uni-
versity may be found in Schofield (1980).
Problems may occur when modelling of piles in
centrifuge, especially of shaft piles in centrifuge,
is affected by scaling effects related to the ratio of
shear band thickness/grain size on the mobilised
shear stress which does not scale in centrifuge. The
thickness of the shear band depends on the sur-
face roughness and varies between 2–5 times d50 for
smooth piles 10–15 times d50 for rough piles being
d50 the mean grain size.
Fioravante (2002) studied the effect of the ratio
of the pile diameter/mean grain size (Dp /d50) on the
shear resistance mobilised τsu and observed that the
scaling effect can be reduced with a ratio Dp /d50 > 50.
Fioravante (2002) also observed that the shear resist-
ance mobilised depends on the normalised roughness
RN = Rt/d50 being Rt the pile surface roughness meas-
ured over a length of 0.8 to 2.5 mm. For monotonic
loading no dilation is expected for RN < 0.02 (fully
smooth) and for RN > 0.1 (fully rough) the interface
dilates and high values of τsu are expected.
The experimental work presented in this paper
was carried out in a 10 m beam centrifuge at
Cambridge University. Two different (Dp /d50) ratios
were tested in centrifuge using sand (Dp /d50 = 60)
and silica flour (Dp /d50 = 100). The aim of the
modelling is to prove whether silica flour can be
used in centrifuge in replacement of sand in the
attempt of reducing scaling effects. Also the influ-
ence of the normalised roughness was analysed
by using two different piles: one rough and one
smooth.
Tests were performed in a circular strong box of
diameter 850 mm and high of 400 mm. The height
of the soil in the tub was 320 mm for all tests. Tests
were performed at a centrifugal acceleration of
50 g and piles were installed in flight by means of
an automatic 2D actuator. Full details of the test
programme and pile installation procedure will be
given in the next paragraphs.
2.1 Test programme
Monotonic and cyclically jacked installations were
performed in saturated silica flour and in dry sand.
In total 14 pile installations were installed, 6 in silica
flour and 8 in sand. A summary of the pile instal-
lations programme is presented in Table 1. For the
cyclically jacked tests the displacement was reversed
at three different depths: 25 mm, 75 mm and 125 mm

Table 1. Test programme.

Installation Installation speed Stroke

Test Soil Pile type method [mm/s] = [mm] = RN =

Si_M01 Silica flour Smooth M 0.1 – 1.60

Si_M02 Silica flour Smooth M 0.5 – 1.60
Si_M03 Silica flour Smooth M 4.0 – 1.60
Si_S01 Silica flour Smooth CJ 0.5 10 1.60
Si_S02 Silica flour Smooth CJ 0.5 2 1.60
Si_S03 Silica flour Smooth CJ 0.5 0.4 1.60
Sd_M01 Sand Rough M 1.4 – 1.00
Sd_S01 Sand Rough CJ 1.4 14 1.00
Sd_S02 Sand Rough CJ 1.4 2 1.00
Sd_S03 Sand Rough CJ 1.4 0.8 1.00
Sd_M02 Sand Smooth M 1.4 – 0.12
Sd_S04 Sand Smooth CJ 1.4 14 0.12
Sd_S05 Sand Smooth CJ 1.4 2 0.12
Sd_S06 Sand Smooth CJ 1.4 0.8 0.12

Note: M = monotonic; CJ = cyclically jacking.

87
for the number of cycles. The stroke is the maxi-
mum displacement in one cycle. Figure 2 shows the
typical displacement history applied during cyclic
jacking, the stroke varies as indicated in Table 1.
The cyclic jacking speed was 1.0 mm/s for sand and
0.5 mm/s for silica flour. The rates of penetration are
comparable to the rate used at prototype scale.
2.2 Materials
2.2.1 Marine Quartz sand
Fine sand and silica flour were used in the attempt
of reducing scale effects. The sand is called Marine
Quartz sand and was supplied by Specialist Aggre-
gates Ltd.1 The sand is mostly silica at 96.2%
and its mean grain size is between 0.1–0.3 mm.
Figure 3 shows the particle size distribution (PSD)
obtained both by sieving and by SPOS (Single
particle Optical Sizing) (White 2003). The PSD
obtained by SPOS overestimates the grain size of
20–30% as described by White (2003). Sand was
poured dry with an automatic sand pourer at a
relative density of 90%. A dense sample was aimed
in order to enhance friction fatigue under cycling
shearing (Mortara 2007).
2.2.2 Silica flour
Silica flour is >99% silica and it has a mean grain
size of 12 μm that classifies it as silt. The mechani-
Figure 2. Displacement vs time (CJ).
Figure 3. Particle size distribution of Marine Quartz sand.
1. www.specialistaggregates.com
88
cal properties of silica flour are summarized in
Table 2. No standard preparation exist for prepara-
tion of centrifuge models with silica flour. Due to
the impossibility of pouring dry silica flour, silica
flour was mixed with water and pumped into the
tub (Silva (2005)). A layer of sand was used as bot-
tom drainage. In the attempt to obtain a very dense
model the tub was vibrated on a vibration table
during filling. The operation also was necessary
for the release of air bubbles eventually trapped in
during mixing.
2.3 Testing apparatus
2.3.1 Model pile
A stainless steel model pile was used for the tests.
The pile has an outside diameter of 12 mm, inside
diameter of 10 mm and length of 160 mm. The
bending stiffness of the model pile is not reliant
on accurate scaling. The purpose of having a very
stiff pile in centrifuge was primarily due to avoid
pile buckling during driving. Strain gauges were
positioned at the pile base and at the pile head
to measure loads experienced during the tests.
The strain gauges consisted of a full Wheatstone
bridge circuit, compensating for any tempera-
ture changes in the pile material. Finally, in order
to obtain the rough pile, on the outside surface
of the pile sand grains were stuck with gel glue.
Figure 4 shows the pile before and after the proce-
dure was terminated.
Values of the maximum surface roughness of the
pile were obtained by using an Taylor & Hobson
Form Talysurf series inductive profilometer over a
pile length of 2.5 mm. The value of Rt obtained for
the smooth pile was 25 μm. Due to the full scale of
Table 2. Silica flour properties (M.F. Silva (2005)).
Property Value
Grain size, d10 6.6 μm
Mean grain size, d50 15.5 μm
Minimum saturated density, γmin 15.5 kN/m3
Maximum saturated density, γmin 19.2 kN/m3
Permeability, k 6 × 10−8 m/s
Specific gravity, Gs 2.652
Critical friction angle, Φ′cv 36.6 degrees
Figure 4. Piles used for centrifuge modelling.
the instrument, it was impossible to measure the
surface roughness of the rough pile in the same way.
Rt can roughly be estimated as Rt = d50. From this
values, values for RN can be calculated as described
in section 2, which are reported in Table 1.

2.3.2 Static actuator

Piles were installed in flight with a 2D actuator
designed by Haigh et al. (2010). The actuator is able
to apply a combination of loads and displacements
in both vertical and horizontal direction. Multiple
installation could therefore be carried out in the
same flight. Piles were installed at centre-to-centre
spacing greater than 11 times D (being D the pile
diameter). As described by Gui & Bolton (1998) the Figure 5. Installation load (M).
effect of the container/pile diameter is negligible for
a distance greater than 10 times the pile diameter.

2.4 Pile installation procedure

Piles were installed in two ways: monotonic and
cyclically jacked. The maximum installation depth
was 160 mm. For cyclically jacked installation the
pile was stopped and displacement reversed at 25,
75 and 125 mm depth for a given number of cycles.
Mortara (2007) showed that for cyclic shearing
in modified shear box, the shear stress degrada-
tion reaches an asymptote after a maximum of
30 cycles. For each depth, therefore, the displace-
ment was reversed 30 times, the first inversion was
for half of thestroke as indicated in Figure 2.

2.5 Instrumentation
Miniature pore pressure transducers (PPT) were
used for measuring excess pore pressure induced
by pile installation in silt. The transducers have a
maximum pressure reading of 700 kPa. Sensors
were placed at 20 mm from the pile centre after
the consolidation phase was terminated. An hol-
low tube was used for drilling holes in the model
and for driving the transducers in the holes easier.
Data from all channels were acquired at a sampling
frequency of 500 Hz.
3 RESULTS
3.1 Monotonic installations
Five monotonic installations were performed in
total: 3 in saturated silt and 2 in dry sand. In silt
the installations were carried out at different instal-
lation rate as indicated in Table 1, in sand an instal-
lation rate of 1.4 mm/s was used in both tests but
two different piles were used.
Figure 5 shows the installation loads recorded
by the strain gauges on the pile head for (a) silica
Figure 6. Mobilised shaft load vs base load.
flour and (b) sand. As expected the loads required
for installation in silica flour were considerably
lower than those recorded in sand. The maximum
load recorded for silt was 289 N whilst in sand was
4.9 kN. The inconstant trend of the installation
loads recorded for Si_M03 in Figure 5(a) is prob-
ably due to the presence of in homogeneities in the
soil body. For the tests in sand, Figure 5(b) clearly
shows that the load required for installing the rough
pile (Sd_M01) is 80% higher than for the smooth
pile (Sd_M02). This is believed to be primarily to the
increase in the effective base area after roughening.
It is interesting to observe the amount of
shaft load mobilised during the installation.
Figure 6 shows the maximum base and shaft load
recorded at the end of each installation normalised
by the total head load. It can be observed that,

89
for all installations, the shaft load is always less
than 50% of the head load and it is maximum for
Sd_M01 (30%). The lowest value is 8% for Sd_M02.
As observed by Uesugi & Kishida (1987) in direct
shear tests, the interface friction angle can be cor-
related to the normalised surface roughness RN of
the interface sand/steel. In the tests performed RN
is 0.12 for smooth piles in sand, 1 for rough piles
in sand and 1.6 for smooth piles in silica flour. The
interface friction angle increases with normalised
roughness.
3.2 Cyclically jacking
As described in Section 2, cyclically jacking
involves pile displacement reversal at a given depth
for a fixed number of cycles. Typical results for
silica flour and sand are shown in Figure 7. For
simplicity only the installations with the largest
stroke are shown.
After displacement reversal it is clear that for
all installation the head load required for installing
the pile increases more than the monotonic value.
The extra load required diminishes when the pile
is pushed deeper and goes toward its monotonic
value. Typically the area influenced by cycli-
cally loading is between 2–3 pile diameter. This is
believed to be due to compaction occurring under
the pile toe during cyclic jacking.
Although Figure 7 clearly shows the effect of
cyclic jacking on the required force for installation,
it does not give information on the behaviour of
shaft and base load evolution during cycling. For
this aim, Figure 8 shows a typical history of head,
base and shaft load during cyclic jacking. For com-
parison base and shaft loads are normalised by the
value at the first cycle. It can be observed that dur-
ing cyclic jacking, the base load increases for each
Figure 7. Installation load (CJ).
90
Figure 8. Head, Base and Shaft load—Sd_S01 at
z = 75 mm.
cycle whilst the shaft load reduces. As the shaft
load is only a small percentage of the total head
load, for closed-ended piles, cyclic jacking leads to
a overall increase in the head load.
Nevertheless it is interesting to analyse shaft load
reduction which becomes important when install-
ing unplugged tubular piles. The authors believe
a phenomenon similar to friction fatigue occurs
when the pile is sheared cyclically. In fact after
the pile is being installed, as described by White &
Bolton (2004), an interface of crushed grains
develops between the pile shaft and the surround-
ing soil. During cyclic shearing the grains in the
interface rearrange and, according to stress level,
there may be additional particles breakage and vol-
ume contraction. If volume contraction occurs, the
horizontal stress acting on the pile shaft decreases
leading to a decrease in the shaft friction.
A parameter called ’degradation ratio’ (R) can
be defined in order to evaluate the reduction of
shaft friction during cyclically jacking. The param-
eter R is the ratio between the peak values recorded
in the shaft load for each cycle n to the peak value
recorded in the first cycle.
Fshafft ( n )
R= (1)
Fshafft ( )
Figure 9 shows the degradation ratio com-
puted for the tests Sd_S01 at 25, 75 and 125 mm
depth. The increase shown in the curve at 25 mm is
believed to be due to the shallow depth of cycling
compared to the cycling stroke. For this depth,
data are not representative of the phenomenon.
For z = 75 and 125 mm, shaft load reduces with
the number of cycles and it reaches an asymptotic
value after about 10 cycles. The observation reflects
the ‘friction fatigue’ mechanism: after the first
10 cycles, the possible grain crushing has occurred,
the volume contracts to its maximum value and the
shear stress reduces to its minimum value. The mini-
mum value for R is 0.08 after 13 cycles at z = 125 mm.
For z = 75 mm the minimum value recorded is 0.35.

Figure 9. Degradation ratio R—Sd_S01.
4 CONCLUSIONS
Monotonic and cyclically jacked installations were
performed in the centrifuge in order to assess the
effects of the installation method on the stress state
of the soil surrounding the pile. An area below the
pile tip 2–3 times the pile diameter was detected to
be affected by cyclic jacking. The compaction at
the pile base lead to an addition head load when
the pile is pushed deeper. The method can be used
in closed-ended piles to improve pile stiffness and
strength.
The effect of the normalised roughness RN was
also investigated by testing a smooth and a rough
pile. The load required for the installation of the
rough pile was larger than that for the smooth pile,
the increase was mainly due to an increase in the
effective base area after pile roughening. Also the
percentage of shaft load mobilised with the rough
pile is greater than for the smooth pile, in accord-
ance with the observation by Fioravante (2002) on
the normalised roughness. RN is therefore demon-
strated to have a significant effect in the increase of
the mobilised shaft load that was increased from
8% in Sd_M02 to 30% in Sd_M01.
During cyclic jacking, conversely to the base load,
the shaft load decreases with the number of cycles
applied. A parameter R called ’degradation ratio’
was defined in order to evaluate the amount of deg-
radation. It was shown that the degradation occurs
for the first cycles after which R reaches and con-
stant value. The minimum value for R is 0.08 when
cyclic jacking is performed at z = 125 mm in Sd_S01.
Cyclic jacking can be used for unplugged tubular
piles for reducing the installation load required.
91
ACKNOWLEDGEMENT
Centrifuge tests were carried out at the Schofield
Centre at Cambridge university. A special thank
must therefore be addressed to all technicians in
the lab, without whom the testing would not have
been possible.
I am also very grateful to Giken Seisakusho Ltd.
for the financial support they are giving me for the
entire course of my Ph.D.
REFERENCES
Balachowski, L. 2006. Scale effect in shaft friction from
the direct shear interface tests. Archives of Civil and
Mechanical Engineering VI(3).
Fioravante, V. 2002. On the shaft friction modelling of
non-displacement piles in sand. Soils and foundations
42(2), 23–33.
Gui, M. & Bolton, M. 1998. Guidelines for cone penetra-
tion tests in sand. Centrifuge 98, Kimura, Kusakaba
and Takemura, 155–160.
Haigh, S., Houghton, N. & Lam, S. 2010. Development
of a 2D servo-actuator for novel centrifuge modelling.
(2001), 239–244.
Mortara, G. 2007. Cyclic shear stress degradation
and post-cyclic behaviour from sandsteel interface
direct shear tests. Canadian Geotechnical … 752(1978),
739–752.
Schofield, A.N. 1980, January. Cambridge Geotechnical
Centrifuge Operations. Géotechnique 30(3), 227–268.
Silva, M. 2005. Numerical and physical models of rate
effects in soil penetration. (July).
Taylor, R. 1995. Geotechnical Centrifuge Technology.
Blackie Academic and Professional.
Uesugi, M. & Kishida, H. 1987, January. Tests of the
interface between sand and steel in the simple shear
apparatus. Géotechnique 37(1), 45–52.
White, D. 2003. PSD measurement using the single par-
ticle optical sizing (SPOS) method. Géotechnique 53,
317–326.
White, D. 2005. A general framework for shaft resist-
ance on displacement piles in sand. Proceedings of
the International Symposium, on Frontiers in Offshore
Geotechnic, 19–21 Sept 2005, Perth, WA, Australia.
White, D.J. & Bolton, M.D. 2004, January. Displacement
and strain paths during plane-strain model pile instal-
lation in sand. Géotechnique 54(6), 375–397.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Influence of installation procedures on the response of capacitance

water content sensors

M. Caruso & F. Avanzi

Politecnico di Milano, Milano, Italy

C. Jommi
Politecnico di Milano, Milano, Italy
Delft University of Technology, Delft, The Netherlands

ABSTRACT: The effects of installation procedures on the performance of capacitance probes for moni-
toring water content changes in surficial soils are analysed numerically. The sensors measure the resonant
frequency of an inductive-capacitive circuit, which includes the surrounding soil as capacitive element.
Literature calibration curves are most often adopted to determine the soil water content from the measured
resonant frequency. Alternatively, specific calibration for a given soil is performed in the laboratory. In both
cases, reference is made to ideal conditions, in which perfect contact is assured between the probes access
tube and the soil sample, prepared at uniform void ratio. Installation procedures in the field affect the sys-
tem and they may hinder correct estimation of volumetric water content in the field. Numerical results are
presented to quantify the influence of soil density changes promoted by installation operations and of the
saturated kaolin/cement paste used to guarantee continuous contact between the access tube and the soil.

1 INTRODUCTION Sensors based on electromagnetic properties

Monitoring water content exchanges in surficial
soils and embankments is becoming increasingly
important in a number of engineering applications
related to soil stability and water management.
The hydraulic state of the soil is completely known
if independent measurements of suction and water
content are performed.
Different suction and water content probes have
been developed to the aim of tracking soil hydraulic
state. Water potential can be measured directly by
tensiometers. Nonetheless, the measurement range
of commercial equipment, without incurring in
cavitation after a short time lapse, is narrow com-
pared with the expected suction changes in most
soils exposed to soil-atmosphere interaction. High
capacity tensiometers are being developed, but in
fine grained soils direct suction measurement still
may be difficult. Therefore, water content deter-
mination appears to be mandatory to follow the
hydraulic state of the soil.
Among current equipment and techniques for
measuring monitoring soil-water status (Scanlon
et al. 2002, Charlesworth 2005), gravimetric water
content determination is the only direct way to
determine the amount of water in the soil. All the
other techniques rely on indirect methods in which
different properties of the soil depending on water
content are measured.
appear to be particularly advantageous, as they
are safe, durable, less invasive compared to other
techniques, and they have very short response time.
The soil dielectric constant of soils, at frequencies
between 1MHz and 1 GHz, is strongly dependent
on volumetric water content (Topp et al. 1980), due
to dielectric constant of water being much higher
than that of the other soil constituents. This obser-
vation has promoted the development of non-
destructive experimental techniques inferring the
amount of water in the soil from measurement of
electrical permittivity. Time Domain Reflectom-
etry (TDR) allows for water content determination
from measurement of the propagation velocity
of a high-frequency electromagnetic pulse signal.
Noborio (2001) offers a general TDR technique
review, putting in evidence its powerful capabilities,
its excellent accuracy, and a measurement range
covering the expected in situ variations for water
content. However, soil electrical conductivity and
salinity problems may be a major issue not easy
to deal with (e.g., Mojid et al. 2003, Schwank &
Green 2007). The main issue in the use of TDR is
its installation procedures, which requires manual
probe insertion into the soil, with direct access at
the depth were the probe is installed.
The dielectric constant of soils can also be derived
based on capacitance. This method includes the soil
as a part of the capacitor in which the permanent

92
dipoles of water in the dielectric medium (i.e. the 2 CAPACITANCE SENSORS
soil) are aligned by electric field and become polar-
ized. The capacitor is arranged to be part of an The EnviroSCAN® capacitance sensors consist
oscillator circuit, so that the frequency of oscilla- of two brass rings (50.5 mm outer diameter and
tion is a direct measure of capacitance (Gardner 25 mm height) mounted on a plastic sensor sup-
et al. 1991). Capacitance probes are a fast, safe and port (Fig. 1). The probes are inserted inside a PVC
relatively inexpensive way of measuring the relative access tube, strictly in contact with the rings and the
permittivity of soils, which in turn can be used to surrounding soil. Up to 16 probes can be mounted
estimate soil water content (Robinson et al. 1998, on a plastic guide at different depths. Minimum
Kelleners et al. 2004). Response time of the probes, relative distance between probes is 10 cm.
less than 1 second, is one of the most attractive The resonant frequency of the inductive-capacitive
features of capacitance probes. circuit, in which the probe is inserted, Fr, is linked to
This contribution focuses on the performance the inductive L and capacitive C elements,
of capacitance sensors (Sentek® 2001), which can
−1
be mounted in series into an access tube plugged
into the soil, down to two meters from the soil sur-
Fr ⎡2 p
⎣ ( LC ) ⎤⎦ , (1)
face. Monitoring depth can reach 5 meters below
the ground level, by means of appropriate install- where L is the circuit inductance and C = Cs + Cp is
ing procedures (Caruso, 2007). the total capacitance, including the soil component,
The sensors measure the resonant frequency Cs, together with probe constants, Cp (Kelleners
of the inductive-capacitive circuit, which includes et al. 2004). The volume of soil which effectively
the surrounding soil as capacitive element. Cali- contributes to the resonant frequency of the probe
bration relationships are then adopted to infer is a cylinder of about 20 cm diameter and 10 cm
soil water content from the measured resonance height, and its capacitance is a function of its rela-
frequency. tive dielectric permittivity, ε r.
As soil capacitance strongly depends on volu-
metric water content more than on any other prop-
erty, literature calibration curves are most often
adopted (e.g. Robinson et al. 1998, Kelleners et al.
2004). Better practice reckons on specific calibra-
tion for a given soil performed in the laboratory,
on samples prepared at uniform void ratio and
water content (Schwank et al. 2006).
In both previous cases, calibration refers to ideal
conditions, in which perfect contact is assured
between the probes access tube and a soil sample,
which is assumed to have a uniform void ratio in
the whole representative volume contributing to
the system response.
When the access tube is installed in situ, two
different procedures may be adopted: tipically, it
can be driven into the soil (to a maximum depth of
one to two meters, depending on the soil), or it can
be inserted into a drilled hole of slightly smaller
diameter. In the latter case, a mortar made of
kaolin, cement and water is added in the borehole
before inserting the access tube, in order to assure
perfect contact between the tube and the surround-
ing soil.
Both procedures create a disturbance in the
soil, which affects the resonant frequency of the
equipment, which in turn may hinder correct esti-
mation of volumetric water content in the field
based on ideal calibration procedures. Relevant
results of numerical analyses performed to quan-
tify the influence of installation procedures on
the response of the system are presented in the Figure 1. Capacitance sensors, internal circuit and
following. plastic support.

93
 Since the sensors output primarily varies with
the air/water ratio, the resonant frequency is scaled
to the maximum measurement range. The scaled
frequency, SF, is defined as
SF = ( Fa Fr ) ( Fa Fw )
−1
(2)
where Fr is the reading in the soil, and Fa and Fw
are the frequency readings when the access tube
is surrounded by air and water, respectively. After
scaling, unique relationships between the scaled
frequency SF and the soil volumetric water con-
tent, θ, are established, which play the role of
calibration curves for water content (Paltineanu &
Starr 1999).
3 IMPLEMENTATION
To analyse the possible effects of installation
procedures on the system response, three differ-
ent experimental configurations were simulated
numerically.
The analyses were performed by means of full
3D finite element discretisation, implemented in
Comsol Multiphysics®. Preliminary analyses had
shown that, due to the shape of the plastic sup-
port (Fig. 2), axisymmetric scheme should not be
adopted in the simulation, because of the relevant
error induced in the simulated electrical response.
Charge conservation throughout the domain
and null electrical fluxes across the external
boundaries were imposed. The radius of the dis-
cretisation domain was fixed in all the analyses to
220 mm (Fig. 3). Different discretisation meshes
were adopted, up to a maximum of roughly
150000 tetrahedral elements. Maxwell laws were
solved with reference to steady state response of
Figure 2. Top view of the inner part of the mesh,
showing the plastic support section.
94
Figure 3. General view of the discretisation domain.
the system imposing a fixed potential difference
of 10 V between the brass rings, and solving for
medium capacitance. Afterwards, the resonant
frequency of the circuit was derived based on its
definition (eq. 1).
A reference system was first analysed
(configuration 1), representing the ideal labora-
tory calibration protocol, where perfect contact is
assured between the access tube and a sample hav-
ing uniform void ratio in the whole representative
volume. The results of this analysis were compared
to the two numerical schemes representing, respec-
tively, the driven access tube (configuration 2),
and the access tube plugged into a pre-drilled hole
filled with mortar (configuration 3).
The resonant circuit of a probe was simulated
by reconstructing the real geometry of the relevant
portion of the system, composed by the sensor, the
access tube and the PVC internal support, with a
total height of 212 mm.
As for configuration 1, the ideal installation
procedure has been modeled by defining a simple
porous media of average properties around the
tube, with a porosity of 0.45.
As for configuration 2, the driven procedure
has been reproduced by decreasing the void ratio
in an inner portion of the domain (Fig. 4), fol-
lowing field and model experience on driven
piles (Randolph et al. 1979, Dijkstra et al. 2012).
Configuration 3 entails the definition of an inner
zone of 1.5 mm filled with kaolin, cement and water
around the access tube, while the surrounding soil
volume was assigned an undisturbed porosity of
0.45, as a first approximation (Fig. 5).
As for the relative dielectric constants, values
of 1 for air (εr,a), of 88 for water (εr,w), of 3 for
 To analyse the behaviour of sensors in contact
with moist soil, a mixture law had to be adopted to
assign dielectric properties to the soil layers. Two
different mixture models were tested, the one pro-
posed by Dirksen (Hilhorst et al. 2000),

ε S (θ ) θ ε + ( − n) ε ( θ) ε , (3)
r r,w r,s r,a

where εr is the relative dielectric constant of the mix-

ture, n is the soil porosity, θ is the volumetric water
content, and S(θ) is the experimental relation:

1
S (θ ) = (4)
Figure 4. Schematic top-view of driven installation
3( n − θ )
scheme.
and the simple model by Topp et al. (1980):

ε = 3.03 + 9.3 θ 146 0 θ 2 − 76.7 θ 3. (5)

r

The resonant frequency of the sensor as function

of the liquid water content is reported in Figure 6.
Simulations have been performed varying θ from 0
to 0.45 in steps of 0.05. The results of the simulation
are compared to experimental data obtained at the
Politecnico di Milano on soil samples of sandy silt
at different θ (0.04, 0.22 and 0.38). Samples were
prepared by moist tamping around the access tube
at constant porosity of 0.45. The model was able
to reproduce the experimental data in the range of
volumetric liquid water content investigated, even
if paucity of experimental data limits the validity
Figure 5. Schematic top-view of drilled installation
scheme.

the soil particles (εr,s) and for PVC, and of 103

for brass rings were imposed. The imaginary
component of the relative dielectric constant of
water was neglected, because the operational fre-
quencies of the probe are far from the value of
frequency at which relaxation of water occurs
(Mojid et al. 2003).

4 REFERENCE MODEL

Preliminary analyses were performed to simulate

the resonant frequencies of the system ideally
inserted in air or water, which bound the response
range of the sensors in contact with soil. The cal-
culated resonant frequency in water and in air were
99.8 MHz and 145.5 MHz, respectively. These val-
ues agree with the typical resonant frequencies of Figure 6. Simulation results for reference conditions:
the capacitive sensors in air and water, as reported mixture models compared with experimental data in the
e.g. by Kelleners et al. (2004). laboratory.

95
 between the soil volume prior to and after instal-
lation was imposed to determine the final aver-
age porosity in the disturbed volume. A porosity
profile for the disturbed area was determine
imposing it had this average value, and the initial
porosity n0 = 0.45 at the external boundary of the
disturbed soil volume. The final porosity profile
reads:

⎛ R2 ⎞
n(r ) = 0.4558 − 0.144 ⎜ 20 ⎟ (6)
⎝r ⎠

The porosity profile was then discretised in lay-

ers, while assigning the dieletric characteristics to
the numerical model. Continuity of the degree of
Figure 7. Electric field in an ideal soil with θ equal to saturation was assumed across the layers having
0.05 (mixture model by Topp et al. 1980). different posrosity.

5.2 Drilled hole with mortar

To simulate the drilling procedure, an additional
volume of kaolin, cement and water was added
between the access tube and the undisturbed soil.
The mixture model was assumed for the mortar
fixing the water content of the mixture to the value
of 0.17, which is the mean value of water content
used in the field.

5.3 Results
Simulated resonant frequencies for the three con-
figurations as a function of volumetric water
content of the undisturbed soil, θ0, are shown in
Figure 9.
Figure 8. Electric field in an ideal soil with θ equal to
0.45 (mixture model by Topp et al. 1980).

of this statement. The mixture model by Dirksen

seems to better reproduce experimental data on the
entire water content range.
In Figures 7 and 8, the calculated electric field
generated by the probes for θ equal to 0.05 and
0.45, respectively is shown. As expected, the vol-
ume of soil significantly affecting the response of
the system decreases at increasing water content.

5 INSTALLATION EFFECTS

5.1 Driven access tube

To model the effects of the driven access tube, a
disturbed inner zone was introduced.
The disturbed zone has a thickness of roughly
110 mm, which corresponds to a ratio between
the disturbed thickness and the access tube radius Figure 9. Simulation results for the three models in
(R) equal to r/R = 5. Soil mass conservation terms of resonant frequency as function of θ0.

96
 As evident, the driven access tube procedure
entails a change in the resonant frequency of the
system, which increases with the undisturbed volu-
metric water content, θ0. The actual resonant fre-
quency is always greater than that of the ideal case,
at the same θ0.
The difference between the estimation of θ0,
with respect to the ideal reference scheme, is in
the range between 10% and 20%. The maximum
difference corresponds to θ0 equal to 0.2, which
means around a degree of saturation of about
0.5.
As for the drilled procedure, 3, the effect of the
kaolin mortar changes with θ0, since it represents
a wetter layer in dry soils, and a drier layer in wet
soils. Therefore, the results for the ideal scheme
and for the drileld scheme intersect, at a value of
θ0 which is roughly equal to the volumetric water
content of the mortar.
The difference between the estimations of θ0 in
this case, with respect to the ideal reference scheme,
may be higher than for the driven procedure, and
ranges from 0 to 40%. The maximum difference
occurs for dry soil.
These conclusions are confirmed also by the
elaboration in Figure 10, in which the comparison
between the actual θ0 and the values of θ one would
estimate relying on a ideal scheme are compared.
The latter were calculated by scaling the simulated
resonant frequencies with the values of resonant
frequency in air and water, as suggested in a stand-
ard interpretation procedure.
Figure 10. Simulation results for the three models in
terms of inferred volumetric water content as function
of θ0.
97
6 CONCLUSIONS
A finite element model was designed to assess
the influence of installation procedures on the
response of capacitance water content sensors.
The results allow for quantifying the sensitivity
of the system to porosity changes, relevant for the
driven installation procedure, and to the presence
of the kaolin mortar adopted in the drilled instal-
lation procedure.
If disturbance induced by installation is disre-
garded, erros up to about 20% of the actual value
may affect the estimation of water content. In
both cases, the estimated water content is gener-
ally lower than the actual one, although for differ-
ent reasons and with different relevance at varying
water content.
The conclusions of this numerical study were
confirmed by recent preliminary data coming
from a dedicated experimental investigation in the
field.
REFERENCES
Caruso, M. 2007. A theoretical, experimental and numer-
ical approach to predict the hydrologic balance in the
upper unsaturated soil layers (in Italian). Doctorate
Thesis, Università di Parma.
Charlesworth, P. 2005. Soil water monitoring. An infor-
mation package. 2nd ed. Irrigation Insight No 1.
Dijkstra, J., Broere, W. & Van Tol, A.F. 2012. Electrical
resistivity method for the measurement of density
changes near a probe. Géotechnique 62(8): 721–725.
Gardner, C.M.K., Bell J.P., Cooper J.D., Dean T.J.,
Hodnett M.G. & Gardner N. 1991. Soil Water Content.
In Smith R.A., Mullings C.E., (eds) Soil Analysis—
Physical Methods. Marcel Dekker, New York.
Hilhorst, M.A., Dirksen, C., Kampers, F.W.H. &
Feddes, R.A. 2000. New dielectric mixture equation for
porous materials based on depolarization factors. Soil
Science Society of America Journal. 64: 1581–1587.
Kelleners, T.J., Soppe, R.W.O., Robinson, D.A., Schaap, M.G.,
Ayars, J.E. & Skaggs, T.H. 2004. Calibration of capaci-
tance probe sensors using electric circuit theory. Soil
Science Society of America Journal. 68: 430–439.
Mojid, M.A., Wyseure, G.C.L. & Rose, D.A. 2003.
Electrical conductivity problems associated with
time-domain reflectometry (TDR) measurement in
geotechnical engineering. Geotechnical and Geological
Engineering 21: 243–258.
Noborio, K. 2001. Measurement of soil water content
and electrical conductivity by time domain reflectom-
etry: a review. Comp and Elec in Agric 31: 213–237.
Paltineanu, I.C. & Starr, J.L. 1997. Real time soil water
dynamics using multisensors capacitance probes. Soil
Science Society of America Journal 61: 1576–1585.
Randoplh, M.F., Carter, J.P. & Wroth, C.P. 1979. Driven
piles in clay—the effects of installation and subse-
quent consolidation. Géotechnique 29(4): 361–393.
Robinson, D.A., Gardner, C.M.K., Evans, J.,
Cooper, J.D., Hodnett, M.G. & Bell, J.P. 1998. The
dielectric calibration of capacitance probes for soil
hydrology using an oscillation frequency response
model. Hydrology and Earth System Sciences 2 (1):
111–120.
Scanlon, B.R, Andraski, B.J. & Bilskie, J. 2002. Miscel-
laneous methods for measuring matric or water poten-
tial. In Dane, J.H. & Topp, G.C. (eds), Methods of Soil
Analysis, part 4, Physical Methods. Soil Sci Soc. Am:
643–670.
Schwank, M. & Green, T.R. 2007. Simulated effects of
Soil Temperature and Salinity on Capacitance Sensor
Measurements. Sensors, 7(4): 548–577.
98
Schwank, M., Green, T.R., Mätzler, C., Benedickter, H. &
Flürer, H. 2006. Laboratory characterization of a com-
mercial capacitance sensor for estimating permittivity
and inferring soil water content. Vadose Zone Journal.
5: 1048–1064.
Sentek 2001. Calibration of Sentek Pty Ltd Soil Moisture
Sensors. Sentek Pty Ltd, Stepney, South Australia.
Topp, G.C., Davis, J.L. & Annan, A.P. 1980. Electromag-
netic determination in soil-water content: measure-
ment in coaxial transmission lines. Water Resources
Research. 16: 574–582.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

The load capacity of driven cast in-situ piles derived from installation
parameters

D. Egan
Keller Foundations, UK

ABSTRACT: The theoretical estimation of the ultimate capacity and serviceability performance of
common displacement foundation systems such Driven Cast In-Situ (DCIS) piles is difficult due to the
huge disruption in soil structure and insitu stress regime caused by the installation process. Even though
much research effort is expended on complex numerical modeling and reduced scale laboratory or centri-
fuge modeling there remains the difficulty of translating the knowledge gained into practical prediction
tools appropriate for routine design and installation of the full size product in the field. So to advance
and validate the conclusions drawn from numerical and small scale research the third strand of measuring
and analysing full size field behaviour must be added. This paper will summarise recent advances made in
the field measurement and analysis of installation parameters to predict the load capacity of driven cast
in-situ piles. The results from installation and testing of a DCIS pile is used to illustrate the methodology
now being routinely by Keller Foundations in the UK. The conclusions drawn from this paper are already
raising the standard of reliability, efficiency and sustainability of DCIS piles on routine projects.

1 INTRODUCTION laboratory or centrifuge modeling there remains

The Driven Cast In-Situ (DCIS) piling method
has been in use for many decades and has over
this period demonstrated its effectiveness and
efficiency.
Examples of more recently completed large
structures supported on DCIS piles installed
by Keller in the UK include the O2 Arena (con-
structed within the Millennium Dome—itself
supported in DCIS piles) and the London 2012
Olympic stadium.
The body of published literature relating to the
design of DCIS piling is not large. Neely (1991)
created a database of load tests on DCIS piles with
expanded bases and developed correlations for
design. Evers et al. (2003) provided a useful study
for the performance of DCIS piles compared to
CFA piles at a site in France & Flynn et al (2012)
present a case study for the performance of a DCIS
test pile at a site in London.
Development of a complete theoretical model
for the design of DCIS piles is complicated by the
very significant disruption in soil structure and
insitu stress regime caused by the installation proc-
ess. Even the selection of appropriate soil param-
eters for use in design equations from ground
investigation tests is problematic due to the state
changes occurring during pile installation.
Even though much research effort is expended
on complex numerical modeling and reduced scale
the difficulty of translating the knowledge gained
into practical prediction tools appropriate for rou-
tine design and installation of the full size product
in the field. This paper considers a method to esti-
mate DCIS geotechnical capacity based on instal-
lation parameters measured during the installation
of a pileby modern instrumented piling rigs.
2 MODERN DCIS PILING EXECUTION
In recent times advances in piling rig instrumenta-
tion and telecommunications have enabled a new
level of sophistication in the recording of installa-
tion parameters and this opens up the potential of
assessing pile capacity in real time.
DCIS design using a static design approach,
based on ground parameters derived from bore-
holes and laboratory testing has for many years
been the approach used to estimate ultimate pile
capacity, but, as noted above, this approach has
its difficulties. At the start of execution of piling,
trial drives, are undertaken to validate the ground
conditions are as envisaged in the initial design.
(Historically with manual counting of the number
of hammer blows to achieve a given penetration.)
The control criteria can then be set for installa-
tion of the production piles. Where required this
can be supplemented with project specific static
load testing to support and validate the achieved

99
pile load capacity. While this approach has been
used satisfactorily for the installation of hundreds
of thousands of piles over many decades, mod-
ern instrumentation now allows both robust and
detailed control and feedback of the pile driving
process, in real time, for every pile installed.
On modern instrumented DCIS piling rigs the
energy delivered to each pile can now be meas-
ured and used to validate that pile’s ultimate load
capacity. As the pile is driven the driving param-
eters (drive energy, drive tube depth and rate of
advancement) are displayed to the rig driver giving
a high degree of control to the installation proc-
ess. Additional benefits of an automated electronic
data capture system include a reduction of manual
recording and on-site paperwork (which requires
a finite manpower resource), robust archiving of
records and real time access to installation data
remote from the work site.
From the experience of developing a large
database of systematically acquired pile instal-
lation data it has been possible to map the proc-
ess required to allow real time assessment of pile
capacity. Initially focused on piles founded in
granular soils an empirical approach has been
developed which has been used to validate, in near
real time, the adequacy of DCIS piles as they are
driven. Of course these advances should be seen
in the context of dynamic pile driving formulae
which have been used on preformed piles for many
decades. Not withstanding widespread the use of Figure 1. Typical borehole from the test site.
set calculations for pre-formed piles, their appli-
cability (and reliability) for driven cast in-situ pil-
ing, where the drive tube is firstly driven into the and Alluvium were not considered as contribut-
ground and then removed and replaced by cast in ing to the pile geotechnical capacity, but rather a
place concrete, is highly questionable. source of potential negative skin friction, they are
The development of real time assessment of ignored in the assessment of ultimate pile capacity
DCIS pile capacity for piles driven into coarse for design purposes.
grained soils is presented in detail below. This
method has been developed over a period of time
and is based on drive data and load test data 4 DRIVEN CAST INSITU PILE
carried out on a series of sites. However to bet- INSTALLATION
ter illustrate the process reference to a single test
pile (TP3) installed at a site in Erith, London is Figure 2 illustrates the pertinent features of the
made. DCIS system. A steel drive tube, closed at the base by
a sacrificial base plate, is top driven to the required
depth. The steel tube is then charged with a free
3 DESCRIPTION OF TEST SITE flowing concrete and withdrawn from the ground.
Steel reinforcement can be inserted in the pile bore
The test site was located within the Thames basin before or after the concrete is placed. The require-
to the south of the river at Erith, London. The ments for the execution of DCIS piles are set out
ground conditions comprised a mantle of Made in EN 12699 ‘Execution of special geotechnical
Ground, overlying soft and very soft clay and works—Displacement piles’. In addition Keller
peat Alluvium. The piles were founded in coarse monitor all aspects of the installation process by
grained River Terrace Deposits (RTD) comprising onboard computer, with the key parameters for
medium dense to very dense gravelly sand. Thanet capacity assessment being the hammer energy
Sand was found beneath the RTD. Figure 1 shows imparted for each blow, the rate of advancement
a typical bore hole for the site. The Made Ground of the tube per blow and the tube and base plate

100
Figure 2. The DCIS installation process.

geometry. The recorded data are instantly fed back

to the design office by modem enabling secure
backup of the information as well as providing the
facility for the design engineer to see feedback on
the ground conditions encountered immediately
the pile is completed.

5 RECORDED INSTALLATION
PARAMETERS

From the recorded parameters the tube drive resist-

ance with depth is calculated, where Ru is the drive
resistance, N is the hammer energy, s is the rate of
tube advancement per hammer blow and c is the
elastic compression of the tube.
f1 f2 N
Ru = (1)
⎛ c⎞
s−
⎝ 2⎠
The factors f1 and f2 are hammer and energy
transfer efficiency factors. The elastic compression
of the drive tube is calculated using the following
equation, which is based on the Danish pile driving
formula, and where N is the hammer energy, L is
the tube length, A is the cross sectional area of the
tube and E the elastic modulus.
2 f1 f2 NL
c= (2)
AE
Figure 3 shows the drive resistance, Ru, plotted
with depth for TP3 and the four associated anchor
piles used for reaction in the maintained load test.
The piles were DCIS piles of nominal diameter,
d, of 340 mm with a base plate diameter, db, of
380 mm. TP3 was driven fractionally over 11.0 m
(11.01 m) and the anchor piles to 15 m.
The Made Ground and Alluvium proved to be
of very low strength which drive tube easily pen-
etrated with little energy required. The RTD was
encountered at around 8.25 m depth as indicated
by a sharp increase in drive resistance to between
Figure 3. Dynamic drive resistance, Ru, plotted with
depth.
500 kN and 700 kN as shown in Figure 3. This
corresponds to medium dense to dense gravelly
RTD. Below 12 m the anchor piles encountered a
further sharp increase in drive resistance to around
2300 kN at 15 m, which is thought to coincide
with the top of the Thanet Sand at this location.
The test pile was intentionally stopped within the
medium dense zone so as to ensure geotechnical
failure during the maintained load test.
6 CORRELATION OF PILE
INSTALLATION PARAMETERS
WITH SOIL PARAMETERS
The installation of each pile where state of the art
instrumentation is used can be considered as a form
of large scale ground investigation test from which
ground parameters may be derived. Dividing the
drive resistance by the area of the base plate, Ab,
gives the bearing pressure, qpc, at the pile toe.
Ru
q pc = (3)
Ab
Taking qpc to be analogous to the cone resist-
ance, qc, obtained by a static cone penetration test
a ready and quick evaluation of the nature of the

101
ground into which the pile is being driven can be
obtained. In developing the analogy between qpc
and CPT qc factors such as energy losses within the
drive/tube system, frictional losses along the drive
tube shaft, rate effects and size effects have all to
be considered.
An extensive series of field tests using dynamic
pile driving analysis has been undertaken to assess
the loss in energy between the point of impact of
the hammer at the top of the drive tube and the
pile base plate.
The ratio of the area of the base plates of differ-
ent sizes of DCIS pile and a standard CPT range
between 26 and 6, and taken in the round this is
considered beneficial. One of the disadvantages
of the CPT, especially in coarse soils, is the pro-
pensity for the tip resistance to be influenced by
larger soil particles (or localised weaker or softer
layers) in the soil. In contrast the larger diameter
of the DCIS base plates stresses a greater depth
of soil leading to somewhat of lower bound (and
safe) smoothing effect when compared to the CPT
qc. Indeed the zone of soil stressed beneath the
pile toe at the end of driving is exactly that which
will be stressed by the finished pile in the working
condition. (This is in the context of a single pile,

Figure 5. Angle of friction, φ′, from drive resistance Ru.

where piles are grouped such that they interact

suitable consideration of group effect must be
made in the design.).
Figure 4 compares qc from a CPT test under-
taken about 4 m from TP3, and pile toe bearing
pressure, qpc.
Once qpc is derived a correlation with effective
angle of friction, φ′, may be made, for example
using the method shown in Lunne et al (1997).
From the measured drive resistance, Ru, the
dynamic bearing pressure, qpc, is derived and cor-
related to in-situ angle of friction for the gravelly
sand founding stratum. Figure 5 shows the derived
angle of friction over the socket length of TP3
driven into the RTD.

7 ESTIMATE OF PILE CAPACITY

FROM INSTALLATION PARAMETERS

The ultimate pile socket capacity, Qult, comprising a

shaft component, Qs,ult, and base component, Qb,ult,
may now be derived using standard soil mechanics
principles, where:

Qult Qs ,ult + Qb,ult (4a)

Figure 4. Comparison of qc an qpc for TP3. Qult β ′

v l + Nq ⋅ σ v′ ⋅ Ab (4b)

102
 The above equation is readily implemented in the
commercially available software or spreadsheets.
The program Oasys Pile, using the Berezantzev
formulation for the bearing capacity factor, Nq,
(Oasys Pile user manual, 2012) was used to give:-
Qult , socket kN kN kN
Test Pile 3 was subjected to a maintained
extended proof load test to failure following the
procedure outlined in the ICE Specification for
Piling and Embedded Retaining Walls (2007).
Figure 6 shows the load settlement response.
A plunging failure of the pile under an applied
load, Qf, of 1947 kN and at a displacement of
45 mm was observed.
In the analysis of the pile test results the tem-
porary hold up resistance generated over the pile
shaft passing through the made ground and allu-
vium was estimated at 80 kN, which must be added
to Qult,socket as calculated in pile to give the true
anticipated theoretical pile capacity of 2060 kN.
Thus the pile under test achieved 94% of its the-
oretical capacity which is considered an excellent
correlation between the calculated capacity and
that shown by the static load test.
The results of the illustration of the installation
and load testing of pile TP3 show that a rational
Figure 6. Load settlement response of test pile 3.
procedure to assess the ultimate capacity of DCIS
piles located in coarse grained soils based in pile
installation data is available.
8 CONCLUSIONS
(5)
Driven cast in-situ piling has proved a popular and
efficient piling method over many decades, which,
in suitable ground conditions, produces piles of
high capacity and efficiency. Recent developments
in rig instrumentation have enabled the routine and
systematic acquisition of high quality installation
records. The benefits of an automated electronic
data capture system include a reduction of manual
recording and on-site paperwork (which requires
a finite manpower resource), robust archiving of
records and real time access to installation data
remote from the work site. In addition Keller has
demonstrated the development of a real time vali-
dation method for DCIS piles is viable. A rational
link between the installation parameters, funda-
mental soil parameters relevant after pile installa-
tion, the physical mechanisms of load generation
and observed load capacity has been established.
These developments in the driven cast in-situ
piling process have lead to improved efficiency
(and hence sustainability) of the DCIS method.
REFERENCES
EN 12699:2001. Execution of special geotechnical works—
Displacement Piles. 2001. BSI. London.
Evers, G. Hass, G. Frossard, A., Bustamante, M. Borel, S. &
Skinner, H. 2003. Comparative perfomances of con-
tinuous flight auger and driven cast in place piles in
sands. Deep foundations on bored and auger piles.
Van Impe (ed) Millpress, Rotterdam.
Flynn, K. McCabe, B.A. & Egan, D. 2012. Proceedings
9th International Conference on testing and design
methods for deep foundations, Kanazawa, Japan.
ICE Specification for Piling and Embedded Retaining
Walls 2007. 2nd Ed. ICE. London.
Lunne T., Robertson P.K. & Powell, J.J.M 1997. Cone
Penetration testing in geotechnical Practice.
Neely, W.J. 1990. bearing capacity of expanded-base piles
with compacted concrete shafts. Journal of geotechni-
cal Engineering, 116(9): 1309–1324.
Oasys Pile 19.2 User manual 2012 http://www.oasys-
software.com/media/Manuals/Latest_Manuals/
Pile19.2_manual.pdf.
103
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

On the numerical modelling and incorporation of installation

effects of jacked piles: A practical approach

H.K. Engin
Geo-Engineering Section, Delft University of Technology, Delft, The Netherlands

R.B.J. Brinkgreve
Geo-Engineering Section, Delft University of Technology, Delft, The Netherlands
Plaxis BV, Delft, The Netherlands

A.F. van Tol

Geo-Engineering Section, Delft University of Technology, Delft, The Netherlands
Deltares, Delft, The Netherlands

ABSTRACT: The installation process of a displacement pile causes a considerable amount of soil dis-
placement and high levels of stresses, and therefore alters the soil state and properties around the pile.
These installation effects may have important consequences on the performance of the pile in its service
life (e.g. load—displacement behaviour) and on the neighbourhood (e.g. vibrations, nuisance). A more
realistic behaviour and therefore an improved design would be achieved by considering the installation
effects in the analyses. In current practice, the installation effects are taken into account by some empiri-
cal design methods in order to estimate the bearing capacity of foundation piles. The objective of this
numerical study is to investigate and model the installation effects of pile jacking in sand in a numerical
framework. In the first part of the study a simplified numerical technique was employed to investigate
the installation effects. The results are approximated by nonlinear regression. Despite the limitations and
simplifications, it was shown that the installation effects can be represented in terms of functional forms
reasonably well. Furthermore, these functions can be easily applied in a standard FE analysis.

1 INTRODUCTION analyses employ small deformation formulations.

The bearing capacity of a pile depends on the soil
properties and the stress state. The stress and density
dependent behaviour determines the soil response.
Depending on the installation method as well as
the geometry of the pile, the mechanical (i.e. stress
and density) and physical (i.e. grain size, sphericity,
roundness) properties are altered due to large stress
and strains dictated by the penetration of a dis-
placement pile. These installation effects have been
investigated throughout field (Lehane 1992; Chow
1996, among others), laboratory (White 2002;
Dijkstra 2009; Liu 2010, among others) and numer-
ical (Baligh 1975; The & Houlsby 1991; Mahutka
et al. 2006, among others) studies. Incorporation of
the installation effects is important in order to make
better predictions on the pile bearing capacity and
its behaviour under different loading conditions.
In geotechnical engineering practice, it is common
to numerically model displacement piles as ‘wished-
in-place’ without simulating the installation process.
Furthermore, most of the FE packages used in such
The installation effects can be obtained by simu-
lating the penetration process. However, it is not
possible to model the penetration process using FE
method with small deformation formulation unless
extra update of the FE mesh is employed (e.g.
Einav & Randolph 2005). Simple techniques such
as the use of prescribed boundary conditions at the
pile soil interface (Broere & van Tol 2006; Said et al.
2008; Pham et al. 2010) could provide case specific
solutions in obtaining the installation effects. There-
fore the solutions could not be generalised.
In this paper first the Press-Replace technique
(Engin & van Tol 2011), which employs a small
deformation FE model, yet accounting for large
deformation effects, is briefly explained. The
method was used to obtain installation effects of
a jacked pile in sand using hypoplasticity to model
the sand behaviour.
Secondly, the possibilities of approximating the
installation effects by mathematical surfaces were
investigated. As the most relevant state for the geo-
technical practice is the state after pile installation,

104
the final equalisation states have been analysed, i.e.
the situation after the jack force has been removed.
For the approximation of the installation effects
a nonlinear regression algorithm is employed. By
the approximated form the state around a pile that
is modelled as wished-in-place, can be altered to
incorporate the installation effects without simu-
lating the installation process. Finally a compari-
son of the results obtained by the approximated
effects and the simulation results using the Press-
Replace technique are given.
2 NUMERICAL MODELLING
OF PILE JACKING
The installation effects around a jacked pile have
been investigated using the the ‘Press-Replace’ (PR)
technique (Engin & van Tol 2011). The technique
was introduced to promote the modelling of large
deformation effects using small deformation for-
mulations. It was shown that the installation effects
could be modelled reasonably well. In Figure 1, a
general view of the model and four subsequent cal-
culation phases are shown. The calculation phases
continue until the desired level of penetration of the
pile is achieved. In order to facilitate the PR tech-
nique, an axisymmetric FE mesh with small slices
is introduced in the region where the pile will be
jacked. Interface elements are also defined between
the pile tip and the underlying soil to model proper
interaction with the continuum. In order to avoid
stress oscillations at the corner of the pile tip, inter-
faces should be extended into the soil volume (van
Langen 1991). Based on the preliminary study, the
vertical as well as horizontal interface extension
int
lengths equal to the slice thickness (lext tslice ) were
suggested. Furthermore, an optimum slice thick-
ness in terms of numerical accuracy and CPU time,
tslice = 0.10 ⋅ Dp was found.
In this study, the PR technique is used to model
continuous pile jacking in sand to investigate the
installation effects for different pile geometry and
soil density variations. Due to lack of space, only
one variation (D = 0.40m; L = 10D; Id = 0.80) is
presented in this paper. The focus is on the change
in state around the pile due to the installation
process, which also dictates the load displacement
Figure 1. Details on the Press-Replace modelling tech-
nique and progress of penetration of the pile (Engin
et al., 2011).
105
behaviour. A simple sketch of the geotechnical
problem considered and the corresponding axisym-
metric FE model employed in the numerical analy-
ses is given in Figure 2. The soil profile consist of
approximately 15m thick very soft clay layer over-
lying a sand layer, which is a typical profile in The
Netherlands and places having similar geological
sequences (e.g. Pleistocene glacial and interglacial
periods). The clay layer is modelled as an elas-
tic layer of 1 m thickness having an effective unit
weight of γ ′ = 100κN/m3. The purpose of modelling
the clay layer as such is to facilitate the generation
of geostatic (K0) stresses in the sand layer. There-
fore the clay layer is considered to generate a proper
stress level on top of the sand layer. The interaction
between the pile and the clay layer is also ignored
due to very low cohesion of the clay layer.
To model sand behaviour, hypoplasticity (von
Wolffersdorff 1996) was used in the analyses due
to its state dependent stiffness and strength for-
mulation. Having a state dependent stiffness and
strength results in different levels of stiffness and
strength mobilisations at different levels of pen-
etration depths. Similarly, different levels of mobi-
lisation of stiffness and strength is obtained for
different pile diameters and different sand densi-
ties. As a result, the effect of change in stiffness is
taken into account in the analyses. Table 1 presents
the hypoplastic model parameters of Baskarp sand
used in the analyses. The soil state (stress and den-
sity) around the pile is altered as a result of the dis-
placement of the soil dictated by the pile jacking
operations. At the same time, the stiffness of the
soil changes. Hence, these three major installation
effects should be assessed in order to quantify the
new state around the jacked pile. A state dependent
Figure 2. a) Sketch of the geotechnical problem
modelled b) General view of the FE model.
Table 1. Hypoplastic soil model parameters for Baskarp
sand (Anaraki, 2008).
ϕc hs
(°) (MPa) n ed0 ec0 ei0 a β
30 4000 0.42 0.548 0.929 1.080 0.12 0.96
constitutive model such as hypoplasticity, reduces All length scales in the model were also
the number of unknowns since the stiffness is normalised with the size of pile diameter, D. This
defined in terms of stress and void ratio. Therefore, normalisation enables a more objective evaluation
to assess the installation effects for each case ana- of the effects of pile diameter and length on the
lysed, the Cartesian stresses at the end of equalisa- Rrr, Rzz, Rθθ, and Rrz and Re values.
tion, σ rr′ f , σ zz
′ f , σ θθ
′ f , and σ rz
′ f and the void ratio, ef
distributions around the jacked pile are normalised
with the corresponding K0 states as: 3 APPROXIMATION OF INSTALLATION
EFFECTS
σ rr′ f σ′f σ′f
Rrr = , Rzz = zz , Rθθ = θθ , In this part of the study the possibility of incor-
p0′ p0′ p0′
porating the installation effects around a wished-
σ rz
′f ef in-place pile to account for the installation effects
Rrz = , Re = (1)
p0′ eK 0 without simulating the whole penetration process
is investigated. A practical way to represent the
For the K0 state, installation effects is using model functions for
each variable and applying a regression analysis to
σ rr′ ,0 σ rr′ ,0 determine the coefficients of the model functions.
Rrr,0 = = Once having the fit, less data, i.e. only the fitting
p0′ 1
( rr,,
rr zz,,0
zz θθ ,0 ) coefficients, are required to represent a continuous
3 field of installation effects.
3K 0
= (2) Based on the preliminary study the following
1 2K0 model functions, in which the length scales are also
normalised by the pile diameter D, are proposed
Similarly, for the rest of the normalised Carte- for the representation of each normalised Carte-
sian stresses the following can be obtained: sian stress and void ratio field:

3 3K 0 ∧
n
Rzz ,0 = , Rθθ ,0 = (3) ψ m = Rm, + ∑ aim ⋅ e − ρ im (5)
1 2K0 1 2K0
i =1

Since σ rz ,0 = 0,
0 and e eK 0 ,
Rrz,0 = 0, Re,0 = 1
Since the initial (K0) state is known, one can
determine the altered state by multiplying the
normalised values by mean effective stress at
K0 state ( p0′ ). Based on the new stress and void
ratio, the hypoplastic model spontaneously alters
the stiffness to a consistent level as a result of its
efficient stiffness formulation.
Figure 3. Description of the surface plots.
and,
(4)
ρim = biim (r / D riim )2 + cim ( z / D + L / D zim )2 (6)
where m (m = 1, 2,…, 5) represents the variable
considered and n represents the number of expo-
nential terms used for that variable; such that
m = 1 represents the radial, m = 2, the vertical,
m = 3, the tangential, m = 4, the shear stress com-
ponents and m = 5 represents the void ratio. The
approximation functions have n = 3, 4, 3, 6, and 4
exponential terms, respectively. Rm,0 is the constant
term calculated for the K0 state of corresponding
variable (see Equations(2)–(4)). aim, bim and cim are
the fitting coefficients representing the magnitude
and spread of the exponential forms and rim and
zim are the fitting coefficients representing the loca-
tion of the Gaussian forms of corresponding vari-
able m. For the sake of compatibility for different
L/D analyses, the vertical position of the Gaussian
form is given relative to the location of pile tip (e.g.
the term in the exponent cim(z/D + L/D + zim)2).
The PR results of each case have a different dis-
cretisation level. Most of the effects are pronounced
near the pile, as expected. To have a more reliable
correction factor and a consistent distribution of
the data points, the PR field data were interpolated

106
onto a uniform grid. The grid spacing is smaller
near the pile, in the influence region, where the
installation field shows clear deviation from the K0
state, and larger otherwise. As a result of using the
new grid the regression analysis is preconditioned
to fit better in the influence region. Furthermore, in
order to eliminate the spurious concentrations and
to obtain the underlying functions that describe
the installation fields, median filtering was applied
on the interpolated PR analysis results.
The model functions have in common two spatial
variables r and z, one constant K0 and five fitting
coefficients aim, bim, cim, rim and zim. Since the model
functions are not linear in r and z, the resulting
regression analysis applied on the surface data is
a bivariate non-linear regression. Figures 4–8 give
a comparison of the approximate surface and the
∧
PR simulation results (data points). Figure 4. Fitting results of normalised radial stresses ψ rr.
After obtaining the fitting coefficients and
substituting the normalised forms back into
Equations (1)–(5), the following forms can be
obtained for the Cartesian stresses, σ rr′ , σ zz ′ , σ θθ
′
and σ rz′ and the void ratio, e:

⎡ 3 ⎤
σ rr′ = p0′ ⋅ ⎢Rrrr,0 + ∑ ai1 ⋅ e − ρi 1 ⎥ (7)
⎣ i =1 ⎦
⎡ 4 ⎤
′ = p0′ ⋅ ⎢Rzzz ,0 + ∑ ai 2 ⋅ e − ρi 2 ⎥
σ zz (8)
⎣ i =1 ⎦
⎡ 3 ⎤
′ = p0′ ⋅ ⎢Rθθθ ,0 + ∑ ai 3 ⋅ e − ρi 3 ⎥
σ θθ (9)
⎣ i =1 ⎦
⎡ 6 ⎤
′ = p0′ ⋅ ⎢Rrrz ,0 + ∑ ai 4 ⋅ e − ρi 4 ⎥
σ rz (10)
⎣ i =1 ⎦
∧
Figure 5. Fitting results of normalised vertical stresses ψzz.
⎡ 4 ⎤
e eK 0 ⋅ ⎢1 + ∑ ai 5 ⋅ e − ρi 5 ⎥ (11)
⎣ i =1 ⎦

Considering the hypoplastic relation of (Bauer

1996) Equation (11) can also be formulated as:
n
⎛ 3′ p0 ⎞
−⎜
⎝ hs ⎟⎠ ⎡ 4
−ρ 5 ⎤
e e0 ⋅ e ⎢1 + ∑ ai 5 ⋅ e i ⎥ (12)
⎣ i =1 ⎦

where, e0 is the void ratio at zero stress level, hs and

n are hypoplasticity model parameters representing
granulate hardness and exponent of the logarithmic
decay function. Figures 9–13 compare the surface
fits of the cases evaluated back on the reference FE
mesh with the PR results.
Alternatively, the accuracy of fits can be checked
by comparing the load—displacement behaviour
of the pile after the installation effects are imposed Figure 6.∧
Fitting results of normalised tangential
stresses ψθθ .

107
 Figure 10. Comparison of the surface fit of σ ′zz with
∧
Figure 7. Fitting results of normalised shear stresses ψ rz. PR data.

Figure 11. Comparison of the surface fit of σθθ

′ with
∧ PR data.
Figure 8. Fitting results of normalised void ratio ψ e .

Figure 9. Comparison of the surface fit of σ ′rr with Figure 12. Comparison of the surface fit of σ ′rz with
PR data. PR data.

108

Figure 13. Comparison of the surface fit of e with PR
data.
Figure 14. Comparison of the calculated pile load test
curves of∧ the imposed installation effects using the fitting
results Ψ with PR result and K0 state.
using the forms given in Equations (7)–(11). The
load tests were performed on the equalised state
of the imposed stress and void ratio to compare
the force displacement curves with the reference
curves obtained from load tests performed on
the directly interpolated PR simulations results.
Figure 14 shows that the imposed states using
the proposed functions of stress and void ratio
underpredict the PR results. It can be seen that the
load displacement curve obtained from the case in
which the proposed formulas were employed yield
an improved response compared to a situation
with simply K0 stress state. As expected, the curve
109
obtained using the formulas was on the conserva-
tive side of the curve obtained from the PR results.
An improvement factor, η at any displacement
level can be defined using the total reaction force,
Ft of each case as:
FtPR − FtK 0 (13)
η= ∧
FTΨ
Using Eqn. (13), a resonable improvement factor
of approximately 0.65 was found at u− y 0.10 D.
4 DISCUSSION OF RESULTS
AND CONCLUSIONS
In this study, first the installation effects in sand due
to pile jacking were investigated. In order to quantify
the installation effects a simplified FE technique—
the Press-Replace technique—was employed. The
change in stress and density at the final equalisa-
tion states was considered as the main installation
effects. In view of the results obtained from FE
simulations, the possibilities of incorporating the
installation effects were investigated. An approxi-
mation of the installation effects, which allows flex-
ibility for imposing the installation field around a
wished-in-place pile to model it as a jacked pile, has
been presented.
The Cartesian stress and void ratio fields result-
ing from pile jacking simulations by the PR tech-
nique, have been approximated by model functions
of Gaussian forms. The fitting results have shown
that the approximated surfaces successfully rep-
resented the stress and void ratio distributions
obtained from the PR simulations. A numerical load
test was performed by imposing the approximated
installation effects around a wished-in-place pile.
A comparable, but conservative load-displacement
curve was obtained. Nevertheless, an improve-
ment of final capacity more than 65% compared
to the load displacement curve of the pile load
test starting from the K0 state has been obtained.
One of the reasons for obtaining reduced capacity
of the imposed states is the approximation error
introduced during the regression analyses. More
importantly, the surface peaks were smoothened
by the median filtering. The approximations also
introduce unbalanced stresses, which are equalised
by a NiL step performed after imposing the instal-
lation effects.
As a next step, the stress and void ratio fields
of the geometry and density variations can be rep-
resented in a unique form. Obtaining the unique
form requires a second regression over the fitting
results of each variation.
REFERENCES
Baligh, M.M. 1975. Theory of deep static cone penetra-
tion resistance. Technical report, MIT Dept. of Civil
Eng.
Bauer, E. 1996. Calibration of a comprehensive hypo-
plastic model for granular materials. Japanese Geo-
technical Society 36(1), 13-26.
Broere, W. & van Tol, A. 2006. Modelling the bearing
capacity of displacement piles in sand. In Proceedings
of the ICE—Geotechnical Engineering, Volume 159,
pp. 195–206.
Chow, F.C. 1996. Investigations into the behaviour of dis-
placement piles for offshore foundations. Ph.D. thesis.
Dijkstra, J. 2009. On the Modelling of Pile Installation.
Ph.D. thesis.
Einav, I. & Randolph, M.F. 2005. Combining upper
bound and strain path methods for evaluating pen-
etration resistance. Int.J. Num. Meth. Engng. 63(14),
1991–2016.
Engin, H.K., B.R.B.J. & van Tol, A.F. 2011. Numerical
analysis of installation effects of pile jacking in sand.
In G. Pietruszczak, S. & Pande (Ed.), In proceedings
of International Symposium on Computational Geome-
chanics, ComGeo-II, pp. 744–755.
Lehane, B. 1992. Experimental investigations of pile
behaviour using instrumented field piles. Ph.D. thesis.
Liu, W. 2010. Axisymmetric centrifuge modelling of deep
penetration in sand. Ph.D. thesis.
110
Mahutka, K.P., Kőnig, F., & Grabe, J. 2006. Numeri-
cal modelling of pile jacking, driving and vibratory
driving. In Proceedings of International Conference
on Numerical Simulations of Construction Processes
in Geotech-nical Engineering for Urban Environment
(NSC06), pp. 235–246.
Pham, H.D., Engin, H.K., Brinkgreve, R.B.J. & van Tol, A.F.
2010. Modelling of installation effects of driven piles
using hypoplasticity. In T. Benz and S. Nordal (Eds.),
Numerical Methods in Geotechnical Engineering,
Numerical Methods in Geotechnical Engineering 2010:
Proceedings of the Seventh European Conference on
Numerical Methods in Geotechnical Engineering,
pp. 261–266. Taylor & Francis.
Said, I., De Gennaro, V., & Frank, R. 2008. Axisymmet-
ric finite element analysis of pile loading tests. Com-
puters and Geotechnics 36(1–2), 6–19.
Teh, C.I. & Houlsby, G.T. 1991. An analytical study of
the cone penetration test in clay. Géotechnique 41(1),
17–34.
van Langen, H. 1991. Numerical Analysis of Soil-Struc-
ture Interaction. Ph.D. thesis.
von Wolffersdorff, P.A. 1996. A hypoplastic relation for
granular materials with a predefined limit state sur-
face. Mechanics of Cohesive-frictional Materials 1(3),
251–271.
White, D.J. 2002. An investigation into the behaviour of
pressed-in piles. Ph.D. thesis.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Analytical and laboratory study of soil disturbance caused

by mandrel driven prefabricated vertical drains

A. Ghandeharioon
Department of Civil Engineering, Faculty of Science and Engineering, Laval University,
Quebec, Canada

ABSTRACT: Analytical investigations and large-scale laboratory experiments were conducted to study the
soil disturbance due to the installation of mandrel-driven Prefabricated Vertical Drains (PVDs) in soft saturated
clays. Considering the mandrels commonly used for installing PVDs in the field, an elliptical Cavity Expan-
sion Theory (CET) was formulated to analyze the shear strain and pore pressure developed in soil during the
mandrel installation. The elliptical CET was developed using modified Cam clay parameters for the undrained
analysis of PVDs installed in soft soil deposits. This formulation identifies a critical zone, a plastic zone, and an
elastic zone around mandrel-driven prefabricated vertical drains. The large-scale laboratory tests consider the
effects of in-situ stresses using a specially designed consolidometer, and a rate controlled installation machine.
The pore water pressure was measured at various locations during the installation of a PVD and withdrawal
of the mandrel. The analytically predicted pore pressures agreed with the measurements in the laboratory. The
results of moisture content tests were also analyzed to verify the concept of an elliptical smear zone around
drains, and to derive a relationship between in-situ effective stresses and the extent of the smear zone.

1 INTRODUCTION In this paper the installation of mandrel-driven

Soil stabilization is an effective method to improve
the engineering properties of soil clay lands and
to alleviate unacceptable differential settlements.
When the soft layer is thick, use of Prefabricated
Vertical Drains (PVDs) combined with pre-load-
ing is an economically viable approach for improv-
ing ground. This technique increases the shear
strength of soft soils by decreasing the moisture
content, which in turn reduces the void ratio.
An analogy between installing mandrel-driven
PVDs and driving cone/pile led to modeling the
installation procedure using Cavity Expansion
Theory (CET). While this theory deals with the
changes in stresses, displacements and pore pres-
sures resulted from the expansion of cavities in
geomaterials, it considers the elastic and plastic
deformations during the penetration process, and
the effect of penetration on the initial state of
stress (Yu and Mitchell 1996). Following Gibson
and Anderson (1961), the CET has been adopted
as the most significant method for interpreting self-
boring pressuremeter tests. As highlighted by Vesic
(1972), and Yu and Houlsby (1991) solutions of the
limit pressure of spherical and cylindrical cavities
can be used to predict the end bearing and shaft
capacities of piles in soil and rock. In addition,
stability and serviceability of tunnels and under-
ground excavations have been analyzed by CET.
PVDs and resulting disturbance in soft saturated
clays are analyzed with a new elliptical CET. The
total and effective stresses and excess pore water
pressure in the soils surrounding the mandrel are
predicted. The theoretical variations of excess pore
pressure are then validated with the measurements
in large-scale laboratory tests. Laboratory studies
have been previously performed to evaluate driv-
ing vertical drains and its effect on soil disturbance
and consolidation. Bergado et al. (1991) con-
ducted tests by installing a PVD in a transparent
cell made of polyvinyl chloride and filled with a
reconstituted soft Bangkok clay. Hird and Moseley
(2000) measured the loss of head in multi-layered
soils and examined the characteristics of the smear
zone to suggest that the radius of inner smear
region would be 3 times the equivalent radius of
the drain. Sharma and Xiao (2000) performed
large-scale laboratory experiments and found that
the smear zone was about 4 times the equivalent
radius of the drain. The laboratory experiments in
this research incorporate a novel machine for driv-
ing mandrels at rates in the range of usual prac-
tice and a large-scale consolidometer specifically
manufactured for this purpose to examine the dis-
turbance in soil and the excess pore pressure while
installing a PVD. At the end, the results of mois-
ture content tests were used to assess the extent of
the smear zone around the PVD.

111
2 ELLIPTICAL CAVITY EXPANSION In terms of effective stresses, Equations (1)
THEORY become:

2.1 Assumptions and definition of the problem ∂ Δu ∂ Δσ r′ 1 ∂ Δ rθ Δσ r′ − Δσ θ′

+ + + +S =0 (2a)
It is assumed that driving a mandrel (with a common ∂r ∂r r ∂θ r
rectangular or rhomboidal cross section) into soil
creates an elliptical cavity, which expands concentri- 1 ∂ u 1 ∂ Δσ θ′ ∂ Δτ rθ Δτ θ
+ + +2 0 (2b)
cally along the horizontal plane. It is also assumed r ∂θ r ∂θ ∂r r
that the soil is homogeneous and that interface
between the mandrel and saturated soft clay is fric- where, Δu is the excess pore water pressure.
tionless. The shape of the anchor is not included in The strain-displacement relationships in the
this analysis, and the mandrel penetration is consid- polar coordinates are:
ered as undrained. Figure 1 illustrates a planar view
of an elliptical cavity (in polar coordinates) with an ∂U
initial semimajor axis of a0 and a semiminor axis of εr = (3a)
∂r
b0. σ0 is the initial internal pressure of the cavity, and
also the uniform pressure acting on the soil bound-
U ∂V
aries at infinity. As the mandrel is driven into the εθ = + (3b)
ground the cavity pressure increases to Pi and the r r ∂θ
axes of cavity expand to a1 and b1, respectively. Con-
sequently, as this elliptical cavity expands the soil is ∂U ∂V V
γ rθ = + − (3c)
assumed to have in-plane displacements. Moreover, r ∂θ ∂r r
it is assumed that Hook’s elasticity law and modi-
fied Cam clay govern, respectively, the elastic and where, εr, εθ and γrθ are the radial, tangential and
plastic behavior of soil. shear strain, respectively. U and V are, respectively,
the components of displacement in a radial and
2.2 Analysis of the elastic state tangential direction.
In view of the undrained plane strain condi-
The equations of equilibrium in polar coordinates,
tion and with reference to Equations (3), the set
expressed as total stresses are:
of Equations (2), in absence of the body forces,
convert into:
∂ Δσ r 1 ∂ Δτ rθ Δσ − Δσ θ
+ + +S =0 (1a)
∂r r ∂θ r
∂ Δu 2G ⎛ ∂ 2U ⎞ 2G ∂U
+ 2 ⎜ 2 +U ⎟ + =0 (4a)
1 ∂ Δσ θ ∂ Δτ rθ Δτ θ ∂r r ⎝ ∂θ ⎠ r ∂r
+ +2 0 (1b)
r ∂θ ∂r r
1 ∂ u 2G ⎛ ∂U ⎞
r ⎝ ∂θ ∫
+ 2 ⎜ + U dθ + f r )⎟
where, Δσr, Δσθ and Δτrθ are variations in radial, r ∂θ ⎠
tangential, and shear stresses, respectively. S and T
are radial and tangential components of body force. ∂ 2U ∂3U
− 8G ∫ d 2Gr ∫ dθ
∂r 2 ∂r 3
d 2 f 2G ⎛ ∂U df ⎞
r ⎝ ∫ ∂r
−2G − ⎜ dθ + ⎟ = 0 (4b)
dr 2 dr ⎠

This system of differential equations is solved

by considering the applicable boundary conditions
and the loading pattern in the developed cavity to
yield (Ghandeharioon 2010):

⎛ P − σ 0 ⎞ ⎛ r12 ⎞
U =⎜ i ⎜ ⎟, V τ rθ 0, u=0 (5)
⎝ 2 G ⎟⎠ ⎝ r ⎠

2
⎛r ⎞
Figure 1. Expansion of an elliptical cavity in an infinite Δσ r Δσ r′ = − Δσ θ′ Δσ θ = − ( Pi − σ ) ⎜ 1 ⎟ (6)
⎝ r⎠
soft saturated clay, shown in polar coordinates.

112
where, r is the position of a soil element measured where, σrp is the total radial stress at the elastic-
from the center of the cavity and r1 is the radius of plastic boundary.
an elliptical cavity. G is the shear modulus of soil. As shown by Cao et al. (2001), a closed form
The pressure-expansion relationship in the elas- solution of the total stresses may be derived by
tic phase is developed as: taking the deviator stress in the plastic zone equal
to the deviator stress at failure (qf). By incorpo-
2G rating the logarithmic strains to justify the large-
r1 = r (7)
2G (Pi − 0 )0 strain effects, the internal pressure in cavity is
expressed as:
in which, r0 (θ ) a0 b0 b02 (a02 b02 ) 2
θ is the ini-
tial radius of the elliptical cavity. np 1 qf 3G (r − r )
Pi p0 − Mp0′ − ln (12)
3 3 r12 q f
2.3 Analysis of the plastic state
As the mandrel penetrates the soil, internal pres- M ′ ( n p /2 ) λ
in which, q f Mp
sure in the cavity increases and as a result, the devi- According to Ghandeharioon (2010), the rela-
ator stress (q) of every soil particle increases. The tionship between radial distance and its associated
total stress parameters are used in this research to deviator stress can be determined by:
analyze the cavity expansion in the context of criti-
cal state. In the plane strain condition: ⎛ r2 − r2 ⎞
ln ⎜1 − 1 2 0 ⎟
⎝ r ⎠
p ( + )/ q = 3( − )/ 2 (8) κΛ ⎡ η⎞ ⎤
1⎛
r r q
= −2 3 ⎢ζ − tan ⎝ M ⎠ np − 1⎥
1
t
tan
3G υM ⎣ ⎦
The constant deformation in soil establishes a
link between q p′ and the initial state of effective (13)
stress, p0′. Based on Cao et al. (2001), the effective
stress path in the p′ : q plane in modified Cam clay The principle of effective stress denotes that
context under an undrained condition is given by: the excess pore water pressure is the difference
between changes in the states of total and effec-
1 tive stresses:
M ′ n p ( p′ p′ )
−
q Mp Λ −1 (9)
Δu = Δp − Δp' (14)
where, n p pp′y0 /pp0′ and Λ = 1 −κ λ are the isotropic
overconsolidation ratio and plastic volumetric Based on Equations (13) and (12), q and p can
strain ratio, respectively. py′0 is the maximum iso- be calculated at any soil element inside the plas-
tropic preconsolidation stress. κ is the slope of tic region. Equation (11) is then used to derive
elastic swelling line and λ is the slope of the normal the state of total stress at that particular posi-
compression line in the ln p′:υ plane where υ is the tion, while noting that p σ r q/ 3 . Finally,
specific volume. by using Equation (14), the value of excess pore
Using Equation (9), the minimum internal pres- pressure can be determined at the location being
sure required to yield the soil elements adjacent to considered.
the wall of the cavity is determined as:

Pimin p0 − Mp0′ ( n p )/3 (10) 2.4 Analysis of the critical state

Driving the mandrel further will ultimately lead
Once the mandrel is driven in further the pressure to the soil being at critical state at the wall of the
in the cavity increases from the value identified by cavity. Thus, the internal pressure will approach its
Equation (10), and the plastic region extends into maximum value at:
the elastic area. In view of the simulated loading pat-
tern and Equation (5), in absence of the body forces, np − 1
Λ
Mp0′ ⎛ n p ⎞
integrating Equations (1) along a particle path start- Pimax p0 − Mp0′ −
ing at the elastic-plastic boundary results in: 3 3 ⎜⎝ 2 ⎟⎠
⎡ ⎛ 2 ⎞ Λ 3G ⎤
2 r dr ln ⎢ − ⎜ ⎟ ⎥ (15)
σr σ rp − ∫ q
3 rp r
(11) ⎢ ⎝ n p ⎠ Mp0′ ⎥
⎣ ⎦

113
 The deviator stress of a particular soil element porous brass tips were used to monitor changes
can then be related to its radial distance from the in pore pressure during initial consolidation and
center of the cavity by: during installation of the mandrel and its with-

r1 ⎛ q κΛ ⎡ −11 ⎛
η⎞ ⎤⎞
= 1 − exp ⎜ −2 3 ⎢ζ − tan ⎜⎝ M ⎟⎠ + ta np − 1⎥ ⎟
1
(16)
r ⎝ 3G υM ⎣ ⎦⎠

1 (η + )( − − ) and η = q . drawal. The transducers were installed at different

where, ζ = ln positions, as depicted in Figure 3, and were con-
2 (η − )( − + ) p′ nected to a digital data logger.
After assembling the cell, a soft clay deposit was
By incorporating the results of laboratory tests prepared, but because the consolidometer is a large
on soil permeability conducted by Sathananthan
and Indraratna (2006), Hird and Moseley (2000)
and Indraratna and Redana (1998), it is proposed
that the plastic shear strain normalized by the
rigidity index can be adopted to characterize the
disturbed soil surrounding the mandrel-driven
PVDs. Normalized plastic shear strain is as large as
0.86%−1.05% at the boundary of soil that is in crit-
ical state, depending on the properties of the soft
clay. The smear zone, within which the horizontal
coefficient of permeability varies rapidly with the
radial distance from the center of the mandrel,
extends to where this ratio is about 0.10%–0.17%.
The magnitude of normalized plastic shear strain
varies between 0.01% and 0.05% at the bound-
ary of the marginally disturbed zone, that region
where the horizontal permeability changes slightly
with distance. The plastic zone of soil experiences
generation of Δu during the installation of mandrel
and covers the area where the plastic shear strain Figure 2. Large-scale consolidometer cell with the
is larger than zero. Finally, there is an outermost doughnut pressure chamber mounted on top of it.
elastic zone with zero plastic shear strain and no
change in the total pore pressure before and after
the mandrel penetrates the soil. Full details of the
analytical derivations and subsequent proposed
theory are provided by Ghandeharioon (2010).

3 LARGE-SCALE LABORATORY TESTS

3.1 Test apparatus and materials

The large-scale laboratory tests, in this research,
utilize a consolidometer cell that is manufactured
from two cylindrical half sections of 7 mm thick
stainless steel, bolted together along their edges
and then placed on a base. With a flange mounted
on top of the cell, the outside diameter would be
664 mm and the height would be 1070 mm. The
surcharge loading is applied to the surface of the
soil by an air jack compressor system via a cali- Figure 3. Radial positions of the fast response pore
brated regulator to the piston mounted on top of pressure transducers (Ts) in relation to the center of the
the consolidometer, as shown in Figure 2. Fast cell at the horizontal plane 0.26 m below the soil surface
response pore pressure transducers equipped with (unit: mm).

114
apparatus, a sample of undisturbed saturated clay
was deemed to be impractical. Therefore, a com-
mercially available reconstituted clay was used for
the tests. The reconstituted soft clay was assessed to
have a moisture content of 60% (1.1 times its liquid
limit) and a plasticity index of 28%. The PVD and
mandrel used in the tests were 100 mm × 4 mm and
125 mm × 28 mm in cross section, respectively.
3.2 Test procedure
The procedure for filling the consolidometer cell
with the reconstituted clay and applying a sur-
charge loading to the piston mounted onto the top
plate is discussed by Ghandeharioon (2012). An
initial consolidation pressure of 37 kPa was applied
to the top plate before the PVD was installed. Once
initial consolidation was 90% complete, the top
plate was removed and replaced by another plate
specifically designed to enable a surcharge load
to be applied while the PVD was being installed.
A novel machine for driving mandrels at a pre-
desired rate was manufactured to study the instal-
lation of mandrel-driven PVDs. Figure 4 shows
the final setup of the laboratory equipments where
the mandrel-driving machine is mounted onto the
large-scale consolidometer.
Figure 4. Final setup for large-scale laboratory tests
where the mandrel-driving machine is mounted onto the
consolidometer.
115
After the PVD was installed, a consolidation
pressure was applied to the soil deposit. Once con-
solidation reached 90%, 54 samples were cored
from various locations along three different direc-
tions, to determine the extent of the smear zone.
More details of the large-scale laboratory tests are
presented in Ghandeharioon (2012).
4 RESULTS OF LABORATORY TESTS AND
VALIDATION OF DEVELOPED THEORY
The pore water pressure has been measured in this
research at various locations of the consolidometer
while different surcharge loads (20 kPa, 32.5 kPa,
and 50 kPa) were applied to the surface of soil dur-
ing installation of a PVD and withdrawal of the
mandrel. The large-scale tests also made it possible
to investigate the effect of in-situ effective stresses
on the size of the smear zone. The variations of
pore water pressure measured in the laboratory at
different locations in the consolidometer (Fig. 3),
while the mandrel was being driven into the soil
and then withdrawn, are shown in Figure 5 for
a surcharge pressure of 50 kPa. The initial pore
water pressure recorded by pore pressure trans-
ducers T1–T6 was 22.6 kPa, 34.9 kPa and 52.2 kPa
Figure 5. Variations of excess pore pressure measured
in the laboratory during installation of a PVD and with-
drawal of the mandrel by the transducers identified in
Figure 3 (surcharge loading = 50 kPa).
for tests conducted under surcharge pressure of
20 kPa, 32.5 kPa and 50 kPa, respectively.
The changes in excess pore water pressure show
that the maximum pore pressure at any given level
develops owing to the compression ahead of the
mandrel during the installation process just before
the base of the mandrel passes the observation
level. When the base of the mandrel passes the
observation level, the vertical compression changes
because of a shearing force developed at the inter-
face of the mandrel and soil, and consequently
the pore pressure drops off until the PVD is fully
installed. Once the mandrel is started to be with-
drawn the pore pressure drops due to the sudden
decrease in mean stress.
As soon as the base of the mandrel passes from
the observation level during withdrawal, the pore
pressure experiences a sudden decrease attrib-
utable to the vacuum created by passing of the
mandrel. The residual excess pore pressure that
exists once the mandrel has been completely with-
drawn will drain over time as a result of radial and
vertical consolidation. These observations were
also used to validate the numerical simulation in
Ghandeharioon et al. (2012). These variations of
pore water pressure reveal that when the vertical
drain was fully installed the excess pore pressure
that built up decreased as the radial distance to the
PVD increased, but when the mandrel was com-
pletely withdrawn the residual pore pressure was
larger further away from the PVD.
Distribution patterns for excess pore pressure
with the radial distance from the center of the
cavity predicted with the developed elliptical CET
and conventional cylindrical CET, and measured
in the laboratory are compared in Figure 6. It is
evident that the developed elliptical CET evaluates
the pore pressure during mandrel installation more
accurately than the conventional cylindrical CET.
Figure 6. Distribution patterns predicted for excess
pore pressure with radial distance using the elliptical CET
and cylindrical CET along the major axis of the mandrel
0.26 m below the soil surface, and measured when base of
the drain shoe passed that horizontal plane, with a sur-
charge loading = 50 kPa.
116
Figure 7. Variations of the moisture content of
soil measured in the laboratory along the 0°, 45° and
90° axes in relation to the installed PVD (surcharge
loading = 50 kPa, consolidation pressure = 80 kPa).
As discussed by Ghandeharioon (2010), install-
ing mandrel-driven PVDs affects both the hori-
zontal coefficient of permeability and the moisture
content of the soil adjacent to the drain. Variations
of moisture content are investigated in this research
to determine the extent of smear zone, and also to
evaluate the effect of in-situ effective stresses on
the size of the smear zone. Figure 7 illustrates the
results of laboratory measurements of the mois-
ture content of 54 samples cored from various
locations along three different directions in the test
corresponding to a surcharge pressure of 50 kPa,
after consolidation was 90% complete. r, rm(θ) and
rsmear are the radial position measured from the
center of the cell, equiv alent radius of the mandrel
and radius of the smear zone, respectively.
Based on the variation of the measured moisture
contents and by implementing the concept devel-
oped by Sathananthan and Indraratna (2006), it is
found that the radius of the smear zone normalized
by the equivalent radius of the mandrel is 2.55 in
that test. The normalized radius of the smear zone
is 2.65 and 2.85 for tests where the surcharge load-
ing applied to the surface of soil during installa-
tion was 32.5 kPa and 20 kPa, respectively. These
results verify the concept of an elliptical smear
zone identified by Ghandeharioon et al. (2010) for
the cases where drains were driven by rectangular
mandrels. Furthermore, these results indicate that
for a given type of soft soil subjected to a particular
rate of mandrel installation, the size of smear zone
decreases as the in-situ effective stresses increase.
5 CONCLUSIONS
The disturbance and pore water pressure in soft sat-
urated clays were analyzed during the installation
of mandrel-driven prefabricated vertical drains.
A new elliptical cavity expansion theory for an
undrained condition was developed by incorpo-
rating the modified Cam clay parameters to study
the installation of PVDs in soft clay deposits. As
a mandrel is driven into the soil, the internal pres-
sure in the cavity that is expanding, and the cor-
responding stresses and excess pore pressure in the
soft soil were predicted. The plastic shear strain
normalized by the rigidity index of the soil was
used to characterize the disturbed soil surround-
ing the mandrel-driven PVDs. Based on this con-
cept, the boundaries of critical zone, smear zone,
marginally disturbed zone, plastic zone, and elastic
zone were identified.
In this research the installation of PVDs was
studied in the laboratory under the field conditions
using a large-scale consolidometer and a mandrel-
driving machine capable of working at realistic
rates. Variations of pore water pressure during the
installation of a PVD and withdrawal of the man-
drel were monitored by pore pressure transducers.
Subsequently, the pattern of distribution calculated
for excess pore pressure by the developed theory
was verified by the measurements performed in the
laboratory.
Using the results of moisture content tests on
samples, which in relation to the installed PVD
were cored from various locations and directions,
the extent of smear zone was measured in the
large-scale consolidometer in each test performed
under a different surcharge loading. The results
verified the concept of an elliptical smear zone
around drains that was identified in previous stud-
ies for cases where PVDs were driven by common
field mandrels. These results also showed that for
a given type of soft soil subjected to a particu-
lar rate of installation, the extent of smear zone
around the mandrel decreases as the in-situ effec-
tive stresses increase.
ACKNOWLEDGEMENTS
This research was supported by a scholarship from
the Cooperative Research Center (CRC Australia)
for Railway Innovation awarded to the author for
his Ph.D. studies at University of Wollongong,
Australia. Their support is acknowledged.
REFERENCES
Bergado, D.T., Asakami, H., Alfaro, M.C., &
Balasubramaniam, A.S. 1991. Smear effects of
vertical drains on soft Bangkok clay, Journal
of Geotechnical Engineering, ASCE, 117(10):
1509–1530.
117
Cao, L.F., Teh, C.I., & Chang, M.F. (2001). Undrained
cavity expansion in modified Cam clay I: Theoretical
analysis, Géotechnique, 51(4): 232–334.
Ghandeharioon, A. 2010. Analytical and numerical study
of soil disturbance associated with the installation of
mandrel-driven prefabricated vertical drains, Ph.D.
Thesis, School of Civil, Mining & Environmen-
tal Engineering, Faculty of Engineering, Univ. of
Wollongong, Wollongong, Australia.
Ghandeharioon, A. 2012. Large-scale laboratory assess-
ment of smear effects in soft soils stabilized by prefab-
ricated vertical drains. Proceedings of 65th Canadian
Geotechnical Conference: GeoManitoba—Building on
the past, Canadian Geotechnical Society, Canada.
Ghandeharioon, A., Indraratna, B., & Rujikiatkamjorn, C.
2010. Analysis of soil disturbance associated with man-
drel-driven prefabricated vertical drains using an ellip-
tical cavity expansion theory, International Journal of
Geomechanics, 10(2): 53–64.
Ghandeharioon, A., Indraratna, B., & Rujikiatkamjorn, C.
2012. Laboratory and finite element investigation of
soil disturbance associated with the installation of
mandrel-driven prefabricated vertical drains, Journal
of Geotechnical and Geoenvironmental Engineering,
138(3): 295–308.
Gibson, R.E., & Anderson, W.F. 1961. In-situ measure-
ment of soil properties with the pressuremeter. Civil
Engineering Public Works Reviews, 56, 615–618.
Hird, C.C., & Moseley, V.J. 2000. Model study of seepage
in smear zones around vertical drains in layered soil,
Géotechnique, 50(1): 89–97.
Indraratna, B., & Redana, I.W. 1998. Laboratory deter-
mination of smear zone due to vertical drain instal-
lation. Journal of Geotechnical and Geoenvironmental
Engineering, 124(2): 180–184.
Sathananthan, I., & Indraratna, B. 2006. Laboratory
evaluation of smear zone and correlation between
permeability and moisture content, Journal of Geo-
technical and Geoenvironmental Engineering, 132(7):
942–945.
Sharma, J.S., & Xiao, D. 2000. Characterisation of
a smear zone around vertical drains by large-scale
laboratory tests, Canadian Geotechnical Journal, 37:
1265–1271.
Vesic, A.S. 1972. Expansion of cavities in infinite soil
mass. Journal of the Soil Mechanics and Foundations
Division, ASCE, 98, 265–290.
Yu, H.S., & Houlsby, G.T. 1991. Finite cavity expansion
in dilatant soil: Loading analysis. Geotechnique, 41,
173–183.
Yu, H.S., & Mitchell, J.K. 1996. Analysis of cone resist-
ance: A review of methods. The University of Newcas-
tle, Australia, Report No. 142.09.1996.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

CEL: Simulations for soil plugging, screwed pile installation

and deep vibration compaction

J. Grabe, S. Henke, T. Pucker & T. Hamann

Institute of Geotechnical and Construction Engineering, Hamburg University of Technology, Hamburg, Germany

ABSTRACT: In many geotechnical applications like pile installation processes or soil improvement
large deformations of the surrounding soil occur. These large deformations in combination with the com-
plex material behaviour of the soil lead to numerical boundary value problems which are often difficult to
solve. These difficulties are related to large mesh distortions and numerical problems due to complex con-
tact conditions. One possibility to overcome these difficulties is to use special numerical techniques which
are especially invented for large deformation simulations. In this work, the Coupled Eulerian-Lagrangian
method (CEL) is identified to be well suited for such boundary value problems involving large distortions
of the surrounding soil. Therefore, the numerical technique is presented and its suitability is shown in dif-
ferent geotechnical applications. The screwed pile installation and the deep vibration compaction process
are investigated using the CEL method.

1 INTRODUCTION process is investigated in a numerical study. The

In many geotechnical applications like pile instal-
lation processes or soil improvement large defor-
mations of the surrounding soil occur. These large
deformations in combination with the complex
material behaviour of the soil lead to numerical
boundary value problems which are often difficult
to solve. These difficulties are related to large mesh
distortions and numerical problems due to com-
plex contact conditions.
One possibility to overcome these difficulties
is to use special numerical techniques which are
especially developed for large deformation simula-
tions. The Coupled Eulerian-Lagrangian method
(CEL) is used to model the following geotechnical
applications:
1. Screwed pile installation: The whole installation
process of screwed piles with variation of the
main parameters influencing the penetration
Figure 1. Deformation of a continuum in a Lagrangian
(left) and an Eulerian analysis (right).
results are discussed regarding the influence of
the screwed pile installation on the surrounding
soil.
2. Deep vibration compaction: In this study the
main parameters frequency and vibration force
are varied to study their influence on the com-
paction success.
2 COUPLED EULERIAN-LAGRANGIAN
APPROACH
2.1 Eulerian and Lagrangian formulation
There are two alternatives to describe the move-
ment of a small volumetric element as a function
of time (Fig. 1): the Lagrangian approach and the
Eulerian approach.
1. Lagrangian approach: The movement of the
continuum is specified as a function of the
material coordinates and time. This is the tra-
ditional approach used in conventional small
strain finite element analyses. The nodes of
the Lagrangian mesh move with the material
as it deforms. The interface between two parts
is accurately tracked. Large deformations may
lead to severe mesh distortions and pose limita-
tions to this approach.
2. Eulerian approach: The movement of the con-
tinuum is specified as a function of the spatial
coordinate and time. This approach is often used
in fluid mechanics. An Eulerian reference mesh,

118
 which remains stationary enables the motion of
the material to be traced. Materials can move
freely through an Eulerian mesh, which remains
undeformed.
2.2 Coupled Eulerian-Lagrangian approach
The Coupled Eulerian-Lagrangian (CEL) method
aims at capturing the advantages both of the
Lagrangian and the Eulerian approach. The CEL-
Method allows an interaction between Lagrangian
elements and Eulerian material by using an
Eulerian-Lagrangian contact formulation. The
Eulerian time integration is realized applying the
“Lagrange-plus-remap” formulation. First of all, a
traditional Lagrangian phase is calculated in each
time increment and the nodes of the Eulerian mesh
are temporarily fixed within the material. Therefore
the Eulerian elements can deform temporarily
in the Lagrangian phase. Second, in the Eulerian
phase, the so called transport phase, elements are
tested for significant deformation. These elements
are automatically remapped and the material flow
through these elements is calculated. The timestep
has to be sufficiently small, so no element distor-
tions occur during one timestep in the Lagrangian
phase. In the analyses the movement of the Eulerian
material through the mesh is tracked by comput-
ing its Eulerian Volume Fraction (EVF). Each
Eulerian element is designated a percentage, which
represents the portion of that element filled with a
material. If an Eulerian element is completely filled
with a material, its EVF is 1; if there is no material
in the element, its EVF is 0. This approach is avail-
able in Abaqus (Dassault Systèmes, 2010).
The application of the CEL-Method on geo-
mechanical problems involving large deformations
has been approved e.g. by Qiu et al. (2009, 2010),
Qiu & Grabe (2011), Henke et al. (2010), Bienen
et al. (2011), Pucker & Grabe (2012), Pichler et al.
(2012).
2.3 Contact formulation
Contact between Eulerian and Lagrangian struc-
tures is enforced using a general contact formu-
lation that is based on a penalty method. The
algorithm does not enforce contact between the
Lagrangian elements and the Eulerian elements.
The Lagrangian elements can move through
the Eulerian mesh without resistance until they
encounter an Eulerian element filled with mate-
rial (EVF ≠ 0), see Figure 2. The penalty contact
method is less strict compared to the kinematic con-
tact method used in the Lagrangian approach. It
approximates hard pressure-overclosure behavior.
This method allows small penetration of the
Eulerian material into the Lagrangian domain.
119
Figure 2. Schematic illustration of the contact between
a Lagrangian and a Eulerian part.
2.4 Hypoplastic constitutive model
The choice of an appropriate constitutive model is
a challenge, because of the complex behaviour of
soils due to large deformation, i.e. critical state or
soil softening. The hypoplastic constitutive model
according to von Wolffersdorff (1996) with the
extension of intergranular strain by Niemunis &
Herle (1997) is used in this paper. The constitutive
model is able to reproduce the non linear and inelas-
tic behaviour of soils realistically. Specific properties
of granular soils are considered, including dilatancy,
different stiffnesses for loading and unloading
paths, barotropy and pycnotropy. The extension of
intergranular strain allows the simulation of effects
such as accumulation of deformation and hystereti-
cal behaviour under cyclic loading. The constitutive
model is rate independent in a rate-type formulation
defined by the tensorial function:
Ο
M (T ):D (1)
Ο
In this equation T is the objective Jaumann stress
rate, D the strain rate and M a fourth order tensor,
which depends on the actual Cauchy stress T, the
void ratio e and the intergranular strain δ. One
advantage of the constitutive model is the consid-
eration of the void ratio. Thus, compaction proc-
esses can be described accurately.
3 SIMULATION OF SCREWED PILE
INSTALLATION
The installation process of piles influences the
surrounding soil depending on the installation
method. The piles can be driven, jacked, vibrated
or drilled into the ground. The effects inside the
soil are still not completely investigated. There-
fore, most methods to predict the pile behaviour
or its bearing capacity are based on empirical
approaches.
The influences of the installation process on
the surrounding soil have been investigated for
driven, jacked and vibrated piles by Mahutka
(2007), Henke (2008), Henke & Grabe (2009) in
numerical studies. In this section, the numerical
simulation of the installation process of drilled
piles is presented. Using this simulation, new
insight can be gained in understanding effects
in the soil nearby installed piles. The installation
of a full displacement pile with lost bit (FDP) is
simulated. Full displacement piles push the soil
sideways without excavating soil. The installation
process is divided into six steps schematically pre-
sented in Figure 3.
In the first step the drilling tool is taken into
position. During the second and the third step, the
drilling tool is drilled to the purposed depth. Dur-
ing the drilling process, the soil is transported to
the displacement body where the soil is compacted.
The reinforcement basket is installed in step four.
The drilling tool is removed in step five and the pile
is concreted simultaneously. The concrete hardens
in step six and the installation is completed. The
presented numerical study simulates only the steps
two and three. A detail of the drilling tool is illus-
trated in Figure 4.
3.1 Numerical simulation
3.1.1 Modeling
Due to the asymmetric drilling tool and the applied
CEL-Method, a three dimensional model has to be
used. The drilling tool is modeled as a rigid body,
because it is assumed, that the deformations of the
drilling tool are neglectable according to the large
deformations of the soil. The displacement body
Figure 3. Schematic illustration of the installation process of full displacement piles (Busch et al., 2010).
120
has a diameter about 51 cm and the drill rod about
32 cm. The diameter of the auger elements relate
to the displacement body.
The soil is modeled as an Eulerian area. The
shape is cylindrical with a diameter about 16 m
and a height of 20 m. Thereby, the distance
between the boundaries and the drilling tool is
more than 15 D (15 times of the pile diameter
D), so that no influences of the boundary con-
ditions on the results are expected. Above the
soil, a 2 m void area is provided, so that the soil
can avoid into this free space during the drilling
process. The whole Eulerian area is discretized
with approx. 163,000 eight-noded elements with
reduced integration. The discretized model is
shown in Figure 5.
The drilling tool penetrates the soil velocity
controlled. Assuming, that the response of the soil
model does not depend on the velocity, the pene-
tration velocity is taken to 1 m/s. The ratio between
the penetration velocity and the rotation velocity
should be approx. 10. Therefore the rotational
velocity is chosen to 10 U/s. The reaction forces
and moments are recorded during the simulation.
3.1.2 Contact and constitutive model
The general contact algorithm is used. The normal
contact is chosen to hard contact and the tangen-
tial contact uses the Mohr-Coulomb model with a
friction angle about δ = 1/3 φ'.
The hypoplastic constitutive model presented in
section 2.4 is used to simulate the drilling process.
The material is chosen as Mai-Liao sand and its
parameters are given in Table 1.

Figure 4. Scheme of the FDP drilling tool invented by
BAUER (Busch et al., 2010) (left); numerical discretiza-
tion of the drilling tool (right).
3.2 Results
3.2.1 Changes of the stress state
The stress state around the drilling tool is signifi-
cantly influenced, see Figure 6. The transport of
the soil to the displacement body causes a compac-
tion of the soil at the height of the displacement
body. After the displacement body passes through,
a gap between the soil and the drill rod occurs
and the soil can relax. Therefore, the horizontal
stresses, illustrated in Figure 6, decrease nearby the
drilling tool. The massive compaction of the soil
in combination with the changing cross-section of
the drilling tool lead to a reduction of the horizon-
tal stresses nearby the drilling tool.
3.2.2 Changes of the soil’s bulk density
The hypoplastic constitutive model is able to predict
changes of the soil’s bulk density, see section 2.4.
These changes caused by the drilling process are
shown in Figure 7. The influence of two different
velocity ratios between the penetration velocity vz
and the rotation velocity vr are investigated. Nearby
the drilling tool, the soil is loosened up, caused by the
changing cross-section of the drilling tool. Between a
distance of 1 D to 3 D, the soil is strongly densified.

Figure 5. Discretized model with void area and mesh
of the soil in detail.
3.3 Conclusion
The CEL-method is able to reproduce the soil’s
behaviour during a drilling process. The changes
of the stress state and of the bulk density can be
investigated. Therefore, this method can be used
to gain deeper insights into the soil behaviour and
the influencing installation parameters, such as
the penetration and rotation velocity as well as the
geometric properties of the drilling tool.

Table 1. Soil parameters for Mai-Liao sand for the

hypoplastic constitutive model. 4 DEEP VIBRATION COMPACTION
SIMULATION
Parameter Value Description
4.1 Introduction
ϕc 31.5 Critical state friction angle (°)
hs 32.0 Granular hardness (MPa) The deep vibration compaction method is a method
n 0.32 Exponent for deep compaction of granular soils to reduce
ed0 0.75 Minimum void ratio settlements and porosity of the soil and to increase
ei0 1.04 Critical void ratio the vertical bearing capacity of foundations.
ec0 1.20 Maximum void ratio The deep vibration compactor consists of a
α 0.40 Exponent steel tube with a length of 2 to 4 m and a diameter
β 1.00 Exponent between 30 cm and 50 cm. Inside the tube one or
Rmax 1 exp (−4) Maximum value of more rotating masses are located. These masses are
intergranular strain rotating around the vertical axis of the compac-
mR 5.00 Stiffness ratio at 180° change tor such that the compactor is oscillating during
of direction the compaction process. Therefore, the vibration
mT 2.00 Stiffness ratio at 90° change
of direction
energy is directly transferred from the tube into
βR 0.50 Exponent the surrounding soil. The compactor is hinged at a
Χ 6.00 Exponent stay tube with the same diameter which is carried
by a crane for example, see Fellin (2000).

121
Figure 6. Horizontal stresses in the soil before (left) and after the drilling process (right).

Figure 7. Bulk density of the soil along horizontal paths at depths of 1 m, 3 m and 4.5 m after completion of the 6 m
drilling process for different drilling ratios.

The deep vibration compaction process is sche-
matically depicted in Figure 8. First, the vibrator
is installed up to the wished penetration depth
by the help of a drilling fluid as it can be seen in
Figure 8. The penetration is achieved only due to
the self-weight of the vibrator and vibration of the
compactor. It is possible to reach depths of up to
60 m with this technique. Afterwards, the vibrator
is pulled stepwise in combination with compaction
of the surrounding soil. The steps are between 0.3

122
Figure 8. Schematic description of the deep vibration compaction process after “Keller Grundbau”.
and 1.0 m and the vibration time at one penetra-
tion depth lies between 30 and 90 s, see Witt (2009).
The vibration energy leads to shear waves in the
soil such that the grain structure is rearranged.
Therefore, cyclic shearing acts on the soil such
that the void ratio is reduced and the soil is com-
pacted. Due to this compaction soil is transferred
into the compacted region such that settlements at
the ground surface become evident. To cope with
these settlements the funnel is refilled with material
(Fellin, 2000). To compact a large area the com-
paction process is redone in discrete distances such
that a narrow grid can be improved. The vibra-
tion frequencies of typical vibrators lie between 30
and 60 Hz such that the maximum deflection of
the vibrator lies between 3 and 21 mm at the toe.
This comes along with horizontal forces between
150 and 700 kN (Witt, 2009). Fellin (2000) points
out that the soil around the vibrator is compacted
up to a dense packing in distances between 0.6 m
and 1.75 m. In greater distances it is not possible to
receive a dense packed sand.
4.2 Numerical modelling
A three-dimensional coupled Eulerian-Lagrangian
model is used to simulate the deep vibration com-
paction process, see Figure 9. The soil continuum
has a diameter of 30 m and a height of 27 m. At
the beginning of the simulation the first two meters
of the soil continuum are modeled material-free to
123
allow material transport into this region during
the analyses. The vibrator itself has a length of
2.5 m and a circular cross-section with a diameter
of 30 cm. At the beginning of the deep vibration
compaction simulation the vibrator is modeled
wished-in-place at a penetration depth of 10 m.
The vibrator is vertically hinged to a tube with the
same cross-section. The vibrator and the stay tube
are modeled linear elastic with the parameters of
steel.
In all simulations the soil continuum is discretized
using eight-noded-Eulerian elements whereas the
vibrator components are modeled with Lagrang-
ian elements (eight-noded-hexaeder elements with
reduced integration). Contact between vibrator
and soil is considered using Coulomb’s friction
law. As a simplification the contact between soil
and vibrator is considered to be frictionless.
All simulations are done using the hypoplastic
constitutive model with soil parameters for loose
Mai-Liao sand.
As loading history first a K0-stress state is
applied on the soil continuum. Afterwards, the
simulation of the deep-vibration compaction proc-
ess is carried out for a total duration of 3 s. The
vibrator’s movement is discretized both displace-
ment and force-controlled. In the displacement
controlled simulations the toe of the vibrator
oscillates at a certain frequency with a predefined
radius around the centre of gravity of the vibra-
tor whereas in the force-controlled simulations the
Figure 9. Three-dimensional numerical model to simulate the deep-vibration compaction process.

force acting at the toe of the vibrator is predefined Table 2. Parameter combinations investigated regard-
and the frequency is considered to be f = 30 Hz. ing the deep vibration compaction process in sand.
The top point of the vibrator is connected to the
stay tube in all simulations. Vibration Oscillation Vibration
frequency f amplitude uH force Fdyn
All varied parameters in this numerical study Name (Hz) (mm) (kN)
are summarized in Table 2.
f30u4 30 4 –
f30u8 30 8 –
4.3 Results f30u12 30 12 –
During the displacement-controlled simulations f30u16 30 16 –
the movement of the vibrator is forced along a f30u20 30 20 –
circular path around the vibrator’s centre of grav- f40u12 40 12 –
ity. In contrast during the force-controlled simu- f50u12 50 12 –
lation the movement of the vibrator depends on f60u12 60 12 –
f30F150 30 – 150
the acting forces and the soil stiffness such that the
f30F300 30 – 300
deflection does not follow an ideal circular path. f30F500 30 – 500
In Figure 10 the deflection of the vibrator’s toe is f30F700 30 – 700
depicted as a plot in the x-y-plane. This line repre-
sents 3 s of vibration.
At the beginning of the deep vibration com-
paction process the soil is in a loose state. The the soil the stiffness increases and the amplitude of
surrounding soil is very compactable and has a the vibrator therefore is reduced. The drift of the
relatively low stiffness such that after start of the vibrator shown in Figure 10 results out of the local
vibration process large deformation amplitudes differences in soil stiffness due to the compaction
can be reached. With increasing densification of of the sand.

124
Figure 10. Deformation of the vibrator in x-y-plane
during the deep vibration compaction process, force-
controlled simulation, Fdyn = 150 kN, f = 30 Hz.
Figure 12. Normalized force and horizontal displacement in time domain during the deep vibration compaction
process, force-controlled simulation, Fdyn = 150 kN, f = 30 Hz.
Furthermore, a phase shift between the direc-
tion of the vibration force and the direction of
vibrator movement can be seen, see Figure 11.
The vibration force is shifted with an angle ψ to
the normalized horizontal deformation. The angle ψ
depends on the soil damping and the eigenfrequency
of the system. The higher the damping the closer this
angle is to 90° (Fellin, 2000). In the present case for a
vibration force of Fdyn = 150 kN and a frequency of
f = 30 Hz the phase shift is ψ = 113°.
Furthermore, the influence of the main parame-
ters vibration frequency and oscillation amplitude
on the compaction process is of main interest. In
Figure 12 the results of displacement controlled
simulations with varying frequency and oscillation
amplitude respectively are shown.
It is evident that mainly the frequency has an
impact on the void ratio distribution around
the vibrator. It can be seen that with decreasing
125
Figure 11. Dependency of maximum oscillation ampli-
tude and vibration force—Comparison between numeri-
cal simulation and values found in literature.
frequency the area of influence where a dense soil
state is reached increases significantly. In contrast,
the oscillation amplitude, at least in the investi-
gated range does not significantly influence the
void ratio distribution around the vibrator.
To judge the quality of the numerical simula-
tions the calculated deflections of the vibrator in
dependency of the vibration forces are compared
to results found in literature (Fellin, 2000), see
Figure 13.
Regarding this graph, it can be seen that the
numerically received results fit well into the data
given by various companies and literature. Only
the simulation with a vibration force of 700 kN
seems to overestimate the oscillation amplitude.
This is caused due to the high force compared to
the comparatively low soil stiffness which leads to
non-uniform movement of the vibrator which was
not judged to be realistic.
Figure 13. Top: Influence of the vibration frequency on the compaction success, Bottom: Influence of the oscillation
amplitude on the void ratio around the vibrator—displacement-controlled simulations; results along a horizontal path
at the height of the centre.
A more detailed discussion regarding the numer-
ical simulation of deep vibration compaction proc-
esses using the Coupled Eulerian-Lagrangian
method can be found in Henke et al. (2011) and
Henke et al. (2012).
5 CONCLUSION
The simulation of large deformations of soil often
cause numerical boundary value problems in many
geotechnical applications like pile installation
processes or soil improvement. These large defor-
mations in combination with the complex material
behaviour of the soil are often difficult to solve
numerically. The Coupled Eulerian–Lagrangian
method (CEL) is well suited to model complex
geotechnical installation processes, like screwed
pile installation and deep vibration compaction, in
126
dry granular soil material using a hypoplastic con-
stitutive model.
In case of screwed piles the CEL method is used
to simulate the drilling process. The CEL-method
is able to reproduce the soil’s behaviour during this
process. The changes of the stress state and of the
bulk density can be investigated. Therefore, this
method can be used to gain deeper insights into
the soil behaviour and the influencing installation
parameters, such as the penetration and rotation
velocity as well as the geometric properties of the
drilling tool.
Another geotechnical application to simulate
with the CEL method is the deep vibration com-
paction. It can be shown that the process of soil
compaction can be simulated realistically. Effects
such as a phase-shift between the direction of
the vibrator force and the direction of motion
of the vibrator and the size of the area of influ-
ence around the vibrator as observed in reality can
be simulated. Furthermore the numerical model
is used to investigate the influence of the main
parameters vibration frequency and oscillation
amplitude on the compaction process.
At the moment research is done to enhance the
existing constitutive model to 2 or 3-phases such
that also the effect of pore pressure in dynamic
analyses can be taken into account. First results
of this research developing an explicit formula-
tion including pore water pressure are presented in
Pichler et al. (2012).
REFERENCES
Bienen B., Henke S. & Pucker T. 2011. Numerical study of
the bearing behaviour of circular footings penetrating
into sand. Proc. of 13th International Conference of
International Association for Computer Methods and
Advances in Geomechanics (IACMAG): 939–944.
Busch, P., Grabe, J., Gerressen, F.W. & Ulrich, G. 2010.
Use of displacement piles for reinforcement of exist-
ing piles, Proceedings of DFI and EFFC 11th Int.
Conf. in the DFI series, Geotechnical Challenges in
Urban Regeneration in London/UK: 113–119.
Dassault Systèmes 2010. Abaqus User Manual Version
6.10.
Fellin W. 2000. Rütteldruckverdichtung als plastody-
namisches Problem. Dissertation, Institute of Geo-
technics and Tunneling, University of Innsbruck.
Advances in Geotechnical Engineering and Tunneling,
Heft 2, 2000.
Henke, S. 2008. Herstellungseinflüsse aus Pfahlrammung
im Kaimauerbau, Dissertation, Veröffentlichungen
des Instituts für Geotechnik und Baubetrieb der TU
Hamburg-Harburg, Heft 18.
Henke, S., & Grabe, J. 2009. Numerical modeling of pile
installation, Proc. of 17th Int. Conf. on Soil Mechanics
and Foundation Engineering (ICSMFE): 1321–1324.
Henke S., Hamann T. & Grabe J. 2012. Numerische
Untersuchungen zur Bodenverdichtung mittels
Rütteldruckverfahren. 2. Symposium “Baugrund-
verbesserung in der Geotechnik”, TU Wien, zur
Veröffentlichung akzeptiert.
127
Henke S., Hamann T. & Grabe J. 2011. Coupled
Eulerian-Lagrangian Simulation of the Deep
Vibration Compaction Process as a Plastodynamic
Problem. Proc. of EURODYN 2011, Leuven.
Henke, S., Qiu G. & Grabe J. 2010. A coupled eulerian-
lagrangian approach to solve geotechnical problems
involving large deformations. Proc. of 7th European
Conference on Numerical Methods in Geotechni-
cal Engineering (NUMGE) in Trondheim/Norway:
233–238.
Herle, I. 1997. Hypoplastizität und Granulometrie
einfacher Korngerüste. Heft 142 Institut für Bod-
enmechanik und Felsmechanik der Universität
Fridericana in Karlsruhe.
Mahutka, K.-P. 2007. Zur Verdichtung von rolligen
Böden infolge dynamischer Pfahleinbringung und
durch Oberflächenrüttler, Dissertation, Veröffentlic-
hungen des Instituts für Geotechnik und Baubetrieb
der TU Hamburg-Harburg, Heft 15.
Pichler, T., Pucker T., Hamann T., Henke S. & Qiu G.
2012. High-performance abaqus simulations in soil
mechanics reloaded—chances and frontiers. Proc.
of International Simulia Community Conference in
Providense, Rhode Island/USA: 237–266.
Pucker, T. & Grabe J. 2012. Numerical simulation of the
installation process of full displacement piles. Com-
puters and Geotechnics, 45: 93–106, DOI: 10.1016/j.
compgeo.2012.05.006.
Qiu, G. & Grabe J. 2011. Explicit modeling of cone and
strip footing penetration under drained and undrained
conditions using a visco-hypoplastic model. Geotech-
nik 34(3): 205–271.
Qiu, G., Henke S. & Grabe J. 2009. Applications of
coupled eulerian lagrangian method to geotechnical
problems with large deformations. Proc of SIMULIA
Customer Conference 2009 in London: 420–435.
Qiu, G., S. Henke, & J. Grabe 2010. Application of a cou-
pled eulerian-lagrangian approch on geomechanical
problems involving large deformation. Computers and
Geotechnics, DOI:10.1016/j.compgeo.2010.09.002.
vonWolffersdorff, P.-A. 1996. A hypoplastic relation for
granular materials with a predefined limit state sur-
face. Mechanics of Frictional and Cohesive Materials,
1: 251–271.
Witt K.J. 2009. Grundbau-Taschenbuch, Teil 2: Geotech-
nische Verfahren. Berlin, Ernst & Sohn Verlag.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Towards a framework for the prediction of installation rate effects

S. Robinson & M.J. Brown

Division of Civil Engineering, University of Dundee, Dundee, Scotland, UK

ABSTRACT: There is a need for an improved understanding of rate effects over a wide range of strain
rates in order to improve the modelling and analysis of installation effects. Using triaxial testing on recon-
stituted kaolin over a wide strain rate range this paper examines the impact of strain rate on the aspects
of soil response which are important in the analysis of installation effects. It is demonstrated that shear
strength, small strain stiffness and the elastic shear strain threshold are rate dependent. The implications
of this for a common stiffness degradation model are analysed and an improved model proposed.

1 INTRODUCTION (Brown 2009), as it is based on a correlation using

Finite Element Analysis (FEA) is increasingly
being used in the industry to analyse complex geo-
technical activities that cannot be easily dealt with
using hand calculations. This increasing range of
activities means that soil behaviour over a wider
range of strain rates needs to be considered, no
more so than in installation effects, where activi-
ties such as pile driving and testing can induce high
strain rates.
Splitting the geotechnical model into a series
of elements forms the basis of FEA. For each
element, knowing the stress-strain response over
the entire range of strains induced is crucial to
the accuracy of the model, especially in the small
strain domain.
The fact that strain-rate effects can have a sig-
nificant impact on this relationship is well known
(Krieg & Goldscheider 1998). However, in order to
improve the ability to use FEA to analyse instal-
lation effects, there is a need for a better under-
standing of how strain-rate effects vary with strain
level.
This paper considers the strain-rate dependence
of rate effects over a wide range of strain magnitudes
at shear strain rates from 0.333 to 60,000%/hr.
only peak shear strengths. Similarly, only und-
rained testing was used, meaning that the impact
of drainage on the determination of the rate effect
was not dealt with.
Drainage effects are significant as the strain
rate influences whether there is time for drain-
age to occur. This means that rate effects form
a “U-shaped” curve as shown in Figure 1. At
extremely slow rates, there is time for pore pres-
sure developed during shearing to fully dissipate.
As the strain rate increases, the reducing time for
drainage means that increasing pore pressures and
hence lower shear strengths are observed in the par-
tially drained region. As the strain rate is increased
further, the soil response becomes almost fully und-
rained, where pore pressures are at their maximum.
After this point, viscous effects cause the observed
shear strength to increase with strain rate up to a
potential viscous limit, which Chow & Airey (2011)
demonstrated using free falling penetrometers. The
form of this shear strength-strain rate relationship
has been verified by studies using a wide range

1.1 The influence of strain rate

The most commonly used assumption in terms of
rate effects is that of Kulhawy and Mayne (1990)
which states that for every log cycle (tenfold)
increase in strain rate, a ten percent increase in
shear strength is expected. However, this has been
shown to be variable, with the rate effects observed
ranging from 9.5 to 20% (Bea 1982). Another key Figure 1. U-shaped curve showing the idealised vari-
consideration is that this assumption does not con- ation of rate effect with strain rate (Quinn and Brown
sider any variation of rate effects with strain level 2011).

128
of test methods from cone penetrometer testing
to large scale shear boxes (Steenfelt 1993, Lehane
et al 2009).
As previously mentioned, small strain behaviour
is important to FEA. Shibuya et al (1996) showed
that the initial small strain stiffness, Emax, is rate
independent and constant for the range of strain
rates investigated (0.6 to 84%/hr). It was also found
that this linearity continued up to a strain defined
as the elastic strain threshold, εEL, which was itself
rate dependent, increasing with increasing strain
rate. These findings were confirmed by Mukabi &
Tatsuoka (1999).
Lo Presti et al (1996) found that soil stiffness is
rate independent below 0.001% shear strain, and
increases with strain rate beyond this point. To
allow comparison of relative stiffnesses at differ-
ent strains and strain rates, Lo Presti et al (1996)
defined the strain rate coefficient, α.
There are numerous geotechnical applications
where rate effects are an important consideration,
including free falling penetrometer tests where
velocities can reach up to 12 m/s and Statnamic pile
tests where the pile is displaced at rates of the order
of 1 m/s (Chow & Airey 2011). In these cases, rate
effects cause the capacities derived from the tests to
exceed static values; an issue which is usually dealt
with by the inclusion of damping co-efficients in
the analyses (Brown 2009). These damping co-
efficients are not always uniform throughout each
test, but may vary with strain level. This highlights
the importance of understanding the strain level
dependence of rate effects in order to improve the
accuracy of dynamic testing and installation effects
in the field, as well as for use in FEA.
2 MATERIAL AND TESTING PROCEDURES
The tests were carried out on reconstituted
speswhite kaolin, the properties of which are
shown in Table 1.
The samples were first prepared as slurry with
a moisture content of 120% using de-aired, de-
ionised water before being one dimensionally con-
solidated to an effective stress of 180 kPa for three
days. These were then trimmed to 200 mm length
and 100 mm diameter to create triaxial samples.
Once installed in the triaxial apparatus, the sample
was saturated to an effective stress of 50 kPa at a
back pressure of 300 kPa and then re-consolidated
to an effective stress of 300 kPa to restore isotropic
conditions. Sample drainage was facilitated by
using vertical filter paper drains on the surface of
the sample, connected to both the top and bottom
drainage valves. These were required as the use of
lubricated end platens in the testing meant that
conventional drainage was not possible, and had
129
Table 1. Properties of the speswhite kaolin used.
Property Value
Plastic limit, wP (%) 32.5
Liquid limit, wL (%) 65.0
Plasticity index, PI (%) 32.5
Clay fraction (%) 80
Activity (%) 40.6
Specific surface area (m2/g)* 36.7
Permeability (mm/s)# 1.17 × 10−6
cv (m2/year)+ 23.52
MCSL 0.9
λ 0.101
N 2.678
*Determined from methylene blue spot testing;
#
Determined at an effective stress of 300 kPa; +Determined
for a 100 kPa stress increment.
the additional benefit of significantly reducing
consolidation times.
2.1 Testing apparatus
The tests were carried out in a GDS advanced elec-
tromechanical dynamic triaxial rig specially modi-
fied to carry out high speed monotonic tests. The rig
is capable of axial displacement rates of 100 mm/s,
and during high speed testing it is controlled by
a GDS digital control system capable of control-
ling the axial displacement within a time interval
of 0.1 milliseconds. Both the back and cell pres-
sures were provided by GDS pressure controllers.
Lubricated end platens of a similar design to those
proposed by Rowe & Barden (1964) were used in
order to minimise the inhomogeneity caused by
end restraint conditions. As these can introduce
errors into the measurement of small strains using
external methods, these were measured using
Hall effect transducers, two axial and one radial,
mounted directly on the sample providing a resolu-
tion of 1 × 10−6% strain. Pore pressures were moni-
tored using a mid-height pore pressure transducer
mounted on the surface of the sample.
2.2 Testing programme
The testing programme consisted of triaxial tests
at shear strain rates from 0.333 to 60,000%/hr in
order to investigate strain rate effects over as large a
range as possible. These were carried out at a com-
paratively low effective stress of 300 kPa as previ-
ous studies have shown that greater rate effects are
observed at higher moisture contents (Bea 1982,
Chow & Airey 2011). Throughout the testing pro-
gramme, the samples were allowed to drain through
the filter paper drains in order to allow rate effects
due to consolidation to be investigated.
3 RESULTS AND DISCUSSION
3.1 Observed rate effects and their modelling
The rate effects observed at the various strain rates
are shown in Figure 2, with a shear strain rate of
100%/hr taken as the reference rate. In order to
allow comparison with other studies using dif-
fering materials and sample sizes, the strain rates
have been converted in the normalised dimension-
less velocities used by Randolph & Hope (2004) as
shown in Equation 1.
vd
V=
cv
where V is the normalised dimensionless velocity, v
is the strain rate applied (in m/year), d is the sample
diameter (in m) and cv is the coefficient of consoli-
dation in m2/year. As can be seen, the curve follows
the behaviour expected with time for consolidation
effects dominating up to V = 11, after which und-
rained viscous effects are significant.
In order to quantitatively assess the data, the
rate effects model proposed by Randolph & Hope
(2004) shown in Equation 2 has been fitted to the
data using least mean square regression.
q ⎛ b ⎞
= 1+
qrefe ⎝ 1 + cV d ⎟⎠
⎪⎧ λ ⎡ ⎛V ⎞ ⎛ Vreef ⎞ ⎤ ⎪⎫
⎨1 + i −1 ⎜ ⎟ − sinh
⎢ sinh i −1 ⎜ ⎥⎬
⎩⎪ ln( ) ⎣ ⎝ V0 ⎠ ⎝ V0 ⎟⎠ ⎦ ⎭⎪
where b, c, and d are curve fitting parameters used
to model the time for consolidation effects and λ is
the rate effect per log cycle increase in strain rate
used to model viscous effects. Vref is the normalised
Figure 2. Graph of rate effect against normalised veloc-
ity at peak strength with the model by Randolph and
Hope (2004) using a shear strain rate of 100%/hr as the
reference rate.
velocity associated with chosen reference rate and
V0 is the point after which time for consolida-
tion effects are negligible. This process was also
repeated for the measured rate effects at selected
strains to identify the influence of strain level on
rate effects as shown in Figure 3. The parameters
obtained from the curve fitting process are shown
in Table 2.
The fitting parameters at peak strength (which
occurs at varying strain levels) show that the rate
effect per log cycle was found to be 22.5% which
is higher than previous studies. This highlights the
need to develop a framework to predict rate effects
(1) based on the current state and properties of the
material in question.
3.2 Strain level dependence of rate effects
Figure 3 shows that in the partially drained domain
rate effects increase with strain level, which is intui-
tive as at greater strains there will be more time for
consolidation to occur. Additionally, at the point
at which the transition to undrained behaviour
occurs, the rate effects were found to be relatively
independent of the strain level. Beyond this point
the rate effect per log cycle reduces from 27.5%
at 0.17% shear strain to 18.1% at 2.67% strain,
indicating that undrained rate effects reduce with
strain level, which is important in areas such as
Rapid Load Pile Testing (RLT) and assessment of
static pile working loads.
(2) Lo Presti et al (1996) developed a method for
comparing rate effects at different shear strain
levels by defining a strain rate coefficient, α
(Equation 3). This is the change in shear modulus
at a given strain over the log cycle increase in shear
strain rate normalised by the shear modulus at a
reference shear strain rate.
Figure 3. Graph of rate effect against normalised
velocity at various shear strain levels shown with the
model proposed by Randolph and Hope (2004).
130
 Table 2. Parameters used to fit the model by Randolph & Hope (2004) to the data at each
shear strain level.

Property 0.17% 0.33% 0.67% 1.67% 2.67% Peak

b 0 0 0.153 0.505 0.888 1.387

c 26.52 25.55 24.61 23.11 23.53 25.09
d 1.865 1.859 2.163 2.589 2.556 2.309
λ 0.275 0.251 0.244 0.212 0.181 0.225
V0 18.17 21.08 17.76 14.10 6.81 23.39
Vref 11.17 11.17 11.17 11.17 11.17 11.17

ΔG (γ )
α (γ ) = (3)
)) G (γ ,γ refe )
Δ (log(γ ))

This method was used to further investigate the

apparent dependence of rate effects on strain level.
Figure 4 shows the variation of the strain rate co-
efficient, α, with shear strain for the shear strain
rates considered. This confirms the earlier findings
that for rates where there is time for consolidation,
the rate effects observed increase with increas-
ing strain. Similarly, the rate effects reduced with
strain for the undrained tests. However, the graph
highlights a significant difference in the strain rate
co-efficient between the partially drained and und-
rained tests at extremely small strains.
Figure 4. Graph of strain rate co-efficient against shear
strain for various shear strain rates using a shear strain
3.3 Small strain behaviour rate of 100%/hr as a reference.
To further investigate the effect of strain level on
rate effects the small strain data from the Hall
effect transducers was examined. Figure 5 shows
the variation of the shear modulus, G, for each of
the rates. Of particular interest is the fact that the
initial small strain shear modulus, G0, appears to
vary with shear strain rate, reducing according to
the correlation shown in Figure 6. One possible
explanation for this is that at slower rates, there
is more time for sample drainage at a micro-scale,
causing an increase in the shear modulus at low
shear strain rates. However, this does not explain
the increased initial strain rate co-efficient at low
strains in the fast tests.
Figure 7 shows the variation of the elastic shear
strain threshold and this confirms that as found by
Mukabi & Tatsuoka (1999), the elastic shear strain Figure 5. Graph of shear modulus against shear strain
threshold increases with rate. A best fit was applied for various shear strain rates from Hall effect small strain
transducer data.
to this data and found that for the soil properties and
stress conditions used, γEL varies as in equation 4.

γEL = 0.003 γ 0.2

(4)
Considered in conjunction with Figure 5, this
shows that at the initial strain levels considered
in Figure 4 (γ = 0.2%), the faster tests exhibited a
greater stiffness than the slower rate tests, explain-
ing the discrepancy in the initial values of the strain
rate coefficients. This is due to the delay in stiffness
degradation caused by the increase in the elastic
shear strain threshold.

131
 (Figs. 8 and 9). This was done for the slowest
(0.333%/hr) and fastest (60000%/hr) tests.
This showed that at the low strain rate, the cur-
rent HS small model matched the measured shear
moduli with relatively good accuracy, with a devia-
tion of less than 12%. However, for the high strain
rate test, the increase in εEL and the more rapid
subsequent degradation meant that this variation
increased to up to 71%. Clearly, this margin of
error is significant.
It is proposed that this can be better modelled
using the form shown in Equation 6.

Figure 6. Graph of G0 against shear strain rate.

Figure 8. Comparison of the current and proposed

stiffness degradation models with the measured data for
Figure 7. Graph of elastic shear strain threshold a shear strain rate of 60000%/hr.
against shear strain rate.

3.4 Modelling of shear modulus degradation

One commonly used method for modelling shear
modulus degradation is the HS small model (Plaxis
2011) shown in Equation 5.

G 1
= (5)
G0 ⎛ γ ⎞
1 + α HHS ⎜
⎝ γ 0.0 7 ⎟⎠

where αHS is equal to 0.384, G is the shear modulus

observed at a given shear strain, γ, and γ0.7 is the
shear strain at which G degrades to 70% of the ini-
tial stiffness, G0.
As small strain degradation is important in
FEA, the predicted values of G/G0 from this model
were compared with those obtained by normalis- Figure 9. Comparison of the current and proposed
ing the observed shear moduli by the value of G0 stiffness degradation models with the measured data for
determined at the corresponding shear strain rate. a shear strain rate of 0.333%/hr.

132
G 1
= β
(6)
G0 ⎛ γ ⎞
1 + α HHS ⎜
⎟
⎝ γ 0.7,dyn ⎠
where αHS = 0.384 as in the existing model. γ0.7,dyn
is the shear strain at which G/G0 = 0.7 in the fast
(dynamic) test, which has the effect of moving the
G/G0 curve to the right to intercept the measured
data. Similarly, the new power, β, allows the steep-
ness of the rate of degradation after the elastic
shear strain threshold to be increased. The pro-
posed model is shown alongside the current HS
small model in Figure 8 and Figure 9. As can be
seen it closely follows the current model at slow
strain rates, while better approximating the data at
faster rates, reducing the variance from the data to
11%. For the soil tested, γ0.7,dyn and β were found
to vary according to the following relationships
(Equation 7 and Equation 8) such that the pro-
posed model resolves into the existing model at low
strain rates. (Figs. 10 and 11).
Figure 10. Graph of γ0.7, dyn/γ0.7 against shear strain rate.
Figure 11. Graph of β against shear strain rate.
133
γ 0.7,dyn = γ 0.7 × 1.5 γ 0.2 (7)
β = 0.043 log( γ ) + 1.13 (8)
where γ is the shear strain rate in%/hr.
4 CONCLUSIONS
Based on triaxial tests on reconstituted kaolin at
strain rates over more than 5 orders of magnitude,
it has been shown that rate effects have numerous
effects on the response of the soil, ranging from
increased deviatoric strength to changes in small
strain behaviour. These effects have also been
shown to be strain level dependent.
a. Peak strength rate effects in the partially drained
domain due to time for consolidation have
been shown to be greater than undrained rate
effects.
b. The undrained rate effects observed have been
shown to be higher than those reported by
other studies at 22.5% per log cycle for peak
strength.
c. The importance of strain level in relation to rate
effects has been highlighted, with rate effects
increasing with strain when partially drained
and reducing with strain when undrained.
d. The applied strain rate was found to have a sig-
nificant impact on the small strain response,
with the elastic shear strain threshold increasing
with rate and G0 appearing to reduce with rate.
e. The need for a new model to accurately rep-
resent shear modulus degradation was high-
lighted, and a potential form for achieving this
described.
It is hoped that the results and discussion in
this paper will be of use in the FEA of installation
effects and in their analysis more generally.
ACKNOWLEDGEMENTS
This work is supported by funding from the Uni-
versity of Dundee Division of Civil Engineering
and the Institution of Civil Engineers Research &
Development Enabling Fund, which is gratefully
acknowledged.
REFERENCES
Bea, R.G. 1982. Soil strain rate effects on axial pile
capacity. Proc. 2nd Int. Conf. on Numerical Methods in
Offshore Eng: 107–132.
Brown, M. 2009. Recommendations for Statnamic use
and interpretation of piles installed in clay. Rapid Load
Testing on Piles: 23–36. London: Taylor & Francis.
Brown, M.J. & Powell, J.J.M. 2013. Comparison of rapid
load test analysis techniques in clay soils. ASCE Jour-
nal of Geotechnical & Geoenvironmental Engineering.
Available online 21/03/12.
Chow, S.H. & Airey, D.W. 2011. Rate effects in free fall-
ing penetrometer tests. Proc. Int. Symp. on Defor-
mation Characteristics of Geomaterials, Seoul, 1–3
September 2011.
Krieg, S. & Goldscheider, M. 1998. Bodenviskotät und
ihr Einfluβ auf das Tragverhalten von Pfählen. Bau-
technik 75: 806–820. Ernst und Sohn.
Kulhawy, F.H. & Mayne, P.W. 1990. Manual on estimating
soil properties for foundation design. Report EL-6800,
Electric Power Research Institute, Pala Alto.
Lehane, B.N., O’Loughlin, C.D., Gaudin, C. &
Randolph, M.F. 2009. Rate effect on penetrometer
resistance in kaolin. Geotechnique 59: 41–52.
Lo Presti, D.C.F., Jamiolkowski, M., Pallara, O. &
Cavallaro, A. 1996. Rate and creep effect on the stiff-
ness of soils. ASCE GSP 61: 166–180.
Mukabi, J.N. & Tatsuoka, F. 1999. Influence of reconsol-
idation stress history and strain rate on the behavior
of kaolin over a wide range of strain. Geotechnics for
developing Africa: 365–377. Rotterdam: Balkema.
134
Plaxis bv. 2011. Material models manual 2011. Delft:
Plaxis.
Quinn, T.A.C. & Brown, M.J. 2011. Effect of strain rate
on isotropically consolidated kaolin over a wide range
of strain rates in the triaxial apparatus. Proc. Int.
Symp. on Deformation Characteristics of Geomateri-
als, Seoul, 1–3 September 2011.
Randolph, M.F. & Hope, S. 2004. Effect of cone velocity
on cone resistance and excess pore pressures. Proc. Int.
Symp. Eng. Practice and Performance of Soft Deposits:
147–152.
Rowe, P.W. & Barden, L. 1964. Importance of free ends
in triaxial testing. ASCE journal of the soil mechanics
and foundations division 90 (SM1): 1–27.
Shibuya, S., Mitachi, T., Hosomi, A. & Hwang, S.C.
1996. Strain rate effects on stress-strain behavior as
observed in monotonic and cyclic triaxial tests. ASCE
GSP: Measuring and modelling time dependent soil
behavior 61: 214–227.
Steenfelt, J.S. 1993. Sliding resistance for foundations on
clay till. Predictive soil mechanics: 664–684. London:
Thomas Telford.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Rate dependent shear strength of silt at low stresses

S. te Slaa & J. Dijkstra

Delft University of Technology, Delft, The Netherlands

ABSTRACT: The penetration resistance of freshly deposited silt is measured. Therefore, a miniature
ball-cone is designed with high accuracy at the applied scale. This paper presents the operation charac-
teristics of the newly designed ball-cone and presents the first results. The first results indicate that subtle
differences in penetration resistance are resolved.

1 INTRODUCTION Additionally, the probe geometry should facili-

tate a reliable relation between the penetration
Recently, researchers start to treat the erosion of data and the undrained shear strength in the sedi-
cohesive sediment from a soil mechanical perspec- ment. The ball-cone fulfills these requirements as
tive, e.g. (Jacobs, 2011; Winterwerp et al., 2012). a ball has a relatively large penetration surface
Such approaches are based on the assumption that and a closed form solution exists between the pen-
the mobilization of shear strength results from etration resistance and the undrained strength.
interaction between sediment particles on a micro- Although strictly speaking the latter only applies
scopic scale. The undrained shear strength Su is for a perfectly plastic material only, it is a reason-
used as the strength parameter in these approaches. able first estimate for this intermediate soil with a
In riverine environments hydrodynamic conditions very low permeability.
are predominantly characterized by shear flow. This For loosely packed intermediate soils, the shear
shear flow exerts a load on the bed. The response strength is depending on the shear rate however
of this bed is a function of the combined hydraulic (Chung et al., 2006; Randolph & House, 2001).
and geotechnical conditions. Only limited research Therefore, variation in shear strength with shear
has been carried out for freshly deposited silt beds. rate is considered to be a function of the failure
These mainly empirical studies lack a thorough mechanism only. Figure 1 shows a range in tran-
underpinning of the underlying physical mecha- sition from undrained to drained penetration
nisms (Roberts et al., 1998). resistance as the normalized velocity is reduced
As opposed to conventional geotechnical research obtained from experiments in a perfect plastic
the focus in this study is on the characterization of
the strength properties in the first 100 mm of a fresh
silt bed. Typically, in this surface layer the material
is loosely packed and the stress levels are very low.
The material behaviour of these sediments in these
conditions is poorly understood and is very hard to
study in classical geotechnical element tests.
Therefore, in the present study we investigate the
soil mechanical surface strength of fresh silt deposits
with the aim of finding a relation between geome-
chanical and hydrodynamic failure mechanisms of
silt. Hence, an in-situ penetration test is scaled to lab-
oratory dimensions as a first approach to measure
the undrained strength properties of the sediment.

2 EXPERIMENTAL

2.1 Ball-cone penetrometer Figure 1. Experimental derived range of constant pen-

etration test results. Shaded area represents the varia-
The very low shear strength in the sediment requires tion in normalized resistance as function of penetration
a penetration test with a very high sensitivity. velocity (Redrawn after Randolph & House 2001).

135
material. The dimensionless number on the x-axis
is a Peclet-like number and is used to express the
variation form drained to undrained response.
Peclet numbers are used as well for assessing
drained and undrained erosion of sediment beds
(Winterwerp & Van Kesteren, 2004).
The dimensions of the ball-cone are chosen
such that the full flow conditions are reached as
early as possible, in order to capture the resistance
in the first 100 mm of the bed, whilst maintain-
ing high sensitivity for low shear strengths. At the
same time scaling considerations require that the
minimum diameter of the ball should be large
enough to consider continuum conditions around
the probe. Hence, the ratio of the ball diameter Dball
and the largest mean diameter of a silty sediment
(D50 ∼ = 60 μm) is chosen to be Dball/D50 ≥ 200.
2.2 Mechanical setup
Following the requirements above the final design
for the penetrometer used in our experiments
is equipped with a 12.0 mm diameter ball. As
opposed to the design of Randolph & Hope (2004)
the ball resistance isn’t measured directly above the
ball but the load is transmitted by a rod to a meas-
uring bridge on top of the probe above the water
table. The rod (ø = 3.0 mm) is guided in a tube
(ø = 4.0 mm), which is not connected to the ball, to
prevent pick up of shaft resistance on the rod and
assure that only the ball resistance is measured. The
normal load in the rod is converted in a bending
moment of the sensing element in the measurement
bridge. This design is more sensitive for registering
small loads and not susceptible for temperature
fluctuations. The design is optimized for loads up
to 150 N with accuracy better than 0.01 N. The
dimensions of the ball-cone are shown in Figure 2.
Figure 2. Ball-cone penetrometer (not to scale).
136
The standard TU Delft 2 axis actuator, which
has originally been designed for the geotechnical
centrifuge, has been used to push the ball-cone into
the sediment. This actuator has sufficient linear
stroke for 100 mm penetration and the penetra-
tion rate has been varied between 0.05 mm/s and
0.3 mm/s.
2.3 Calibration procedure
Calibration of the ball cone penetrometer is carried
out on a regular basis using a calibration rig com-
prising of a lever system. Figure 3 shows a schema-
tization of this calibration set up. The ratio of the
length between the hinge and loading point (L1) and
the hinge and the ball cone (L2) is 1:5. Care is taken
that the ball cone is loaded vertically at all times.
Next a stepwise loading cycle is applied varying
within the range of 0–100 N. At the end of a load-
ing cycle the lever was removed again and a gently
tap against the shaft is given in order to check for
remaining friction between the tube and the con-
nection rod. This turned out to be negligible.
Calibration results show a linear relation between
the output voltage and the applied mass during the
loading cycle and some small hysteresis during the
subsequent unloading cycle (Fig. 4). A linear fit
Figure 3. Schematization of calibration set-up. Ratio
L1:L2 = 1:5.
Figure 4. Characteristic calibration results. Dark and
light blue dots indicate mean measured output volt-
ages during the loading cycle and unloading cycle,
respectively.

Figure 5. Grain size distribution of used silt,
D50 = 18 μm.
of the loading cycle shows a small offset near the
origin for which a correction is applied during the
actual measurement.
2.4 Sample preparation
Silt is defined as granular material with particle
sizes ranging from 2 to 63 μm with a non-cohesive
base mineral like quartz or feldspar. This material
is refrained from true cohesion.
For present experiment, an artificially produced
silt sized material is obtained from industry. The
preparation method should result in a freshly
deposited and saturated sample. In order to prevent
entrapment of air, the silt is sprinkled into the sam-
ple boxes which are filled with de-aired water. An
amount of silt is added at which the concentration
of the silt-water mixture was low enough so it was
still easy to mix up, but high enough te prevent seg-
regation of particles after mixing: the density of the
mixture was still below the structural density. Next,
the silt-water mixture was allowed to settle and con-
solidate for some time. This time was varied.
3 RESULTS AND ANALYSIS
3.1 Bearing factor N
The bearing factor (N = qnet/su) of a penetrometer
is of paramount importance to relate the net cone-
resistance to the shear strength. The cone resist-
ance can be determined directly by dividing the
measured force by the projected area. The used
ball-cone has a radius of 6.0 mm, corresponding
with a projected area of 113.1 mm2. Randolph et al
(2000) analyses the upper and lower bound solu-
tions for the bearing factor of a spherical object
penetrating a rigid-plastic material and compare
137
them with the solutions for the more classical
T-bar. The lower and upper bound solution follow
from using either a Tresca (upper bound) or Von
Mises (lower bound) yield criterion. Their solu-
tions lead to a non-dimensional bearing capacity
factor that varies in the range of 11.80–15.54 as the
ball interface condition varies from fully smooth
to fully rough. The corresponding values for a Von
Mises yield criterion are about 8% lower. Corre-
sponding lower bound solutions range from 10.98
for a smooth interface to 15.10 for a rough inter-
face as can be seen in Figure 6 (Randolph et al.,
2000). Here, only the theoretical solutions for pen-
etration resistance in isotropic Tresca soil will be
considered. The interface friction coefficient α is
assumed to be 0.3 (Low et al., 2011), resulting in
a bearing factor of 12.8. Now the net cone resist-
ance can directly be diverted to the measured shear
strength.
3.2 Vertical bed strength profiles
The first results for tests without the outer tube
installed show that the performance for the mini-
ature ball-cone in water is excellent. Both a high
sensitivity and low hysteresis is obtained for the
setup (Fig. 7).
This small hysteresis of 0.01 N (or 0.1 kPa) is
similar to the measured hysteresis during calibra-
tion (Figure 4). The maximum reading agrees well
with the buoyancy of the small ball cone at 8 mm
below the water surface. However, when the ball-
cone is used to penetrate a silt water mixture with
initial densities of 2020 kg/m3 (Fig. 8) a remaining
offset is observed while retracting the ball-cone.
This offset is not only the hysteresis in the system,
since this is only 0.35 kPa. Also, when ingress of
material between the inner rod and the outer tube is
minimized by sealing the opening with some grease
the offset reduces but does not vanish (Fig. 9).
Figure 6. Variation of bearing factor for ball and T-bar
with interface roughness (Randolph et al., 2000).
Figure 7. Net upward force due to buoyancy measured
during immersing in water.
Figure 8. Results of initial penetration tests with con-
stant penetration velocity. A remaining offset exists while
retracting the ball-cone. All test al performed on the same
sample.
A cross calibration of the probe performance
in silt is made by performing a penetration test
in silt on a balance (+/−1 g). The results shown in
Figure 10 show that an immediate constant pick
up of shaft resistance between rod and tube (result-
ing from some inclined penetration) is mobilized
(black line in Figure 10). This offset of about 3 kPa
remains when pulling the ball-cone out of the silt.
This offset will subsequently be used to correct
the penetration data in results shown below. The
source of is offset is found in the friction between
the outer tube and inner rod, since the results
from penetration tests without the outer tube are
refrained from this remaining offset (blue line in
Figure 10).
138
Figure 9. Results of penetration tests after friction
reduction between the inner rod and outer. All test al
performed on the same sample.
Figure 10. Cross calibration results for penetration tests
with and without tube with additional response measure-
ments on a balance.
4 CONCLUSIONS
The newly developed ball-cone penetrometer
shows to be a valuable tool for the characterization
of the first 100 mm of a silt bed. Subtle differences
in penetration resistance are resolved with a high
sensitivity by using a measurement bridge which
is configured for bending. This high sensitivity
resulted into the need to solve some subtle issues in
the mechanical system, before being able to use the
probe for the intended application.
REFERENCES
Chung, S., Randolph, M. and Schneider, J., 2006. Effect
of Penetration Rate on Penetrometer Resistance in
Clay. Journal of Geotechnical and Geoenvironmental
Engineering, 132(9): 1188–1196.
Jacobs, W., 2011. Sand-mud erosion from a soil mechani-
cal perspective, Ph.D. thesis, Delft University of
Technology.
Low, H.E., Randolph, M.F., Lunne, T., Andersen, K.H.
and Sjursen, M.A., 2011. Effect of soil characteristics
on relative values of piezocone, T-bar and ball pen-
etration resitances. Géotechnique, 61(8): 13.
Randolph, M.F. and Hope, S., 2004. Effect of cone
velocity on cone resistance and excess pore pressures,
Int. Symp. on Engineering Practice and Performance
of Soft Deposits, Osaka, Japan, pp. 147–152.
Randolph, M.F. and House, A.R., 2001. The complemen-
tary roles of physical and computational modelling.
IJPMG—International Journal of Physical Modelling
in Geotechnics, 1: 01–08.
139
Randolph, M.F., Martin, C.M. and Hu, Y., 2000.
Limiting resistance of a spherical penetrometer in
cohesive material. Géotechnique, 50(5): 573–582.
Roberts, J., Jepsen, R. and Gotthard, D., 1998. Effects of
particle size and bulk density on erosion of quartz par-
ticles. Journal of Hydraulic Engineering, 124: 1261.
Winterwerp, J. and Van Kesteren, W., 2004. Introduction
to the physics of cohesive sediment in the marine envi-
ronment. Elsevier Science Ltd.
Winterwerp, J.C., van Kesteren, W.G.M., van Prooijen, B.
and Jacobs, W., 2012. A conceptual framework for
shear-flow induced erosion of soft cohesive sediment
beds. Journal of Geophysical Research—Oceans.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Control of excess pore pressure development during pile

installations in soft sensitive clay

T. Tefera, G. Tvedt & F. Oset

Norwegian Public Roads Administration, Norway

ABSTRACT: During the construction of the 148 m long Øvre Sund Bridge on soft sensitive clay in
Drammen, in the south-eastern part of Norway, the stability of the river banks was a challenge. The
installation of displacement piles strongly compresses the adjoining soils and leads to build up of excess
pore water pressure. This temporary build up of excess pore water pressure, coupled with the sensitivity of
the clay soil, causes the soil to lose a good fraction of its shear strength in the short term. Dissipation of
the excess pore pressure generated during pile installations may allow pore pressure to rise in the vicinity
of the river bank slope, thus leading to failure of the slope, even where the soil has not been remoulded.
A strict construction control mechanism during the construction of the Øvre Sund Bridge in Drammen
was planned. The follow up of the development of excess pore pressure during piling activity was one of
the measures taken. Based on stability analyses of the slope along the river banks a criterion for excess
pore pressure margins due to pile installations was established. This paper discusses the result of this strict
pore pressure control for the piling activity. The result shows a systematic monitoring of the development
of excess pore pressure during pile installations in sensitive clay helps to observe the safety margins of
the slopes continuously during pile installation, saves waiting time due to excess pore water pressure and
avoids extra cost of the project due to the build up of excess pore pressure related to pile installations.

1 INTRODUCTION of soft, silty clay with an undrained shear strength

determined by vane shear test in the range of
The Øvre Sund Bridge project in Drammen, 10–20 kPa.
south-eastern Norway, is the recently completed The detailed geotechnical analysis of the slide
bridge construction across the Drammen River showed that the piling activity reduced the stability
which was officially open to public on the 10th of of the riverbank by 23% from 1.3 to 1.0.
September 2011. The bridge connects the Grønland To follow up and control the stability problem
part with the Hamborgstrøm part of the city. The during construction of the Øvre Sund Bridge
bridge is 148 m long, 23.1 m wide and has 5 spans of in this relatively sensitive site condition differ-
25.8 m and 32 m long as shown in Figure 1. The soil ent measure had been taken to limit the effect of
conditions on both sides of the river and in the river pile installations on the river banks. This action
bed are relatively similar. The soil is composed of
three layers. The relatively thin layer of silty sand soil
is overlying a relatively thick layer of soft clay. The
thick soft clay is overlying dense moraine. A total
of 191 concrete piles on the Grønland side, 32 steel
pipe piles in the river and 100 concrete piles on the
Hamborgstrøm side of the river were installed.
According to the geotechnical report of the
project there had been many slides along the riv-
erbanks of the river. One of the slides and a more
dramatic one that had occurred related to pile
driving for a factory building was on December 9,
1971 (Aas 1975). The piling activity triggered a
large slip at which approximately 1000 m2 of the
site subsided and slid out into the Drammen
River.
The soil condition where the slide had occurred Figure 1. Plan and section view of the Øvre Sund
composed of 5 m of sand and silt over 15–20 m Bridge.

140
includes measures taken before piling started as
part of site preparation, during piling activity and
after piling. These measures taken are discussed in
section 3.
2 SITE CONDITION
The Øvre Sund Bridge construction site is located
in the city of Drammen in the south-eastern part
of Norway about 50 km from Oslo, across the river
Drammen. The general soil condition on both sides
of the river and in the river bed is relatively similar.
The site condition consists of mainly three differ-
ent soil layers. The thin top layer of silty sandy soil
with some gravel is overlying a relatively thick layer
of soft sensitive clay. The thick soft sensitive clay
is overlying dense moraine. The depth to the dense
moraine layer is dipping in the direction of the
Hamborgstrøm side of the river (Fig. 1). The depth
to the bottom dense moraine layer varies from about
10 m on the upstream side of Grønland to about
25 m on the downstream of the Hamborgstrøm side
of the river. The clay layer is normally consolidated
on both sides of the river and has an undrained
shear strength of 10–15 kN/m2 at the top of the
layer and increases with depth to 20–30 kN/m2 at
the bottom of the layer. The remolded undrained
shear strength is in the order of 1 kN/m2, the unit
weight is in the order of 19–20 kN/m3 and the sen-
sitivity is 6–35. The transition between the soft clay
layer and dense moraine layer consists of stratified
sand and gravel.
3 MEASURES TAKEN TO CONTROL
THE STABILITY OF THE RIVER BANKS
The geotechnical evaluation of the river bank’s
slope stability shows that the safety factor is mar-
ginal. Any additional loading on the river banks
and construction activities which will increase the
pore pressure in the soil will adversely affect the sta-
bility of the slopes. This implies that there is a need
to take different counter measures in design and
during construction of the abutments foundation.
The bridge abutments foundation was therefore
designed in such a way that the load will be trans-
ferred deep down to the dense moraine layer. Pre-
cast concrete piles with dimension 270 × 270 mm
were chosen for the abutments foundation. Fur-
thermore different actions were taken to control
these marginally stable river bank slopes. These
actions include measures before piling started as
part of site preparation, during piling activity
and after piling completed. Based on economical
aspects and practical considerations of the river
system the following measures were taken.
141
3.1 Before the pile driving started
In order to improve the safety factor of the river
banks slope during pile installation, the river
banks top terrain was unloaded by excavating the
top soil. This unloading improved the safety factor
of the riverbanks slope before starting the instal-
lation of piles by 10–20%. Figure 2 shows the site
preparation which is mainly unloading the top soil
on the Grønland side of the river bank.
3.2 During pile driving
The installation of displacement piles strongly
compresses the adjoining soils and leads to build
up of excess pore water pressure. The dissipation
of the excess pore pressure generated during pile
installations may allow pore pressure to rise in the
vicinity of the river bank slope, thus leading to fail-
ure of the slope, even where the soil has not been
remoulded. The development of excess pore pres-
sure during pile installation for the construction of
Øvre Sund Bridge was a major concern for the sta-
bility of the river banks. To reduce this build up of
pore water pressure due to pile installation proper
selection of pile type, special installation technique
on the riverbanks and strict monitoring of the
build up of pore water pressure were carried out.
The foundation for axis 5 pillar, in the river, which
is close to the river bank on the Hambergstrøm side
of the river (Fig. 1) was designed using 813 mm
diameter open ended steel pipe piles to reduce the
effect of soil displacement during pile installation.
Clogging of these open ended pipe piles during
installation was a main concern which will affect
the main intended purpose to reduce the build up
of excess pore pressure. The development of pore
water pressure during pile installation in axis 5 was
also followed by piezometers.
Vertical drains (Mebradrain) were also installed
on the outside wall of the open ended steel pipe
piles (Fig. 3). A technique commonly adopted in
Scandinavia to reduce the effect of soil displace-
ment during installation by preboring a slightly
Figure 2. Unloading of the top terrain on the Grønland
side of the river bank to improve the stability situation
before piling.

Figure 3. The four vertical drains installed on the
outside wall of the open ended steel pipe piles in axis 5.
Figure 4. The soft clay material (clay sausage) drilled/
sucked out of the hole before driving the concrete pile.
undersize hole (clay sausage) before driving piles
was also considered. This technique was carried
out for 6 concrete piles in the Hambergstrøm side
of the river bank (Fig. 4).
Furthermore the development of excess pore
pressure during piling activity was monitored by
piezometers systematically installed at different
depths in the ground. A total of 8 piezometers in
the Grønland side of the river bank, 6 piezometers
in the river and 8 piezometers on the Hamborg-
strøm side of the river bank were installed. Based
on thorough evaluation of the critical slip surfaces
of the river bank slopes maximum allowable excess
pore water pressure during pile installation was
established. This maximum allowable excess pore
142
water pressure value, alarming boundary, varies
for each piezometer according to their correspond-
ing locations in the ground. The excess pore water
development due to pile driving activity in the
Grønland side of the riverbank is discussed more
in detail in section 4.
3.3 After pile driving is completed
The main concern for the stability of the river-
bank slopes after the installation of piles for the
abutment foundations were the additional terrain
loads from backfilling and the possible erosion at
the foot of the river bank slops. The geotechnical
evaluation shows that these activities reduce the
safety factor of the river banks below the accept-
able margins. Based on economical and practical
considerations to reduce the effect of backfill-
ing on the stability of the riverbanks light weight
materials were used. Accordingly the top layer of
the river bed was carefully replaced by properly
designed gravel material to protect the river bed
from erosion.
4 EXCESS PORE WATER PRESSURE
DEVELOPMENT DURING PILING
To predict the excess pore water development
related to pile driving is a complex task. Further-
more prediction of excess pore water pressure
and rate of dissipation is, however, still not suf-
ficiently reliable for an evaluation of the stability
to be made in advance of the construction. The
preferred method is of course to couple the theo-
retical predictions and evaluations made by meas-
urements and following up out in the field during
construction. Based on the field measurement of
the excess pore pressure it is, however, possible
to evaluate the stability of the slope during con-
struction at any time. This is the measure that has
been taken during the construction of Øvre Sund
Bridge in Drammen to evaluate the stability of
the river bank slopes during pile installation. The
same principle was adopted on both sides of the
river bank. In the Grønland and Hamborgstrøm
sides near the edges of the river banks (Fig. 1) 8
piezometers on each side, and in the river near the
edges of the river banks 6 piezometers, all together
22 piezometers were installed. Based on the con-
tractual agreement piling activity has to stop when
there is high excess pore water pressure build up
in the soil. Furthermore it is the client’s respon-
sibility to cover the downtime (waiting time) cost
due to excess pore water pressure. The follow up of
excess pore pressure development on the Grønland
side of the river bank is discussed more in detail in
this paper.
 The development of pore water pressure was
monitored by piezometers at four locations in each
side of the river bank and at three locations in the
river bed. Because of the relative proximity to the
river bank, pile driving activities in axes G4, G5,
G6 and river front foundations (Figs. 1 and 5) were
critical for the stability of the river bank slope.
Figure 5 shows the locations of piezometer sta-
tions in the Grønland side of the river bank.
Table 1 shows the elevation at which the tips of
the eight piezometers where placed in the ground.
The piezometers at station GP4 were installed two
years before the startup of the construction of the
bridge. The two years measurement shows that the
pore water pressure in the soil is close to hydro-
static and follows both the seasonal as well as tidal
variation of the river water level. The piezometers
at stations GP1, GP2 and GP3 were installed two
weeks before the piling activity started. The tips
of the piezometers in these three stations were
located along the critical failure plane where a
change in shear strength of the soil has a signifi-
cant effect on the stability of the river bank slope.
Figure 6 shows the follow up of the development
of excess pore water pressure during pile driving
for the riverfront foundation near station GP4.
Pile driving on land started near the river bank.
This piling activity near piezometer stations GP3
and GP4 was a very good chance to document
recorded data for the specific site condition. Based
on these data a pore pressure chart was developed.
In the chart the maximum build up of excess pore
water pressure vs. distance from the pile can be
correlated.
Figure 5. Locations of piezometer stations in the
Grønland side of the river. GP1, GP2, GP3 and GP4 are
piezometer stations.
143
Table 1. Elevation of the tip of the piezometers in the
Grønland side of the river bank.
Station Piezometers Elevation, masl
GP4 G-71306 −4.9
G-71606 −9.9
G-71706 −13.1
GP3 G-53108 −10.0
G-52808 −14.0
GP2 G-53008 −9.0
G-52908 −14.0
GP1 G-52508 −7.0
Figure 6. Station GP4 with three installed piezometers
in the Grønland side of the river and driving pile G184.
Note that the river was frozen due to cold temperature
in January.
As anticipated during installation of piles into
the soft clay layer excess pore water pressures were
generated at both shallow and deep piezometer
locations. The excess pore pressure was gener-
ated within approximately 13 m radius from the
center of the pile. No significant changes in pore
water pressure were observed beyond this distance.
Figure 7 shows the relative location of piezometer
tips and length of pile driven into the ground.
The effects of pile installation alone on the
development of excess pore water pressure with-
out preboring a slightly undersize hole were care-
fully monitored before, during and immediately
after pile installations in the Grønland side of the
river bank. Table 2 and Figure 8 show the excess
pore water pressure recorded due to driving pile
number G184 and G182. At station GP4 the pore
water pressure increase due to driving the two piles
G182 and G184 was relatively large, 80 kPa in the
middle and 45 kPa at the shallow depth of the soft
clay layer. At the deeper piezometer location near
to the permeable moraine layer the pore pressure
 Figure 8. Measured excess pore water pressure build up
at piezometer G-71606 station GP4 during driving pile
Figure 7. Relative location and length of Pile G182 and G182 and G184.
Pile G184 installed into the ground near station GP4.

Table 2. Measured excess pore-water pressure due to

driving pile nr G184 and G182 near station GP4.

Distance from Excess pore

piezometer pressure
Piezometer Pile nr. (m) (KPa)

G-71306 G184 4.1 8.2

G182 2.0 25.2
G-71606 G184 5.1 11.6
G182 0.5 68.1
G-71706 G184 4.7 1.0 Figure 9. Measured excess pore water pressure per pile
G182 1.5 −1.0 vs. horizontal distance from the pile at station GP2, GP3
and GP4.

increases were much less pronounced. In Figure 8
the reference numbers are: number 1—before pil-
ing activity, 2—after driving pile G184 about 12 m
into the ground, 3—after driving pile G184 about
26 m into the ground, 4—after driving pile G184
and before driving pile G182, 5—after driving pile
G184 and driving pile G182 about 12 m into the
ground, 6—after driving pile G184 and driving pile
G182 about 24 m into the ground and 7—one day
after both piles were driven.
Furthermore the recorded excess pore water
pressure for the deepest piezometer at station GP4
confirms that the piezometer tip is located just
on top of the dense moraine layer which consists
of stratified sand and gravel. As is clearly seen
in Table 2 the recorded excesses pore water pres-
sure depends not only on the horizontal distance
between the pile and the piezometer but also on the
depth in the soft clay layer where the piezometer
tip is located. Based on the recorded excess pore
water pressure at stations GP2, GP3 and GP4 an
excess pore water pressure chart was developed.
Figure 9 shows the development trend of the antic-
ipated excess pore water pressure in the 10–15 m
thick soft clay layer due to driving 270 × 270 mm
concrete pile.
5 CONTROL OF EXCESS PORE WATER
DEVELOPMENT DURING PILING
Based on stability analysis critical failure planes of
the river bank slopes were identified. Further analy-
ses were made to determine the maximum allowable
excess pore water pressure in the slope beyond which
the safety factor drops below the acceptable margin.

144

Figure 10. The location of piezometer tips and the
critical failure planes. L1is the sandy, silty soil with some
gravel, L2 is the soft sensitive silty and sandy clay layer
and L3 is the dense moraine layer.
Figure 10 shows the critical failure planes and the
locations of the piezometer tips at GP1, GP2 and
GP3. The established alarming boundary zones for
the excess pore water pressure at station GP1, GP2
and GP3 were 10, 50 and 60 kPa respectively. When
the build up of excess pore water pressure along
the critical failure plane at station GP3, GP2, and
GP1 reaches a critical value of 60 kPa, 50 kPa and
10 kPa respectively piling activity has to stop until
it dissipates. The stability analyses also show that
a localized build up of excess pore water pressure
more than 60 kPa can be tolerated if there is no
substantial change on the adjacent stations along
the critical failure plane. Based on the excess pore
water pressure development trend line (Fig. 9) a
detailed schedule for pile driving was prepared. A
minimum distance between piles of 8 m for piles to
be driven per day was planned (Tvedt 2009). Based
on the measured excess pore water pressure the
safety margins of the river bank slopes were con-
tinuously checked during construction.
6 CONCLUSION
As was anticipated during driving of piles into the
soft clay layer excess pore water pressures were gen-
erated at both the shallow and deep piezometers.
145
It was also observed that during pile driving the
excess pore pressure developments were measured
within approximately 13 m from the center of the
pile. No significant changes in pore water pres-
sure were observed beyond this horizontal distance
from the pile center immediately after driving. Sig-
nificantly high pore water pressure increases were
not recorded by deep piezometers when the piles
were driven into the underlying dense moraine.
The development of excess pore water pressure
was successfully controlled by limiting the mini-
mum distance between piles to be driven per day.
The pile driving schedule prepared based on mini-
mum distance between piles to be driven per day
was very effective. It was possible to keep the devel-
oped excess pore water pressure within the allow-
able range, continuously control the safety margin
of the river bank slopes and save the client from
waiting time cost. Total lengths of about 2980 m of
pile length were driven into the ground in 28 days
which was about 106 m pile length per day.
ACKNOWLEDGEMENT
The authors would like to thank the Norwegian
Public Roads Administration, NPRA, for provid-
ing resources for this publication.
REFERENCES
Aas, G. 1975. Skred som følge av peleramming i bløt leire.
NGI Publication 110. 49–54 (In Norwegian).
Eigenbord, K.D. & Issigonis, T. 1996. Pore-water
pressures in soft to firm clay during driving of piles into
underlying dense sand, Can. Geotech. J. 33, 209–218.
Fleming, K., Weltman, A., Randolph, M. & Elson, K.
2008. Piling Engineering, Taylor & Francis, New York.
Johansen, S. & Finstad, J.A. 2009. Øvre Sund bru
alarmgren-se poretrykksmålere Grønland, RIG
026-Rev A. (In Norwegian).
Kirkebø, S. 2006. Forslag til sikkerhetsfilosofi, RIG 001.
(In Norwegian).
Nr. 115513-4, Rv 283 HP: 400, Parsell Øvre Sund bru,
Geoteknisk rapport, 1. februar 2007 (In Norwegian).
Tvedt, G. & Tefera, T. 2009. Øvre Sund bru erfaringer
med poretrykk og bæreevne under peleramming på
Grøn-land, Notat. (In Norwegian).
Tefera, T. Tvedt, G. & Oset, F. 2011. Excess pore pressure
during pile driving in soft sensitive clay. Proceedings
of 15th European Conference on Soil Mechanics and
Geotechnical Engineering, Athens. 1285–1290.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Seabed pipelines: The influence of installation effects

D.J. White
Shell EMI Offshore Engineering, University of Western Australia, Australia

ABSTRACT: Seabed pipelines exemplify the type of geotechnical challenge addressed by the GEO
INSTALL project. Pipeline laying involves complex soil-structure interaction. The seabed undergoes
large deformations that are accompanied by changes in strength. Pipeline design requires assessment of
the available pipe-soil interaction forces after installation, and the potential mobility and scour of the
surrounding soil. These all depend on the installation process. This paper summarises recent research into
the geotechnical aspects of pipeline installation, including numerical and physical modelling as well as
field observations.

1 PIPELINE GEOTECHNICS
1.1 Offshore pipeline networks
Pipelines form the vital transport network of off-
shore hydrocarbon developments. They carry the
hydrocarbon product and other fluids between
wells and in-field processing facilities, and also to
shore. A pipeline network is accompanied by vari-
ous forms of infrastructure to support, connect and
stabilize the pipelines, as illustrated in Figure 1.
‘Rigid’ pipelines are made from steel tubes, typi-
cally 0.1–1.5 m in diameter, with layers of internal
and external coating for corrosion, abrasion and
impact protection, thermal insulation, and to add
weight for improved stability. ‘Flexible’ pipelines
are composites of metal and polymer, and range
in diameter from 0.1–0.5 m. ‘Risers’ are separate
tubes or continuations of a pipeline that provide
the connection to the floating or fixed structure at
the ocean surface.

Figure 1. Pipeline networks and associated infrastructure.

146
 As offshore developments extend into deeper
waters located further from shore, pipelines and
risers represent an increasingly important part
of the development infrastructure. Typical costs
of gas export pipelines on the North-West Shelf of
Australia now exceed $4 million per km, a signifi-
cant proportion of which is for measures to stabi-
lise the pipe on the seabed. Stability is the major
focus of geotechnical design for pipelines, both
under the action of hydrodynamic loading from
waves and currents, and due to expansion and con-
traction arising from the elevated temperature and
pressure of the contents.
Even if long export pipelines are not required,
because the product is exported via tankers from
a floating facility, a large network of infield pipe-
lines may be needed. BP’s Greater Plutonio devel-
opment provides a well-documented case study
(Jayson et al., 2008). This project is located 165 km
north-west of Luanda off the coast of Angola in
a water depth of 1400 m. A Floating Production,
Storage and Offloading (FPSO) vessel is moored
adjacent to a top-tensioned riser tower, which is a
column of vertical pipes, or risers that carry prod-
ucts between the FPSO and the seabed. The risers
are maintained in tension, to prevent buckling,
by a buoyancy unit at the top. The wells are dis-
tributed over a wide area, and are each supported
on subsea foundations. Connecting the wells are
more than 100 km of pipelines. These are laid
directly on the seabed and are subject to signifi-
cant expansion loads during cycles of startup and
shutdown, as the temperature and internal pres-
sure change.
1.2 Pipeline installation effects
The GEO INSTALL project, which this confer-
ence forms the closing event of, has focused on a
research area that is extremely relevant to the anal-
ysis of seabed pipelines. The research activities in
GEO INSTALL mirror some of the research efforts
underway in our own group at the University of
Western Australia (UWA). Quoting from the GEO
INSTALL programme outline:
“Infrastructure construction involves the installa-
tion of structural elements [such as pipelines]. The
installation process itself can be quasi-static or
dynamic, and generally involves very large defor-
mations and changes in pore pressure. … the influ-
ence of installation on key design parameters, such
as mobilised strength at the soil-structure interface
and soil stiffness, is difficult to quantify and, as yet,
impossible to model.”
Pipeline engineering exemplifies precisely these
geotechnical challenges, and this keynote provides an
opportunity to expose pipeline geotechnics—which
147
is a rather niche area of research and practice—to
a wider geotechnical audience.
The uncertainties on which GEO INSTALL
has focused, associated with soil-structure inter-
action and the resulting large deformations and
changes in soil properties, are shown for pipelines
in Figure 2.
The laying process for a seabed pipeline is
dynamic. The pipe is either stored in short sections
and welded together as part of the lay process, or
(for smaller diameter pipes) it is stored on a reel
and ‘unwound’ as the vessel sails along the route.
In regions where over-trawling is not a hazard,
pipelines are usually laid directly on the seabed
without burial.
Depending on the lay method and soil conditions,
the process of pipe embedment may be drained,
undrained or partially-drained. Motions of the ves-
sel and hydrodynamic loading on the hanging pipe
section lead to slow dynamic motions between the
pipe and the seabed. The pipe is not simply lowered
statically onto the seabed. The seabed is disturbed
by these dynamic motions, and in soft soils heave
and remoulding occur around the pipe.
After laying, there is typically a period of sev-
eral months before the pipeline enters operation.
During this time, any excess pore pressure gen-
erated during the lay process dissipates and the
surrounding soil changes in strength. It is this
post-equalisation consolidated strength—which
varies spatially around the pipe—that controls the
subsequent resistance of the pipe to lateral and
axial movements.
These issues will be familiar to piling engineers.
Pile design methods for axial capacity in clay soils
Figure 2. Pipeline installation effects.
are often couched in terms of the post-installation
strength or lateral effective stress on the pile,
accounting for changes in strength from installa-
tion and equalization (e.g. Randolph et al. 1979,
Jardine et al. 2010).
1.3 Pipeline design for global expansions
During operation, the internal temperature and
pressure cause a pipeline to expand. This expan-
sion is opposed by the axial resistance between the
pipe and the seabed. Excessive compressive force
may lead to buckling of the pipeline, depending
on the soil resistance and the pipe weight. Once a
buckle forms, the axial force drops significantly as
a length of pipe feeds axially into the buckle. Exces-
sive feed-in—due to low axial resistance—leads to
high bending strains within the buckled section.
In deep water, where there is no requirement
to bury the pipe for protection from trawling, on-
bottom buckling in the horizontal plane is now a
widely adopted solution to accommodate temper-
ature and pressure-induced expansions. However,
controlled lateral buckling requires very careful
design in order to ensure that buckles form as
planned, and that the fatigue response within the
buckle is acceptable throughout the operating life
of the flowline (Bruton et al. 2007, 2008). Thermal
cycles can also lead to the accumulation of axial
movement, which is termed pipeline walking (Carr
et al. 2006). The walking arises from asymmetry in
the heat-up and shutdown processes, or from other
asymmetries in the pipeline such as a high riser ten-
sion at one end or a seabed slope. Over many ther-
mal cycles, walking can lead to significant global
displacement of the pipeline. Walking is not a limit
state for the pipeline itself, but can lead to failure at
the mid-line or end connections.
Buckling and walking must be assessed during
design, in order to control the stresses within the
pipeline and the movements relative to any asso-
ciated structures. Structural analyses of pipelines
incorporate interaction forces between the pipe and
the seabed in the vertical, axial and lateral direc-
tions. The conventional approach is to construct
independent models of the force-displacement
response in each direction—in a manner analo-
gous to the ‘t-z’ and ‘p-y’ models used for axial and
lateral pile-soil analysis. The pipe-soil resistance
in each of these directions depends on the pipe-
line embedment and the surrounding soil strength.
Correct modelling of installation effects is required
to assess these forces accurately.
1.4 Pipeline design for hydrodynamic stability
During the period between installation (laying) and
operation, the embedment of the pipeline may also
148
change due to seabed mobility. In shallow water
(<150 m depth) the action of waves and tides can
lead to sediment transport at the seabed, and scour.
Depending on the metocean and seabed conditions,
as well as the installed (or ‘as-laid’) embedment of
the pipeline, this seabed mobility may lead to net
erosion or burial around the pipe (Figure 2).
Installation of a pipeline on the seabed causes an
increase in the local sediment transport capacity due
to the change in flow conditions around the pipe-
line. A consequence of the increase of local sedi-
ment transport capacity is the scour of sediments
away from an area surrounding the pipeline. Once
local scour is initiated at a cross-sectional location,
it can spread longitudinally along the pipeline to
form a free span. The pipeline may then sag into the
scour hole as the result of the structural deflections
and extra penetration into the soil at the span shoul-
ders (which attract increased load from the span-
ning lengths of pipeline). As the pipeline sags into
a scour hole, it deflects more flow over the top of
the pipeline. This leads to a reduction of sediment
transport capacity directly below the pipeline. The
scour hole is then backfilled with sediments and this
induces the natural self-burial of the pipeline. Once
the pipeline is buried or sags into the scour hole, its
stability is enhanced through hydrodynamic shield-
ing and the increase of lateral soil resistance.
Accurate modelling of the installation process
and the resulting as-laid embedment is an important
input to assessments of seabed mobility processes.
These processes are illustrated in Section 6.
1.5 Soil mechanics context
The stress levels relevant to pipeline geotechnics are
significantly lower than found in conventional foun-
dation engineering. The bearing pressure imposed
by pipeline is typically 0.5–5 kPa, and the soil in the
shallowest 0.5 m of the seabed is generally very soft
(if fine-grained) or mobile (if coarse-grained).
Pipeline laying, and any subsequent movements
across the seabed, lead to large strains in the sur-
rounding soil. The consequent remoulding and
reconsolidation of the seabed soil leads to changes
in topography and strength. Site investigation
technology is advancing rapidly, to provide bet-
ter tools to quantify these near-surface seabed
characteristics.
Two important aspects of the constitutive
response of fine-grained seabed soils, which affect
the interaction forces on seabed pipelines, are:
1. The curvature of the effective stress failure enve-
lope at low stresses.
2. The generation and dissipation of excess pore
pressure during and in between episodes of
shearing to very high strains.
 This first effect is illustrated by the two types
of data shown in Figure 3. The red lines are effec-
tive stress paths determined from load and pore
pressure data gathered by the ‘SMARTPIPE’—an
instrumented model pipe that operates in situ, at the
seabed. As the pipe was slowly cycled axially, alter-
nating shear stresses were mobilized, tracing out the
failure envelope. The cross-shaped markers are the
results from model tests performed in a large tank
of the same soil, reconstituted at NGI’s laboratory.
The second effect is best demonstrated by cyclic
penetrometer tests with periods of reconsolida-
tion between each cyclic episode, such as the T-bar
test shown in Figure 4. The reduction in strength
Figure 3. Failure envelope curvature at low stresses:
Comparison of residual effective stress states measured
during axial pipeline sliding on very soft clay (Hill et al.
2012).
Figure 4. Undrained strength through episodes of
remoulding and reconsolidation: a T-bar test in lightly
overconsolidated kaolin in the UWA beam centrifuge
(White & Hodder 2010).
149
during the initial packet of cycles is partly due to
the generation of positive excess pore pressure
in this lightly over-consolidated material. As this
positive pore pressure dissipates and the effective
stress rises back to the geostatic state the mate-
rial densifies and the subsequent undrained shear
strength is higher.
After just three episodes of full remoulding and
reconsolidation, the current remoulded strength is
comparable to the original intact strength. These
results illustrate the contrasting effects of remould-
ing and reconsolidation that are shown schemati-
cally around a pipeline in Figure 2.
This behaviour is best explored in a critical state-
type framework, since this provides an explicit
link between moisture content (which reduces as
positive pore pressures dissipate) and undrained
strength. White & Hodder (2010) describe a critical
state-type model that produces the back-analysis
indicated by the line ‘Model’ in Figure 4.
2 PIPELINE INSTALLATION GEOMETRY
2.1 Pipeline laying processes
The first and most well-recognised installation
effect relevant to the assessment of pipeline embed-
ment is simply the geometry of the lay process.
Pipelines are usually installed by laying from a ves-
sel, using an S-lay, J-lay or reel-lay configuration.
In an S-lay arrangement, the pipe is assembled in
a horizontal position and then offloaded over the
rear of the vessel in short sections, concurrent with
a forward motion of the vessel. The pipe steepens
in an overbend behind the vessel, and is supported
by a frame extending from the vessel, called a
stinger.
In a J-lay arrangement the pipe is assembled on
the vessel at a steep inclination, close to vertical,
which eliminates the overbend at the offloading
point. The departure angle from a reel lay arrange-
ment may be steep like a J-lay system, or more
shallow, over a stinger, like an S-lay system.
2.2 Touchdown stresses
The static bearing capacity of a pipeline on und-
rained soil is well established, based on plasticity
solutions (Murff et al. 1989, Randolph & White
2008a, Martin & White 2012), finite element
analyses (Aubeny et al. 2005, Merifield et al. 2009,
Chatterjee et al. 2012a), and model testing (Dingle
et al. 2008).
Example vertical penetration responses are
shown in Figure 5. These were obtained using
different software but the same large deforma-
tion finite element technique (Hu & Randolph
Figure 5. Typical undrained static vertical penetration
response (Randolph & Gourvenec 2011).
Figure 6. Pipeline laying notation (Randolph & White
2008b).
1998a, b), which is described in Section 4.1. The
variation in normalized vertical load V/su-invertD
with embedment, w/D fits a power law (where V is
the vertical pipe-soil load, su-invert is the in situ und-
rained shear strength at the pipe invert level, and D
is the pipe diameter).
However, assessment of as-laid pipe embed-
ment is not simply a matter of equating the pipe
submerged weight, W′, with the bearing capacity,
V, to determine the embedment, w. For a pipeline
being laid from a vessel, the maximum value of V
within the touchdown zone is greater than the sub-
merged pipe weight W′ due to the catenary effect.
This stress concentration can be estimated from
the lay tension, pipe bending stiffness and the sea-
bed response (Fig. 6) (Pesce et al. 1998, Lenci &
Callegari 2005, Randolph & White 2008b, Palmer
2009).
Parametric solutions for the static lay condi-
Figure 7. Effect of seabed stiffness on the static pipe-
tions have been presented by Randolph & White soil contact stress through touchdown zone (Westgate
(2008b). For most practical values of lay tension, et al. 2013).
analytical solutions (Lenci & Callegari 2005) and
numerical analysis using OrcaFlex (Orcina 2008)
all converge to unique design lines. The touchdown interaction in pipeline laying: the stiffness of
stress concentration, flay, can be expressed approxi- the seabed affects the loads applied to the struc-
mately as: ture. The range of stiffnesses, k = V/w, shown in
Figure 7 is wider than the usual uncertainty faced
Vmax ⎛ EIk ⎞
0.25 in bearing capacity calculations. This is because
f lay = ≈ 0.6 + 0.4 ⎜ (1) the lay process introduces additional uncertainty
W′ ⎝ T0 ⎟⎠ in the operative seabed strength, which is discussed
in Section 3.
where Vmax is the maximum vertical pipe-soil
load and the other parameters are as shown in
Figure 6. 3 PIPELINE INSTALLATION DYNAMICS
The stiffness of the seabed affects both the max-
imum stress concentration and also the length of While a pipeline is being laid, the vessel is sub-
the elevated contact stress through the touchdown jected to waves and the hanging length of pipe also
zone. Example profiles of stress concentration are attracts hydrodynamic load in the water column.
shown in Figure 7 for a range of seabed stiffnesses. These effects cause the pipe to oscillate at the
This illustrates the influence of soil-structure seabed, softening the soil and entraining water.

150
Also, in rough sea states, transient reductions in 4 SOIL DISTURBANCE FROM PIPE
lay tension can occur which increase the vertical LAYING
pipe-soil load (Westgate et al. 2009, Westgate et
al. 2010a). 4.1 Numerical modelling technique
The combination of both vertical and horizontal
Assessments of the degree of soil remoulding
pipe-soil loading reduces the available vertical
during pipeline laying can be made using Large
bearing capacity of the soil, as shown through
Deformation Finite Element (LDFE) analysis.
theoretical plasticity analysis for planar contacts
Our group at UWA has made extensive use of the
(Green 1954) and also shallowly-embedded pipes
‘RITSS’ form of LDFE analysis, which was set
(Randolph & White 2008a). Lastly, horizontal
out by Hu & Randolph (1998a, 1998b) and has
oscillations can displace the seabed soil laterally,
recently been coupled with the commercial soft-
creating a trench by lowering the seabed elevation
ware ABAQUS.
adjacent to the pipe, as shown via experimental
The RITSS scheme is illustrated in Figure 9. In
simulations in a geotechnical centrifuge (Cheuk &
this method, large displacements of the pipe are
White 2011) and via field observations (Lund 2000,
divided into many small steps in which the pipe
Westgate et al. 2009).
is moved to a very small incremental distance
During pipe laying, remotely operated vehi-
and small strain Lagrangian calculations are per-
cles are used to survey the touchdown zone dur-
formed. Stresses and other material parameters
ing pipelaying. Video observations show that the
are then recovered from the integration points to
amplitude of pipe oscillations are typically a frac-
the nodes following the superconvergent patch
tion of a pipe diameter, with a period comparable
recovery technique (SPR, Zienkiewicz & Zhu
to the wave period (Westgate et al. 2009). Struc-
1992).
tural modelling of the vessel response and the pipe
At the end of each step, the displaced soil
catenary can be used to estimate these movements
boundary is updated and the whole soil domain
a priori (Westgate et al. 2010a).
is remeshed. After that, the stresses and other
The amplitude of the pipe oscillations in the
parameters are interpolated from the old nodes to
touchdown zone increases with the seastate.
the integration points of the new mesh. With the
Westgate et al. (2010a) back-analysed the observed
interpolated stresses as initial conditions for the
embedment of a pipeline installed across uniform
next step, another small strain step is performed.
soil conditions but in a range of seastates. These
This process is repeated until the pipe is moved the
data show a strong correlation between seastate—
required distance. Mesh generation, small strain
and thus vessel motion and pipe oscillations in the
calculations and post processing of the data are
touchdown zone—and embedment (Fig. 8).
performed using ABAQUS. Python, the in-built
Experimental and numerical simulations have
scripting language of ABAQUS, is used to auto-
been performed to explore these dynamic installa-
tion effects. The critical questions are:
1. How does the installation process alter the
strength of the soil surrounding the pipeline?
2. What is the consequent effect on the as-laid
embedment and the breakout resistance?

Figure 8. Effect of metocean conditions on as-laid pipe Figure 9. RITSS FE analysis scheme (after Wang et al.
embedment (Westgate et al. 2010a). 2010).

151
mate the process. The whole process is executed 4.2 Numerical simulations of pipeline laying
by a main Fortran program that calls different
This numerical technique has been used to explore
subroutines and Python scripts to carry out the
the pipe-seabed interaction during laying. An
entire task.
example analysis is shown in Figure 10. The in
Within each small strain analysis, the soil is
situ soil strength profile was su0 = (sum + kz) with
modelled as an elastic perfectly-plastic material,
sum = 0.75 kPa, k = 1.6 kPa/m, and the soil sof-
with failure according to the Tresca yield criterion.
tened according to Equation (2) with parameters
The particular value of the Tresca shear strength
ξ95 = 10, δrem = 0.25, μ = 0. The pipe was held under
for each element within the mesh is calculated
a constant vertical load of V/sumD = 3.7 and oscil-
when each small strain analysis is initialised. This
lated laterally with amplitude of +/−0.05 D.
calculation accounts for the influences on the soil
These parameters were chosen to represent a
of the cumulative strain (i.e. the level of distur-
centrifuge model test reported by Cheuk & White
bance or remoulding) strength, as well as the cur-
(2011), which is discussed in Section 4.3. The
rent strain rate.
amplitude of horizontal oscillation is comparable
The cumulative strain induced in a soil element
to the video observations of pipe laying reported
by any previous pipe movement will cause soften-
by Westgate et al. (2009). The resulting pipe embed-
ing of that soil and a decrease in the operative shear
ment and the measured horizontal resistance agree
strength—akin to the behaviour shown during the
well with the centrifuge model test results.
initial cyclic episode shown in Figure 4.
The calculated patterns of lateral resistance
The effects of strain rate on the shear strength
and embedment and the local soil remoulding are
of soil have also been studied extensively by many
shown in Figure 10. Even small lateral motions
researchers (e.g. Casagrande & Wilson 1951,
of just +/−0.05 D lead to a remoulded zone that
Graham et al. 1983, Biscontin & Pestana 2001,
extends by al-most one pipe diameter to each side,
Lunne et al. 2006, Low et al. 2008).
and the pipe it-self rests on fully remoulded soil.
To combine these softening and strain rate
The horizontal resistance initially rises through
effects within an undrained analysis, the simple
the first few cycles, as the pipe embedment and
Tresca soil model is used. After each analysis step,
soil contact width increase. With further cycles the
the original shear strength of the soil at every node,
effect of soil softening on either side of the pipe
su0, is modified to an updated value, su, according
dominates, as plastic strain accumulates. This leads
to the following formula (Einav & Randolph 2005,
to a reduction in the normalized lateral resistance
Zhou & Randolph 2007, 2009, Chatterjee et al.
in the later cycles. The vertical embedment after
2010, 2012a):
20 cycles is significantly greater than the 0.1 D
penetration achieved under purely vertical loading
⎡ ⎛ ⎛ ( ε1 − ε3 ) v p / D ⎞ ⎞ ⎤ prior to the lateral cycles.
s u = ⎢1 + μ log ⎜ max ⎜1, ⎥
⎢⎣ ⎝ ⎝ δ D γ ref ⎟⎠ ⎟⎠ ⎥⎦ The pipe embedment has almost stabilized after
30 cycles. Any additional cycles do not signifi-
e + ( − rem ) e
× ⎣⎡δ rem −3 ξ ξ95
⎤⎦ s u 0 (2) cantly deepen the pipe, because the soil influenced
by the lateral motions is already fully softened,
and the vertical bearing capacity is provided by
The first part of the relationship captures the
deeper soil that is less influenced by the motions.
increase in strength due to the strain rate effect and
the second part reflects a strength decrease due to
strain softening. Δε1 and Δε3 are the major and
minor principal strains, respectively, resulting from
a displacement increment, δ. vp is the pipe velocity
and μ is the rate of strength increase per decade
of strain rate, typically in the range of 0.05–0.2
(Biscontin & Pestana 2001, Lunne & Andersen
2007). γ ref is the reference strain rate taken as 1%
per hour or 3 × 10−6 per second. Softening is mod-
elled as an exponential decay function of the cumu-
lative absolute plastic shear strain ξ. δrem is the ratio
of fully remoulded strength to the initial strength
and is essentially the inverse of sensitivity, St. The
parameter ξ95 indicates the relative ductility of soil
and is the value of ξ at which the soil strength has
reduced by 95% of the way from the initial to the Figure 10. LDFE simulation of soil strength after
fully remoulded values. dynamic pipe laying (after Wang et al. 2009).

152
This stabilization is a useful concept for the assess-
ment of dynamic embedment in practice, since it
narrows down the input parameters necessary for
a simplified analysis of the lay process. It may not
be necessary to specify precisely the number of
cycles that will be imposed during the lay process
if that number is sufficient for the embedment to
approach the steady value.

4.3 Geotechnical centrifuge simulations

To complement the numerical LDFE studies of
pipeline laying, several physical model testing stud-
ies have been performed using the UWA geotech-
nical beam centrifuge. The typical arrangement for
a pipe-soil interaction test is shown in Figure 11.
The UWA centrifuge is equipped with a motion
control system that allows load-controlled or
displacement-controlled operation on independent
vertical and horizontal axes. The software control
system allows sequential operations to be pre-
programmed, providing a flexible basis for mim-
icking the behaviour of an element of pipe during
the laying process. Details of the control and data
acquisition systems are given in De Catania et al.
(2010) and Gaudin et al. (2010).
An example result from the study reported by
Cheuk & White (2011) is shown in Figure 12. The
soil properties and simulated pipe weight, V, were
the same as the LDFE simulation described in Sec-
tion 4.2. The settlement under the static load was
∼0.1D (point A in Figure 12). The imposed lat-
eral motions were 20 cycles of ±0.05 D (as for the
LDFE simulation, Figure 10) followed by further
sets of 20 cycles of amplitude ±0.1, ±0.2, ±0.1 and
finally ±0.05.
Like the LDFE simulation, the pipe initially
settled at a rapid rate. On each occasion that the

Figure 12. Physical modelling of the response of a pipe-

line element during laying (Cheuk & White 2011).

amplitude of lateral motions steps up, the lateral

Figure 11. Typical arrangement for a pipe-soil interac-
tion test in the UWA beam centrifuge.
resistance jumps up then falls back as the newly-
disturbed soil becomes remoulded (points B and
C in Figure 12).
A visual indication of the remoulding that takes
place during dynamic pipeline penetrated into a
soft clay seabed is provided by images from a sub-
sequent study (Westgate et al. 2013). In these tests
the model seabed was lightly over-consolidated
kaolin clay with a mudline strength of ∼1 kPa.
After penetration of the pipe monotonically into
this seabed, the disturbed soil appears as a blocky
rubble (Fig. 13a). In contrast, after dynamic

153
Figure 13. Disturbed soil after (a) monotonic
penetration and (b) dynamic embedment of a model pipe
(Westgate et al. 2013).

embedment the surrounding soil has the appear-

ance of being fluidized (Figure 13b).

4.4 Calculation model for softening and embedment

The simulations reported by Cheuk & White
(2011) and Westgate et al. (2013) have been used
to devise approximate methods for assessing pipe-
line embedment accounting for the progressive
remoulding due to small-amplitude oscillations
within the touchdown zone.
These methods use as inputs an assumed pattern
of lateral oscillations within the touchdown zone and
the concentrated vertical load that emerges from the
catenary solution. The lateral movements are com-
bined with a ‘macroscopic’ version of Equation (2) to
capture the progressive softening of the soil. Rather
than considering the strains within a soil element,
the ‘macroscopic’ approach lumps the softening
of the surrounding soil into a single quantity that is
linked to the cumulative lateral movement—which is
analogous to the cumulative shear strain in the soil.
This process provides an estimate of the changing
strength and therefore bearing capacity of the sur-
rounding soil through the assumed number of oscil-
lations within the lay process. Based on this changing
bearing capacity, theoretical failure envelopes in
Vertical-Horizontal (V-H) load space are used to
assess the consequent settlement of the pipe, based
on the assumption of associated flow.
This embedment calculation method can be
applied in a cycle-by-cycle manner (Cheuk & White
2011) or through the full displacement response of
the pipe element (Westgate et al. 2013).
Figure 14. Comparisons of calculated (vertical lines)
4.5 Comparison with field surveys and observed pipe embedment for three field sites (West-
gate et al. 2013) (sites referred to as A-C. Case 2 calcu-
Three sets of as-laid field survey data have been lations use average lay conditions. Cases 1 and 3 use
back-analysed by Westgate et al. (2013) using this calmest and heaviest seas respectively).

154
embedment calculation method. The pipelines
ranged from 0.33 to 0.63 m in diameter, with
submerged pipe weights between 0.22 and 0.95
kN/m. The sediments were fine-grained sediments
with undrained shear strength gradients between
4 and 38 kPa/m (and negligible strengths at the
mudline).
Estimates of pipe embedment were made for
each site using the pipeline properties, soil con-
ditions, and model parameters representative of
the site-specific lay conditions. Three calculations
of the embedment process were performed for
each site, representing the range in sea state (wave
height) measured during laying, and therefore the
amplitude of oscillations in the touchdown zone.
The ranges of calculated embedment values are
compared to histograms of as-laid embedment for
each site in Figure 14. In all cases, the agreement
between the calculated and observed embedment is
excellent, capturing both the average embedment
(when using the best estimate parameters) and
the range of embedment (when using the extreme
minima and maxima parameters).
A less accurate but simpler approach to assess
as-laid pipeline embedment on fine-grained soils, is
to assume that the soil is fully remoulded and apply
this strength within the vertical bearing capacity
equation (Westgate et al. 2010b; Westgate et al.
2012). This approach is often sufficiently accurate
for typical lay conditions, but probably only suc-
cessful due to compensating errors: the reduction
in bearing capacity under V-H loading is ignored,
but the loss of strength is probably over-estimated
through the assumption of full remoulding.
5 POST-LAYING BREAKOUT RESISTANCE
5.1 Equalisation
The seabed softening that results from the dynamic
movements during laying of a pipeline is princi-
pally due to positive excess pore pressure being
generated as the soil is sheared. After laying, this
excess pore pressure, and any excess pore pressure
generated simply in response to the static pipeline
self-weight, will dissipate.
During this dissipation period, the effective
stress around the pipe-soil interface will rise and the
strength of the surrounding soil will also increase.
These ‘set-up’ processes will cause the available
axial and lateral pipe-soil resistance to rise.
There is typically a period of several months
between the laying of a pipeline and the initial
operation. It is therefore the consolidated axial and
lateral resistance that is usually relevant to design,
but some assessment of the pore pressure dissipa-
tion rate is required to confirm this.
155
Krost et al. (2011) and Gourvenec & White
(2010) present numerical solutions for consolida-
tion around pipelines ‘wished-in-place’ in elastic
soil. The dimensionless time factors, T = cvt/D2,
for 50% and 90% dissipation of the average excess
pore pressure around the pipe surface are T50 ∼ 0.1
and T90 ∼ 2 respectively. Subsequent solutions,
using the Modified Cam Clay constitutive model,
have considered the effect of the installation proc-
ess (Chatterjee et al. 2012b).
5.2 Lateral breakout resistance: Numerical results
Theoretical solutions for the combined vertical-
horizontal bearing capacity of a pipeline embed-
ded in uniform undrained soil are well-established
(Randolph & White 2008a, Merifield et al. 2008,
Martin & White 2012). However, the dissipation of
installation-induced excess pore pressure changes
the strength of the surrounding soil, with a conse-
quent enhancement of the V-H bearing capacity.
Numerical simulations of pipe penetration
and lateral breakout in Modified Cam Clay soil
are reported by Chatterjee et al. (2013). The pur-
pose of this study was to investigate the potential
enhancement of lateral breakout resistance due to
consolidation hardening of the surrounding soil.
Two types of analyses were performed: undrained
vertical penetration followed by either unconsoli-
dated or consolidated undrained lateral breakout.
After the initial penetration (modelled using the
LDFE approach—Section 4.1), the pipe was either
immediately broken out laterally (the unconsoli-
dated case), or full pore pressure dissipation was
permitted before breakout (the consolidated case).
The unconsolidated and consolidated fail-
ure envelopes for an undrained embedment of
w/D = 0.1 and 0.5 are compared in Figure 15.
Figure 15. Unconsolidated and consolidated undrained
yield envelopes for w/D = 0.1 and 0.5 (Chatterjee et al.
2013).
The initial undrained strength was uniform with
depth and the soil was normally consolidated.
After consolidation, the failure envelopes have
expanded by approximately 50% for purely verti-
cal loading, and the horizontal load at the apex
point (where H is maximum) is approximately
doubled. The changes in strength in the surround-
ing soil, and the additional embedment that arises
during the consolidation process, are shown in
Figure 16 for the w/D = 0.5 case. Clearly the post-
installation consolidation process has a significant
influence on the lateral breakout resistance, and
‘set-up’ effects of this magnitude should not be
neglected.
However, it should be noted that in these analy-
ses the simulated pipe weight was maintained at the
unconsolidated undrained vertical bearing capac-
ity during the dissipation process, thus maximiz-
ing the potential strength gain. In practice, most
pipelines will be ‘overloaded’ on laying, because
the self-weight will be lower than the maximum
vertical load applied in the touchdown zone (by at
least a factor of flay, Section 2.2). The changes in
V-H bearing capacity in this example are therefore
upper bounds for set-up in normally consolidated
Modified Cam Clay.
5.3 Lateral breakout resistance: Centrifuge results
Centrifuge model tests provide complementary
data of the ‘set-up’ around pipelines installed into
soft soil. Using the general experimental arrange-
ment shown in Figure 11 and Figure 13, a set of
model tests has recently been performed to compare
the lateral breakout response from different initial
conditions that involve varying levels of genera-
tion and dissipation of excess pore pressure during
Figure 16. Contours of ratio of consolidated to original
shear strength after pore pressure dissipation (Chatterjee
et al. 2013).
156
installation (Gourvenec & White 2013). Details of
a subset of these tests are set out in Table 1. The
purpose of these tests was to explore the changes
in lateral breakout resistance that can result from
installation effects in soft clay, associated with dis-
turbance and reconsolidation. The model seabed
was normally—consolidated kaolin clay.
In all three cases, the initial embedment was
w/D = 0.5 and immediately after reaching this
embedment the simulated pipe weight was reduced
to V/D = 3 kPa. Two alternative ‘laying’ methods
were considered: undrained monotonic vertical
penetration and a specified pattern of oscillations
that remould the surrounding soil. The adopted
‘zig-zag’ pattern of lateral movement during pen-
etration (visible in Fig. 17) mimics the disturbance
and remoulding associated with the real lay proc-
ess, and the results from case RemCU are therefore
the most representative of field conditions.
For the monotonic installation method, two
cases involving full (90%) and negligible dissipa-
tion were performed (cases UU and CU). The
same pipe weight was maintained during breakout,
Table 1. Lateral breakout comparison cases.
Case V/D ‘Laying’ Consolidated?
UU 0.9 kPa Push→w/D = 0.5 No: immediate
breakout
CU 0.9 kPa Push→w/D = 0.5 Yes: 90%
dissipation
RemCU 0.9 kPa Specified Yes: 90%
oscillation dissipation
Figure 17. Pipe invert trajectories during lateral
breakout for different installation and ‘set-up’ cases
(Gourvenec & White 2013).
and the pipe was able to rise or fall to maintain
this load.
The displacement paths during lateral breakout
are shown in Figure 17. In all cases, the trajectory
is approximately the same with the pipe moving
upwards at a slope of approximately 10°. The
mobilised lateral resistance is significantly differ-
ent in each case.
The UU case involves no consolidation effects so
the original undrained strength might be assumed to
apply throughout, notwithstanding some softening
due to the accumulated strain. In this case, the
penetration and breakout resistance should be pre-
dictable using the in situ su profile and theoretical
solutions, such as failure envelopes based on plas-
ticity limit analysis (Randolph & White 2008a,
Martin & White 2012). The line marked LA in
Figure 18 shows this calculated breakout resistance,
ignoring heave (i.e. assuming a horizontal sea-
bed beside the pipe) and using the in situ strength
profile. There is indeed good agreement with the
UU result, with the ∼10% under-prediction being
attributable to the neglect of heave.
The more realistic CU case shows an 80%
increase in peak breakout resistance, which is com-
parable to the LDFE results shown in Figure 15.
This gain in capacity is highly brittle, which is
also consistent with the small size of the zone of
enhanced strength shown in Figure 16. However,
even beyond this initial peak, the CU case still
shows higher resistance than the UU case, as the
strengthened soil continues to be mobilised and
pushed ahead of the pipe.
The most realistic case—RemCU—shows an
even greater breakout resistance, more than double
Figure 18. Lateral breakout response for different instal-
lation and ‘set-up’ cases (Gourvenec & White 2013).
157
that observed in case UU. This gain in resistance
is less brittle, reflecting a wider zone of strength-
ened soil. The remoulding during ‘laying’ leads
to a larger zone of elevated pore pressure and a
correspondingly higher gain in undrained strength
after consolidation. The outcome—which is ini-
tially counter-intuitive—is that more softening
during laying leads to more net hardening after
consolidation.
5.4 Summary of post-laying breakout observations
These complementary simulations of pipeline
laying and breakout, using numerical modelling
(Section 5.2) and centrifuge testing (Section 5.3)
show a strong installation effect. This effect results
from post-laying consolidation, and is enhanced
if the lay process involves remoulding. Three key
conclusions emerge, which provide some answers
to the questions raised in Section 3 for pipelines
laid on soft clay seabeds:
1. Pipeline installation can significantly enhance
the strength of the surrounding soil due to con-
solidation under the pipe self-weight.
2. The dynamic pipe movement during installa-
tion creates additional shear-induced excess
pore pressure in the soil surround the pipeline.
Although this remoulding weakens the soil in
the short term, the subsequent strength gain
during consolidation is greater than following
monotonic installation.
3. The subsequent breakout resistance is signifi-
cantly enhanced (‘set-up’) by these changes
in strength, and suitable modifications to the
operative strength are required when using the-
oretical calculations for breakout resistance.
It should however be noted that in these tests
the soil was initially normally consolidated. This
state provides the greatest tendency to contract
and harden in response to shearing. At a higher
OCR these ‘set-up’ effects will be smaller, and the
effects of remoulding and reconsolidation may
balance, meaning that use of the in situ strength
may be appropriate in assessments of breakout
resistance.
6 POST-LAYING SEABED MOBILITY
The studies described in Sections 4 and 5 relate
principally to fine-grained soils, in which the
response during laying is undrained. These soils
are not usually mobile under environmental
hydrodynamic loading. However, coarser materi-
als, such as sands and silts, do move under wave
and current action in sufficiently severe storm or
tide conditions.
 When pipelines are installed on mobile soils,
there is generally a risk of global instability—i.e.
gross movement of the pipeline—under hydro-
dynamic action. The design must ensure that any
such movements are tolerable.
Current design practice to assess global pipe-
line stability generally neglects the mobility of the
seabed. The analysis instead focusses on simulat-
ing fluid-pipe and pipe-seabed interaction without Figure 19. General view of UWA large O-tube facility.
including the fluid-seabed interaction that leads to
the scour and erosion processes that were intro-
duced in Section 1.4 (e.g. Zeitoun et al. 2008).
Significant research efforts have been devoted
to pipeline on-bottom stability and seabed mobil-
ity at UWA over the past three years, supported
by the ARC1 and industry partners Woodside and
Chevron through the STABLEpipe Joint Indus-
try Project. An innovative Large O-Tube (LOT)
testing facility has been established to simulate
ocean-seabed interactions at large scale (Cheng et Figure 20. Images from a large O-tube test showing
al. 2010, An et al. 2011, Jas et al. 2013). The LOT scouring and settlement of a pipeline on a mobile seabed,
followed by breakout.
is a recirculating water tunnel that can generate
cyclonic and tidal flow conditions that cannot be
modelled in traditional wave flumes of similar shallowly embedded, a scour tunnel opens beneath
dimensions (Fig. 19). the pipe, widening into a scour hole that grows
Precise control of the inline propeller allows longitudinally. As the pipe sinks into this hole is
maximum wave and current conditions of approx- becomes more stable, because of the shielding
imately 3.0 m/s with a peak period of 13 s and from the flow and the increased soil resistance that
4.0 m/s respectively. Irregular flow can also be gen- results from the trench.
erated, and typical 100-year return period tropical If the pipe is initially too deeply embedded for
storm conditions for Australia’s NWS are com- a scour tunnel to open up, then this process is
fortably within the performance envelope of the delayed or may not occur. Consequently, a pipeline
facility. that is installed at a deeper embedment can be less
The facility is also equipped with an instrumen- stable than a pipeline that is initially resting on the
ted model pipe mounted on a sophisticated actuator seabed at zero embedment.
system to record the applied horizontal and vertical Figure 20 shows a sequence of images from a
forces. The pipe is also equipped with a network of large O-tube test in which an irregular storm was
surface pressure cells to record the hydrodynamic imposed on a pipe initially at a very shallow embed-
load around the pipe circumference. The actuator ment. The images show this sequence of (i) tunnel
system prevents model pipe movements in unreal- scour, (ii) propagation of the tunnel (away from
istic degrees of freedom such as rolling. The feed- the window), (iii) settlement of the model pipe and
back system can provide neutral horizontal control, then (iv) breakout at the height of the storm. In
allowing the pipe free lateral movement in response other tests, different responses have been observed
to the natural balance between hydrodynamic load- for different installed embedments.
ing and soil resistance.
Model tests in the large O-tube have high-
lighted the influence of pipeline installation—and 7 CONCLUSIONS
the resulting as-laid embedment—on the subse-
quent on-bottom stability. Some of the results are Seabed pipelines are their own foundation. The
counter-intuitive, and show the reverse trends rela- geometry of this foundation—the depth and
tive to conventional design practice. width—is set by the installation process, which is
For example, on mobile sandy soils the gen- subject to uncertainty even until the day of installa-
eral response is for the seabed to begin scouring tion when the prevailing weather determines the lay
before the storm is sufficiently intense to dislodge vessel motions and the consequent oscillations as the
a pipe resting on the seabed surface. If the pipe is pipeline touches down on the seabed.

1
Australian Research Council Linkage Project LP0989936.

158
 The installation of seabed pipelines exemplifies the
particular class of geotechnical engineering that the
GEO INSTALL project has been focused on. Pipeline
installation causes large seabed deformations and
changes the properties of the surrounding soil.
These effects alter the subsequent pipe-soil interac-
tion behaviour.
This paper has summarised recent research
efforts that have involved numerical analysis, exper-
imental testing and field observations. The over-
arching aim has been to create an understanding of
installation effects to allow more reliable estimates
of the subsequent in-service behaviour to be made.
Large deformation finite element analysis is a
mature research tool that is increasingly being used
in industry practice. The particular LDFE method
used at UWA provides a flexible basis for incorpo-
rating the constitutive behaviour that is most rele-
vant to the problem being simulated. Softening and
rate effects have been superimposed on a conven-
tional Tresca model. Modified Cam Clay has been
used to capture consolidation and drainage effects.
An equally important research tool is well-
controlled physical modelling. Examples from
both the UWA geotechnical centrifuge facilities
and the new O-tube flume have been presented.
Where soil-structure interaction involves large
deformations and many episodes of loading or
displacement, physical modelling remains essential
to uncover effects that could not be reliably fore-
seen using only numerical modelling. In offshore
engineering, physical modelling often provides the
only basis for gathering validation data for compu-
tations and design calculations, due to the limita-
tions of scale and location associated with offshore
infrastructure.
For pipelines on fine-grained soils, an impor-
tant theme is the contrasting effects on soil
strength of remoulding and reconsolidation.
Classical solutions for undrained penetration
and lateral breakout resistance have recently been
derived, using traditional plasticity limit analysis
and complementary numerical tools. Field and
laboratory testing has shown that the use of these
solutions requires careful selection of the opera-
tive strength. During dynamic pipe laying on soft
soils, the strength is reduced due to remoulding,
but during subsequent breakout it can be higher
due to reconsolidation.
In coarse-grained soils the installation process
and the resulting embedment affect the propensity
for seabed mobility around the pipe and subse-
quent scour and self-burial. This cross-disciplinary
topic is an important area of current research. It
offers the potential for a more realistic methodol-
ogy for assessing pipeline stability on mobile sea-
beds. UWA’s O-tube is a novel tool that allows full
ocean-pipeline-seabed interaction to be simulated
159
at large scale, to provide data to underpin these
new approaches.
ACKNOWLEDGEMENTS
This work forms part of the activities of the Cen-
tre for Offshore Foundation Systems (COFS),
currently supported as a node of the Australian
Research Council Centre of Excellence for Geo-
technical Science and Engineering. Support from
the ARC Future Fellowships program and Shell is
acknowledged.
The pipe-soil interaction studies described here
have been performed in collaboration with many
colleagues at UWA, including Mark Randolph,
Susan Gourvenec, Dong Wang, Zack Westgate,
Santiram Chatterjee and Matt Hodder. The O-tube
team includes Liang Cheng, Hongwei An, Cheng-
cai Luo and Scott Draper.
Industry support from BP (Andy Hill), Subsea 7
(Paul Brunning), Woodside (Andrew Pearce, Nino
Fogliani) and the SAFEBUCK and STABLEpipe
JIPs is also acknowledged.
REFERENCES
An, H., Luo, C., Cheng, L. White, D. and Brown, T. 2011.
Introduction to the O-tube. Proc. Sixth International
Conference on Asian and Pacific Coasts. Hong Kong,
China.
Aubeny, C.P., Shi, H. & Murff, J.D. 2005. Collapse load
for a cylinder embedded in trench in cohesive soil.
International Journal of Geomechanics, 5(4): 320–325.
Bruton, D., Carr, M. & White, D.J. 2007. The influence
of pipe-soil interaction on lateral buckling and walk-
ing of pipelines: the SAFEBUCK JIP. Proc. 6th Int.
Conf. on Offshore Site Investigation and Geotechnics,
London. 133–150.
Bruton D.A.S., White D.J., Carr M.C. & Cheuk, C.Y.
2008. Pipe-soil interaction during lateral buckling
and pipeline walking: the SAFEBUCK JIP. Proc. Off-
shore Technology Conference, Houston, USA. Paper
OTC19589.
Biscontin, G. & Pestana, J.M. 2001. Influence of periph-
eral velocity on vane shear strength of an artificial
clay. ASTM Geotech. Test. J. 24(4): 423–429.
Carr, M.C., Sinclair, F., & Bruton, D.A.S. 2006. Pipeline
walking—understanding the field layout challenges,
and analytical solutions developed for the SAFE-
BUCK JIP. Proc. Offshore Tech. Conf., Houston,
Paper OTC17945.
Casagrande, A. & Wilson, S.D. 1951. Effect of rate of
loading on the strength of clays and shales at constant
water content. Géotechnique 2(3): 251–263.
Chatterjee, S., Randolph, M.F., White, D.J. & Wang, D.
2010. Large deformation finite element analysis of
vertical penetration of pipelines into the seabed. Proc.
2nd Int. Symp. on Frontiers in Offshore Geotechnics.
Perth. 785–790.
Chatterjee S., Randolph M.F. & White D.J. 2012a. The
effects of penetration rate and strain softening on the
vertical penetration resistance of seabed pipelines.
Géotechnique, 62(7):573–582.
Chatterjee S., Yan, Y., Randolph M.F. & White D.J.
2012b. Elastoplastic consolidation beneath shallowly
embedded offshore pipelines. Géotechnique Letters,
2(2):73–79.
Chatterjee S., White D.J. & Randolph M.F. 2013. Cou-
pled consolidation analysis of pipe-soil interactions.
Canadian Geotechnical Journal, in review.
Cheuk, C.Y. & White D.J. 2011. Modelling the dynamic
embedment of seabed pipelines. Géotechnique. 61(1):
39–57.
Cheng, L. White, D.J., Brown, T. An, H. and Luo, C.
2010. O-tube Pipeline Stability Project, Development
of the Large O-tube Facility, Commissioning Report.
UWA Report GEO: 10500 v2. 115 pp.
De Catania S., Breen J., Gaudin C. & White D.J., 2010.
Development of multiple-axis actuator control system
Proc. Int. Conf. on Physical Modelling in Geotechnics.
Zurich, Switzerland, 325–330.
Dingle, H.R.C., White, D.J., & Gaudin, C. 2008. Mecha-
nisms of pipe embedment and lateral breakout on soft
clay. Canadian Geotechnical Journal, 45(5): 636–652.
Einav, I., & Randolph, M.F. 2005. Combining upper
bound and strain path methods for evaluating pene-
tration resistance. International Journal for Numerical
Methods in Engineering, 63(14): 1991–2016.
Gaudin C., White D.J., Boylan N., Breen J., Brown, T. &
De Catania S. 2010. Development of a miniature high
speed wireless data acquisition system for geotechni-
cal centrifuges Proc. Int. Conf. on Physical Modelling
in Geotechnics. Zurich, Switzerland, 229–234.
Gourvenec S.M. & White D.J. 2010. Elastic solutions for
consolidation around seabed pipelines. Proc. Offshore
Technology Conference, Houston. Paper 20554.
Gourvenec S.M. & White D.J. 2013. Softening and con-
solidation around seabed pipelines: centrifuge model-
ling. In review.
Graham, J., Crooks, J.H.A. & Bell, A.L. 1983. Time
effects on the stress-strain behaviour of natural soft
clays. Géotechnique 33(3): 327–340.
Green, A.P. 1954. The plastic yielding of metal junctions
due to combined shear and pressure, J. Mechanics and
Physics of Solids, Vol. 2, pp. 197–211.
Hill A., White D.J., Bruton D.A.S., Langford T., Meyer V.,
Jewell R. & Ballard J-C. 2012. A new framework for
axial pipe-soil interaction illustrated by a range of
marine clay datasets. Proc. Int. Conf. on Offshore Site
Investigation and Geotechnics. SUT, London.
Hu, Y. & Randolph, M.F. 1998a. A practical numerical
approach for large deformation problems in soil. Int.
J. Numerical and Analytical Meth. Geomech. 22(5):
327–350.
Hu, Y. & Randolph, M.F. 1998b. H-adaptive FE analy-
sis of elastoplastic non-homogeneous soil with large
deformation. Computers and Geotechnics. 23(1–2):61–83.
Jardine, R.J., Chow, F.C., Overy, R. & Standing, J.R.
2010. ICP design methods for diven piles in sands and
clays. Thomas Telford.
Jas E., O’Brien D., Fricke R., Gillen A., Cheng L. &
White D.J. 2013. Pipeline stability revisited. Journal of
Pipeline Engineering. in press.
160
Jayson, D., Delaporte, P., Albert, J-P, Prevost, M.E.,
Bruton, D. & Sinclair, F. 2008. Greater Plutonio
project—Subsea flowline design and performance.
Offshore Pipeline Technology Conf., Amsterdam.
Krost K., Gourvenec S.M. & White D.J. 2011. Consoli-
dation around partially-embedded submarine pipe-
lines. Géotechnique, 61(2): 167–173.
Lenci, S. & Callegari., M. 2005. Simple analytical models
for the J-lay problem. Acta Mechanica, 178:23–39.
Low, H.E., Randolph, M.F., DeJong, J.T. & Yafrate, N.J.
2008. Variable rate full-flow penetration tests in intact
and remoulded soil. Proc. 3rd Int. Conf. on Geotech.
Geophys. Site Characterization, Taipei, Taiwan
1087–1092.
Lund, K.M. 2000. Effect of increase in pipeline soil pene-
tration from installation. Proc. of ETCE/OMAE 2000
Joint Conference; Energy of the New Millennium Paper
OMAE2000/ PIPE-5047.
Lunne, T., Berre, T., Andersen, K.H., Strandvik, S. &
Sjursen, M. 2006. Effects of sample disturbance and
consolidation procedures on measured shear strength
of soft marine Norwegian clays. Can. Geotech. J.
43(7): 726–750.
Lunne, T. and Andersen, K.H. 2007. Soft clay shear
strength parameters for deepwater geotechnical
design. Proc. 6th Int. Conf. Offshore Site Investiga-
tion and Geotechnics: Confronting New Challenges and
Sharing Knowledge, SUT, London, Vol. 1: 151–176.
Martin C.M. & White D.J. 2012. Limit analysis of the
undrained capacity of offshore pipelines. Géotech-
nique, 62(9):847–863.
Merifield, R.S., White, D.J. & Randolph, M.F. 2008a.
Analysis of the undrained breakout resistance of
partially embedded pipelines. Géotechnique 58(6),
461–470.
Merifield R., White D.J. & Randolph M.F. 2009. The
effect of surface heave on the response of partially-
embedded pipelines on clay. ASCE J. Geotechnical &
Geoenvironmental Engineering, 135(6):819–829
Murff, J.D., Wagner, D.A. & Randolph, M.F. 1989.
Pipe penetration in cohesive soil. Géotechnique 39(2),
213–229.
Orcina. 2008. OrcaFlex software, Orcina Ltd, Ulverston,
UK.
Palmer, A. 2009. Touchdown indentation of the seabed,
Applied Ocean Research, Vol. 30, No. 3. pp. 235–238.
Pesce, C.P., Aranha, J.A.P. and Martins, C.A. 1998. The
soil rigidity effect in the touchdown boundary layer of
a catenary riser: Static problem, Proc. 8th Int. Offshore
and Polar Engineering Conf., Montreal, pp. 207–213.
Randolph, M.F., Carter, J.P., & Wroth, C.P. 1979. Driven
piles in clay—the effects of installation and subse-
quent consolidation. Géotechnique Vol. 29, No. 4,
361–393.
Randolph, M.F. and White, D.J. 2008a. Upper bound
yield en-velopes for pipelines at shallow embedment in
clay. Géotechnique, 58(4):297–301.
Randolph, M.F. and White, D.J. 2008b. Pipeline embed-
ment in deep water: processes and quantitative
assessment. Proc. Offshore Technology Conference,
Houston, USA. Paper OTC19128-PP.
Randolph M.F. & Gourvenec S.M. 2011. Offshore
Geotechnical Engineering. ISBN 978-0415477444.
560 pages. Taylor & Francis.
Wang D., White D.J. & Randolph M.F. 2009. Numerical
simulations of dynamic embedment during pipelaying
on soft clay. Proc. Conf. on Offshore Mechanics and Arc-
tic Engineering, Honolulu. Paper OMAE2009-79199.
Wang, D, White, D.J. & Randolph, M.F. 2010. Large
deformation finite element analysis of pipe penetra-
tion and large-amplitude lateral displacement. Cana-
dian Geotechnical Journal, 47(8): 842–856.
Westgate, Z.J., White, D.J. and Randolph, M.F. 2009.
Video ob-servations of dynamic embedment dur-
ing pipelaying on soft clay. Proc. Conf. on Offshore
Mechanics and Arctic Engineering, Honolulu. Paper
OMAE2009-79814.
Westgate, Z.J., Randolph, M.F., White D.J. & Li, S.
2010a. The effect of seastate on as-laid pipeline
embedment: a case study, Applied Ocean Research,
Vol. 32, pp. 321–331.
Westgate, Z.J., White, D.J., Randolph, M.F. & Brunning, P.
2010b. Pipeline laying and embedment in soft fine-
grained soils: field observations and numerical simula-
tions, Proc. Offshore Tech. Conf., Paper OTC20407.
Westgate, Z., White, D.J. & Randolph, M.F. 2012. Field
observations of pipeline embedment in carbonate
sediments. Géotechnique, 62(9):787–798.
161
Westgate Z.J., White D.J. & Randolph M.F. 2013.
Modelling the embedment process during offshore
pipe laying on fine-grained soils. Canadian Geotechni-
cal Journal, in press.
White, D.J. & Hodder, M. 2010. A simple model for the
effect on soil strength of remoulding and reconsolida-
tion. Canadian Geotechnical Journal, 47: 821–826.
Zeitoun H.O., Tornes K., Cumming, G., & Branković, M.
2008. Pipeline stability—state of the art. Proc. 27th
Int. Conf. on Offshore Mechanics and Arctic Engineer-
ing, OMAE2008, Estoril, Portugal.
Zhou, H. & Randolph, M.F. 2007. Computational tech-
niques and shear band development for cylindrical
and spherical penetrometers in strain-softening clay.
Int. J. Geomechanics 7(4): 287–295.
Zhou, H. & Randolph, M.F. 2009. Penetration resistance
of cylindrical and spherical penetrometers in rate-
dependent and strain-softening clay. Géotechnique.
59(2):79–86.
Zienkiewicz, O.C. & Zhu, J.Z. 1992. The superconver-
gent patch recovery and a posterior error estimates.
Part 1: The recovery technique. International
Journal of Numerical Methods in Engineering, 33(7):
1331–1364.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

LDFE analysis of installation effects for offshore anchors

and foundations

L. Andresen & H.D.V. Khoa

Norwegian Geotechnical Institute, Oslo, Norway

ABSTRACT: During installation of offshore anchors and foundations the soil undergoes large defor-
mations. Effects of installation such as penetration resistance, punch-through, changes in contact stresses
and strengths should be accounted for in the design and planning. Typically such effects have been
assessed by semi empirical methods and engineering judgement skills obtained from field and model
testing. Numerical methods for modelling large deformation and penetration problems in geomateri-
als are now available, and although research into new developments and improvements is still ongoing,
these methods can already be used for some design problems. This paper presents and briefly reviews two
modelling approaches: the Arbitrary Lagrangian-Eulerian method and the Coupled Eulerian-Lagrangian
method of Abaqus/Explicit. The CEL method is found to be the most promising of the two and this
method is used to model the penetration of a spudcan foundation. Excellent agreement is obtained with
the penetration resistance and failure mechanism measured in a centrifuge test.

1 INTRODUCTION often used to describe the governing equations are

the Lagrangian and the Eulerian frames.
Installation effects are considered in the design of
– Lagrangian framework: The mesh is attached on
offshore foundations and anchors. Some of the
the material so that they move together in the
design issues related to the installation are: the pre-
entire computational process. When the mate-
diction of total penetration resistance and penetra-
rial deforms, the mesh deforms accordingly.
tion depth, the effect of multilayered soil profiles
The advantages of the Lagrangian formulation
and punch-through failure during the penetration
are: the time history of the field variables at
process, and the build-up of pore pressure and
a material point can be easily tracked and an
strength weakening caused by the large deforma-
irregular mesh can be conveniently modelled.
tions imposed during installation.
The main disadvantage is when the mesh is
The objective of this paper is to present the Large
heavily distorted, the accuracy of the solution
Deformation Finite Element (LDFE) analysis as a
will be severely affected. In addition, the time
suitable method to model penetration and installation
step, which is controlled by the smallest element
problems in offshore geotechnics. A brief overview of
size, can become too small to be efficient for the
numerical methods used for large deformation mod-
time marching, and may even lead to the diver-
elling is given in section 2. Section 3 contains a brief
gence of the computation procedure.
review and comparison between two different model-
– Eulerian framework: Contrary to the Lagrangian
ling techniques: the Arbitrary Lagrangian-Eulerian
mesh, the Eulerian mesh is fixed in space and
method and the Coupled Eulerian-Lagrangian
does not change with time while the material
method. Modelling issues such as boundary effect,
flows across the mesh. Therefore, it is difficult
discretisation errors and loading rate effect are dis-
to analyze the time history of field variables at
cussed and recommendations given.
a fix point on the moving material. Expensive
In particular we have found the Coupled
numerical mapping is required to accurately
Eulerian-Lagrangian method of Abaqus to be a suit-
treat the irregular mesh. The major advan-
able tool and this is demonstrated in section 4 by back-
tage of this method is, since the Eulerian mesh
analysis of a centrifuge test of spudcan penetration.
remains during motion, large deformations of
the material do not cause any deformation of
the mesh itself.
2 COMPUTATIONAL METHODS
Various techniques have been developed
In traditional continuum mechanics based finite ele- based on the two frameworks in order to over-
ment formulations, two fundamental frameworks come numerical difficulties in LDFE analysis.

162
Among the techniques proposed used in whereas the soil is discretised with the Eulerian
geotechnical engineering are: Updated Lagrangian elements. The effect of different FE discretisations
(UL), Material Point Method (MPM) and Smooth will be considered in section 3.3.
Particle Hydrodynamics (SPH). The different but The interaction between the pile and the soil is
complementary features of the Lagrangian and modelled as frictionless (smooth).
Eulerian frame suggest that it would be compu- If not otherwise specified, all analyses are
tationally beneficial to combine these two so as conducted with a constant penetration rate of
to strengthen their advantages and to avoid their 0.1 m/s.
disadvantages. This idea has led to the develop- Generally the authors advocate a stepwise mod-
ment of three approaches that apply both the elling approach from a simple 2D/Axisymmetric
Lagrangian and Eulerian formulation: Arbitrary FE-model which preferably can be validated
Lagrangian-Eulerian (ALE), Remeshing and Inter- against known analytical results towards a full 3D
polating Technique with Small Strain (RITSS) and FE-model with the full complexity of soil layering
Coupled Eulerian-Lagrangian (CEL). etc. Table 1 summarises the different LDFE analy-
This paper mainly focuses on application of the ses performed in this section. It’s worth noting that
ALE and CEL methods which are available in the the CEL method implemented in Abaqus/Explicit
commercial finite element program Abaqus. can only be used with the three-dimensional brick
elements.
3 LDFE MODELLING APPROACH
3.2 Effect of boundary
3.1 General description
An ALE axisymmetric FE-model with a radius of
A 5 m long cylindrical “anchor pile” with diameter 10 m (equivalent to 25D) from the centre line and
D = 40 cm and a tapered tip, which is penetrating a depth of 10 m is established to study the bound-
into a weightless uniform clay domain, is mod- ary effect. Figure 1 plots the radial displacements
elled using both the ALE and the CEL methods calculated at two different depths of 4 m and 5 m
in Abaqus/Explicit. During the penetration proc-
ess the clay is assumed to be undrained with an
undrained shear strength su of 5 kPa and rigidity Table 1. Summary of different LDFE analyses of
index G/su of 200. The clay material behaviour is anchor pile.
described by the Tresca constitutive model.
Both axisymmetric and full three-dimensional FE-models Modelling Different effects
FE-models are considered. The objective is to (Abaqus/Explicit) techniques studied
compare the two methods and to present and Axisymmetric ALE Boundary
discuss some modelling issues common to the Axisymmetric ALE Discretisation
LDFE analysis. Owing to axisymmetry, only a Three-dimensional ALE Penetration rate
15°-domain of the 3D problem is modelled so that Three-dimensional CEL CEL versus ALE
the computational time is optimized without caus-
ing excessive skewing of the elements close to axis
of symmetry.
In all the FE-models fully fixity is applied at the
base of the soil domain and roller conditions at its
vertical sides. At the axisymmetric line while the
roller conditions are used in the CEL FE-model,
a zipper-technique (Cudmani & Sturm 2006) is
employed in the ALE FE-model. This technique is
necessary to improve contact between pile and sur-
rounding soil during large penetration. For the 3D
FE-models, the symmetric boundary conditions
are in addition imposed on the two planes of sym-
metry by constraining the two in-plane rotations.
In the ALE FE-models the anchor pile is mod-
elled by analytical rigid elements. The soil domain
is discretised with Lagrangian elements except for
the soil within a depth of 7.5 m and a radius of
3 m from the axisymmetric line where the ALE
elements are applied. In the CEL FE-model, the Figure 1. Effect of boundary on LDFE analysis of
pile is modelled by Lagrangian rigid solid elements anchor pile.

163
versus the radius of the FE-model at the end of
the anchor pile installation, i.e. the tip at 5 m deep.
It can be seen that from a radius of 5 m (12.5 D)
the effect of the side boundary is considered to
be insignificant with respect to the soil responses.
Therefore, the radius of 5 m will be adopted for all
following simulations.
3.3 Effect of discretisation
The effect of discretisation is investigated using an
axisymmetric FE-model with the ALE method.
A total of six mesh densities with element size
ranging from 0.05 D to 0.3 D as presented in
Figure 2 are generated and run with the same
constant penetration velocity of 0.1 m/s. Figure 3
plots the calculated end bearing pressures which all
agree fairly well with the analytical solution for the
bearing capacity of a cone penetration in cohesive
Figure 2. Six different FE meshes used for studying
effect of discretisation on LDFE analysis of anchor pile.
Figure 3. Computation times (hours) and calculated
end bearing pressures for six different discretisations.
164
soils given by Liyanapathirana (2009) and Yu et al.
(2000). From the results presented in Figure 3, it is
seen that the coarser meshes give faster run times
but also less accurate solutions. The mesh density
of 0.1 D below the pile tip is considered to be suffi-
cient fine in order to satisfy both the accuracy and
the efficiency of the simulations. Thus, it will be
used in the following 3D FE-models.
3.4 Effect of penetration rate
In most real installation problems the rate of pen-
etration is so slow that dynamic effects can be
disregarded and the problem can be regarded as
being static. Solving a static problem with Abaqus/
Explicit requires choosing a penetration velocity
that is artificial but which does not introduce iner-
tia effects of significance for the solution. Slower
velocity, however, requires longer computation
time. The aim is to find the highest penetration
rate that gives acceptable accuracy, i.e. gives insig-
nificant inertia.
An ALE 3D FE-model shown in Figure 4(a)
is created. Its geometric dimensions and discreti-
sation are decided based on the calculated results
discussed in sections 3.2 and 3.3. Three installa-
tion rates of 1, 0.75 and 0.5 m/s are considered.
Figure 3 reveals that the corresponding computa-
tion time required for a penetration depth of 5 m is
approximately 59, 74 and 108 minutes, respectively.
Figure 5 plots the calculated end bearing pressures
versus the penetration depth. It is shown that all
Figure 4. Geometry and discretisation of ALE 3D FE-
model (a) and CEL 3D FE-model (b).

Figure 5. Comparison of empirical and calculated bear-
ing pressures of anchor pile. Effect of loading rate.
ALE 3D results are almost identical to the ALE
axisymmetric results obtained with a relatively
slow installation rate of 0.1 m/s. Hence the pen-
etration rate of 1 m/s will be adopted in all subse-
quent analyses.
It has been demonstrated that the ALE adap-
tive meshing in Abaqus/Explicit is suitable to deal
with this pile penetration large deformation prob-
lem. Other valuable advantages of using the ALE
method in Abaqus are:
– The method is available for both Abaqus/Explicit
and Abaqus/Standard, which largely expands its
application fields such as: steady-state transport
analysis, couple pore fluid flow and stress analy-
sis as well as couple temperature-displacement
analysis;
– It can be used in planar, axisymmetric as well as
three-dimensional boundary value problems;
– Mass scaling is supported so that the computa-
tional efficiency can be significantly improved
while retaining the necessary degree of accuracy
of solutions.
The main limitation of the ALE method in
Abaqus is that the “physical” boundary between
two different materials is always associated with a
Lagrangian boundary region, i.e. a non-adaptive
mesh boundary. In other words, no material can
flow through the boundary regions. Therefore, this
method cannot handle layered soils where large
deformations can occur at the material bound-
ary regions. To overcome this drawback the CEL
method in Abaqus/Explicit will be used.
Before applying the CEL method to LDFE
analysis including a more complex structural
geometry and multilayered soils, the same anchor
165
pile installation problem as used with the ALE FE-
model is re-analysed using the CEL approach. The
comparison results are discussed in next Section.
3.5 CEL method versus ALE method
Figure 4(b) presents a CEL 3D FE-model of the
anchor pile. By disregarding the void layer, the dis-
cretisation in the remaining part of the FE-model
is nearly identical to the ALE 3D FE-model (8911
compared to 9044 elements). The purpose of the
void layer is just to capture the soil heave during
the pile installation process. The penetration rate
of 1 m/s is used.
Figure 5 plots the calculated end bearing pres-
sure of the pile versus the penetration depth. It can
be seen that the CEL result agree rather well with
the both the empirical prediction and the ALE
result. With regard to computational efficiency,
the CEL analysis takes slightly longer time than
the ALE one (64 compared to 59 minutes).
Through the comparisons above, it can be con-
cluded that similar to the ALE method, the CEL
method is suitable for the LDFE analysis. Hence
in the next section, the CEL technique in Abaqus/
Explicit will be applied to back-analyse the cen-
trifuge test E2UU-II-T5 of a spudcan foundation
penetrating into stiff over soft clay layers carried
out by Hossain and Randolph (2010a).
4 APPLICATION TO A SPUDCAN
PENETRATION PROBLEM
4.1 General description
The spudcan is a circular or polygonal conical
foundation commonly used in the offshore oil and
gas industry as footings for the three independ-
ent legs of mobile jack up rigs. Different types of
spudcans and installation methods are described in
Menzies & Spoor (2008). The spudcan foundation is
penetrated by up to three diameters into the seabed
and pre-loaded by ballasting the hull of the rig.
LDFE analyses of spudcan penetration have
previously been presented by Hossain & Randolph
(2010b), Wang & Carter (2002) and Yu et al.
(2012) using the Remeshing and Interpolating
Technique with Small Strain (RITSS) developed
by Hu & Randolph (1998). Qiu & Henke (2011)
and Tho et al. (2012) have recently applied the
CEL method in Abaqus to simulate large instal-
lation of spudcan foundations. These contribu-
tions study, among other effects, the effect of
punch-through failure of spudcans on stiff clay
overlaying soft clay and loose sand overlaying clay.
Qiu et al. (2011) from the group at the Hamburg
University of Technology have also demonstrated
the applicability of the Abaqus CEL method for
geotechnical large deformation problems.
This section focuses on applying the CEL
method in Abaqus/Explicit to back-analyse
the centrifuge test E2UU-II-T5 carried out by
Hossain & Randolph (2010a). In this laboratory
test, a spudcan of prototype diameter 6 m was
installed into a stiff clay (Clay 1), with a thick-
ness of 4.5 m, overlaying a soft clay (Clay 2).
Figure 6 shows a CEL 3D FE-model of the centri-
fuge test and its discretisation.
The undrained shear strengths and the rigidity
indexes of the two clay layers are adopted from the
T-bar test given by Hossain & Randolph (2010a).
The soil behaviour is described by the Tresca consti-
tutive model of which the parameters are detailed in
Table 2. In order to account for the effect of strain
softening due to the soil gradually being remoulded
as it flows around the spudcan and the effect of strain
rate, Randolph et al. (2008) suggested to reduce the
undrained strength by about 15%. Therefore, only
Figure 6. Geometry and discretisation of CEL 3D FE-
model of spudcan.
Table 2. Tresca parameters of two clay layers used in
CEL 3D FE-model for back-analysis of centrifuge test
E2UU-II-T5.
Value
Parameter Unit Clay 1 Clay 2
Submerged unit weight, γ ′ kN/m3 8.0 7.4
Rigidity index, G/su – 500 500
Poisson’s ratio, ν – 0.49 0.49
Undrained shear strength, su kPa 38.3 11.0
166
85% of the strengths given in Table 2 are applied to
the CEL 3D FE-model.
The contact between the soil and the spudcan
foundation is modelled as smooth. The numeri-
cal penetration rate of 1 m/s is adopted, which
is assumed slow enough such that the calculated
results are not affected by inertia effects.
4.2 Results
Figure 7 presents a comparison of the numerically
predicted bearing pressure against the centrifuge
result. The penetration depth d is defined as zero
from the level at which the maximum contact area
of the spudcan is at the original soil surface. Hence
the normalised penetration depth d/D starts from
negative value. It can be seen from the figure that
the CEL FE-model predicts very well not only
the magnitude of the punch-through load (peak
load) but also the depth where the punch-through
failure occurs (i.e. stage A). An excellent agreement
between the numerical and experimental spudcan
responses is observed until the normalised depth
d/D = 1. For penetration depth beyond d/D = 1
the two spudcan responses gradually diverge from
each other. This mismatch could be explained by
the fact that when the spudcan is displaced closer
to the container base at 11.5 m (equivalent to
d/D ≈ 1.92), the bottom boundary effect become
more and more prominent. The bottom bound-
ary in the CEL FE-model is, however, taken deep
enough (d/D = 6.67) so that the spudcan response
is not affected.
Figure 8 shows the deformed soil layers and flow
mechanisms depicted by the velocity fields at three
installation stages: stage A (d/D = 0.07), stage B
(d/D = 0.75) and stage C (d/D = 1.6). The numeri-
cal results are also compared with the centrifuge
Figure 7. Comparison of measured and calculated
bearing pressures of spudcan.

Figure 8. Comparison of experimental and numerical
spudcan failure mechanisms at three different penetra-
tion depths for test E2UU-II-T5 performed by Hossain
and Randolph (2010a).
model test results. Excellent agreement in terms
of the flow mechanism including surface heave,
soil displacements, failure pattern can be found at
stages A and B. Figure 8(a) shows that at the shal-
low penetration depth stage A, a stiff clay wedge
of truncated cone shape is formed from the edge of
the spudcan based towards the interface between
167
the two clay layers. Potential punching failure sur-
faces are observed and depicted by the dark red
solid lines. It is recalled that Stage A corresponds
to the peak spudcan resistance (see Fig. 7). In con-
sequence of the punch-through failure, the subse-
quence penetration leads to a significant reduction
of the calculated bearing capacity of the spudcan
foundation. When continuing to displace the spud-
can deeper down till Stage B at which the spud-
can base is at the original stiff/soft clay interface,
the stiff clay plug trapped beneath the spudcan is
gradually reformed in an inverted truncated cone
as shown in Figure 8(b). The underlying soft clay
is correspondingly forced to squeeze outward
because of the inward sidewalls of the inverted
truncated cone containing the stiff clay. The stiff
clay wall above the spudcan edge is seen to remain
standing thanks to its high cohesion. Figure 8(c)
shows that at Stage C, a complete deep penetration
mechanism is revealed. Very good agreement in
terms of the soil back-flow pattern and the cavity
depth are observed from the CEL FE-model and
the centrifuge test model results. It is also found
from Figure 8(c) that at this Stage the numerical
simulation predicts a localize failure mode while
it is a more diffuse type of failure observed in the
laboratory test. The difference of the failure modes
can be attribute to the fact that since the penetra-
tion depth is very close to the container base of the
centrifuge test, the measurement results, especially
beneath the spudcan base, are most likely affected
by the boundary condition. The observation
clarifies further the divergence of the two bearing
pressure results obtained beyond the normalised
penetration depth d/D = 1, which is presented in
Figure 7.
5 CONCLUSIONS
The capability of the Large Deformation Finite
Element (LDFE) in modelling of penetration
and installation effects of offshore anchors and
foundations has been investigated in this paper.
Both the Arbitrary Lagrangian-Eulerian (ALE)
method and the Coupled Eulerian-Lagrangian
(CEL) method available in Abaqus/Explicit were
successfully used to simulate deep penetration of
a typical anchor pile from the free surface into
uniform clay. The pile resistance calculated by the
two methods agrees very well with the empirical
prediction. Through the comparison between the
two approaches, it was found that in the Abaqus
program, the ALE method is more flexible since it
can be applied in planar, axisymmetric and three-
dimensional boundary value problems whereas
only the later is available for the CEL method. In
addition, for the same discretisation and loading
rate, the ALE simulation runs slightly faster and
predicts more accurately the pile bearing pressure
than the CEL one. However, the ALE method in
Abaqus does not support the boundaries between
different materials as they are considered as non-
adaptive mesh (or Lagrangian) boundaries. There-
fore, in order to back-analyze the centrifuge test
E2UU-II-T5 of a spudcan foundation penetrating
into stiff clay overlying soft clay, which is a rather
complex boundary value problem in terms of struc-
tural geometry and layered soils, the CEL method
has been applied instead. Through the compari-
son results it was found that except for some later
stages in the centrifuge test where the measured
results were most likely affected by the container
base boundary, the numerical predictions were in
excellent agreement with the experimental observa-
tions. In fact, the CEL 3D FE-model was able to
not only calculate the spudcan foundation resist-
ance but also predict and capture well the failure
mechanisms experimentally observed during the
penetration process.
ACKNOWLEDGEMENT
The research has been performed as a part of
“GEO-INSTALL” (Modelling Installation Effects
in Geotechnical Engineering) project. The work
is partially funded by the European Community
through the program “Marie Curie Industry-
Academia Partnership and Pathways”, under
Contract No PIAP-GA-2009-230638 and partially
supported by the Norwegian Research Council.
All these supports are gratefully acknowledged.
REFERENCES
Cudmani, R. & Sturm, H. 2006. An investigation of the
tip resistance in granular and soft soils during static,
alternating and dynamic penetration. In H. Gonin, A.
Holeyman, and F. Rocher-Lacoste (Eds.), TransVib
2006: International Symposium on Vibratory Pile
Driving and Deep Soil Compaction 221–231.
168
Hossain, M.S. & Randolph, M.F. 2010a. Deep-
penetrating spudcan foundations on layered clays:
Centrifuge tests. Géotechnique 60(3): 157–170.
Hossain, M.S. & Randolph, M.F. 2010b. Deep-
penetrating spudcan foundations on layered clays:
Numerical analysis. Géotechnique 60(3): 171–184.
Hu, Y. & Randolph, M.F. 1998. A practical numerical
approach for large deformation problems in soil. Int J
Numer Anal Methods Geomech 22(5): 327–350.
Liyanapathirana, D.S. 2009. Arbitrary Lagrangian
Eulerian based finite element analysis of cone pen-
etration in soft clay. Computers and Geotechnics 36:
851–860.
Menzies, D. & Roper, R. 2008. Comparison of jackup rig
spudcan penetration methods in clay. Proc Offshore
Technology Conference OTC-19545.
Qiu, G. & Henke, S. 2011. Controlled installation of
spudcan foundations on loose sand overlying weak
clay. Marine Structures 24(4): 528–550.
Qiu, G., Henke, S. & Grabe, J. 2011. Applications of a
Coupled Eulerian-Lagrangian approach on geotechni-
cal problems involving large deformations. Computers
and Geotechnics 38(1): 30–39.
Randolph, M.F., Wang, D., Zhou, H., Hossain, M.S., &
Hu, Y. 2008. Large deformation finite element analy-
sis for offshore applications. Proc 12th Int Conf of Int
Association for Computer Methods and Advances in
Geomechanics 3307–3318. Goa, India.
Tho, K.K., Leung, C.F., Chow, Y.K. &
Swaddiwudhipong, S. 2012. Eulerian finite-element
technique for analysis of jack-up spudcan penetration.
Int J Geomech 12(1): 64–73.
Wang, C.X. & Carter, J.P. 2002. Deep penetration of
strip and circular footings into layered clays. Int J
Geomech 2(2): 205–232.
Yu, H.S., Herrmann, L.R. & Boulanger, R.W. 2000.
Analysis of steady state cone penetration in clay.
J Geotech Geoenviron Eng ASCE 126(7): 594–605.
Yu, L., Hu, Y., Liu, J., Randolph, M.F. & Kong, X. 2012.
Numerical study of spudcan penetration in loose sand
overlying clay. Computers and Geotechnics 46: 1–12.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Investigation into the effect of pile installation on cyclic lateral

capacity of monopiles

T. de Blaeij & J. Dijkstra

Delft University of Technology, Delft, The Netherlands

ABSTRACT: Monopiles are being used more extensively for offshore wind turbine foundation. Research
into the behaviour of these large open-ended piles generally do not examine the installation effect on
lateral capacity. This paper presents the results of model pile load tests on monopiles in the geotechni-
cal centrifuge. The effect of monotonic pile installation on the subsequent cyclic lateral capacity were
investigated. In order to study these effects, a novel actuator was developed which simultaneously allows
to install the piles in-flight and is able to load the piles laterally without interrupting the test. A series of
initial tests on open-ended monopiles, which show the effectiveness of the setup, are presented. These
initial tests investigated the effect of 1 ⋅ g and N ⋅ g pile installation on the subsequent lateral capacity in
two way cycling for two different initial densities. The preliminary results indicate that N ⋅ g installation
has a small positive effect, which decays with the number of cycles, on the lateral capacity of short stiff
monopiles. Hence, for open-ended model piles the installation effects are negligible. At the present stage,
given the accuracy of the experimental setup, no conclusive results for the initial stiffness response were
gathered. However, tentatively, the conclusions for the lateral capacity also hold for the initial stiffness.
The latter needs to be confirmed in future experiments, where the setup is optimized for the measurement
of the small strain response of the monopile and a large number of loading cycles. These conclusions
are for tests on open-ended model piles only. If for similitude a close-ended model pile will be used the
pile installation stage could be omitted in order to better approach the negligible installation effects as
presented in this paper.

1 INTRODUCTION The p-y method (Reese et al. 1974, Murchison &

In general, offshore structures are used for the
oil, gas and offshore wind farm industry. Due to
placement of wind farms at an increasing dis-
tance from the coast, environmental conditions
have become harsher in the recent years (Clauss
et al. 1988). Several types of foundations exist for
offshore wind turbines. The choice of which foun-
dation to be used depends on the site and loading
conditions. Nowadays non-slender (stiff) mono-
piles piles are used with a diameter in the range of
4–6 m driven 20–30 m into the seabed. These piles
are non-slender because of their low length (L)/
diameter (D) ratio of about 5.
Wind turbines are extremely sensitive to transla-
tion and vibrations of the wind turbine. The latter
could easily affect the efficiency and the service-
ability of the turbine. Therefore, the serviceability
limit for rotation is generally 0.5° where 0.25° is
allowable rotation from the installation and 0.25°
rotation is due to loading (Vattenfall 2008). In
order to meet these requirements an adequately
designed foundation is required to transfer loads
from the wind turbine to the surrounding soil.
O’Neill 1983) is often used for the design of these
monopile foundations. However, this method is
based on long slender piles with diameters < 2 m
and a length over diameter ratio L/D of 34.4.
Several other design standards such as API (2007),
DNV (2011) and GL (2005) rely on this method.
However, large diameter monopiles are considered
to be short and rigid which rotate rather than bend
when subjected to lateral loads (Byrne et al. 2010,
Achmus et al. 2007). As a result recently, numer-
ous studies have been performed on the response
of monopile foundations. These studies typically
carried ouut the experiments in the laboratory at
1 ⋅ g (Dührkop & Grabe 2008, Byrne et al. 2010,
Peng et al. 2011, Cuéllar et al. 2012) and at N ⋅ g
in the geotechnical centrifuge (Barton et al. 1983,
Oldham 1985, Dyson & Randolph 2001, Brant &
Ling 2007, Rosquoet et al. 2007, Klinkvort et al.
2010, Li et al. 2010, Alderlieste et al. 2011, Bienen
et al. 2011). Only in few studies the piles were
installed in-flight and subjected the pile to lateral
load without stopping the centrifuge. However, as
shown by Craig (1985), White (2004) and Dijkstra
(2009) the installation effects of single close-ended

169
piles are significant and should be incorporated in 2.2 Mechanical features
the tests. Unfortunately, the latter is complicated
The aim for the new design of the actuator is to
by the scaling conditions. Especially, in cases where
allow for the simulation of the installation stage,
a closed-ended pile is used instead of an open-
whilst at the same time offering possibility to apply
ended pile, in order to properly scale the lateral
a lateral load on the pile head after installation
stiffness, this could have an over pronounced effect
has fin-ished. The load application of the mecha-
on the resulting lateral capacity. Hence, the effect
nism on the free pile head is designed such that no
of the pile installation on the subsequent lateral
bending moment is transferred to the pile head. To
pile response is still not completely understood.
accomplish this a transversal rod through the top
Current paper presents novel results of the
of the model pile which fits in the template of the
effect of the installation of an open-ended pile
actuator acts as a hinge at the pile head. The pin
in sand on the subsequent cyclic lateral capacity.
only falls into place in the template after the full
A series of geotechnical centrifuge tests were per-
pile installation stage has been accomplished. The
formed to investigate the effect of pile installation
actuator is illustrated in Figure 2. This new actua-
at low stress level (1 ⋅ g) and at large stress level in
tor allows for 106 mm vertical penetration of the
the geotechnical centrifuge (N ⋅ g). The cyclic lat-
model pile and 55 mm lateral displacement, which
eral load tests have in both cases been executed at
is about two times the outer pile diameter (Do is
N ⋅ g. These tests require a novel actuator, capable
27.36 mm). The displacement of the pile head is
of both pile installation and lateral loading with-
measured by registering the absolute position of
out fixing the pile head and without interruption
the actuator. At present no load cell is applied in
of the test.
between the actuator and the pile head to maxi-
mize the penetration length.
The template in the actuator which catches the
2 EXPERIMENTAL SETUP pile is instrumented with strain gauges to meas-
ure the lateral load on the pile head. The loca-
2.1 Geotechnical centrifuge tion of the strain gauges is chosen such that the
The geotechnical centrifuge at Delft University load measurement is insensitive for vertical loads
of Technology is a small beam centrifuge with
a radius of 1.22 m and is equipped with all the
necessary data acquisition and camera facili-
ties for contemporary centrifuge testing (Fig. 1).
The flight computer (ITX computer board with
National Instruments PCI-6221 data acquisition
card), strain gauge amplifiers (custom made 16
channels, 1x-10000x gain, <100 kHz) and motor
control (3 channels DC-servo) are all hosted on
the centrifuge. The communication to the com-
puter is by means of a wireless network link or
wired over the slip rings. The network camera has
its own set of slip rings.

Figure 2. Load mechanism installed on load frame,

Figure 1. Delft University of Technology centrifuge. scaled monopile and actuator.

170
and bending moments in the horizontal and ver-
tical plane. After construction of the frame it has
been calibrated for static loads in the laboratory.
However, the complete system as constructed in
the centrifuge, is more extensively calibrated at 1 ⋅ g
and N ⋅ g in order to deal with internal variation and
hysteresis of the stress and strain in the mechanism
resulting from cyclic loading at 1 ⋅ g and N ⋅ g. The
cyclic loading in compression has been calibrated
with a mechanical spring, due to the remaining hys-
teresis in the mechanism, the calibration of the test
data has been done with a typical hysteresis loop
as presented in Figure 3. Practical difficulties only
allowed the frame to be calibrated in compression.
However, the linearity of the static calibration of the
mechanism in tension and compression give confi-
dence to use this cyclic calibration also for tension.
Besides hysteresis also the stiffness of the actuator
is taken into account for the correction of displace-
ment which is measured on the actuator spindle.
The brass model pile, depicted in Figure 4,
is designed to properly scale the lateral bending
stiffness (Alderlieste 2011) and to prevent plugging
during installation (large Do/D50). The maximum
height of the roughness profile (Rz) of the pile Figure 4. Model pile.
is 1.54 μm and the mean deviation of the rough-
ness profile (Ra) is 0.37 μm. The low normalized
roughness (Rn) of 0.007 implies a smooth inter-
face and dilative behavior of the interface shear
band is expected to be low (Paikowsky et al. 1995,
Garnier & König 1998). The strongbox with actua-
tor and relevant physical dimension are depicted in
Figure 5. Dimensions such as eccentricity (e, 0.71 ⋅
Do), distance to vertical boundary (B1, 3.85 ⋅ Do),
distance between bottom and pile tip (B2, 1.3 ⋅ Do),
embedded pile length (Lemb, 5 ⋅ Do) and installed
pile length (A1, 3.9 ⋅ Do) are kept constant during
the experiments.

Figure 5. Schematic cross-section of the strongbox.

3 TEST PROCEDURE

3.1 Sample preparation

Figure 3. Specific frame hysteresis loop during cyclic
calibration.
The model pile tests are performed in a homogene-
ous single layer of dry sand with two different rela-
tive densities of 60 ± 2% and 80 ± 2% (Table 2). The
sand consist of uniformly graded quartz grains,
with a D50 of about 230 μm, the latter is a reason-
able approximation for North-Sea soil conditions.
Medium dense samples with a relative den-
sity Id of 60 ± 2% are prepared by incrementally

171
Table 1. Test execution sequences for different installa- where in each cycle the pile head displacement is act-
tion conditions. ing in two opposing directions, has been applied.
1 ⋅ g installation 48 ⋅ g installation
3.3 Test program
Monotonically jack the Spin up of the centrifuge
pile 106 mm at 0.5 mm/s to 48 ⋅ g The test program was designed such that the dif-
Retract the actuator Monotonically jack the pile ference in lateral cyclic response between 1 ⋅ g and
106 mm at 0.5 mm/s N ⋅ g installation of an open-ended pile can be
Spin up of the centrifuge Retract the actuator identified. The initial density of the sample has
to 48 ⋅ g been varied to gain more insight into the influ-
ence of the initial soil state on the installation
Cyclical loading with a displacement amplitude of 1 mm, effect in subsequent lateral loading. Each test is
a rate of 0.23 mm/s and 200 cycles. performed in a newly prepared sample. Given the
very large pile diameters of the monopiles where
Table 2. Test details.
plugging is not likely to occur, at the present stage
plugging of the open-ended pile should not occur
Test code Installation Id Cycles in the model test either. Preliminary tests show
that even at higher acceleration levels no plugging
T01-80-1 ⋅ g 1⋅g 60 ± 2% 200 did occur.
T02-80-N ⋅ g 48 ⋅ g 60 ± 2% 200
T03-60-N ⋅ g 48 ⋅ g 80 ± 2% 200
T04-60-1 ⋅ g 1⋅g 80 ± 2% 200 4 RESULTS

The results of the cyclic lateral loading stage for

filling the strongbox using a funnel. Material is
dispersed on top of a sieve which is subsequently
retracted after an incremental filling step, cre-
ating a very loose sample. After the necessary
amount of soil is dispersed in the strongbox, the
box is vibrated in order to densify the soil to the
desired target density. There is no seggregation
observed in the samples, due to the uniformity
of the sand. The reproducibility and consistency
of this method proves to be satisfactory for these
relatively small sample sizes. Dense samples with
a relative density of Id of 80 ± 2% are prepared
in a similar manner but now by vibrating the
filled strongbox for a longer duration with a sur-
charge on top of the material until the material
sufficiently densified within the required target
volume.
each test are summarized in Figure 6. Of each
cycle the peak load and peak displacement (cor-
rected for deformation of the actuator) are plotted
against the number of cycles. The peak loads show
that, as expected, the maximum capacity increases
with increasing density. Also the main gain in lat-
eral capacity is yielded in the first 20 cycles (see
Fig. 7). The latter indicates a progressive stiffen-
ing of the lateral response. Differences between the
1 ⋅ g and N ⋅ g loading are modest, i.e. a pile which
is installed at 48 ⋅ g yields up to 5–10% more lateral
capacity which decays as function of the number
of loading cycles to about 0–5%. The displacement
peaks indicate that the actuator has a small drift in
its zero position. The position error grows to about
0.007Do after 200 load cycles. Also, due to the

3.2 Test stages

During the tests two different execution sequences,
in which mainly installation conditions vary, have
been carried out. Table 1 presents the sequence for
both low stress level installation (1 ⋅ g) as well as full
stress level installation (48 ⋅ g in the middle of the
sample). The aim of the tests was to compare the
effect of 1 ⋅ g installation to that of N ⋅ g installation.
Therefore, as presented in Table 1, after installation
in all tests a similar cyclic lateral loading scheme
has been implied.
The loading was by means of a computer con-
trolled displacement amplitude, loading frequency Figure 6. Peak loads (black) and displacements (gray)
and number of cycles. In all tests two-way cycling, for tests presented in Table 2.

172
correction of the displacement data for the stiff-
ness of the actuator, the corrected displacement
which is actually applied on the pile head reduces
as the load increases.
In order to better understand the effect of
installation on the lateral pile response the full 1st,
2nd, and 150th loading loops have been presented
in detail in Figures 8–10. Each Figure shows the
same cycle number for each test as function of
the initial relative density and the pile installation
method (1 ⋅ g and 48 ⋅ g). All these cycles have been
corrected for stiffness of the loading mechanism
and hysteresis. For ease of comparison the initial
offset at the beginning of the 150th cycle has been
zeroed.
The results indicate that at the start of the test,
i.e. in the first load cycle (Fig. 8) the differences Figure 9. Cycle 2 for tests presented in Table 2.
between 1 ⋅ g installation and 48 ⋅ g are large. The
negative load at the start of the cycle, for the 48 ⋅ g
test, most probably originate from pre-loading
of the actuator as a result of inclined pile instal-
lation during in flight installation. Although the

Figure 10. Cycle 150 for tests presented in Table 2.

Figure 7. First 20 peak loads (black) and displacements
(gray) for tests presented in Table 2.
Figure 8. Cycle 1 for tests presented in Table 2.
maximum loads are quite similar for all tests, the
stiffness of the pile-soil at the end of the primary
displacement do show a higher stiffness for the
piles installed at 48 ⋅ g. Load reversal in backward
direction shows a higher lateral resistance in case
of 48 ⋅ g installation. The latter is still visible for
backward loading direction in the second cycle,
as presented in Figure 9. Figure 10 shows a very
consist response for both 1 ⋅ g and 48 ⋅ g installa-
tion, i.e. the differences in response resulting from
pile installation decay as function of cycle number.
However, the differences between the tests per-
formed at different initial densities remain. Also,
a progressive stiffening of the pile-soil response
becomes apparent when the second cycle within
one test is compared to the 150th cycle.
Typical for all the cycles is the shape of the
loop. It seems that the passive earth pressure in
loading is being mobilized gradually whereas

173
the active earth pressure decreases more rapidly
after reversal of displacement and thus loading
direction.
5 CONCLUSIONS
A series of model pile tests on open-ended mono-
pile in medium dense and dense sand have been
performed. The effect of pile installation at 1 ⋅ g and
at 48 ⋅ g (in flight during the centrifuge test) on the
subsequent cyclic lateral capacity at elevated stress
level (48 ⋅ g) in sand have been studied for two dif-
ferent initial densities (Id = 60 ± 2% and 80 ± 2%).
Similar to previous model pile tests the results
clearly show a progressive stiffening of the lateral
pile response as function of the number of load
cycles. The N ⋅ g installation has a small positive
effect, which decays with the number of load cycles,
on the lateral capacity of short stiff monopiles.
Hence, for open-ended model piles installation
effects are negligible. At the present stage, given the
accuracy of the experimental setup, no conclusive
results for the initial stiffness response are gathered.
However, tentatively, the conclusions for the lateral
capacity also hold for the initial stiffness. The latter
needs to be confirmed in future experiments, where
the setup is optimized for the measurement of the
small strain response of the monopile and a large
number of loading cycles.
These conclusions are for tests on open-ended
model piles only. If for similitude a closed-ended
model pile will be used the pile installation stage
could be omitted in order to better approach the neg-
ligible installation effects as presented in this paper.
ACKNOWLEDGEMENTS
The first Author of this paper acknowledges the
received financial support from Deltares for pre-
paring the manuscript.
REFERENCES
Achmus, M., Abdel-Rahman, K., & Kuo, Y. 2007.
Numerical modelling of large diameter steel piles under
monotonic and cyclic horizontal loading. In Tenth
International Symposium on Numerical Models in Geo-
mechanics, pp. 453–459. Taylor & Francis London.
Alderlieste, E. 2011. Experimental modelling of lateral
loads on large diameter mono-pile foundations in
sand. M.sc. thesis, Delft University of Technology.
Alderlieste, E.A., Dijkstra, J., & van Tol, A.F. 2011.
Experimental investigation into pile diameter effects
of laterally loaded mono-piles. ASME Conference
Proceedings 2011(44397), 985–990.
174
API 2007. Recommended Practice for Planning, Design
and Constructing Fixed Offshore Platforms—Work-
ing Stress Design. American Petroleum Institute.
ERRATA AND SUPPLEMENT 3, AUGUST
2007.
Barton, Y., Fin, W., Pary, R., & Ikuo, T. 1983. Lateral pile
response and p-y curves from centrifuge tests. Offshore
Technology Conference Paper number OTC 4502.
Bienen, B., Dührkop, J., Grabe, J., Randolph, M., &
White, D. 2011. Response of piles with wings to monot-
onic and cyclic lateral loading in sand. Journal of Geotech-
nical and Geoenvironmental Engineering 138(3), 364–375.
Brant, L. & Ling, H. 2007. Centrifuge modeling of piles
subjected to lateral loads. In H.I. Ling, L. Callisto,
D. Leshchinsky, J. Koseki, and G.M.L. Gladwell (Eds.),
Soil Stress-Strain Behavior: Measurement, Modeling
and Analysis, Volume 146 of Solid Mechanics and
Its Applications, pp. 895–907. Springer Netherlands.
10.1007/978-1-4020-6146-2_68.
Byrne, B., Leblanc, C., & Houlsby, G. 2010. Response of
stiff piles in sand to long-term cyclic lateral loading.
Gotechnique 60(2), 79–90.
Clauss, G., Lehmann, E., & Östergaard, C. 1988.
Meerestechnische Konstruktionen. Springer.
Craig, W. 1985. Installation studies for model piles.
Publication of: Balkema (AA).
Cuellar, P., Georgi, S., Baeßler, M., & Rücker, W. 2012.
On the quasistatic granular convective flow and sand
densification around pile foundations under cyclic lat-
eral loading. Granular Matter, 11–25.
Dijkstra, J. 2009. On the Modelling of Pile Installation.
Ph. D. thesis, Technische Universiteit Delft. ISBN:
9789085704324.
DNV 2011. Offshore standard dnv-os-j101—design of
offshore wind turbine structures.
Dührkop, J. & Grabe, J. 2008. Monopilegrundungen
von offshore-windenergieanlagen–zum einfluss
einer veränderlichen zyklischen lastangriffsrichtung.
Bautechnik 85(5), 317–321.
Dyson, G. & Randolph, M. 2001. Monotonic lateral load-
ing of piles in calcareous sand. Journal of Geotechnical
and Geoenvironmental Engineering 127, 346–352.
Garnier, J. & König, D. 1998. Scale effects in piles and
nails loading tests in sand. In Centrifuge 98.
GL 2005. Guideline for the certification of offshore wind
turbines.
Klinkvort, R., Leth, C., & Hededal, O. 2010. Centrifuge
modelling of a laterally cyclic loaded pile. Physical
Modelling in Geotechnics, 959–964.
Li, Z., Haigh, S.K., & Bolton, M.D. 2010. Centrifuge
modelling of mono-pile under cyclic lateral loads.
7th International Conference on Physical Modelling in
Geotechnics 2, 965–970.
Murchison, J. & O’Neill, M. 1983. An Evaluation of P-y
Relationships in Sands. Report (American Petroleum
Institute). University of Houston-University Park.
Oldham, D. 1985. Experiments with lateral loading of
single piles in sand. Publication of: Balkema (AA).
Paikowsky, S., Player, C., & Connors, P. 1995. A dual
interface apparatus for testing unrestricted friction of
soil along solid surfaces. ASTM Geotechnical Testing
Journal 18(2), 168–193.
Peng, J., Clarke, B., & Rouainia, M. 2011. Increasing Vattenfall 2008. Kriegers flak offshore wind farm-design
the resistance of piles subject to cyclic lateral load- basis foundations. Technical report, Vattenfall Vind-
ing. Journal of Geotechnical and Geoenvironmental kracft AB.
Engineering 137(10), 977–982. White, D.J. & Lehane, B. 2004. Friction fatigue on dis-
Reese, L., Cox, W., & Koop, F. 1974. Analysis of laterally placement piles in sand. Geotechnique 54, 645–658.
loaded piles in sand. Offshore Technology Conference
OTC 2080, 473–486.
Rosquoet, F., Thorel, L., Garnier, J., & Canepa, Y. 2007.
Lateral cyclic loading of sand-installed piles. Soils and
Foundations 47(5), 821–832.

175
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Development of a coupled FEM-MPM approach to model a 3D

membrane with an application of releasing geocontainer from barge

F. Hamad & C. Moormann

Institute of Geotechnics, Stuttgart, Germany

P.A. Vermeer
Deltares, Delft, The Netherlands
Institute of Geotechnics, Stuttgart, Germany

ABSTRACT: In a wide spectrum of geotechnical applications, materials undergo large deformations and/
or large displacements. On modeling these problems with a Lagrangian finite element method, the mesh
can become too distorted and re-meshing is essential. In the past decades, considerable efforts have been
made to adopt what is called meshfree methods to mitigate the problems related to mesh distortion. One
of these methods is the Material Point Method (MPM) that represents the continuum field as Lagrangian
material points (particles), which can move through the fixed background of an Eulerian mesh.
The objective of this paper is to formulate and validate a coupled FEM-MPM approach for the numer-
ical simulation of large deformation of a membrane containing soil e.g. dumping of a geocontainer. In
this approach the membrane is discretised by a surface mesh with accurate computation of the membrane
stresses. This membrane mesh is free to move through a 3D mesh of non-structured tetrahedral elements.
Furthermore, the proposed approach is applied successfully to model geotextile of a geocontainer being
released from a split barge. Frictional contact is defined between the geotextile and the barge. On the other
side, rough contact is assumed between the geotextile and the Mohr-Coulomb soil type inside.

1 INTRODUCTION fall velocity and placing the container (Bezuijen

In recent years, traditional forms of river and
coastal works became very expensive to built and
maintain. Various coastal engineering applications
can be of use from traditional construction materi-
als to the more novel materials such as geotextile.
By the late eighties, a new application for the geo-
textile has been developed then called Geocontain-
ers. A geocontainer is a prefabricated geotextile
placed in a split barge and mechanically-filled with
sand or slurry up to several hundred cubic meters
then dumped from the scow bed in the desired
position. Single and structure of geocontainers
can be constructed for many applications in shore
protection such as breakwaters or to store and iso-
late contaminated materials. These applications of
geotextile and geosynthetics in coastal engineering
are extensively addressed by Pilarczyk (2000).
Generally speaking, the geotextile would expe-
rience high stresses on releasing it from the barge
and when it is hitting the ground. Therefore, study-
ing installation effects of a geocontainer is an
important aspect to be considered. In this regards,
several field and lab tests were accomplished at
Deltares to tackle this issue e.g. measuring strains,
et al. 2002). Nevertheless, numerical modeling of
this problem still has a gap of handling the entire
process.
2 REVIEW OF PREVIOUS WORK
Numerical methods such as the finite element
method (FEM) have the capability for modeling
membranes by using a formulation that resembles
to some extent the formulation of shell elements;
obviously without rotational degrees of freedom.
In many engineering problems large displacements
and deformations occur. When these problems are
modeled with a Lagrangian finite element method,
the mesh can become too distorted and re-meshing
is needed.
Large deformation problems can, however, be
solved using so-called meshfree methods. These
methods trace the history of the state variables at
material points, which are not related to the ele-
ment mesh. The mesh does not deform, and the
problem of severe mesh distortion is overcome.
One of these methods is the particle-in-cell method
(PIC) which was used to model fluid dynamics in

176
the 1960s. Sulsky & Schreyer (1993) adopted the
PIC for solid mechanics and gave it later the name
Material Point Method (MPM). MPM has been
successfully used to model several problems related
to geotechnical applications, such as the discharge
of granular material from a silo (Wieckowski et al.
1999) and pile hammering (Jassim et al. 2011).
York (1997) developed MPM for the modeling
of thin membranes, in which a 2D membrane is
discretised by a collection of material points on its
surface.
By definition, the membrane is a thin-walled
structure that has stiffness in its plane and no stiff-
ness in bending. In addition, the stress over thick-
ness is assumed to be constant. The condition of
constant stress through the thickness of the mem-
brane is automatically enforced by using a single
layer of material points as shown in Figure 1. In this
approach, standard MPM procedure is adopted
except that the stresses of the membrane material
points are computed on local coordinate (ξ, η, ς)
where a plane-stress condition is satisfied, rather
than compute it on global (x, y, z) coordinate.
Fernandes (2004) shows that linear shape func-
tions may result in unphysical stress oscillations
when a material point crosses from one element to
another. Furthermore, the accuracy of the integra-
tion depends mainly on the considered number of
material points inside the element. As an improve-
ment, (Zhang et al. 2011) proposed a hybrid finite
element-material point (FEMP) method for mod-
eling reinforced concrete being subjected to impact
loading. The essential of this method is to intro-
duce a hybrid bar element into MPM, where the
nodal variables are updated from background
nodal displacements of the grid and the stresses
are updated on the element.
Similar to the above FEM/MPM combination
scheme, the membrane can be introduced accu-
rately in a new computational framework that is
valid for low order tetrahedral mesh. To have such
a scheme, the integration should be done accu-
Figure 1. 3D membrane with its MPM representation.
177
rately over the membrane using surface discretisa-
tion. Due to deformation, this mesh would move
through the 3D mesh producing stresses in local
plane. Then, the in-plane stresses are integrated
accurately to have the membrane forces acting on
the membrane nodes. Finally, the computed forces
are mapped on the original tetrahedral mesh, where
momentum equation is solved, by the location of
the considered membrane node.
As in classical finite elements, MPM formu-
lation starts with the weak formulation of the
momentum equation as formulated in Section 3.
Subsequently spatial and time integration form
the general solution scheme of the method. Sec-
tion 4 details the representation of the membrane
using a new approach. Furthermore application of
the new approach on a real problem of a releas-
ing geocontainer from barge is investigated in the
Section 5. Finally, the conclusion and outlook is
summarized in the last section 6.
3 BRIEF REVIEW OF MPM
In MPM the continuum body is discretised by
Lagrangian material points as shown in Figure 2.
The momentum equation is solved on the back-
ground Eu-lerian mesh which provides a conven-
ient means of calculating discrete derivatives.
3.1 Spatial discretisation
The conservation of linear momentum is given by
Šu ∇ + Šg (1)
where σ (x, t) is the Cauchy stress tensor at posi-
tion x and time t, (x, t) is the mass density, g (x, t)
is the gravitational acceleration vector, u (x, t) is
the displacement with the superposed dot denoting
differentiation with time.
Figure 2. Continuum body represented by MPM.
 By taking the virtual displacement δu as test np
function for a domain of volume V surrounded
by boundary S, the weak form of the momentum
mi ∑ mpN (xp ) (7)
p =1
equation can be written as
Referring to Equation 4, the internal force vec-
∫V δ ∫V δδεε dV ∫V δ u
T T T
Š dV ŠggdV tor is given by,
+ ∫S δ uT tdS (2)
t np

where t = σ ⋅ n is the prescribed traction on bound-

F int ∑Vp BT ( x p ) σ p (8)
p =1
ary St, n is the outward unit normal and the script
T denotes the transpose. Similar to the standard where the quotient of the material point mass
finite element method, the value of a variable and density is the volume of the material point,
inside the element can be based on the nodal values Vp = mp / p and B is the strain-displacement matrix,
and the nodal shape functions. For example, the as also used in standard finite element method
displacement vector can be written as (Wieckowski et al. 1999), σp is a vector containing
the stress components at the material point p. The
u( ,t)) d δ u = Nδ a
Na and
Na (3) external nodal force vector is given by,
where N is a matrix containing global shape func- np
tions over the whole computational grid and a is
the nodal displacement. Using these definitions
F ext ∑ m p N T (xx p g ∫S t
N T t dS (9)
p =1
and discretising the momentum Equation 2, it
takes the form (Wieckowski et al. 1999),
3.2 Time integration
M a = F ext F int (4)
So far the momentum Equation 4 is defined and
where M is the consistent mass matrix, ä the nodal needs to be solved for discrete time intervals. With
acceleration vector, F ext and F int are the external the mass matrix being a diagonal matrix, solving
and internal nodal force vectors respectively. The the system of equations becomes trivial. The Eul-
mass matrix is given by er-forward time integration scheme is adopted to
solve this equation, i.e.
np
M ∑ m p N T ( x p )N ( x p ) (5) a t Δt
a t + Δt at , at = [ M lt ] 1 F t (10)
p =1

where np is the number of material points and
mp is the mass of material point p at location xp.
However, in practice, the lumped mass matrix is
preferred over the consistent mass matrix. This
simplifies the computations at the expense of
introducing a slight amount of numerical dissipa-
tion (Burgess et al. 1992). The lumped mass matrix
Ml is the diagonal matrix. The structure of the
lumped mass matrix is as follows
⎡ m1 0 0 ⎤
⎢0 m … 0 ⎥
Ml = ⎢ ⎥
2
(6)
⎢    ⎥
⎢ ⎥
⎣0 0 … mn ⎦
where n is the total number of degrees of freedom
which could increase or decrease within the calcu-
lation as particles move between elements. How-
ever, each component of the diagonal term (mi) in
Equation 6 being the corresponding row sum of
the consistent matrix is performed as
where Δt is the current time increment, at and
at Δt are the nodal velocities at time t and t + Δt,
respectively. The incremental nodal displacement is
obtained by integrating the nodal velocity by the
Euler-backward rule (Jassim et al. 2011)
Δt Δt
at tat (11)
The positions of the particles are subsequently
updated from
Δt Δt
xpt xpt + N p at (12)
where xpt and xpt Δt are the particle positions at time
t and (t + Δt) respectively.
For the present MPM solution procedure, a
slightly different algorithm has been adopted for
updating the particles velocity following Sulsky
et al. (1995).
By solving the equation of motion for the nodes,
the elements deform and the material points in the
interior of the element move in proportion to the

178
motion of the nodes, based on the nodal basis
functions. The position of the material points is
updated using a single-valued continuous velocity
field and, hence the interpenetration of material is
precluded. This automatic feature of the algorithm
allows simulations of no-slip contact between dif-
ferent bodies without the need for special interface
tracking and contact algorithms.
After getting the grid node velocities, the strain
increment Δεp of the material point p at time
(t + Δt) is calculated from,
Δ Δt
Δ εpt Δt B ( x p )aet (13)
where aet Δt is the vector of nodal velocities of
the element e which the material point belongs
to. Based on the calculated strain increment, the
stress increment Δσ pt Δt at each material point
is updated by applying the appropriate constitu-
tive model. The constitutive models are applied at
material points and this allows easy evaluation and
tracking of history-dependent variables. It also
allows computations with more than one material,
since each material point retains its material prop-
erties throughout the computation, for example in
the modeling of soil/membrane interaction.
As the mass of the material point mp is kept
constant during the calculation, the density of the
material is implicitly updated by updating the vol-
ume of the material point through
Δt t Δt
Vpt Vpt ( + Δ v p ) (14)
where Δεv,p is the incremental volumetric strain of
the material point p computed from the incremen-
tal strain components (Δεv = Δεx + Δεy + Δεz) in
(x, y, z) coordinates.
At the end of time (t + Δt) all the material point
variables are updated and a new cycle is begun
using the information carried by the material
points to initialise nodal values on the grid. Note
that at this stage, a new grid can be defined since
all the state variables are carried by the material
points. In practice however, the most efficient way
is to keep the original grid.
4 MPM REPRESENTATION
OF A MEMBRANE
In this section, the representation of the MPM
membrane is improved by introducing new method.
Within this method, an accurate integration of the
membrane force is achieved, in addition to reduc-
ing the oscillations related to particles crossing ele-
ment boundaries. For more clarification about the
approach, a schematic diagram of this procedure is
179
presented for a rectangular element with the rein-
forcement of a bar/membrane element embedded
through the width as shown in Figure 3. In the ini-
tial configuration of this figure, an external force is
applied on top. As a result, the membrane particles
would follow the deformations kinematically pro-
ducing in-line strain εξ which in turns produces the
stress σξ. Accordingly, this stress is integrated along
the length to get the membrane forces acting on
opposite direction to the deformation. Lastly, the
internal membrane force is transferred to the original
mesh by weighting taking into account the location
of the computational nodes relative to the membrane
nodes.
4.1 Membrane formulation with accurate
integration
In the new membrane representation, the surface of
the membrane is discretised by 2D triangular mesh
as shown in Figure 4. The mesh connectivity is main-
tained during the simulation where the membrane
nodes are free to move through the 3D mesh. The
displacement of these nodes is described by the solu-
tion of the momentum equation on the 3D compu-
tational mesh, but the increment in membrane strain
and stress is based on the deformation of the 2D
triangular mesh. The membrane strains and stresses
are calculated at the integration point as shown in
Figure 4. As long as 3-noded triangular element is
adopted in this scheme, the location of the integra-
tion point could be anywhere inside the triangle and
does not need to correspond to the Gaussian point.
In standard MPM integration approach the
internal force vector is calculated using the strain-
displacement matrix of the 3D element and the
integration is performed by summing over material
point volumes (Equation 8). In the new approach,
the strain-displacement matrix is based on the 2D
linear triangular element and integration point is
used to obtain the internal force vector on the 2D
mesh. The internal force is then mapped from the 2D
grid nodes to the 3D grid nodes where the momen-
Figure 3. Schematic diagram of the new approach.
 (Equation 17) and the derivates of the shape func-
tions. At this stage, the matrix β is transformed to
local coordinate βl using the transformation ten-
sor T which is defined by the directional cosines
as given by (Bathe 1996). The tensor T should be
adjusted to satisfy the membrane theory, i.e. the
out-of-plane stresses are zeros, which is done by
eliminating rows three, five and six. The reduced
rotation tensor Tl then takes the form,

T
Figure 4. Membrane discretisation using 2D triangular ⎡ tx2ξ tx2η 2txξ txη ⎤
element. ⎢ ⎥
⎢ ty2ξ ty2η 2tyξ tyη ⎥
⎢ ⎥
l ⎢ tzξ ⎥
2
tum Equation 4 is solved. The mapping is based on tz2η 2tzξ tzη
the shape functions of the 3D mesh and not on the T ⎢ ⎥ (18)
gradient of the 3D shape functions. Hence, there is ⎢txξ tyξ tx tyη t ξ tyη + txηtyξ ⎥
⎢ ⎥
no change in the sign of the material points contri- ⎢tyξ t ξ tyηtzη tyξ t η + tyηt ξ ⎥
bution to the internal force vector when the point ⎢t t tzηt tzξ t η + tzηtxξ ⎥⎦
crosses an element boundary in the 3D mesh. Apart ⎣ zξ ξ η
from the membrane, the other materials are mod-
eled using the standard MPM procedure. where txξ is the directional cosine between x and ξ.
The general formulation for 2D elements in 3D The local strain-displacement matrix βl is given by
space with three translational degrees-of-freedom
per node were used to model the membrane. The l l
local coordinate system of an element is shown in 4. (19)
The displacement field v is approximated in terms
of interpolation functions H and nodal displace- Subsequently, the incremental local strain Δεt+Δt
ments d, v = H d. Hence, according to Bathe (1996) of the membrane element is calculated from
the x-component of the nodal displacements is
Δε l t +ΔΔt t Δt
Δ β l am with ε l [εξξ εηη εξη ]T (20)
nn nn
ζ
x(ξ ,η,ζ ) ∑ Hi xi ∑ hi Hi nxi
2 i =1
(15) t Δt is the nodal velocities vector of the
where a m
i =1
membrane element m. This vector is updated con-
where, Hi is the interpolation functions of node i tinuously by finite element interpolation of the 3D
represented at location (ξ, η, ς), nn is the number nodal velocity a t Δt .
of nodes per element, h is the membrane thick- Afterward, the membrane constitutive model
ness, and nx is the x-component of the unit normal is applied to compute the incremental local stress
n. For specific application of 3-noded elements tensor Δσl at the integration point of the triangular
with constant membrane thickness h, Equation 15 element,
reduces to,
l t +Δ
Δ Δt
Δσ Δε l ,t , (21)
nn
ζ hhn
x(ξ ,η,ζ ) ∑ Hi xi + 2 x (16)
where D is the constitutive tensor given for linear
i =1
elastic material by,
The Jacobian matrix J which involves the deriv-
atives of global coordinates (x, y, z) with respect to ⎡( − v ) v 0 ⎤
the local coordinates (ξ, η, ς) can be constructed D=
E ⎢ v ( − v) 0 ⎥⎥ (22)
as follows ( + v )( − v ) ⎢
⎢⎣ 0 0 ( − v ) ⎥⎦
∂ ∂
= Jiij where iand j = 1, 2, 3 (17) The total stresses tensor at the integration point
∂ξi ∂x j
σl is then updated using the increment given by
Equation 21
The global strain-displacement matrix β of the
membrane element can be formed in the usual
σ l t Δt σ l t σ l ,t Δt
(23)
way of using the inverse of the Jacobian matrix

180
The membrane force F2D is calculated on the
membrane mesh, using standard finite element
integration,

F 2D = ∫ A β l ,T σ l h dA
d (24)

where A is the membrane surface area. This force

can now be transfered to the 3D mesh using the
following mapping procedure,

nm
f i3 D ∑ i
m
f m2 D (25)
m =1

where fi3D is the membrane force vector of node i in

the 3D mesh, nm is the total number of membrane
nodes, Nim is the shape function of node i evaluated
at the location of the membrane node m, and fm2 D Figure 5. Geocontainer initial configuration.
is the force vector of m. After getting this force for
the entire 3D mesh, its contribution is added to the
global internal force Fint in Equation 4.

5 APPLICATION: RELEASING
GEOCONTAINER FROM BARGE

The releasing of a geocontainer from a split barge

is a quite challenging application. Most traditional
numerical methods can not model a large deforma-
tion soil-geotextile interaction problem dynami-
cally. In the present section, the potential of MPM
to handle this problem is demonstrated.

Figure 6. Equilibrium Forces on geocontainer.

5.1 Analyis of releasing geocontainer
The releasing process of a geocontainer from a
barge has been modeled using MPM. The dimen-
sions of the barge, as described by De Groot &
Bezuijen (2000), with the initial MPM mesh dis-
cretisation is illustrated in Figure 5. Owing to sym-
metry, only a half of the problem is modeled. The
plane-strain problem is represented in a 3D code
using 4-noded tetrahedral elements. Sand is mod-
eled by the Mohr-Coulomb model with the prop-
erties: Elastic modulus of 820 kPa, Poisson’s ratio
of 0.33, cohesion of 1 kPa, friction angle of 30°
and unit weight of 18 kNm3. Linear elastic geotex-
tile is considered with (Young’s modulus × thick-
ness) = 400 kN/m and Poisson’s ratio being zero.
Frictional contact is assumed between the barge
and the geotextile with a friction coefficient of
(μ = 0.3). In the present calculations, the effect of
water is not considered and the soil inside the geo-
container is assumed to be dry.
Initially, the gravity solution is carried out by
keeping the barge closed. After solving gravity, the
barge opening is controlled by a prescribed angular
velocity. Trying to get an overlap with De Groot &
Bezuijen (2000) who present the quasi-static solu-
tion of a simplified geocontainer, the problem needs
to be modeled without the dynamical effects. There-
fore, the opening process is modeled in this paper
by small loading steps of an angle of opening θ = 3°
which is applied gradually during the step. In addi-
tion, some artificial damping is added to the system
to get the quasi-static solution faster. At the end of
each loading step equilibrium state should be satis-
fied, otherwise, the calculations proceed with keep-
ing the angle of opening θ constant. Equilibrium is
illustrated in Figure 6. In this figure, the force (Fh)
results from the horizontal stress in soil along the
symmetry plane representing the load that squeeze
the soil. Moreover, the normal reaction force to the
barge (Fn) which produce the frictional force (μ Fn).
All these forces plus gravity and the geotextile tensile
force (TGT) should keep the geocontainer in place as
soon as the opening stopped.
At the beginning of the opening process and
due to the location of the hinge, part of the soil is

181
lifted up by the barge. On the contrary, the center
part of the geocontainer will move downward as it
loses the support underneath. Hence, the deforma-
tion pattern of the soil will take a circular shape,
whilst its center does not have any deformation as
shown in Figure 7 for an angle of opening θ = 14°.
As the barge opening increases, tensile forces in
the geotextile (TGT) increase as the membrane keeps
the soil together. Eventually, these forces pull the
soil away from the barge which in turn decreases
the frictional forces between the geotextile and the
barge. As a result, the geocontainer will lose con-
tact gradually with increasing angle of opening.
The top left part of Figure 9 shows the vertical dis-
placement of the very bottom point as a function
of the opening angle. The last mark in this curve
represents the state where there is not equilibrium Figure 8. Principal stress directions showing the arch-
state possible anymore which is corresponds to ing effect.
θ = 14°. For this state, the total displacement of the
soil is already shown in Figure 7.
The stress field inside the soil is redistributed
during loading; showing the effect of arching as
illustrated in Figure 8. During this process, the
horizontal force (Fh) increases gradually and simul-
taneously with the geotextile tensile forces (FGT) as
shown in Figure 9. Holding quasi-static equilib-
rium, the difference between these curves increases
with loading, however, this is due to the loss con-
tact during the releasing shown in Figure 6. Since
the frictional force (μ Fn) and the normal force to
the barge (Fn) decrease, the term (Fh − TGT ) would
increase to insure equilibrium state.
The last part of this analysis is done to simulate
the geocontainer beyond quasi-static equilibrium
state. Accordingly, the opening angle increase fur-
ther by opening the barge continuously. Similar to
the previous quasi-static solution phase, the open-
ing is done slowly to reduce the dynamic effects. As

Figure 9. Displacement and forces of the geocontainer.

a result, the vertical displacement of the bottom

increases gradually until θ = 22° at which point
the geocontainer accelerates gradually out of the
barge as shown in the right side of Figure 9. In
this figure, the forces (Fh) and (TGT) increase par-
allel to another for the quasi-static side while the
horizontal force Fh decreases suddenly because the
soil becomes more free to dilate outside the barge.
At this point, the calculation is stopped as the solu-
tion is dominated by dynamic effects. It is impor-
tant to notice in this figure that the two forces (Fh)
and (TGT) eventually approach each other, which
indicates that the barge does not apply any signifi-
cant force on the geocontainer. The shape of the
Figure 7. Norm of total displacement |u|. geocontainer at (θ = 22°) is shown in Figure 10.

182

Figure 10. Total displacement |u| at θ = 22°.
For the Mohr-Coulomb soil inside the geocon-
tainer the critical opening angle, to maintain equi-
librium is θ = 14°; whereas the maximum computed
geotextile force is FGT = 157 kN/m at an opening
angle of θ = 22°. These values deviate from those
got by De Groot & Bezuijen (2000). However, the
later used different approach to analyze the releas-
ing process by subdividing the soil body by rigid
blocks that slide on each other with friction.
6 CONCLUSIONS AND OUTLOOK
In the present paper, an important application
in the geotechnical field has been tackled, i.e. the
releasing of a geocontainer from a split barge was
modeled with MPM. In this modeling, a new for-
mulation for the geotextile has been introduced
and successfully validated. The opening process
of the barge is controlled by prescribing rotational
velocity of the barge in the presence of frictional
contact with the geocontainer.
This paper shows the potential of MPM to
model such an application of combing geotextile
with soil under large deformations. However, it
is important to mention that the modeling of the
entire process of releasing a geocontainer from a
barge into water requires more elaborate work. For
example, the soil should be modeled using a more
advanced constitutive description which considers
the effect of density change.
Furthermore, the realistic modeling of geo-
textile should take into account the permeability.
183
Allowing water to penetrate into the geocontainer
and including the effect of water drag would be
more attractive for real geocontainer applications.
Currently, modeling free water with MPM is under
development.
ACKNOWLEDGMENT
This research was carried out as a part of the
“GEO-INSTALL” project (Modeling Installa-
tion Effects in Geotechnical Engineering). It has
received funding from the European Community
through the program (Marie Curie Industry-
Academia Partnerships and Pathways) under grant
agreement no PIAP-GA-2009-230638.
REFERENCES
Bathe, K. 1996. Finite element procedures, Volume 2.
Prentice hall Englewood Cliffs, NJ.
Bezuijen, A., R.R., S., & Klein Breteler M., Berendesen
E., P.K. 2002. Field tests on geocontainers. In Pro-
ceedings of 7th International conference on Geosythet-
ics, Nice.
Burgess, D., Sulsky, D., & Brackbill, J. 1992. Mass matrix
formulation of the flip particle-in-cell method. Jour-
nal of Computational Physics 103(1), 1–15.
De Groot, M. & Bezuijen, A. 2000. Forces in geocon-
tainer geotextile during dumping from barge. In Pro-
ceedings of 2nd EuroGeo2, pp. 623–627.
Fernandes, I. 2004. Exploring quadratic shape functions
in material point method. Masters thesis, Department
of Mechanical Engineering, The University of Utah.
Jassim, I., Hamad, F., & Vermeer, P. 2011, 27–29 April.
Dynamic material point method with applications
in geomechanics. In 2nd International Symposium on
Computational Geomechanics (COMGEO II), Cavtat-
Dubrovnik, Croatia.
Pilarczyk, K. 2000. Geosynthetics and geosystems in
hydraulic and coastal engineering. Taylor & Francis.
Sulsky, D. & Schreyer, H. 1993. A particle method with
large rotations applied to the penetration of history-
dependentmaterials. Advances in Numerical Simu-
lation Techniques for Penetration and Perforationof
Solids, ASME, AMD AMD-Vol 171, 95–102.
Sulsky, D., Zhou, S., & Schreyer, H. 1995. Application of
a particle-in-cell method to solid mechanics. Compu-
ter Physics Communications 87(1), 236–252.
Wieckowski, Z., Youn, S., & Yeon, J. 1999. A particle-
in-cell solution to the silo discharging problem. Inter-
national journal for numerical methods in engineering
45(9), 1203–1226.
York, A.R. 1997. The development of modifications to
the material point method for the simulation of thin
membranes, compressible fluids, and their interactions.
Ph.D. thesis, The University of New Mexico.
Zhang, D., Ma, X., & Giguere, P. 2011. Material point
method enhanced by modified gradient of shape
function. Journal of Computational Physics 230,
63796398.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Leg penetration assessments for self-elevating tubular leg units in sand

over clay conditions

D.A. Kort
Norwegian Geotechnical Institute, Oslo, Norway

S. Raymackers
GeoSea NV, Zwijndrecht, Belgium

H. Hofstede
GustoMSC B.V., Schiedam, The Netherlands

V. Meyer
Norwegian Geotechnical Institute, Oslo, Norway

ABSTRACT: Leg penetration assessments were performed for two self-elevating units with tubular legs
operating at 21 locations within an offshore wind farm. The assessments were performed on the basis
of available CPT data, experience-based soil parameters and the calculation methods recommended by
SNAME T&RB 5-5A. In general the soil conditions are dense sand over stiff clay with strong variations
of the sand layer thickness across the wind farm. Initially at some locations the assessment indicated a
significant risk of punch-through, which could potentially lead to problems with insufficient leg length
and create difficulties for leg extraction after installation of the wind turbines. To reduce the uncertain-
ties of the leg penetration assessments, preload trials were performed at three critical locations. The three
preload trials provided a useful basis to revise the input soil parameters and thereby enhance the leg pen-
etration predictions. No indications of large penetrations or rapid penetrations were encountered during
the offshore field campaign.

1 INTRODUCTION

For the Thornton Bank offshore wind farm located

in the North Sea leg penetration assessments were
performed for the jack-up vessels Neptune and
Vagant, both owned by GeoSea (DEME group)
and designed by GustoMSC (see Fig. 1). Both ves-
sels are four-legged self-elevating units with tubu-
lar legs without spudcans with diameters of 3.5 m
(Neptune) and 2.0 m (Vagant). The soil conditions
at 21 selected locations at the wind farm consisted of
dense sand over stiff clay with strong variations of
the sand layer thickness across the wind farm. Initial
leg penetration assessments indicated a significant
risk of punch-through. As a result, possible large leg Figure 1. Self-elevating units Neptune and Vagant side-
penetrations in the deep clay were predicted, which by-side installing an offshore wind turbine. © Photo:
could potentially lead to problems with insufficient GEOSEA/DEME.
leg length and create difficulties for leg extraction
after installation of the wind turbines. To reduce
the uncertainties of the leg penetration assessments, The first part of this paper presents the basis
preload trials were performed with the Neptune at and the methodology for the leg penetration
three critical locations. The preload trials provided a assessments. In the second part of this paper soil
useful basis to revise the input soil parameters and data, field measurements and back-analyses from
thereby enhance the leg penetration predictions. three jacking trials at selected windmill locations

184
are presented. The methodology of the leg penetra- 2.2 Available geotechnical data
tion assessments was then used to predict leg pen-
The available geotechnical data consisted of one
etrations at other locations at the Thornton Bank
CPTU at each windturbine location. Sampling
wind farm. The third part of this paper compares
was performed at selected locations and index and
predicted leg penetrations to field measurements
advanced laboratory testing (triaxial testing and
from 21 different locations.
DSS testing) were performed on samples from the
underlying clay layer.
2 BASIS FOR THE LEG PENETRATION
ASSESSMENTS 2.3 Dredging activities
Dredging activities had taken place at the windtur-
Basis for the leg penetration assessments include
bine locations after the site investigations, resulting
the leg specifications and the available geotechnical
in updated reference seabed levels.
data. Relevant data are briefly summarised below.

2.1 Leg specifications 3 JACKING TRIALS

The legs of both the vessels Neptune and Vagant are
Initial leg penetration assessments carried out before
close-ended piles with constant diameter. The legs
the jacking trials suggested a possibility of rapid leg
have a number of stiffners fixed on a closed-ended
penetrations for a number of locations due to punch-
bottom plate, creating a slightly conical shape at
through—dense sand over soft clay. Those assess-
the leg tip. A similar leg is shown in Figure 2 and
ments were based on the methodology presented in
Table 1 summarises the main leg specifications of
Section 4, although lower friction angles were used
both vessels.
than those presented in Figure 12, combined with a
The legs of four-legged self-elevating vessels
smaller load spread angle (larger n; Fig. 13).
are typically penetrated stepwise: Two diagonal
As the initial leg penetration predictions were
legs are slightly pushed into the subsoil and subse-
somewhat careful because they were based on
quently unloaded to less than 50% of the platform
relatively few site-specific soil data and to further
weight to provide the reaction forces needed for the
evaluate the risk of rapid leg penetrations, it was
opposite penetrating diagonal legs. This process is
decided to perform preload tests at three critical
repeated until (1) the maximum preload is reached
locations with regard to rapid penetrations.
and (2) when the additional penetrations in a load
Figures 3 to 5 present profiles of corrected cone
step are negligibly small.
resistance (qt), undrained shear strength (su) and

Figure 2. Example of a leg.

Table 1. Summary of main leg specifications.

Neptune Vagant

Leg diameter (m) 3.5 2.0

Max. preload at leg tip (MN) 30.1 8.4
Distance leg tip to end plate (m) 0.75 0.70
Figure 3. Interpreted soil profile at Trial 1 location.

185
Figure 4. Interpreted soil profile at Trial 2 location. Figure 6. Predicted and measured penetrations at Trial 1.

Figure 5. Interpreted soil profile at Trial 3 location. Figure 7. Predicted and measured penetrations at Trial 2.

relative density (Dr) interpreted from the CPT
data, and simplified profiles used in the penetra-
tion analyses. The profiles are related to the water
depth in TAW (Tweede Algemene Waterpassing).
The leg penetration depths presented are
related to the leg tip. However, due to the open
conical shape of the leg tip, penetration resistance
is only built-up when the soil becomes in contact
with the end bearing plate of the pile (Figure 2).
The applied foundation load is composed of the
buoyant weight of the leg and the load applied to
the leg.

186

Figure 8. Predicted and measured penetrations at Trial 3.
The results of the jacking trials could be con-
sidered as an additional soil investigation on the
basis of which the soil input parameters could be
revised. Because none of the jacking trials resulted
in rapid penetrations and the measured leg penetra-
tions appeared to be relatively low compared to the
predictions, it was obvious that the assumed fric-
tion angles of the upper sand could be increased,
as presented in Section 4.
4 METHODOLOGY OF THE LEG
PENETRATION ANALYSES
The leg penetration analyses have been performed
in accordance with the industry standard guidelines
published by SNAME (2008) for spudcan penetra-
tion analyses. The method is based on conventional
bearing capacity analyses. Punch-through—dense
sand over soft clay—has been analysed using the
method described in the commentary of SNAME.
The leg penetration analyses consisted of two
main activities:
1. Assessment of soil parameters.
2. Leg penetration calculations.
4.1 Assessment of soil parameters
Soil parameters needed for the leg penetration
analyses include unit weight, equivalent friction
angle for the sand layers and undrained shear
187
strength profile for the clay layers. As the friction
angles are assessed from correlations with relative
density, assessment of the relative density is also
required. The significant soil parameters for the
bearing capacity analyses include relative density,
friction angles and undrained shear strength pro-
files; soil parameter assessment is discussed below.
4.1.1 Assessment of relative density
A best estimate of the relative density was assessed
from the corrected CPT data using the method of
Baldi et al. (1986). The CPT data was corrected
for in-situ stress relief due to dredging following
the procedure for normalising the cone resistance
given by Gregg (2007), which includes some minor
updates for interpretation of silt-sand mixtures
and clay-silt mixtures compared to the procedure
developed by Robertson & Wride (1998).
Although the Robertson & Wride (1998) proce-
dure for normalising cone resistance was originally
developed for liquefaction analyses, the method is
widely supported and frequently used world-wide.
The stress-corrected cone resistance due to dredging
was significantly lower than the original cone resist-
ance in the upper metre but the effect on the calcu-
lated relative density appeared to be insignificant.
To account for a possible loosening of the sea-
bed due to dredging and other effects (such as tidal
current), a maximum relative density of 80% was
assumed in the upper 0.5 metre.
4.1.2 Assessment of friction angles
In the present method the relevant friction angles to
be considered for very dense sand are peak friction
angles, φ′p, and constant volume friction angles,
φ′cv, from drained triaxial testing in sand. Drained
peak and constant volume friction angles are con-
sidered applicable for the following reasons:
– The process of leg penetration may involve very
large shear strains (say 100–1000%) and shear
strain rates in the soil underneath the leg tip.
Experimental data show that the density of very
dense sand around a penetrating pile tip will
decrease during penetration to a certain criti-
cal density state (Dijkstra, 2009). In that case
the sand underneath the pile tip will behave in a
drained manner.
– Peak values of the drained friction angles are
associated with shear strains of order of mag-
nitude 1% and constant volume drained fric-
tion angles are applicable from say 10% shear
strain and more. Hence, peak values of the
drained friction angles may be applicable in
the outcropping zone of the failure mechanism
whereas constant volume friction angles may
be applicable underneath the leg tip. This state-
ment is supported by the simplified FE analysis
in Figures 9–11.

Figure 9. Model detail of a simplified bearing capacity
FEM analysis of a tubular leg.
Figure 10. Typical shear strain contour diagram at
failure.
– Appendix E of ISO 19905 (2012) proposes use
of critical state friction angles (constant volume
friction angles) increasing by up to 5% with
increasing relative density on the basis of back-
analyses of field penetration records and cen-
trifuge tests.
The weighting between peak and constant vol-
ume friction angles is based on a simplified back
analysis of the field measurement from Trial 2
(Section 4) using Plaxis (2008).
Figure 9 presents a detail of an FE mesh show-
ing the most relevant geometry and soil parameter
input and the calculated principal stresses. The
model is based on a 3.5 m diameter pile with a rigid
base at 2.1 m depth, which is a simplified represen-
tation of Trial 2 at 2.85 m.
The FE result in Figure 10 confirms that rela-
tively large shear strains below the leg tip may be
expected, and thus applicability of constant vol-
ume friction angles, and that relatively small shear
strains outside the leg tip may be expected, and
thus applicability of peak volume friction angles.
Figure 11 shows three load-displacement
curves:
– Full end bearing: φ′ = 40.4º is taken underneath
and outside the leg tip
– Reduced end bearing: φ′ = 34.0º is taken under-
neath the leg tip and φ′ = 40.4º outside the leg
tip
– Equivalent friction angle: φ′ = 37.5º is taken
underneath and outside the leg tip.
Figure 11 shows that the reduced end bearing
capacity can be assessed with equivalent friction
angles as defined in Figure 9. Based on the simpli-
fied model it is suggested to assess the equivalent
friction angle by taking 45% of the friction angle
underneath the leg tip and 55% of the friction
angle away from the leg tip.
Low estimate and high estimate friction angles
used in the calculation are therefore based on a
weighted combination of peak friction angles and
constant volume friction angles:

φ ′ LE φ ′ cv , LE + 0.55 ⋅ φ ′ cv , HE (1)

φ ′ HE φ ′ cv , HE + 0.55 ⋅ φ ′ p BE (2)

The low estimate parameters are based on con-

stant volume friction angles for high consolidation
stresses underneath the leg tip and on constant vol-
ume friction angles for in situ consolidation stress
conditions outside the leg. This refers to the case
where the soil underneath the leg tip is experienc-
ing a significant consolidation pressure prior to
loading to failure and the soil outside the leg tip
Figure 11. Load displacements curves. has experienced large shear strains (disturbance

188
from shallower penetrations). On the other hand, 4.2 Leg penetration calculations
the high estimate parameters are based on con-
The leg penetration calculations were performed
stant volume friction angles for in situ consolida-
for the characteristic foundation loads given in
tion stress conditions underneath the leg tip and
Table 1. Low and high estimated characteristic soil
on peak friction angles for in situ stress conditions
strength parameters have been selected to evaluate
outside the leg. This refers to the case where the
the range in leg penetration for a given location.
soil underneath the leg tip is experiencing low
The punch-through—dense sand over soft
consolidation pressure prior to loading to fail-
clay—analyses require input of a load spread fac-
ure and the soil outside the leg tip is experiencing
tor (Fig. 13).
small shear strains (no disturbance from shallower
A load spread factor of n = 2.5 is used, which
penetrations).
seems to be justified by the FEM analyses pre-
Figure 12 provides guidance on the assessment
sented in Figure 10. A load spread of 1:2.5 is out-
of the various friction angles from the relative
side the 1:3 to 1:5 range recommended in SNAME
density, which are based on the empirical database
T&RB 5-5A and ISO 19905-1, but can be justified
from Andersen & Schjetne (2012).
for these locations on the basis that:
4.1.3 Assessment of undrained shear strength 1. Trial installations have been undertaken at these
in clay locations on which the analyses have been sub-
The undrained shear strength for triaxial com- sequently calibrated.
pression in clay was assessed from the CPT data 2. A load spread of 1:3 to 1:5 has been reported,
using suC = (qt − σv0)/Nkt as presented by e.g., under certain conditions, to under-predict peak
Lunne et al. (1997). Nkt ranging from 15 to 20 punch-through loads for both centrifuge tests
was concluded for the purpose of the leg penetra- (Teh et al., 2010; Lee, 2009) and from back-
tion assessments at the Thornton Bank site on analysis of field data (Baglioni et al. 1982).
the basis of a study of the available laboratory Note, however, that this is not always the case.
data and literature. 3. A load-spread of 1:2.5 is within the range sug-
The deep clay layer was described as highly plas- gested in DNV Classification Note 30.4 (DNV,
tic with plasticity index Ip > 100% and on the basis 1992) for spudcan penetration analysis (1:2 to
of Bjerrum (1973) the undrained shear strength of 1:3).
the clay is assumed isotropic. Hence, the average
As the bearing capacity formulae and empirical
undrained shear strength accounting for strength
relationships given in by SNAME (2008) have been
anisotropy is suavg/suC = (suC+suDSS+suE)/(3 suC) = 1.0.
derived specifically for spudcans, the contribution
Here is suC = undrained shear strength measured
in triaxial compression; suDSS = undrained shear
strength measured in direct simple shear; suE = und-
rained shear strength measured in triaxial exten-
sion (Bjerrum 1973).

Figure 12. Assessment of friction angles from relative Figure 13. Load spread method according to SNAME
density. (2008).

189
of the leg shaft friction capacity has been added to field campaign, predictions of leg extractions have
the calculated penetration resistance. The leg shaft not been included in this paper.
friction capacity has been determined using
5 PENETRATIONS DURING WINDFARM
Fs ∫ s Odz (3) INSTALLATION

The methodology of the leg penetration assessments

where Fs = leg friction capacity; τs = leg shaft
explained in Section 4 was used to improve the pre-
friction capacity at depth z, in clay: α ⋅ su; in sand
diction of the leg penetrations at other locations at
K ⋅ σ′v ⋅ tan δ; α = remoulding factor; su = average
the Thornton Bank windfarm. Because of the het-
undrained shear strength (suavg); K = earth pressure
erogeneity of the soil layering and soil conditions
coefficient; σ′v = effective vertical stress; δ = leg-soil
across the wind farm site, use was made of the site
interface friction angle; O = leg circumference at
specific CPT data for the individual locations.
depth z; z = depth below seafloor.
The predictions were made those legs that were
Typical values of K = 1 and δ = 0.67 φ′ have been
located in the dredged zone and they were consid-
used for the leg shaft friction capacity assessments
ered most susceptible for punch-through.
in sand (e.g., Lambe & Whitman, 1979). For the
For a large number of locations the revised pre-
clay, α = 0.5 is estimated which is based on the fol-
dictions resulted in the conclusion that the risk of
lowing simplified assumptions:
rapid penetration during installation was signifi-
– During penetration, a zone of remoulded clay cantly reduced. Only for two locations (both vis-
is assumed to be developed along the shaft. For ited during the jacking trials) the predictions still
installation of suction anchors α-factors have indicated a risk for rapid penetrations.
been back-calculated of α = 1/St, where St is Figures 14 and 15 show the predicted and meas-
the sensitivity of the clay (e.g., Schroeder et al. ured leg penetrations at 21 selected critical windfarm
(2006). Using a typical sensitivity of St = 3 gives locations for the Neptune and the Vagant vessels,
α = 0.33. respectively. At all locations the measured leg pene-
– After stopping penetration the shear strength trations fit very well within the predicted range. This
in the remoulded zone may quickly increase means that the methodology proposed in Section 4
due to both thixotropy (thixotropy is the worked well for the Thornton Bank wind farm.
regain of remoulded undrained shear strength The leg installation at all locations and in par-
with time when the soil is allowed to rest under ticular at the critical locations went according to
constant water content) and reconsolida- plan. The penetrations measured during preload at
tion of the remoulded clay. An estimated the “critical” locations were comparable to those
factor 1.5 accounting both for thixotropy and measured during trials. No deep penetrations were
reconsolidation during leg penetration yields encountered.
α = 1.5 × 0.33 = 0.5.
Additional set-up and thixotropy effects may
lead to a significant increase in leg friction capacity,
even after a week, which is especially relevant for
leg extraction. Purwana et al. 2009 report that dif-
ficulty with leg extraction is often experienced dur-
ing departure from location, especially in the case
of deep leg penetrations or extended operations,
which confirms the relevance of the additional
set-up and thixotropy effects to leg extraction. Spe-
cial purpose laboratory testing is recommended for
projects where extraction forces in the clay need to
be predicted. Such laboratory testing may include:
– Standard DSS testing of intact and remoulded
samples
– Special-purpose DSS testing on remoulded,
subsequently reconsolidated samples
– Thixotropy testing
– Standard classification tests.
As no indications of large penetrations or rapid Figure 14. Predicted and measured penetrations with
penetrations were encountered during the offshore Neptune during windfarm installation for 20 locations.

190

Figure 15. Predicted and measured penetrations with
Vagant during windfarm installation for 21 locations.
6 CONCLUSIONS AND
RECOMMENDATIONS
For the installation of an offshore wind farm three
field trials of leg penetrations for a self-elevating
unit with tubular legs have been performed. The
soil conditions at the trial locations comprise dense
sand over stiff clay. The three preload trials pro-
vided a useful basis to revise the input soil param-
eters and thereby enhance the leg penetration
predictions. No indications of large penetrations
or rapid penetrations were encountered during the
offshore field campaign.
The measured leg penetrations at 21 selected
locations for two vessels with different leg
geometries fit very well to the revised prediction
of the leg penetrations based on the methodology
presented in this paper. The number of locations
for which a significant risk of punch through and
deep penetrations was predicted reduced to only
two (both also visited during the trials), whereas
the initial leg penetration assessments indicated
this risk for a larger number of penetrations.
The leg penetration predictions such as carried
out in this paper are strongly dependent on the
input soil parameters. It is therefore recommended
to monitor leg penetrations on the job site and,
where necessary, adjust predictions on the basis of
measured penetrations.
Monitoring leg penetrations and extractions can
be considered as a form of pile load testing. From
this point of view it would be beneficial for future
wind-farm projects to instrument the legs with
strain gauges to measure the shaft friction along the
depth. The costs for such testing are believed much
lower than special purpose pile load tests on which
input parameters are based now) and the results are
believed to be highly relevant for the industry.
191
ACKNOWLEDGEMENT
GeoSea and GustoMSC are greatly acknowledged
for giving permission to publish this material.
REFERENCES
Andersen, K.H. & Schjetne, K. 2012. Data base of fric-
tion angles and consolidation characteristics. Accepted
for publication in ASCE, Journal of Geotechnical and
Geoenvironmental Engineering.
Baglioni, V.P., Chow, G.S. & Endley, S.N. 1982. Jack-up
rig foundation stability in stratified soil profiles. Jour-
nal of Offshore Technology, OTC 4409, pp. 363–383.
Baldi, G., Bellotti, R., Ghionna, V. Jamiolkowski, M. &
Pasqualini, E. 1986. Interpretation of CPTs and CPTUs;
2nd part: drained penetration of sands. Proceedings of
the 4th International Geotechnical Seminar Singapore:
143–56.
Bjerrum, L. 1973. Problems of Soil Mechanics and Con-
struction on Soft Clays. State-of-the-Art report to Ses-
sion IV, Proceedings 8th ICSMFE, Moscow. Vol. 3,
pp. 111–159.
Dijkstra, J. 2009. On the modelling of pile installation.
Dr. Thesis Delft University of Technology. Zutphen:
Wöhrmann Print Service.
DNV 1992. DNV Classification Note 30.4. Det Norske
Veritas.
Gregg 2007. Cone Penetration Test (CPT) Interpretation.
http://www.greggdrilling.com/PDF_files/Technical-
MethodologyPDF/cptinterpretationsummary.pdf.
Downloaded 15 December 2011.
ISO 19905-1 2012. Petroleum and natural gas industries.
Site-specific assessment of mobile offshore units—
Part 1: Jack-ups.
Lambe, T.W. & Whitman, R.V. 1979. Soil Mechanics.
New York: John Wiley & Sons.
Lee, F.H. 2009. Investigation of potential punch-through
failure on sands overlying clay soils. Ph.D. Thesis, The
University of Western Australia.
Lunne, T., Robertson, P.K. & Powell, J.J.M. 1997. Cone
Penetration Testing in Geotechnical Practice. London:
Blackie Academic & Professional.
Plaxis 2008. Plaxis Version 9.02. www.plaxis.nl.
Purwana, O.A., Quah, M. Foo, K.S. Nowak S. &
Handidjaja, P. 2009. Leg Extraction/Pullout
Resistance—Theoretical and Practical Perspectives.
Proceedings of the 12th International Conference The
Jack-Up Platform Design, Construction & Operation,
London.
SNAME 2008. Technical and Research Bulletin 5-5A.
Guidelines for Site Specific Assessment of Mobile
Jack-Up Units. Society of Naval Architects and Marine
Engineers, Jersey City, New Jersey. August 2008.
Schroeder, K., Andersen, K.H. & Jeanjean, P. 2006. Pre-
dicted and observed installation behavior of the Mad
Dog anchors. Proceedings of the Offshore Technology
Conference, Houston, OTC 17950.
Teh, K.L., Leung, C.F. Chow, Y.K. and Cassidy, M.J.
2010. Centrifuge model study of spudcan penetration
in sand overlying clay. Proceedings of the Offshore
Technology Conference, Houston, OTC 20060.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Investigating the scales of fluctuation of an artificial sand island

M. Lloret-Cabot
Centre for Geotechnical and Materials Modelling, The University of Newcastle, Newcastle, Australia
Department of Geoscience and Engineering, Delft University of Technology, Delft, The Netherlands

M.A. Hicks & J.D. Nuttall

Department of Geoscience and Engineering, Delft University of Technology, Delft, The Netherlands

ABSTRACT: Artificial sand islands were constructed in the Canadian Beaufort Sea for use as
hydrocarbon exploration platforms in the 1970s and 1980s. For some of these islands, extensive Cone
Penetration Test (CPT) data are available for characterising the hydraulically placed sand during and after
the construction process. Tarsiut P-45 was the first island using the ‘Molikpaq’ concept, which consisted
of a mobile arctic caisson system to provide the temporary structure for the exploitation. Two main sand
fills were constructed: (a) a sandfill berm on which the caisson system was founded; and (b) the body of
the island structure (island core). This paper presents an investigation of the variability of the sand in the
berm in terms of the vertical and horizontal scales of fluctuation. This geo-statistical investigation is car-
ried out using CPT data from the berm before and after the founding of the caisson system, and sets the
basis for a preliminary discussion on the potential soil variability changes caused by the installation and
infilling of the caisson structure when placed on the berm.

1 INTRODUCTION study. This paper uses in situ Cone Penetration

The spatial nature of soils is of particular
importance in geotechnical engineering because
the mechanical response of a geotechnical structure
is typically governed by how ‘strong’ and ‘weak’
zones vary throughout the soil domain (Griffiths &
Fenton 2001, Spencer 2007, Hicks & Spencer 2010).
In the light of this, a fundamental statistical char-
acteristic is the scale of fluctuation θ, which is a
measure of the distance between zones of similar
‘strength’ (Hicks & Onisiphorou 2005). In other
words, θ describes the distance over which proper-
ties are significantly correlated (Vanmarcke 1984)
and it may be used to describe the spatial vari-
ability. Indeed, the scale of fluctuation is the key
parameter in the correlation model used to gener-
ate numerical predictions of spatial variability. It is
therefore crucial, in order to accurately represent
the soil variability of a particular site, to obtain
representative estimates of θ for the site.
Although a number of analyses are available in
the literature for the vertical scale of fluctuation,
very little information is generally found for the hor-
izontal scale of fluctuation. The horizontal scale of
fluctuation, however, is an important consideration
in geotechnical computations, as demonstrated in
a number of studies (Hicks & Samy 2002, Hicks &
Onisiphorou 2005, Hicks & Spencer 2010, Nuttall
2011) and has therefore been incorporated in this
Test (CPT) data from an artificial sand island to
estimate the scales of fluctuation in the vertical θv
and horizontal θh directions during two different
stages of the construction process. In particular,
the available CPTs are used to describe the vari-
ability of the sand in the berm before (Stage 1) and
after the founding of the caisson and subsequent
core infilling (Stage 2). A comparison of the results
obtained is used to provide a preliminary assess-
ment of the potential effects that the caisson instal-
lation and core infilling may have on the vertical
and horizontal scales of fluctuation.
2 ISLAND CONSTRUCTION
Numerous artificial islands were constructed
during the 1970s and 1980s in the Canadian
Beaufort Sea, to provide temporary structures
for hydrocarbon exploration. Data from one of
these islands, Tarsiut P-45, is investigated in this
paper. Tarsiut P-45 was the first island to use
the ‘Molikpaq’ concept, a mobile arctic caisson
system designed to reduce the required fill vol-
umes (Hicks & Smith 1988). Two main sand fills
were constructed: (a) a sand fill berm on which
the caisson system was founded; and (b) the body
of the island structure (referred to as the core).
The fill material used in most of the constructed

192
 3.1 Scale of fluctuation
Based on the concept of the variance function
discussed in Vanmarcke (1977), Wickremesinghe &
Campanella (1993) proposed an alternative strategy
to estimate θv which has since been used in a number
of studies (Wong 2004, Hicks & Onisiphorou
2005, Lloret-Cabot et al. 2012). Their strategy is
very convenient for analysing the vertical soil vari-
ability, not only because the method can be eas-
ily automated (Wickremesinghe 1989) but also
Figure 1. Side view of Tarsiut P-45 (Wong 2004). because many data points are typically available
in the vertical direction of a CPT. In addition, in
the current study, it has been considered useful to
also include results using an alternative method to
islands was sand, which was typically dredged
allow for comparison. Both approaches are out-
from local borrow pits.
lined in the following paragraphs.
The island of Tarsiut P-45 was constructed
during 1984 in a water depth of 25.5 m. The
3.1.1 Approach A
berm was constructed from the seabed up to
Approach A assumes that the available data are
19.5 m below Mean Sea Level (MSL) using a
statistically homogeneous (or stationary). The nec-
bottom dumped placement method. The caisson
essary requirements for a function to be statistically
was then positioned on the constructed berm
homogeneous are: (a) constant mean and constant
and the core was filled-in with sand using a pipe-
standard deviation; (b) an autocorrelation func-
line placement technique. The sand specifica-
tion independent of the location and dependant
tion for the berm and core was for a D50 greater
only on the separation distance (or lag distance,
than 300 μm and a silt content less than 4%
τ). For geotechnical engineering purposes, this is
(Wong 2004). Figure 1 illustrates a cross-section
equivalent to saying that the mean and standard
through Tarsiut P-45 in which the two sand fills
deviation do not spatially vary and that the cor-
are shown.
relation between property values at two different
locations is only a function of their separation dis-
tance. A constant mean can be obtained after de-
3 STATISTICAL EVALUATION
trending the data as illustrated in Figure 2b, where
the identified linear trend of qc with depth (Fig. 2a)
Two analyses are included in this study. In the
has been removed from the data. The approach of
first analysis, the cone tip resistances qc from
removing a linear trend of qc with depth from the
four CPTs are used to estimate the vertical and
horizontal scales of fluctuation of the berm prior
to set-down of the caisson. The second analysis
uses six CPTs for the calculation of the scales of
fluctuation of the berm after core infilling. All
the CPT data comprise tip resistance readings
at 2 cm intervals. In each analysis two comple-
mentary methods are used to determine the verti-
cal scale of fluctuation θv. The first of these two
methods will be referred to as Approach A, and
uses the strategy proposed by Wickremesinghe &
Campanella (1993). The second method will be
referred to as Approach B and estimates the ver-
tical scale of fluctuation by best fitting the the-
oretical correlation model to the experimental
correlation model. Due to the limited data avail-
able, Approach B is the only technique used for
the estimation of the horizontal scale of fluctua-
tion θh. A brief description of each approach is
given below, further details can be found else-
where (Wickremesinghe & Campanella 1993,
Baecher & Christian 2003, Wackernagel 2003, Figure 2. Example of a CPT profile: (a) original data,
Fenton & Griffiths 2008). (b) detrended data, (c) normalised data.

193
data, provides a useful approximation that has
been applied in a number of prior studies (Uzielli
et al. 2005, Hicks and Onisiphorou 2005, Lloret
et al. 2012). A constant standard deviation and the
condition (b) stated above, are likely to be reached
if the data are extracted from the same soil layer, as
approximate uniform fluctuations are likely to be
observed if the analysed data are contained within
the same soil type layer (Phoon & Kulhawy 1999).
Note that this is a reasonable assumption for the
case studied here, as all the berm sand was dredged
from the same location and deposited in the same
manner (Wong 2004).
Once the linear depth trend of the soil layer
investigated has been removed, the next stage in
Approach A is to normalise the detrended data as
shown in Figure 2c. Then, the normalised data are
considered in pairs (n = 2) and the moving average
of the data is calculated with an averaging length
equal to the spacing between data points (τn = 2). The
variance of this series is then calculated (σ2n = 2).
This procedure is repeated for the case of triplets
n = 3, with the corresponding variance (σ2n = 3) and a
spacing average (τn = 3) being equal to twice the spac-
ing between the original data points. The process
can be further extended until the total number of
data used in the averaging are n = N. A consequence
of spatial averaging is that the variance tends to
decrease with increasing n, due to the cancelling
out of fluctuation as a result of spatial averaging.
For each n, the variance function can be estimated
from (Wickremesinghe & Campanella 1993):
σ 2n
Γ ( τ) ≈ (1)
σ2
where σ2 is the variance for the original data. On the
other hand, for large τ the variance function Γ(τ)
can be estimated as (Vanmarcke 1977, Vanmarcke
1984):
Note that various alternative expressions for
the correlation model are also possible (Fenton &
Griffiths 2008) but, for simplicity, only the expo-
nential correlation model is used in this paper, i.e.
⎧ −2 τ ⎫
ρ ( τ ) = exp ⎨ ⎬ (3)
⎩ θ ⎭
where θ is the scale of fluctuation in the vertical or
horizontal direction. The experimental correlation
function is given by
n − j +1
1
ρˆ (jΔτ ) = ∑ (X i − μˆ )(X i + j − μˆ ) (4)
σˆ (n − j) i =1
where μ̂ and σ̂ are the estimated mean and stand-
ard deviation from the in situ CPT data. Note
that for this estimator it is desirable that the data
be equispaced (Fenton & Griffiths 2008) at a
spacing Δτ.
4 RESULTS
This section presents the results obtained for the
estimation of θv and θh, before and after core infill-
ing. The results are presented in two parts. The
first part describes the soil variability of the sand
in the berm before caisson installation. It includes
the estimation of the vertical scale of fluctua-
tion (using Approaches A and B) followed by the
results obtained for the horizontal scale of fluctua-
tion (using Approach B). The second part of the
analysis includes the results for the soil variability
in the berm after core infilling. All results are sum-
marised in Tables 1 and 2.
4.1 Variability of the sand in the berm
before caisson installation

θ ≈ Γ ( τ) τ (2) Four aligned CPTs along the same straight hori-

zontal line are used in this section to assess the soil
Wickremesinghe & Campanella (1993) found
that Equation 2 reaches a peak at a large value of Table 1. Estimation of the scales of fluctuation in
τ and that this maximum gives a good approxima- Tarsiut P-45 berm before caisson set down.
tion of θ.
θv (m)* θv (m)** θh (m)**
3.1.2 Approach B
A useful alternative approach (Approach B) is to CPT01 0.37 0.44 –
simply estimate θ by best fitting the theoretical cor- CPT02 0.21 0.24 –
relation function ρ(τ) (see Equation 3) to the exper- CPT03 0.47 0.46 –
imental correlation function ρ̂ (τ ) (see Equation 4) CPT04 0.35 0.37 –
using statistically homogeneous data (section 3.1.1). Mean, m 0.35 0.38 12.7
In the analyses presented here, the experimental Variance, m2 0.11 0.10 –
correlation model has been estimated using the
same normalised data as used in Approach A. *
Values using Approach A. **Values using Approach B.

194
Table 2. Estimation of the scales of fluctuation in
Tarsiut P-45 berm after caisson set down and subsequent
core infilling.

θv (m)* θv (m)** θh (m)**

CPT05 0.89 0.88 – Figure 3. Plan view of CPT locations for berm before
CPT06 0.62 0.61 – caisson installation.
CPT07 0.29 0.26 –
CPT08 0.42 0.44 –
CPT09 0.32 0.30 –
CPT10 0.66 0.63 –
Mean, m 0.53 0.52 18.1
Variance, m2 0.05 0.05 –
*
Values using Approach A. **Values using Approach B.

variability in the berm prior to set-down of the

caisson. Each CPT has an approximate depth of
8 m. A plan view of the CPTs (indicated as AA’)
is plotted in Figure 3, where the distance between
CPTs is also indicated.
Figure 4 shows the consistency of the results
obtained for the vertical scale of fluctuation when
using Approach A. The maximum fluctuation
is generally reached when the lag distance τ is
between 0.8 and 1.2 m (out of a total vertical CPT
length of approximately 8 m), giving values of θv
ranging from 0.21 m and 0.47 m (see Table 1). The
average vertical scale of fluctuation for this case is
0.35 m, with a variance of 0.11 m2.
Figure 5 shows a typical result for the estima-
tion of the vertical scale of fluctuation when using
Approach B. The solid line indicates the theoretical
correlation model that uses the optimised value of
θv and the dashed line indicates the experimental
correlation function computed from CPT02. Simi-
lar results are obtained for the other CPT profiles
(i.e. CPT01, CPT03 and CPT04). Note that in order
to better illustrate the important parts of the curves
plotted (i.e. values of τ that give a correlation higher
than 0) the horizontal axis has been intentionally
limited to 3.5 m (although the total vertical length
of the CPTs used is about 8 m in all cases).
Comparing Figures 4 and 5, it is interesting to
see the agreement between the values of θv obtained
for the profile CPT02 when using the two methods.
This consistency between the estimated values of
θv, when using the two different approaches, is also
observed for the other three CPT profiles as sum-
marised in Table 1. In fact, both methods give a
very similar mean θv of about 0.4 m and a very
similar variance of 0.1 m2. Table 1 also includes the
result obtained for the horizontal scale of fluctua-
tion when using Approach B (i.e. θh = 12.7 m). Note
that only one value can be estimated from the four
aligned CPTs used here. Hence, the result for θh
should be treated with caution, because only a few
Figure 4. Estimated vertical scales of fluctuation from
the berm before caisson installation using Approach A.
Figure 5. Typical result for the estimation of the vertical
scale of fluctuation using Approach B (berm before
caisson installation).
points of the experimental correlation model can
be estimated when using only 4 CPTs (see Fig. 6).
Furthermore, the result estimated for θh is likely to
be dependent on the horizontal distance between
CPTs. This aspect has not been investigated
here because only one single line (section AA’ in
Figure 3) has been used for the analysis of the
berm before caisson installation. Further studies
in this area are required in order to obtain more
conclusive results.
In Figure 6 the solid line indicates the theoreti-
cal correlation model that uses the optimised value
of θh and the dotted line represents the estimated
experimental correlation model. The symbols in
the figure represent the estimated values of the
experimental correlation model, which are com-
puted using Equation 4 and the CPT data at the
horizontal spacing considered.
A useful parameter can now be defined by divid-
ing the horizontal and vertical scales of fluctuation
(i.e. the degree of anisotropy of the heterogeneity

195
 Figure 7. Plan view of CPT locations for berm after
caisson infilling.

Figure 6. Typical result for the estimation of the hori-

zontal scale of fluctuation using Approach B (berm
before caisson installation).

ξ = θh/θv). Using the values shown in Table 1, the

degree of anisotropy for the berm before caisson
installation is approximately 35.

4.2 Variability of the sand in the berm

after caisson installation and subsequent
core infilling
Figure 8. Estimated vertical scales of fluctuation from
Six aligned CPTs along the same straight horizon- the berm after caisson infilling using Approach A.
tal line are used in this section to assess the soil
variability in the berm after positioning the caisson
and infilling of the core with sand. In this case, the
CPTs were taken from the core surface and pen-
etrated the core fill before arriving in the sand of
the berm. Hence, the initial 19.5 m measurements
are not considered in the analysis (see Figure 1)
as they correspond to the core. All CPTs have
different penetration depths through the berm
ranging from approximately 3 m to 6 m. A plan
view of the six aligned CPTs (indicated as BB’) is
plotted in Figure 7, where the distance between Figure 9. Typical result for the estimation of the ver-
CPTs is also indicated. Figure 8 shows the results tical scale of fluctuation using Approach B (berm after
obtained for the vertical scale of fluctuation using core infilling).
Approach A, and highlights the different CPT
depths considered.
The results obtained in this second analysis for
the vertical scale of fluctuation are again consistent
between Approaches A and B. Figure 8 shows the
results obtained for θv when using Approach A. The
maximum fluctuation is generally reached when
the lag distance is between 0.5 m and 1.6 m, giv-
ing values of θv ranging from 0.29 m and 0.89 m
(see Table 2). The average vertical scale of fluctua-
tion has slightly increased in this second analysis.
When using Approach A, θv increases from 0.35 m Figure 10. Typical result for the estimation of the hori-
to 0.53 m whereas the increase in Approach B goes zontal scale of fluctuation using Approach B (berm after
from 0.38 m to 052 m. A significant reduction in the core infilling).
variance is observed when comparing Tables 1 and
2, probably due to more CPTs being used in this sec-
ond analysis. Figure 9 shows a typical result for the obtained for this CPT when using Approach A, i.e.
estimation of the vertical scale of fluctuation when θv = 0.62 m (see Figure 8 and Table 2).
using Approach B. In particular, Figure 9 shows A slight increase of the vertical scale of fluctua-
the estimated correlation model for CPT06 giving tion after the installation of the caisson and sub-
θv = 0.61 m which, again, is very similar to the value sequent core infilling is observed when comparing

196
the results from Tables 1 and 2. Furthermore,
when looking at the results for the horizontal scale
of fluctuation, a significantly larger value of θh is
obtained following Stage 2 of the construction. In
this second analysis, the estimated horizontal scale
of fluctuation gives a value of 18.1 m as illustrated
in Figure 10 and the degree of anisotropy of the
heterogeneity ξ is once again approximately 35.
The increases in the estimated scales of fluc-
tuation seem to suggest that the installation, and
subsequent infilling of the caisson system, lead to
less variability across the site. However, it has been
already highlighted that the values of the hori-
zontal scale of fluctuation should be treated with
caution due to the limited data. In fact, the results
from the horizontal scale of fluctuation when using
Approach B are expected to be influenced by the
distance between CPTs, and this influence has not
been studied in here. Thus, a deeper investigation
involving a larger number of CPTs with smaller
horizontal spacing is desirable.
5 CONCLUSIONS
The vertical and horizontal scales of fluctuation have
been estimated for the berm of Tarsiut P-45 before
and after caisson set-down. Two methods have been
used for the computations of the vertical scale of
fluctuation and these mainly give consistent results.
This suggests that both approaches provide good esti-
mates of θv. The average value for the vertical scale of
fluctuation of the berm before caisson installation is
0.35 m when using Approach A and 0.38 m if using
Approach B. On the other hand, the average θv of
the berm after core infilling when using Approach A
is 0.53 m and 0.52 m when using Approach B.
The estimated horizontal scale of fluctuation for
the berm prior installation of the caisson system is
12.7 m whereas θh for the berm after core infilling is
18.1 m. The degree of anisotropy of the heterogene-
ity is therefore approximately 35 for both analyses.
The increase in the estimated scales of fluctuation
suggests that the installation, and subsequent infill-
ing of the caisson system, lead to a more uniform
material. The computed results, however, should be
treated with caution due to the limited data consid-
ered here. Further studies would be desirable for
more conclusive results, particularly with regard to
the horizontal scale of fluctuation. In this investi-
gation θh may have been overestimated due to the
spacing between CPTs being too large.
ACKNOWLEDGMENTS
The authors gratefully acknowledge the support
provided by the EU FP7 programme under the
project: ‘Geo-Install’ (PIAP-GA-2009-230638).
197
REFERENCES
Baecher, G.B. & Christian, J.T. 2003. Reliability and
statistics in geotechnical engineering. John Wiley &
Sons Inc.
Fenton, G.A. & Griffiths, D.V. 2008. Risk assessment
in geotechnical engineering. John Wiley & Sons,
New Jersey, USA.
Griffiths, D.V. & Fenton, G.A. 2001. Bearing capacity
of spatially random soil: the undrained clay Prandtl
problem revisited. Géotechnique 51(4): 351–359.
Hicks, M.A. & Onisiphorou, C. 2005. Stochastic evalua-
tion of static liquefaction in a predominantly dilative
sand fill. Géotechnique 55(2): 123–133.
Hicks, M.A. & Samy, K. 2002. Influence of heterogeneity
on undrained clay slope stability. Quarterly Journal of
Engineering Geology and Hydrogeology 35(1): 41–49.
Hicks, M.A. & Smith, I.M. 1988. Class A prediction
of Arctic caisson performance. Géotechnique 38(4):
589–612.
Hicks, M.A. & Spencer, W.A. 2010. Influence of het-
erogeneity on the reliability and failure of a long 3D
slope. Computers and Geotechnics 37(7–8): 948–955.
Lloret, M., Hicks, M.A. & Wong, S.Y. 2012 Soil char-
acterisation of an artificial island accounting for
soil heterogeneity. GeoCongress 2012, R.D. Hryciw,
A. Athanasopoulos-Zekkos & N. Yesiller (Eds.), San
Francisco, 2816–2825.
Lloret-Cabot, M., Hicks, M.A. & Eijnden, A.P. van den.
2012. Investigation of the reduction in uncertainty
due to soil variability when conditioning a random
field using Kriging. Géotechnique letters 2: 123–127.
Nuttall, J.D. 2011. Parallel implementation and applica-
tion of the random finite element method. PhD thesis,
, UK.
Phoon, K-K., & Kulhawy, F.H. 1999. Characterization
of geotechnical variability, Canadian Geotechnical
Journal 36(4): 612–624.
Spencer, W.A. 2007. Parallel stochastic and finite element
modelling of clay slope stability in 3D. PhD thesis,
, UK.
Uzielli, M., Vannucchi, G. & Phoon, K-K. 2005.
Random field characterisation of stress-normalised
cone penetration testing parameters. Géotechnique
55(1): 3–20.
Vanmarcke, E.H. 1977. Probabilistic modeling of soil
profiles, Journal of the Geotechnical Engineering Div.
ASCE 103(11): 1227–1246.
Vanmarcke, E.H. 1984. Random Fields: Analysis
and Synthesis. The MIT Press, Cambridge,
Massachusetts.
Wackernagel, H. 2003. Multivariate geostatistics: An
introduction with applications. Springer, Germany.
Wickremesinghe, D.S. 1989. Statistical characterization
of soil profiles using in situ tests. PhD thesis, University
of British Columbia, Canada.
Wickremesinghe, D.S. & Campanella, R.G. 1993. Scale
of fluctuation as a descriptor of soil variability.
Proc. Conf. Probabilistic Methods in Geotechnical
Engineering, Canberra, 233–239.
Wong, S.Y. 2004. Stochastic characterisation and
reliability of saturated soils. PhD thesis, University of
Manchester, UK.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Volume averaging technique in numerical modelling of floating deep

mixed columns in soft soils

P. Becker
Kempfert und Partner Geotechnik, Hamburg, Germany, previous University of Strathclyde, Glasgow, Scotland, UK

M. Karstunen
Chalmers University of Technology, Göteborg, Sweden
University of Strathclyde, Glasgow, Scotland, UK

ABSTRACT: The deformation behaviour of deep mixed columns in anisotropic soft soil is a three-
dimensional problem which has to be considered adequately in numerical modelling. However, in litera-
ture simplifications are often to be found due to computational costs of fully coupled 3D analyses by
either modifying geometry of the problem or material parameters in 2D-plainstrain or -axisymmetric
conditions. This paper uses an enhanced 2D volume averaging technique for numerical modelling of deep
mixed columns. The method enables mapping the 3D problem in two dimensions, and yet modelling the
two constituents (column and soft soil) appropriately within a homogenized material. The performance
of the technique is demonstrated by considering unit cell simulations of floating columns in soft soil, in
which the results from 2D finite element simulations utilizing volume averaging technique are compared
against conventional and fully 3D coupled finite element analyses. It is shown that the quality of the
match is very good but depending on the number of columns and stiffness ratio between columns and
soft soil.

1 INTRODUCTION take the complex stress-strain behaviour of natural

The properties of very soft clays, silts and organic
soils can be improved with deep mixing, a soil
improvement technique in which stabilizing agents,
such as lime and/or cement are mixed into the soil
in situ by using auger-type mixing tools. Deep
mixed columns are nowadays extensively used to
reduce settlements and to improve the overall sta-
bility of road and railway embankments and foun-
dations on soft soils.
Whilst the most typical applications are embank-
ments, increasingly the method is used under foun-
dations. The first category of applications is very
common in Scandinavian countries, whilst the
latter application has been pioneered in Poland as
a cost effective alternative for piling. Due to eco-
nomic, sustainability and environmental reasons
there is an increasing interest to the technique in
the rest of Europe. With deep mixing, the strength
and deformation properties of the soft soil can be
improved and the risk of large horizontal or verti-
cal deformations is substantially reduced.
Numerical methods, such as Finite Element (FE)
analysis, can be used as an alternative to conven-
tional design methods. They are particularly suit-
able for serviceability limit state design. FE analyses
allow adopting advanced constitutive models that
soil and stabilized columns into account. The prob-
lems involving a grid of circular columns under an
embankment or a strip footing are fully three-dimen-
sional problems. As 3D analyses are computation-
ally very expensive an enhanced 2D technique using
the so-called Volume Averaging Technique (VAT) is
adopted. The basic idea is to describe the column-
improved ground as a homogenized composite
material and map the true 3D problem into 2D.
Once the constitutive relations of both composites
are defined, the response of the column improved
ground can be studied in two dimensions subject to
arbitrary loading and boundary conditions.
In this paper the volume averaging technique by
Vogler & Karstunen (2007, 2009) is used in which
two advanced constitutive models for soft soil
and deep mixed column were implemented: the
S-CLAY1S model (Karstunen et al. 2005) to rep-
resent the soft soil and the MNhard model (Benz
2007) to represent the deep mixed columns.
The performance of the volume averaging tech-
nique is demonstrated by comparing the results of
discrete 2D axisymmetric and full 3D models of
unit cell simulations of floating deep mixed col-
umns, i.e. column and soil are modelled separately,
with enhanced 2D analyses using the volume aver-
aging technique.

198
2 VOLUME AVERAGING TECHNIQUE local equilibrium between the soil and the column
material in each integration point can be formu-
2.1 Introduction and fundamental assumptions lated with the following equilibrium conditions,
which assure that there is no stress discontinu-
The basic idea of the volume averaging technique
ity between soil and column material in terms of
is to model the periodic system as a homogenous
radial and shear stress, see also Figure 2:
material instead of modelling columns and natural
soil separately (Fig. 1). The principles adopted by
σ eq
x σ sx = σ cx (2a)
Vogler & Karstunen (2007, 2009) are based on the
ideas of Schweiger & Pande (1986), further refined
by Lee & Pande (1998). The formulation has been σ eq
z σ sz = σ cz (2b)
extended to three dimensions and a new solution
routine has been developed to cope with the highly τ eq
xy τ sxy τ cxy (2c)
non-linear constitutive models.
Within the volume averaging technique a peri- τ eq τ syz = τ cyz (2d)
yz
odic distribution of the columns in the natural soil
is assumed. Furthermore perfect bonding, in other
words no slip between natural soil and columns, is Furthermore, perfect bonding between the col-
assumed. The method allows for adopting any elas- umns and the soft soil is assumed, and hence no
to-plastic constitutive model to the two constitu- slip is permitted between the two materials. This
ents: natural and improved soil. Local equilibrium can be achieved with the following kinematic con-
between soil and column as well as compatibility ditions, see also Figure 3:
and validity of the constitutive relations are satis-
fied through stress/strain redistribution within a ε eq
y ε sy = ε cy (3a)
sub-iterating procedure.
γ zx
eq
γ zx
s
= γ zx
c
(3b)
2.2 Equivalent material stiffness matrix
The constitutive equations for the constituents
Homogenisation is carried out by determin- can be described in terms of effective stress incre-
ing the strain increment and the stress increment ments as:
in the homogenised equivalent material according
to the following averaging rules:
(σ s )′ Ds ε s (4a)
σ eq
Ωs σ + Ωc σ
s c
(1a)

ε eq Ωs ε s + Ωc ε c (1b)

where Ω is a volume fraction. The superscripts eq,

s and c refer to the homogenized material, the soil
and the column material, respectively. σ̇ and ε̇ are the
(total) stress and strain rate tensors, respectively.
In the following it is assumed that the y-axis is
in the vertical direction. The initial assumption of
Figure 2. Local equilibrium conditions between column
and soil for radial (left) and shear stress (right).

Figure 3. Kinematic conditions of perfect bonding

Figure 1. Discrete (left) and homogenised (right) repre- between column and soil for axial (left) and shear strain
sentation of embankment problem. (right).

199
(σ c )′ Dc ε c (4b)
where Dc,s represent the appropriate elasto-plastic
material (or elastic) stiffness matrices for the soil
and the columns, expressed naturally in terms of
effective stress.
In principle any elasto-plastic constitutive law
can be chosen for either of the two constituents.
Here, the soft soil has been modelled with the
S-CLAY1S model (Karstunen et al., 2005) and the
columns with the MNhard model (Benz, 2007),
which are described in Section 3. Considering the
averaging rules (Eq. 1), the equilibrium and kin-
ematic conditions (Eqs. 2 & 3) and the constitutive
relations (Eq. 4), the constitutive relation for the
averaged material can be written as:
(σ eq )′ Deq ε eq (5)
with the equivalent stiffness matrix defined as
D Ωs Ds S1 c Dc S1c (6)
3 CONSTITUTIVE MODELING
3.1 Modelling of soft soil with S-CLAY1S
The S-CLAY1 model (Wheeler et al. 2003) is a
critical state model that is based on the Modified
Cam Clay (MCC) model. Similarly to the MCC
model, S-CLAY1 is assuming isotropic elastic-
ity within the yield surface. Additionally to the
MCC model, S-CLAY1 is capable of simulating
anisotropic soil behaviour, induced by the geologi-
cal and mineralogical history, and the subsequent
loading of natural soil deposits. This anisotropy is
modelled by inclining the yield surface to represent
experimentally observed yield points for soft clays.
Furthermore, changes of anisotropy due to plastic
straining are taken into account by an additional
hardening law. S-CLAY1 is capable of simulating
accurate yield points and the correct development
of volumetric strains and shear strains for recon-
stituted clays, as demonstrated by Karstunen &
Koskinen (2008).
The S-CLAY1S model (Karstunen et al. 2005) is
an extension to the S-CLAY1 model which allows
additionally for modelling destructuration of
bonds as necessary for natural soft clays. The effect
of bonding is described by an “intrinsic yield sur-
face”, which is of the same shape and inclination
as the yield surface of the natural soil (Fig. 4), but
with a size p′mi that is related to p′m of the natural
clay by the amount of bonding × (p′m = p′mi(1 + x)).
The initial value for the amount of bonding x0 can
be estimated based on the sensitivity of the clay.
200
Figure 4. Yield surface of S-CLAY1S (triaxial stress
space).
The values for the model parameters for the
soft soils considered in the unit cell simulation, the
Swedish Noedinge clay are given in Tables 1 to 3.
For definitions of the soil constant and state vari-
ables, the reader should refer to Karstunen et al.
(2005).
3.2 Modelling of deep mixed columns
The deep mixed columns have been modelled with
the MNhard model (Benz 2007). MNhard is for-
mulated in the classical theory of plasticity. Dif-
ferent stress-dependent stiffnesses are assumed
for both elastic unloading/reloading and primary
shear loading. The hyperbolic stress-strain relation-
ship for primary loading is defined by the secant
stiffness modulus E50 (Fig. 5). E50 (and similarly
Eur representing unloading/reloading) are defined
stress dependent:
m
⎛ c ′ cos ϕ ′ + σ 3′ sin ϕ ′ ⎞
ref
E 50 E ⎜ (7)
i ϕ ′ ⎟⎠
ϕ ′ + p′ sin
50
⎝ c′
ref
where E50,ref is the secant stiffness modulus at iso-
tropic reference pressure p′ref, c′ and σ′ are cohesion
and friction angle, σ′3 is the effective minor princi-
ple stress and the power m is defining the amount of
stress dependency. Failure criterion of the MNhard
model is the Matsuoka-Nakai failure criterion,
which is closer to the observed failure behaviour of
granular materials than Mohr-Coulomb. However,
the pre-failure behaviour for serviceability limit
states is not affected by the failure criterion.
As there are not many triaxial tests done on in-
situ column material, the parameter values for deep
mixed columns have been chosen to correspond to
a series of triaxial tests (Aalto, 2003) performed on
excavated columns, which have been fabricated in
soft Finnish clay. The input parameters for the col-
umn are given in Table 4 using the equation E = 2G
(1 + ν′) based on the elasticity theory where G is
the shear modulus.
Table 1. Initial values for state parameters and K0. integration scheme. Because both constituents
exhibit highly non-linear behaviour and an implicit
Layer e0 OCR K0 α0 x0 integration scheme is used, a sub-iteration scheme
was necessary: If the initially predicted internal
Soft clay 2.9 1.25 0.56 0.458 10
strain distribution between soft clay and improved
columns leads to an violation of the equilibrium
conditions (Eq. 2), the strains between this two
Table 2. Conventional soil constants.
materials are redistributed in an iterative scheme
Layer γ (kN/m3) κ ν′ λ M until equilibrium is satisfied (more details in
Vogler & Karstunen, 2007).
Soft clay 15 0.02 0.15 1.15 1.2

4 UNIT CELL SIMULATION

Table 3. Additional soil constants.
4.1 2D and 3D unit cell FE modell and analysis
Layer λi β μ a b
The performance of the volume averaging tech-
Soft clay 0.275 0759 30 10 0.2 nique is demonstrated with the unit cell simulation
of floating deep mixed columns. The soft soil is
improved with lime-cement column of 10 m length
and 0.6 m diameter and a centre to centre spacing
of 1.0 m and 1.5 m, resulting in an improvement
ratio, i.e. volume ratio, of Ωc = 28.3% and 12.6%
respectively.
Furthermore, the performance of the volume
averaging technique is demonstrated by varying
the stiffness of the deep mixed column with values
of Gur,ref = 10, 20 and 50 MPa with a constant ratio
G50,ref = Gur,ref/2.25. The influence of an embank-
ment was modelled using a distributed load of
10 kPa and 50 kPa above a layer of dry crust
with 1 m height. Full three-dimensional and two-
dimensional discrete axisymmetric finite element
Figure 5. Hyperbolic stress-strain relationship in pri- calculations have been compared with enhanced
mary loading for a standard drained triaxial test. axisymmetric calculations using volume averag-
ing. The 3D mesh contains about 4,500 15-noded
wedge-elements, whereas for the axisymmetric
Table 4. Input parameters for the columns.
analysis 480 15-noded triangles were sufficient,
γ G50ref Gurref c′ see also Figure 6. For both calculations first the
Layer (kN/m3) (kPa) (kPa) m ϕ′ (kPa) ν′ stresses were initialized using a K0-procedure. Then
the embankment load was brought up undrained,
Column 15 4444 10000 0.8 35 15 0.35 before simulating consolidation to a maximum
remaining excess pore water pressure of 1 kPa.
Reference stress for stiffness, p′ref = 100 kPa.

4.2 Local equilibrium conditions

Table 5. Input parameters for the dry crust.
Local equilibrium conditions (Eq. 2) between col-
γ Ε c′ umn and soft soil are fulfilled if radial stresses of
Layer (kN/m3) (kPa) ϕ′ (kPa) ν′ column and soil are equal at a virtual column-soil
interface.
Dry crust 15 750 30 6 0.15
The radial stresses of column (C) and soil (S)
from discrete modelling of the unit cell with
axisymmetric (2D-Disc) and full 3D simulations
3.3 Solution strategy and implementation
(3D-Disc) are compared with the Volume Aver-
into finite element code
aging Technique (VAT) in Figure 7. The results
The volume averaging technique is implemented are presented for a volume ratio Ωc = 0.283,
into the PLAXIS (Brinkgreve et al. 2010) finite Gur,ref = 50 MPa and a variation of loading condi-
element code using an implicit backward Euler tions with Δσ = 10 kPa and 50 kPa.

201

Figure 6. Unit cell FE models (2D and 3D) of floating
columns.
Figure 7. Radial stresses in column, soil and VAT.
Local equilibrium is perfectly achieved for
Δσ = 10 kPa. However, higher loads result in
slightly over predicted radial stresses within the
columns whereas the stresses in the soil and VAT
are almost equal.
202
Figure 8. Shear stresses in column, soil and VAT.
Figure 9. Radial strains in column and VAT.
For shear stresses local equilibrium is achieved
for all loading conditions, variations of input
parameters and volume ratios. Figure 8 shows for
example the results for higher loads. Scatter at the
top and toe of column is due to the load transfer
and is negligible.
The radial strains have to achieve equilibrium
for radial stresses between column and soil. It can
be seen from Figure 9 that radial strains of the vir-
tual column within VAT are in accordance with the
discrete simulations. But the radial strains of VAT
are increasing with higher load or reduced volume
ratio.
4.3 Kinematic conditions
The kinematic conditions of the volume averaging
technique are based on perfect bonding between
column and soft soil, i.e. no differential displace-
ments between column and soft soil.
Perfect bonding is achieved for the presented
variations. Figure 10 shows the axial strains with
a perfect match for 10 kPa loading conditions.
There is a small overestimation of axial strains
of VAT with increasing vertical load which is still
acceptable. Whereas the vertical strains in column
and soil are showing perfect bonding with the dis-
crete simulation of floating columns.
4.4 Settlements
The time dependent settlements in the centre of
the column and the soil at the unit cell boundary
are shown in Figure 11 for 10 kPa loading and in
Figure 12 for 50 kPa loading with variations of
column stiffness.
The previous described sensitivity of the volume
averaging technique related to increasing loading
conditions and their influences on the equilibrium
and kinematic conditions can be seen in the settle-
ments as well. The VAT provides very good results
for the 10 kPa load case and slightly overestimates
the settlements for 50 kPa loading.
Figure 10. Axial strains in column, soil and VAT.
Figure 11. Settlements (load Δσ = 10 kPA).
203
Figure 12. Settlements (load Δσ = 50 kPA).
5 CONCLUSIONS
The performance and accuracy of the volume
averaging technique has been demonstrated with
simulations of floating columns in a unit cell.
The technique has been verified against conven-
tional, fully three-dimensional analyses, where
soft soil and deep mixed columns were modelled
separately.
The volume averaging technique provides a good
agreement with conventional and full 3D analyses.
Equilibrium and kinematic conditions have been
shown to satisfy the requirements within the natu-
ral range of applicability of ground improvements
with deep mixed columns in soft soil.
Furthermore, the volume averaging technique
allows for fully coupled analyses. The implemen-
tation allows for the determination of the time-
settlement behaviour of improved areas.
ACKNOWLEDGEMENTS
The research leading to these results has received
funding from the European Community’s Seventh
Framework Programme FP7/2007-2013 under
grant agreement n° PIAG-GA-2009-230638.
REFERENCES
Aalto, A. 2003. Full scale tests in field using dry mixing
method. In Vermeer et al. (eds), Proc. Int. Workshop
on Geotechnics of Soft Soils—Theory and Practice.
Noordwijkerhout, The Netherlands, Essen: VGE,
505–510.
Benz, T. 2007. Small strain stiffness of soils and its
numerical consequences. PhD Thesis, University of
Stuttgart, Germany, book 55.
Brinkgreve, R.B.J., Swolfs, W.M. & Engin, E. 2010. Plaxis
2D 2010, manual.
Karstunen, M. & Koskinen M. 2008. Plastic anisotropy
of soft reconstituted clays. Canadian Geotechnical
Journal 45: 314–328.
Karstunen, M., Krenn, H., Wheeler, S.J. Koskinen, M. &
Zentar, R. 2005. Effect of anisotropy and destructu-
ration on the behaviour of Murro test embankment.
ASCE International Journal of Geomechanics, Vol. 5,
No. 2, 87–97.
Koskinen, M. & Karstunen, M. 2004. The effect of,
structure on the compressibility of Finnish clays.
SGF Report 3:2004, Proc. of XIV Nordic Geotechni-
cal Meeting 19–21 May 2004, Ystad, Sweden, Vol. 1,
A-11–A-22 Linköping: Swedish Geotechnical Society.
Lee, J.-S. & Pande, G.N. 1998. Analysis of stone-column
reinforced foundations. Int. J. for Numer. Anal. Meth.
Geomech., Vol. 22, 1001–1020.
Schweiger, H.F. & Pande, G.N. 1986. Numerical Analysis
of stone column supported foundation. Computers
and Geotechnics, 347–372.
204
Vogler, U. & Karstunen, M. 2007. Numerical model-
ling of deep mixed columns with volume averaging
technique. In Pande, G.N. and Pietruszcak, S. (eds.),
Proc. of the 10th intern. Symp. on Numerical Mod-
els in Geomechanics (NUMOG X), Rhodes, Greece,
25–27 April 2007, London: A.A Balkema, 495–503.
Vogler, U. & Karstunen, M. 2009. Application of volume
averaging technique in numerical modeling of deep
mixing. In Karstunen, M. & Leoni, M. (eds), Geo-
technics of Soft Soils – Focus on Ground Improve-
ment, London: Taylor & Francis Group, 189–195.
Wheeler, S.J., Näätänen A., Karstunen M. &
Lojander M. 2003. An anisotropic elasto-plastic
model for natural soft clays. Canadian Geotechnical
Journal 40(2):403–418.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Comparison between theoretical procedures and field test results

for the evaluation of installation effects of vibro-stone columns

E. Carvajal & G. Vukotić

Kellerterra S.L., Madrid, Spain

J. Castro
University of Cantabria, Santander, Spain

W. Wehr
Keller Holding GmbH, Offenbach, Germany

ABSTRACT: Several theoretical procedures to estimate the soil improvement produced by installation of
vibro-stone columns are described. Particularly, finite element model and analytical solutions of a cylindrical
cavity expansion were compared with results from an actual field test which was performed in silty sand and
clayey soil treated with a column group. The results show that after dissipation of pore pressure the instal-
lation effects produce considerable improvement due to a large increase of the horizontal effective stress
and due to densification process of sand. The load settlements response of the tested column group has
been analyzed and compared with theoretical estimation of the improvement with and without installation
effects, and with Priebe’s analytical solution. It is observed that the column group installation effects have an
important influence that should be evaluated with more advanced modelling or directly with in situ testing.

1 INTRODUCTION installation effects. However, densification effect

could be considered in design on the bases of the
1.1 Soil grain size influence to installation effects properly evaluated final soil characteristics (Priebe
1995). Experience has demonstrated that evaluation
The main goal of the vibrocompaction method is
of the compaction suitability is not related only with
to increase the relative density of soil (Dr). To allow
the grain size distribution but also with an indicator
the adequate transmission of vibratory energy it is
of soil strength, e.g. cone penetration resistance. In
necessary to fluidise the soil by controlled water jets
this sense, Figure 1 shows empirical relation pro-
that increase pore water pressure and reduce the fric-
posed by Massarsch (1991), for the evaluation of
tional contact between the soil particles. Then, after
compaction suitability as a function of friction
a certain time of compaction, rearrangement of the
ratio and cone penetration tip resistance.
particles in a denser state will be achieved. More
Several authors analyzed effects of the
details related to vibrocompaction and its effects
vibro-stone column installation based on field
can be seen in the literature (Sonderman & Wehr
2004, Greenwood & Kirsch 1984, Slocombe et al.
2000, Kirsch & Kirsch 2010). Fines content superior
of 10% (Degen 1997) and permeability lower than
10−3 cm/s (Greenwood & Kirsch 1984) are consid-
ered as limits for the soil densification by the appli-
cation of vibrocompaction due to strong damping
of vibratory energy and time effects. Vibro-stone
columns represent a technological development
that has allowed extension of the application limits
of deep vibration techniques to fine cohesive soils
(Kirsch & Kirsch 2010).

1.2 Assessment of installation effects

Standard procedures for the stone column
design commonly do not take into account any Figure 1. Compactability of soils. Massarsch (1991).

205
measurements (Watts et al. 2001, Kirsch 2006, Gäb
et al. 2007, Castro 2008), but related to a specific
cases and hence, cannot be generalized in a straight-
forward manner (Castro & Karstunen 2010).
On the other hand, there have been various
attempts with different approaches for theoreti-
cal modelling (Kirsch 2006, Elshazly et al. 2007,
Guetif et al. 2007, Castro & Karstunen 2010,
Castro et al. 2012). In general, field measurements
and theoretical approaches have shown that instal-
lation of vibro-stone columns in saturated clays
cause its remolding close to the vibrator, and pro-
duces increase of pore pressures and horizontal
stress. However, after a relative short period of
consolidation, soil tends to gain in stiffness and
additionally provides greater confinement to the
columns. Whereas, in intermediate soils with vari-
able fines content such as silty sands or sandy silts,
vibrations could also achieve a significant den-
sification by rearranging the soil particles into a
denser state. This would be the case of soils in the
marginal compactable zone shown in Figure 1.
In the following sections some theoretical pro-
cedures that are considered as reasonably suitable
to evaluate the vibro-stone column installation
effects will be described and compared with field
trial measurements obtained during column
installation.
2 MODELLING PROCEDURES
2.1 Vibro equipment performance
A rigorous modelling should consider both, the
actions imposed by vibrator and the soil response.
Due to the complex interaction between vibrator
and soil, densification modelling of vibro tech-
niques is still a challenging task (Heibrock et al.
2006). However, various attempts have been done
by several authors (Fellin 2000, Cudmani et al.
2003, Arnold et al. 2008, Arnold & Herle 2009).
Ranges of vibrators basic characteristics, and ones
usually considered for modelling are shown in
Table 1.
Usually dynamic effects are substituted by
quasi-static loading; one-dimensional or 2D
axisymmetric conditions are often adopted, as
well as dashpot-spring and full numerical models.
Furthermore, in order to simulate the shear waves
Table 1. Vibrator characteristics.
Freq. Amplitude Horiz. force Weight Diameter
(Hz) (mm) (kN) (kN) (mm)
25–60 6–50 150–700 15–45 300–500
206
induced into the ground, hypoplastic model has
been considered, which fits more accurately to the
sand behavior. In general, results obtained from
models are in accordance with the experiences on
actual observed effects, stressing the importance
of the compaction time in order to achieve better
results. It is also important to emphasize the model
performed by Arnold et al. (2008), which confirms
the existence of a zone with low densification
immediately close to the vibrator, a second zone
with best compaction results that extends from
approximately 0.5 to 3 m from vibrator axis, and
a third zone at a distance of more than 3 m, where
soil compaction does not occur.
Mentioned findings are directly related to the
vibrocompaction process and may be applied
to vibro-stone column installation for predomi-
nantly sandy soils. For predominantly fine grained
soils the densification is insignificant using
vibrocompaction.
2.2 Cavity expansion
As it was mentioned, if vibro-stone column are
performed in saturated clays the vibration energy
has a negligible effects on the densification of the
natural soil, but effects caused by the displacement
of soil due to column installation can be evaluated
by the application of the cylindrical cavity expan-
sion theory. As columns are built in a reasonably
short period the undrained conditions should be
considered even for relatively permeable soils, e.g.
sandy silts. Due to difficulties to determine the
internal cavity pressure produced by the vibrator
it is more suitable to simulate column expansion
considering radial displacement (Castro 2008).
Egan et al. (2008) summarized the development
of analytical solutions in three aspects: soil behav-
ior, consideration of finite or infinite medium,
and the influence of the initial radius of the cav-
ity. Concerning mentioned aspects there are sev-
eral solutions proposed by various authors, e.g.
Vesić (1972), Carter et al. (1986), Yu & Houlsby
(1991), Yu (2000). For complex soil models numer-
ical analysis very frequently is selected as more
adequate, but large strains should be enabled and
considered.
In any case the key aspects to be analyzed will
be the rapid increase of pore pressure and its dis-
sipation over time, as well as the final stress state in
terms of the at rest earth coefficient K0.
2.3 Stress distribution
When the cavity expands to a certain radius, a
plastic boundary is obtained at a radial distance
ρF measured from the cavity axis. Stresses between
cavity and ρF are in the plastic zone, whereas
beyond ρF stresses are in the elastic zone. Thus, the
radial stress pF at point ρF represents fully plastic
behavior. For a cohesive and frictional soil in an
infinite medium and assuming Mohr Coulomb
behavior Baguelin et al. (1978) proposed the
expression (1) to determine the plastic radius,
based on the assumption of no volume change and
considering total stresses:
a 2 (p0 + c ⋅ cot φ ) ⋅ sin φ
= (1)
ρF2 G
where a is the radius of the cavity; p0 is the pre-
existence horizontal stress; and G is the shear
modulus. φ is the angle of internal friction and c
the cohesion of the soil.
It can be noted that for purely cohesive clay with
φ = 0, or adopting Tresca criterion, equation 1 turns
into equation 2, which is the same solution devel-
oped by Randolph & Wroth (1979) for large strains:
a2 cu
= (2)
ρF2 G
where cu is the undrained shear strength.
Since plastic radius is known, and combin-
ing the equilibrium equation of cylindrical cavity
expansion with Mohr Coulomb failure criterion,
the solution for distribution of major principal
stress σr in the radial direction within the plastic
zone can be obtained for purely cohesive soil with
equation 3, and for cohesive-frictional soils with
equation 4, as a function of radial distance ρ.
ρF
σ r = pF + c u ⋅ ln
ρ strength for standard columns grid.
where pF = p0 + cu
1− K a
⎛ ρ2 ⎞ 2 As it was stated by contractors experience and
σ r = (pF + c ⋅ cotφ ) ⋅ ⎜ F2 ⎟ − c ⋅ cot φ
⎝ρ ⎠
where pF = p0 ⋅ (1 + sinφ) + c ⋅ cotφ; and Ka = tan2(45-
φ/2).
Whereas in the elastic zone the well-known
solution can be used to determine radial stress
distribution:
ρF2
σr p0 (pF − p0 ) ⋅
ρ2
2.3.1 Pore pressure
Considering that excess pore pressures are caused
only by the variation of mean total stresses, all
pore pressure increments take place only in the
plastic zone. In the same way, if we take into
account the equilibrium equation of cylindrical
cavity expansion and adopting zero pore pressure
in the elastic zone as a boundary condition, it can
be obtained the equation 6 to determine de excess
pore pressure distribution as function of ρ.
⎛ρ ⎞
Δu 2c u ln ⋅ ⎜ F ⎟
⎝ ρ⎠ (6)
2.3.2 Influence of soil constitutive model
In the previous sections the elastic perfectly-plastic
models have been applied in order to highlight
the possibilities of available and relatively simple
analytical solutions to reach a simple approach
of most important effects of vibro-stone columns
installation. In fact, saturated clays with low sensi-
tivity may be well modelled with these approaches.
With hardening plastic behavior a first enhance-
ment of the modelling could be made consider-
ing the stress state modification due to plastic
strains and stress dependency of stiffness. This
model is suitable for the application of both, sandy
and clayey soils subjected to a cavity expansion,
although calculations usually have to be done with
numerical model.
On the other hand, Castro & Karstunen (2010)
presented a numerical modelling of Bothkennar
clay based on S-CLAY1 and S-CLAY1S, which
are Cam clay-type models that consider anisotropy
and destructuration. Results agree with practical
experience, showing a great remolded zone close
to vibrator. Authors recommend for practical pur-
(3) poses a reduction of 15%–20% of initial undrained
2.4 Back-calculation procedures
(4) several field measurements, during vibro-stone
columns installation a certain heave on the
surface could appear (Egan et al. 2008), and is
more important for closer distances between
columns. It confirms that soil stress state var-
ies, and from the heave measurement data it is
possible to perform back-calculation of final soil
characteristics.
The real scale load test on treated soil might be
very useful for back analysis regarding the addi-
(5) tional stiffness of the soil. This procedure can be
used to estimate the installation effects of column
groups, e.g. Kirsch (2006) proposed the evaluation
of an enhancement zone placed around column
group by means of stiffness variation and its com-
parison with measured settlements.
207
3 FIELD TEST
3.1 Field measurements
In order to evaluate the installation effects of
vibro-stone columns executed with dry bottom
feed method in a profile of intermediate soil i.e.
silty sand and clayey silt, field test consisting of
a group of 13 vibro-stone columns, 20 m to 25 m
long and with diameter between 0.90 and 1.00 m
was carried out. Instrumentation campaign was
composed by 12 piezometers distributed along
4 lines and located at depths of 6 m, 10 m and
16 m each. To control vertical displacements and
stresses 1 extensometer was extended up to depth
of 40 m. Figure 2b shows a section and plan view
of the field test. In order to compare measurements
with theoretical procedures, the pore pressure was
monitored during the installation of one column
(see column A, Fig. 2b).
Piezometer radial distances from the axis of the
column A can be seen in Figure 2b (distance of
1.25 m for piezometer Pz3, 1.8 m for piezometer
Pz4, whereas both piezometer Pz1 and Pz2 were set
at radial distance of 5 m).
The first evaluation step of the effects of col-
umns installation can be seen in Figure 2a. Typical
cone penetration resistance between columns
increase 8 to 14 times comparing with tip resist-
ance before the treatment. From the combination
of friction ratio FR% and tip resistance it can be
observed that soil most probably experienced a
certain improvement due to densification accord-
ing to the marginal compactable zone shown in
Figure 1.
Regarding excess pore pressure, in Figure 3
are presented both, the peak values measured by
Figure 2. Field test characteristics, (a) cone penetration resistance after and before the treatment, (b) plan view and
section of field test and (c) geometry of finite element modelling.
208
piezometer and its comparison with the theoretical
predictions that are exposed in next chapter.
Maximum values of 65 kPa and 110 kPa were
measured at depths of 10 m and 16 m respectively,
and at radial distance of 1.8 m. At a radial dis-
tance of 5 m were observed peaks values of 50 kPa
and 68 kPa, probably related to the presence of
sand layer that is able to transmit the generated
vibrations.
Figure 4 shows that after column installation
the excess pore pressure is rapidly dissipated and
after 5 hours (300 minutes) almost no excess pore
pressure was detected. Moreover, the same Figure
shows that consolidation rate estimated with finite
element method reasonably agrees with the meas-
ured consolidation rates.
3.2 Modelling of field measurements
Finite element modelling with Plaxis v8 code was
performed and results were compared with analyti-
cal solutions presented in chapter 2.2. Modelling
comprises an axisymetric model of 4 soil layers
with Mohr-Coulomb behavior and extrafine mesh
of 15-noded elements close to the column axis. The
whole geometry is indicated in Figure 2c.
Column expansion was modeled by a prescribed
displacement. According to Carter et al. (1979) the
expansion from a finite radius can be related to the ide-
alized expansion on an infinite medium where radius
starts from 0, by the relationship r2f,fin − r20,fin = r2c – 0.
Therefore, adopting initial radius r0,fin of 0.1 m and
actual column radius rc of 0.55 m, the magnitude of
expansion to be considered is rf,fin − r0,fin = 0.46 m.
To enable the soil free movements roller bound-
aries were assumed on all sides. Soil properties are

Figure 3. Predicted and measured peaks of excess pore
pressures.
Figure 4. Measurements of excess pore pressure
dissipation.
listed in the Table 2. The water table was situated
at 1.20 m depth. Because the modelling is aimed to
study the effects on the soil, the column was con-
sidered only as a void with infinite permeability.
It was adopted Plaxis K0-procedure corresponding
to normally consolidated soil.
Calculations were performed in two phases,
with undrained conditions for the first phase and
consolidation analysis for the second phase. The
“up-dated” option of Plaxis software was activated
to take into account large strains.
Figure 5 shows the distribution of initial excess
pore pressures Δu after cavity expansion, obtained
at the same depths of piezometers. It is observed
that results from finite elements modelling and the
analytical solution (equation 6) are quite similar.
The undrained shear strength, adopted to estimate
the plastic radius ρF, was determined combin-
ing equations (1) and (2) and considering effec-
tive stresses and effective soil parameters (c’, φ’).
Results show that the plastic radius ρF coincides
with the influence zone of Δu, and its values
209
decrease with depth, reaching approximate radial
distances of 4 m, 3 m and 2.5 m at 6 m, 10 m and
16 m depths respectively. It is also in accordance
with the increase of undrained shear strength cu,
while the maximum values of Δu increase with
depth. At a radial distance close to 1.75 m the pore
pressures are almost equalized at the three ana-
lyzed depths (6 m, 10 m and 16 m), and the Δu
values that fits best with theoretical procedures are
close to this zone, near normalized radial distance
of 4 column diameters.
On the other side, Figure 6 shows the results
from finite element method for the distribution
of normalized effective mean stress p and normal-
ized coefficient of earth pressure K at the end of
consolidation. Therefore, it is supposed that the
stress state associated to increased K and p values
will be suitable to estimate the stiffness generated
due to column installation. Although the increase
of the stiffness is dominated by the great increase
of the radial stress, the reduction of the circum-
ferential stress in the plastic zone has to be taken
into account.
Consequently, for the estimation of the enhanced
modulus of soil, instead of the coefficient of lateral
earth pressure K, the mean effective stress should
be considered. In general, a power law in the form
E = E0 (p’/p’0)m might be used. However, K is still
the best indicator to assess the extent to which the
soil stiffness increases due to columns installation.
Figure 6 also shows that the zone influenced by
K/K0 = 1.5–2 could be very suitable for the esti-
mation of the stiffness increase. This zone is in
between 4 and 6 column radii at 16 m depth, while
at 6 m depth is between 10 and 14 column radii.
It is important to emphasize that in the analyzed
case less difference between K/K0 and p/po was
observed with the increase of the depth.
3.3 Field load testing
An equivalent load of 75 kPa was placed above
the field-test site shown in Figure 1b. Loading
was made of 4 m height backfill material, widely
larger than the testing zone. In Figure 7 the load
settlements behavior of a group composed of
13 columns are indicated.
Several theoretical estimations are compared
with the actual measurements, taking into account
the situation without improvement and situations
considering the improvement with and without
any installation effects.
Thus, a settlements estimation with coefficient of
lateral earth pressure at rest K0 is compared with the
case of improved lateral earth pressure K* = 1.75
according to the modelling results shown in
Figure 6, and with the analytical solution proposed
by Priebe (1995) which represents the case of K = 1.
 Table 2. Soil properties.

Depth γ’ c’ φ’ Eoed* kh
Layer (m) (kN/m3) (kPa) (o) (kPa) (m/s)

1. Silty sand 0–3 8.50 1 33 20,000 1.15 ⋅ 10−8

2. Silty sand 3–8 8.50 5 25 4500 1.15 ⋅ 10−8
3. Clayey silt 8–17 8.50 10 17 3600 2.30 ⋅ 10−9
4. Silty sand 17–25 8.50 5 28 7200 1.15 ⋅ 10−8

*confined modulus.

Figure 5. Estimation of excess pore pressures due to

vibro-stone column installation.

Figure 7. Load settlements results of field test.

Figure 6. Distribution of effective mean stress and

coefficient of lateral earth preasure.

Figure 8 shows the accumulated compressibility
monitored by the incremental extensometer
installed in the middle of field test, together with
the theoretical estimation of improvement consid-
ering the same cases described in Figure 7. During
the load testing the extensometer only provided
data up to a maximum foundation load of 35 kPa.
The reduction of settlements obtained with the
consideration of K* improved by cavity expansion
is about 20–30% greater than the case of improve-
ment without any installation effects (K0). The
results closer to the actual measurements are those
estimated according to Priebe (1995).
Figure 8. Accumulated compressibility results of field
test.
Furthermore, as have been stated by Kirsch
(2006), the settlement reduction in terms of
improvement factor “n” depends on the load
level, which can be noted comparing results from
Figures 7 and 8. It has to be stressed that the dif-
ference between actual measurements and theo-
retical estimation could be attributed to the global
installation effects of the column group, possible
densification effects and the global increase of

210
horizontal stresses. The evaluation of these global
effects should be made with more advanced model-
ling or directly by means of in situ testing.
4 CONCLUSIONS
Installation effects should be considered in the
design of vibro-stone column treatments, and can
be estimated according to cavity expansion theory
with reasonable accuracy. The improvement induced
by individual column installation can be expressed
by the increase of coefficient of earth pressure. For
the evaluation of global effects induced by a column
group installation, advanced modelling should be
performed to consider more realistic characteristics
of vibro-stone column execution and its effects.
REFERENCES
Arnold, M., Herle, I. & Wehr, J. 2008. Comparison of
vibrocompaction methods by numerical simulations.
In Karstunen et al. (eds), Geotechnics of Soft Soils:
Focus on Ground Improvement: University of Strath-
clyde, Glasgow.
Arnold, M. & Herle, I. 2009. Modelling of vibro-
compaction using hipoplasticity with intergranu-
lar strains. Proceedings of the 17th International
Conference on Soil Mechanics and Geotechnical
Engineering—ICSMGE.
Baguelin, F., Jezequel, J.F. & Shields, D.H. 1978. The
Pressuremeter and Foundation Engineering. TransTech
Publications.
Carter, J.P., Randolph, M.F. & Wroth, C.P. 1979. Stress
and pore pressure changes in clay during and after
the expansion of a cylindrical cavity. International
Journal for Numerical and Analytical Methods in
Geomechanics.
Carter, J.P., Booker, J.R. & Yeung, S.K. 1986. Cavity
expansion in cohesive frictional soils. Geotechnique
36. No. 3:pp. 349–358.
Castro, J. 2008. Análisis teórico de la consolidación y
deformación alrededor de columnas de grava. Ph.D.
thesis, University of Cantabria, Santander, Spain.
Castro, J. & Karstunem, M. 2010. Numerical simulations
of stone column installation. Canadian Geotechnical
Journal 47(19):1127−1138.
Castro, J., Kamrat-Pietraszewska, D. & Karstunem, M.
2012. Numerical modelling of stone column installation
in Bothkennar clay. International Symposium on Ground
Improvement IS-GI Brussels; ISSMGE-TC 211.
Cudmani, R.O., Osinov, V.A., Bühler, M.M. & Gudehus, G.
2003. A model for de evaluation of liquefaction sus-
ceptibility in layered soils due to earthquakes. 12th
Panamerican Conference on SMGE. Cambridge.
Degen, W. 1997. Vibroflotation ground improvement.
Altendorf, unpublished.
Egan, D., Scott, W. & McCabe, B. 2008. Installation
effects of vibro replacement stone columns in soft
clay. In Geotechnics of Soft Soils—Focus on Ground
Improvement. Glasgow, 3–5 September 2008.
211
Elshazly, H., Elkasabgy, M. & Elleboudy A. 2008. Effect
of Inter-Column Spacing on Soil Stresses due to
Vibro-Installed Stone Columns: Interesting Findings.
Geotech Geol Eng 26:225–236. Springer Science—
Business Media.
Fellin, W. 2000. Rütteldruckverdichtung als plastodyna-
misches Problem. Advances in Geotechnical Engineer-
ing and Tunnelling Vol 3.
Gäb, M., Schweiger, H.F., Thurner, R., & Adam, D.
2007. Field trial to investigate the performance of a
floating stone column foundation. In Proceedings
of the 14th European Conference on Soil Mechan-
ics and Geotechnical Engineering, Madrid, Spain,
24–27 September 2007. Millpress, Amsterdam, The
Netherlands. pp. 1311–1316.
Greenwood, D.A. & Kirsch, K. 1984. Specialist Ground
Treatment by Vibratory and Dynamic Methods.
Piling and Ground Treatment. The institution of Civil
Engineers: Tomas Telford, London.
Guetif, Z., Bouassida, M. & Debats, J.M. 2007.
Improved soft clay characteristics due to stone col-
umn installation. Computers and Geotechnics 34(2):
104–111.
Heibrock, G., Kebler, S. & Triantafyllidis, TH. 2006. On
modelling vibro-compaction of dry sands. Numerical
Modelling of Construction Processes in Geotechnical
Engineering for Urban Environment—Triantafyllidis
(ed). Taylor & Francis Group. London. ISBN 0 415
39748 0.
Kirsch, F. 2006. Vibro stone column installation and
its effect on ground improvement. In Proceedings
of Numerical Modelling of Construction Processes
in Geotechnical Engineering for Urban Environment,
Bochum, Germany, 23–24 March 2006. Taylor and
Francis, London: 115–124.
Kirsch, K. & Kirsch, F. 2010. Ground Improvement by
Deep Vibratory Methods. New York: Spon Press.
Massarsch, K.R. 1991. Deep Soil compaction Using
Vibratory Probes in Deep Foundation Improvement.
STP1089 ASTM.
Priebe, H.J. 1995. Design of vibro replacement. Ground
Engineering 28(10):31–37.
Randolph, M.F. & Wroth, C.P. 1978. An analytical
solution for the consolidation around a driven pile.
International Journal for Numerical and Analytical
Methods in Geomechanics. Vol. 3: pp. 217–229.
Slocombe, B.C., Bell, A.L. & Baez, J.L. 2000. The den-
sification of granular soils using vibro methods.
Geotechnique 50. No. 6. 715–725.
Sonderman, W. & Wehr, W. 2004. Deep vibro techniques.
Ground Improvement. 2nd Edition Ed. Moseley &
Kirsch.
Vesić, S.A. 1972. Expansion of Cavities in Infinite Soil
Mass. Journal of Soil Mechanics and Foundation
Engineering Division. ASCE. 98(3):pp. 265–290.
Watts, K.S., Johnson, D., Wood, L.A. & Saadi, A.
2000. An instrumented trial of vibro ground treat-
ment supporting strip foundations in a variable fill.
Géotechnique 50(6):699–708.
Yu, H.S. & Houlsby, G.T. 1991. Finite cavity expansion
in dilatant soils: loading analysis. Geotechnique. 42(2):
pp. 173–183.
Yu, H.S. 2000. Cavity expansion methods in geomechan-
ics, Dordrecht, Boston; Kluwer Academic Publishers.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Numerical analyses of stone column installation in Bothkennar clay

J. Castro
University of Cantabria, Santander, Spain

M. Karstunen
Chalmers University of Technology, Gothenburg, Sweden
University of Strathclyde, Glasgow, UK

N. Sivasithamparam
University of Strathclyde, Glasgow, UK
Plaxis BV, Delft, The Netherlands

C. Sagaseta
University of Cantabria, Santander, Spain

ABSTRACT: The paper presents the results of numerical simulations studying the installation effects
of stone columns in a natural soft clay. Stone column installation is modelled as an undrained expansion
of a cylindrical cavity, using the finite element code PLAXIS that allows for large displacements. The
properties of the soft clay correspond to Bothkennar clay, a soft Carse clay from Scotland (UK). The
complexity of this material is simulated via two advanced recently developed constitutive formulations
able to account for the soil structure, namely S-CLAY1 and S-CLAY1S. Modified Cam Clay model is also
used for comparison purposes. The paper shows the new stress field and state parameters after column
installation and the subsequent consolidation process. This sets the basis for including installation effects
in studying the settlement reduction caused by stone columns.

1 INTRODUCTION The study is restricted to those cases where

Stone columns are a common improvement tech-
nique for foundation of embankments or struc-
tures on soft soils. The gravel columns have a
higher strength, stiffness and permeability than the
natural soft soil. Therefore, they improve the bear-
ing capacity and the stability of embankments and
natural slopes, reduce the total and the differential
settlements, accelerate the soil consolidation and
reduce the liquefaction potential. Besides, column
installation also modifies the properties of the sur-
rounding natural soft soil. However, design of stone
columns does not usually consider those installa-
tion effects and is typically based on their perform-
ance as rigid inclusions (Barksdale & Bachus 1983,
Balaam & Booker 1985, Castro & Sagaseta 2009).
Some authors (Priebe 1995) account for a certain
changes in the stress state during installation by
using higher values of the earth pressure at rest
than that for the natural soil. This paper discusses
the alteration of the stress state and the initial state
parameters of the soil caused by the column instal-
lation, which is nowadays one of the major con-
cerns in an accurate design (Egan et al. 2008).
columns are used in purely cohesive soils and the
main effect is considered to be the cavity expansion
induced by the vibrator penetration. Stone columns
may also be used in soils that have an important
granular fraction and in those cases, the densifica-
tion caused by the vibration is probably the main
installation effect. That process is beyond the scope
of the paper and has been analysed using mostly field
measurements (Slocombe et al. 2000, Massarsch &
Fellenius 2002), as the numerical modelling is quite
complicated (Arnold & Herle 2009).
Experimental studies have shown some of the
effects of column installation. For example, the
increase of pore pressures and horizontal stresses,
and the remoulding of the surrounding soil have been
measured in the field (Watts et al. 2000, Watts et al.
2001, Kirsch 2004, Gäb et al. 2007, Castro & Sagaseta
2012). There have also been attempts to investigate
these effects through physical modelling of the proc-
ess by means of centrifuge testing (Lee et al. 2004,
Weber et al. 2010), but the soils used are reconstituted
and hence, not fully representative of natural clays.
Numerical modelling is a useful tool that may help
to derive some conclusions or recommendations

212
about installation effects for column design, if the bonding and degradation of bonds, using an
assumptions made in the model are validated by intrinsic yield surface and a hardening law describ-
experimental measurements. Furthermore, few ing destructuration as a function of plastic straining.
attempts (Kirsch 2006, Guetif et al. 2007) had been The models have been implemented as User-defined
made in this field, using simple soil models. There- soil models in Plaxis. An implementation that uses
fore, the authors recently decided to study installa- an implicit integration scheme (Sivasithamparam
tion effects numerically using advance soil models 2012) has been used, instead of an explicit previous
to reproduce the behaviour of natural structured version in Castro & Karstunen (2010).
soft soils (Castro & Karstunen 2010). The results The values for S-CLAY1 model param-
were satisfactory, as they compared well with field eters (soil constants) and the initial state vari-
measurements (e.g. Roy et al. 1981, Kirsch 2006). ables for Bothkennar clay are listed in Tables 1
Here, a detailed analysis of those numerical results and 2, respectively. The additional parameters for
once the excess pore pressures have been dissipated is S-CLAY1S are detailed in Table 3. S-CLAY1 model is
presented. The new state of the soil that is obtained formulated to be a hierarchical model, which reduces
after column installation and full consolidation sets to MCC model by assuming μ = 0 and α0 = 0.
the basis of a future study of the influence that stone The geometry of the numerical model and
column installation has on the ground improvement, the finite element mesh is shown in Figure 1.
especially on the settlement reduction. Parametric studies were carried out to check how
wide the model should be to have a negligible influ-
ence of the outer boundary. Mesh sensitivity stud-
2 NUMERICAL MODEL ies were performed to confirm the accuracy of the
mesh. Calculations accounted for large displace-
The finite element code Plaxis v9 (Brinkgreve ments using the “updated mesh” option in Plaxis,
2008) was used to develop a numerical model of which uses an updated Lagrangian formulation
a reference problem to study installation effects of described by McMeeking & Rice (1975).
stone columns. The installation of only one stone Column installation is modelled as the expan-
column was considered, to simplify the problem sion of a cylindrical cavity, which is considered to
to an axisymmetric two-dimensional geometry. In occur in undrained conditions, because columns
order to consider a realistic situation, properties are usually installed in a short period of time. The
of Bothkennar clay were used for the soft soil. The expansion of the cavity is modelled as a prescribed
Bothkennar soft clay test site has been the subject of displacement from an initial radius, a0, to a final
a number of comprehensive studies (Géotechnique one, af. Values of a0 = 0.1 m and af = 0.41 m rep-
Symposium in print 1992). The soil at Bothkennar resent the installation of a column with a radius
consists of a firm to stiff silty clay crust about 1.0 m of rc = 0.4 m (Carter et al. 1979). After undrained
thick, which is underlain by about 19 m of soft clay.
The ground water level is 1.0 m below the ground
surface. Typically, in a structured soil, the in situ Table 1. S-CLAY1 parameters for Bothkennar clay.
water content is close to the liquid limit.
Depth γ
Stone columns have been applied in Bothkennar (m) (kN/m3) κ ν′ λ M μ β
clay (Watts et al. 2001, Serridge & Sarsby 2008)
or other Carse clays (Egan et al. 2008). For the 0–1 18.0 0.02 0.2 0.48 1.4 30 0.94
numerical model in this paper, a column length of 1–10 16.5 0.02 0.2 0.48 1.4 30 0.94
10 m is used. The untreated clay underneath is not
modelled, because the installation effects in this
part of the soil are not particularly significant and Table 2. S-CLAY1 initial state variables.
furthermore, modelling the tip of the column may
lead to some numerical instabilities. Depth POP
The behaviour of Bothkennar clay was modelled (m) e0 α0 OCR (kPa) K0
using two advanced constitutive models, namely
S-CLAY1 (Wheeler et al. 2003) and S-CLAY1S 0–1 1.1 0.539 – 30 1.35
(Karstunen et al. 2005). The Modified Cam Clay 1–10 2 0.539 1.5 – 0.544
model (MCC) (Roscoe et al. 1958) is also used for com-
parison purposes. S-CLAY1 is a Cam Clay type of
model with an inclined yield surface to model inherent Table 3. S-CLAY1S additional parameters.
anisotropy, and a rotational component of harden-
ing to model the development or erasure of fabric λi χ0 a b
anisotropy during plastic straining. The S-CLAY1S
0.18 5 11 0.2
model accounts, additionally, for interparticle

213
Figure 1. Model geometry and finite element mesh.

expansion of the cavity, the generated excess pore

pressures are dissipated through the permeable col-
umn and the surface. Further details of the numer-
ical model can be found in Castro & Karstunen
(2010). The study here focuses on the results after
full consolidation.

3 STRESS FIELD

The aim is to study how the stress field in the natural

soft soil around the column has changed. Horizon-
tal stresses increase after column installation and,
therefore, the changes in the stress field are usually
considered through an increase in the lateral earth
pressure coefficient, K0. For example, Priebe (1995)
already assumed in his method a value of K0 = 1,
which is higher that the initial value at rest for most
soils. Kirsch (2006) presented field measurements
of K0 and showed that after column installation
values between 1 and 1.7 times the initial one have
been measured, depending on the distance to the
column. Elshazly et al. (2006, 2008) have numeri-
cally back-calculated, from field measurements of
load-displacement curves, values of K0 between 0.7
and 2.5, with average values around 1.3.
Assuming an increase of the K0 value is a use-
ful approach for column design but it is just a sim-
plification of the changes in the stress field during
column installation. Figure 2 shows the results
for the three soil models considered. The vertical,
radial and hoop (circumferential) stresses may be Figure 2. Effective stresses after column installation
still considered as the principal stresses as the shear and full consolidation: (a) MCC, (b) S-CLAY1 and
stresses are negligible. The stress changes are dif- (c) S-CLAY1S.
ferent at different depths but they are directly pro-
portional to the undrained shear strength, cu, and
therefore, also to the depth and the initial vertical column, but for S-CLAY1S vertical stress decreases
stress in this case. because of destructuration. Interestingly, the hoop
As expected, radial stresses increase near the stresses are quite different to radial stresses. Three
column, but vertical and hoop stresses also change. different areas may be distinguished: (1) an elas-
Vertical stresses increase in the part nearest to the tic one (beyond 11.5–13.5 column radii depending

214
on the soil model) where soil behaviour is always
elastic and hoop stresses decrease, (2) an area that
is plastic during undrained expansion of the cav-
ity but is not after consolidation, where vertical
stresses change only slightly and (3) points that
are on the yield surface also after consolidation
(closer than 4.5–6 column radii), where densifica-
tion and the increase in mean effective stresses are
important.
Randolph et al. (1979) presented similar results
for normally consolidated Boston Blue clay using
the MCC model and a similar numerical model
for driven piles. The results from this study with
the MCC model are compared with those results
(Fig. 3). The comparison shows that, although
there are differences in the OCR, the soil param-
eters and the initial stresses used in the current
study, the values near the column/pile are very sim-
ilar and well correlated with the initial undrained
shear strength. The vertical and hoop stresses are
Figure 3. Effective stresses after column installation and
full consolidation. Comparison with a previous study by
Randolph et a. (1979) using MCC model.
215
around 3 times the initial value of the undrained
shear strength and the radial stresses are between
4 and 5. Randolph et al. (1979) predicted a slightly
higher value because their analysis is in perfect
plane strain conditions while in the present analy-
sis there are some vertical strains since the model
realistically considers the soil surface. That also
causes subtle differences in the shape of the curves.
In the far field, the initial stresses of the present
analysis are lower because of the lower value of the
σ'z/cu relationship.
The stress field described in Figure 2 is the basis
for an ongoing study on the influence that the instal-
lation effects have on the settlement reduction. As the
finite element mesh gets very distorted after column
installation, it is advisable to input directly the stress
field as the initial one in a new model. That requires
curve fitting of the stress field and the modification
of the soil model input parameters to allow for this
particular initial stress field. Special care should be
taken to ensure that equilibrium is fulfilled:
∂σ ′ r σ ′ r − σ ′θ
+ =0 (1)
∂r r
neglecting shear stresses to avoid unnecessary
complexity.
In the present model, the installation of only
one column is considered. Obviously, the installa-
tion of several columns will cause some interaction
and overlapping of the stress changes.
4 VOID RATIO AND
OVERCONSOLIDATION
Column installation not only alters the stress field
but also the values for the state parameters of the
soil, such as the void ratio (Fig. 4). The void ratio
decreases just near the column, less than 4.5–6
column radii (Zone 3). However, the densification
is especially important in the area closer than 2
column radii (Zone 3a). Weber et al. (2010) meas-
ured a similar value of 2.5 column radii for the
densification area. They fitted the data points of
porosity and density using a hyperbolic function;
a similar function may be here proposed to fit the
void ratio:
a1
e a3 − (2)
r rc − a2
The coefficient a3 must be equal to the initial
void ratio prior to column installation, in this case
a3 = 2.0. Furthermore, the coefficient a2 must be
lower than 1 because the void ratio cannot be nega-
tive in the soil domain.
Figure 4. Void ratio after column installation and
consolidation.
The results for MCC and S-CLAY1 are quite sim-
ilar but soil destructuration due to column installa-
tion causes a greater densification of the soil.
The value of the void ratio is directly related to
the changes in the mean effective stresses and the
mobilized soil stiffness. The mean effective stresses
and the size of the yield surface, given by p′m, are
plotted in Figure 5, where the distance to the col-
umn axis is in logarithmic scale to amplify the zone
of interest near the column. The different zones
previously distinguished are also visible here: (1) far
from the column, the mean effective stresses does
not change, (2) the mean effective stress increases
but without expanding the yield surface because
the overconsolidation ratio is OCR = 1.5, (3) the
current stress point is on the yield surface and (3a)
the yield surface is notably expanded through strain
hardening, p′m increases. Note that the extension
of these zones is slightly different depending on the
soil model. The current mean effective stress, p′, do
not necessarily coincide with p′m when the point is
on the yield surface because p′m is the mean effec-
tive stress at the apex of the yield surface on the
right (Wheeler et al. 2003). For MCC, the differ-
ences are noticeable but for S-CLAY1 type models,
they are very similar. The relative expansion of the
yield surface is similar for MCC and S-CLAY1,
which explains the similar results for the void ratio.
For S-CLAY1S, the intrinsic yield surface, p′mi, is
significantly expanded in Zone 3a.
For normal column spacings, the soil is in
Zone 3, but the densification or the strain harden-
ing is only important in Zone 3a, i.e. for closely
spaced columns.
The changes in the stress field and the initial
state variables, if anisotropy and destructuration
are not considered, have an influence on the ground
improvement through the following features: (a) the
increase in the radial stress, which has a positive effect
216
Figure 5. Size of the yield surface and mean effective
stresses after column installation and consolidation.
on the improvement as the column is better laterally
confined and (b) the increase in the mean effective
stresses, which may be positive if the yield surface is
expanded and the soil hardens (Zone 3a) or may be
negative if the yield surface is not clearly expanded
as the soil loses its overconsolidation and for a sub-
sequent loading process, there is not an elastic region
anymore and the soil has not hardened either.
5 ANISOTROPY
The advanced soil models S-CLAY1 and
S-CLAY1S can be used to reproduce the changes
in soil fabric due to column installation. The incli-
nation of the yield surface changes, as shown in
Figure 6, where the components of the fabric
tensor, {αi}, are plotted. The two models predict
almost identical changes in anisotropy.
To help to visualize the changes in soil fabric,
Figure 7 shows the (αz − αθ, αr − αθ) vector in arbi-
trary points. This vector changes from horizontal
direction for an initial vertical cross anisotropy
towards a nearly vertical one for radial cross ani-
sotropy. Its length is equal to α0 = 0.539 for the ini-
tial situation and changes only slightly.
The components of the fabric tensor after col-
umn installation and consolidation (Figure 6) must
be input as the initial values to study the influence
that those changes have on the settlement reduc-
tion. That requires curve fitting of those compo-
nents. By definition of the fabric tensor, the fitting
of those components must fulfilled that
α r + αθ + α z = 3 (3)
So, for the sake of simplicity, it seems sensible to
keep αθ constant and decrease αz the same amount
that αr increases.

Figure 6. Components of the fabric tensor after col-
umn installation and consolidation in S-CLAY1.
Figure 7. Visualization of the soil fabric: (a) before col-
umn installation and (b) after column installation and
consolidation.
6 CONCLUSIONS
The paper describes the changes in soil stresses
and initial state variables after stone column instal-
lation. They will be used as the initial ones in an
on-going study of the influence that those installa-
tion effects have on the settlement reduction. The
stresses and the void ratio may be fitted by hyper-
bolic curves that fulfil some conditions, such as
equilibrium.
The stresses near the column are related to
the initial undrained shear strength of the soil.
For normal column spacings, the soil is in a plas-
tic state, which may be negative for a subsequent
loading process because it has lost its overcon-
solidation or positive if the yield surface has been
notably expanded (strain hardening). The latter
217
case occurs for closely spaced columns. In any
case, the increase of the radial stress improves the
lateral confinement of the column. The initial ver-
tical cross anisotropy of the soil changes towards
a radial one when approaching the soil-column
interface.
ACKNOWLEDGEMENTS
The research was carried out as part of a GEO-
INSTALL project by the European Community
through the programme Marie Curie Industry-
Academia Partnerships and Pathways (PIAP-GA-
2009-230638).
REFERENCES
Arnold, M. & Herle, I. 2009. Comparison of vibro-
compaction methods by numerical simulations. Int.
J. Num. Analytical Methods in Geomechanics 33(16):
1823–1838.
Balaam, N.P. & Booker, J.R. 1985. Effect of stone column
yield on settlement of rigid foundations in stabilized
clay. Int. J. Num. Analytical Methods in Geomechanics
9(4): 331–351.
Barksdale, R.D. & Bachus, R.C. 1983. Design and
construction of stone columns. National Technical
Information Service Report FHWA/RD-83/026,
Springfield, Virginia.
Brinkgreve, R.B.J. 2008. Plaxis 2D—Version 9. Plaxis,
Delft.
Carter, J.P., Randolph, M.F. & Wroth, C.P. 1979. Stress
and pore pressure changes in clay during and after the
expansion of a cylindrical cavity. Int. J. Num. Analyti-
cal Methods in Geomechanics 3(4): 305–322.
Castro, J. & Karstunen, M. 2010. Numerical simulations
of stone column installation. Canadian Geotechnical
Journal 47(10): 1127–1138.
Castro, J. & Sagaseta, C. 2009. Consolidation around
stone columns. Influence of column deformation.
International Journal for Numerical and Analytical
Methods in Geomechanics 33(7): 851–877.
Castro, J. & Sagaseta, C. 2012. Pore pressure during stone
column installation. Proc. of ICE—Ground improve-
ment 165(2): 97–109.
Egan, D., Scott, W. & McCabe, B. 2008. Installation
effects of vibro replacement stone columns in soft
clay. In Karstunen and Leoni (eds.) Geotechnics of
Soft Soils—Focus on Ground Improvement, Glasgow,
3–5 September 2008. Taylor and Francis: London,
pp. 23–29.
Elshazly, H.A., Hafez, D. & Mosaad, M. 2006. Back cal-
culating vibro-installation stresses in stone columns
reinforced grounds. Proc. of ICE—Ground improve-
ment 10(2): 47–53.
Elshazly, H.A., Elkasabgy, M. & Elleboudy, A. 2008.
Effect of inter-column spacing on soil stresses due
to vibro-installed stone columns: interesting findings.
Journal of Geotechnical and Geological Engineering
26: 225–236.
Gäb, M., Schweiger, H.F., Thurner, R. & Adam, D. 2007.
Field trial to investigate the performance of a float-
ing stone column foundation. In Proc. of the 14th
European Conf. Soil Mech. Geotech. Eng., Madrid,
24–27 September 2007. Millpress: Amsterdam,
pp. 1311–1316.
Géotechnique Symposium in print. 1992. Bothkennar
soft clay test site: characterisation and lessons learned.
Géotechnique 42(2).
Guetif, Z., Bouassida, M. & Debats, J.M. 2007. Improved
soft clay characteristics due to stone column installa-
tion. Computers and Geotechnics 34(2): 104–111.
Karstunen, M., Krenn, H., Wheeler, S.J., Koskinen, M. &
Zentar, R. 2005. Effect of anisotropy and destructu-
ration on the behaviour of Murro test embankment.
ASCE International Journal of Geomechanics 5(2):
87–97.
Kirsch, F. 2004. Experimentelle und numerische Unter-
suchungen zum Tragverthalten von Rüttelstopfsäu-
lengruppen. Dissertation, Technische Universität
Braunschweig.
Kirsch, F. 2006. Vibro stone column installation and its
effect on ground improvement. In Proc. of Numeri-
cal Modelling of Construction Processes in Geotech-
nical Engineering for Urban Environment, Bochum,
Germany, 23–24 March 2006. Taylor and Francis:
London, pp. 115–124.
Lee, F.H., Juneja, A. & Tan, T.S. 2004. Stress and pore
pressure changes due to sand compaction pile installa-
tion in soft clay. Géotechnique 54(1): 1–16.
Massarsch, K.R. & Fellenius, B.H. 2002. Vibratory com-
paction of coarse-grained soils. Canadian Geotechni-
cal Journal 39(3): 695–709.
McMeeking, R.M. & Rice, J.R. 1975. Finite-element for-
mulations for problems of large elastic-plastic defor-
mation. International Journal of Solids and Structures
11: 606–616.
218
Priebe, H.J. 1995. Design of vibro replacement. Ground
Engineering 28(10): 31–37.
Randolph, M.F., Carter, J.P. & Wroth, C.P. 1979. Driven
piles in clay-the effects of installation and subsequent
consolidation. Géotechnique 29(4): 361–393.
Roscoe, K.H., Schofield, A.N. & Wroth, C.P. 1958. On
the yielding of soils. Géotechnique 8(1): 22–53.
Roy, M., Blanchet, R., Tavenas, F. & La Rochelle, P.
1981. Behaviour of a sensitive clay during pile driving.
Canadian Geotechnical Journal 18(2): 67–85.
Sivasithamparam, N. (2012). Development and imple-
mentation of advanced soft soil models in finite
elements. PhD thesis, University of Strathclyde,
Scotland.
Slocombe, B.C., Bell, A.L. & Baez, J.I. 2000. The densifi-
cation of granular soil using Vibro methods. Géotech-
nique 50(6): 715–726.
Watts, K.S., Chown, R.C. & Serridge, C.J. 2001. Vibro
stone columns in soft clay: A trial to study the influ-
ence of column installation on foundation perform-
ance. In Proc. of the 15th Int. Conf. Soil Mech. Geotech.
Eng., Istanbul, 28–31 August 2001. Taylor and Francis:
London, pp. 1867–1870.
Watts, K.S., Johnson, D., Wood, L.A. & Saadi, A. 2000.
An instrumented trial of vibro ground treatment sup-
porting strip foundations in a variable fill. Géotech-
nique 50(6): 699–708.
Weber, T.M., Plötze, M., Laue, J., Peschke, G. &
Springman, S.M. 2010. Smear zone identification and
soil properties around stone columns constructed in-
flight in centrifuge model tests. Géotechnique 60(3):
197–206.
Wheeler, S.J., Näätänen, A., Karstunen, K. & Lojander, M.
2003. An anisotropic elastoplastic model for soft clays.
Canadian Geotechnical Journal 40(2): 403–418.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Execution of Springsol® deep mixed columns: Field trials

S. Melentijevic, F. Martin & L. Prieto

Grupo Rodio-Kronsa S.L.U., Madrid, Spain

ABSTRACT: The reparation and underpinning of existing structures and infrastructure, due to differ-
ent post-constructive pathologies, often needs the ground improvement of man-made fills and soft soils
in general. The increasing growth of deep soil mixing methods has evolved into the development of the
special Springsol® tool, which permits the installation of deep mixed soil-cement columns under existing
superstructures by the application of new procedures linked with a controlled opening system. Based on
the results of different recently executed field trial tests and projects, this paper describes some advantages
of the Springsol® tool used for the construction of deep mixed columns. The execution parameters and
methods of quality control during and after column installation are described and analyzed, taking into ac-
count installation effects and its influence on the geo-mechanical characteristics of the improved soil. Full
scale test results are presented and compared with results obtained by laboratory tests on core samples.

1 INTRODUCTION investigation. These field tests were conducted as

a base for an R&D project in progress dealing with
The Springsol® tool was initially developed as a new numerical modeling of the process of deep soil-
wet deep mixing technique for improvement of the mixing by the Springsol® tool.
ground under existing railway infrastructure, with-
out interference of the binder with existing ballast
and sub-ballast layers and without removing exist- 2 EXECUTION OF SPRINGSOL® COLUMNS
ing rail-tracks (Le Kouby et al., 2010).
This technique can also be applied in any under- The typical process of construction of the
pinning works for existing structures due to the Springsol® column is presented in Figure 1 and
spreadable form of the Springsol® tool, which rep- assumes:
resents the main advantage when comparing it to
conventional single axis deep mixing techniques. – Positioning of the drilling machine, drilling of
It permits the installation of the column through the existing superstructure, placing the steel
existing structures (foundations, both slabs and casing to the required depth of the head of the
footings), roads and railway embankments, by intro- column.
duction of casings in existing small diameter bore- – Introducing the mixing tool in the folded
holes of 165 mm. After the passage of the casing, form through the casing to the end of the steel
the tool is expanded to a larger diameter that allows casing.
the formation of columns up to 700 mm. This tech- – Opening the tool to a larger diameter once out-
nology requires the installation by small basic drill- side the casing and introduction of the binder in
ing machines that provide its suitability for works in slurry form through nozzles located at the tip of
reduced spaces and under low headroom conditions. the tool to the required depth of the column.
It also requires small batching plants that provide
necessary flow rates and pressures.
As with all other deep mixed soil-cement col-
umns, the Springsol® column has a higher strength,
lower compressibility and lower permeability than
the original ground. Its construction reduces set-
tlements, increases bearing capacity and slope sta-
bility, and remediates liquefaction potential, etc.
In this paper, some of experiences during field
trials of the execution of Springsol® columns are
presented, with the emphasis on geotechnical
properties obtained after detailed laboratory Figure 1. Construction process of the Springsol® column.

219

Figure 2. Original and modified Springsol® tool.
Figure 3. Diameter variation of the Springsol® column.
– The Springsol® tool is pulled up after reach-
ing the final depth of the soil-cement column.
The mixing during withdrawal is continued in
order to homogenize the soil-cement mixture.
The tool is closed once the casing is reached in
order to be pulled out.
The injection of the slurry can be done during
the penetration or during the withdrawal of the
tool, or by combining both phases of the construc-
tion of the column.
The original Springsol® tool was developed for
enlarged diameters of 400 mm, with the retracted
tool of only 150 mm that significantly reduces
interference with existing foundations. The con-
tinuous development of the Springsol® tool due
to recent requirements for the application of soil-
cement columns of larger diameters, evolved in
modifying the tool for opened diameters up to
700 mm (see Fig. 2). The automatic system for
opening and closing the arms of the tool in order
to change the diameter of the column along the
depth has also been introduced. Nowadays the
Springsol® tool involves the successful execution
of columns with diameters ranging from 400 to
700 mm. Figure 3 presents an example of the exe-
cuted column varying in diameter along the length
of the column, obtained during one of the field
trial tests described in the following sections.
220
Part of the column with the maximum diam-
eter of 600 mm and part of the column with the
reduced diameter of approximately 300 mm can
be observed. It is important to emphasize that the
variation was executed in a programmed and con-
trolled way. The installation of both vertical and
inclined columns is possible.
3 QUALITY AND CONTROL PARAMETERS
DURING EXECUTION
The quality of the soil-cement column, regarding
its homogeneity and strength, is influenced by the
mixing efficiency defined as follows:
⎛ω ω ⎞
Im = N ⎜ b + s ⎟ (1)
⎝ U b Us ⎠
where Im = blade rotation number is the total
number of mixing blades passing along 1.0 m of
tool penetration used as an indirect measure of the
degree and efficiency of mixing (rev/m); N = total
number of mixing blades where a full diameter
blade is counted as two blades; ωb = rotational
speed of the mixing tool during penetration (rpm);
ωs = rotational speed of the mixing tool dur-
ing withdrawal (rpm); Ub = penetration velocity
(m/min); Us = withdrawal velocity (m/min). The
recommendation of a minimum required value of
360 rev/m is established in Japan for the wet deep
mixing method for ensuring a low coefficient of
variation of the unconfined compressive strength
(CDIT, 2002).
Another important parameter for the quality and
geomechanical characteristics of the column is:
Wc C co Q
Ii = = (2)
(π ⋅ Φ 4) ⋅ U
Vs 2
p
where Ii = cement quantity introduced per m3 of
the treated soil (kg/m3); Wc = the cement consump-
tion quantity per 1.0 m of the column (kg/m);
Vs = volume of the column in 1.0 m (m3/m);
Cco = kg of the cement per m3 of the slurry (kg/m3);
Q = flow grout of the slurry (l/m); Φ = diameter
of the column (m) and Up = penetration velocity
(m/min).
During construction of Springsol® columns
there are 3 important systems or levels that permit
control during execution:
– The system for automatic flow grout injection
as a function of the penetration velocity (see
Fig. 4). By normal rate of advancement the
pump is functioning according to the established
flow rate, with it being possible to control it

Figure 4. Pumping and automatic flow rate control
equipment.
Figure 5. Spoil collection system, peristaltic bomb,
spoil container.
automatically in order to meet the increasing
or decreasing resistance of soil layers that could
affect the penetration velocity.
– The monitoring by the continuous registering of
the execution working parameters in real time
(depth (m), penetration velocity (m/h), torque
pressure (bar), rotational speed (rpm), flow rate
(L/min) and grouted volume (L/m)).
– The system for the management of spoil (see
Fig. 5). The construction of Springsol® columns
is characterized by the “clean” environment after
its execution due to the system of spoil collection
installed at the base of the mast of the drilling
rig, which is connected to the peristaltic pump
that is drawing the spoil directly to the container.
Apart from the prevention of contamination of
the working platform with spoil, it allows indirect
control of homogeneity and geomechanical char-
acteristics of the treated or improved soil.
4 COLUMN GEOMATERIAL
CHARACTERIZATION
Some of the field trials that were carried out
in Spain are presented in this section, both for
221
predominantly cohesive and granular types of
soils. The cement used for the construction of
these columns in both field trial tests presented
was the Portland type CEM II/B-M 32.5 R. More
details on experiences based on various projects
performed in Spain with the Springsol® tool appli-
cation are given in Melentijevic et al. (2012).
Considering that the soil-cement columns are
stiffer than the surrounding soil, the stress is con-
centrated on them so that the main factor for the
design is the compressive strength. Unconfined
Compressive Tests (UCS), being the most com-
monly used test for cement treated soil, were con-
ducted both on core samples and full scale tests,
observing the typical crushing failure for these
types of tests. Core samples (cylinders of 2″ diam-
eter with the relation diameter/height = 0.5) were
obtained in the laboratory by the coring of blocks
that were cut from excavated columns at the field
trial test site. The relationship between axial strain
at failure (εf) and UCS is shown in Figure 14, mostly
covering typical low values for cement treated col-
umns of εf < 1%. The stress-strain curves of the
cement stabilized soil are generally characterized by
high strength and low values of axial strain at fail-
ure, while the native soil is defined by small strength
and large strain at failure. The modulus of elasticity
(E50) is an important parameter for design and is
mostly determined by the relationship with uncon-
fined compressive strength. These correlations
obtained during the tests are given in Figure 13.
The compressive strength is taken at peak value and
the value of E50 as a secant stiffness modulus at a
stress equal to 50% of the failure stress.
Triaxial compression tests (CU type—consoli-
dated undrained) have also been carried out on core
samples, presenting clear shear bands developed
during failure. In triaxial compression tests, with
the increase of the confining pressure the stress-
strain relationships change from strain softening
to strain hardening. The typical stress-strain corre-
lations of the soil-cement columns are comparable
to ones observed in the cases of overconsolidated
clays or dense sands, considering the increase up
to the peak value and then decreasing to residual
value at the post peak state (see Figure 15).
4.1 Field trial test 1
Field trial test 1 comprises 12 test columns of 5.0 m
length and 400 to 530 mm diameter, performed in
sandy clay. The Springsol® tool applied for the test
is shown on the left side of Figure 2 and represents
its basic form or mechanism. The consumption of
cement varied from 250 to 350 kg/m3 of the treated
soil, while the average blade rotation number was
650, ranging from 350 to 1000, basically due to the
penetration velocity that was influenced by the soil
density/consistency. The lower value of the blade
rotation number corresponds to the upper part
of the column and in general to a softer original
soil layer. The rotational speed of the blades dur-
ing penetration was maintained at 50 rpm, while
the mixing penetration blade velocity ranged from
10 to 25 cm/min. The soil-cement columns were
exposed after 28 days of their execution, allowing
observation of the column shape, diameter, homo-
geneity, etc. (see Fig. 6).
A few columns were completely extracted for
full scale tests. The extracted columns were also
cut in blocks in order to obtain core samples for
laboratory testing by unconfined compression
tests and triaxial tests. The full scale test and the
failure cracks of the column after the test can be
observed in Figure 7.
Considering that core samples are most prob-
ably the principal source for verification of the
properties of cement treated soils, and taking into
account that coring is an invasive method that pro-
vokes micro fractures influencing the quality of
results, less affected or better core samples were
selected with the objective to provide more realistic
results. The aspect of some core samples of field
Figure 6. Excavated Springsol® column (Diam. = 400 mm).
Figure 7. Failure of the column (Diam. = 400 mm)
before and after UCS full scale tests.
222
test 1 for the laboratory testing (UCS and triaxial
CU tests) is presented in Figure 8.
4.2 Field trial test 2
A total of 5 trial test columns were performed,
with a minimum length of 3.0 m and 600 mm
diameter. The ground was defined as silts (ML)
and silty well graded sands (SW-SM), whose sieve
analysis chart is presented in Figure 9. The average
blade rotation number varied from 250 to 750, the
lower value being representative to the first meter
of the column corresponding to soft materials and
higher penetration rate. The average consump-
tion of cement was 300 kg/m3. Trial columns were
executed with the new modified and improved tool
shown in the right photo of Figure 2. The rota-
tional speed of the blades during penetration var-
ied from 55 to 110 rpm in accordance with the soil
density/consistency. The mixing penetration veloc-
ity ranged from 10 cm/min in a stiffer sandy layer
to 45 cm/min in silts and automatic flow grout
injection control was applied.
The soil-cement columns were fully exposed after
28 days of construction to be verified geometrically
and geotechnically. Some of the columns were
extracted to be tested in the laboratory at full scale
in order to determine the unconfined compressive
Figure 8. Core samples of field trial test 1.
Figure 9. Sieve analysis chart.
strength and its corresponding elastic modulus 4.3 Results
(E50) after 28 days. The general appearance of the
The correlation of E50 and UCS varies between
excavated columns is given in Figure 10. This shows
50 and 1000, as reported by several published
the column with the uniform diameter of 600 mm
investigations (Bruce, 2001; CDIT, 2002; etc.) for
and the one with the controlled diameter variation
different soils, basically depending on the natural
achieved by the application of the system for open-
soil grain size. This relationship for the field trial
ing and closing blades.
tests is given in Figure 13, varying from 50 to 420
The UCS tests on columns of 600 mm diam-
for core samples and 90 to 380 for full scale tests.
eter were performed at full scale (with the rela-
The summary of values of failure axial strain for
tion of height to diameter greater than 2). The
both field trial tests is presented in Figure 14, with
cracks observed after failure of the columns are
the majority of strains at failure being low values
presented in Figure 11. The UCS and triaxial tests
(εf < 1%).
were also conducted on cored samples of the trial
The results of triaxial tests (CU) for the core
test columns. The appearance of some of the core
samples of field trial test 2 are shown in Figure 15.
samples before and after the failure during tests is
The undrained residual strength parameters are
presented in Figure 12.
5 to 40% lower than the peak values.
The correlation between the blade mixing
number and the UCS is presented in Figure 16,
with the proposed relationship taking into account
both field trial tests.
The quantity of cement in some of the con-
structed Springsol® columns during field trial test 2

Figure 10. Excavated Springsol® column (Diam. =

600 mm).

Figure 13. Correlation between E50 and UCS for both

field trial tests on Springsol® columns.

Figure 11. UCS full scale tests on Springsol® column

(Diam. = 600 mm).

Figure 12. The aspect of core samples before and after Figure 14. Correlation between axial failure strain and
UCS tests for field trial test 2. UCS for core samples and full scale tests.

223
 Table 1. Parameters of Springsol® columns.

Diameter 400–700 mm
W/C 0.6–1.2
Perforation velocity 15–50 cm/min
Im min 350 rev/m
Ii 150–350 kg/m3
UCS 0.5–6.0 MPa
E50 (50–500) UCS
Shear strength 20–40% UCS
Bending strength 8–15% UCS

was verified by the technique of X-ray fluorescence

Figure 15. Triaxial test results on core samples
(Column diam. = 400 mm).
(UNE-EN 15309:2007), resulting in a total amount
of cement of 25 ± 4% (Rodriguez et al, 2012).
These results are in very good agreement with the
average amount of cement of 300 kg/m3 used for
the construction of columns, thus confirming the
high efficiency of the automatic grout injection
and mixing system.
Figure 17 shows the appearance of cross sec-
tions of extracted Springsol® columns. A certain
difference in the homogeneity along the cross sec-
tion is observed, with the higher cement content in
the center of the column.

5 CONCLUSIONS

The application of the Springsol® tool for the

construction of soil-cement columns is specially
focused for the underpinning of existing structures
(slabs and footings), roads and railway embank-
ments due to its geometrical characteristics. The
technique makes use of existing soil, thus reducing
the needed amount of cement. In general there is
Figure 16. Correlation between UCS and Im. no limit regarding depth of columns, but until now
the maximum applied depth in different projects
and field trial tests is up to 15 m.
The retractable mixing blades can be rotary
drilled to the desired depth of the column’s head,
allowing the treatment to a larger diameter down to
specific depths without interference of the binder
with the subgrade layers and reducing the effect on
existing foundations.
In practice the unconfined compressive strength
is the mostly used geotechnical parameter for the
evaluation of the deep soil mixing column resist-
ance or capacity. Unconfined compressive strength
tests are widely used in practice as other important
design parameters can be estimated using this
parameter. Experimental relationships obtained in
the described field test are given in Table 1.
The results of the field tests are in agreement
with worldwide published data for deep mixed col-
Figure 17. The cross section of extracted columns. umns (Bruce, 2001; CDIT, 2002; etc.).

224
 The relationship between axial failure strain and
UCS obtained in presented tests covering typical
low values of εf < 1% for cement treated columns.
The stress-strain curve is characterized by high
strength and low values of axial strain at failure.
The results of UCS tests on core samples, influ-
enced by the disturbance produced during their
coring, are lower than values of UCS obtained on
full scale tests. It confirms that, for more repre-
sentative results, the dimension of the test sample
is of major importance. Actually more tests are
being performed in order to verify the influence of
test sample dimension, which can be of additional
importance in heterogeneous soils, very character-
istic for the Iberian Peninsula.
ACKNOWLEDGEMENTS
The authors are grateful to the personnel of
Rodio-Kronsa that participated in the execution
of the Springsol® columns, with special assist-
ance from Juan Manuel Dimas, Esteban Casado
and José Luis Arcos. The research program is
partially financially supported by CDTI con Ref:
IDI-20110066.
REFERENCES
Bruce D.A. 2001. An introduction to deep mixing meth-
ods as used in geotechnical applications, Volume III:
The verification and properties of treated ground.
U.S. Department of Transportation, Federal Highway
Administration, report FHWA RD-99-167.
225
CDIT (Coastal Development Institute of Technol-
ogy), Japan. 2002. The Deep Mixing Method,
A.A. Balkema.
Laboratorio de Geotecnia. Cedex. 2011. Informe de lab-
oratorio para Grupo Rodio-Kronsa. Columnas suelo-
cemento para el proyecto de investigación de nuevas
técnicas y herramientas de soil-mixing.
Le Kouby, A., Bourgeois, E. & Rocher-Lacoste, F. 2010.
Subgrade improvement method for existing railway
lines—an experimental and numerical study. EJGE
Vol. 15: 461–494.
Melentijevic, S., Prieto, L. & Arcos, J.L. 2012. Aplica-
ciones de columnas suelo-cemento tipo Springsol®. 9º
Simposio Nacional de Ingeniería Geotécnica. Ciment-
aciones y Excavaciones Profundas. Proc. Symp.,
Sevilla, 17–19 October 2012: 175–189.
Rodriguez Abad, R. & Estaire Gepp, J. 2012. Determi-
nación mediante WD-XRF del contenido de cemento
en suelos inyectados y en mezclas de suelo-cemento. 9º
Simposio Nacional de Ingeniería Geotécnica. Cimenta-
ciones y Excavaciones Profundas. Proc. Symp., Sevilla,
17–19 October 2012: 255–268.
UNE-EN 15309:2007. Characterization of waste and
soil. Determination of elemental composition by
X-ray fluorescence.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

A method of modelling stone column installation for use in conjunction

with unit cell analyses

B.G. Sexton & B.A. McCabe

College of Engineering and Informatics, National University of Ireland, Galway, Ireland

ABSTRACT: Traditionally, the majority of numerical studies investigating stone column behaviour
have studied the problem under unit cell (axisymmetric) conditions, in which the granular columns tend
to be ‘wished-in-place’ (no installation effects). In this study, cylindrical cavity expansion is used to work
out post-installation lateral earth pressure coefficients (and hence a post-installation stress-regime in the
ground) arising due to the lateral expansion and subsequent soil remoulding caused by the vibrating
poker as columns are installed in a soft clay. Two sets of two-dimensional axisymmetric finite element
analyses have then been carried out using PLAXIS 2D to examine load-settlement behaviour, the first set
assuming the coefficient of earth pressure to be unaffected by column installation, while the second set
have been conducted using increased earth pressure coefficients based on the cavity expansion procedure.
The Hardening Soil Model (no viscous effects) has been used to model the behaviour of the granular
column material and the soft clay. Settlement improvement factors calculated using both approaches have
been compared to establish the effect of column installation. Predicted improvement factors have been
put into context by comparison with existing analytical settlement design approaches. The results indicate
that this approach can be used as a realistic means of accounting for column installation in conjunction
with unit cell analyses, with larger improvement factors predicted when installation (increased K) is taken
into account.

1 INTRODUCTION and subsequent compaction of the columns,

i.e. K values remain at the at-rest (K0) values.
Settlement considerations tend to govern the However, field measurements by Kirsch (2006)
design of stone columns in soft or marginal soils, have indicated K/K0 values in excess of unity post-
and the extent of settlement reduction achiev- construction. Some authors have aimed to cap-
able is quantified by a dimensionless ‘settlement ture this effect with global increased K values; e.g.
improvement factor’ (n = suntreated/streated), where Gäb et al. (2008), based on approaches adopted
suntreated and streated are the settlements of untreated by Priebe (1976, 1995) and Goughnour & Bayuk
ground and the ground treated with stone col- (1979) have adopted K = 1, while Domingues et al.
umns respectively. Most analytical formulations (2007) have used K = 0.7 (between the conserva-
are based on a unit cell approach, which repre- tive K0 = 1 − sin ϕ′ for normally consolidated soils
sents the behaviour of a single column within an and K = 1 approaches).
infinite grid of regularly spaced columns (Fig. 1).
The area replacement ratio, Ac/A, is a measure of
the amount of in-situ soil replaced with granular
column material and is a function of the column
spacing (s), the column diameter (D = 2Rc, where
Rc is the column radius) and a parameter k cap-
turing the arrangement. The value of Ac/A has a
major bearing on the value of n.
Several authors have analysed stone columns
using 2D (e.g. Ambily and Gandhi 2007) and 3D
(e.g. Killeen and McCabe 2010) finite element
analyses. Some of the analyses have involved
‘wished-in-place’ columns, in which the coeffi-
cient of lateral earth pressure (K) is assumed to
be unaffected by the vibratory action of the poker Figure 1. Typical column grids encountered in practice.

226
 In this paper, PLAXIS 2D (Brinkgreve et al.
2010) is used to provide an indication of the possi-
ble effect of column construction on K values sur-
rounding the columns, and in turn, the influence of
the new K values on the settlement improvement
factors for an infinite grid. Benchmark (Case A)
unit cell analyses were performed on columns that
were wished-in-place. Case B unit cell analyses
involved a modified initial K profile derived from
prior Cylindrical Cavity Expansion (CCE) analy-
ses of single columns. The approach used in Case
B is novel, as the prescribed displacements in the
CCE stage would cause both excessive heave and
numerical problems (due to boundary proximity)
for the size of unit cell required to give typical Ac/A
ratios.
The improvement factors derived have been put
in the context of those derived from existing ana-
lytical settlement design methods based on the unit
cell approach (i.e. Priebe 1995, Castro & Sagaseta
2009, Pulko et al. 2011).
2 CCE BACKGROUND
Although not entirely representative of the vibra-
tory action of the poker and the progressive com-
paction of the columns from the base upwards,
CCE nevertheless provides a convenient means
of simulating the lateral expansion of a granular
column into the surrounding soil. In practical situ-
ations, column installation involves the ‘creation’
of a cavity in the soil, i.e. modelling column instal-
lation using CCE should thus involve expand-
ing a cavity from a zero initial radius to the final
radius of the column. However, numerical simu-
lations must begin with a finite cavity radius to
avoid the development of infinite circumferential
strain. Consequently, it must be ensured that the
internal cavity pressure reaches the limit pressure
(plim), as defined by Gibson & Anderson (1961)
in Equation 1 (p0 is the original in-situ horizontal
total stress, cu is the undrained shear strength of
the soil, E is the Young’s Modulus of the soil and ν
is its Poisson’s ratio).
⎧ ⎡ E ⎤⎫
plim = p0 + cu ⎨1+ ⎢ ⎥⎬ (1)
⎩ ⎣ u 2c
c ( ) ⎦⎭
Carter et al. (1979) have described the use of the
cavity expansion technique in two different types
of elastoplastic soil, and report that doubling the
cavity size is sufficient in most cases (the internal
cavity pressure will be within 6% of the ultimate
limit pressure; any further expansion beyond this
will only cause further growth of the annular
region of yielded soil). Since the cavity expansion
227
process must start from a finite radius, the final
cavity radius, af, should be obtained by rearranging
Equation 2 (i.e. observing volume conservation),
where a0 is the initial cavity radius.
af 2 a 2 = Rc 2 (2)
Elshazly et al. (2006) have backfigured the post-
installation coefficient of lateral earth pressure, K,
by matching measured load-settlement behaviour to
finite element analyses (by changing K values) car-
ried out using PLAXIS and found that K/K0 may
range from 1.1 to 2.5, with best estimates of approx-
imately 1.5. Guetif et al. (2007) have used CCE to
evaluate the improvement to the Young’s Modulus
of a soft clay as a result of column installation.
Castro & Karstunen (2010) have applied the cavity
expansion technique to the undrained installation
of a single stone column. The authors confirm that
column installation generates excess pore pressures
and increases the horizontal stresses in the soil.
3 NUMERICAL MODELLING
The numerical modelling carried out consists of
two distinct parts. The first part involves the use
of CCE to establish post-installation lateral earth
pressure coefficients generated by column installa-
tion in a soft clay soil. The second part involves
carrying out two sets of wished-in-place unit cell
analyses under an applied load for (i) a soft clay
assuming the clay properties to be unaffected by
column installation (Case A) and (ii) a soft clay
with increased lateral earth pressure coefficients
calculated using the CCE analyses (Case B).
Essentially, the analyses carried out in this paper
are intended as a progression of some of the afore-
mentioned work by Guetif et al. (2007), Castro &
Karstunen (2010), etc., albeit with a simpler soil
model in the latter case.
3.1 Finite Element (FE) model parameters
The Hardening Soil (HS) Model has been used to
model the behaviour of both the granular column
material and the soft clay soil. The HS model is a
hyperbolic elastoplastic model that takes account
of the stress dependency of stiffness moduli but
does not account for viscous effects such as creep
or stress relaxation. The formulation of the model
has been described by Schanz et al. (1999). The
clay has been modelled as an undrained material
(PLAXIS Undrained A approach), with param-
eters defined in terms of effective strength and
effective stiffness. The stone has been modelled as
a highly permeable drained material (permeability,
k = 1.7 m/day).
 The soft clay properties (Table 1) are representa- boundary is located 30 m away from the axis of
tive of a simplified single layer profile (no crust) symmetry (to ensure that results are unaffected
loosely based on parameters for the Bothkennar by boundary proximity). For the unit cell radii
soft clay test site (e.g. Nash et al. 1992). A reference required for typical Ac/A ratios encountered in
cohesion of 1 kN/m2 has been used for numerical practice, the boundary would have been too close
stability. The power, m, dictating the stress depend- (leading to numerical problems). Roller bounda-
ency of stiffness moduli with depth has been set ries have been applied to all sides. The finite ele-
equal to 1 in order to simulate logarithmic com- ment mesh (Fig. 2), consisting of approximately
pression behaviour (e.g. Brinkgreve et al. 2010). The 5000 6-node triangular elements, has been refined
soil stiffnesses have been defined so that the ratio in the region surrounding the cavity (largest strains
of the oedometric/constrained moduli between the in this region).
granular material and the clay (commonly referred Simulating column installation in PLAXIS
to as the modular ratio Ec/Es) is approximately 20. 2D using a cavity expansion technique will
The ratio of λ (=Cc/ln10) to κ (=Cs/ln10) has been involve three phases after initial stress generation
fixed at approximately 7 (representative of Both- (initial stresses have been generated using the K0
kennar clay, e.g. Allman & Atkinson 1992) where procedure):
Cc and Cs are the compression and swelling indices
a. Install ‘dummy material’ (e.g. Guetif et al.
respectively.
(2007) suggest modelling the cylindrical hole
The parameters for the granular column mate-
created by the poker using a ‘dummy material’
rial have been chosen based on Killeen & McCabe
with a weak Young’s modulus) over the entire
(2010). The friction angle, ϕ′, has been set equal to
column length to a radial extent, a0.
45o (representative of bottom feed columns). The
b. Apply a prescribed displacement (undrained
dilatancy angle has been calculated as ψ = ϕ′ − 30o.
conditions) from the initial radius, a0, to a final
The oedometric modulus, Eoedref, was assumed equal
radius, af (Fig. 2). Three different a0 values of
to the secant modulus, E50ref and the unload-reload
0.10 m, 0.15 m and 0.20 m have been used for
modulus, Eurref, was calculated as Eurref = 3E50ref (the
which af values have been calculated accord-
default setting in PLAXIS). A complete listing of
ing to Equation 2 as af = 0.316 m, 0.335 m and
model parameters is given in Table 1.
0.361 m respectively.
c. This cavity expansion stage is followed by a
3.2 Installation (CCE) stage for Case B consolidation phase to allow excess pore pres-
sures to dissipate to establish the long-term
A column diameter (D) of 0.6 m (column radius,
stress changes in the ground caused by column
Rc = 0.3 m) has been adopted for this study, which
installation.
is typical of columns at soft cohesive soil sites, e.g.
Watts et al. (2000). The unit cell model is 5 m long This approach is described in detail by Castro &
(i.e. column length, L = 5 m) and the external far Karstunen (2010). The application of a prescribed
displacement in this manner was deemed to be the
best option for numerical stability purposes (Kirsch
Table 1. FE model parameters. 2006, Castro and Karstunen 2010), as opposed to
the application of a volume strain to an expand-
Clay (drained) Stone backfill (drained) ing soil cluster. Based on the increased horizontal
stresses (after the consolidation phase), it will be
γunsat (kN/m3) 16.5 19.0
possible to work out post-installation lateral earth
γsat (kN/m3) 16.5 19.0
pressure coefficients, K, as detailed in Equation 3,
kx (m/day) 1 × 10−4 1.7
ky (m/day) 6.9 × 10−5 1.7
einit 2.0 0.5
φ′ (o) 34 45
ψ (o) 0 15
cref (kPa) 1.0 1.0
K0nc 0.441 0.296
Eoedref (kPa) 3500 70000
E50ref (kPa) 4350 70000
Eurref (kPa) 21874 210000
m (power) 1.0 0.3
pref (kPa) 100 100
ν′ur 0.2 0.2
K0 0.441 0.296
OCR 1.5 –
Figure 2. Finite element mesh for CCE stage.

228
where σ′xx, σ′yy and σ′zz are the effective radial, ver-
tical, and tangential (hoop) stresses respectively.
⎛ σ ′ + σ ′ zz ⎞
K = ⎜ xx ⎟ (3)
⎝ 2.σ ′ yy ⎠
The variation of K/K0 with r/Rc (r is the radial
distance from the axis of symmetry) calculated by
the HS model is shown in Figures 3a and 3b. K/K0
is independent of the amount of lateral expan-
sion and it has also been verified that the K/K0
values are independent of depth (K0 was set equal
to 1 − sin ϕ′ = 0.441 for the normally consolidated
clay considered in this study).
The K/K0 values calculated using the HS model
have been compared to field values of K/K0
reported by Kirsch (2006) mostly beyond r = 5Rc
and numerical predictions of K/K0 presented by
Castro & Karstunen (2010) in Figures 3a and 3b
respectively. The first set of field data (‘Field 1’)
pertains to 9 m long 0.8 m diameter stone columns
installed in a silt layer (K0 = 0.91) while the sec-
ond set of data (‘Field 2’) was obtained for 6 m
long columns installed in a sandy silt (K0 = 0.57);
the latter set are more comparable to the situation
Figure 3. (a) K/K0 versus r/Rc (comparison with field
data). (b). K/K0 versus r/Rc (comparison with numerical
data).
229
modelled. Field values of K/K0 are slightly greater
than HS model values for r/Rc > 8, but the agree-
ment, for the most part, is relatively good for the
‘Field 2’ data. Field values are below those pre-
dicted by the HS model for r/Rc < 8 (Castro &
Karstunen (2010) have suggested that this ‘dip’
close to the column may be due to remoulding and
dynamic effects).
The numerical predictions by Castro &
Karstunen (2010) have been obtained using two
‘user-defined’ anisotropic (the HS model is an iso-
tropic soil model) constitutive soil models imple-
mented in PLAXIS 2D (L = 10 m, D = 0.8 m). The
S-CLAY1 model, based on the Modified Cam Clay
(MCC) model, employs an inclined yield surface
and a rotational component of hardening to model
anisotropy of plastic behaviour. The S-CLAY1S
model is an extension of the S-CLAY1 model
that also takes account of bonding and destructu-
ration (in addition to anisotropy). The computed
values of K/K0 are lower for S-CLAY1S than for
S-CLAY1 owing to the destructuration caused by
column installation (Castro & Karstunen, 2010).
The K/K0 values obtained using the S-CLAY1
and HS models are in excellent agreement for r/
Rc < 5 with deviation at higher r/Rc values. K/K0
values obtained using S-CLAY1S appear to be in
good agreement with the HS model for r/Rc > 7.
However, it is clear that the model predictions will
not (and should not) be in perfect agreement owing
to different assumptions inherent in their develop-
ments (in addition, the different geometries and
soil types employed in the different studies will
lead to different predictions). The purpose of the
comparisons carried out was simply to indicate
that the HS model K/K0 values are in reasonable
agreement with previous studies.
The K/K0 decay with r/Rc is approximated by a
stepwise reduction as shown in Figure 4; average
K/K0 values have been interpreted at intervals of
0.10 m representing concentric rings around the
axis of symmetry. Extrapolation has been used
to estimate K/K0 values for r/Rc < 1.15 (numeri-
cal discontinuities in this region lead to difficul-
ties in establishing accurate values, e.g. Carter
et al. (1979)). These values are used as input
parameters for the unit cell (Case B) models.
Also demarcated on Figure 4 is the radial extent
(r/Rc = 3.16) relevant to the largest unit cell mod-
elled (A/Ac = 10).
3.3 Unit cell settlement behaviour
The axisymmetric unit cell model consists of
a 0.6 m diameter (Rc = 0.3 m) column 5 m in
length (Fig. 5). The modelling approach adopted
in this paper is similar to approaches adopted by
Debats et al. (2003) and Ambily & Gandhi (2007).
Figure 4. Adopted K/K0 profile.
Figure 5. Axisymmetric unit cell model.
The base is fixed in all directions while the vertical
boundaries are fixed only in the radial direction.
The unit cell analyses are performed for a range of
reciprocal area-replacement ratios (A/Ac) between
2 and 10 (typical values used in practice based on
a database of n values compiled by McCabe et al.
(2009)). The desired A/Ac ratio has been attained
in each case by altering the unit cell diameter (i.e.
column spacing), with Rc fixed at 0.3 m.
The stages in each analysis are as follows:
a. Initial stress generation using the K0 procedure.
b. Install (wish-in-place) stone column.
c. Place a plate (loading platform to prevent dif-
ferential settlement - arbitrary properties) over
the soil/column surface.
d. Apply a 100 kPa load in undrained conditions.
e. Consolidation phase to allow the excess pore
pressures to dissipate.
Cases A and B follow the same steps for stages (b)
to (e). However, initial stress generation is different
for each case. Initial stresses in the clay for Case A
are generated using a uniform K value (e.g. mean
effective stresses (p’) are plotted in Fig. 6a) while
initial stresses for Case B are generated using the
profile generated from the CCE analyses in sec-
tion 3.2 (Fig. 6b).
230
Figure 6. (a) p’ with depth for Case A and A/Ac = 10.
(b) p’ with depth for Case B and A/Ac = 10.
4 RESULTS
4.1 Case A—Comparison with analytical
solutions
Settlement improvement factors for Case A are
compared with a selection of analytical settlement
design methods based on the unit cell approach (i.e.
Priebe 1995, Castro & Sagaseta 2009, Pulko et al.
2011) in Figure 7. These methods are based on
elastic-plastic theory (Mohr-Coulomb failure crite-
rion) assuming the soil to remain elastic as the col-
umn yields. Priebe (1995) assumes constant volume
deformation during yield while Castro & Sagaseta
(2009) and Pulko et al. (2011) have accounted
for dilation of the granular column material as
it yields. In spite of its empirical nature, Priebe’s
(1995) n2 (correction to Priebe’s basic improvement
factor, n0, accounting for column compressibility
and the bulk densities of the soil and column mate-
rials) tends to be widely used in European geotech-
nical practice. The methods derived by Castro &
Sagaseta (2009) and Pulko et al. (2011) are analyti-
cal solutions based on Balaam & Booker’s (1981)
elastic solution. Castro & Sagaseta (2009) have con-
sidered an undrained loading situation followed by
a consolidation process whereas Pulko et al. (2011)
have considered a drained loading situation. As is
evident from Figure 7, the agreement between the
PLAXIS-predicted improvement factors and the
improvement factors predicted by the analytical
methods is excellent. This gives confidence in the
unit cell modelling procedure used in this study.
4.2 Case A versus Case B
Settlement improvement factors generated for
Cases A and B are compared in Figure 8. As
can be seen, accounting for column installation

Figure 7. n versus A/Ac for Case A (K = 1 − sin ϕ′ =
0.441).
Figure 8. n versus A/Ac for Case A and B.
(i.e. larger K values) leads to larger settlement
improvement factors, i.e. larger K values leads
to increased horizontal stresses in the soil, hence
providing more resistance to lateral bulging of the
granular material.
5 CONCLUSIONS
• Cylindrical cavity expansion has been used to
calculate post-installation coefficients of lateral
earth pressure following stone column installa-
tion in a soft clay. The modelling captures a vari-
ation of K/K0 with radius not usually considered
by other methods.
• The K/K0 profile generated as a result of the
cavity expansion procedure has been imported
successfully as an input profile for a unit cell
axisymmetric model.
• Load settlement behaviour has been examined
for two Cases A and B with K/K0 = 1 and K/K0
as predicted by CCE respectively.
• The results show that the approach can be used
as an alternative to the conventional ‘wished-in-
place’ column installation technique (with larger
settlement improvement factors predicted when
installation has been taken into account).
231
ACKNOWLEDGEMENTS
The authors would like to acknowledge the Irish
Research Council for Science, Engineering, and
Technology (IRCSET) for funding this research.
REFERENCES
Allman, M.A. & Atkinson, J.H. 1992. Mechanical prop-
erties of reconstituted Bothkennar soil. Géotechnique
42(2): 289–301.
Ambily, A.P. & Gandhi, S.R. 2007. Behavior of Stone
Columns Based on Experimental and FEM Analysis.
Journal of Geotechnical and Geoenvironmental
Engineering 133(4): 405–415.
Balaam, N.P. & Booker, J.R. 1981. Analysis of rigid rafts
supported by granular piles. International Journal for
Numerical and Analytical Methods in Geomechanics
5(4): 379–403.
Brinkgreve, R.B.J., Swolfs, W.M. & Engin, E. 2011.
PLAXIS 2D 2010 Material Models Manual, PLAXIS
B.V.
Carter, J.P., Randolph, M.F. & Wroth, C.P. 1979. Stress
and pore pressure changes in clay during and after the
expansion of a cylindrical cavity. International Journal
for Numerical and Analytical Methods in Geomechanics
3(4): 305–322.
Castro, J. & Karstunen, M. 2010. Numerical simulations
of stone column installation. Canadian Geotechnical
Journal 47(10): 1127–1138.
Castro, J. & Sagaseta, C. 2009. Consolidation around
stone columns. Influence of column deformation.
International Journal for Numerical and Analytical
Methods in Geomechanics 33(7): 851–877.
Debats, J.M., Guetif, Z. & Bouassida, M. 2003. Soft
soil improvement due to vibro-compacted columns
installation. Proceedings of the International Work-
shop “Geotechnics of Soft Soils. Theory and Practice”,
Noordwijkerhout, 551–556.
Domingues T.S., Borges J.L., & Cardoso A.S. 2007. Stone
columns in embankments on soft soils. Analysis of the
effects of the gravel deformability. Proceedings of the
14th European Conference on Soil Mechanics and Geo-
technical Engineering, Madrid, 1445–1450.
Elshazly, H., Hafez, D. & Mossaad, M. 2008.
Back-calculating vibro-installation stresses in
stone-column-reinforced soils. Proceedings of the
ICE—Ground Improvement 10(2): 47–53.
Gäb, M., Schweiger, H.F., Kamrat-Pietraszewska, D. &
Karstunen, M. 2008. Numerical analysis of a float-
ing stone column foundation using different consti-
tutive models. Proceedings of the 2nd International
Workshop on the Geotechnics of Soft Soils—Focus on
Ground Improvement, Glasgow, 137–142.
Gibson, R.E. & Anderson, W.F. 1961. In situ measure-
ments of soil properties with the pressuremeter. Civil
Engineering and Public Works Review 56(658): 615–618.
Goughnour, R.R. & Bayuk, A.A. 1979. A Field Study of
Long Term Settlements of Loads Supported by Stone
Columns in Soft Ground. Proceedings of the Interna-
tional Conference on Soil Reinforcement: Reinforced
Earth and Other Techniques (Coll. Int. Renforcements
des Sols.), Paris, 1: 279–285.
Guetif, Z., Bouassida, M. & Debats, J.M. 2007. Improved
soft clay characteristics due to stone column installa-
tion. Computers and Geotechnics 34(2): 104–111.
Killeen, M.M. & McCabe, B.A. 2010. A Numerical
Study of Factors Affecting the Performance of Stone
Columns Supporting Rigid Footings on Soft Clay.
Proceedings of the 7th European Conference on Numer-
ical Methods in Geotechnical Engineering, Trondheim,
833–838.
Kirsch, F. 2006. Vibro Stone Column Installation and
its Effect on Ground Improvement. Proceedings of
the International Conference on Numerical Modelling
of Construction Processes in Geotechnical Engineering
for Urban Environment, Bochum, 115–124.
McCabe, B.A., Nimmons, G.J. & Egan, D. 2009. A review
of field performance of stone columns in soft soils.
Proceedings of the ICE—Geotechnical Engineering
162(6): 323–334.
Nash, D.F.T., Powell, J.J.M. & Lloyd, I.M. 1992. Initial
investigations of the soft clay test site at Bothkennar.
Géotechnique 42(2): 163–181.
232
Priebe, H.J. 1976. Evaluation of the settlement reduction
of a foundation improved by Vibro-Replacement.
Bautechnik 2: 160–162.
Priebe, H.J. 1995. The design of vibro replacement.
Ground Engineering 28(10): 31–37.
Pulko, B., Majes, B. & Logar, J. 2011. Geosynthetic-
encased stone columns: Analytical calculation model.
Geotextiles and Geomembranes 29(1): 29–39.
Schanz, T., Vermeer, P.A. & Bonnier, P.G. 1999. The
hardening soil model: Formulation and verification.
Beyond 2000 in Computational Geotechnics—Ten Years
of PLAXIS International, Amsterdam, 281–290.
Vesic, A.S. 1972. Expansion of cavities in infinite soil
masses. Journal of the Soil Mechanics and Foundations
Division 98(4): 265–290.
Watts, K.S., Johnson, D., Wood, L.A. & Saadi, A.
2000. An instrumented trial of vibro ground treat-
ment supporting strip foundations in a variable fill.
Géotechnique 50(6): 699–708.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Cement grout filtration in non-cohesive soils

X.A.L. Stodieck
Federal Waterways Engineering and Research Institute, Karlsruhe, Germany
Norwegian University of Science and Technology, Trondheim, Norway

T. Benz
Norwegian University of Science and Technology, Trondheim, Norway

ABSTRACT: This paper is concerned with cement grout filtration during installation of pressure
grouted ground anchors in non-cohesive soils. It is aimed for a simple model that can predict the build up
of filter cake during grouting. A series of laboratory tests were carried out to investigate the time depend-
ent expulsion of water from cement grouts at different grouting pressures and water contents. Addition-
ally, permeability tests were performed on filter cake material. Different approaches to model the filtration
process of cement grouts are compared and applied in back calculation of the performed tests. Material
parameters such as permeability and consolidation coefficients are provided for the cement grouts tested.
Amethod to estimate the time required for filter cake formation is proposed.

1 INTRODUCTION device. The applicability of the two models is then

The ability to predict filter cake formation and
its mechanical behaviour is a vital first step in
understanding the effects of anchor installation.
Pressure grouting may increase the load bearing
capacity of grouted anchors in non-cohesive soils.
A modelthat can predict the build up of filter cake
in such soils is therefore essential in determining
installation parameters such as grouting pressure
and duration. Today, these parameters are typi-
cally chosen on an empirical basis. It is assumed
that during grouting water, is filtered from the
grout while the majority of cement particles are
retained by the soil. If a sufficient amount of water
is expelled from the grout, cement particles contact
each other and form a filter cake. This filter cake
material enables the transfer of effective stresses to
the soil, leading to expansion of the borehole walls
and increase of the radial effective stresses in the
soil. Given that a stiff filter cake material exists, the
stress increase will be partially maintained.
In this paper a study on the filtration behaviour
of cement grout is presented. First, two models to
calculate the expulsion of water from cement grout
are reviewed, a two-dimensional filtration approach
and classical consolidation theory. Then, a series of
laboratory tests on cement grout filtration is pre-
sented, which was conducted to validate both mod-
els. Filtration tests at different grouting pressures
were carried out on cement grout with varying ini-
tial water contents, using a simple pressure filtration
validated by back calculation of the laboratory tests.
Model parameters are determined for each model.
These parameters are then used to find correlations
between the initial water content of cement grout,
grouting pressure and filtration time. Finally, both
models are compared with respect to the capability
to predict the filtration rate of cement grouts.
In the presented study cement grout is consid-
ered as granular material with constant material
properties in time. A retarding agent was used to
delay the hydration process and to ensure constant
grout properties during the test duration. Tests
conducted with identical grout mixture at different
ages showed very similar behaviour.
1.1 Filter cake formation
A filtration process is the separation of solids and
liquids. Pressure is applied to a suspension in order
to drain the liquid part through a filter while the
solid particles are retained. Particles pile up at the
filter medium and build a so-called ‘filter cake’
which then acts as filter medium itself. Within the
filter cake particles are in contact to each other
and effective stresses are transferred between the
grains. The concentration of particles where they
start to contact each other might be compared to
the loosest packing of a material. The bulk den-
sity ρ without compaction of the used dry Ordi-
nary Portland Cement ranges between 1.23 kg/l
and 1.38 kg/l. With a specific grain density ρs of

233
3.15 kg/l the respective void ratio ranges from
e = 1.56 to e = 1.28.
The concentration of cement grouts is usually
m
specified by the mass ratio w / z = mwc . Assuming
all voids are fully saturated, the relation between
e and w/z is
V m γ
e= w = w c, (1)
Vc mc γ w
where mw and mc are the masses of water and
cement respectively.
The void ratio of the dry cement without com-
paction corresponds to a water content between
w/z = 0.5 and w/z = 0.4. The highest bulk density
which can be achieved by vibration ranges from
1.6–1.9 kg/l, which corresponds to e = 0.96 − 0.66
and w/z = 0.30 − 0.21 for full saturation.
Warner (2004) states that cement grout is a non-
Newtonian fluid that is a viscous fluid for water
contents above w/z = 0.6, but has paste consistency
for water contents below w/z = 0.4. In between it
flows to some extent but is also capable to store
some energy.
Therefore, it is assumed that the particles start
to contact each other for water contents between
0.4 and 0.5, but might rearrange easily as only little
contact forces are established between the grains
and most stresses are transferred by pore pres-
sure. In this study it is however looked for a filter
cake with a stable grain skeleton and consider-
able strength and stiffness. For grouting pressures
between 5 kPa and 60 kPa McKinley (1993) found
filter cakes with water contents between 0.3 and
0.5, while Picandet et al. (2011) and Lee et al. (2012)
report filter cake water contents of around 0.3.
1.2 Filter cake formation analysis
The one dimensional filtration process of a suspen-
sion can be modelled by a simple two-phase filtra-
tion approach. McKinley (1993) used this model
to describe the filtration process of cement grouts
and states that the formation of filter cake can be
considered as a distinct phase change from a liquid
to a solid phase. The cement particles retained by
the filter medium form a filter cake with constant
material properties while the suspension retains the
initial concentration. The build up of filter cake is
defined by the amount of water expelled from the
suspension.
The rate of water expulsion is limited by the
hydraulic resistance of the filter cake and filter
medium. McKinley (1993) found that for a suf-
ficient large permeability of the filter medium, it
only influences the very first part of the filtration
process and can thus be disregarded. Darcy’s Law
234
gives the discharge rate Q in relation to the filter
cake’s current thickness Lc and its permeability kc
kc σ
Q= A (2)
γ w Lc
where A is the drainage area and σ the total stress
acting on top of the sample.
Assuming a constant concentration of cement
particles within each phase, the calculation of filter
cake formation with time is straight forward. The
piston deformation during filtration δ, shown in
Figure 1, can be calculated with the initial void ratio
of the grout eg and the final filter cake void ratio ec:
eg − ec
δ = Lc . (3)
1 + ec
If water and cement particles are regarded
incompressible, the volume change of the sample
is equal to discharge rate of water. Dividing
Equation (2) by A gives
dδ kc σ
= . (4)
dt γ w Lc
Combining equations (3) and (4) and integrating
finally gives the filter cake thickness with time
2σ kc ⎛ ec ⎞
L2c = ⎜ ⎟ t. (5)
γ w ⎝ eg ec ⎠
The filtration process ends, when the piston
reaches the filter cake at Lc,fin = h0 − δfin.
1.3 Terzaghi consolidation
Other authors used Terzaghi’s classical consolida-
tion theory to describe the filtration process of
cement grouts (Lee et al. (2012), Kleyner & Krizek
(1995) and Picandet et al. (2011)) and they claim
that the process could be modelled reasonably well
with this approach.
Classical consolidation theory is based on the
assumption of homogeneous material properties
within the sample, constant over time. The dissipa-
tion of excess pore pressures over sample height is
du d 2u
= cυ 2 , (6)
dt dz
where cv is the consolidation coefficient, related to
the permeability k and stiffness of the material
k
cυ = . (7)
γ w mv
Figure 1. Filtration test set-up.
The compressibility coefficient mv is defined as
∈
mv = (8)
σ′
where ∈ is the vertical strain caused by the effective
stress σ ′.
2 FILTRATION TESTS
2.1 Materials
For all tests Norcem Standard Portland Cement
with a specific grain density of 3.15 kg/l was used.
The retarding agent (Sika® Retarder) was added to
prevent setting effects during the test procedure.
2.2 Test procedure
The experimental set-up for a filtration test is shown
in Figure 1. A 54 mm diameter steel cylinder is filled
with cement mix up to a height of h0 = 80 mm and
a piston is placed on top of the sample.
The piston is loaded until the desired constant
pressure is reached. Then a valve is opened at the bot-
tom of the test apparatus to allow drainage through
a filter paper and porous stone. During the filtration
test water is drained out of the cement grout while
cement particles are retained by the filter medium.
The piston deformation δ and pore pressure u are
recorded at top of the sample during the test.
Tests with grouting pressures of 100, 200, 500,
1000 and 1400 kPa were carried out with water
contents of w/z = 0.4, 0.45, 0.5 and 0.6. With each
mixture, a minimum of two tests were conducted
for each grouting pressure.
2.3 Results
Results of a filtration test with w/z = 0.5 and a
grouting pressure of 1000 kPa is shown in Figure 2.
The piston deformation is approximately linear
with time until it declines at a certain point after
235
Figure 2. Typical deformation and pore pressures
measured at the piston during filtration test with grout-
ing pressure = 1000 kpa and w/z = 0.5.
Figure 3. Typical piston deformation compared to the
two phase cake filtration solution and classical consolida-
tion theory.
which only little further deformations takes place.
The pore pressure at top of the sample remains
constant in a fist phase until it suddenly drops at a
certain point in time (filtration time tfilt). This shows
that in the first phase of the tests some parts of the
grout are still liquid and do not transfer effective
stresses. The drop of pore pressures at tfilt indicates
that cement particles are in contact and trans-
fer effective stresses from piston to the bottom.
Considering cake filtration theory this would be
the time when the piston reaches the top of the fil-
ter cake material and all excess water is filtered out.
However, more water is expelled after the filtration
point which shows that after a first phase of filtra-
tion the filter cake material consolidates further.
Figure 3 shows the deformation curve from the
test results lying between the deformations calcu-
lated with Terzaghi’s consolidation theory and the
two-phase filtration model. The two approaches
represent an upper and lower limit. The two
phase filtration model allows a clear distinction
between grout and filter cake while in consolida-
tion approach the expulsion is associated with the
dissipation of excess pore pressure which happens
gradually. While in the filtration approach the pore
pressures on top would theoretically stay constant
until the piston reaches the filter cake, they decrease
gradually from the beginning when using the Based on the filtration model the coefficient of
consolidation approach. Although it seems that the permeability of the filter cake kc was determined
filtration approach describes the test results more by rearranging (5) to
accurately, using consolidation theory would be
very handy as it is readily available in common finite h2fiin γ w ⎛ eg ec ⎞
kc = . (10)
element programmes for geotechnical applications. 2σ t ⎜⎝ 1 + e ⎟⎠
fiin c

2.3.1 Water content

The water content of the grout was determined The time tfin represents the point where the pis-
directly before and after the filtration test by drying ton touches the filter cake and was determined by
a small sample in a microwave oven until the weight extrapolating the linear part of the deformation
remained constant. Although drying in the micro- − t curve to get tfin = 4 × t50
wave requires much less time than conventional Assuming consolidation theory, the mean coef-
drying in an oven, the method is not very accu- ficient of permeability km was determined by rear-
rate as water may be bound chemically. Therefore, ranging Equation (7)
the water content of the filter cake material was
cv
alternatively calculated from the final void ratio. km = (11)
The filter cake’s final void ratio ec was assessed by γ w mv
rearranging Equation (3), and using Lc = h0 − δfin.
and inserting
⎛ ( + eg )δ ffin ⎞
ec eg − ⎟⎠ (9) ∈ δ ffin
i
⎝ h0 mv = = . (12)
σ h0σ
Water contents determined by drying in the
microwave oven were on average 8% lower than the For some samples permeability tests were carried
theoretical w/z values. out after filtration. A constant hydraulic head was
The filter cake water contents w/zfin versus grouting applied at top of the probe and the flow through the
pressure are presented in Figure 4. Taking the mean sample was recorded. Based on Darcy’s law, the coef-
values for all tests at a certain grouting pressure gives ficient of permeability was then determined using
a final water content of around w/zfin = 0.29 for pres- the specific discharge q and the applied gradient i.
sures above 200 kpa and w/zfin = 0.32 for pressures at
dh p
200 kpa, based on the calculated values. i= = (13)
dz γ w h fin
2.3.2 Permeability
dh γ w h fin
The permeability of the filter cake material was q k →k q (14)
determined directly from permeability tests and dz p
compared to the values achieved from back calcu-
lation, using the two-phase filtration model as well The filter cake’s permeabilities k determined
as classical consolidation theory. directly from permeability tests and by back

Figure 5. Coefficient of permeability k, versus final

water content. Determined directly from permeability tests
Figure 4. Final water contents of filter cake versus ( ); from filtration tests, using filtration theory ( )
grouting pressure. Measured values ( ) and calcu- and consolidation theory ( ). Compared to values
lated values ( ). from Picandet et al. (2011) (dashed line).

236
calculation are presented in Figure 5. The
permeabilities calculated with the consolidation
approach are in general higher than the values from
filtration approach, as the mean value of the entire
sample is taken into account while the filtration
approach only considers the permeability of the
filter cake. The results form permeability tests are
close to the results derived from the filtration tests
using two-phase filtration theory. Figure 5 shows
a clear dependency between permeability k and
final water content of the filter cake. This agrees
with the understanding of a void ratio dependent
permeability of granular materials. Picandet et al.
(2011) propose to describe the void ratio dependent
permeability of cement grouts with the function
e 3
k Cx (15)
1+ e
which was originally introduced by Carman (1956),
or with an logarithmic approach
Δe
Δ (log10 k ) = (16)
Ck
suggested first by Taylor in 1948. The Carman
equation fitted for permeability tests on cement
grout filter cake carried out by Picandet et al. (2011)
is plotted in Figure 5. For the permeability tests
of this study a better fit was found with Taylor’s
logarithmic relation given in Equation (16).
3 BACK CALCULATION OF TEST
RESULTS
3.1 Filtration rate (Filtration model)
Based on the two-phase filtration model, the filter
cake formation can be calculated directly from the
filtration rate, which is considered constant with
t . The rate of filter cake formation is defined
as the change of filter cake thickness Lc with t
and was determined from test results using the
filtration approach. Assumingg a linear relation
between deformation δ and t , the inclination of
the filtration curve was approximated with
dδ δ
≈ 50 . (17)
d t t50
Combining Equation (17) and (3) yields filtra-
tion rate
dL
Lc δ 50 ⎛ 1 + ec ⎞
= ⎜ ⎟. (18)
d t t50 ⎝ eg ec ⎠
237
The filtration rate for different initial water con-
tents and grouting pressures is illustrated in Figure 6.
Filtration rates clearly increase with grouting pressure,
due to higher pressure gradients causing higher dis-
charge rates (Equation (2)). Although the test results
show some scatter, the filtration rate decreases with
increasing initial water content. This finding is plau-
sible, as with higher initial water content, more water
must be drained to reach the final water content.
The theoretical filtration rate was back calcu-
lated for different initial water contents, using the
filtration model. Rearranging Equation (5) and
differentiation with respect to t gives
dL
Lc ⎛ 2kk ⎞⎛ ec ⎞
= ⎜ c⎟⎜ ⎟σ. (19)
d t γ
⎝ w ⎠ ⎝ g ec ⎠
e
In reasonable agreement with Figure (4), the back
calculation was conducted with a constant, represent-
ative filter cake void ratio ec = 0.95, corresponding to
a water content of w/zc = 0.3. For w/zc = 0.3 the perme-
ability kc = 7.5 × 10−8 m/s was then determined from
the correlation presented in Figure (5). The back cal-
culated filtration rates fit well with the test data.
3.2 Coefficient of consolidation
(Consolidation Model)
Using Terzaghi’s consolidation theory the defor-
mation rate is defined by the coefficient of con-
solidation cv and compressibility coefficient. The
compressibility coefficient mv was estimated with
Equation (8), assuming full consolidation was
achieved at the end of the test. The coefficient
of consolidation cv was estimated from t50 at 50%
consolidation, assuming that the consolidation
curve is linear with t . The theoretical solution of
dL
L
c
Figure 6. Filtration rate versus grouting pressure.
d t
Test results (symbols) and calculated values from
filtration approach—assuming kc = 7.5 × 10−8 m/s and
w/zfin = 0.3 (lines).
Terzaghi’s consolidation theory at this point gives Figure 9 shows the degree of consolidation U = δ/
the time parameter T = cvt50 /h2 = 0.197 and therefore δfin of a typical filtration test compared to consoli-
the consolidation parameter is determined with dation curves for different values of cv. The consoli-
dation curve for the cement grout passes through
t50 different Terzaghian consolidation curves, which
cv = 0.197 . (20) means cv changes gradually with degree of consoli-
h2
dation U and therefore with effective stress σ′.
This method is a simplification of the ‘log(t)- The filtration tests were back calculated with a
method’ (Casagrande & Fadum 1940), which addi- modified Terzaghi consolidation approach, using
tionally includes the determination of an initial the finite difference method. The sample height was
and final deformation. For this study it was found divided in 30 elements and in each element the con-
adequate to assume δ0 = 0 and take the last meas- solidation coefficient cv was updated for each time
ured deformation as δfin. step. A linear interpolation between the initial cv,ini
The coefficients of consolidation cv versus and final cv,fin consolidation coefficient was used:
grouting pressure determined from this method
are plotted in Figure 7. An increase of cv with cv = cv,ini + (cv, fin
i cv,ini )U . (21)
increasing grouting pressure is shown, while no
clear correlation of initial water content and cv can Unfortunately the correlation presented in
be identified. A linear correlation of cv and grout- Figure 7 can not be used for this purpose as the
ing pressure was hence used for all test results: parameters were determined from t50 and represent
cv = 6.07 × 10−8 ⋅ σ + 7.79 × 10−6. Using this cor- the cv value at 50% consolidation for the given pres-
relation for cv, tests at different grouting pressures sure. The result of this back calculation is shown in
were back-calculated with Terzaghi’s consolidation Figure 9 (solid line).
theory. Results are presented in Figure 8. The cal-
culated consolidation curves fit reasonably well
with the test results.
The assumption of a constant cv value during the
test, as used in classical consolidation theory, might
not be appropriate for this material, as it exhibits large
changes in permeability and stiffness during the con-
solidation process. Both parameters, permeability km
and compressibility mv, decrease with decreasing void
ratio σ ′ or increasing effective stress cv. Looking at
Equation (7), cv could either decrease, increase or stay
constant with effective stress, dependent on which
parameter changes more. It is usually assumed that
the change of km and mv cancels each other out and
the assumption of a constant cv is appropriate. This
does not apply for all materials, and Abuel-Naga &
Pender (2012) proposea modified consolidation
approach with linear effective stress dependent cv. Figure 8. U versus t for w/z = 0.4, 0.45, 0.5 and 0.6
for different grouting pressures. Back calculation of test
results with Terzaghi consolidation theory (red line).

Figure 7. Consolidation parameter cv versus grouting Figure 9. Consolidation curve of filtration test in com-
pressure. Best fit for all values cv = 6.07 × 10−8σ + 7.79 × 10−6 parison to back calculation with constant cv and effective
with R2 = 0.976. stress dependent cv (cv,ini = 2.5e − 5; cv,fin = 8.0e − 5).

238
3.3 Estimation of filter cake thickness
In the two phase filtration model consolidation
of the filter cake is not taken into account and the
extension of the final filter cake material is directly
calculated from Equation (5). A clear distinction
between filter cake and unfiltered grout is made.
In contrast, no distinction is made between fil-
ter cake and grout in classical consolidation the-
ory, as the whole sample consolidates gradually
with time. Material properties such as stiffness and
permeability are regarded constant. However, if
we assume that a stable contact of grains is estab-
lished at a certain void ratio, the expansion of filter
cake could then be related to the degree of consoli-
dation U.
δ
U= . (22)
δ ffin
i the material. The compressibility mv can be deter-
The consolidation ratio U at tfilt, at which pore
pressures at top of the sample start to decrease,
ranges between 0.75 to 0.95, which corresponds
to water contents between 0.28 to 0.36. Theoreti-
cally the deformation and therefore degree of con-
solidation at tfilt depends mainly on the initial water
content of the grout, as this defines the amount of
deformation needed to reach a certain mean water
content in the sample. In addition the applied
grouting pressure defines the compression of the
filter cake material.
However, no significant correlation between
initial water content, grouting pressure and U
was found. This could be due to variations of
the test results and inaccuracies of the interpreta-
tion. Assuming that all tests result in a final water
content of 0.29 as indicated in Figure 4, a mean
water content at the filtration time of w/zfilt = 0.32
would give Ufilt values of 0.75, 0.83, 0.87 and 0.91
for initial water contents of 0.75, 0.83, 0.5 and 0.6
respectively. Although the compression of filter
cake for different grouting pressures is ignored in
this approach, it fits in the range of test results.
4 SUMMARY
This paper presents a series of filtration tests on
cement grouts with water contents ranging from
0.4 to 0.6. The influence of initial water content
and grouting pressure was investigated and test
results were back calculated with a two-dimen-
sional filtration model and classical consolidation
theory.
Both models could be used to back-calculate
the laboratory tests. While the filtration model is
designed to describe filtration of solid particles
from a suspension, consolidation theory considers
the expulsion of water from a solid. The water con-
239
tents of the cement grouts used range between the
limit for liquid and plastic consistency and in this
case the two models represent an upper and lower
limit to describe the filtration process.
Required parameters for the filtration model are
the initial and final water contents and the perme-
ability of the filter cake. The filter cake material
exhibits a void ratio dependent permeability kc
between 3.81 × 10−8m/s and 1.97 × 10−7m/s. The val-
ues calculated on the basis of filtration theory com-
ply well with the directly measured values, which
confirms that the filtration model is applicable.
Assuming a final water content of w/zfin = 0.3 and
the corresponding permeability kc = 7.5 × 10−8m/s
from figure (5) the filtration rate could be back-
calculated for all tested grouts.
The consolidation model requires the consoli-
dation coefficient cv and compressibility mv of
mined from the strain required to achieve the final
water content and the applied grouting pressure.
Values for cv were found between 6.86 × 10−6 m2/s
and 1.33 × 10−4 m2/s, based on Terzaghi’s consoli-
dation theory. The mean values of all tests can be
described by a linear correlation between cv and
grouting pressure. With this correlation tests could
be back calculated using consolidation theory. To
determine the filtration point, it is proposed to use
the degree of consolidation U. By assuming a final
water content w/zfin = 0.29, a mean water content
at the filtration time w/zfilt = 0.32 was found as
maximum value to set limits to the determined U
values.
The determined calculation parameters for both
models are dependent on the final water content
of the filter cake material which was found to vary
between 0.25 and 0.35. No correlation between
w/zfin and grouting pressure could be identified.
Therefore it is not possible to determine the void
ratio at which the grout actually acts as a solid
and deformation is stress dependent. It seems that
the scatter of test results exceed the influence of
grouting pressure and initial water content for the
range of water contents and grouting pressures of
this study. Where no correlations could be identi-
fied, mean values of all tests were taken for back-
calculation of test results, which gave reasonable
results. These parameters might not be applicable
for water contents and grouting pressures outside
the test range.
The filtration model allows a straight-forward
determination of the filter cake thickness with
time, but for an investigation of the filtration proc-
ess in borehole geometry the use of consolidation
theory could be favourable, as it is already imple-
mented in common geotechnical FEM software.
When using consolidation theory it is not straight-
forward to define the filtration time, especially as
calculated deformations are inaccurate in the final
part of the curve. However, a rough estimation of
the minimum filtration time is possible.
REFERENCES
Abuel-Naga, H.M. & Pender, M.J. 2012. Modified terza-
ghi consolidation curves with effective stress-dependent
coefficient of consolidation. Gotechnique Letters 2(2),
43–48.
Carman, P.C. 1956. Flow of gases through porous media.
London: Butterworths Scientific Publications.
Casagrande, A. & Fadum, R.E. 1940. Notes on soil testing
for engineering purposes. Cambridge, Mass.: Harvard
University, Graduate School of Engineering.
Kleyner, I. & Krizek, R.J. 1995. Mathematical Model for
Bore-Injected Cement Grout Installations. Journal of
geotechnical engineering. 121(11), 782–788.
240
Lee, S.-W., Kim, T.-S., Sim, B.-K., Kim, J.-S. & Lee, I.-M.
2012. Effect of pressurized grouting on pullout resist-
ance and group efficiency of compression ground
anchor. Can. Geotech. J. Canadian Geotechnical
Journal 49(8), 939–953.
McKinley, J.D. 1993. Grouted ground anchors and the soil
mechanics aspects of cement grouting. Ph. D. thesis.
Picandet, V., Rangeard, D., Perrot, A., & Lecompte, T.
2011. Permeability measurement of fresh cement
paste. Cement and Concrete Research 41(3), 330–338.
Taylor, D.W. 1948. Fundamentals of soil mechanics. New
York: J. Wiley.
Warner, J. 2004. Practical Handbook of Grouting—Soil,
Rock, and Structures. Norwich, NY: John Wiley &
Sons.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

The undrained cohesion of the soil as a criterion for column installation

with a depth vibrator

J. Wehr
Keller Holding GmbH, Offenbach, Germany

ABSTRACT: Piles, vibro concrete columns and rigid inclusions cannot be installed in liquid media like
water with cu = 0 kPa, because a cone shaped slope will form instead of a cylindrical body. In international
standards a limit of cu = 15 kPa is currently used as a lower boundary. Additionally the minimum center
to center distance between fresh concrete displacement piles without permanent casing is specified in
EN 12699 between 6–10 times the diameter depending on the cu-value. These limits should be applied
to all kinds of displacement piles like rigid inclusions in order to avoid damaging neighbouring columns
during installation.
However, recent world-wide site experiences reveal that these limits are not valid for granular columns like
vibro stone columns. The limit of the undrained cohesion should be reduced to cu ≈ 5 kPa.

1 INTRODUCTION
It is common knowledge that vibro stone columns
and vibro concrete columns cannot be installed in
liquid media like water with cu = 0 kPa, because a
cone shaped slope will form instead of a column.
In the German recommendation (FGFS 1979) for
soil improvement with depth vibrators, dated 1979,
limits of cu = 15–25 kPa are given. Since that time
the lower limit value has been, and still is, used in
all kinds of German and international standards.
However, more than 10 years of site experiences of
the Keller company around the world reveal that
this limit value is far too conservative and should
be reduced to cu = 5 kPa.
The aim of this publication is to find out the limit
value of the undrained cohesion where a column
installation is no longer possible. Site experiences
of Keller have been compiled and summarized and
furthermore systematic model tests are presented
varying the cu-value.
2 PREVIOUS SITE EXPERIENCES
The cu-values of Keller sites have been com-
piled, where an installation of vibro stone col-
umns (VSC) or vibro concrete columns (VCC)
with a depth vibrator was possible in soils with
cu < 15 kPa. Details may be found in the following
references.
Austria:
• Klagenfurt, VSC, metro-market, cu = 5–10 kPa
(Marte et al., 2005)
Germany:
• Lübeck-Herrenwyk, VSC, metal factory,
cu = 11–26 kPa (Völzke, 2001)
• Mering, VCC, railroad embankment,
cu = 5–20 kPa (Borchert et al. 2004)
• Rethen, VSC, sugar plant, cu = 12–18 kPa
• Zossen, VCC, dam for ring road, cu = 4–8 kPa
(Zimmermann, 2003)
Malaysia:
• Shah Alam Expressway, VSC, (Raju/Hoffmann
1996, Raju 1997)
Kinrara Interchange, VSC, cu = 6–40 kPa
Sri Petaling Interchange, VSC, cu = 8–50 kPa
Shah Alam West, VSC, cu = 5–15 kPa
• Putrajaya Boulevard,VSC, cu = 5–20 kPa
(Raju, 2002)
Poland:
• VSC, highway embankment, cu = 5 kPa
(Marte et al., 2005)
In summary, the experience of all the above pre-
vious sites show that the installation of vibro stone
columns and vibro concrete columns is possible
well below the old limit value of cu = 15–25 kPa,
and that the actual limit value should be approxi-
mately cu = 4–5 kPa.

241
3 MODEL TESTS AT KAISERSLAUTERN
UNIVERSITY

The aim of these model tests is to find a new cu

limit value for dynamic excitation using a liquefi-
able very well defined silt. Here the results of a
Diploma thesis are presented (Gundacker 2004).

3.1 Test materials

All tests were executed in silt with different water
contents. For each water content the correspond-
ing undrained cohesion was determined with labo-
ratory shear vane tests, see Table 1.
Perlea (2000) describes the liquefaction of silt
under dynamic excitation with impressive case his-
tories, i.e. earthquakes, and emphasizes that not
only sand can liquefy.
He gives the following conditions for liquefac-
tion of silts:
• Clay content < 20%
• Plasticity index Ip ≤ 13
• Liquid limit wL < 33.5% and water content
w > 30% or w > 0.87 wL and w < 30%
• Sensitivity S > 4
A mixture of 97.5% crushed stone powder and
2.5% caolinite fulfills all these criteria, except
for the sensitivity of 1.6 being below the recom-
mended value of 4. This mixture is used for all fur-
ther tests.
In order to model the stone material a uniform
coarse sand is used with d50 = 0.77 mm, emax = 0.875
and emin = 0.434.
3.2 Test layout
The model scale is 1:10 and results mainly from the
size of the PVC-tube modeling the depth vibrator.
A PVC container with an inner diameter of
29 cm and a height of 60 cm was filled with silt up
to a height of 50 cm, see Figure 1.
In order to install the sand columns, a PVC-
tube with an outer diameter of 5.0 cm and an
inner diameter of 4.6 cm was used. This PVC-tube
replaces the vibrator and follower tubes which have
a diameter of approximately 40 cm in reality. The
PVC-tube is necessary to keep the hole open dur-
Table 1. Undrained cohesion depending on the water
content of the silt.
Water content in% 29.9 30.5 32.0 35.5
Undrained cohesion in kPa 5.5 4.0 3.5 2.0
Figure 1. PVC container.
ing sand filling, because no water or pressurized
air is used as in reality.
To model the vibrations of a real vibrator,
a small vibrator was fixed to the lower end of a
second PVC-tube (“vibrator tube”) with an outer
diameter of 4.5 cm. The vibrator tube fits exactly
into the first tube. Horizontal vibrations were
applied. Details are given in Table 2.
Being able to observe the shape of the model
columns after the tests, the sand was mixed with
cement in the ratio of 4:1. This changes the behav-
ior of the sand column only in a negligible way.
The cement is able to harden by sucking water out
of the surrounding silt.
3.3 Column installation and definition of failure
Although static penetration has been included in
the test program, in the following sections only
dynamic penetration will be reported. In this
way the reality is modelled best. Applying a static
weight of 40 N on both tubes this assembly creates
a hole in the soil. In reality static pull down forces
of up to 200 kN are used.
After reaching the desired depth the vibrator
tube is removed. Subsequently sand with a height
of 3 cm is filled into the outer tube and then the
tube is pulled upwards resulting in a 2 cm high
sand column only supported by the soil. As a next
step the vibrator tube is reintroduced with the
static weight on top. The vibrations are stopped,

242
Table 2. Comparison of vibrators.
Reality Model
Frequency 40–60 hertz 50 hertz
Diameter 30–45 cm 4.5 cm
Double amplitude 3–15 mm 1–2 mm
if the settlement of the column does not change
anymore. Empirically this procedure ends after a
maximum time of 2 minutes. However, if the sand
settles more than 1 cm, the vibrations are stopped,
a new layer of sand is added and finally the vibra-
tor is reintroduced and started again. Before the
start of the tests it has been calculated how much
sand is necessary to double the diameter of the
sand column. Such an increase is regarded as fail-
ure during the installation of a column. If it is pos-
sible to introduce even more sand, only a negligible
resistance of the soil can be expected. In this case
the test is ended.
3.4 Test results
Test no. 1: cu = 5.0 kN/m2 − w = 29.0%
In the literature there are several examples where
the undrained cohesion of 4 to 5 kN/m2 was suffi-
cient to install vibro stone columns, see section 2.
For this reason the first test was executed with a
cu = 5 kN/m2. The aim of this test was to prove
that this undrained cohesion of a soil which is liq-
uefiable according to Perlea (2000) is sufficient to
support a vibro column. No failure was observed
during installation.
Test no. 2: cu = 4.0 kN/m2 − w = 30.5%
In the following tests the water content is sub-
sequently increased by 1.5%. With a water content
of 30.5% an undrained cohesion of 4.0 kN/m2 is
reached. Here as well there was no failure during
column installation.
Test no. 3: cu = 3.5 kN/m2 − w = 32.0%
In this test the water content was again raised by
1.5%. Here failure was observed with an enlarge-
ment of the diameter by 100%, see Figure 2. The
closer the installation comes to the surface the
smaller are the stresses acting as soil resistance and
the faster failure is reached.
Test no. 4: cu = 4.0 kN/m2 − w = 30.5%
In this test the same cu-value as in test no. 2 was
chosen, but the static weight was doubled from
40 N to 80 N. Under this higher load more sand
was vibrated into the surrounding soil, but no fail-
ure was observed.
243
Figure 2. Column with cu = 3.5 kN/m2, where failure
has occurred.
4 CONCLUSION
Site experiences and systematic model tests dem-
onstrate that the installation of vibro stone col-
umns and vibro concrete columns is possible with
a depth vibrator, even in a very soft soil with an
undrained cohesion of cu ≥ 4 kPa. The old limit
value of cu = 15 kPa is no longer valid.
However, vibro columns in such soils should
only be installed by specialist contractors with
intense monitoring (depth, amperage of motor,
stone/concrete consumption, vibration frequency),
preferably online. This is absolutely necessary,
because the above criterion can only serve as an
indicator of the soil before the ground improve-
ment starts, but does not cover aspects of the
dynamic behavior of the system, i.e. centrifugal
force of the vibrator, tendency for soil liquefac-
tion, etc.
In the case of vibro concrete columns it may
be necessary to install a vibro gravel drain first in
order to improve the existing soil. Then the vibro
concrete column is protected by a ring of gravel.
The drain and the column can be installed with the
same bottom feed vibrator unit.
REFERENCES
Borchert, K.-M. & Kattner, G. 2004. Prüfbericht der
Probebelastungen an 3 Fertigmörtelstopfsäulen, inter-
nal report.
FGFS, Forschungsgruppe für Straßenwesen. 1979.
Merkblatt für die Untergrundverbesserung durch
Tiefenrüttler.
Gundacker, S. 2004. Anwendungsgrenzen bei der Herstel-
lung von Schottersäulen, Diploma-thesis, Technical
university of Kaiserslautern.
Marte, R. & Schuller, H. 2005. Verbesserung sehr weicher
Seesedimente und Torfe durch Schottersäulen—zwei
Fallbeispiele, Bauingenieur, Volume 80, 430–440.
Perlea, V. 2000. Liquefaction of cohesive soils, soil
dynamics and liquefaction, ASCE geotechnical special
publication, no. 107, 58–76.
Raju, V.R. & Hoffmann, G. 1996. Treatment of tin mine
tailings in Kuala Lumpur using vibro replacement. Proc.
12th SEAGC, May 1996, Kuala Lumpur.
Raju, V.R. 1997. The behaviour of very soft cohesive soils
improved by vibro replacement, Ground Improvement
Conference, London.
244
Raju, V.R. 2002. Vibro replacement for high earth
embankments and bridge abutment slopes in Putra-
jaya, Malaysia, International Conference on Ground
Improvement Techniques, Malaysia, p. 607–614.
Völzke, B. 2001. Böschungssicherung für die Sanierung
des Metallhüttengeländes in Lübeck-Herrenwyk, 8.
Darmstädter Geotechnik Kolloquium, 233–239.
Zimmermann K.-U. 2003. Gründung von Verkehrswege-
bauten in Feuchtgebieten mit organischen Böden
geringer Scherfestigkeit, Mitteilungen des Instituts für
Grundbau und Bodenmechanik, Universität Braunsch-
weig, Pfahlsymposium, Issue 71, 71–81.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Modeling of rock fall impact using Discrete Element Method (DEM)

G. Grimstad & O. Melhus

Oslo and Akershus University College of Applied Sciences, Norway

S. Degago & R. Ebeltoft

The Norwegian Public Roads Administration, Norway

ABSTRACT: Buried corrugated steel culverts, cushioned on top by a layer of granular material, are
sometimes used to protect infrastructure from rock falls. A crucial point in design of such structure is
the cushion material ability to absorb the kinetic energy from the rock block. This is evaluated by two
important parameters, i.e. the resulting impact load and penetration of the falling rock. In this article a
Discrete Element Model (DEM) is used in simulation of the impact from a falling rock block on such a
cushion layer. The simulation is compared to measurement data from a full-scale field test. Applicability
and potential of DEM to simulate rock fall impacts on cushion are assessed and discussed in the light
of measurements from full scale test and estimations using empirical relations proposed by Norwegian
guideline.

1 INTRODUCTION classical mechanics (Gupta and Saigal 2003);

contact mechanics (Kawahara et al. 2006, Kishi,
Norway is a country with several types of poten- 1999, Montani et al. 1999, Jacquemound 1999);
tial geohazards such as quick clay slides, snow ava- semi-empirical approaches (Yoshida et al. 1988).
lanches, debris flows and rock falls. The country is Detailed discussions on the above methods can be
also characterized by its mountainous features and found in e.g. Degago (2007).
scattered population. The population spread calls In this article the penetration depth and impact
for a need of infrastructures in potentially hazard- forces from a spherical object falling on a cushion
ous areas. One of such hazards is rock falls, which layer, consisting of spherical discrete element (DE)
poses a threat to adjacent infrastructures as well as particles, is presented. Numerical simulations, by
human lives using the infrastructures. In areas with means of the DE method, have previously been
high risk of rock fall incidence, roads or railways
are sometimes shielded with concrete or steel pro-
tection shelters (rock sheds). Rock sheds are usually
preferred in situations where the bounce height is
too high for other mitigation options and in cases
where there is a confined rock fall zone with fairly
high frequency of rock falls. These structures typi-
cally consist of a load-carrying primary structure
with a soil cushion layer on top. This layer is oper-
ating as an energy-absorbing and load-distributing
system (Labiouse et al. 1995, Pichler et al. 2005).
Figure 1 shows a sketch of such a protection sys-
tem. For more info on rockfall and rockfall protec-
tion see e.g. Volkwein et al. (2011).
With regard to design of rock sheds, much
emphasis has been given to estimation of impact
load. However, penetrations related to rockfall
constitute an important part of rockfall study. Dif-
ferent estimation methods have been employed to
estimate peak impact loads on rock sheds that is
the primary input in design of rock fall protec- Figure 1. Sketch of typical rock shed and soil cushion
tion structures. These methods are based on e.g. (Labiouse, et al. 1995).

245
applied for rockfall impact analysis on e.g. embank- large scale odeometer has diameter 500 mm and a
ments (see Plassiard and Donzé, 2009). The experi- height of 400 mm. The tests were run by loading
mental study presented by Ebeltoft et al. (2006) to several load levels with unloading between each
and Ebeltoft and Larsen (2006) is used as a basis load increment. Figure 3 shows results of such
for the numerical study presented in this paper. two oedometer tests, where preloading to 25 kPa
was done prior to the presented data. During the
preloading as much as 10 mm of displacement was
measured. The two tests shows some differences,
2 BACKGROUND
however, one can estimate the oedometer stiffness
to be between 18 and 26 MPa for a stress increment
2.1 Experimental work
of 500 kPa loading from 50 kPa.
Ebeltoft and Larsen (2006) studied an alternative Figure 4 shows high speed–camera documenta-
protection system based on a buried steel structure. tion of the rock fall during impact.
Figure 2 shows an illustration of the experimen-
tal setup. A 15.0 m long structural MP200 arch
type metal culvert was installed. Backfill material
in the near zone consists of 8–16 mm gravel. The
outer zone consists of 23–64 mm size gravel. The
top 1.2 m of the fill made the so-called cushion
layer, which was constructed without compaction.
The experimental study used natural rock boul-
ders with different weights and drop heights. The
test discussed in this paper involves a 200 kg rock
dropped from a 10 m height.
Large scale oedometer tests were ran on the
gravel to find oedometer stiffness properties. The

Figure 3. Oedometer tests results.

Figure 2. Illustration of experimental setup (Ebeltoft Figure 4. High speed-camera documentation of

et al. 2006). impact.

246
2.2 Semi empirical approaches
The Norwegian practice for calculating penetra-
tion depth and maximum contact force (SVV,
2011) is adopted from the Swiss recommendations
given by the “Bundesamt für Strassen” (ASTRA,
2008). Accordingly, Equation 1 and 2 are adopted
to estimate a maximum contact force, Pmax [kN],
and penetration depth, t [m], respectively.
R 02
Pmax = 2.8 R tan ϕ ⋅ Ek 0.6
M E 0.4 tan
e d
2Ek
t=
Pmax
where, R is the radius of falling boulder [m], e is
the soil layer thickness [m]. ME is the soil modulus
(Modulus of subgrade reaction) [Pa], ϕ is the fric-
tion angle [°], and Ek is the energy of the falling
boulder during impact [J].
According to SSV (2011), ME can be estimated as:
Eoed
ME =
2
where, Eoed is the oedometer modulus.
If one assumes ME = 10 MPa, Ek = 19.6 kJ,
tanϕ = 1.0, R = 0.26 m, and e = 0.6–1.2 m, then
Pmax = 240–330 kN and t = 0.12–0.16 m. Figure 5
gives the deceleration history for the 200 kg stone
in the field test. From the figure one can estimate
the maximum force to be about 220 kN, when
multiplying the deceleration with the weight of
the stone. The measured penetration depth after
impact was 0.20 m.
2.3 Discrete element method
In this work, the Star-CCM+ program v. 7.04 (CD-
adapco, 2012) is used for the dynamic discrete ele-
ment simulations. DEM was established by Cundall
Figure 5. Deceleration history for the 200 kg stone.
and Strack (1979). STAR-CCM+ uses a classical
mechanics method to model DEM, and is based on
soft-particle formulation where particles are allowed
to develop an overlap. The calculated contact force
is proportional to the overlap, as well as to the parti-
cle material and geometric properties (CD-adapco,
2012). In addition to the standard linear momen-
tum, the formulation incorporates the angular
momentum conservation equation (equation 4).
(1) ⎛ ⎡ ω p ⎤⎞
(
Ip p ) ∑ ⎜ rc ⋅ ⎢ Fc Crr Fc ⋅ n c ⋅ ⎥⎟
dt contacts ⎜
⎝ ⎢⎣ ω p ⎥⎦⎟⎠
(2) (4)
where, Ip is the particle moment of inertia, ωp is the
particle angular momentum, rc is the vector from
the particle center of gravity to the contact point,
Fc is the contact force vector, Crr is the Rolling Fric-
tion Coefficient and nc is the is a particle surface
normal at the contact point. The contact force vec-
tor is controlled by the contact theory.
In the work presented in this article, the Hertz
Mindlin contact theory is used for the simula-
(3) tions. The inputs to the simulations are then the
Static and Rolling Friction Coefficient (μ and
Crr respectively) and the Normal and Tangen-
tial Coefficients of Restitution (COR) (εn and εt
respectively), both for particle-particle interac-
tions and particle-wall interactions. In addition to
these properties, the elastic stiffness of the parti-
cles, E, Poisson’s ratio, ν and particle density, ρ,
are analysis inputs. In this study spherical parti-
cles with normally distributed diameters of the
particles were used.
3 DEM SIMULATIONS
3.1 Discrete element properties
The properties used in the simulations are as given
in table 1 and table 2. The spheres used in the sim-
ulations were normally distributed with a mean
diameter of 43 mm and a standard deviation of
9 mm. In addition, the minimum and maximum
diameters were set to 23 mm and 64 mm, respec-
tively. This gives a cumulative weight distribution
curve as presented in figure 6. The particles were
loosely packed with a random distribution that
gives a void ratio of about 0.9.
Table 1. DEM particle properties.
ρ [kg/m ]
3 E [GPa] ν
2700 50 0.3
247
 Table 2. DEM particle interaction
properties.

μ Crr εn εt

0.7 0.2 0.8 0.8

Figure 7. Determination of friction angle by measuring

Figure 6. Cumulative weight distribution curve for the
angle of repose.
particles.

Imre et al. (2008) found a value for the CORs

of dynamic collision of granite with granite to
be around 0.8 and this value was adopted in this
work. The Rolling Friction Coefficient was set to
a “high” value of 0.2 to take into account that the
actual gravel particles were of non-spherical shape
(see equation 4).

3.2 Simulated odeometer stiffness

and friction angle
Two analyses were run in order to estimate proper-
ties of the discrete element assembly.
The first analysis involves a simple estimation of
the friction angle by measuring the angle of repose,
θ, under 1 g condition. Thereafter the assumption
of tanϕ = tanθ is used. Figure 7 shows such analy-
sis where the friction angle was determined to be
around 50° (tanϕ = 1.2).
The second analyses is an oedometer test sim-
ulation where a box of 400 mm by 100 mm by
400 mm was used. The elements were loaded on
top with an initial load of 10 kPa before a 500 kPa
stress increment was applied. Figure 8 gives the
results of the oedometer simulation. For the time
interval of 1.4 s, the oedometer stiffness can be
estimated to 18 MPa. The time used for the stress
increment was chosen such that the strain rate
before and after application of the 500 kPa was
the same.
According to equation 1 and 2, a penetration of
0.102 m and a maximum force of 384 kN (or an
estimated maximum deceleration of 192 g) for a
cushion layer of 0.6 m thickness are calculated.
Figure 8. Simulated oedometer results.
3.3 Modelling of impact
To model the impact on a cushion layer, a 2 m by
2 m box filled with 0.6 m thick gravel was used.
This resulted in about 50 000 particles. The parti-
cles were allowed to settle to rest under the action
of gravity alone. Then the boulder was dropped
with an impact speed of 14 m/s. The boulder was
modeled as a spherical object, with a radius of
0.26 m and weight of 200 kg.
The global time steps, Δt, during this analysis
were 1e-5 s/step. However, the program does sub-
stepping based on three criteria: The 1st criteria is
the time it takes the Rayleigh wave to propagate
across the surface of the sphere to the opposite
pole (for this the sphere with smallest radius in the
problem is used in the calculation), the 2nd criteria
is that the duration of a collision of two perfectly
elastic spheres should be done in a minimum of
10 steps, finally the 3rd criteria is that it should

248
take at least 10 time steps for the particle to move
the full length of the radius. The three criteria are
given in equation 5 to 7 below.

π ⋅ Rmin
Δtt1 = (5)
vRayleigh

⎛ 5 2 ⋅π ⋅ ρ 1− ν 2 ⎞ R
Δt2 = 0.1 ⋅ 2.94 ⋅ ⎜ ⋅ ⎟ ⋅ (6)
⎝ 4 E ⎠ vimpact
5

R
Δt3 = 0.1 ⋅ (7) Figure 10. Penetration depth versus time simulated
v particlel with DEM.

where R is radius, vRayleigh is the Rayleigh wave

velocity, vimpact is the velocity at impact and vparticle
is the particle velocity. The theoretical background
of equation 5 is found in Timoshenko (1951).
The penetration and deceleration of the boulder
was monitored during the analysis. The analysis
was run with an implicit solver. The simulations
were parallelized to 8 processes, which gives a rea-
sonable calculation time for 50 000 particles.

4 RESULTS OF THE DISCRETE ELEMENT

ANALYSIS
Figure 9 shows the final configuration of the dis-
crete element simulation of the impact. As can be
seen the boulder is about one radius into the cush-
ion layer. Slight heave of the cushion layer around
the boulder can also be noticed.
4.1 Penetration depth
Figure 10 shows the calculated penetration depth
in the DEM simulation where the final penetration
was 0.229 m. This is quite similar to the measured
penetration in the field test (i.e. 0.2 m). However,
as can be seen in the figure much of the penetra-
tion happens after the deceleration of the test was
finished (i.e. after approx. 0.02 s). This indicates
that much of the penetration in this simulation is
Figure 11. Deceleration simulated with DEM.
actually attributed to settlement due to the dead
weight of the boulder close to rest (static).
4.2 Deceleration history
Figure 11 shows the deceleration of the boulder in
the DEM simulation. As can be seen from the plot
is the calculated deceleration twice the measured
deceleration (Figure 5). However, when compared
to the estimated maximum deceleration using the
Swiss formula that gave 192 g, the values from the
DEM simulation are quite comparable.
5 DISCUSSION

The numerical simulation, with DEM, and the

measured values, in terms of deceleration were
quite different. Parts of the reasons for the differ-
ences can be attributed to discrepancies between
idealizations in the DEM simulation and the actual
test conditions, which can be briefed as follows;
− The cushion layer thickness was different. The
DEM simulations were done on the top 0.6 m
layer, in the test the cushion layer consisted of
two uncompacted layers with 0.6 m thickness
Figure 9. Final configuration of the DEM simulation. each.

249
− The test was done with a flexible culvert. The
idea with the flexible culvert is to reduce the
impact load. The numerical study support such
an effect. Hopefully further studies will support
this hypothesis.
− The shapes of the boulders in the tests were not
as spherical as in the DEM simulation. Degago
et al. (2008) gave attention to the shape effect
on penetration depth and maximum peak load.
They found that spherical objects give larger
impact forces and less penetration depths than
pyramidal shaped objects.
− For peak deceleration it is believed that the hori-
zontal distance to the boundaries has no effect
on the results. However, for final settlement and
the deceleration process a box of 2 m x 2 m is
small, both when it comes to reflections from the
boundaries and particles sliding/rolling along
the boundaries.
The effect of particle properties in the DEM
simulation, when it comes to both shape and mate-
rial properties, such as stiffness etc. should explain
some of the differences. Finally the DEM simula-
tion done in this study does not take into account
particle crushing.
6 CONCLUSIONS AND
RECOMMENDATIONS
This paper presents some initial studies to use
DEM to simulate rock fall onto a cushion layer.
There are several uncertainties in the DEM simula-
tions that should be studied further. However, the
paper demonstrates that DEM can be used to sim-
ulate rock fall onto a cushion layer in a dynamic
simulation.
The maximum deceleration found in the DEM
simulation is comparable with the one obtained
using the equation proposed by the guidelines
used in Norway and Switzerland. However, when
it comes to penetration depth the analysis gives
higher results. When DEM results are compared
to the field test differences are observed. This is
mainly due to the assumptions involved in ideal-
izing the test to DEM simulation.
The next steps in DEM simulation would be to
find improved parameters or models for the parti-
cles, to include non-spherical particles and particle
breakage and to investigate importance of parti-
cle parameters such as stiffness and coefficient of
restitution. All this can be done with the present
version of Star CCM+ software. It is worthwhile
to note that softer particles will increase the size
of the necessary time steps and hence potentially
decrease calculation time. However, non-spheri-
cal shapes (particle clusters) and particle break-
250
age will increase the demand for computational
resource.
The DEM simulations should be coupled with
the finite element or the finite volume method
for the simulation of the culvert and far field parts
of the model. This is not possible in the present
version of the software. However, it is in the future
plans of the software developers. For details on
such coupling one can see e.g. Rojek and Oñate
(2008). Such analysis will potentially provide a bet-
ter documentation on the effects rock fall on flex-
ible culverts and will potentially be used in design
of rock shield structures.
Finally back calculation of other drop heights
and boulder masses should be carried out. In the
tests mass up to 2650 kg and drop heights of 20 m
were used. Such large boulders with such high
kinetic energy would require that a large zone
should be modeled with DEM, again increasing
calculation time.
ACKNOWLEDGEMENTS
Authors are grateful to acknowledge the Norwe-
gian Public Roads Administration for financial
support for the experimental studies. The authors
also wish to thank the International Centre for
Geohazards (ICG) and the Research Council of
Norway for supporting this research.
REFERENCES
ASTRA, 2008. Einwirkungen infolge Steinschlags auf
Schutzgalerien. Richtlinie,ASTRA 12 006, V2.03.
Cundall, P.A. & Strack, O.D.L. 1979. A discrete numeri-
cal model for granular assemblies. Geotechnique, 29,
47–65.
Degago, Samson A. 2007. Impact tests on sand with
numerical modeling, emphasizing on the shape of
the falling object, Master Thesis, NTNU, Trondheim,
Norway.
Degagao, S.A, Ebeltoft, R. & Nordal, S. 2008. Effect of
rock fall geometries impacting soil cushion: A numeri-
cal procedure. The 12th International Conference of
International Association for Computer Methods and
Advances in Geomechanics (IACMAG), Goa, India.
Ebeltoft, R. & Larsen, J.O. 2006. Instrumentation of
buried flexible structure subjected to rock fall loading.
Joint international conference on computing and deci-
sion making in civil and building engineering. Montreal,
Canada.
Ebeltoft, R., Gloppestad, J. & Nordal, S. 2006. Finite ele-
ment analysis of buried flexible culverts subjected to
rockfall loading. Numerical Methods in Geotechnical
Engineering, NUMGE06, Graz, Austria.
Gupta, A. & Saigal K.R. 2003. Simple Formulation to
Evaluate Surface Impacts on Buried Steel Pipelines,
Vessel and Piping Stresses, Welding research Council
Bulletin no. 479.
Imre, B, Räbsamen, S. & Springman, S.M. 2008. A coef-
ficient of restitution of rock materials. Computers &
Geosciences, 34(4): 339–350.
Jacquemound, J. 1999. Swiss Guideline for the Design
of Rockfall Protection Galleries: Background, Saftey
concept and case histories, Joint Japan-Swiss Scientific
Seminar on Impact Load by Rock Falls and Design of
Protection Structures, Kanazawa, Japan, 95–102.
Kawahara, S. & Muro, T. 2006, Effects of Dry Density
and Thickness of Sandy soil on Impact Response due
to Rockfall, Journal of Terramechanics, 43: 329–340.
Kishi, N., Ikeda, K., Mikami, H. & Takemoto, S. 1999,
A proposed New Design Procedure for RC Rock-
Sheds, Joint Japan-Swiss Scientific Seminar on Impact
Load by Rock Falls and Design of Protection Struc-
tures, 103–112.
Labiouse, V., Descoueudres, E. & Montani, S. 1995,
Numerical Analysis of Rock Blocks Impact a soil Cush-
ion, Numerical Models in Geomechnics-NUMOG V,
Pande & Pieetruszczak (eds), 645–650.
Masuya, H., Tanak, Y., Onda, S. & Ihara, T. 1999,
Evaluation of Rock Falls on Slopes and Simulations
of the Motion of Rock Falls in Japan, Joint Japan-
Swiss Scientific Seminar on Impact Load by Rock
Falls and Design of Protection Structures, Kanazawa,
Japan, 21–28.
Montani, S.S., Labouse, V. & Descoeudres, F. 1999,
Action of Falling Blocks Impacting Rocksheds Cov-
ered with a Soil Cushion, Joint Japan-Swiss Scientific
Seminar on Impact Load by Rock Falls and Design of
Protection Structures, Kanazawa, Japan, 51–57.
251
Pichler, B., Hellmich, C. & Mang, H.A. 2005. Impact of
rocks onto gravel, design and evaluation of experi-
ments. International Journal of Impact Engineering
31(5): 559–578.
Plassiard, J.-P. & Donz´e, F.-V., 2009. Rockfall Impact
Parameters on Embankments: A Discrete Element
Method Analysis, Struct. Eng. Int., 19, 333–341.
Rojek, J. & Oñate, E., 2008. Multiscale analysis using a
coupled discrete/finite element model. Interaction and
Multiscale Mechanics, 1(1): 1–31.
SVV, 2011. Sikring av veger mot steinskred, Grunnlag for
veiledning, VD rapport, No. 32.
Timoshenko, S. 1951. Theory of Elasticity. McGraw-Hill
Book Company.
Volkwein, A., Schellenberg, K., Labiouse, V., Agliardi, F.,
Berger, F., Bourrier, F., Dorren, L.K.A., Gerber, W. &
Jaboyedoff, M. Rockfall characterisation and struc-
tural protection—a review. Nat. Hazards Earth Syst.
Sci., 11, 2617–2651.
Yoshida, H., Masuya, H. & Ihara, T. 1988, Experimen-
tal Study of Impulsive Design Load for Rock Sheds,
IABSE Proceedings P-127/88, 61–74.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Investigation into the factors affecting the shaft resistance of driven

piles in sands

D. Igoe, K. Gavin & L. Kirwan

School of Civil, Structural and Environmental Engineering, University College Dublin, Dublin, Ireland

ABSTRACT: The paper presents the results of field tests performed to study the effects of the installa-
tion technique, degree of plugging, cyclic loading and ageing on the shaft resistance developed on open-
ended piles in sand. Two instrumented model piles were jacked and driven into an artificially created
loose sand deposit in Blessington, Ireland. The results from these tests indicated that the equalized radial
effective stresses which are suggested to control the shaft capacity vary strongly with the degree of plug-
ging and number of load cycles experienced during installation. A comparison of jacked and driven
installations suggest similar radial stresses were developed provided the jacked pile had experienced a suf-
ficient number load cycles. The degree of plugging experienced during installation controlled the radial
stresses near the bottom of the pile, with closed-ended or plugged piles developing high stresses near
the pile base and exhibiting friction fatigue up the shaft, compared with open-ended coring piles which
exhibited relatively low stresses along the length of the pile shaft. A comparison with full scale 340 mm
diameter pipe piles driven into the dense sand in Blessington noted comparable radial stresses when the
pile was fully coring but exhibited a larger increase in radial stress near the pile toe as the pile became
plugged. Further tests on the effects of ageing show a pile shaft capacity increase of 260% over 220 days
after driving. Further research is underway in Blessington to investigate the mechanisms controlling this
ageing behaviour.

1 BACKGROUND In addition, the ICP tests suggested that the σ′rf

Significant advances in the design of axially loaded
piles in sand have been made over the past 20 years.
Field tests using instrumented piles and database
studies have shown traditional earth pressure based
design methods for predicting a piles axial capacity
to have poor reliability. Research campaigns at sev-
eral research centers around the world were under-
taken to better understand the mechanisms which
control this behavior. Some of the most important
insights into pile behavior were provided by Impe-
rial College (Lehane, 1992, Chow, 1997, Jardine
et al. 2005 and others) using a highly instrumented,
101 mm diameter, closed-ended model pile, known
as the Imperial College Pile (ICP). The pile, which
was jacked into both medium-dense and dense
sand deposits, used an array of instruments to
simultaneously measure the radial (horizontal)
stresses acting at the pile-soil interface and the
shear stresses at multiple levels along the pile shaft.
The model tests confirmed that the shaft friction
was controlled by the Mohr-Coulomb failure cri-
terion relating the ultimate unit shaft friction, τf,
to the radial effective stress at failure, σ′rf, and the
interface friction angle, δf, where:
τf = σ′rf ⋅ tanδf (1)
could be separated into two components namely;
the stationary radial effective stress measured after
installation before loading (σ′rc) and the increase in
radial effective stress occurs due to dilation during
loading (Δσ′rd).
σ′rf = σ′rc + Δσ′rd (2)
It was found that the σ′rc (which was suggested
as the primary component controlling τf) at each
level along the shaft clearly mirrored the CPT cone
resistance, qc, and reduced with distance from the
pile toe (h). This corroborated previous research
in clays by Hereema (1980), who introduced the
term friction fatigue to describe the reduction in
the mobilized shear stress developed in a given soil
horizon during driving, as the distance to the pile
toe increased. Chow (1997) postulated that (i) free
surface effects (ii) pile driving whip (iii) the geom-
etry of the steady flow around the pile tip and (iv)
the cyclic loading imposed by jacking (or driving)
were the mechanisms which principally caused
friction fatigue, with the last two being the most
important. Based on database studies of driven
piles which were seen to match quite reasonably
with the radial stress predictions from the jacked
ICP, Jardine et al. (2005) considered the geometri-

252
cal spreading and cyclic degradation (the primary
causes of friction fatigue) to vary principally with
h/R, where R is the pile radius, for closed-ended
piles.
Chow (1997) sought to extrapolate from the
closed-ended ICP tests to establish an approxi-
mate link which could be used for the design of
open-ended piles. It was suggested that using a
modified R* term, where R* is equal to the radius
of an equivalent closed-ended pile of the same
solid area, would result in a higher rate of friction
fatigue degradation which could be applicable for
open-ended piles. The aforementioned research led
to the development of the Imperial College Pile
design method (ICP-05) shown in equation 3:
σ′rc = 0.029 ⋅ qc ⋅ (σ′v0/Pa)0.13 ⋅ (h/R*)−0.38 (3)
where σ′v0 is the in-situ vertical effective stress
and Pa is atmospheric pressure (≈100 kPa). The
ICP-05 method includes an expression to predict
the stress increase caused by dilation, Δσ′rd, based
on the work of Lehane and Jardine (1994):
Δσ′rd = (4G/D) Δy (4)
where G is the operational shear modulus of the
soil (which can be correlated with CPT qc) and Δy
is the radial displacement occurring during pile
loading (≈ 0.02 mm for lightly-rusted steel piles).
The Δσ′rd component is inversely proportional
to pile diameter and as such was considered to be
relatively small for field scale piles. The method
for predicting local shear stress, which was seen to
be primarily controlled by the CPT qc and (h/R*)
components, was a major advancement over tra-
ditional design methods and has been shown to
offer significantly improved reliability in database
studies.
The extrapolation of the measurements of
radial stress on closed-ended piles to the design of
open-ended piles has caused much debate. Gavin
and Lehane (2003), Lehane et al. (2005), Igoe et al.
(2011) and others have suggested that using the
R* parameter to extrapolate to open-ended pile
behavior does not account for the complex con-
ditions such as soil plugging experienced during
the installation of an open-ended pile. Gavin and
Lehane (2003) suggested that an open-ended pile
would have a lower installation base resistance
(qb), which in turn would result in lower shaft
resistance. The qb was shown to depend on the qc
and the amount of plugging experienced during
installation as defined by the Incremental Filling
Ratio (IFR):
IFR = ΔLp/ΔL (5)
253
where ΔLp is the change in plug length for a given
change in pile penetration, ΔL (IFR = 0 means fully
plugged, IFR = 1 means fully coring). It was sug-
gested that using qb instead of qc as a design input
parameter in equation 3 (and using R instead of
R*) could provide a more logical extrapolation from
closed-ended behaviour. White et al. (2005) pre-
sented cavity expansion analyses which suggested
a weaker dependence on IFR than that proposed
by Gavin and Lehane (2003). The UWA-05 design
method was developed by modifying the original
ICP equations slightly, by dropping the vertical
effective stress function and using Ar,eff instead of
R* to account for open-ended behaviour, as seen
below:
σ′rc = 0.03 ⋅ qc ⋅ (h/D)−0.5 ⋅ Ar,eff0.3 (6a)
Ar,eff = 1 – IFR (Di/D)2 (6b)
where D is the external pile diameter and Di is the
internal pile diameter.
While both the UWA-05 and ICP-05 methods
offer comparably good performances in database
studies, Gavin et al. (2011) showed that when these
were used to design piles for a typical North Sea
oil platform, each method gave significantly differ-
ent pile length requirements. Recognizing that the
ICP-05 and UWA-05 methods for open-ended pile
design were fundamentally based on the results of
instrumented closed-ended model piles and which
were subsequently adapted and calibrated against
databases of field scale open-ended piles, the
UCD Geotechnical Research Group conducted a
prolonged research campaign to further improve
understanding of open-ended pile behavior.
2 EXPERIMENTAL DETAILS
2.1 UCD instrumented piles
Model open- and closed-ended piles were used
in the tests described in this paper. The stainless
steel open-ended pile had an external diameter of
168 mm and a wall thickness (t) of 9 mm, giving a
D/t ratio typical for pipe piles. The lower instru-
mented section, 2 m in length, was constructed
using the twin-walled technique, where two strain
gauged steel pipes with slightly different diameter
are joined (at the top) thus allowing separate meas-
urement of the internal and external shaft resist-
ance and the base load developed on the annulus of
the pile toe itself. Three instrumented units which
housed the radial total stress sensors and pore
pressure transducers were located at h/D = 1.5,
5.5 and 10.5. Spacer sections machined from the
same pipe, which were 1.0 m in length and of equal
diameter, could be used to allow pile installation to
6 m. The stress sensors were TML (Tokyo Sokki
Kenkyujo Co.) PDA-500 kPa, and had a rated
capacity of 500 kPa. The same sensors were used
to measure pore pressure response (located diamet-
rically opposite the total stress sensors). A porous
ceramic disc was mounted flush with the outer pile
surface in front of each of the pore pressure sen-
sors. The location of the instrumentation, includ-
ing the electrical resistance strain gauges are shown
in Fig. 1. Full details of the design, construction
and calibration of the model pile are contained in
Igoe et al. (2010).
The model closed-ended pile used in the field
tests was a 73 mm, stainless steel pile with radial
total stress and pore-water pressure sensors at the
same relative locations (h/D = 1.5, 5.5 and 10.5) as
the open-ended piles and a similar level of overall
instrumentation. This instrumentation is described
in detail by Igoe et al. (2010).
In addition to the model piles, a larger scale,
340 mm diameter, 7 m long, steel, open-ended
pile was instrumented with four radial total stress
sensors at relative locations (h/D = 1.5, 5.5, 10.5
and 17.5). The sensors used were the same make/
type as those used in the model piles only with a
higher rated capacity of 3 MPa (to survive driving
and denser soil conditions). Details of the design
Figure 1. UCD open-ended model pile (after Igoe et al.,
2011).
254
of the pile have been described by Kirwan et al.
(2012). For continuity with previous publications
the open—and closed-ended model piles are des-
ignated pile OE and CE respectively while the full
scale pile is designated pile S5.
2.2 UCD test site
The experimental pile tests were performed at the
University College Dublin (UCD) sand test site
which is situated in Blessington, 25 km southwest
of Dublin city. The ground conditions comprise
of glacially deposited, very dense, fine siliceous
sand with a CPT qc resistance in the range of 15 to
20 MPa, and small strain stiffness (G0) in the range
75 to 150 MPa. Sand replacement tests indicated
that the relative density of the in-situ sand deposit
was between 90% and 100%. Particle size distri-
bution analyses performed on samples indicated
that the mean particle size, D50, was approximately
0.15 mm. The equilibrium ground water table was
approximately 13 m bgl. The natural water content
was relatively uniform at 10–12% above the water
table and was found not to vary seasonally.
To allow for comparison of pile behaviour in
conditions where only the sand state varied, a
trench 2.5 m wide, 10 m long and 6 to 7 m depth
was excavated at the site. Immediately following
excavation the material was backfilled into the
trench at its natural moisture content (10–12%)
using an excavator bucket maintaining a drop-
height of between 1 and 2 m, with the intention
that this air pluviation technique would produce
a uniform loose sand deposit. Four CPT profiles
were performed at 2 m spacings along the trench.
The CPT qc profiles are compared with profiles
measured in the in-situ, dense sand in Fig. 2a. The
CPT profiles were consistent at all locations in
the trench, with qc decreasing from 2 MPa near the
ground surface to 1.0 MPa at depths below 1.5 m
bgl until the cone reached the base of the trench
at 6–7 m bgl. This compared to CPT qc values of
15–20 MPa in the dense in-situ soil over this depth
range. The friction ratios (Fr = ratio of CPT sleeve
friction fs to cone resistance qc) were similar at
1–2% in the loose and dense sand deposits. Profiles
of the small strain shear modulus, G0, determined
by Multi-channel Analysis of Surface Waves
(MASW) are shown in Fig. 2c. The G0 values in
the trench increased from 20 MPa near the ground
surface to 40 MPa at depth.
2.3 UCD pile testing program
The model pile tests described in this paper have
been described in further detail in Igoe et al. (2011).
All model pile tests were conducted in the loose
backfilled trench at Blessington. The piles were
Figure 2. CPT profile at Blessington.
installed through initial starter holes, varying from
0.85–1.9 m in depth. The first test (OE1) involved
jacking the open-ended model pile into the sand,
using the CPT truck as reaction, in 100 mm jacking
strokes to a final depth of 5.9 m. The installation
was paused between each jacking stroke to moni-
tor the plug development and record the stationary
radial stresses. The second test (OE2) involved driv-
ing the pile to a depth of 4.1 m, using a specially fab-
ricated 25 kg hammer falling through a drop-height
of 0.5 m. The pile was jacked a further 1.45 m into
the ground using the CPT truck with 50 mm jack-
ing strokes. The third test involved the installation
of the 73 mm diameter closed-ended model pile
(CE), which was installed using the CPT truck in
50 mm jacking strokes to a final depth of 3.55 m.
At the end of installation for each model test (OE1,
OE2 & CE) a series of cyclic tests were performed
to assess the effects of cyclic loading on the radial
stresses. A final pile test involved driving the full
scale pile S5 into the dense natural sand deposit
using a Junttan PM16 pile driving rig to a final
depth of 7 m bgl. Driving was paused every 1 m to
assess the IFR and stationary radial stresses.
3 EXPERIMENTAL RESULTS
During installation of the model piles, the station-
ary radial effective stresses (σ'rs), which were equiva-
255
lent to the short term equalized radial stresses (σ'rc),
were measured between jacking strokes. The jacked
closed-ended model pile (CE) developed large σ'rs
values near the pile toe during installation, and
exhibited friction fatigue with much smaller values
noted further up the pile shaft. Similar behaviour
was noted for the jacked OE model pile when it
was experiencing full plugging during installation.
The driven OE model pile, which was virtually
fully coring throughout installation (IFR > 80%),
exhibited very little friction fatigue, with low σ'rs
values noted along the entire shaft. Compression
cyclic load tests performed on the CE and OE
jacked piles showed a significant reduction in σ'rs
over the first 20–50 load cycles, reducing to values
similar to those measured on the driven model pile
(at sensor locations remote from the pile toe). The
combined effect of plugging and cyclic loading on
the model piles is shown in Fig. 3. It is evident that
for closed-ended piles or fully plugged open-ended
piles (IFR = 0) that the σ'rs values near the pile toe
(at h/D = 1.5) are significantly higher than further
back up the shaft, even after 100 or more cycles (i.e.
friction fatigue is evident). This is related to high
stress mobilized in the vicinity of the pile toe and
the geometry of the steady flow around the pile tip.
For a fully coring pile (IFR = 100%), much lower
stresses are noted at h/D = 1.5 and these are in line
with the minimum threshold values further up the
pile shaft (i.e. no friction fatigue is evident).
 ICP-05, UWA-05 and UCD-11 design methods
compared to the radial stresses measured from
the full scale pile, S5 driven into Blessington. Due
to damage to two of the sensors during driving,
radial stress data is only available from sensors at
h/D = 1.5 and 5.5. The IFR during installation of
pile S5 reduced steadily from 95% over the first 1 m
of penetration to 45% at the end of installation. It
is evident that ICP-05 and UWA-05 methods do
not capture the full effects of plugging near the pile
toe. The UCD-11 method predicts a larger range

Figure 3. Effect of cycling on stationary radial effective

stresses along the pile shaft (after Igoe et al. 2011).

The effect of cyclic loading is evident (repre-

sented by the arrows in Figure 3) and shows that
jacked piles, which experience a small number of
cycles during installation, can develop significantly
higher stationary radial stresses during installation
than driven piles or piles which have experienced a
large number of load cycles. However it should be
noted that the benefits of these high radial stresses
can be removed after only a small number of load
cycles.
From this research, Igoe et al. (2011) proposed
a new design method, referred to as UCD-11, in
order to predict σ'rc for the end of driving condi-
tion (after many cycles had been experienced):

σ′rc = qc ⋅ [0.025 – 0.0025(h/D)] ⋅ Ar,eff > σ′rc,min (7a)

σ′rc,min = γ ⋅ qc (7b)

where γ accounts for the minimum threshold value

which was thought to depend on the sand state.
Based on the limited available data, tentative val-
ues of γ = 0.003 and 0.006 were proposed for loose
and dense sand, respectively, with higher values
possible in very dense over-consolidated deposits.

4 COMPARISON TO FULL SCALE PILES

Figure 4. (a) Comparison of UCD-11 and (b) UWA-05
Figure 4 shows a comparison of the equalized and ICP-05 design methods against measured radial
radial effective stresses determined using the stresses from full scale pile S5.

256

Figure 5. Ageing tension static load tests on previously
untested piles full scale piles at Blessington (after Gavin
et al., 2013).
of σ′rc values near the pile toe for the varying IFR
values but over predicts the σ′rc at high IFR values
provided γ = 0.006 is used (as suggested for dense
sands). From the limited available data it appears
a γ = 0.003 may be applicable to predict the end of
installation radial stresses in both loose and dense
sands, however this research is still ongoing and
further testing using a new instrumented pile is cur-
rently underway to verify this. The effect of ageing
or pile ‘set-up’ also has a very large influence on
the pile capacity. Tests conducted at Blessington by
Gavin et al. (2013) on three piles (identical to pile
S5) show the effect of ageing on the pile capacity.
The piles were load tested in tension at differ-
ent time intervals after driving. It is evident from
Fig. 5 that a 260% increase in pile capacity was
noted over the 220 days after driving. Jardine et al.
(2005) included the Intact Ageing Characteris-
tic in the ICP-05 design method which suggested
similar increases in capacity over time. Axelsson
(2000) measured the radial stresses on precast con-
crete driven piles in sand and suggested that the
effect of pile ageing was primarily due to enhanced
constrained dilation at the pile soil interface. Fur-
ther research is required to investigate the ageing
phenomenon.
5 CONCLUSIONS
The paper presents field data from model and full
scale open—and closed-ended piles jacked and
driven into loose and dense sand deposits. Recent
CPT based methods which were developed to bet-
ter capture the mechanisms controlling pile behav-
ior were discussed. Radial stress measurements
from a full scale open-ended driven pile in Bless-
257
ington show that while recent CPT based design
methods provide significantly improved radial
stress predictions, they are still unable to capture
the full combined effects of installation plug-
ging, cyclic loading and ageing. Further research
is being undertaken at UCD using a new highly
instrumented full scale pile to reliably measure the
radial and shear stresses during installation, ageing
and load testing.
REFERENCES
Chow, F., Jardine, R.J., Brucy, F. and Nauroy, J.F., 1997.
“Time related increases in the shaft capacities of
driven piles in sand,” Geotechnique, Vol. 47, No. 2,
pp. 353–361.
Gavin, K.G. and Lehane, B.M., 2003. “The shaft capac-
ity of pipe piles in sand,” Canadian Geotechnical Jour-
nal, Vol. 40, pp. 36–45.
Gavin, K.G., Igoe, D. and Doherty, P., 2011. “Use
of open-ended piles to support offshore wind tur-
bines: A state of the art review,” Proceedings of the
ICE—Geotechnical Engineering, Vol. 164, No. GE4,
pp. 245–256.
Gavin, K.G, Igoe, D. and Kirwan, L., 2013. “The effect
of ageing on the axial capacity of piles in sand”, ICE
Geotechnical Engineering Special Edition, In Press.
Heerema, E., 1980. “Predicting pile driveability: heather
as an illustration of the friction fatigue theory,”
Ground Engineering, Vol. 13, pp. 15–37.
Igoe, D., Doherty, P. and Gavin, K.G., 2010. “The devel-
opment and testing of an instrumented open-ended
model pile,” Geotechnical Testing Journal, Vol. 33,
No. 1, pp. 1–11.
Igoe, D., Gavin, K.G. and O’Kelly, B.C., 2011. “The shaft
capacity of open-ended piles in sand,” Journal of Geo-
technical and Geoenvironmental Engineering, ASCE,
Vol. 137, No. 10, pp. 903–913.
Jardine, R.J., Chow, F.C., Overy, R.F. and Standing, J.,
2005. ICP Design Methods for Driven Piles in Sands
and Clays. T. Telford. London, University of London
(Imperial College).
Jardine, R.J. and Chow, F.C., 2007. “Some Recent Devel-
opments in Offshore Pile Design,” 6th International
Offshore Site Investigation and Geotechnics Confer-
ence, London.
Lehane, B.M., 1992. “Experimental investigations of pile
behaviour using instrumented field piles,” PhD Thesis,
University of London (Imperial College).
Lehane, B.M. and Jardine, R.J., 1994. “Shaft capacity of
driven piles in sand: a new design approach,” Confer-
ence on the Behaviour of Offshore Structures, Boston,
Mass.
Lehane, B.M., Schneider, J.A. and Xu, X., 2005. “The
UWA-05 method for prediction of axial capacity of
driven piles in sand,” Frontiers in Offshore Geotech-
nics: ISFOG, Perth, University of Western Australia.
White, D.J., Schneider, J.A. and Lehane, B.M., 2005.
“The influence of effective area ratio on shaft friction
of displacement piles in sand,” Frontiers in Offshore
Geotechnics, ISFOG, University of Western Australia,
Perth.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Monitoring and risk assessment in EPB TBM’s in urban environments:

High speed railway tunnel Sants-Sagrera running next to Sagrada
Familia Basilica (World Heritage)

J.E. París Fernández & J. Gómez Cabrera

Sener Ingeniería y Sistemas, S.A., Barcelona, Spain

ABSTRACT: In this paper we are going to summarize the procedures used to manage the operation
of a TBM (Tunnel Boring Machine) type EPB (Earth Pressure Balance) with continuous and “real time”
monitoring of working and parameters. To show the characteristics of this method we chose a relevant
example, Sagrada Familia Basilica designed by Antoni Gaudi, which is a UNESCO World Heritage Site.
Some control parameters have been followed in order to detect possible damage to the building structure.
During this time, Sener Ingenieria y Sistemas supported the project with direct monitoring and risk as-
sessment to the TBM, reaching an advanced level in order to detect anomalies in the TBM operation,
ground movements or building failure.

1 INTRODUCTION

During the last decade the use of EPB TBM’s to

create underground communication networks in
our cities has grown considerably. In parallel, the
risk of an accident or incident associated with
underground activities has increased because of the
very sensitive environment. When talking about a
world architectural symbol like the iconic Sagrada
Familia Basilica by Antoni Gaudi, the red alarm
light flashes immediately. For this reason, continu-
ous monitoring and follow up to the TBM’s opera-
tions are done to give a high grade of confidence
about the work undertaken.

1.1 The scenario

One of the most recognizable symbols of Barcelona
is the Sagrada Familia Basilica, located between
Sardenya and Marina Streets looking from west to
east. From north to south the location is between
Provença and Mallorca Streets (see Figure 1). The
high-speed railway tunnel runs along Mallorca Figure 1. Sagrada Familia location and tunnel position.
Street affecting principally the south façade of the
Temple from P.K. + 3.650 to P.K. + 3.540, as ref-
erenced in the Construction Project. In terms of at a depth of 18 to 20 m. In the location that is the
excavated rings, the affected zone is approximately focus of this article the principal component of the
between ring 950 and ring 1050. soils is sand with random layers of brown argil-
laceous marlstones.
1.2 The geology A pile wall was driven all along Mallorca Street,
in order to give protection to the foundations and
The geology of the tunnel route is principally structure of the historic building. It is possible to
quaternary materials comprising a mix of clay, see a descriptive section in Figure 2. The pile wall
sand, gravel and lime. The water table is located was completed with three concrete beams and soil

258

Figure 2. Typical cross section showing tunnel location
with respect to the temple.
treatment in the surface. During the execution of
the pile wall the settlements detected in the Temple
were around 2 mm.
1.3 The construction project
This urban tunnel is part of the High Speed Railway
Line Madrid-Zaragoza-Barcelona-France border
and it crosses Barcelona city. The total length of the
track is 4849 m and the average cover around 28 m.
The geometry design parameters are as it follows:
− Minimal free surface: 85 m2
− Internal radius: 5.2 m
− Free height of the section: 7.36 m
− Cover in considered section: 25.5 to 26.5 m
Taking account of the geotechnical properties of
the excavated material, it was decided to perform
the job using an EPB (Earth Pressure Balance)
TBM (Tunnel Boring Machine) manufactured
by Herrenknecht, with an excavation diameter of
11.55 m, 105 m length, 2300 ton weight and named
“Barcino”.
A very complex monitoring system for the struc-
ture was installed. Level sensors were placed at dif-
ferent heights, and extension gauges, piezometers,
inclinometers and a robotic survey complete a huge
network that can detect any single movement.
2 EPB TBM WORKING OPERATION
AND MONITORING
2.1 EPB working operation
An EPB TBM was operated with continuous front
support using the excavated material to keep the
259
pressure inside the excavation chamber. To regulate
the pressure and the amount of material inside, a
screw conveyor is used with two relevant param-
eters: the rotation speed and the gate aperture
percentage. Additional to these parameters is the
treatment of the excavated soil by foam and differ-
ent kinds of polymers.
When an EPM TBM is working, the operator
always tries to find a balance between the material
entering the chamber, the screw conveyor rotation
and the aperture of the gate at the end of the screw
conveyor. This state is very difficult to reach in
mixed soils with different densities and behaviors.
It is accepted that the safest way to operate an EPB
is with the chamber full of material, but sometimes
this is not possible. Think, for example, of a mixed
ground formed by sand and clay. The friction of
the clay against the internal wall of the chamber
raises the temperature and it is necessary to cool it
down through injecting water that can wash away
the sand very fast, making a chaotic scenario of
densities inside the chamber.
A critical moment for the front support occurs
when the TBM must stop. There are different
causes. During normal operation for a single shield,
when the stroke of the hydraulic jacks reach the
end a new ring has to be installed. At this moment
the cutter head stops its rotation, the gate of the
screw conveyor is closed and some excavated mate-
rial remains in the chamber. To prevent unexpected
loss of pressure during the ring installation, the
TBM has been equipped with a bentonite injection
system which is prepared to actuate if the pressure
drops below the reference level. The engineer takes
care during this moment and, in the case of exces-
sive pressure loss, informs the operator to turn on
the bentonite system. In this way, during stops it is
possible to maintain the pressure.
Another important issue is the hyperbaric oper-
ation. Human intervention at the front is one of
the most dangerous and complex situations that
can happen during tunnel driving. The different
phases to form the bentonite “cake”, that gives the
stability to the ground and allows work inside the
chamber, must be followed and monitored with
special attention. Also, during long time mainte-
nance stops the bentonite “cake” can modify the
surrounding ground behaviour.
2.2 Monitoring EPB operation
The complexity of the operations requires a big
control of parameters, in some cases around 200.
The monitoring system hereby proposed and
described is not only oriented to the operator or
the Contractor. The system gives support to all
parties implicated in the job from a global and
dedicated point of view.
Figure 3. Scheme of monitoring and data acquisition
in real time.
The principle of the system is taking data from
the EPB computer and, at the same time, data from
the monitoring net placed along the track. All the
data collected in one control center are analyzed
in “real time” by an expert team that has direct
contact with the TBM’s operator. The scheme is
shown in Figure 3.
In the first instance, the system could look very
simple, but there some characteristics that make it
very efficient. These characteristics are:
− Total independence of the monitoring team from
the Owner and Contractor. This fact avoids sub-
jective decisions.
− Global vision in “real” time of the TBM
operation.
− Possibility to improve TBM performance by
analyzing all the data collected.
− Continuous monitoring 24/7 and specially
dedicated surveillance during the hyperbaric
interventions.
3 MONITORING DURING CROSSING
UNDERNEATH SAGRADA FAMILIA
Considering the importance of the Temple and
the evident risk of damage of its structure, special
attention was focused from ring 950 to ring 1050.
This does not mean that for the rest of the drive
less attention was paid, but this part really has a
special relevance. To better understand the work
executed we are going to reduce the study field
from ring 970 to 990.
3.1 Excavation chamber
We can divide the excavation chamber into different
parts as a function of the height. To determine the
pressure at those heights the bulkhead is equipped
with several pressure gauges (see Figure 4).
260
Figure 4. Pressure sensor distribution inside the excava-
tion chamber.
Figure 4 shows the sensor configuration and
the optimal values of pressure considered for this
interval of rings. The caption is directly taken
from a tool used by the monitoring engineer to
determinate if there is an air bubble in process. In
“Level 1”, at the top of the chamber, the ground
cell shows a value of 1.4 m3 and the cell “Air Bub-
ble” 0.15 m. This means that, at this moment and
with these working conditions, the air bubble in the
top of the chamber is reduced to 15 cm and, in
terms of EPB operation, completely depreciable.
Measuring the difference of pressures between the
sensors (the height difference is known) it is pos-
sible to calculate the density at every instant in
the zone under consideration. In this case we are
going to pay special attention to the sensors P1 and
P2-P3 that control the top of the chamber.
3.2 Monitoring FER (Foam Expansion Rate)
The optimal working condition considers a density
of 1.4 t/m3 in the top of the chamber. In section 2
it was mentioned that the ground treatment was by
foam and polymers. The purpose of this treatment
is to get a better mixture of the material inside the
chamber, but the excess of compressed air used to
inject this component into the chamber can accu-
mulate in the top of the chamber producing an air
bubble, around 15 m3 over the sensors P1-P2P3, a
situation that is completely unacceptable for safe
operation. To correct this excess of air, the bulk-
head is equipped with a vent valve in the top, just
over the sensors level P1-P2P3. When the moni-
tored parameters indicate the presence of air, the
vent valve is immediately opened, evacuating the
air and filling the empty space in the chamber with
the injection of bentonite as a face support until the
conditions of the excavation allows the top to be
filled with excavated material. Of course, the effect
of the purge is not immediate; the shift monitoring
engineer calculates the amount of material needed
to refill the chamber and, in the case of the density
not recovering in a reasonable stroke length, the
engineer informs the EPB operator.
Taking into account that the scales are not the
same in Figure 5, we can check the importance
of a good soil treatment in terms of operational
safety. It is represented by the density measured
in the top of the chamber together with the FER
(Foam Expansion Rate). FER is a parameter that
gives a good idea about the quality of the foam
used at each moment. The value comes from the
expression:
V compressed air
E =
FER (1)
V foam solution
where:
− Vcompressed air indicates the volume of compressed
air injected
− Vfoam solution indicates the total volume of foam
obtained.
A normal working ratio of FER is between 8
and 20, but these values must be adapted to the
geological conditions every time. For example, a
saturated soil needs dry foam which will be indi-
cated with a high FER value. On the other hand, a
lower expansion rate is normal for wet foams.
The dashed line represents the optimal density
for operation with minimum risks. It is clearly
demonstrated that, when the FER increases, the
densities in the top of the chamber decrease with
Figure 5. Graphical representation of FER (Foam
Expansion Rate) vs rings excavated. Density P1-P2P3 is
represented as a reference.
261
consequent formation of an air bubble. After con-
veying the situation to the operator the vent valve
starts to drain the air and the FER is corrected. Of
course, this operation is not a matter of seconds; it
takes at least, in the case shown in the figure, eight
rings (ring 976 to ring 984) excavated to recover the
right density. Eight rings means, in terms of dis-
tance, 32 m. This is one of the principal reasons for
this dedicated survey. Without noticing this situa-
tion the distance could have been 2–3 times more,
causing an enormous risk of suffering a “blow-
out” in the ground.
3.3 Monitoring densities
A good tool for the monitoring engineer is to com-
pare the densities between the different levels in the
chamber. If we pay attention to Figure 6, which is
representing the density at two levels of the cham-
ber, it is possible to recognize an anomaly after ring
978; the density at the medium level remains around
the same value without significant variations and
the density in the top of chamber (P1-P2P3) goes
down. This is a signal that something is happen-
ing at the face; a soil change, or, in the TBM, a
mechanical problem or a bad operation.
3.4 Monitoring bentonite injection
Another parameter to take account of to control the
presence of an air bubble in the top of the chamber
is the bentonite injection through the shield. The
function of this injection is to fill the gap gener-
ated by the cutter head. This gap could be suscep-
tible to filling with air, causing a risk just behind
the cutter head and before the installation of the
Figure 6. Graphical representation of the two levels of
density inside excavation chamber.
concrete ring. It is a volume without any apparent
support and means a big risk of settlements in the
surface. The injection of bentonite must create a
pressure balance with the material inside the cham-
ber; in fact, if the top part of the chamber is empty
the bentonite without pressure will displace to the
front filling the empty space. In Figure 7 we can see
a good example.
Using the same ring interval as before and
comparing the cubic meters of bentonite injected
through the shield with the density in the top of
the chamber, we can observe that when the density
in the top of the chamber decreases, the amount
of bentonite injected rises dramatically to 4.6 m3,
ring 980, before going down to around 1.0 m3 when
the density reaches again the optimal balance. In a
good operation the injection is reduced to a mini-
mum volume, which is needed to fill the gap and
give a certain pressure to it. From ring 981 to 987
the volume of bentonite is more or less constant
around 1 m3 and the density is maintained at over
1.4 t/m3 with a good margin of reaction. In ring
987 it comes back to the same situation. The FER
rises very fast (Figure 5), the density drops off
and the volume of bentonite injected is doubled.
This example focuses on a few rings demonstrat-
ing that these three parameters are directly linked
and that a good interpretation can solve dangerous
situations.
3.5 Monitoring the extraction of material
The next parameter which focuses the attention of
the engineer in charge is the amount of material
extracted from the chamber. For each part of the
tunnel a theoretical quantity is calculated based on
Figure 7. Graphical representation of bentonite vol-
ume injection.
262
the ground composition and TBM characteristics.
For the Sagrada Familia crossing, a theoretical
weight of 368 tons extracted per ring excavated
was calculated.
Of course there is an interval of admittance in
the weight. The upper alarm limit was established
as 405 tons and the lower alarm limit as 331 tons.
These two limits are represented in Figure 8.
Coming back to ring 980, the density in the top of
the chamber goes down and the extracted amount
of material shares this tendency (this is also possi-
ble to detect two rings before). The reason for this
behavior is the necessity to refill the chamber with
material after the wrong treatment occurred in ring
978 (FER rises to 15). The TBM operator receives
from the engineer advice that an air bubble is in
progress and immediately two actions take place:
− The volume of bentonite is increased to help
during the refilling of the chamber with exca-
vated material.
− The extracted material reduces its amount oper-
ating over the screw conveyor. In this case the
rotation speed is reduced in order to keep the
material in the chamber. See Figure 9.
The fast detection of an air bubble progress
makes changes in the density possible in only three
rings of the advance using a combination of oper-
ations in the TBM:
− Using the vent valve to drain the air retained in
the top of the chamber
− Filling the empty space with bentonite to com-
pensate for the absence of material
− Controlling the extracted material with the oper-
ation of the screw conveyor.
Figure 8. Graphical representation of the material
extracted from the chamber.

Figure 9. Graphical representation of screw conveyor
rotation speed.
3.6 Results obtained with continuous monitoring
An air bubble in the excavation face could be a
big problem in the TBM operation, especially in
urban environments like in this case. From inside
the TBM, this is not easy to detect due to the
complexity of working with this kind of machine,
but, from an external point view, collecting the
data in real time and analyzing the situation with
dedicated personnel and tools is possible to offer a
more effective risk control management. Also, we
have to take account that it is not necessary to have
a massive soil failure to create a dangerous situa-
tion in the surface and in the tunnel. With a small
settlement a structure can suffer irreversible dam-
age, and not only in the surface, because today our
city undergrounds are saturated in services, metro
lines, sewages lines, storm channels, etc. With con-
tinuous monitoring by specialized personnel, it is
possible to avoid these settlements and get a better
performance of the TBM.
If we take a look to the final numbers, the result
is that the settlements obtained in the crossing
underneath. The Temple are lower than 2 mm, five
times less than the project calculation. Considering
the diameter of the EPB, 11.55 m, the soil condi-
263
tions, sand with clay layers and saturated in water,
we can conclude that this kind of monitoring is an
essential tool to avoid the associated risks with this
kind of job.
4 CONCLUSIONS
“Real time” monitoring improves risk assessment
in TBM operations, to avoid dangerous situations
involving structures and people. The supervision
of the different parameters by a dedicated and
specialized team of engineers makes it possible to
detect the anomalies and to correct these imme-
diately and with good practice. All scenarios are
taken into account, ranging from bad practice of
the TBM operator to a geological change.
The results obtained in Sagrada Familia and
in the rest of the drive reinforced the needed of
monitoring, not only to avoid the essential risk of
tunneling practice, but also to improve the TBM
performance through advising about soil treat-
ments and operation procedures.
REFERENCES
Anagnostou, G & Kovari, K. 1994. Stability analysis for
tunneling with slurry and EPB shields. Mir 94 “Gal-
lerie in condizioni difficilli”, Torino.
Gómez, J. 2009. Excavation monitoring in tunnels exe-
cute by EPB. SENER’s experience in Oporto and Lis-
boa metro lines. Obras Urbanas. Febrero 2009, nº 13.
Gómez, J. 2011. Follow up and risks control in TBM EPB
operation in “real time”. Sants-Sagrera Tunnel. I Foro
Internacional Ferroviario. Bcn Rail 2011, Barcelona.
Gomez, J. & Roldan, J. 2012. Follow up and risk assess-
ment in EPB TBM operations in urban environments.
Sants-Sagrera tunnel crossing underneath Sagrada
Familia Temple. Second Colombian and First Andean
and Central American Congress and Exhibition of
No—Dig Technologies and Underground Infrastructure
2012, Cartagena de Indias, Colombia.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Effect of roughness on keying of plate anchors

D. Wang, C. Han & C. Gaudin

Centre for Offshore Foundation Systems, The University of Western Australia, Crawley, WA, Australia

ABSTRACT: Suction Embedded PLate Anchors (SEPLAs) are a relatively new type of anchorage in
deep waters. The SEPLA is inserted vertically into clayey seabed and then pulled to rotate until it becomes
nearly perpendicular to the loading inclination. The keying response has been studied by means of centri-
fuge tests and large deformation finite element (LDFE) analyses. However, the predicted ultimate losses
of embedment were lower than the experimental results in most scenarios. The effect of anchor rough-
ness on the keying process is then investigated using LDFE approach with contact algorithm. A friction
coefficient of 0.3 in typical normally consolidated kaolin clays is determined by comparing the numerical
results with three groups of experimental data measured. The loss of embedment during keying depends
on the combined influences of anchor roughness, anchor thickness and loading eccentricity. When the
anchor thickness ratio is less than 0.7 and the loading eccentricity ratio not larger than 0.5, the anchor
roughness needs to be considered.

1 INTRODUCTION

The over-consumption of hydrocarbon reserves

has forced a transition of offshore explorations
from shallow waters to deep and ultra-deep waters
up to 2000 m, which necessitated the replacement
of traditional fixed platforms by floating facilities
moored to the seabed. Suction Embedded PLate
Anchors (SEPLAs), as shown in Figure 1a, are one
of recently developed anchoring systems, which
are increasingly used in the Gulf of Mexico and
West Africa due to their advantages of low cost,
accurate positioning and short installation time.
The plate anchor is normally rectangular in
shape, and is initially slotted vertically inside a
suction caisson. The suction caisson is penetrated
to the target depth and then retrieved, with the
SEPLA being left in the soil. The mooring chain
attached to the padeye is tensioned and the anchor
is led to rotate until an inclination approximately
perpendicular to the chain direction. The installa-
tion process is termed keying. During the keying
process, the SEPLA tends to translate upward,
which induces a loss of embedment referencing to
the original centroid of the plate top. For seabeds
comprised of normally consolidated clay, with
strength increasing with depth, the loss of embed-
ment results in the reduction of anchor pull-out
capacity. Therefore, it is crucial for routine design
to predict the embedment loss during keying
installation. Figure 1. Geometry of SEPLAs.

264
 The embedment loss of SEPLA was firstly
explored through in-situ and centrifuge model tests
(Wilde et al. 2001, O’Loughlin et al. 2006, Gaudin
et al. 2008, Gaudin et al. 2010). The disadvantage
of the in-situ test is that the evolved inclination and
embedment depth of SEPLAs cannot be measured
in a straightforward and accurate way. In the centri-
fuge tests using the advanced Particle Image Veloci-
metry (PIV) technique (White et al. 2003), the soil
flow mechanism around the keying anchor was cap-
tured and the trajectory of the anchor was tracked
continuously and reliably (O’Loughlin et al. 2006).
To implement the PIV technique in the centrifuge, a
transparent side is set for the chamber containing soil
sample, and both ends of the anchor are placed in
contact with the sides of the chamber, i.e. the anchor
is under plane-strain condition. The pull-out resist-
ances measured in the PIV tests are not reliable due to
the friction of the chamber sides. However, the posi-
tions of anchor are quantified to demonstrate the his-
tories of the anchor inclination and embedment loss.
Apart from the model tests, the keying processes
of strip and rectangular SEPLAs were investigated
by Song et al. (2009) and Wang et al. (2011) using a
large deformation finite element (LDFE) approach
named ‘remeshing and interpolation technique
with small strain’ (RITSS). The RITSS approach is
based on frequent mesh regeneration to overcome
mesh distortion around the keying anchor. Two
simple equations were presented to predict the ulti-
mate loss of embedment through a large amount
of parametric studies varying the anchor geom-
etry, soil strength profile and loading eccentricity
ratio. However, the SEPLAs modelled in these
large deformation analyses were assumed as fully
rough, i.e. no relative tangential movement between
the anchor and ambient soil was permitted. The
embedment loss tends to be underestimated, which
is on the unsafe side for practical applications.
In this paper, the influence of anchor rough-
ness on the keying response and embedment loss
is studied using two-dimensional RITSS approach.
An appropriate range of friction coefficient is sug-
gested by comparison with previous vertical pull-
out tests using the PIV. The inclined pull-out tests
were abandoned since the chain-soil interaction
would introduce more complexity on determin-
ing the embedment loss. The focus of this paper
is limited to the roughness of anchor, which will
be pursued by offsetting the uncoupled factors.
The combined effect of anchor roughness, anchor
thickness and loading eccentricity ratio on the ulti-
mate embedment loss is also studied.
2 METHODOLOGY
The RITSS approach can be regarded as a special
Arbitrary Lagrangian-Eulerian method. In analy-
265
ses using the RITSS, the keying process is divided
into typically dozens to a hundred small steps. In
each step, the translation and rotation of SEPLA
must be small enough to avoid severe mesh distor-
tion around the SEPLA. Then the deformed soil is
re-meshed and updated Lagrangian calculation is
performed for each step, followed by field variables
(e.g. stresses and material properties) mapped to
the new mesh at the commencement of next step.
Here the RITSS strategy is coupled with com-
mercial package ABAQUS/Standard (SIMULIA,
2010). The flow chart of ABAQUS-based RITSS
and mapping techniques are detailed in Randolph
et al. (2008) and Wang et al. (2010, 2011).
The SEPLA’s installation is usually completed
within a few hours, therefore, the process can be
regarded as undrained due to low permeability of
clay. An elastic-perfectly plastic model with Tresca
yield criterion was used to describe the soil, with
Young’s modulus of 500su and Poisson’s ratio of
0.49. The undrained shear strength of clay su var-
ied linearly with depth as follows:
su = sum + kz (1)
Where sum represents the undrained strength at
seabed surface, k is the strength gradient and z
denotes the soil depth. The anchor was idealised as
a rigid body since its stiffness is much higher than
soil. The coefficient of earth pressure at rest was
taken as 1.
The schematic geometry of the anchor models
used in the centrifuge is shown in Figure 1b. The
initial embedment depth of the SEPLA prior to
keying (Hi) must be sufficiently deep, Hi ≥ 3B, to
guarantee deep failure mechanism in the subse-
quent daily operation stage (Song et al. 2009). B
represents the anchor width. The SEPLA is led to
rotate by a vertical or inclined pulling force applied
at padeye. The distance from the padeye to the
front face of plate is indicated as loading eccentric-
ity, e. It was found in previous experimental and
numerical researches that the loading eccentricity
ratio, e/B, is a dominant factor for keying response
(Gaudin et al. 2008, Wang et al. 2011). For practical
SEPLAs, the padeye is connected to the fluke via
shanks. However, the shanks were replaced with a
slender shaft in most centrifuge tests using PIV, in
order to lower the complexity of manufacturing
the anchor models. The shaft was ignored in the
numerical analyses, since the area and self-weight
of the shaft are marginal compared with those of
the fluke. This may lead to a slight overestima-
tion of the embedment loss. It was assumed that
no separation occurs between the SEPLA and soil
during the entire keying process. The assumption is
supported by the phenomena witnessed in almost
all centrifuge tests (Gaudin et al. 2006, O’Loughlin
et al. 2006, Gaudin et al. 2008, Gaudin et al. 2010,
Song et al. 2009). This is due to the negative pore
pressures (e.g. suction) generated on the back
face of the SEPLA given the anchor embedment
depth is sufficiently deep. This ‘bond’ mechanism,
however, does not necessarily mean the soil can-
not slide frictionally along the anchor faces. The
frictional soil-anchor interactions was simulated as
Coulomb friction contact, but the frictional resist-
ance is limited by a maximum shear stress, τmax,
which was defined as the product of the intact soil
strength at anchor centre (suc) and friction coeffi-
cient α
τmax = αsuc
A rough contact state, in geotechnical sense,
is achieved as α = 1. Note that the anchor centre
does not coincide with fluke centre when a keying
flap is attached (see Figure 1). Previous centrifuge
tests concerning offshore installations of suction
caissons and pipelines have indicated that under
static loads, the friction coefficient between steel
and kaolin clay ranges from 0.3 to 0.5 (Chen and
Randolph, 2007, Dingle et al. 2008). In ABAQUS-
based LDFE analysis, the maximum shear strength
at soil-anchor interfaces was assumed to be con-
stant in each small step, but it was updated at the
end of each step according to the current embed-
ment depth of anchor centre. The uncertainties are
then limited to the determination of the friction
coefficient.
3 COMPARISON WITH MODEL TESTS
Two-dimensional LDFE analyses were carried out
to reproduce the keying responses of strip anchors
in 3 groups of centrifuge tests by O’Loughlin et al.
(2006), Gaudin et al. (2008) and Gaudin et al.
(2010), respectively. The 3 groups are ordered as
Case A, B and C in Table 1. Within our knowl-
edge, these are all the published tests to measure
embedment loss of SEPLAs using the PIV tech-
nique. These tests were conducted in the drum cen-
Table 1. Centrifuge tests reproduced numerically (vertical pull-out).
SE PLA
Submerged unit Keying
Case Material weight (kN/m3) flap
A Stainless steel 67 No
B Aluminum 17 No
C Stainless steel 67 Yes
trifuge of the University of Western Australia, and
the soil samples were composed of kaolin clay. The
model anchors were inserted vertically into the soil
manually. The moderate soil disturbance induced
by the insertion was almost fully recovered during
spinning to the high g-level for the pull-out test,
therefore, the insertion process is not mimicked
in the LDFE analyses. Only vertical pull-out tests
were replicated, with the prototype geometries of
SEPLAs and soil properties listed in Table 1. In the
table, the thickness of SEPLAs and the submerged
unit weight of soil are represented with t and γ ′,
respectively.
(2)
3.1 Case A
Three model tests were reported by O’Loughlin
et al. (2006), with loading eccentricity ratio
e/B = 0.17, 0.5 and 1. For e/B = 0.17, the meas-
ured loss of embedment was as high as 2.2B, given
the initial embedment depth was Hi/B = 3. In addi-
tion, the soil was observed to be separated from
the back face of the anchor before the rotation
was completed. This ‘separation’ mechanism is not
allowed in practical applications. Hence the test
with e/B = 0.17 was abandoned. For e/B = 0.5 and
1, the experimental and numerical keying responses
are compared in Figure 2, where Δz represents the
embedment loss of fluke centre. When e/B = 0.5,
the numerical curve of α = 0.3 approach good
agreement with the experimental curve until the
anchor inclination is reduced to 40° to the horizon-
tal. The experimental ultimate loss of embedment
is reached at anchor inclination of 27° rather than
∼0 obtained from LDFE calculation. The differ-
ence may be attributed to the experimental errors
on loading inclination at the padeye and non-uni-
formity of the soil sample. If the anchor is assumed
as rough (α = 1), the embedment loss predicted is
much lower than the experimental and numerical
result with α = 0.3. When the loading eccentricity
ratio is increased to e/B = 1, the predicted keying
responses in terms of α = 0.3 and 1 are close, indi-
Soil
B t sum k γ′
(m) (m) Hi/B e/B (kPa) (kPa/m) (kN/m3)
3 0.2 3 0.17, 0.5, 1 0 0.7 6.5
2 0.3 3 0.25, 1 0 1 6.5
3.2 0.3 3.125 0.625 0 2.72 6.5
3.2 3.125 0.391
266

Figure 2. Comparison for Case A.
cating that the effect of anchor roughness is mini-
mal and the keying process dependent mainly on
the eccentricity ratio. The model anchor ceased to
rotate at anchor inclination of 23° to the horizon-
tal, which is also earlier than those predicted by the
LDFE analyses.
The influence of the anchor roughness can be
explained using the plasticity model developed by
Cassidy et al. (2012). The plasticity model incor-
porates combined loads into a yield envelop as
follows:
1/ p
⎛ ν ⎞
q ⎡⎛ | M | ⎞ m ⎛ | H | ⎞ n ⎤
f =⎜ − 1 + ⎢⎜ +⎜ ⎥ =0
⎝ νmax ⎟⎠ ⎢⎝ Mmax ⎟⎠ ⎝ Hmax ⎟⎠ ⎥
⎣ ⎦
(3)
where V, H and M are equivalent normal, sliding
and rotational loads at anchor centre and f defines
a plastic potential, of which the derivatives against
equivalent loads represent the corresponding local
normal (δw), sliding (δu) and rotational (δβ) dis-
placements (Figure 3), given an associated flow
rule is assumed. The fitting parameters within this
equation, p, q, m and n, are pre-decided at current
depth of embedment.
Figure 3. Schematic of global and localised loads and
displacements.
267
At the start of vertical pull-out, V = 0, whereas
H = Ta and M = eTa. Ta is the vertical pulling force
at padeye. These initial values satisfy Eq. 3. The
ratio of rotation and upward sliding displace-
ment δβ/δu at the start of keying is then deter-
mined as:
δβ ∂f ∂M m ⎡⎛ Hmax ⎞ ⎤
n
= = ⎢⎜ ⎟ − 1⎥ (4)
δ u ∂f / ∂H en ⎢⎝ | H | ⎠ ⎥⎦
⎣
It can be postulated from the preceding equa-
tion that the reduction of Hmax, resulting from a
decreased friction coefficient α as all other param-
eters are maintained, leads to a smaller rotation-
sliding displacement ratio, indicating the rotational
motion has slowed down with respect to soil-fluke
movement. On the other hand, when the influence
of loading eccentricity e is pursued from Eq. 4, it is
shown that a smaller e value corresponds to a more
significant change rotation-sliding ratio. Both pos-
tulations are in agreement with observations in
Figure 2.
3.2 Case B
Gaudin et al. (2008) conducted 10 experimental
tests with load inclination varying from 30° to
120°, of which two were vertical pull-out cases with
loading eccentricity ratio e/B = 0.25 and 1. The
experimental and numerical results of the vertical
pull-out cases are compared in Figure 4. For lower
eccentricity ratio e/B = 0.25, the numerical curve
representing α = 0.3 matches the experimental data
very well, whilst the assumption of rough contact
tends to grossly underestimate the loss of embed-
ment. As the loading eccentricity ratio is as large
as e/B = 1, similar to that observed in Figure 2, the
keying responses are nearly independent of the
friction coefficient.
Figure 4. Comparison for Case B.
3.3 Case C It is noteworthy that the strength gradient in
Case C is as high as k = 2.72 kPa/m, which is 3.86
The model anchors used in Case A and B are flap-
and 2.72 times of that in Case A and B, respectively.
less anchors. Gaudin et al. (2010) investigated the
An interesting observation is that the fitting fric-
activation of keying flap in centrifuge, with flap
tion coefficient α = 0.3 for Case A and B is 3 times
width of 0.5 or 0.375 times of the fluke width. It
of that for Case C. Therefore, it is deduced that
was observed in their tests that the flap remains
the maximum shear strength along the anchor-soil
aligned with the fluke until the fluke is nearly
interface is not proportional to the local intact soil
perpendicular to the pull-out force. This unex-
strength at the anchor centre, instead, the real fric-
pected phenomenon can be explained through the
tion coefficient reduces with the local soil strength.
moment applied on the keying flap (Wang et al.
For offshore normally consolidated clays, the typi-
2012). Therefore, the fluke and flap were simpli-
cal strength gradient is in the range 0.9–1.3 kPa/m
fied as a rigid body in the LDFE analyses, with the
and friction coefficient is suggested as 0.3 for pre-
anchor width being the sum of the fluke width,
dicting the keying of SEPLAs (see Figs. 2, 4). If
flap width and the width of gap between the fluke
the strength gradient is apparently higher than this
and flap. The intact soil strength suc in Eq. 2 is rela-
typical value, it is conservative to select a lower
tive to the centre of whole anchor. Gaudin et al.
value, say, α = 0.1.
(2010) reported two experimental keying curves
In the above comparisons (Figs. 2, 4, 5), the
both for B = 3.2 m (see Table 1). The numerical
LDFE analyses with friction coefficient of 0.3 or
keying responses are plotted in Figure 5, where the
0.1 predicts more reliable keying responses than
embedment loss refers to the fluke centre. In the
those with α = 1. As the loading eccentricity ratio
test with e/B = 0.625, the flap is activated as fluke is
e/B < 0.5, the embedment loss is significantly
rotated to ∼20° to the horizontal. The experimen-
underestimated if the friction coefficient is taken
tal curve is located in a narrow range determined
as 1. The influence of friction coefficient is mini-
by the numerical curves with α = 0.1 and 0.3. The
mal as the loading eccentricity ratio is increased to
anchor with α = 1 is predicted to rotate signifi-
e/B = 1.
cantly faster than the measured. For e/B = 0.391,
the experimental loss of embedment is at least
twice of the prediction in terms of α = 1. The fric-
tion coefficient α = 0.1 provides better agreement 4 FACTORS AFFECTING
with experimental curve than α = 0.3. EMBEDMENT LOSS

The resistances to the upward movement of the

anchor during keying consist of the frictional resist-
ances along the front and back faces and the resist-
ances at anchor ends. The former depends strongly
on the friction coefficient, but the latter is almost
independent of the friction coefficient. The end
resistance is directly related to the anchor thick-
ness. Thus, given the influence from eccentricity
ratio as discussed before, the importance of anchor
roughness is also a function of the ratio between
the frictional and end resistance. To explore the
sensitivity of these mutually affected factors in
respect to embedment losses, a series of LDFE
studies were conducted by varying anchor thick-
ness, loading eccentricity and friction coefficient.
For a typical flapless anchor (B = 4 m, Hi/B = 5) in
normally consolidated clay (sum = 0, k = 1 kPa/m),
the friction coefficient is selected as α = 0.3 or 1.
Figure 6 presents a quantified interpretation on
the combined effects of the loading eccentricity
ratio, anchor thickness ratio and frictional coeffi-
cient, in which Δzu represents the ultimate loss of
embedment. It can be seen that:
1. The frictional effect becomes more remarkable
with the reduction of loading eccentricity ratio.
Figure 5. Comparison for Case C. For e/B = 0.5, the predicted loss of embedment

268

Figure 6. Effect of the anchor roughness on the ulti-
mate embedment loss.
with α = 0.3 is 1.39 times (t/B = 0.1) to 3.37 times
(t/B = 0.05) of that with α = 1. The difference is
reduced to 1.11 times (t/B = 0.1) from 1.36 times
(t/B = 0.05) when the loading eccentricity ratio
is enhanced to e/B = 1.
2. The frictional effect is impaired with the
increase of thickness ratio, especially for lower
loading eccentricity ratio. The reason is that a
larger magnitude of resistances is transferred
to the bearing at the anchor ends as the anchor
becomes thicker.
3. Once the loading eccentricity ratio increases to
1 and anchor thickness ratio is larger than 0.07,
effects of the friction coefficient and anchor
thickness on the ultimate loss of embedment
can be ignored.
5 CONCLUSIONS
The keying installation process of frictional
SEPLAs was investigated using a large deforma-
tion finite element approach. A friction coef-
ficient was defined as the ratio of the maximum
shear strength along the ‘anchor-soil’ interface to
the intact soil strength at the anchor centre. The
friction coefficient determined is far less than 1 by
comparing numerical results with 3 groups of cen-
trifuge tests in kaolin clay. The keying responses
based on a friction coefficient of α = 0.3 agree well
with all the experimental curves, for a soil strength
gradient of 0.9–1.3 kPa/m, typical for normally
consolidated clays. The estimated friction coef-
ficient may be reduced to as low as 0.1 when the
strength gradient rises to 2.7 kPa/m. For the sea-
bed other than kaolin clay, the friction coefficient
can be determined in a similar way. The effect
of frictional resistance on the keying response is
combined with the effects of loading eccentricity
269
ratio and anchor thickness ratio. The friction effect
becomes dominant for loading eccentricity e/B ≤
0.5 or for anchor thickness t/B ≤ 0.07.
REFERENCES
Cassidy, M.J., Gaudin, C., Randolph, M.F., Wong, P.C.,
Wang, D. & Tian, Y. 2012. A plasticity model to
assess the keying of plate anchors. Geotechnique, 62,
825–836.
Chen, W. & Randolph, M.F. 2007. External radial stress
changes and axial capacity for suction caissons in soft
clay. Geotechnique, 57, 499–511.
Dingle, H.R.C., White, D.J. & Gaudin, C. 2008. Mecha-
nisms of pipe embedment and lateral breakout on soft
clay. Canadian Geotechnical Journal, 45, 636–652.
Gaudin, C., O’Loughlin, C.D., Randolph, M.F. & Low-
mass, A.C. 2006. Influence of the installation proc-
ess on the performance of suction embedded plate
anchors. Geotechnique, 56, 381–391.
Gaudin, C., Simkin, M., White, D.J. & O’Loughlin, C.D.
2010. Experimental investigation into the influence of
a keying flap on the keying behaviour of plate anchor.
Proc. 20th Int. Offshore and Polar Engineering Conf.,
533–540.
Gaudin, C., Tham, K.H. & Ouahsine, S. 2008. Plate
anchor failure mechanism during keying process.
Proc. 18th Int. Offshore and Polar Engineering Conf.,
613–619.
O’Loughlin, C.D., Lowmass, A., Gaudin, C. & Ran-
dolph, M.F. 2006. Physical modelling to assess keying
characterics of plate anchors. Proce. of 6th Int. Conf.
on Physical Modelling in Geotechnics, 659–665.
Randolph, M.F., Wang, D., Zhou, H., Hossain, M.S. &
Hu, Y. 2008. Large Deformation Finite Element Anal-
ysis for Offshore Applications. Proc. 12th Int. Conf.
International Association for Computer Methods and
Advances in Geomechanics, 3307–3318.
Simulia. 2010. ABAQUS 6.10 Manuals. Simulia.
Song, Z., Hu, Y., O’Loughlin, C. & Randolph, M.F. 2009.
Loss in Anchor Embedment during Plate Anchor
Keying in Clay. Journal of Geotechnical and Geoenvi-
ronmental Engineering, 135, 1475–1485.
Wang, D., Gaudin, C. & Randolph, M.F. 2012. Large
deformation finite element analysis investigating the
performance of anchor keying flap. Ocean Engineer-
ing, under revision.
Wang, D., Hu, Y. & Randolph, M.F. 2010. Three-dimen-
sional large deformation finite-element analysis of
plate anchors in uniform clay. Journal of Geotechnical
and Geoenvironmental Engineering, 136, 355–365.
Wang, D., Hu, Y. & Randolph, M.F. 2011. Keying of
Rectangular Plate Anchors in Normally Consolidated
Clays. Journal of Geotechnical and Geoenvironmental
Engineering, 137, 1244–1253.
White, D.J., Take, W.A. & Bolton, M.D. 2003. Soil defor-
mation measurement using particle image velocime-
try (PIV) and photogrammetry. Geotechnique, 53,
619–631.
Wilde, B., Treu, H. & Fulton, T. 2001. Field testing of
suction embedded plate anchors. Proc. 11th Offshore
and Polar Engineering Conf., 544–551.
 Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4

Author index

Andresen, L. 162 Haigh, S.K. 86 Oset, F. 140

Andreykiv, A. 24 Hamad, F. 176
Avanzi, F. 92 Hamann, T. 118 Paris Fernández, J.E. 258
Han, C. 264 Petalas, A. 65
Becker, P. 198 Heaney, C.E. 8 Prieto, L. 219
Beijer Lundberg, A. 79 Henke, S. 118 Pucker, T. 118
Benz, T. 233 Hicks, M.A. 8, 31, 192
Beuth, L. 1 Hofstede, H. 184 Raspini, F. 51
Bonnier, P.G. 8, 43, 72 Huybrechts, N. 58 Raymackers, S. 184
Brinkgreve, R.B.J. 8, 24, 43, 65, Robinson, S. 128
72, 104 Igoe, D. 252 Rohe, A. 31
Brown, M.J. 128 Ishihara, Y. 86 Rozos, D. 51
Burali d’Arezzo, F. 86
Jassim, I. 15 Sagaseta, C. 212
Caruso, M. 92 Jommi, C. 92 Sexton, B.G. 226
Carvajal, E. 205 Sivasithamparam, N. 43, 72, 212
Castro, J. 205, 212 Karstunen, M. 43, 72, 198, 212 Stodieck, X.A.L. 233
Coetzee, C. 15 Khoa, H.D.V. 162
Kirwan, L. 252 Tan, S. 31
de Blaeij, T. 169 Kort, D.A. 184 te Slaa, S. 135
De Cock, F. 58 Tefera, T. 140
Degago, S. 245 Lameire, B. 58 Tvedt, G. 140
Denies, N. 58 Lim, L.J. 24
Dijkstra, J. 79, 135, 169 Lloret-Cabot, M. 192 Van Lysebetten, G. 58
Loupasakis, C. 51 van Tol, A.F. 79, 104
Ebeltoft, R. 245 Vardon, P.J. 38
Egan, D. 99 Maertens, J. 58 Vermeer, P.A. 1, 15, 176
Engin, H.K. 104 Martin, F. 219 Vervoort, A. 58
McCabe, B.A. 226 Vukotić, G. 205
Galavi, V. 65 Melentijevic, S. 219
Gaudin, C. 264 Melhus, O. 245 Wang, D. 264
Gavin, K. 252 Meyer, V. 184 Wehr, J. 241
Ghandeharioon, A. 111 Moormann, C. 176 Wehr, W. 205
Gómez Cabrera, J. 258 Moretti, S. 51 White, D.J. 146
Grabe, J. 118
Grimstad, G. 245 Nuttall, J.D. 38, 192

271