Finite element analysis in

geotechnical engineering
Application

David M. Potts and Lidija Zdravkovid

Imperial College of Science, Technology and Medicine

With contributions from:

Trevor I. Addenbrooke
Kelvin G . Higgins
NebojSa KovaCevid

Published by Thomas Telford Publishing, T h o r n s Telford Ltd, 1 Heron Quay, London

E14 4JD.
URL: http://\?.ww.thomastelford.com

Distributors for Thomas Telford books are

USA: ASCE Press, 180 1 Alexander Bell Drive, Reston, VA 20191-4400, USA
Japan: Maruzen Co. Ltd, Book Deparhnent, 3-10 Nihonbashi 2-chome, Chuo-ku,
Tokyo 103
Australia: DA Books and Journals, 648 Whitehorse Road, Mitcharn 3 132, Victoria

First published 2001

Also available from Tbomas Telford Books
Finite elemenl analysis in geotechnical engineering: theory. ISBN 0 7277 2753 2

A catalogue record for this book is available from the British Library

ISBN: 0 7277 2783 4

O David M. Potts and Lidija ZdravkoviC, and Thomas Telford Limited, 2001

All rights, including translation, reserved. Except as permitted by the Copyright, Designs
and Patents Act 1988, no part of this publication may be reproduced, stored in a retrieval
system or transmitted in any form or by any means, electronic, mechanical, photocopying or
otherwise, without the prior written permission of the Publishing Director, Thomas Telford
Publishing, Thomas Telford Ltd, 1 Heron Quay, London E14 4JD.
This book is published on the understanding that the author are solely responsible for the
statements made and opinions expressed in it and that its publication does not necessarily
imply that such statements andlor opinions are or reflect the views or opinions of the
publishers. While every effort has been made to ensure that the statements made and the
opinions expressed in this publication provide a safe and accurate guide, no liability or
responsibility can be accepted in this respect by the authors or publishers.

Contents
Preface
Authorship

xvi

Acknowledgements

xvi

1.

Obtaining geotechnical parameters

1.1 Synopsis
1.2 Introduction
1.3 Laboratory tests
1.3.1
Introduction
1.3.2
Oedometer test
1.3.3
Triaxial test
1.3.4
True triaxial test
1.3.5
Direct shear test
1.3.6
Simple shear test
1.3.7
Ring shear test
1.3.8
Hollow cylinder test
1.3.9
Directional shear cell
1.3.10
Geophysical techniques
1.3.11
Pemeameters
1.4 In-situ tests
1.4.1
Introduction
1.4.2
Standard penetration test (SPT)
1.4.3
Cone penetration test (CPT)
1.4.4
Pressuremeter testing
1.4.5
The plate loading test
1.4.6
Pumping tests
1.5 Summary

2.

Tunnels
2.1
Synopsis
2.2
Introduction
2.3
Tunnel construction

ii / Finite element analysis in geotechnical engineering: Application

2.3.1
2.3.2
2.3.3

2.4

2.5

2.6

2.7

2.8
3.

Earth
3.1
3.2
3.3

3.4

Introduction
Open faced shield tunnelling
Tunnel Boring Machines (TBM), including slurry
shields and Earth Pressure Balance (EPB) tunnelling
The sprayed concrete lining (SCL) method
2.3.4
Ground response to tunnel construction
2.3.5
Simulation of the construction process
2.4.1
Introduction
2.4.2
Setting up the initial conditions
2.4.3
Important boundary conditions
2.4.4
Modelling tunnel excavation
2.4.5
Modelling the tunnel lining
Modelling time dependent behaviour
2.5.1
Introduction
Setting up the initial conditions
2.5.2
2.5.3
Hydraulic boundary conditions
2.5.4
Permeability models
A parametric study of the effect of permeable and
2.5.5
impermeable tunnel linings
Choice of soil model
2.6.1
Introduction
Results from a parametric study
2.6.2
Devices for improving the surface settlement
2.6.3
prediction
Interaction analysis
2.7.1
The influence of building stiffness on tunnel-induced
ground movements
The Treasury building - a case study
2.7.2
2.7.3
Twin tunnel interaction
Summary
retaining structures
Synopsis
Introduction
Types of retaining structure
3.3.1
Introduction
3.3.2
Gravity walls
ReinforcedJanchored earth wall
3.3.3
3.3.4
Embedded walls
General considerations
3.4.1
Introduction
3.4.2
Symmetry
3.4.3
Geometry of the finite element model
3.4.4
Support systems

Contents 1 iii

3.4.5

Choice of constitutive models

3.4.5.1
Structural components
3.4.5.2
Soil
3.4.6
Initial ground conditions
3.4.6.1
General
3.4.6.2 'Greenfield' conditions
3.4.6.3 Modified initial soil stresses
3.4.7
Construction method and programme
3.4.7.1 General
3.4.7.2 Construction method
3.4.7.3 Time related movements
3.4.7.4 Ground water control
3.5
Gravity walls
3.5.1
Introduction
3.5.2
Earth pressure due to compaction
3.5.3
Finite element analysis
3.6
Reinforced earth walls
3.6.1
Introduction
3.6.2
Finite element analysis
3.7
Embedded walls
3.7.1
Introduction
3.7.2
Installation effects
3.7.2.1 General
3.7.2.2 Field measurements
3.7.2.3 Analysis
3.7.2.4 Comments
3.7.3
Modelling of walls
3.7.3.1 Element type
3.7.3.2 Wall stiffness
3.7.3.3 Interface behaviour
3.7.3.4 Wallpermeability
3.7.4
Support systems
3.7.4.1
Introduction
3.7.4.2 Support stiffness
3.7.4.3 Connection details
3.7.4.4
Active support systems
3.7.4.5 Berms
31.7.4.6 Ground anchors
3.7.4.7 Relieving slabs
3.7.5
Long term behaviour and post construction effects
3.7.6
Adjacent structures
3.8
Summary
Appendix 111. l

84
84
85
88
88
88
89
91
91
91
92
93
93
93
94
95
96
96
99
103
103
104
104
104
105
106
107
107
109
111
111
112
112
112
113
114
115
115
116
1 18
119
122
123

iv / Finite element analysis in geotechnical engineering: Application

4.

Cut slopes
4.1
Synopsis
4.2
Introduction
4.3
'Non-softening ' analyses
4.3.1
Introduction
Cut slopes in stiff 'non-softening' clay
4.3.2
4.3.2.1 Introduction
4.3.2.2
Soil parameters
4.3.2.3
Finite element analyses
4.3.2.4
Results of analyses
Cut slopes in soft clay
4.3.3
4.3.3.1
Introduction
4.3.3.2 Soil parameters
4.3.3.3 Finite element analyses
4.3.3.4 Results of analyses
4.4
Progressive failure
4.5
'Softening' analyses
introduction
4.5.1
4.5.2
Choice of constitutive model
4.5.3
Implications for convergence
Cut slopes in London Clay
4.5.4
4.5.4.1
Introduction
4.5.4.2 Soil parameters
4.5.4.3 Finite element analyses
Results of a typical analysis
4.5.4.4
4.5.4.5 Effect of coefficient of earth
pressure at rest
Effect of surface boundary suction
4.5.4.6
4.5.4.7 Effect of slope geometry
4.5.4.8 Effect of surface cracking
4.5.4.9 Effect of subsequent changes to slope
geometry
4.5.4.10 Further discussion
Construction of cut slope under water
4.6
4.7
Summary

5.

Embankments
5.1
Synopsis
5.2
Introduction
Finite element analysis of rockfill dams
5.3
5.3.1
Introduction
5.3.2
Typical stress paths
5.3.3
Choice of constitutive models
5.3.3.1 Linear elastic analysis

Contents 1 v

5.4

5.5

5.3.3.2 'Power law' models

5.3.3.3 Hyperbolic model
5.3.3.4 K-G model
5.3.3.5 Elasto-plastic models
Layered analysis, stiffness of the simulated layer
5.3.4
and compaction stresses
5.3.5
Example: Analysis of Roadford dam
5.3.5.1 Introduction
5.3.5.2 Material parameters
5.3.5.3 Finite element analysis
5.3.5.4 Comparison with observations
Example: Analysis of old puddle clay core dams
5.3.6
5.3.6.1 Introduction
5.3.6.2 Dale Dyke dam
5.3.6.3 Ramsden dam
Finite element analysis of earth embankments
5.4.1
Introduction
5.4.2
Modelling of earthfill
Example: Road embankments on London Clay
5.4.3
5.4.3.1
Introduction
5.4.3.2 Material properties
5.4.3.3
Finite element analyis
Example: Failure of Carsington embankment
5.4.4
5.4.4.1 Introduction
5.4.4.2 Material parameters and soil model used
5.4.4.3
Finite element analysis
5.4.4.4 Original Carsington section
5.4.4.5 Effect of the core geometry on
progressive failure
5.4.4.6 Effect of berm in improving the stability
Finite element analysis of embankments on soft clay
5.5.1
Introduction
5.5.2
Typical soil conditions
5.5.3
Choice of constitutive model
5.5.4
Modelling soil reinforcement
5.5.5
Example: Effect of a surface crust
5.5.5.1 Introduction
5.5.5.2 Soil conditions
5.5.5.3 Finite element analysis
5.5.5.4 Results
5.5.6
Example: Effect of reinforcement
5.5.6.1 Introduction
5.5.6.2 Soil conditions
5.5.6.3 Results

vi / Finite element analysis in geotechnicai engineering: Application

5.5.7

5.5.8

5.6

6.

Example: Staged construction

5.5.7.1 Introduction
5.5.7.2 Soil conditions
5.5.7.3 Finite element analysis
5.5.7.4 Results
Example: Effect of anrsotropic soil behaviour
5.5.8.1
Introduction
5.5.8.2 Geometry
5.5.8.3 Soil conditions
5.5.8.4 Finite element analysis
5.5.8.5 Results

Summary

Shallow foundations
6.1
Synopsis
6.2
Introduction
6.3
Foundation types
6.3.1
Surface foundations
6.3.2
Shallow foundations
6.4
Choice of soil model
Finite element analysis of surface foundations
6.5
6.5.1
Introduction
6.5.2
Flexible foundations
6.5.3
Rigid foundations
6.5.4
Examples of vertical loading
6.5.4.1 Introduction
Strip footings on undrained clay
6.5.4.2
6.5.4.3 Effect of footing shape on the bearing
capacity of undrained clay
6.5.4.4 Strip footings on weightless drained soil
6.5.4.5 Strip footings on a drained soil
6.5.4.6 Circular footings on a weightless drained
soil
6.5.4.7 Circular footings on a drained soil
Undrained bearing capacity of non-homogeneous
6.5.5
clay
6.5.5.1 Introduction
6.5.5.2 Constitutive model
6.5.5.3 Geometry and boundary conditions
6.5.5.4 Failure mechanisms
Undrained bearing capacity of pre-loaded strip
6.5.6
foundations on clay
6.5.6.1 Introduction
6.5.6.2 Constitutive model

Contents / vii

6.6

6.7
7.

Deep
7.1
7.2
7.3

7.4

6.5.6.3 Geometry and boundary conditions

6.5.6.4
Results of the analyses
6.5.6.5 Concluding remarks
6.5.7
Effect of anisotropic strength on bearing capacity
6.5.7.1 Introduction
6.5.7.2 Soil behaviour
6.5.7.3 Behaviour of strip footings
6.5.7.4 Behaviour of circular footings
Finite element analysis of shallow foundations
6.6.1
Introduction
6.6.2
Effect of foundation depth on undrained bearing
capacity
6.6.3
Example: The leaning Tower of Pisa
6.6.3.1 Introduction
6.6.3.2 Details of the Tower and ground profile
6.6.3.3 History of construction
6.6.3.4 History of tilting
6.6.3.5 The motion of the Tower foundations
6.6.3.6
Stability of tall towers
6.6.3.7 Soil properties
6.6.3.8 Finite element analysis
6.6.3.9 Simulation of the history of inclination
6.6.3.10 Temporary counterweight
6.6.3.1 1 Observed behaviour during application
of the counterweight
6.6.3.12 Permanent stabilisation of the Tower
6.6.3.13 Soil extraction
6.6.3.14 The response of the Tower to
soil extraction
6.6.3.15 Comments
Summary
foundations
Synopsis
Introduction
Single piles
7.3.1
Introduction
7.3.2
Vertical loading
7.3.3
Lateral loading
Pile group behaviour
7.4.1
Introduction
7.4.2
Analysis of a pile group
7.4.3
Superposition
7.4.3.1 Simple superposition

viii / Finite element analysis in geotechriical engineering: Application

7.5

7.6

8.

7.4.3.2 Pile displacements with depth

Load distribution within a pile group
7.4.4
7.4.4.1 Obtaining an initial trial division of the
applied loads
7.4.4.2 Evaluating pile head displacements
7.4.4.3 Checking the rigid pile cap criterion
7.4.5
Pile group design
7.4.5.1 Matrix formulation of the pile group
response
7.4.5.2
Superposition of loads
7.4.5.3 Evaluating the solution displacements
and rotations
7.4.6
Magnus
7.4.6.1
Introduction
7.4.6.2 Soil properties and initial conditions
7.4.6.3 Finite element analyses
7.4.6.4
Design of Magnus foundations
7.4.6.5 Environmental loading
Bucket foundations
7.5.1
Introduction
7.5.2
Geometry
7.5.3
Finite element analysis
Modelling
of the interface between top cap and soil
7.5.4
7.5.5
Isotropic study
7.5.5.1 Soil conditions
7.5.5.2 Parametric studies
7.5.5.3 Results
7.5.6
Anisotropic study
7.5.6.1
Introduction
7.5.6.2 Results
7.5.7
Suction anchors
7.5.7.1
Introduction
7.5.7.2 Geometry
7.5.7.3 Results
Summary

Benchmarking
8.1
Synopsis
8.2 Definitions
8.3
Introduction
Causes of errors in computer calculations
8.4
8.5
Consequences of errors
8.5
Developers and users
8.6.1
Developers

Contents I ix
8.6.2
Users
Techniques used to check computer calculations
Benchmarking
8.8.1
General
8.8.2
Standard benchmarks
8.8.3
Non-standard benchmarks
8.9
The INTERCLAY I1 project
8.10 Examples of benchmark problems - Part I
8.10.1
General
Example l: Analyses of an ideal triaxial test
8.10.2
Example 2: Analysis of a thick cylinder
8.10.3
Example 3: Analyses of an advancing tunnel
8.10.4
heading
Example 4: Analysis of a shallow waste disposal
8.10.5
Example 5: Simplified analysis of a shallow waste
8.10.6
8.1 1 Examples of benchmark problems - Part II
(German Society for Geotechnics benchmarking exercise)
8.1 1.1
Background
8.1 1.2
Example 6: Construction of a tunnel
8.1 1.3
Example 7: Deep excavation
8.1 1.4
General comments
8.12 Summary
Appendix VIII.1 Specification for Example 1: Analyses of an
idealised triaxial test
VIII.1 .l Geometry
VIII. 1.2 Material properties and initial stress conditions
VIII. 1.3 Loading conditions
Appendix VIII.2 Specification for Example 2: Analysis of a thick
cylinder
VIII.2.1 Geometry
VIII.2.2 Material properties
VIII.2.3 Loading conditions
Appendix VIII.3 Specification for Example 3: Analysis of an
advancing tunnel heading
VIII.3.1 Geometry
VIII.3.2 Material properties
VIII.3.3 Loading conditions
Appendix VIIIA Specification for Example 4: Analysis of a shallow
waste disposal
VIII.4.1 Geometry
VI11.4.2 Material properties
VIII.4.3 Loading conditions
Appendix VIII.5 Specification for Example 5: Simplified analysis
of a shallow waste disposal

8.7
8.8

/ Finite element analysis in geotechnical engineering: Application

VIII.5. I
VIII.5.2
VIII.5.3
VII1.5.4
Appendix VIII.6
VIII.6.1
VIII.6.2
Appendix VIII.7
VI11.7. I
VIII.7.2
VIII.7.3
9.

Geometry
36 1
Material properties
36 1
Loading conditions
36 1
Additional boundary conditions
362
Specification for Example 6: Construction of a tunnel 362
Geometry
362
Material properties
362
Specification for Example 7: Deep excavation
362
Geometry
362
Material properties
362
Construction stages
363

Restrictions and pitfalls

9.1
Synopsis
9.2
Introduction
9.3
Discretisation errors
Numerical stability of zero thickness interface elements
9.4
9.4.1
Introduction
9.4.2
Basic theory
9.4.3
111-conditioning
9.4.4
Steep stress gradients
Modelling of structural members in plane strain analysis
9.5
9.5.1
Walls
9.5.2
Piles
9.5.3
Ground anchors
Structural members in coupled analyses
9.5.4
9.5.5
Structural connections
9.5.6
Segmental tunnel linings
Use of the Mohr-Coulomb model for undrained analysis
9.6
9.7 Influence of the shape of the yield and plastic potential
surfaces in the deviatoric plane
Using critical state models in undrained analysis
9.8
9.9 Construction problems
9.10 Removal of prescribed degrees of freedom
9.1 1 Modelling underdrainage
9.12 Summary
References

396

List of symbols

410

Index

4 15

Preface
While the finite element method has been used in many fields of engineering
practice for over thirty years, it is only relatively recently that it has begun to be
widely used for analysing geotechnical problems. This is probably because there
are many complex issues which are specific to geotechnical engineering and which
have only been resolved relatively recently. Perhaps this explains why there are
few books which cover the application of the finite element method to geotechnical
engineering.
For over twenty years we, at Imperial College, have been working at the
leading edge of the application of the finite element method to the analysis of
practical geotechnical problems. Consequently, we have gained enormous
experience of this type of work and have shown that, when properly used, this
method can produce realistic results which are of value to practical engineering
problems. Because we have written all our own computer code, we also have an
in-depth understanding of the relevant theory.
Based on this experience we believe that, to perform useful geotechnical finite
element analysis, an engineer requires specialist knowledge in a range of subjects.
Firstly, a sound understanding of soil mechanics and finite element theory is
required. Secondly, an in-depth understanding and appreciation of the limitations
of the various constitutive models that are currently available is needed. Lastly,
users must be fully conversant with the manner in which the software they are
using works. Unfortunately, it is not easy for a geotechnical engineer to gain all
these skills, as it is vary rare for all of them to be part of a single undergraduate or
postgraduate degree course. It is perhaps, therefore, not surprising that many
engineers, who carry out such analyses andfor use the results from such analyses,
are not aware of the potential restrictions and pitfalls involved.
This problem was highlighted four years ago when we gave a four day course
on numerical analysis in geotechnicalengineering.Although the course was a great
success, attracting many participants from both industry and academia, it did
highlight the difficulty that engineers have in obtaining the necessary skills
required to perform good numerical analysis. In fact, it was the delegates on this
course who urged us, and provided the inspiration, to write this book.
The overall objective of the book is to provide the reader with an insight into
the use of the finite element method in geotechnical engineering. More specific
aims are:

xii / Finite element analysis in geotechnical engineering: Application

- To present the theory, assumptions and approximations involved in finite

element analysis;
To describe some of the more popular constitutive models currently available
and explore their strengths and weaknesses;
To provide sufficient information so that readers can assess and compare the
capabilities of available commercial software;
To provide sufficient information so that readers can make judgements as to the
credibility of numerical results that they may obtain, or review, in the future;
To show, by means of practical examples, the restrictions, pitfalls, advantages
and disadvantages of numerical analysis.

The book is primarily aimed at users of commercial finite element software both
in industry and in academia. However, it will also be of use to students in their
final years of an undergraduate course, or those on a postgraduate course in
geotechnical engineering. A prime objective has been to present the material in the
simplest possible way and in manner understandable to most engineers.
Consequently, we have refrained from using tensor notation and have presented all
theory in terms of conventional matrix algebra.
When we first considered writing this book, it became clear that we could not
cover all aspects of numerical analysis relevant to geotechnical engineering. We
reached this conclusion for two reasons. Firstly, the subject area is so vast that to
adequately cover it would take many volumes and, secondly, we did not have
experience with all the different aspects. Consequently, we decided only to include
material which we felt we had adequate experience of and that was useful to a
practising engineer. As a result we have concentrated on static behaviour and have
not considered dynamic effects. Even so, we soon found that the material we
wished to include would not sensibly fit into a single volume. The material has
therefore been divided into theory and application, each presented in a separate
volume.
Volume 1 concentrates on the theory behind the finite element method and on
the various constitutive models currently available. This is essential reading for any
user of a finite element package as it clearly outlines the assumptions and
limitations involved. Volume 2 concentrates on the application of the method to
real geotechnical problems, highlighting how the method can be applied, its
advantages and disadvantages, and some of the pitfalls. This is also essential
reading for a user of a software package and for any engineer who is
commissioning andlor reviewing the results of finite element analyses.
Volume 1 of this book consists of twelve chapters. Chapter I considers the
general requirements of any form of geotechnical analysis and provides a
framework for assessing the relevant merits of the different methods of analysis
currently used in geotechnical design. This enables the reader to gain an insight
into the potential advantage of numerical analysis over the more 'conventional'
approaches currently in use. The basic finite element theory for linear material
behaviour is described in Chapter 2. Emphasis is placed on highlighting the

Preface I xiii

assumptions and limitations. Chapter 3 then presents the modifications and

additions that are required to enable geotechnical analysis to be performed.
The main limitation of the basic finite element theory is that it is based on the
assumption of linear material behaviour. Soils do not behave in such a manner and
Chapter 4 highlights the important facets of soil behaviour that ideally should be
accounted for by a constitutive model. Unfortunately, a constitutive model which
can account for all these facets of behaviour, and at the same time be defined by
a realistic number of input parameters which can readily be determined from
simple laboratory tests, does not exist. Nonlinear elastic constitutive models are
presented in Chapter 5 and although these are an improvement over the linear
elastic models that were used in the early days of finite element analyses, they
suffer severe limitations. The majority of constitutive models currently in use are
based on the framework of elasto-plasticity and this is described in Chapter 6.
Simple elasto-plastic models are then presented in Chapter 7 and more complex
models in Chapter 8.
To use these nonlinear constitutive models in finite element analysis requires
an extension of the theory presented in Chapter 2. This is described in Chapter 9
where some of the most popular nonlinear solution strategies are considered. It is
shown that some of these can result in large errors unless extreme care is exercised
by the user. The procedures required to obtain accurate solutions are discussed.
Chapter 10 presents the finite element theory for analysing coupled problems
involving both deformation and pore fluid flow. This enables time dependent
consolidation problems to be analysed.
Three dimensional problems are considered in Chapter 11. Such problems
require large amounts of computer resources and methods for reducing these are
discussed. In particular the use of iterative equation solvers is considered. While
these have been used successfully in other branches of engineering, it is shown
that, with present computer hardware, they are unlikely to be economical for the
majority of geotechnical problems.
The theory behind Fourier Series Aided Finite Element Analysis is described
in Chapter 12. Such analysis can be applied to three dimensional problems which
possess an axi-symmetric geometry but a non axi-symmetric distribution of
material properties andtor loading. It is shown that analyses based on this approach
can give accurate results with up to an order of magnitude saving in computer
resources, compared to equivalent analyses performed with a conventional three
dimensional finite element formulation.
This volume ofthe book (i.e Volume 2) builds on the material given in Volume
1. However, the emphasis is less on theory and more on the application of the finite
element method in engineering practice. It consists of nine chapters.
Chapter 1 considers the problems involved in obtaining geotechnical
parameters. These are necessary to define the constitutive models and initial
conditions for an analysis. The relative merits of laboratory and field testing are
discussed and the parameters that can be obtained from the various tests examined.
The analyses of tunnel construction is considered in Chapter 2. Emphasis is

xiv / Finite element analysis in geotechnical engineering: Application

placed on simulating the construction process and how this can be achieved in a
two dimensional analysis. Modelling of the tunnel lining, the choice of an
appropriate constitutive model for the soil and the selection of appropriate
hydraulic boundary conditions are considered.
Chapter 3 considers the analysis of earth retaining structures. In particular the
analysis of gravity, embedded and reinforced/anchored walls are examined.
Emphasis is placed on modelling the structural elements, choosing appropriate
constitutive models and simulating construction.
Cut slopes are considered in Chapter 4. The concepts behind progressive failure
are introduced. Its role in slope stability is then examined and in particular its
interaction with the long term dissipation of excess pore water pressures.
The analysis of embankments is discussed in Chapter 5. Embankments built of
earthfill and rockfill and those built on weak and strong foundations are
considered. The choice of appropriate constitutive models is discussed at some
length as are the appropriate hydraulic boundary conditions and the role of
progressive failure. For embankments on soft ground, single and multi-staged
construction and the benefits of reinforcement are examined.
Chapter 6 considers shallow foundations. To begin with, simple surface
foundations are considered and comparisons with the classical bearing capacity
solutions made. The ability of numerical analysis to advance the current state ofthe
art is then demonstrated by considering some of the weaknesses in current bearing
capacity theory. For example, the effect of self weight on drained bearing capacity,
the effect of foundation shape and its depth below the soil surface are considered.
The effects of anisotropic soil strength and of pre-loading on bearing capacity are
also examined. The analysis of tall towers and the difference between bearing
capacity failure and leaning instability is discussed. Analysis of the leaning Tower
of Pisa is then used to demonstrate the power of numerical analysis.
Deep foundations are considered in Chapter 7. The analyses of single piles and
pile groups subjected to combined vertical, lateral and moment loading are
considered. The behaviour of suction caissons and the possible detrimental effects
of neglecting anisotropic soil strength are discussed.
Benchmarking and validation of numerical analyses are discussed in Chapter
8. The various options, their deficiencies and results from some recent
benchmarking exercises are described.
Chapter 9 describes many of the restrictions and pitfalls that the authors have
experienced. In particular, restrictions implicit in modelling problems as plane
strain, problems associated with initial conditions and pitfalls associated with the
use of some of the more common constitutive models are discussed.
Emphasis throughout this volume of the book is placed on explaining how the
finite element method should be applied and what are the restrictions and pitfalls.
In particular, the choice of suitable constitutive models for the various geotechnical
boundary value problems is discussed at some length. To illustrate the material
presented, examples from the authors experiences with practical geotechnical
problems are used.

Preface / xv
All the numerical examples presented in both this volume and Volume 1 of this
book have been obtained using the Authors' own computer code. This software is
not available commercially and therefore the results presented are unbiased. As
commercial software has not beenused, the reader must consider what implications
the results may have on the use of such software.
London
March 200 1

David M. Poas
Lidija Zdravkovic

Authorship
This volume has been edited and much of its content written by David Potts and
Lidija Zdravkovic. Several of the chapters involve contributions from colleagues
at Imperial College and the Geotechnical Consulting Group (GCG). In particular:
Dr Trevor Addenbrooke (Imperial College) wrote a large part of Chapter 2
(Tunnels);
Mr Kelvin Higgins (GCG) wrote Chapter 8 (Benchmarking) and contributed to
Chapter 3 (Earth retaining structures);
Dr NebojSa KovateviC (GCG) wrote large parts of Chapter 4 (Cut slopes) and
Chapter 5 (Embankments).

Acknowledgements
We would like to acknowledge our colleagues Professors John Burland and Peter
Vaughan and all the past and present research students at Imperial College, without
whose interest and involvement this book would not have been possible. In
particular we would like to acknowledge Dr Dennis Ganendra whose PhD work on
pile groups (supervised by David Potts) forms part of Chapter 7 (Deep
foundations).
David Potts
Lidija Zdravkovic

1.

Obtaining geotechnical parameters

1. l

Synopsis

One of the essential ingredients for a successful finite element analysis of a

geotechnical problem is an appropriate soil constitutive model. As explained in
Volume 1 of this book, there is not a single constitutive model currently available
which can reproduce all aspects of real soil behaviour. It is therefore important to
recognise in the analysis what aspects of the problem are of major interest and to
choose a model accordingly. However, to employ a particular soil model in an
analysis, appropriate laboratory andfor field tests are required from which to derive
the necessary model parameters. This chapter describes the standard laboratory and
field experiments used in geotechnical practice and the parameters that can be
obtained from each of them.

1.2

Introduction

Since none of the currently available soil constitutive models can reproduce all of
the aspects of real soil behaviour, it is necessary to decide which soil features
govern the behaviour of a particular geotechnical problem (e.g. stiffness,
deformation, strength, dilation, anisotropy, etc.) and choose a constitutive model
that can best capture these features. Another factor that governs the choice of soil
models for finite element analysis is the availability of appropriate soil data from
which to derive the necessary model parameters. This often limits the use of
sophisticated soil models in practice, because their parameters cannot be readily
derived from standard laboratory or field tests.
The aim of this chapter is to give a brief description of standard and special
laboratory and field tests and to review the soil parameters that can be derived from
each of them. Due to restrictions on space it is not possible to give a detailed
account of how the parameters are deduced from the raw laboratory and/or field
data. For this information the reader is referred to the specialist texts on this
subject. It is recognised that a good site investigation should combine the strengths
of both laboratory and field testing.

2 / Finite element analysis in geotechnical engineering: Application

1.3

Laboratory tests

1.3.1

Introduction
A laboratory investigation is a key feature of almost all geotechnical projects. The
activities in the laboratory can be divided into several groups:

- Soil profiling, which involves soil fabric studies, index tests (e.g. water content,
grading, Atterberg limits), chemical and organic content, mineralogy, etc.;

- Element testing to characterise the mechanical behaviour of the soils (e.g.

stress-strain properties, yield, strength, creep, permeability, etc.);

- Derivation of parameters for empirical design methods (e.g. CBR for road
pavements);

- Physical modelling (e.g. centrifuge tests, calibration chamber tests, shaking

table tests)
For the purpose of deriving parameters for constitutive models, element testing
is the most appropriate investigation. However, limited use is also made of the
results from soil profiling. The aim of these element tests is to measure, as
accurately as possible and in a controlled manner, the response of a soil element
to imposed changes in stresses, strains and/or pore pressures. This ability of
accurate control and measurement of soil response is the main advantage of
laboratory over field testing. On the other hand, the main disadvantage of
laboratory element testing is the difficulty in obtaining undisturbed samples (i.e.
samples with preserved initial fabric and state). However, this can be overcome,
to some extent, by the use of advanced sampling techniques (e.g. Lava1 samplers,
Hight (1993))). Difficulties,however, still remain ifsandy and weakly bonded soils
are to be sampled.
The essential requirement for element testing is to simulate field conditions as
closely as possible. This involves simulation of:

Initial stresses (e.g. state of consolidation);

Imposed stress changes;
Sequence and rate of changes;
Field drainage conditions.

If a natural material is tested, experiments are normally performed on a set of soil

samples from different depths across a particular site, so that a full range of soil
response parameters can be collected. In a similar manner, if reconstituted soil is
tested, a group of samples is normally prepared with an identical set-up procedure
and tested at different initial stress levels so that again a full range of material
response parameters can be obtained.
The following paragraphs of this section will describe the majority of soil
laboratory equipment currently available and the experiments that can be
performed in it.

Obtaining geotechnical parameters 1 3

1.3.2 Oedometer test

A standard oedometer apparatus (see
Figure 1.1) consists of a circular
metal ring containing a soil sample
which is usually 70mm in diameter
and 20mm high. Porous discs are
placed at the top and bottom end of
the sample, thus allowing free
drainage of the sample in the vertical
direction. Radial drainage is not
possible because the metal ring is
impermeable. The sample is loaded
only in the vertical direction, via the
Porous stones
Sample
top platen, by applying increments of
load until a desired stress level is Figure I . I: Schematic presentation
reached. After each load increment is
of an oedometer apparatus
applied sufficient time is usually
allowed for full dissipation of any excess pore water pressures.
The complete stress state of a sample in the oedometer apparatus is not known.
The only known stress is the axial total stress, o;. Radial stress and pore water
pressure are not normally measured in the standard oedometer. Regarding strains,
the axial strain E, is measured as the displacement of the top cap divided by the
sample height, while the lateral deformation does not exist because of the
confinement imposed by the metal ring (i.e. &,=O). Therefore the total volumetric
strain of the sample equals the axial strain (i.e. E,,=&,). Due to these constraints
samples are said to undergo one dimensional compression.
Results from oedometer tests on clay soils are normally presented in the form
of a void ratio-axial effective stress (i.e. e-logo,') diagram, see Figure 1.2. As noted
above, sufficient time is allowed after the application of each increment of load to
ensure full dissipation of any excess pore water pressures. Consequently, at this
stage o~,'=o,,.With increasing axial effective stress the volume of a sample (i.e. its
void ratio) reduces and the loading curve from the initial sample conditions follows
the path 'abcd', see Figure 1.2. This path is usually assumed to become straight
once the pre-consolidation pressure is exceeded (i.e. o,'>(o,'),) and is called the
virgin compression line. The slope of this line is known as the compression index
C,, which is calculated as:

To examine the swelling potential of a soil, the oedometric sample can be

unloaded axially, which causes an increase in the void ratio (i.e. path 'ce' in Figure
1.2). In the same e-logo,' diagram this line is called a swelling line and again is
usually assumed to be straight, with a gradient C, which is called the swelling
index.

4 / Finite element analysis in geotechnical engineering: Application

The pre-consolidation e
pressure is usually associated
Virgin consoi~dat~on
with the maximum previous
vertical effective stress that the
sample has ever been
he
subjected to. Hence the
original overconsolidation
ratio of the sample, OCR, can
be calculated by dividing(onl),
by the vertical effective stress
Figure 1.2: Typical results from an
existing in the field at the
oedometer test on clay soil
location the sample was taken
from. It should be noted that for struchred soils this approach may lead to an
overestimate of OCR, as the process of ageing has the effect of increasing the stress
associated with point 'b'.
To be able to use the oedometer data in general stress space, its is necessary to
express the gradients of the compression and swelling lines in terms of invariants.
By knowing the ratio of the radial to axial effective stress in one dimensional
compression (i.e. K,, = orf/o,'), it is possible to estimate the mean effective stress
p' (=(otI1+2K,,o,,')/3) and replot compression and swelling behaviour in terms of
specific volume, v (=l+e), and mean effective stress, p1(i.e. v-lnp' diagram). The
gradient of the virgin compression line is then usually denoted as A and is
calculated as:

The slope of the swelling line, K, in the same diagram is calculated in a similar
manner. These two parameters, A and K, are essential for critical state type models,
such as modified Cam clay, bounding surface plasticity and bubble models (see
Chapters 7 and 8 of Volume 1).
The difficulty is estimating the value of K, which is not measured in the
standard oedometer test. When the soil is on the virgin compression line it is
termed normally consolidated and it is often assumed that K,=K,N"=(l -sinpl) (i.e.
Jaky's formula), where p' is the angle of shearing resistance. If v' is known, then
K,, and hence p' can be estimated. As K T is approximately constant along the
virgin compression line, then if this line is straight in e-logo,' space it will also be
straight in v-lnp' space. In fact, it can be easily shown that CC=2.3R.However, on
a swelling line K,, is not constant but increases as the sample is unloaded. While
":,
varies, there is no
there are some empirical equations describing how K
universally accepted expression. Consequently it is difficult to calculate p' on a
swelling line. In addition, if K,OCvariesand the swelling line is straight in e-logo,,'
space, then it will not be straight in v-lnp' space and vice versa. Clearly the
determination of K from an oedometer test is therefore difficult and considerable
judgement is required.

Obtaining geotechnical parameters / 5

The third parameter, which is needed to define the position of the virgin
compression and swelling curves in a v-lnp' diagram, is the value of the specific
volume at unit mean effective stress or, for the MIT models, at p'=100kPa. Care
must be exercised here as usually this parameter is required for the virgin
compression line obtained under conditions of isotropic compression (i.e. no
deviatoric stress). As noted above in the oedometer test the horizontal and vertical
stresses are unlikely to be equal (i.e. K,+ l), consequently the virgin consolidation
line, although parallel to the equivalent line obtained under isotropic conditions,
%willbe shifted towards the origin of the v-lnp'diagram. For most constitutive
models it is possible to transform the values of the specific volume on the
oedometer virgin consolidation line to the appropriate values on the isotropic
virgin consolidation line.
To complicate matters even further some of the more advanced constitutive
models plot isotropic consolidation data in Inv-lnp' space and assume that the
virgin consolidation and swelling lines are straight in this space, for example the
'41Tabbaa and Wood model described in Section 8.9 of Volume 1 of this book.
The results from the oedometer test can also be used to estimate the coefficient
of permeability in the vertical direction, k,. If, after each increment of load is
applied, the change in sample height is monitored with time this data can be
combined with one dimensional consolidation theory to estimate k,, (Head (1994)).
For sands the behaviour in the
e
oedometer test is more complex, as
the initial density of the sample
affects its behaviour. If two samples
of the same sand are placed in an
oedorneter, one in a loose and one in
a dense state, their behaviour under
increased vertical loading might be as
indicated in Figure 1.3. As the
samples have different initial densities
they will have different initial voids
ratios. With increasing vertical
Typical result from an
effective stress the samples will Figure
initially follow different normal
oedometer test on sand soil
compression lines. At some point
these normal compression lines will merge with a unique virgin compression line,
see Figure 1.3. This will happen at a lower vertical effective stress the looser the
sample. The magnitude of the vertical effective stress at which the virgin
compression line is reached is much higher than for clay soils and is often
considerably larger than the stress levels experienced in many practical situations.
If unloaded at any time the sample will follow a swelling/reloadingline. Behaviour
is therefore characterised either as being on a normal compression line, on the
virgin compression line or on a swellingtreloading line. These lines are often
assumed to be straight in either e-loga,', v-lnp' or Inv-lnp' space.

6 1 Finite element analysis in geotechnical engineering: Application

The data from the oedometer test can also be used to calculate the constrained
modulus as:

If the soil is assumed to be isotropic elastic, this can be equated to the one
dimensional modulus to give:

If a value for the Poisson's ratio p is assumed, it is possible to estimate the

Young's modulus E from E,.

1.3.3 Triaxial test

T? loading frame
The triaxial apparatus
is the most widely used
.piece of laboratory equipment for soil testing. It
is described in detail by Bishop and Uenkel
(1962) in their standard text book on triaxial
testing of soil. A more advanced version of the
triaxial apparatus is the stress path cell described
by Bishop and Wesley (1975).
A conventional triaxial apparatus (Figure 1.4)
incorporates a cylindrical soil sample which has
a diameter of either 38mm or 100mm. The larger
diameter apparatus is usually employed for
testing natural clays, because of the fissuring that
is often present in these materials. With a smaller
diameter sample this feature might be missed.
The sample is enclosed in a thin rubber
membrane which is sealed at the top and bottom
platens by rubber 0-ring seals. The membrane
Back pressure
gives flexibility to radial deformation of the Cellpxssm
sample. It also separates pore pressures
Figure 1.4: Schematic
generated inside the sample from total radial
stresses applied to the outside of the sample.
presentation of a triaxial
The sealed sample is placed on a pedestal in
apparatus
a water-filled cell. An all-around cell pressure,
a,, applies radial total stress, a,, to the vertical sides of the sample and a uniform
vertical stress to the top rigid platen, see Figure 1.5. An additional axial force, F,,,
is applied to the top platen via a loading frame. If the cross-sectional area of the
sample is A , then the total axial stress on the sample is:

Obtaining geotechnical parameters / 7

It is important to note that the force in the
loading frame is not equivalent to the axial
stress, but to the stress equal to (a,-a,), which is
known as the deviator stress, g. The pore water
pressures can be measured in the triaxiai sample
and therefore effective axial and radial stresses
can be evaluated:
0;

= 0, - p j

0;= 0, - pj

(1.6)

Pore pressure can be measured in the end

platens adjacent to the bottom andlor top end of
the sample, or by a probe placed at
approximately mid-height of the sample.
act;ng
Originally axial strains were deduced from ~i~~~~ 1.5: stresses
measurements of axial displacements made with
on a tr;axia,
dial gauges positioned outside the triaxial cell
and therefore remote from the sample. However, strain measurement has advanced
in the last twenty years so that it is now possible to attach instruments on the
sample itself. This avoids errors associated with the compliance of the testing
system. Both axial and radial strains can be measured in this way.
Triaxial apparatus can be used to perform several different tests. Some of the
more common tests are:

- If o;, and ar are increased together such that the radial strain c,=O, the sample
will deform in one dimensional compression, similar to oedometer conditions.
The advantage over the oedometer test is that, by measuring pore water
pressure and radial stress, effective stresses can be calculated and the
coefficient of earth pressure at rest, K?, estimated as:

In a similar manner, if rr, and a, are reduced together such that zr=O, the
coefficient of earth pressure in overconsolidation, K?, can be estimated for
different OCRs. The results of such one dimensional compression and swelling
tests can be plotted in a e-loga,' diagram and parameters C, and C, estimated
as in Equation (1.1). The total volumetric strain, E,, in both cases equals the
axial strain, E,. It should be noted that these values can only be related to insitu conditions if the OCR is known. An alternative procedure for estimating
the value of K,, from undisturbed sample is discussed subsequently.
If the loading frame is locked such that it is not in contact with the top platen
of the sample and the sample is loaded only with an all-around cell pressure
(i.e. o,=or), it is deforming under isotropic compression. If the all-around cell
pressure is reduced, the sample is said to undergo isotropic swelling. The

8 1 Finite element analysis in geotechnical engineering: Application

results of such tests can be plotted in a v-lnp' diagram and used to obtain values
for the parameters A and K, see Equation (1.2).
After initial isotropic or anisotropic compression to a certain stress level,
samples can be subjected to either drained or undrained shearing. In this case
the total radial stress is held constant, while the axial stress is either increased
or decreased. Ifo;>o, the sample is undergoing triaxial compression and in this
case a,,=a, and a,=ae=q=a,. The parameter b (=(a,-a3)l(o,-a,)), which
accounts for the effect of the intermediateprincipal stress, is therefore zero. On
the other hand, if the sample is axially unloaded, such that au<ar,it undergoes
triaxial extension and in this case a,=q and o;=a,=a,=q, resulting in b=I .O. In
Volume 1 of this book the Lode's angle B was used to express the relative
magnitude of the intermediate principal stress. Using Equation (5.3) of Volume
1 it can be shown that 8=tan-,[(2b- 1)1J3] and consequently in triaxial
compression b=O and I%-30" and in triaxial extension b=l and e 3 0 " .

The results from both drained and

g(t)i
undrained triaxial shearing are g
normally plotted in terms of stressstrain diagrams (g-E,) and stress path
E.
diagrams (g-p' or t-S', where
:,I,,
t=(a,,'-arf)/2 and s'=(o,'+a,')/2), see
:,'
I,'
Figure 1.6. In addition, for undrained
P' 6')
shearing the change in pore water
pressure, pf, is usually plotted against
axial strain, E,, , whereas for drained Figure 7.6: Schema tic presentation
shearing the volumetric strain, E,,,,,, is of stress paths and stress-strain
curves from a triaxial test
plotted against axial strain, E,. Both
clay and sand soils may experiencepeak deviatoricstress, before softening towards
a constant deviatoric stress which doesn't change with fbrther straining. This state
is known as the critical state for a particular material. Parameters of importance for
modelling and design are values of the angle of shearing resistance at peak, p,,',
and at critical state, p,,', and they can be different in triaxial compression and
triaxial extension. They are calculated from the known stress state at peak and
critical state as:

E.

Another parameter of interest in geotechnicalpractice is the stiffness of the soil.

From triaxial tests both drained, E', and undrained, E,, , Young's moduli can be
estimated in both triaxial compression and extension. From stress-stain curves
these parameters can be calculated as either secant or tangent values, see Figure
1.6:

Obtaining geotechnical parameters 1 9

A

The triaxial test provides a complete

history of the degradation of soil (E")
stiffness with increase in strain level,
as shown in Figure 1.7. If the
resolution of the local strain
instrumentation is sufficientlyhigh so
that it can detect the initial elastic
behaviour
curve will of
have
a material,
an initialthehorizontal
stiffness

\\\

E,

plateau associated with the maximum

Figure 1.7.- Soil stiffness curve
stiffness value. This small strain
behaviour is an essential part of constitutive modelling if soil deformations are of
interest in the analysis, because soil stiffness is much higher in the early stages of
deformation than it is at large strains. Ignoring this can result in the prediction of
patterns of movement considerably different to those observed in the field, see
Chapters 2 and 3.
If the small strain Young's modulus, E',,, , can be obtained, the other elastic
parameter that can be calculated from a drained triaxial compression test is
Poisson's ratio p,'for straining in the radial direction due to changes in axial
stress, which is calculated as:
p;r = --A&r

Ago

The ability to measure the radial strain to a very high resolution is essential for this
parameter to be estimated.
I-lowever, instead of using parameters E' and p, soil elasticity can be expressed
in terms of shear modulus G and bulk modulus K, see Section 5.5 of Volume 1.
Both can be calculated from triaxial compression or extension tests, although K
only from a drained test:

where:

AJ is the incremental invariant deviatoric stress:

AE, is the incremental invariant deviatoric strain:
AEd = *y((&, - &2)2 +(% - s3)2 +(E, - c3)2 = A(&"Er)

JS

and

where:

hp' is the incrmental mean effective stress and

Ap' = (Am,' + 2 A o ; ) 13 = (ACT;+ 2 A q : ) 1 3

10 / Finite element analysis in geotechnical engineering: Application

AE,, is the incremental volumetric strain and

A&, = AE, + ~ A E ,= AE, + 2A.5,
Again, the complete degradation of shear and bulk stiffness from small to large
strains can be obtained from a triaxial test.
It is worth noting that it is not possible to obtain many anisotropic stiffness
parameters from a triaxial test. For example, if the parameters for the linear
transversely isotropic model were required, see Section 5.6 of Volume l , then only
the values of the axial Young's modulus, E,=AIS,'/AE, and Poisson's ratio
p,,.==A&JAt;,
can be obtained from the standard tests described above. If dynamic
probes such as bender elements, see Section 1.3.10, are mounted on the sides and
ends ofthe sample it is possible to obtain estimates of a further two parameters, but
it is still not possible to obtain a value for the fifth parameter.
The angle of shearing resistance and stiffness parameters are essential for
nearly all of the currently existing constitutive models.
Although not common, the triaxial cell can be used to obtain an estimate of insitu values of K,, in clay soils. Usually, when soil samples are taken during a site
investigation, they are immediately sealed to ensure no transfer of water from or
to the sample. This seal is only broken when a smaller test sample is required for
testing. If it is assumed that during the whole sampling and storage process, from
extraction from the ground until placement in the triaxial cell, the sample remains
undrained, then the sample will retain its mean effective stress,pkl.The sample on
placement in the triaxial cell will have a zero mean total stress applied to it and
consequently the initial pore water suction will be equal to p,'. If a cell pressure
(i.e. U,) is then applied to the sample until a positive pore water pressure is
recorded, (note that most pore water pressure transducers used for triaxial testing
cannot measure suction), the initialp,' can be calculated as the difference between
the applied cell pressure and measured pore water pressure. This can then be used
to determine the in-situ value of K,, using Equation (1.13) (see Burland and
Maswoswe (1982)):

where A , accounts for the change in pore water pressure due to the reduction in
deviatoric stress which occurs during sampling in the field. A value of A,=1/3
corresponds to an isotropic elastic material, whereas a stiff clay (e.g. London clay)
typically has a value of approximately 112; 0,' is the original vertical effective
stress in the sample in-situ and can be estimated knowing the depth of the sample,
the bulk unit weights of the overlying materials and the pore water pressure.
Accurate estimates of K,, from this approach require that the sample remains
undrained at all times. On setting up the sample in the triaxial cell care must be
exercised so that it does not come into contact with water. This can be difficult as
it is usual practice to saturate the porous stones at the top and bottom ofthe sample
and consequently special testing techniques are required, see Burland and

Obtaining g60technical parameters 1 1 1

Maswoswe (1982). The approach also neglects any changes in pore water pressure
that occur when the sample is forced into and out of the sampling tube in which it
is stored.
The triaxial cell can also be used to obtain estimates of permeability. A sample
can be subjected to a cell pressure, a,, and then a difference in pressure applied
between the drainage lines at the top and bottom platens. This will promote flow
of water through the sample. By knowing this pressure difference and measuring
the quantity of water flowing through the sample in a set time, once steady state
conditions have been reached, enables the permeability to be estimated using
Darcy's law. By performing a series ofpermeability tests at different cell pressures
enables a relationship between permeability and voids ratio (or mean effective
stress) to be determined. As the time required for these tests depends on the
drainage path length, which in turn is related to the sample height, short stubby
samples are usually used.
For soils with a low permeability (i.e Ko<lO-'mlsec) it is also possible to
estimate the permeability k from a dissipation stage of a triaxial test. This involves
applying a change in cell pressure quickly and then allowing consolidation during
which time the decay in excess pore water pressure with time is monitored. One
dimensional consolidation theory can then be used to estimate k.

1.3.4 True triaxial test

In contrast to a conventional triaxial
apparatus, the true triaxial equipment tests a
cubical soil sample. The difference between
the two experiments is that in the true triaxial
equipment all three principal stresses, a , , a,
and q , can be applied and controlled
independently, see Figure 1.8, while in a
conventional triaxial cell two principal
stresses are always equal (i.e. ar=a,).
The advantage of this experiment is that
the values of the angle of shearing resistance,
v', can be obtained for the full range of b
between 0 and 1, while conventional triaxial
tests with a cylindrical sample can only have
b=O (triaxial compression) or b=l(triaxial
Figure 1.8: Stresses
extension). This is useful if the value of v' is controlled in a true triaxia/
needed for conditions in the ground other
apparatus
than triaxial (for example plane strain, for
which 0.2<b<0.4, depending on soil type).
It is also possible to measure the soil stiffness in three directions. For example
it is possible to obtain estimates of four of the five parameters necessary to define
the linear transversely anisotropic model. If dynamic probes (i.e. bender elements)

1 2 / Finite element analysis in geotechnical engineering: Application

are installed in the sample it is also possible to obtain an estimate for the fifth
parameter.
While clearly an advantage over the conventional triaxial equipment, it is not
easy to obtain and set up the samples. Consequently true triaxial apparatus are
rarely used for commercial testing. Only a limited number of apparatus exist and
these are mainly located in universities.

1.3.5 Direct shear test

The direct shear test is performed in
F, Movable top part
the s h e a r box a p p a r a t u s ,
of the sample
schematically illustrated in Figure 1.9.
The sample is typically 6Ox60mm in
plane and about 25mm high. It is
placed between rigid and rough top
and bottom platens, in a square box
which is split horizontally. Vertical
load is applied via the top cap and is
ShearIplane
Fixed
h&mpfut
normally kept constant during the
of
the sample
test. A horizontal force is applied to
the top half of the box with the
Figure 1.9: Schematic presentation
bottom half fixed, thus forcing the
of a direct shear box apparatus
two halves of the box to move
horizontally relative to each other.
Rough top and bottom platens transfer
shear stress into the soil. The main
purpose of the test is to examine soil
strength. It is unsuitable for
measuring soil stiffness because of
the non-uniformities imposed in the
sample by the loading arrangement.
The only known stresses on the
sample are the average vertical stress
and the average shear stress, see
Figure 10- The
a Figure 1. 10: Stresses acting on a
shear stress causes the major principal
in a direct shear box
stress, a, , to rotate continually away
from the vertical, but there is no means of measuring this rotation. Because the
sample is confined in a box, it is subjected to significant stress and strain nonuniformities during shearing.
Interpretation ofthe direct shear box is based on assuming that the deformation
of the sample conforms to that of ideal simple shear, see Potts et al. (1987). Such
an idealised condition is indicated in Figure 1.1 1 which shows the top boundary
of a layer of soil displaced parallel to the bottom boundary, under plane strain

Obtaining geotechnical parameters / 1 3

conditions. The behaviour of a typical
element is indicated. The soil is
nent v
subjected to a prescribed shear strain
yxy and is constrained to have zero
soil
direct strain in the x direction (&,=O).
As a result of these prescribed
boundary conditions, the soil is
subjected to a shear stress 5 ,a direct
strain E, and a change in stress in the
x dirkction (Aax*O). Because
Figure I . I I: Simple shear
horizontal planes remain horizontal
and are inextensible (cX=O), at
conditions
ultimate plastic failure these planes
/
'
are a set of velocity characteristics
(Davis (1968)). Such planes are ?_
inclined at f (45O- v/2) to the direction
of the major principal plastic strain
increment. In an isotropic elastoplastic material the directions of the
principal stresses and the principal
plastic strain increments coincide.
Consequently at failure, when all
deformation is plastic, any horizontal
plane is a velocity characteristic and Figure I. 12: Mohr's circle of stress
is inclined at f(45'-vI2) to the
at ultimate conditions
direction of the major principal stress
g, and hence is inclined at f(45O+v/2) to the plane on which o, acts. The Mohr
circle of stress for this state is shown in Figure 1.12. Point P is the pole ofthe circle
and point H represents the stresses (r,), and a,' which act on the horizontal plane.
From the geometry of the circle it can be shown that:

'1

($lf

sin p' cos v

= I-sinplsinv

which reduces to (rXJay')~sinq1

when v=O and (tX,/oy')ptanqfwhen v=q'. It should
be noted that (rxJa,,')~sinpris the interpretation advocated by Hill (1 950), whereas
(rXJoyr),=tany' is the more common interpretation. In reality, due to the nonuniformities inherent in the direct shear box test, it is unlikely that the condition of
idealised simple shear exists throughout much of the sample.
The usual presentation of direct shear results is in terms of a shear stresshorizontal displacement diagram (i.e. r-d), see Figure 1.13. The shear stress
normally increases up to a peak value during the early stages of horizontal
displacement, and then gradually decreases until reaching a residual value after a
large horizontal displacement of the shear box. From this test both peak and
residual angles of shearing resistance can be evaluated, although to obtain the latter

14 / Finite element analysis in geotechnical engineering: Application

''
several reversals of loading may be
necessary. Residual strength can be p,.,
very important in some boundary value
problems involving high plasticity
clays, see Chapter 4. Because clay I-'
particles are elongated and, in their
natural condition, oriented in some
D~splacement
structure, after large deformations they
can become aligned with the direction Figure 1.13: Typical shear stressof shearing. such a situation could displacement diagram from direct
shear box test
occur, for example, during a landslide.
Low plasticity clays and granular materials have a less pronounced drop from
critical state to residual strength.
The direct shear box apparatus is also used for interface shear testing. In this
situation soil is sheared against some other material (e.g. concrete, steel, etc.) in
order to examine the angle of friction, 6, at the interface of the two materials. This
parameter is used, for example, in pile or retaining wall analysis, where it is
necessary to know the maximum value of the angle of friction between the pile or
wall material (concrete or steel) and soil (Jardine and Chow (1996), Day and Potts
(1998)). For such investigations the interface between the soil and the material
against which it is to be sheared is positioned at the split between the upper and
lower halves of the shear box.

1.3.6 Simple shear test

The simple shear apparatus works on a similar principle to the direct shear box in
that the top platen is moved horizontally with respect to the bottom platen.
However, the apparatus is designed so that the sample is allowed to deform more
uniformly. There are currently two designs of simple shear apparatus: one
developed at Cambridge University and the other at the Norwegian Geotechnical
Institute (NGI) in Oslo. The NGI device is the more common design and can be
found in many laboratories, both academic and commercial, around the world.
The NGI device is schematically drawn in Figure 1.14a. It has a cylindrical
sample, 80mm in diameter and about lOmm thick, contained within a wirereinforced rubber membrane. This membrane maintains a constant sample
circumference, but allows uniform vertical deformation and rotation of the vertical
sides of the sample. The whole sample is placed between top and bottom platens.
Axial and shear forces are applied via the top platen. Usually the axial force is kept
constant, while the shear force increases during the test, similar to the direct shear
box test. The vertical and shear strains of a sample can be measured by observing
the vertical and horizontal displacement of the top platen. Although the pore
pressure can be measured, the full stress and strain states of the sample are not
known in the NGI device and again the results are interpreted assuming ideal
simple shear conditions, as discussed above in Section 1.3.5.

Obtaining geotechnical parameters / 15

I m"

'~

Top platen

tom

Top and bonom

Bonom platen

u e
pressure

a) NGI simple shear device

b) Cambridge simple shear device

Figure l. 14: Schematic presentation oP a) NGI

and bl Cambridge simple shear devices

The Cambridge University device is schematically presented in Figure 1.14b.

The sample here is a rectangular prism, 100 by lOOmm in plan and about 20mm
thick. It is placed within rough rigid top and bottom platens which allow vertical
deformation and rotation of the side platens. From the measurements of these
external deformations it is possible to estimate the average strains within the
sample. As the side platens are designed to allow vertical movement (i.e. they are
smooth) they do not provide complementary shear stresses on the vertical
boundaries of the sample. Consequently the stress and strain state is not uniform
within the sample. Usually normal and shear stress transducers are placed in the
platens and stresses are measured over the middle third of the sample. As it is more
difficult to obtain, test and instrument these prismatic samples only a few pieces
of equipment of this type exist.
Both types of simple shear apparatus are mainly used for measuring soil
strength (i.e. undrained strength or angle of shearing resistance). Due to the more
uniform behaviour of the sample in the Cambridge apparatus, it can also be used
to obtain pre-failure stress strain data.
1 -3.7 Ring shear rest
The ring shear apparatus is another device suitable only for measuring soil
strength. The principle of operation is similar to the direct shear box apparatus,
except that the sample has the shape of a squat hollow cylinder and the top half of
the box is rotated with respect to the bottom. The sample, about lOOmm in outer
diameter, about 20mm thick and of 15mm wall thickness (see Figure 1.15), is
placed within rough top and bottom platens in a horizontally split metal ring.
Constant vertical load is applied via the top platen, while the bottom part of the
ring is rotated with respect to the top part, thus applying a horizontal shear stress
and creating a horizontal slip surface in the sample. No other stresses or strains can
be measured in the standard ring shear apparatus and consequently interpretation
of the results is difficult.
The results from a ring shear test are normally plotted in terms of shear stress
versus displacement (i.e. 7 - 4diagram, as for the direct shear box, see Figure 1.13.

1 6 / Finite element analysis in geotechnical engineering: Application

The shear stress can experience a peak

value (i.e. associated with q,,'), after which
the soil strength drops and reaches a
residual value (i.e. (ore,'). Again strength
parameters are determined assuming
behaviour is similar to ideal simple shear.
The difference between this test and shear
box test is that much larger displacements
can be achieved in a ring shear device (e.g.
several metres) and therefore a better
estimate of the residual strength can be
obtained. Another advantage of the ring
Figure 1.15: Schematic
shear apparatus is that it can be used to
presentation
of a ring shear
examine the effects of very high rates of
apparatus
shearing on residual strength, which is
useful when investigating soil behaviour
under earthquake loading.
The ring shear apparatus can also be used for interface testing.

1.3.8 Hollow cylinder test

From the previous discussion it can be concluded that in the oedometer, triaxial
and true triaxial apparatus the principal stress and strain directions applied to the
sample are parallel with the sarnpIe boundaries. It is only possible for the relative
magnitudes of the principal stresses and strains to change. Their directions are
fixed. In contrast the direct, simple and ring shear apparatus allow rotation of the
principal stress and strain directions, but because not all of the stress components
can be measured it is not possible to measure the magnitude of the principal
stresses, or more importantly control their directions. The only exception might be
a highly instrumented Cambridge type simple shear apparatus, where sufficient
load cells are used to enable sufficient stress components to be measured to
calculate the principal effective stresses, however, it is still not possible to control
or alter the direction of the principal stresses, as this is implicitly set by the manner
in which the apparatus works.
The hollow cylinder apparatus does not suffer from the above shortcomings,
as it allows full control of both the magnitude and the rotation of principal stresses.
It is therefore extremely suitable for investigating soil anisotropy. The sample is
a hollow cylinder (see Figure 1.16) which can have various dimensions. For
example, the large and small hollow cylinder apparatus in the Imperial College
laboratory test samples with the following dimensions: the large one is 25cm high,
25cm outer diameter and 2.5cm wall thickness; the smaller device is I9cm high,
lOcm outer diameter and l .5cm wall thickness.
Four independent loads can be applied to a hollow cylindrical sample (see
Figure 1.16), as opposed to only the one, two or three that can be applied to

Obtaining geotechnical parameters / 17

samples in the other laboratory Hollow cylhder coordinates:
devices. These loads are the inner
cell pressure, p, , the outer cell
Element component
pressure, p,, the vertical load, W,
stresses:
and the torque, M,. This
combination of loads allows
control of four component stresses,
namely the normal stresses a:, o;
and on and the torsional shear
stress r,, (note the remaining shear
Element principal
stresses rrzand rrHare zero). This
then permits control over the
magnitudes and directions a of the
~rincipalstresses. Because all three Element component
major stresses can be controlled, it
strains:
is also possible to independently
control the parameter b. So, with a
a
0,-a
well designed hollow cylinder
b =-L
apparatus it is possible to perform
=1-~3
a set of shear tests with a constant
value of b while changing a, and
vice versa, with a constant value of Figure 1.16: Stresses and strains in a
hollow cylinder apparatus
a while changing b, thus enabling
an independent evaluation of the
effects of each of these two parameters on soil strength, stiffness, yielding, etc.
All component strains can be measured locally in a hollow cylinder device: the
three normal strains E,, and E,, and the torsional shear strain yzo, so that the strain
state in the sample is completely known (note yr,=yrn=O).
Because of the flexibility of a hollow cylinder apparatus, a variety of stress
paths can be applied to soil samples. However, for design purposes its use is
usually most valuable for investigating initial soil anisotropy, where samples are
initially K,, consolidated to a certain stress level (thus simulating initial green-field
conditions in the ground after the sedimentation/erosionprocess) and then sheared
with different orientations of the major principal stress (i.e. a set of samples where
each sample is sheared with a different a value in the range from 0' to 90'). The
results from these tests are normally plotted in terms of stress-strain (J-E,) and
stress path (J-p') diagrams, the latter being a two dimensional projections of the
three dimensional J-p'-a stress space, see Figure 1.17. This figure shows results
from a typical series of hollow cylinder tests performed on a granular soil (see
ZdravkoviC and Jardine (2000)). In particular, the results show how the peak
deviator stress reduces with an increase of a. Typical results derived from such a
set of experiments gives the variation of both drained, p,,', and undrained, S,,,soil
strength with a, shown in Figure 1.18. These results clearly indicate a degree of
soil anisotropy.
B

72 / Finite element analysis in geotechnical engineering: Application

excavation inducing a lengthening
horirond diameter
of the vertical diameter and
Lower tunnel distodon due to passage of upper:
shortening of the horizontal
0.3 o venial diameter
diameter. It is clear that the closer
o.2
o
diameter . horirornldiameter
the two tunnels, the greater the
induced distortion. For all the :V
analyses, the lengthening of the
vertical diameter is greater than .g' -.o.,
*.;.:,.'
the associated shortening of the
horizontal diameter, and the
4.2 -;
sholfening
induced distortions are never as
I
I
0
2
4
6
8
1
0
severe as those for side-by-side
Pillar depth (number of diameters)
tunnels at the same pillar
dimension (the curves for sideFigure 2.43: Deformation of Is' tunnel
by-side tunnels are reproduced
in response to 2ndtunnel for piggy back
from Figure 2.42). Sensible
geometry
extrapolations of the numerical
predictions for the distortion to the upper tunnel on passage of the lower tunnel are
in good agreement with the field data from Kimmance et al. (1996). Not
surprisingly this construction sequence causes more distortion than the alternative
of the second tunnel passing above the existing tunnel. Indeed, it is clear that there
is very little influence when the second tunnel passes above the existing tunnel, and
for pillar depths greater than 3 diameters the influence is effectively zero.
The results presented here demonstrate that the relative position (above, below,
or to the side) and the physical spacing ofthe tunnels has a significant effect on the
tunnel lining response. The results presented quantify the expected distortion
induced in an existing tunnel lining when an adjacent tunnel is excavated.
Addenbrooke and Potts (2001) draw further conclusions with respect to the surface
settlement profiles expected above twin tunnel projects, and the influence of the
rest period between construction of the two tunnels.
V

.l!
4:

2.8

Summary

1. Tunnel construction is a three dimensional engineering process. If restricted to

two dimensional analysis, then one must consider either plane strain or axially
symmetric representations, depending on what the analysis aims to achieve.
2. Methods of simulating tunnel construction in plane strain require at least one
assumption: the volume loss to be expected; the percentage of load removal
prior to lining construction; or the actual displacement of the tunnel boundary.
3. If severe distortions of the tunnel lining are expected (i.e. in response to the
passage of an adjacent tunnel in close proximity) then a model can be used
which allows segmental linings to open or rotate at their joints, or allows
sprayed concrete linings to crack.
4. The recognition of permeability's dependence on stress level has led to
nonlinear models for permeability. Using such a model in place of a linear

Tunnels 1 73

5.

6.

7.

9.

alternative will alter the long term pore water pressure regime, for the same
hydraulic boundary conditions. This will alter the ground response during
consolidation and swelling.
The intermediate and long term behaviour is governed by many factors. In
particular, whether the tunnel acts as a drain or is impermeable, and whether the
initial pore water pressure profile is close to hydrostatic or not. It is important
to be aware of the dependencies, and so to view any prediction of intermediate
and long term behaviour with a critical eye.
It is important to select constitutive models capable of reproducing field
behaviour. For example, in a situation where pre-yield behaviour dominates the
ground response, it is essential to model the nonlinear elasticity at small strains.
Devices for improving settlement predictions can be developed. These
questions must be asked: What is the influence of this adjustment on the soil
behaviour? What are the knock on effects? For example, if one is adopting a
device to match a surface settlement profile, how does this alter any prediction
of sub-surface movement, the pore pressure response, or the lining stresses and
deformations.
The finite element method can be used to quickly assess the impact of different
influences on tunnelling-induced ground movements. Parametric studies can
prove extremely useful in the development of design charts and interaction
diagrams.
One of the great benefits of numerical analysis to the tunnel engineer is that an
analysis can incorporate adjacent influences. For example existing surface
structures, or existing tunnels. It is also possible to reproduce the effects of
compensation grouting to protect surface structures during tunnelling projects.
This chapter has demonstrated the power of the finite element method in this
respect.

3. Earth retaining structures

3.1

Synopsis

This chapter discusses the analysis of earth retaining structures of various forms.
It draws heavily on issues that have to be considered when analysing real
structures. After a brief description of the main types of retaining structures in
current use, general considerations, such as choice of constitutive model,
construction method, ground water control and support systems, are discussed.
Attention is then focussed on the specific analysis of gravity, embedded and
reinforcedlanchored retaining walls.

3.2

Introduction

The purpose of an earth retaining structure is, generally, to withstand the lateral
forces exerted by a vertical or near vertical surface in natural ground or fill. The
structural system usually includes a wall, which may be supported by other
structural members such as props, floor slabs, ground anchors or reinforcing strips.
Alternatively, or additionally, the wall may be supported by ground at its base or
into which it penetrates. In most situations the soil provides both the activating and
resisting forces, with the wall and its structural support providing a transfer
mechanism.
The design engineer must assess the forces imposed on the wall and other
structural members, and the likely movements of both the structure and retained
material. Usually these have to be determined under working and ultimate load
conditions, see Chapter 1 of Volume 1 of this book. In addition, estimates of the
magnitude and extent of potential ground movements arising from construction of
the structure, both in the short term and in the long term as drainage within the
ground occurs, are required. This may be because of the effect construction may
have on existing, or planned, services or structures (buildings, tunnels, foundations,
etc.) in close proximity. Potential damage could occur which has to be assessed and
methods of construction considered which minimise these effects.
Design of retaining walls has traditionally been carried out using simplified
methods of analysis (e.g. limit equilibrium, stress fields) or empirical approaches.
Simplified methods have been developed for free standing gravity walls, embedded
cantilever walls, or embedded walls with a single prop or anchor. Some of these
are described in BS 8002 (1994) and Padfield and Mair (1984). Because of the

Earth retaining structures / 75

statically indeterminate nature of multiple propped (or anchored) walls, these have
often been dealt with using empirical approaches such as those suggested by Peck
(1 969). Simplified methods of analysis are also available for reinforcedlanchored
earth, for example see BS 8006 (1995).
Because all of these traditional design methods are based on simplified analysis
or empirical rules, they cannot, and do not, provide the engineer with all the
desired design information. In particular, they often only provide very limited
indications of soil movements and no information on the interaction with adjacent
structures.
The introduction of inexpensive, but sophisticated, computer hardware and
numerical software has resulted in considerable advances in the analysis and design
of retaining structures over the past ten years. Much progress has been made in
attempting to model the behaviour of retaining structures in service and to
investigate the mechanisms of soil-structure interaction. This chapter will review
some of these advances and discuss some of the important issues that must be
addressed when performing numerical analysis of earth retaining structures. It
begins by describing the different types of retaining structure and the general issues
which must be considered before starting an analysis. It then goes on to consider
the three main categories of retaining structures (i.e. gravity, embedded and
reinforcedlanchored earth walls) in more detail.

3.3

Types of retaining structure

3.3.1

Introduction
The complexity and uncertainty involved in design and analysis increase with the
degree of soil-structure interaction and thus depend on the type of retaining
structure to be employed. It is therefore appropriate to categorise the types of
retaining structure on the basis of the soil structure interaction problems that arise
in design. In Figure 3.1 the main wall
types are shown in order of increasing
complexity o f soil-structure
interaction.

3.3.2 Gravity walls

Gravity, counterfort and cantilever
walls are stiff structures for which the
soil-structure interaction is relatively
simple. For overall stability, the earth
forces on the back of a wall have to
be balanced by normal and shear
stresses at its base. The magnitudes of
these resisting forces are, to a large
extent, controlled by the weight of the

Gravity

Counterfort

Cantilever

Reverse cantilever

Figure 3.1: Types of retaining

structure (cont.. .)

76 1 Finite element analysis in geotechnical engineering: Application

structure, hence the term gravity wall.

Such walls can be formed of mass
concrete, reinforced concrete, or precast units such as crib walls or
gabions. Movement arising from
deformation of the ground beneath
the wall is often negligibly small, but
exceptions occur if the wall and
backfill are constructed on a deep
layer of compressible soil.

Soil nails

3.3.3

Reinforcedlanchored
earth walls
Embedded cantilever
Propped cantilever
Reinforced earth walls and walls
involving soil nails essentially act as
large gravity walls. However, their
internal stability relies on a complex
interaction between the soil and the
reinforcingelements (Or nails]. These
Multi-propped
walls generally have non-structural
level
membrane facing units which are
intended to prevent erosion, or are
purely aesthetic. The membrane has
to be designed to resist any bending
moments and forces that occur, but
this is not the primary method of earth
Multi-anchored
retention. The resistance to ground
movement is provided by soil nails,
Figure 3- 1: (...cant) Types of
anchors and ties of various forms (not
retaining structure
ground anchors) or strips of metal or
geo-fabric.

3.3.4 Embedded walls

When soil movements are important andor construction space is limited,
embedded cantilever walls may be used, with or without props or anchors. Sheet
pile walls, diaphragm walls, contiguous bored pile walls and secant pile walls are
all examples of this type of wall. To maintain stability the walls rely on the
resistance of the ground below excavation level and on the resistance forces
provided by any props or anchors. The flexibility of embedded walls vary within
a wide range and this has considerable effect on the distribution of earth pressures.
The more flexible walls often have smaller bending moments in the structural
elements, but may lead to larger deformations, particularly for embedded
cantilevers with no props or anchors. The complexity of soil-structure interaction

Earth retaining structures l 77

increases with the number of levels of props andlor anchors and hence the
structural redundancy.

3.4
General considerations
3.4.1 Introduction
Before starting any numerical analysis it is important to address a number of issues
to ensure that the most appropriate methods of modelling the soil and structure are
used. It is also important that the correct boundary conditions (e.g. displacements,
pore water pressures, loads, etc.) are applied in any analysis. This section briefly
outlines some of the more important issues. A number of these are discussed in
greater detail in subsequent sections.

3.4.2 Symmetry
In reality all geotechnical problems involving retaining structures are three
dimensional and, ideally, three dimensional analyses, fully representing the
structure's geometry, loading conditions and variations in ground conditions across
the site, should be undertaken. With current computer hardware this is not a
practical proposition for all, but a number of very limited and extremely simple
cases. To analyse any structure it is therefore necessary to make a number of
simplifying assumptions. Most commonly two dimensional plane strain or axisymmetric analyses are undertaken, see Chapter 1 of Volume 1 of this book.
For two dimensional and axi-symmetric
(E
analysis the assumption is frequently made
I
that there is an axis of symmetry about the
I
centre line of an excavation and that only a
'half section' needs to be modelled. In the
case of a three dimensional analysis two
planes of symmetry are often assumed and a
'quarter section' is considered. For example,
Figure 3.2 shows a plane strain excavation
supported by two parallel walls propped near
to the ground surface. If there is symmetry
I
about the vertical line passing through the
Of axicentre of the excavation, it is only necessary Figure 3.2:
symmetric
geometry
to analyse half the problem, either that half of
the problem to the right of the plane of symmetry or that half to the left. This
clearly reduces the size of the problem and the number of finite elements needed
to represent it. However, for such an analysis to be truly representative there must
be complete symmetry about the centre line of the excavation. This symmetry
includes geometry, construction sequence, soil properties and ground conditions.
In practice it is rarely the case that all of these have symmetry and therefore
analyses using a 'half section' usually imply further approximations.

78 / Finite element analysis in geotechnical engineering: Application

As an example consider the

situation shown in Figure 3.3.
This shows the cross section of a
road tunnel forming part of a road ,,
improvement scheme. The tunnel
is formed of two secant pile
retaining walls, a roof slab which
.,
""'Pi"
is connected to the outer walls by
a full moment connection, a base
slab which is not connected to the
walls and a central wall that also
provides SUPPO* to the roof slab.
Figure 3.3: Cross-section of a road
The central piles support the
tunnel
central wall, but neither are
connected to the base slab.
The road alignment at the location of the tunnel is on a bend and consequently
the tunnel roof and base slab are inclined to the horizontal. However, the fall across
the structure does not appear great, less than Im across the excavation which is
between 25 to 26m wide. Consequently, it is tempting to simplify the analysis by
ignoring the fall across the structure and assuming symmetry about its centre. Only
a 'half section' would then need to be considered in any analysis.
To investigate whether such an
approximation is reasonable, a finite
element analysis was undertaken
considering the full cross section of
the tunnel as shown in Figure 3.3.
Figure 3.4 shows the predicted
displaced shape of the two retaining
+-I
walls at an exaggerated scale (solid
L-1
profile; the undeformed profile of the
walls, as represented in the finite
L,:
element mesh, is shown as a dotted
line). It is apparent that there is some
sway of the tunnel structure and wall
L-1
displacements are not the same on
either side of the excavation. The top
I-.!
of the right hand wall of the tunnel is
pushed back into the soil behind it.
Figure 3.4: Displaced shape of walls
An analysis of a 'half section'
of road tunnel
would not have predicted this
asymmetrical behaviour. It would have predicted that both walls move
symmetrically inwards towards the centre ofthe excavation. Because the displaced
shape of one, or both, walls was not correct, the predicted bending moments and
forces in the walls would also have been incorrectly predicted.

[-:

Earth retaining structures / 79

Obviously, no analysis can model geometry in detail, or for that matter every
construction activity, but this example illustrates the care that must be taken when
making simplifying assumptions of any form.

3.4.3 Geometry of the finite element model

Another decision that arises when
performing a numerical analysis is the
choice of the depth and lateral extent i
of the finite element mesh. For
example, when considering the simple i
excavation problem shown in Figure
3.5. a decision must be made as to the i
depth of the mesh below the ground
'
surface and the lateral extent of the
EI= 2 . 3 * 1 0 h ~ ' 1 ~
right hand boundary of the mesh. As,
p =0.15
in this case, the excavation is
symmetric, the position of the left
hand boundary of the mesh is fixed.
3.5: Propped retaining wall
When considering the bottom
boundary of the mesh, the soil stratigraphy often provides an obvious location. For
example, the occurrence of a very stiff and strong layer (e.g. rock) at depth
provides an ideal location for the bottom boundary of the mesh. Also, because soil
strength and stiffness usually increase with depth, analyses are not so sensitive to
the location of the bottom boundary as long as it is not unreasonably close to the
bottom of the wajl. The location of the far field vertical (i.e. the right hand)
boundary is more problematic and will, in part, depend on the constitutive model
employed to represent soil behaviour.
To illustrate these points analyses
have been performed for the
excavation problem shown in Figure
3.5. A typical mesh is shown in
Figure 3.6. Also shown are the
displacement boundary conditions
imposed on the bottom and vertical
boundaries of the mesh. Both the soil
and the wall are modelled using eight
noded isoparametric elements. Wider
meshes were obtained by adding
elements to the right side of the mesh
shown in Figure 3.6. Shallower
meshes were obtained by deleting
elements from the bottom of the
mesh. In the vicinity of the excavation Figure 3.6: Typical finite element
all meshes were similar.
mesh for excavation problem

80 1 Finite element analysis in geotechnical engineering: Application

In the first set of analyses the soil was represented by a linear elastic MohrCoulomb constitutive model in which the soil stiffness increased with depth. The
parameters used are shown on Figure 3.5. The procedure and boundary conditions
for all the analyses were the same. Only the dimensions of the mesh varied with
boundaries of differing depths and distances from the rear of the wall. This set of
analyses was then repeated modelling the soil with the same strength parameters,
but with the linear elasticity replaced by the small strain stiffness model described
in Section 5.7.5 of Volume 1. The parameters for the small strain stiffness model
are given in Table 3.1. In both sets of analyses the soil parameters are typical of the
same stiff clay and are therefore consistent with each other. All analyses simulated
undrained excavation to a depth of 9.3m, with a rigid prop at the top of the wall.
The surface settlement troughs
lwall
behind the wall predicted by analyses
with a lateral extent of lOOm (five
times the half width of the
excavation) and with depths of 36m,
52m and lOOm (1.8, 2.6 and 5 times
the half width of the excavation) are
presented in Figure 3.7. The results of
the analyses using linear elastic preyield behaviour of the soil shown in
- coordinate from the wall to the far side of the mesh (m)
Figure 3.7a indicate that the depth of
a) Linear elastic-plastic model
the mesh has a significant effect on
1""
the predictions. In contrast, the
l0
analyses with the pre-yield small
strain soil behaviour show a much
smaller dependency on the mesh Q
21 0
depth (Figure 3.7b).
The surface settlement troughs
behind the wall predicted by analyses g
with a depth of lOOm and with lateral m$ .'"
extents if loom, 180m and 340m (5,
,,
20
30
40
50
70
80
90
IW
9 and 17 times the half width of the
x - coordinate born the wall to the far side of the mesh (m)
excavation) are presented in Figure
b) Small strain stiffness - ~lasticmodel
3.8. These predictions indicate ;hat
for the analyses performed using the Figure 3.7: Effect of the depth of
small strain stiffness model, see the mesh on the surface settlement
trough
Figure 3.8b, there is little difference
between the surface settlement troughs predicted by the analyses using meshes
with 180m and 340m lateral extent. However, for the analyses employing linear
elastic behaviour pre-yield, see Figure 3.8a, but with stiffness increasing with
depth, the predicted surface settlement troughs differ in all three analyses. These
results therefore indicate that if the small strain stiffness model is being used to
represent elastic behaviour, the mesh with a lateral extent of 180m is sufficient.
However, if the elastic behaviour is linear then a much wider mesh must be used.
X

--

M)

Earth retaining structures 1 81

Table 3.1: Small strain stiffness parameters

Clearly it is not possible to make a

general statement as to the
appropriate vertical and lateral extent
of the finite element mesh. This will
depend on the problem being
analysed, the constitutive models
employed and what facet of behaviour
is under investigation. In practice it is
therefore sensible to experiment with
20
60
IW
140
180
220
264
)W
340
meshes of different size, unless
coordinate
ffom
the
wall
to
the
far
side
of
the
mesh
(m)
experience has already been gained of
a) Linear elastic-plastic model
analysing a similar problem in the
past.
l Wall
Another decision that must be
made when performing the finite
element analysis is the nature of the
boundary conditions to be applied to
the boundaries of the mesh. In the
above analyses, zero horizontal
displacements and zero vertical forces
were applied to the vertical
20
60
IW
140
I80
220
260
300
340
boundaries of the mesh. While this is
- coordinatefromthe wall to the fa^ side of the mesh (m)
the correct boundary condition to
b) Small strain sti&ess - plastic model
model the axis of symmetry
represented by the left hand vertical
Figure 3.8: Effect of the lateral
boundary, the alternative of fixing extent of the mesh on the surface
both vertical and horizontal
settlement trough
X

8 2 / Finite element analysis in geotechnical engineering: Application

~ W ~ I
displacements to zero on the far field
10
vertical right hand boundary is
equally valid. Likewise, for the 3
2 5
bottom boundary, fixing either the $a
a8 o
vertical displacement or both the -,G
vertical and horizontal displacements
;
,
is equally valid. Whether or not the
choice affects the results depends on 8 -lo
the lateral and vertical extent of the
,,1 40
80
100
120
IM
160
180
mesh and, in particular, how far they
x - mardimate fmm the wall to the far side of the mesh (m)
are from the excavation.
a) Linear elastic-plastic model
To investigate the influence of the
boundary condition applied to the
right hand side of the mesh, the
analysis presented above, using the 3
mesh 18Om wide and 100m deep, was
0
repeated, fixing both the vertical and
horizontal incremental displacements $ j!
'I
on the far field vertical right hand
boundary of the mesh (i.e. Au = AV =
0). The predicted surface settlement
-lo
.I5
troughs are shown in Figure 3.9.Also
20
40
M
80
IW
1.70
140
160
180
- coordinate from the wall to the far side of the mesh (m)
shown, for comparison, are the results
b) Small swain stiffness-plastic model
from the analysis in which only the
horizontal displacement on the right
hand boundary was fixed
= 0). It FiL7~re3.9:Effect of displacement
boundary conditions on the far
can be seen that in the vicinity of the
boundary of the mesh On the
excavation the surface settlementsare
not affected by the choice of
Surface sett'ement trough
boundary condition. It is only near to the vertical right hand boundary that the
surface settlement trough is influenced by the choice of boundary condition.

M)

HZ

3.4.4 Support systems

The methods of supporting walls, or any other earth retaining structure, also need
to be addressed (details of any props, ties, anchors, berms, etc.). The details of any
connections between structural members need to be examined to ensure that in the
analysis the structure is modelled adequately.
These differing methods of supporting the structure act in different ways and
can therefore have a considerable influence on the behaviour of the structure.
Propping systems (struts, slabs, etc.) provide passive resistance to movement, the
resistance to movement is intrinsic. On the contrary, ties, anchors and soil
reinforcement rely on stresses within the ground being mobilised to retain the
structure. Changing ground conditions, such as drainage, can reduce the
effectiveness of the support.

Earth retaining structures 1 83

When horizontal props are used to
E
support a wall, see Figure 3.10, analyses are
often undertaken in which the props are
either assumed to be rigid, with the wall
restrained from any horizontal movement at
the position of the prop, or represented by
springs suitably grounded (i.e. fixed in space)
at some distance from the wall, see Figure
3.10. Such assumptions imply symmetry in Horizontal struts
geometry and loading conditions and
therefore that an identical retaining structure
exists as a mirror image to that being
analysed. Such a condition rarely exists in
practice as soil and loading conditions are
likely to vary. However, even if such a
situation did exist, the construction process is
unlikely to be carried out in a symmetrical
manner. For example, the walls themselves
are likely to be constructed sequentially, the
wall one side of the excavation may be
constructed before the other, thus imposing
non-symmetric conditionswhich are likelyto
affect subsequent behaviour, Finno et al.
Ground anchors
(1991).
Raking props, see Figure 3.10, also
Figure 3.10: Types of
promote complex soil structure interaction.
structural support
Part of the load from the retained soil is
transferred via the prop to be resisted by soil at formation level. The magnitude of
the prop load is therefore dependent on the interaction between the walls and the
soil behind and at formation level. Depending on the position of the raking props,
additional soil pressures may be generated on the front of the wall and resistance
to movement will be dictated by soil behaviour.
Ground anchors, see Figure 3.10, involve a complex system of soil interaction,
as both the load in, and resistance of, the anchor are derived from the retained soil.
Consequently, the behaviour of an anchored wall is likely to be different to that of
a propped wall.
The type of connection between wall and support can also affect behaviour and
increase the complexity of any design analysis. Three types of connection are
shown in Figure 3.1 1. Most current design procedures often implicitly assume a
'simple' connection, whereas in reality most connections are either 'pin-jointed',
'full moment' or somewhere in between.
These details can have an enormous influence on bending moments, forces and
displacements. Any propping system must move with the structure unless it is
anchored independently. Most propping systems prevent differential movement
l

84 / Finite element analysis in geotechnical engineering: Application

and not absolute movement and it is important to draw this distinction. This was
illustrated by the example given in section 3.4.2 above, the roof slab influenced the
differential movement of the walls relative to each other, but did not prevent the
sway of the structure.
Alternative ways to model the different types of support connection, shown in
Figure 3.1 1, and to model props and ground anchors, are discussed in Chapter 9.
Wall
Shear = 0
=0

Suppo~t

Support

Axial thrust + 0

Axial thrust + 0

SIMPLE

PIN-JOINTED

Moment + 0
Axial thrust + 0
FULL MOMENT

Figure 3. 1 I : Types of wall-support connection

3.4.5 Choice of constitutive models

It is obviously important that in any analysis realistic, and appropriate, constitutive
models are used to represent the behaviour of the structural components and the
behaviour of the ground.

3.4.5.1 Structural components

When modelling structural components the assumption is frequently made that
these elements behave linear elastically and that tensile stresses can be
accommodated. In many situations such an idealisation is perfectly adequate. The
structure may be designed to sustain these forces. However, under certain
circumstances, particularly the case of unreinforced (mass) concrete structures, the
analysis may have to account for the structure having limiting tensile andlor
compressive strengths. Likewise, when modelling props it may be necessary to
restrict their tensile capacity if in practice they have insufficient fixity to resist
tensile loads. For such situations a constitutive model with limiting tensile and
compressive strengths will have to be employed.
Because most analyses of retaining structures assume plane strain or axisymmetric conditions (i.e. a two dimensional idealisation), structural members will
have to be represented in the same way, even though their geometry may be clearly
three dimensional. For example a line of discrete props, each with a circular cross
section, will have to be modelled as a continuous slab of constant thickness in the
out of plane direction, see Figure 3.12. For structural elements modelled in this
manner an equivalent stiffness in bending, compression or tension has to be used.
Further details can be found in Chapter 9.

Earth retaining structures / 85

Figure 3. 12: Plane strain idealisation of

discrete props

3.4.5.2 Soil
Regarding the properties of the soil, it is important to use realistic constitutive
models to represent the soil's behaviour. Appropriate methods of modelling soil
behaviour are discussed in detail in Chapters 5 to 8 of Volume 1, and obtaining the
relevant soil properties has been discussed in Chapter 1 of this book.
Two categories of soil can be identified, those soils which are used as backfill
material, and those that are found in-situ. While the latter category of soils is
present in all retaining structure problems, the former category is only relevant to
a limited set of structures. The significance of the behaviour of each category
depends on the type of retaining structure. For example, backfill will have a
dominant effect on gravity and reinforced earth walls, whereas in-situ soils will
dominate the behaviour of excavations supported by embedded and nailed walls.
Soils used for backfill purposes are usually free draining sands or gravels.
However, occasionally clays are used. These soils are often modelled using either
an elastic Mohr-Coulomb model, or a model of the Lade type (see Chapters 7 and
8 of Volume 1). If the former constitutive model is used, the elastic part of the
model should preferable benonlinear, with both bulk and shear stiffness dependent
on stress and strain level (see Chapter 5 of Volume 1).
In-situ soils should be modelled using the most appropriate constitutive model
that is available and that can be justified from the site investigation data. Ideally,
this model should account for both nonlinearity at small strains and soil plasticity.
The former being required to enable realistic predictions of displacements and the
latter to limit the magnitudes of the active and passive earth pressures.
In cases where a relatively large amount of field data has been collected
concerning movements around retained excavations, such as in London (e.g.
Burland and Kalra (1986)), empirical correlations have been derived for soil
stiffness. These are often obtained by back analysis. However, these relationships
are usually linear elastic and are based on observations of movements made around
deep excavations during construction and for only a very limited period after
completion. Consequently, such relationships are not necessarily relevant to other
ground conditions, the construction of other types of retaining structure or to long
term conditions.

86 1 Finite element analysis in geotechnical engineering: Application

When long term performance is important, or where there is insufficient field

data available to derive empirical relationships, such as those derived by Burland
and Kalra (1986), then the only option is to rely on laboratory experiments on soil
samples and in-situ field tests to derive suitable parameters for use in analysis. In
most practical cases this is the situation facing the design engineer. For example,
the coefficients and limits needed for the small strain model described in Section
5.7.5 of Volume I can all be derived from high quality triaxial tests which are now
commonly undertaken by commercial laboratories. Alternatively, in the absence
of such laboratory tests, these parameters can be derived from self boring
pressuremeter tests, see Chapter 1.
Regarding the relative performance of the differing categories of constitutive
model, the following points should be considered:

It is apparent that in the overwhelming majority of cases simple, linear elastic

analyses, using either isotropic or anisotropic constitutive models, are entirely
inappropriate and can be misleading. In such analyses there is no restriction on
the tensile stresses that can develop within the soil, nor the magnitude of the
active and passive earth pressures.
Considering a simple embedded cantilever wall with an excavation in front
of it, tension in the soil behind the top of the wall will tend to hold the wall
back. The displacement profile of the wall will be affected and consequently
bending moments and forces induced in the structure. There is also no limit on
the passive resistance that may develop in front of the wall.
This can be illustrated by
considering the example of the
embedded cantilever wall shown i
E' = 6000 + 60M) z (kPa)
in Figure 3.13. For simplicity the
water table is assumed to be at :
considerable depth below the i
bottom of the wall and the soil is i
assumed to behave in a drained i
manner. Two analyses have been
W
h
E =28*1o6 kpa
E =28*106 kpa
A = 1.0 m2/m
performed using the finite element
~ 1 = 2 . 3 * 1 0 ~ ~ r n ~1=0.0833
/rn
m4/m
P =o.15
EI= 2.3*lo6 mm2im
mesh shown in Figure 3.6, one
with an elasto-plastic (MohrCoulomb) and the other with a Figure 3.13: Embedded cantilever
purely elastic constitutive model.
wall
The parameters for both analyses
are given on Figure 3.13. In both analyses the elastic properties were identical.
Excavation was carried out to a depth of 9.3m and the predicted horizontal
displacement of, and bending moments in, the wall are shown in Figures 3.14a
and 3.14b respectively. Clearly, there are large differences in the predictions
from the two analyses, with the elastic analyses predicting unrealistic
behaviour. The reason for this can be, partly, seen from Figure 3.15 which
shows the predicted earth pressure distributions acting on the back and front of

Earth retaining structures 1 87

the wall. The elastic

"
analysis predicts tensile
horizontal pressures and
pressures below the active
limit on the back of the
10
wall, while on the front of g
d
the wall, just below
excavation level, the ,,
Is
predicted pressures greatly
exceed the passive limit.
The results from the elastic
analysis would have been
HO"On"'""d'sp'"""'
(mm)
wall bendmg moment (mm)
even worse if the elastic
stiffness had been constant Figure 3.14: Comparison of elastic and
and not increased with elasto-plastic analysis of cantilever wall
depth.
Linear-elastic perfectly plastic
O
constitutive models do limit the tensile
stresses in the soil and the active and
passive pressures that can develop.
However, they generally give poor
predictions of both the extent and the
distribution of ground movements
adjacent to the structure under
construction.
Beyond the zones immediately
adjacent to the earth retaining structure
where active and passive earth pressures
.M
may develop (the soil starts to behave
plastically), or where loading causes the
soil to yield, the soil's response is entirely
elastic and reasonably uniform
m m zoo loo o loo zoo 3oo
displacements are obtained. This is
Horizontal stress, a,' (Wa)
illustrated in Figures 3.7a and 3.8a where
it can be seen that the use of linear elastic Figure 3. 5: Earth pressures
behaviour pre-yield, albeit with stiffness
predicted by elastic and
increasing with depth, results in surface elasto-plastic analyses of a
settlement troughs behind the wall that
can tile ver wall
are of a large and unrealistic extent. As
noted in Chapter 4 of Volume 1, it is generally recognised that soil stiffness
changes with varying levels of stress and strain, but these models take no
account of such effects.
The use of nonlinear elastic perfectly plastic or strain hardeninglsoftening
plastic constitutive models generally gives much better predictions of
I

88 / Finite element analysis in geotechnical engineering: Application

movements around retaining structures. Many of the constitutive models

described in Volume 1 account for the variations in soil stiffness due to varying
levels of both stress and strain.
Such models generally give much more realistic distributions ofmovements
around retaining structures than the orher two categories of models discussed
above, because the extent of movement is limited. This is important when
considering issues such as potential building damage of adjacent structures. It
is the shape of the displacement profile, as much as the magnitude of
displacements, that dictates the strains induced within the structure in question.
Predictions from such a constitutive model are shown in Figures 3.7b and
3.8b. The settlement troughs shown are muchmore realistic than those obtained
from a constitutive model assuming linear elastic behaviour pre-yield.

3.4.6
Initial ground conditions
3.4.6.1 General
With the exception of embankments retained by reinforced earth wails or gravity
walls, there is usually some excavation associated with the construction of earth
retaining structures. Because the loading of the structure depends, in part, upon
stress changes associated with this process, it is important that these stresses are
estimated reasonably accurately.
Even if there is no excavation associated with construction of the structure, the
initial stresses in the ground are important. For a simple gravity structure
constructed on the ground surface the performance of the foundation, and
consequently the structure, will be influenced by these stresses.
It is important to know the state of stress in the ground if any form of nonlinear
constitutive model is being used to model soil behaviour. This includes linear
elastic perfectly plastic models.
3.4.6.2 'Greenfield' conditions
'Greenfield' conditions are the un-modified initial stresses that exist within the
ground, arising from the various geological processes that it has been subjected to.
These stresses have not been modified by construction or any other man made
activity. The earth retaining structure may have to resist these stresses either in
whole or in part (some stress may be partially relieved before the structure is
formed).
The geological processes that the soil has been subjected to will determine the
ratio of the horizontal to vertical effective stress (i.e. the value of K,,). This ratio
can have an enormous influence on the soil behaviour and consequently the
performance of any structure within it, as demonstrated by Potts and Fourie (1 986).
Consider an element of soil with an effective vertical stress of 400kPa. Points
A, B and C on Figure 3.16 show the position in effective stress space for K,, values
of 1.O, 1.5 and 2.0 respectively. The stress paths shown on this figure are those
associated with a reduction of effective vertical stress under one dimensional
conditions.

Earth retaining structures / 89

Considering the soil beneath an - l w (oa'+a,')/2 (kPa)
excavation and in front of an
100
200
300
400
S00
600
4
embedded wall, there is likely to be a
O
' A
I
greater extent of soil that has failed in 3
B
passive stress relief, because if K , is -l*-..._
-. ,
high then less straining is required for
-K , = l
C
- - - - - .~ ~ - 1 5
the soil to yield than if K , was lower
...,
- - - - - K,=2
(i.e. in Figure 3.16 point C is closer to
the failure envelope than point A). -300This has important implications for Figure 3.16: Stress paths for the
the long term performance of the swelling of clay from different K,
structure and the short term behaviour
values
if the rate of excavation is slow and
rates of drainage are high. These effects will be discussed in subsequent sections.

3.4.6.3 Modified initial soil stresses

The presence of adjacent structures
(e.g. existing tunnels, deep basements,
N
ground bearing or piled foundations,
etc.) will inevitably modify the state
of stress within the ground and these
effects must be accounted for in any
analysis. Such effects may influence
the loads that the retaining structure
has to carry. Before the earth
retaining structure is formed, another
structure may have loaded the soil or Figure 3-17:Pian of the site for the
in the case of a tunnel there may have Jubilee Line station at Westminster
been some unloading.
To illustrate this point, consider
the example of the Jubilee Line ,
Extension underground station at
Westminster (Higgins et al. (1996)).
Figures 3.17 and 3.18 show a plan j
and a cross section through the site j
respectively. The excavation for the j
5.7 London clay
~
~
station is 37m deep and is supported
by l m thick diaphragm walls, which
in turn are supported by a series of ,
Lambeth Group Clay
temporary and permanent props.
52.0~
Lambeth &up Sand
Close to the excavation are two j
41.0~
Thancl sand
tunnels. Also in very close proximity
is the Big Ben clock tower, a historic Figure 3. 18: Cross-section AA of
structure of great significance.
the Westminster site

90 1 Finite element analysis in geotechnical engineering: Application

D/
A series of analyses were undertaken
C
which modelled the construction of both the I 'Dtunnels and the station box. A second series
of analyses were also undertaken which
modelled the construction of the station box
alone. Figure 3.19 shows contours of stress
level within the ground before and after
-D079
tunnel construction. The stress level is a
-E099
a)
dimensionless measure of the available shear
strength mobilised at a particular mean
effective stress. It is defined in Appendix
111. I. A stress level of zero corresponds to a
state of stress in which a,=a,=qand a value
of unity corresponds to a stress state on the
failure envelope. The small excavation
evident in the top left hand corner ofthe plots
in Figure 3.19 is a temporary excavation
performed prior to tunnel construction and is
not the main excavation for the station box.
Inspection of Figure 3.19 clearly shows
,cigure3. 19: contours
of
that construction of the tunnels has modified
stress level
before and
the stresses within the ground close to the 6 ) after tunnel
position where the main walls for the station
box are to be constructed (at this stage these
,,,,,,,ll,isplacement
walls are not present in the analysis). The soil
60
40
20
o
in this location has been unloaded and
consequently the loads imposed on the walls
when they are constructed will be affected.
Figure 3.20 shows the wall displacemcnts
-for comparable analyses with and without the
tunnels present. These correspond to the without
condition where construction of the station
box is complete. It is apparent that the
presence of the tunnels affects the wall
displacements. This is not an uncommon
situation in a dense urban environment such
as London, with an extensive underground
system. This example demonstrates the
Figure 3.20: Horizontal
importance of modelling the complete predisplacements of station box
construction history.
walls with and without
Groundwater conditions may also have
tunnels in analysis
been modified as a result of the presence of
an adjacent structure. There may be seepage around an impermeable basement
wall, there may be a tunnel which is not watertight and tends to drain the

Earth retaining structures I 91

surrounding ground, or there may even be a supply of water from some source.
These effects also have to be considered, particularly when using the more
complex constitutive soil models.

3.4.7
Construction method and programme
3.4.7.1 General
Consideration must be given to how the structure is to be constructed (temporary
as well as permanent works) and the time taken to construct it. Different sequences
of construction mean that soil may experience differing stress paths, yield may
occur in soil elements at differing times or be suppressed completely. As a
consequence additional forces, such as swelling pressures, might be imposed on
the structure. Further, if construction is slow some drainage may occur within the
ground, which might reduce long term effects. However, if significant drainage
does occur there may be some implications for the design of temporary works.

3.4.7.2 Construction method

In order to illustrate the effect of differing methods of construction, the example
of the road tunnel, shown in Figure 3.3, is considered. Analyses have been
performed in which the tunnel has been built in two ways, firstly by 'top down'
and secondly by 'bottom up' construction. In both types of construction the walls
were constructed first. For the top down construction excavation to a level just
below the underside of the roof slab was performed, the roof slab was then
constructed before further excavation. In the bottom up construction excavation to
the underside of the road slab was performed and the road slab cast with some
temporary propping provided. The roof slab was then cast and then the temporary
propping removed.
The different methods of construction necessitated differing methods of
supporting the walls. In the analysis with the bottom up method there was actually
a discontinuous temporary prop (a series of tubular steel sections along the length
of the tunnel) between the walls during excavation, with a nominal stiffness
approximately 10 times less than that of the roof slab.
Table 3.2 summarises the predicted bending moments in the walls for both
methods of construction at the stage in the analyses where the structure has been
completed, but drainage in the soil has yet to occur. Even for this relatively simple
problem, a relatively shallow excavation of between 8 to 10 m depth with a single
prop for the walls, there is a variation in bending moments of up to 20%
attributable to differing construction methods. Without the temporary prop the
differences would have been even greater.
The analysis of the George Green cut and cover tunnel (Potts and Knights
(1 985)) is another good example of how the construction method influenced the
form of the analysis. Figure 3.21 shows a cross section through the tunnel. The
geometry of the tunnel is approximately symmetrical about the centre line of the
tunnel and, as with the example discussed in Section 3.4.2, it is very tempting to
model this as a 'half section'. Doing so would have been a very serious mistake.

92 / Finite element analysis in geotechnical engineering: Application

Table 3.2: Summary of bending moments for alternative methods of
construction

Although the geometry

was symmetric, the method of
construction was not. The
embedded walls were installed
first. There was a preliminary
excavation to allow the roof
slab to be constructed, but soil
was then excavated out to the
underside of the base slab to
form one half of the tunnel.
The base slab for this half of Figure 3.2 1: Cross-section of the George
Green Tunnel
the tunnel was then cast before
excavation started for the other
half of the tunnel. The effect
4 -immediatelyaflcr conshuction
1 - immediately afla wmhuction ofroof slab 1
of this
procedure 2 -imediakly after wmhuction of 1b.a.m e 1 aftbs. m e 1
5 -a& full dissipation of excess p.w.p.
on the central wall is shown in 3 - ~ ~ a koftcrcamhuctionof
l y
roofslab 2
mof slab
....
Figure 3.22. An analysis of
only half the section would
have suggested that there was
4 4 - 2
3
no bending o r lateral
b
.' .."*"
displacement of this wall.
8.0 H
.....
/

10.0

3.4.7.3 Time related

movements

2.0 1.6

1.2 0.8 0.4

11.9

Horizontal wall displacement(cm)

3
-800 -400

0 400 800 1200

Bending moments (mm)

To make an assessment of
Figure 3.22: Bending moments and
time related movements,
analysis need to be time
displacements for the central wall
related and therefore an
analysis with fully coupled consolidation is generally required, see Chapter 10 of
Volume 1. For such analysis it is necessary to specify pore water pressure andlor

Earth retaining structures / 93

flow boundary conditions in addition to displacements andlor forces. In addition,

the permeability of the soil has to be known in order to make a reasonable estimate
of how much drainage may occur during construction. Unfortunately, this is a
notoriously difficult parameter to measure and therefore it is normally prudent to
consider a range of values and perform a small parametric study.
Not only must the magnitude of the permeabilities be correct, but also their
distribution, particularly where under-drainage exists. The permeability distribution
has to be consistent with the initial steady state pore water pressure profile.
Otherwise, during the course of the analysis drainage will occur as the pore water
pressures try to come into equilibrium with the boundary conditions. This problem
is discussed in greater detail in Chapter 9.

3.4.7.4 Ground water control

Measures taken for groundwater control,
during construction and in the long term,
are also an important consideration. Dewatering may be necessary to form an
excavation. For example, consider the
case shown in Figure 3.23 in which an
excavation is to be made at a site in
which clay overlies sand and the water
table is near to the ground surface. Unless Figure 3-23:An example of the
the sand within the excavation is denecessity for de-watering
watered, a base heave failure will occur
during the excavation process, when the total vertical stress imposed by the
remaining clay equals the pore water pressure in the sand at the sand-clay interface.
Such de-watering must also be simulated in any analysis.
Once construction is complete, measures may need to be taken to minimise the
effects of long term heave beneath a structure. Voids may be left beneath ground
slabs to allow for the ground to swell without imposing any load on the structure.
In such situations water would have to be pumped from the void. Alternatively
pumping or drainage systems may be installed to minimise pore water pressure
changes and the consequent loading on the structure.
Another important consideration is the boundary condition applied to excavated
surfaces. Is there a supply of water which will help promote swelling within the
soil, or do excavated surfaces remain impermeable? In such situations the
precipitation boundary condition described in Section 10.6.6 of Volume 1 may be
more appropriate than either a pore water pressure or no-flow boundary condition.

3.5
3.5.1

Gravity walls
Introduction

When the retained height is not large, typically less then 5m, soil movements are
not important and construction space is not limited, gravity type retaining walls are

94 1 Finite element analysis in geotechnical engineering: Application

often used. There is a variety of forms of gravity wall and some examples are
given in Figure 3.1. To maintain stability this category of retaining walls relies on
the weight of the wall (and any enclosed soil) and the shear resistance between the
base of the wall and underlying soil to resist the earth pressures generated by the
retained ground.
During design these walls
should be checked for overall
stability, sliding, overturning and
bearing capacity failure, see
Figure 3.24. As this type of
structure is not usually used where
b) Slldlng farlure
there are adjacent structures a)bL""OfOvemlisfabrli~
andlor services which are sensitive
to ground movements,
displacements are not usually
explicitly calculated. In many
d) Bearlng capaclty farlure
design codes it is often implicitly c ) O v e m g f a r l u r e
assumed that if there is a sufficient
margin o f safety against Figure 3.24: Limit states for gravity
instability, then movements will
walls
be acceptable. If movements are
of concern it may be better to use a different form of retaining structure, such as
an embedded wall, see Figure 3.1.
As there exists considerable experience with the design ofthis type of wall, and
as movements are not usually of concern, numerical analysis is rarely performed
for standard gravity wall types. However, there may be situations where
movements are of concern, or where the soil conditions are unusual or the structure
has dimensions outside current experience. In such cases numerical analysis can
be used. Such analyses can investigate all the potential instabilities shown in Figure
3.24, as well as provide an estimate of movements.
The rest of this section will consider some of the important issues concerned
with the numerical analysis of gravity retaining walls. Although reinforced earth
and nailed walls are gravity walls, they have additional complications associated
with internal stability. Consequently, this type of retaining structure is discussed
separately in Section 3.6.

3.5.2

Earth pressure due to compaction

In many situations gravity walls are built on a foundation, with backfill placed
behind them. Therefore it is necessary to consider not only the properties of the
structure and the behaviour of the foundation (as noted in Figure 3.24 bearing
capacity may be an issue), but also the nature of the backfill. As most backfills are
compacted in layers, stresses will be imposed on the retaining wall due to the
compaction process. However, there are very few reliable measurements of these
stresses from which suitable values may be deduced.

Earth retaining structures 1 95

Although numerical procedures do exist for estimating compaction stresses
(e.g. Duncan and Seed (1 986)), these are generally quite complicated and may be
problematic with highly nonlinear soil models. Because of the lack of reliable field
data, validating these procedures is also difficult. It is much simpler to specify the
horizontal stresses in layers of fill, as a proportion of the vertical stress, as they are
constructed during an analysis, see Section 3.7.9 of Volume1.
In practice the backfill may be constructed in many layers. Simplifications will
therefore often have to be made in an analysis and much fewer layers used. A
discussion on the problems of selecting the number of layers and choosing the
horizontal stress ratio is given in Section 5.3.4 in relation to embankment
construction.

3.5.3
Finite element analysis
The main issues involved with the numerical analysis, apart from selecting
appropriate constitutive models for the structure, foundation and backfill, are the
simulation of the construction procedure and the modelling of the interface
between the soil and structure.
As noted above, the stability of a gravity
wall is mainly derived from the shear stresses
mobilised along its base. This in turn is a
function of the vertical force on the base and
the cohesion and angle of interface friction
between the base and the underlying soil.
The vertical force is derived from the weight
of the wall (including any inclosed soil) and
the vertical shear force imposed on its back
by the retained soil. The latter force is
dependent on the cohesion and angle of
interface friction between the back of the
Figure 3.25: Position of
wall and the retained soil. Clearly, the .
interface elements in gravity
interface properties between the wall and the
wall analysis
soil, both foundation and retained, can have
a dominating effect and must be accurately modelled in any analysis. The most
appropriate way of doing this is to include interface elements within the mesh, see
Figure 3.25. However, there are drawbacks and limitations associated with the use
of such elements. These are discussed further in Chapter 9.
Once the construction procedure has been established, and in particular the
number of layers to be used in backfill construction, the finite element mesh can
be developed. Clearly, each layer of backfill must consist of a finite set of elements
(i.e. elements cannot span more than one construction layer). During the analysis
problems are often encountered due to inadmissible stresses arising in elements
that have just been constructed. This is usually a consequence of the elements
'hanging' on the wall during the construction process resulting in tensile stresses.
This problem, and some solutions, are discussed further in Chapter 9.

96 / Finite element analysis in geotechnical engineering: Application

Gravity walls may be constructed

in cuttings. Certainly this was a
common technique before modem
plant enabled alternative methods of
construction to be developed (e.g.
diaphragm walls, etc.). In this
instance the analysis must model the
cutting which may be backfilled
Figure 3.26: Cutting behind the
completely behind the wall or only
gravity wall
partially, see Figure 3.26. Depending
on the time that the cutting has to
stand, there may be some drainage within the foundation and softening towards the
base of the slope. This may affect the subsequent performance of the structure.
On occasions these walls can have piled foundations simply to resist a large
overturning moment. An example is a quay wall, where piles may be installed to
increase the capacity of the quay to resists forces from ship impact and mooring
lines. Other measures to increase capacity, such as anchors, may be included for
a variety of reasons. Again these piles or additional means of support have to be
modelled realistically.

3.6
3.6.1

Reinforced earth walls

Introduction

Following its first modern applications in France in the 1960's, reinforced earth has
become an increasingly popular alternative to gravity retaining walls. The method
has been shown to be significantly more economic in many situations, due to the
efficient use of material and its favourable distribution of load on soft foundations.
The mechanical properties of the soil are supplemented by the addition of
reinforcement with a tensile capacity, to allow the soil mass to become self
supporting.
Reinforced earth walls generally have
a facing membrane which is normally
non-structural and reasonably flexible, its
purpose being to prevent local
unravelling of soil. In their original
format reinforced walls were constructed
by backfilling in layers, see Figure 3.27.
A typical situation might be as shown in
Figure 3.28. A membrane unit is placed
on a purposely built foundation and is
strips
temporarily supported while a layer of
soil is backfilled behind it. When this
layer reaches mid-height of the Figure 3.27: Typical construction
membrane unit, a reinforcing strip is details of a reinforced earth wa//

Earth retaining structures / 97

placed on top of it and is connected to the membrane unit. Further backfill is
placed until it reaches the top of the membrane unit. The temporary support to the
unit is then removed and used to support the next membrane unit, which is
positioned on the top of the previous one. The process is repeated until the desired
wall height is achieved. In practice there have been many variations on this basic
construction process.
,---------

l1\

Proposed

Figure 3.28: Typical constructjon procedure for a reinforced

earth wall

Originally the reinforcement consisted of

thin strips of metal which were connected to
the facing membrane and run back 5m to
10m into the backfill. These strips are
typically lOOmm to 400 mm wide and are not
continuous in the out of plane direction. They
rely on interface friction for their pull-out
resistance. Nowadays sheets of geotextile are
sometimes used. These are continuous in the
out of plane direction. The same sheets can
also be used to form the facing membrane,
see Figure 3.29. Some of these geotextiles
have a mesh like appearance and their pull
out capacity is a function of both interface
friction between the fabric and the soil and
bearing capacity failure of the soil in front of
the individual components making up the
edges of the holes in the geotextile, see
Figure 3.30.
Further developments have been to
replace the reinforcing strips or sheets by
discrete anchors (Harris et al. (1993)), see
Figure 3.3 1 . The anchors can be made from
bent steel bars, blocks ofconcrete or even old
car or lorry tyres. The anchor cables can be

/ Geotextile sheet

the wall

Figure 3.29: Geotextile used

as facing membrane

11
1
2

Figure 3.30: Example of

geotextile membrane

98 / Finite element analysis in geotechnical engineering: Application

metal or some form of rope material (e.g.

Paraweb). The reinforcing effect is now
derived from the pull out resistance of the
anchor blocks.
Both the reinforced and anchored
walls described above are constructed
using back fill. An alternative approach
sometimes used to form a reinforced wall
in-situ is to use soil nails. This technique
involves excavating a shallow depth in
front of the wall (typically 1 to 2m),
applying a layer of shotcrete over a
reinforcing mesh and installing a row of
soil nails. The wall is formed from the top
Figure 3.3 7: Example of a
downwards by a sequence of such steps,
anchored earth wall
see Figure 3.32. The spacing and type of
nails used are dependent on the nature of the soil and the size and geometry of the
structure. There are different types of nails available: bored and grouted, jetgrouted, driven and more recently ballistically fired steel rods. Grouted nails have
been used most extensively for retaining wall construction, as they provide a
greater surface area for mobilisation of friction and can be installed to a greater
distance into the soil mass than driven or fired nails. They are usually angled
slightly downwards to facilitate grouting operations.
Reinforced soil walls can fail by a number of mechanisms, see Figure 3.33.
Assuming that the reinforced earth mass is suitably reinforced, failure can occur
as a result of global instability, in much the same way as discussed above for
gravity walls.
In addition, failure can occur as a result of internal instability. For example, the
reinforcement itself can
rupture resulting in a loss of
tensile strength and possibly
triggering a global failure of
the reinforced earth wall.
Alternatively, as the integrity
b) Re~nforcedshotcrete
of the reinforced soil mass is
dependent on the stress
transfer between the soil and
the reinforcement, contact
failure can result in pull-out
of the reinforcement or block C) tostalilng the nails
d) Excavahon
sliding of the soil along the
if Figure 3.32: Construction sequence for a
planes
the reinforcement forms
reinforced wall using soil nailing
continuous sheets.

Earth retaining structures 1 99

A further mechanism induced within the reinforced mass is caused by the use
of reinforcement of insufficient stiffness. In this situation, if sufficient strains have
accumulated within the soil mass before full strength ofthe reinforcement has been
mobilised, a displacement induced failure can occur. This mechanism is
particularly important when considering the long term effects of plastic creep in
geotextile reinforcement. An example of this type of failure for a reinforced
embankment is given in Section 5.5.6.

Settlement

Rotation

Block sliding
on foundation

Block sliding
on reinforcement

Reinforcement
stiflhess failwe

Slip surface

Reinforcement
pull-out

Reinforcement
strength failure

Localised
unravelling

Figure 3.33: Possible failure mechanisms

for reinforced soil wall

3.6.2

Finite element analysis

From the above discussion it is clear that reinforced earth walls come in many
different forms, each of which will have its own particular modelling problems.
Within this text it is not possible to discuss all the issues involved. They are very
complex and would probably justify a text on their own. Instead, finite element
analysis of a conventional reinforced earth wall will be discussed in detail. Many
of the modelling problems associated with this type of structure will also be
applicable to other forms of reinforced earth walls. Some of the more important
issues with these other wall types will then be discussed. A plane strain situation
is assumed.

100 1 Finite element analysis in geotechnical engineering: Application

Figure 3.34 shows part of a finite

element mesh used to analyse a
reinforced earth wall with a maximum
height of 13m. The figure shows the
detailed part of the mesh in the
vicinity of the wall. The complete
mesh extended further in both the
lateral direction and with depth. The
reinforcing elements extended 7m
into the backfill and had Im vertical
spacing. Consequently there is only a
single row of solid elements between
each reinforcing level. While this is

Lengthof
reinforcement $:-;-L-:-::$-:-;-:-;--~
F=inp

-----.
-----.--------;
-.--;

----7----7---...

-,-,
fi-> -.-:-:l
$-;..:...-.
*-C-;-r

,L,-.-,-

,---------

,-----,----..a

d--I

$.-.-;-r-n+-,-.-;-:--;-----;--------.

+b

J"-"-;tl:;-;-;-+-1--4

$;-;-l -,-,-,

I.-.-

Foundation

-.

-.-

r , - r -:-;-I-:+-,-.-;
~ - ~ - ~ 4 - : - : - 4-----;
-4awb-4
,A-'--,-+-:$-;-:-.-'--.-----.--.-------.
$:-L-L-I--+-,-+-J-:--2
----- J. X'--- - - --- - - -. - - - - - .
(.-.-:-:-::$-8-I-;-.--:--&$;-;-~-l-l-~~-:-:-,-

'--A

-----

Figure 3.34,, Finite element mesh

for t h e present
for a reinforced earth wall
demonstration, in practice it is
advisable to use more than one row of
elements.
The reinforcing layers were
simulated using membrane elements
which had a capacity to resist only
tensile forces. The facing membrane
was modelled using discrete Mindlin
beam elements, while the interfaces
between them and the soil, and
between the membrane elements and
the soil were modelled using zero
thickness interface elements, see
Figure 3.35. It should be noted that
C
the beam elements representing the
facing units were not continuous but
were discrete, in this example a single
element for each facing unit. It is Figure 3.35: Modelling details of
reinforcing layer and facing
often necessary to use beam elements
membrane
as opposed to membrane elements to
model the facing units, as these
elements can be given some bending and shear stiffness to resist the local
unravelling of the soil immediately adjacent to the facing unit.
The stability of a reinforced earth wall is dependent on the mobilisation of the
reinforcing element strength. This can only be achieved once the reinforcement is
in a position to withstand tensile forces which do not occur until it is attached to
the facing units and has normal stresses applied along its length. Consequently,
there is a stage of the construction process during which the uppermost soil layer
is unsupported and in the absence of cohesive strength will suffer local failure
unless temporary support is provided.

Earth retaining structures 1 101

To avoid this local instability the following construction procedure could be
employed, see Figure 3.36:

- The first soil layer along with the

beam, membrane and interface
elements, representing the facing
unit, and the first layer of
reinforcing are constructed, see
Section 3.7.9 of Volume 1.
During this process zero
horizontal displacement
boundary conditions are applied
to the nodes of the beam element
representing the facing unit to
provide temporary support. To
Figure 3.36: Finite element modelling
simulate the connection between
of construction sequence
the reinforcing strip and the
facing unit the horizontal displacements of the node on the membrane element
nearest to the facing unit, (i.e. node A in Figure 3.35) are tied to those of the
nearest node on the facing unit (i.e. node B). The vertical displacements of
these nodes are also tied.
- The next layer of backfill along with the beam, membrane and interface
elements representing the second facing unit and the second layer of
reinforcing are then constructed. Zero horizontal displacement boundary
conditions are applied to the nodes of the beam element representing the new
facing unit to provide temporary support. The displacement boundary
conditions that were imposed on the first facing unit are removed. This
involves turning this boundary condition off and removing the nodal reactions
that were created in the previous stage at these nodes.
- This process is repeated for the additional layers ofbackfill. However, it should
be noted that to model the connection between the reinforcing strips and the
facing units, the horizontal displacements at nodes A, B and C (see Figure 3.35)
should be tied for each reinforcing strip. Similarly, so should the vertical
displacements at these same nodes. It should be noted that as the facing
elements are not connected to each other and as their rotations are not tied, they
cannot directly transmit moments to each other. Clearly these tied boundary
conditions might change depending on the finer details of the construction
process and the connection between reinforcing and facing unit.
Specifying the material properties for the membrane and interface elements is
also problematic. If the reinforcement is a geotextile, it is usually continuous in the
out of plane direction and therefore compatible with the plane strain assumption.
However, as noted previously, its pull out resistance is derived from friction
generated on the soil-reinforcement boundary and the in plane bearing capacity of
the soil within the holes of the geotextile, see Figure 3.30. As these holes are not

102 / Finite element analysis in geotechnical engineering: Application

modelled by the combination of membrane and interface elements, an

approximation has to be made. There are several empirical methods for accounting
for both components. A convenient method is to assume that the maximum shear
stress that can be mobilised in the interface elements is given by:

where p' is the angle of shearing resistance of the backfill and ,u accounts for both
bearing and frictional effects. ,u can be found from shear box interface tests
performed on the reinforcement and can have values greater than unity.
If the reinforcement is steel strips then its pull out resistance is mobilised from
interface friction with the soil. However, as the reinforcement is not continuous in
the out of plane direction, it is not consistent with the plane strain assumption and
again an approximation has to be made.
One approach to the problem is to model the interface elements as having
equivalent frictional contact strength, by considering the relative areas a n d p values
of the soil to soil and soil to reinforcement interfaces. However this approach does
not provide logical results. For example, if the width of the strip increases the
proportion of soil to reinforcement surface area (p<l) in&-easesrelative to the soil
to soil surface area (p=l) and there is a corresponding drop in the effective contact
strength.
In addition, failure of the interface
elements will result in block sliding as
noted in Figure 3.33. This does not
accurately represent pull-out failure as it
disregards the soil to soil connectivity in
the discontinuities between the
reinforcement elements in the horizontaI
out of plane direction.
A possible solution to the problem is
to tie adjacent nodes of the soil elements
together across the reinforcement. For
example, the vertical displacement of
nodes such as E and G in Figure 3.35 are
Detail of a 3D slice
.,
,' ,
tied to nodes F and H respectively.
- - - - - - - - -f ' --.
Likewise the horizontal displacements of
II
:
'
;
these nodes are also tied. This approach
I
I
has the advantage of modelling the soil
I
continuity while still allowing failure
I
I
'
/
along the soil-reinforcement interface.
- - - - ; ,-~ ~
A better solution to the problem
would be to perform three dimensional
analyses. If the wall satisfies most of the Figure 3-37;30 modelling of a
reinforced earth wall
assumptions of plane strain, apart from
- - - - - - - - - - - - - - - - - - - - - - - - - - - - V - - - -

- - - - _ _ _ _ _ _ _ _ 7 _ _ _ _

Earth retaining structures 1 103

the reinforcement not being continuous in the out of plane direction, then use can
be made of planes of symmetry and only a three dimensional slice of the wall
needs to be analysed, see Figure 3.37.
If the wall is reinforced with anchors as shown in Figure 3.3 1, plane strain
conditions are again not really appropriate and in principle full three dimensional
analyses are required. However, with present computer hardware this is not
practical and further assumptions are made to model the anchors. This problem and
some of the solutions available are discussed further in Chapter 9.
Additional problems arise with
,
a
,
,
'
t
nailed walls. Not only are the nails
not really applicable to a plane strain :
Future face
ofnailedwai
assumption, but problems also arise j
because at the start of the analysis j Bottomofexcavation
,
they have not been installed. For 1
example, consider the situation shown j
c
slots leA for
in Figure 3.38. At the beginning of 1
future nails
the analysis the membrane (or beam) I.........................
and interface elements are not present
in the mesh. Consequently, there are Figure 3.38: Difficulties in modelling
slots (holes) in the mesh at the nail
soil nails
positions. Clearly it would be unwise
to perform analysis with these holes in the mesh. To get around the problem, the
adjacent nodes on either side of a slot must have their displacements tied. These
ties are removed once the membrane (or beam) and interface elements representing
the anchor are constructed.

I
lI

I
I
I

3.7

Embedded walls

3.7.1
Introduction
The most common use of concrete
embedded walls (such as diaphragm,
secant pile and contiguous pile walls,
see Figure 3.39) is to support the sides
of deep excavations, or tunnels which
are formed by various means (cut and
cover, top down, etc). While steel
sheet piles can be, and are, used for
secantpilewall
Conliguous pile wall
this purpose, concrete walls are also Diaphragmwall
used to form bridge abutments and
quay walls.
Figure 3.39: Types of embedded
Under some circumstances these
waN
walls may also be used to limit
ground movements within another structure such as an embankment or cutting, or
to control groundwater (slurry walls to cut off water flow into an excavation or
cutting). However, the following discussion is aimed solely at the first category.

104 / Finite element analysis in geotechnical engineering: Application

3.7.2
Installation effects
3.7.2.1 General
This is an extremely complex issue which is generally ignored in most numerical
analyses. In most instances the walls are 'wished in place'. This is the term usually
used to indicate that installation effects have been ignored and implies the
assumption that there is no increase or reduction in stresses associated with
installation of the walls.
In reality, concrete piled walls are constructed as a series of individual bored
piles and diaphragm walls are constructed as a series of individual panels. During
construction the holes in which the piles or diaphragm wall panels are concreted
maybe supported temporarily (e.g. by bentonite slurry). This is obviously a very
three dimensional process and not easily approximated by a two dimensional
analysis. There is some stress redistribution around these individual elements, both
horizontally and vertically (Huder (1972), Davies and Henkel (1982)).
Likewise the installation of steel sheet piles will alter the stresses within the
adjacent ground.

3.7.2.2 Field measurements

Intuitively it can be expected that wall installation will cause some change in the
stresses within the ground surrounding walk, and some stress changes associated
with this process have been measured for concrete walls (Symons and Carder
(1993)). Despite this work the significance of these stress changes is difficult to
quantify.
Symons and Carder (1993) suggest that total stress reductions of up to 10%
may occur close to bored piles in a stiff over-consolidated clay, whereas for a
diaphragm wall this may be as high as 20%. A value of 30% was inferred from
measurements at the Bell Common tunnel (Tedd et al. (1984)), but there are some
doubts concerning the reliability of these particular observations.
Current evidence is that, with reasonable workmanship, installation effects are
reasonably localised. Inevitably on some occasions effects are more widespread,
particularly when there are significant losses of fluid supporting the excavation for
a pile or panel to the surrounding ground.
One example of where significant movements, and presumably corresponding
stress changes, associated with wall installation did occur, was presented by Davies
and Henkel(1980). The diaphragm walls were intended to support the sides of an
excavation for Chater Station in Hong Kong. During installation of the panels
settlements of up to 78mm were observed adjacent to the wall. By the time the wall
was completed movements were detected at up to 50 m away from the wall (the
panels were up to 37m deep). The conclusion was that:
"Horizontal movement adjacent to individual panels was a result of swelling of
the decomposed granite which created a compressible zone around the
diaphragm wall panel. On construction of adjacent panels, the arching around
the compressible zone broke down and re-compression occurred as the earth

Earth retaining structures 1 105

pressures built up. This caused horizontal ground movements to extend back
from the diaphragm wall, which resulted in large displacements".
Obviously this is a very complex mechanism and very difficult to reproduce in
any analysis. Such large movements were not expected before construction started,
but to try to minimise these effects subsequently panel lengths were reduced and
slurry pressures increased.
There is little field information available concerning displacements and stress
changes associated with the installation of sheet piling.

3.7.2.3 Analysis
The issue is whether or not installation effects should be represented in any
analyses, or should it be assumed that measures will be taken during construction
to mitigate such influences. It should be remembered that in many situations
embedded walls are being used because movements are critical and therefore every
effort will be made to minimise movements during wall installation. As stress
changes within the soil only occur ifmovements are allowed, minimising the latter
will also limit the former.
A number of researchers have tried to analyse these effects (for example
Higgins (1983), Ng (1992), Gunn et al. (1993) and De Moor (1994)).
Unfortunately none ofthese approaches has been entirely satisfactory because they
have not realistically considered three dimensional effects, see St John et al.
More recently some results of three dimensional analyses involving wall
installation have been published (e.g. Schweiger and Freiseder (1994) and
Gourvenec (1998)), but these analyses have generally used very simple constitutive
models and very crude meshes. Such simplifications were necessary because, as
noted in Chapter l I of Volume 1, very large computer resources are needed to
perform full three dimensional analyses. It is therefore questionable how realistic
were the predictions of remote movements associated with wall installation. Since
the purpose of trying to model wall installation is normally to assess remote
movements, and because allowing soil stresses to relax is generally not
conservative as far as the design of structures is concerned, it is questionable how
valuable these studies were. To get a better understanding of the process more
sophistication is required.
One example where this issue had to be addressed was the design of a pin-joint
connection between an inclined road slab and a contiguous bored pile wall for an
underpass (Higgins et al. (1999)). The connection between the slab and the wall
was a pre-cast concrete hinge with a rounded stainless steel bearing, see Figure
The designers were extremely concerned about the ratio of the horizontal to
ertical thrust in this joint under long term conditions and they placed a limit on
his 'thrust ratio'. Assuming that the wall was 'wished in place' would allow no
tress relief during installation. For any given assumption concerning the

106 / Finite element analysis in geotechnical engineering: Application

coefficient of earth pressure at rest,

K,,, it was considered necessary to
examine what reduction in horizontal
stress may occur by installation of the
bored pile wall. This was considered
likely to affect the thrust ratio.
To make an assessment of these
effects, a series of plane strain finite
element analyses were undertaken,
which modelled the complete
construction sequence and long term
conditions. These analyses started
with the finite elements representing
Figure 3.40: Cross-section of an
the wall being excavated and
underpass
simultaneously replaced by a fluid
pressure equivalent to wet concrete. This allowed some relaxation of the stresses
within the soil and enabled the 'thrust ratio' to be evaluated. Because these
analyses were two dimensional plane strain analyses they actually represented the
excavation of an infinitely long slot. Since this method is likely to overestimate the
effects of installation, it was actually a conservative approach providing a bound
to the problem. The other bound came from an analyses in which the piles were
'wished in place'.

3.7.2.4 Comments
To model wall installation realistically a fully three dimensional approach is
required. Even then the sequence of pile or panel installation will need to be
known, in order to correctly account for the effects such as arching around the
individual excavations (piles or panels).
This is apparent from the Chater Station case history described above. If it had
been possible to analyse construction of the wall realistically (i.e. a fully three
dimensional analysis, a representative soil model and realistic boundary
conditions), then some indication of the problems might have been obtained.
Relatively simpler procedures, such as those currently proposed, are unlikely to
have been of much assistance.
Each case has to be considered on its merits. If it is necessary to account for
installation effects, care must be taken in any analysis to ensure that the results are
not simply a consequence of the applied boundary conditions, as discussed by St
John et al. (1995). Care must be taken to ensure that these effects are not overestimated, since this could have serious implicationsfor the design of the structure.
A cautious approach is required.

Earth retaining structures / 107

3.7.3

Modelling of walls
3.7.3.1 Element type
A major concern when modelling embedded walls is the choice of finite element
type to represent the wall itself. If the wall is a diaphragm, secant pile or
contiguous pile wall, see Figure 3.39, then it is relatively thick (usually greater than
0.8m) and solid elements are probably appropriate. However, if the wall is made
from steel sheet piles, then the use ofbeam (or shell) elements is more appropriate.
However, in either case solid or beam (or shell) elements can be used.
As noted in Chapter 3 of Volume 1, beam (or shell) elements provide an
efficient but approximate way of modelling structural members. They are
approximate in the sense that they reduce one dimension of the structural member
to zero. For example, if used to model an embedded retaining wall, the wall is
essentially modelledas havingzero thickness. At first sight this approximation may
seem reasonable and consequently the approximation is often overlooked in
practice. However, the approximation can have a significant effect on the analysis
of an embedded retaining wall.
As an example consider the embedded cantilever wall shown in Figure 3.13.
The wall is Im thick and 20m deep and is made of concrete. The properties of both
the soil and wall are given on Figure 3.13. To simplify matters the water table is
assumed to be at depth below the base of the wall.
Analyses have been
performed in which the wall
was modelled using a single
column of solid elements, with
the finite element mesh shown
in Figure 3.6. Two analyses
have been performed, one with 8
K,, = 2.0 and the other with K,,
= 0.5. The analyses have also
been repeated with a similar
mesh, but with Mindlin beam .20
-30 -25
-20
.l0
-5
0
-5W
0
1030
IJW
2030
elements used to model the
Wall b e n h g moment (kNmim)
m
,,,,,,,
wall. The predicted horizontal
wall displacements and Figure 3.4 1: Effect of wall element type
in a K, = 2.0 soil
associated bending moments

z,,
'"

-13

JM

for the analyses with K,) = 2.0 are given in Figure 3.41. These predictions
correspond to an excavation depth of 9.3m. Corresponding results from the
analyses with K,, = 0.5 are given in Figure 3.42.
For both K , values the use of beam elements to represent the wall results in
larger horizontal movements and bending moments. This is particularly so for the
analyses with K, = 0.5 were the maximum wall displacement and the maximum
bending moment are 33% and 20% respectively greater forthe analysis using beam
elements compared to the analysis using solid elements to represent the wall. The
differences are smaller for the analyses with K,, = 2.

108 / Finite element analysis in geotechnical engineering: Application

"

The reason why there are

differences between the
analyses modelling the wall
with solid and beam elements
is due to the vertical shear 2
stresses mobilised on the back
of the wall. For the analyses
using solid elements, these
shear stresses act downward at
a distance of 0.5m (half the
wall thickness) from the
Wall bendrog moment n;Nmlm)
H o m W wall displacnncnt (cm)
neutral axis of the wall. They
therefore provide a clockwise Figure 3.42: Effect of wall element type
moment which partially
in a K, = 0.5 soil
offsets the anticlockwise
moment generated by the horizontal stresses acting on the back of the wall. While
these same shear stresses are mobilised on the wall in the analyses using beam
elements to represent the wall, the wall has no thickness and therefore there is no
lever arm and no clockwise moment. Hence the difference in results.
The same potential error is likely
to arise when modelling sheet pile
walls using beam elements. Although
the steel sheet piles are made of
t - wail th~ckness
relatively thin steel sheet, the shape of
the sections in plan gives the wall an
Figure 3.43: Detail of a sheet pile
equivalent thickness, see Figure 3.43.
wall
Clearly care must be exercised
when using beam (or shell) elements.
As noted above, the relative
magnitude
of the effect depends on
Secant pile wall
the value of K,, . It will also depend on
the interface properties between the
back of the wall and the soil and on
the number of props or anchors. The
effect is likely to be smaller if the
Contiguous pile wall
wall is propped.
Another source of approximation
concerns the shape of thkkall in plan.
A simple diaphragm wall usually has
a constant thickness and therefore its
T-shape diaphragm wall panels
properties may be constant along the
wall. However, sheet pile walls (see
Figure 3.43), secant pile, contiguous Figure 3.44: Different wall crosssections
pile and T-shaped diaphragm walls

Earth retaining structures / 109

(see Figure 3.44) do not have a constant cross section in plan. Such geometries
violate the basic assumption of plane strain conditions and consequently further
approximations are involved when performing plane strain analyses. This problem
is discussed further in Chapter 9.

3.7.3.2 WaN stiffness

The effective stiffness of the wall is obviously an important consideration for short
and long term conditions. It is common practice to assume that there is a reduction
in stiffness in the long term for concrete walls, due to factors such as creep and
cracking. Some structural engineers suggest that the stiffness should be reduced by
50% to account for this. In a finite element analysis this can be modelled by
changing the stiffness of the concrete after construction and prior to simulation of
long term pore water pressure equilisation.
Potts and Day (1990) undertook a
detailed study of the effects of a
variation in wall stiffness. They
performed a series of plane strain
finite element analyses of three
schemes: Bell Common tunnel, the
George Green tunnel and the House a) Bell Common tunnel
of Commons car park. Figure 3.45
shows cross sections through these
structures. The Bell Common tunnel
is propped at roof level only, the
George Green tunnel is propped at
both roof and road level and the
House of Commons car peak has
seven levels of support.
Each of these schemes was
analysed with retaining walls of
varying stiffness. For each scheme
and for each wall they looked at the
maximum horizontal displacement
and the maximum bending moment in c) House ofCommons car park
the wall in the short term and in the
tong term. Because ground conditions Figure 3-45:Cross-sections of three
and the propping arrangements VaV structures with different support
between schemes they could not
define unique relationships. The results of this study are summarised in Figures
3.46 and 3.47, where for each scheme the results are normalised by those predicted
for the stiffest wall section (i.e. the l m thick concrete wall). They show that the
bending moments in the wall are highly dependent on the wall stiffness, see Figure
3.46. There was a fivefold reduction in bending moment when the wall stiffness
was reduced from that of a l m wide concrete section to that of a Frodingham IN

1 10 1 Finite element analysis in geotechnical engineering: Application

sheet pile. This occurred for all three schemes. This clearly has implications on the
design of the wall.
End of conshuchon
10-

A BeUCommon

GmrgeOrem

m H o w of Commons

2Y

-R
10

15

10

Wall staRncss (In EI)

Wall sllffness

(In El)

l5

Figure 3.46: Variation of bending moment with

stiffness for the three schemes
3.0

End ofcoashuction
'

Wall S O ~ U (h
S EI)

Lnng term

Wall s h 5 w s (In@

Figure 3.47: Variation of lateral wall displacement

with stiffness for the three schemes

In contrast, the lateral wall displacements increased as the stiffness of the wall
reduced. The magnitude of the increase in displacement depended on the scheme.
For the Bell Common tunnel, which had a single prop, the maximum displacement
increased by as much as a factor of three in the long term. However, as the number
of propping levels increased, the increase in displacement reduced. For the House
of Commons car park there was only a very small increase in the maximum wall
displacement as the stiffness of the wall reduced.
Hight and Higgins (1994) undertook a much simpler study of a hypothetical
scheme thought to be reasonably typical of deep excavations in central London.
When looking at wall stiffness, they only considered short term effects. Figure 3.48
shows a cross section of the problem they analysed. Their results are shown in
Figure 3.49. The wall was l m wide and in the standard case the Young's modulus,
E, of the section was 28GPa. Two other cases were also considered, one with E

Earth retaining structures 1 11 1

reduced by a factor of 10 and
-erCnvcl
LTAU
one with it increased by a
Lo*
Clay
factor of 10. The results show
3
that reducing the stiffness of @
E:
Loodan Clay
the wall by a factor of l00 led
,E
to a doubling of the maximum
wall movement. There were
also important changes in the
displaced shape of the wall.
Vertical displacements and the Figure 3.48: Cross-section of a typical
curvature of the settlement
excavation and notation for results
trough behind the wall were
affected, but not far field
settlements.

AvT+ry

Unwcalbonsd

3.7.3.3 Interface
- IOE
behaviour
To model the interface
behaviour between the soil and
the wall, interface elements
can be placed around the Wall.
Horizontal wall
Day and Potts (1998) have
shown that the zero interface
elements described in Secti0n Figure 3.49: Predicted displacemen ts for
3.6 of Volume 1 work well in
different wall stiffness
this situation. They also show
that results are not too sensitive to the stiffness used for the interface elements. This
is good news, as interface stiffness is difficult to determine accurately. However,
they do show that the angle of dilation of the interface elements can have a
significant effect on soil movements.
3.7.3.4 Wall permeability
One important consideration which is often overlooked is the relative permeability
of the wall in relation to the surrounding ground. Obviously, some walls are likely
to be more permeable than others and measures can be taken to minimise seepage
through the wall.
An assessment of the effectiveness of the wall needs to be made and
appropriate boundary conditions applied in any analysis. The wall may be treated
as being completely impermeable, completely permeable, or even given some
finite permeability.
Solid elements representing the wall can easily model any of these conditions,
but if beam (or shell) elements are used more care must be taken. If elements with
zero thickness are used no distinction is made between nodes on either side of the
wall. The wall is therefore permeable. To make the wall impermeable a distinction

1 1 2 1 Finite element analysis in geotechnical engineering: Application

must be made between nodes on either side of the wall. This can be done by
introducing interface elements between the soil and the wall. However, a beam
element cannot be used to model a wall of finite permeability. This problem is
discussed in more detail in Chapter 9.

Support systems
3.7.4.1 Introduction
Section 3.4.4 discussed support systems in general tenns. Embedded walls can be
supported in a variety ofways. Different methods might be chosen to support walls
during construction to those methods adopted for the permanent condition. The
permanent support system may be used to support walls during construction, but
measures such as berms, anchors, grillages or props might be used which are
removed as the permanent system is installed. Generally, these temporary measures
are necessary to provide greater working space or more flexibility in the method
of working.
At the design stage uncertainties will exist over:

3.7.4

The effective stiffness of the support system;

- The time between excavation and the installation of supports;

- The level of prestress applied to the props or anchors.
All these factors need to be considered in any analysis.

3.7.4.2 Support stiffness

The effective stiffness of the support systems can be significantly reduced by
effects such as temperature, creep, bedding, concrete shrinkage, openings in floor
slabs, packing between the prop and the wall, etc. These effects are discussed by
St John (1975).
At the Bell Common tunnel,
Kork Pak
see Figure 3.50, a 'Kork' packing
was placed between the roof slab sheetp,lewall
and the walls to reduce thermal
btwall
effects. It was concluded by
Higgins et al. (1 993), based on
~ s ~ ~ ~ ~ ~ ~ c w a ~ :
S e ~ r c dun
e surround
finite element analyses of the
i
Central support
w a i l
tunnel, that the roof slab would
IOm
provide some support to the walls
O
+..+during the later stages of
construction and in the long term.

These analyses
used stiffness Figure 3.50: Cross-section of rhe Bell
values obtained from laboratory
Common tunnel
t e s t s p e r f o r m e d by t h e
manufacturer ofthe 'Kork' packing. However, field monitoring indicates that even
four years after construction little support is being provided by the roof slab. This

Earth retaining structures 1 1 13

is due to the much higher compressibility of the packing in the field which has
significantly reduced the effective stiffness ofthe support to the wall. This example
denlonstrates the difficulties in determining appropriate support stiffness.
The simple study of a
typical d e e p basement
undertaken by Hight and
Higgins (1994), see Figure
3.48, considered the effect of
variations in prop stiffness on
wall and ground displacements
during construction. An initial
analyses was performed with a
prop stiffness K=50 MNImlm.
Horizontal wall
This value is typical of that
provided by a concrete slab.

The analysis was then repeated Figure 3.51: Predicted displacernents for
with prop stiffness values of
differing prop stiffness
0.2K, 5K and 10K. The results
are shown in Figure 3.5 1. It is apparent that there is a very nonlinear relationship
between stiffness of the props and the predicted displacements. Because the
displaced shape of the wall was modified, there was a consequential effect on the
bending moments in it.
It is apparent that the effective stiffness of the support system can have a
significant influence on the observed and predicted behaviour of walls. Care must
be taken not to over-estimate the effectiveness of this support. However, in some
instances struts are pre-stressed in order to minimise problems associated with
bedding effects, thereby maximising the effectiveness of the support. This prestressing can, and should, be modelled in any analysis.

3.7.4.3 Connection details

The way in which structural components are joined together influences how forces
and moments are transmitted between them. They also influence how the structure
can distort and therefore the behaviour of the ground. Three alternative types of
wall-support connection are shown in Figure 3.11. The various options for
modelling these in numerical analyses are discussed in Chapter 9.
However, to illustrate the importance of the connection consider the example
shown in Figure 3.3. Figure 3.52 shows displaced shapes ofthe walls and roof slab
with two different connection details between walls and the roof slab. In one
nalysis there was a full moment connection between the roof slab and the walls,
ut in the second analysis relative rotation between the walls and the roof slab was
llowed (i.e. pin-jointed connection).
Figure 3.53 shows a comparison between the bending moments in the left wall.
here is a significant effect due solely to the anticipated connection detail. It is
apparent that such issues need very careful attention in any analysis.

1 14 / Finite element analysis in geotechnical engineering: Application

The ability of the connections

between supports and a wall to
withstand tension must also be
considered. It may be that if a support
is simply resting against the wall then
there can be no tension and the
analysis must not allow this to
develop. Therefore some form of
'cut-off has to be used which
prevents tensile forces from
developing. While this is possible if
the support is represented by finite
elements (e.g. solid, bar or beam
elements), it is not so easy if it is
represented by a displacement or
spring boundary condition.

Pm-jomted connection

3.7.4.4 Active support

Full moment connection
S ys terns
In order to control ground Figure 3.52: Displaced shapes of
for different corrnection
movements, active propping systems
details
are installed on occasions. During
construction ground movements are
monitored and if they exceed set
'trigger levels' the supports are
loaded to prevent any further
distortion or at least minimise it.
Either a specified load is applied to
the support or load is continually
adjusted to prevent movement.
Obviously, if such an active system is
mamenteo~ecnon
being considered, then any analysis
o
500
1000
ISM)
has to reproduce these effects and the -lrn
B h p moment ( W d m )
details of the controls (i.e. 'trigger
levels') have to be known.

FUII

Figure 3.53: Bending moments in

If the specified
of left wall for different connection
supporting the wall is that a constant
details
load is maintained in a prop or a
series of props (i.e. load control system), then the forces that result in the support
will have to be continually assessed and additional forces continually applied, or
removed, throughout the analysis to maintain this load. Alternatively, if in a
specification there is a limit on the maximum displacement of the wall
(displacement controlled system), then when this limit is reached a displacement
boundary condition can be applied to 'lock off any hrther movement. However,

Earth retaining structures 1 1 15

the correct boundary condition must be applied and the possibility that the wall will
want to move away (displacements falling below the specified maximum level)
must be considered.
The most complicated situation ofall is where the loads applied to the props are
controlled entirely by the displacement of a remote point such as a building. In
anything other than a simple structure there are obviously many different
combinations of loading that could be applied to the system.
Unless a finite element (or finite difference) program has an 'automatic'
method of modelling these systems, they can be extremely difficult to analyse.

3.7.4.5 Berms
Potts et al. (1993) described a study of the use of berms to provide temporary
support for retaining walls which considered variations in the geometry of the
berms. They related 'displacement efficiency' and 'normalised bending moment'
to the volume of the berm. There were a number of interesting results from this
study concerning the influence of the berm's geometry and the effectiveness of
berms. However, one of their conclusions was that:
"The horizontal resistance provided by the berm has a substantial effect on wall
behaviour. It is therefore not accurate to use an equivalent surcharge load to
model the effect of a berm. If such an approach is used, the effectiveness of the
berm is likely to be under estimated."
Berms are comparatively easy to model, although their stability, as drainage
occurs during construction, may need to be considered. There is also an issue
concerning the correct pore water pressure boundary conditions that should be
applied to their surface if they are left to stand for some time, see Section 3.4.7.4.
Problems can arise when berms are removed and a permanent support is
installed. The berm may be removed in sections and replaced with another form
of support, such as a slab cast in sections on the floor of the excavation. This
procedure is very difficult to model in a two dimensional plane strain analysis and
therefore approximate methods have to be considered. This may involve partial
berm excavation before installing the other form ofsupport or, considering a worst
case scenario, complete removal of the berm without another means of support.

3.7.4.6 Ground anchors

Ground anchors may be modelled in several different ways and this is discussed
in some detail in Chapter 9. However, in order to model the anchor it is necessary
to consider how the components that form the anchor act.
Generally, there is a fixed length and a free length ofthe anchor. Over the fixed
length shear stresses develop between the anchor and the soil, whereas over the
free length there is no frictional component. It is therefore necessary not to
overestimate the frictional forces that might develop between the soil and the
anchor as movement occurs.

1 16 / Finite element analysis in geotechnical engineering: Application

The most successful method of modelling an anchor is to use a combination of

beam elements and interface elements to model the fixed length. A limit placed on
the friction that can develop along the interface ensures that the influence of the
anchor is not overestimated and the elastic extension of the tendon is represented
by the beam element. To model the free length, a beam, bar or spring may be used,
but this should be disconnected from the soil in order to ensure that no friction
develops.
Clearly, anchors have a three dimensional geometry. Modelling them in a plane
strain analysis therefore involves severe approximations, see Chapter 9.

3.7.4.7 Relieving slabs

Relieving slabs may also be used to support a wall and these are quite easily
modelled as long as the soil behaviour is realistically represented. These slabs may
be placed either in front of a wall or behind it. In some circumstances slabs on both
sides of the wall may be used.
St John et al. (1993) described a
design study for a cantilever bored
pile wall with relieving slabs in front
of the wall and behind it. Figure
3.54 shows the effect of the
relieving slab behind the wall on
bending moments and wall
deflectlon Moment
deflecbon Moment
displacements. The beneficial effect
l FILLING
2 EXCAVATION
of the rear slab is in creating a
restoring moment to the wall prior to
excavation, which reduces the peak Figure 3.54: Effect of rear relieving
slab
bending moment and pulls back the
head of the wall. Both short and
long term stability is enhanced.
Figure 3.55 shows the effect of a
slab in front of the wall on the long
term behaviour. Since the front
relieving slab can only be installed
deflechon Moment
deflecbon Moment
after excavation is completed, it can
only have an effect on long term
SHORTTERM
LONG TERM
behaviour. The benefit is again in
reducing the maximum wall bending Figure 3.55: Effect of front relieving
moment.
slab
The cross section used for St
John et al.'s study is shown in Figure 3.56 and the effect of these slabs on
horizontal wall displacements is summarised in Figure 3.57. Run 5 considered a
cantilever wall with no relieving slabs, in Run 2 there was a slab behind the wall,
while in Run 7.1 only a slab in front of the wall was constructed. The stages of
analyses, marked on Figure 3.57, correspond to:

Earth retaining structures / 1 17

Construction of bored pile wall (i.e. 'wished in place');

Constructioh of slab behind the wall (where appropriate);
Construction of fill behind the wall;
Application of bridge loads on the wall;
Excavation to El. +19.7m;
Ground drainage in front of the wall;
Construction of front relieving slab (where appropriate);
Long term dissipation.
Clearly the slabs, especially the one at the rear of the wall, substantially reduce
wall movements.
A structure of this form can only be analysed using a full numerical approach.

+3

1I

Folkstone beds

I
I

Figure 3.56: Cross-section analysed for

the design study
Run S

Run 2

Run 7.1

Horizontal displacement(mm)

Figure 3.57: Horizon tal wall

displacements

1 18 / Finite element analysis in geotechnical engineering: Application

3.7.5

Long term behaviour and post construction effects

In stiff over consolidated clays, such as London Clay, long term effects can be
significant and can dominate the behavnour of a structure taking many years to
occur. London Clay is a relatively well understood soil (emphasis should be placed
on the word relatively), but even so there are few case records with measurements
of long term heave against which the predictive capability of constitutive models
can be assessed.
In any design possible long term effects need careful consideration and
measures taken to mitigate them or to cope with them. Ifconstruction is slow, some
drainage will occur during this period and this will tend to reduce post-construction
effects.
Apart from changes in earth 6
pressures that act on a wall as
drainage occurs and excess pore water
pressures dissipate, one of the most 8 4
Small stram shfhess-plastrc
significant long term effects is the 3
g
Oedometerdata
swelling of clays beneath deep 3
mples from 8-24m depth)
excavations. Figure 3.58 shows $ 2
Linear elashc-plast~c
measurements of swelling in
oedometer tests on London Clay,
together with two predictions. One of o
100
200
300
400
these predictions is based on a linear
Verttcal effectwe stress (&a)
elastic-plastic constitutive model. The
other has the same plastic behaviour
Figure 3, 58: PIedi.tjons and
but the elasticity is based on the small
measurements of S welling behaviour
strain model described in Section
of London Clay
5.7.5 of Volume l . It is apparent that
at low effective stress the test data indicate that large volumetric strains occur and
that large vertical displacements result. The predictions based on the linear elasticplastic constitutive behaviour are unable to reproduce this behaviour. However,
reasonable predictions are obtained from the small strain model.
To minimise the effects of these large volumetric strains and pore water
pressure changes, the swelling potential of the clays may need to be controlled by
drainage, or by other means. Alternatively, a void may be left beneath a basement
to allow heave to occur without imposing any stresses on the base slab of a
structure. Modelling ofthis is reasonably straightforward, but the depth ofthe void
needs to be known before the analysis is performed. Throughout the analysis the
heave is monitored at nodes along the soil surface beneath the base slab until the
displacement of the soil equals the depth of the void. Subsequently, dispiacements
at nodes along the soil surface become linked (i.e. tied) to the corresponding nodes
on the slab (this is not a zero displacement boundary condition, see Section 3.7.4
of Volume l), and forces transferred from the soil to the structure. Unfortunately,
this will not occur evenly across the base of an excavation due to the nonuniformity of stress conditions.

Earth retaining structures 1 1 19

An alternative solution may be to disconnect the base slab from the walls and
the rest of the structure and allow it to move freely with the soil. In this case care
needs to be taken to ensure that the correct connection detail is modelled. It is all
too easy to have a mesh that implies a full moment connection, in which case the
base slab will not move freely as the soil swells but will resist movement and
transfer forces to the rest of the structure.
As in the example of the road tunnel shown in Figure 3.3, there may be piles
beneath the base of the structure. These will inhibit vertical heave due to swelling.
However, these piles have a three dimensional geometry and consequently, when
modelled in a plane strain analysis, assumptions are necessary, see Chapter 9. It is
important that the frictional effects between the ground and the pile are not
overestimated, equally they should not be underestimated.
Even in fast draining soils postconstruction effects may be
significant. It may be necessary to dewater the ground, or certain strata
within it, in order to form an
-5
excavation, see Section 3.4.7.4. ';;
Completion of the structure and -lo
turning off of the de-watering system
.,,
may subject the structure to
significant pressures which have to be
accommodated. The analysis of a ,
deep excavation at Langley Point,
was
Fernie et Figure '3.59: ('ross-section for the
a'. (1996) Figure 3.59 shows a cross marine treatment works at Langle
section through this excavation. The
Point
base slab was restrained by deep piles
which restricted its movement.
During excavation a de-watering system was installed to reduce pore water
pressures within the area of excavation. When the de-watering system was turned
off, after construction was complete, the soil beneath the base slab swelled and
pushed the walls back (the authors termed this the 'balloon effect'). This had the
effect of increasing the maximum bending moment in the walls by 50% and
subjecting the connection between base slab and walls to tensile forces. In these
analyses very careful consideration had to be given to the modelling of the
connection between the base slab and the walls.

3.7.6 Adjacent structures

One of the main reasons for undertaking a soil-structure interaction analysis is to
make an assessment ofthe effect of construction of a particular scheme on adjacent
structures andlor services. As discussed in Section 3.4.6, the presence of such
structures may have influenced ground conditions and the subsequent behaviour
of it.

18 / Finite element analysis in geotechnical engineering: Application

125

.;a loo

% 75
?

50
25
0
'

40

80

120

160

200

-gl*

140
120

60
40
20

10

Figure I . 1 7: Typical results from a

hollo W cylinder experiment: a ) J-p '
projection of c ) J-p '-a
S tress paths
12
and b) J-E,,
stress-strain curves

Figure 1.18: Anisotropy of a) undrained and

b) drained strength

In addition to information on soil strength,hollow cylinder experimentsprovide

information on the directional variation of soil stiffness, i.e. stiffness anisotropy.
For example, Figure 1.19 shows results comparing the stiffness of samples sheared
in drained conditions by changing only the vertical effective stress (E,,'=EI1=
AoZ1lAcZ),
the horizontal effective stress (E,,'=EOt=AG~'/A&~),
or the torsional shear
stress (G,,,, = G,,, = AtZJAy,,,), see ZdravkoviC and Jardine (1997).
Drained directional shearing can also be used for estimating Poisson's ratios.
From a drained test where only the vertical effective stress is changing, the
Poisson's ratio p,' (horizontal straining due to a vertical direct stress) can be
calculated as:

Obtaining geotechnical parameters / 19

From a drained test where only the horizontal effective stress is changing,
Poisson's ratios v,' (vertical straining due to horizontal stress) and',v (horizontal
straining due to horizontal stress) can be calculated as:

p,',, = p& = --&r

&S

From these results the anisotropic compliance matrix can be constructed:

--PL

--'4%

E;,

E;,

--P&
E;

This compliance matrix is symmetric in the linear elastic region (i.e. elastic
plateau) of soil behaviour and in that case:

P;@ P;
---

E;

However, at larger strains, but

still within the small strain
region, nonlinear parameters
do not necessarily produce a
symmetric compliance matrix
(see Zdravkovid and Jardine
(1997)).
Hollow cylinder testing is
highly desirable if anisotropic
soil behaviour is to be
simulated through models
such as MIT-E3.

E;,

Figure 1.19: Stiffness anisotrop y obtained

from a hollow cylinder test

20 1 Finite element analysis in geotechnical engineering: Application

1.3.9 Directional shear cell

The directional shear cell (DSc) is an
example of a piece of specialist
ressure
bags
equipment capable of investigating
soil anisotropy. It was developed at
University College London (UCL) by
Arthur and Menzies (1972) for testing
samples under plane strain conditions.
The plane strain conditions are
imposed on two opposite faces of a
cubical sample by end platens. On the
other four sides both direct and shear
stresses are applied via flexible
membranes and pressure bags (see
Membrane
sheets
Figure 1.20) so that the direction of
the major principal stress can be
Figure 1.20: Schematic presentation
rotated in a controlled manner. The
of a directional shear cel/
UCL version of the DSc is capable of
testing only dry samples, whereas the MIT version (Germaine (1982)) has been
developed to conduct undrained tests too.
Strain measurement is normally performed via photometry andlor radiography,
which is a labourious process and prevents the testing from being conducted using
servo-control. Because the apparatus is restricted to testing samples under plane
strain conditions it is not as a flexible as the hollow cylinder apparatus.
As with the hollow cylinder apparatus tests performed in this apparatus provide
information on both the anisotropic stiffness and strength of the soil samples.
1 -3.10 Geophysical techniques
Recently the bender element technique has been developed for very small strain
(i.e. on the elastic plateau) shear stiffness measurements in the laboratory. Bender
elements are plates made of a piezoelectric material which has the capability to
either convert an applied electric voltage into mechanical motion, or convert an
applied mechanical motion into an electric voltage. These plates are placed onto
two opposite sides of a sample: one plate acts as a transmitter that generates a shear
wave which propagates through the soil until it reaches the other plate which acts
as a receiver. From the measured time that it takes for the shear wave to traverse
the sample and the distance between the bender plates it is possible to calculate the
velocity of the shear wave and hence estimate the value of the elastic shear
modulus G,, (=pV2, where p is the material density).
Schultheiss (1981) was the first to install bender elements in triaxial and
oedometer cells, and others then followed suite (e.g. Lo Presti et al. (1993),
Jamiolkowki et al. (1994), Viggiani (1992), Porovid (1995), JoviEiC and Coop
( 1 998), Kuwano (1999)). The use of bender elements in a conventional triaxial test

Obtaining geotechnical parameters / 2 1

allows three different shear moduli to be determined. For triaxial samples with
bedding planes positioned horizontally,the shear modulus G,,can be measured by
propagating, from the top of the sample, a verticaf shear wave which oscillates in
the horizontal direction, Figure 1.21a. The other two moduli, Gh,,and G,, , can be
measured by placing two more pairs of bender elements on radially opposite sides
of the sample, see Figure 1.2lb, or by trimming a sample such that the bedding
planes are vertical and placing it in two positions relative to the axial bender
elements, as shown in Figure 1.21~.

DP - direction of wave
propagation
DO - direction of wave
oscillation
BE - bender element

Figure 1.2 1: Small strain stiffness

measurements using bender elements

22 1 Finite element analysis in geotechnical engineering: Application

1 -3.1 1 Permeameters
The most common means of measuring
permeability in the laboratory is to use a
permeameter. However, because of the very
large range of permeability in soils, no
simple piece of equipment is suitable for all
cases. The types ofequipment most generally
used are:

-- - -----Waste

The constant head permeameter, for

permeabilities, k, down to about
10-4m/sec, and
The falling head permeameter, for
values of k between 10-4 and
10-'m/sec.

Below about 10-7m/sec permeability can

generally only be measured by indirect
means in a consolidation test in the
oedometer, or in a dissipation test in the
triaxial apparatus.

Water out'

Figure 1.22: Schematic

presentation of a constant
head permeameter

The constant head permeameter

This apparatus is illustrated diagrammatically
in Figure 1.22. A constant head of water is
maintained across the sample during the test.
Once steady state conditions are reached, the
rate of flow is measured. From the
measurements of flow rate and the hydraulic
gradient, Ahll, it is possible to use Darcy's
law to determine the permeability, k.
The falling head permeameter
When the rate of flow through the sample is
too small to be accurately measured in the
constant head permeameter, the falling head
apparatus is used, see Figure 1.23. In this
case the water flows through the sample from
a standpipe, and the head across the sample
~eefohted
base
at any moment is taken to be the difference
Figure 1.23: Schematic
between levels in the standpipe and in the
container surrounding the sample. Noting the presentation of a falling head
rate of drop of the water level in the
permeameter
standpipe and using Darcy's law enables the
permeability to be estimated.

Obtaining geotechnical parameters / 2 3

1.4

In-situ tests

1.4.1 Introduction
In-situ soil testing offers complementary capabilities to laboratory testing by
providing information about current conditions in the ground (e.g. piezometric
levels, strength variation with depth, soil description,etc.). The principal advantage
of in-situ tests is that they assess soil behaviour in its natural condition, thus
avoiding sample disturbance. They can also be more economical than laboratory
tests and in some cases can provide more representative results through a higher
density of data collection.
However, the disadvantage of in-situ testing is that most geotechnical
parameters (e.g. strength, stiffness, etc.) are estimated from empirical correlations,
rather than from direct measurement. Also, in-situ tests load the ground in different
ways to conventional laboratory tests. Because soil properties are nonlinear,
anisotropic and rate dependent a wide range ofresults can be obtained from similar
tests in the same deposit. Therefore great care is required in test preparation.
In many practical situations the only information that exists about a particular
site comes from in-situ testing. This often restricts any finite element analysis to
the use of simple constitutive models.
This section describes the most common in-situ equipment and procedures for
soil investigation and the modelling parameters that can be obtained from them.

1.4.2 Standard penetration test (SPT)

The standard penetration test (SPT) is performed at the bottom of a pre-drilled
borehole, by driving a standard sampling tool through a prescribed distance (450
mm), with a hammer of known weight (63.5 kg). The hammer is dropped from a
fixed height of 760 mm onto an anvil placed on the top of the drilling rods and the
number of blows required for it to penetrate 450mm recorded. The drilling rods are
connected to the back of the sampler and transfer the driving energy from the
hammer to the sampler. The borehole diameter is normally within the 65 to 1 151nm
recommended in most guidelines, because it is known that larger boreholes result
in lower blow counts. Another recommendation is that the SPT test is performed
in a water filled borehole.
The number of hammer blows required to drive the sampler through the first
150mm is normally discarded (as being affected by drilling disturbance). The
number of hammer blows required to drive the sampler through the last 300mm is
taken as the blow count, N. Penetration and blow-counting continues until the
desired depth of the borehole is reached and a typical borehole log is shown in
Figure 1.24.
The SPT is mainly used in granular soils because of the difficulties associated
with taking undisturbed samples for laboratory testing. However, it is also used for
testing clays. It is a simple and cheap test, but not easy to interpret. The reason for
this is the variety of sampling tools, anvils, hammers and methods of hammer
release that exist. They all affect the energy transmitted from the hammer into the

24 / Finite element analysis in geotechnical engineering: Application

sampler and consequently affect the blow count.

Skempton (1986) suggested that the N values
measured by a particular method should be
normalised using a standard procedure. If the
hammer energy, E,, delivered into the rod stem can
be expressed as a ratio of the theoretical free-fall
energy, E,, such that:

: Depth of

sample (m) SPT test

N=3 1

where ER, is the 'rod energy ratio', and if the

standard ER, is adopted to be 60%, then the Nvalues
measured with a known or estimated ER, value can
be normalised to this standard by the conversion:

where N6, is the normalised blow count.

Stroud (1989) summarised the existing
correlations between N6, and soil density, strength
and stiffness, derived for both clays and sands. Some
of his findings are outlined below.
Sand - strength

Skempton (1986) suggested that Terzaghi and

Peck's (1948) correlation between blow count, N,
and relative density, D,, should be corrected for the
above normalisation. The new expression then
becomes:

Figure 1.24: Typical

SPT log

where a and b are material constants and g,,'is the effective overburden pressure.
He further suggested that it was convenient to normalise the blow count in the
above expression to a standard effective overburden pressure of 100 kPa (i.e.
l kg/cm2):

and C, is a correction factor for overburden pressure, given in

where (Nl)60=CNN60
Figure 1.25.
He then derived the relationship between relative density, D,, and standardised
SPT value, (NI),, , for sands shown in Figure 1.26. Bolton (1986) introduced the
critical state value of the angle of shearing resistance, p,', into the correlation and,
on the basis of data collected for different sands, drew a revised relationship.

Obtaining geotechnical parameters 1 25

Stroud (1989) then replotted all the available data to obtain the variation of 9' vs.
(N,),, and (NI), vs. D, for different OCRs and different cp,,', as shown in Figure
1.27.
Correction factor C,

8
d

l00

01

NC coarse sands

'ES!

e
8 200

m NC coarse sands
Laboratow tests

.M
+a

Relative density, D, (%)

Figure 1.25: Correction for

effective overburden pressure
(after Skempton / 198611

Figure 1.26: Effect of relative

density, based on field data
(after Skempton ( 1986))

Sand - stiffness
It is now widely accepted that soil stiffness is strain level dependent, i.e. it
gradually reduces as strains increase. To accurately predict movements of
structures in the ground (e.g. footings, retaining walls, tunnels etc.) it is necessary
to obtain a good estimate of the stiffness of the ground.
Vesid (1973) performed tests with a loaded foundation of a constant breadth in
sands of varying density. He found a unique relationship between settlement, p,
and the degree of loading, qlq,,, , expressed as a ratio of the design load, g, to the
ultimate load, g,,, . This suggested that qlq,,, could be used as an indirect measure
of shear strain and therefore linked to the average stiffness E' (expressed in terms
of Young's modulus).
For this purpose data from a wide range of strip footings, raft foundations and
plate tests on both normally and overconsolidated sands have been collected
(Stroud (1989)). They have been taken from case histories where SPT tests have
been carried out and a correlation, shown in Figure 1.28, has been derived between
the E'IN,, and qlq,,, values. Knowing the design loading and estimating the
ultimate bearing capacity, it is possible to use this graph to estimate the average
ground stiffness at working load on the basis of SPT blow counts.

26 1 Finite element analysis in geotechnical engineering: Application

Clays - strength
The undrained shear strength, S,,, of overconsolidated clays can be related to SPT
blow counts using the simple correlatio~lproposed by Stroud (1 974):
= jfi

(1.23)

N60

where the variation off, with plasticity index is shown in Figure 1.29.

Clay - stiffness
The drained stiffness, E', again expressed in terms of Young's modulus, of
overconsolidated clays in vertical loading can be estimated from case histories, in
a similar manner as for sands. Figure 1.30 shows such data plotted as E'/&, against
qlq,,,. Again, a trend of decreasing stiffness with increasing degree of loading (i.e.
increasing strain level) is evident.

consolidated
40

--35'
33'
31'

30

Figure 1.27a: Variation of 4' and IN,),, with

&'and OCR (after Stroud (79891)

F w sand
skempton
CO- sand) (1986)

Well @add
sand
and gravel
D Sand

0 Nonvich Crag - Skempton (1986)

V Bagshot saod Barton (1988)

o F i saod

Figure 1.276: Variation of IN,), and D, with

&, ' and OCR (after Stroud (7989))

Obtaining geotechnical parameters / 27

?
Average for overconsolidated
sands and gravels
Average for normally
consolidated sands

0.1

0.2

0.3

Figure 1.28: Stiffness vs.

degree of loading for sand
(after Stroud (1989))

Figure 1.29: Variation of f , with

plasticity index
(after Stroud (6 989))

Figure 1.30: Stiffness vs,

degree of loading for clays
(after Stroud ( 1989))

1.4.3

Cone penetration test (CPT)

The cone penetration test (CPT)consists of pushing into the ground a series of
cylindrical rods with a cone at the base. During the push the penetration resistance,

28 / Finite element analysis in geotechnical engineering: Application

q,, and, if required, the frictional resistance on the side (sleeve) of the cone,f;, are
recorded at depth intervals. A typical CPT result is shown in Figure 1.31. The ratio
A/qCis termed the friction ratio.
Empirical correlations are available between these measured quantities and the
angle of shearing resistance and relative density for sands and undrained shear
strength for clays. These correlations should be used with care because they may
only apply to a particular set of conditions (e.g. penetrometer dimensions,
penetration rate, soil type).
Cone resistance, g, (MPa)
Local side friction,

L (~~lm')

Cone resistance,
9, m / m 2 )

Figure 1.3 1: Typical CPT log

Figure 1.32:Cone resistance

vs. peak 4' for sand
(after Robertson and
Campanella / 1983))

Sand
The existing correlations of cone resistance with v', E and G have been derived
from tests performed in a calibration chamber.
Robertson and Campanella(1983) presented relationships between g, and peak
9' for normally consolidated quartz sands at increasing vertical effective stress, see
Figure 1.32. The angles of shearing resistance were obtained from triaxial tests
performed at confining stresses approximately equal to the horizontal effective
stress in the calibration chamber prior to cone penetration. Using the relationships
given in Figure 1.32 for overconsolidated sands will tend to slightly overestimate
9'-

Robertson and Campanella (1983) also presented relationships between cone

resistance, q,, and Young's and shear moduli at increasing vertical effective stress
levels, see Figure 1.33.

Obtaining geotechnical parameters / 29

Cone resistance,g, (XIIPa)

300 -I

Cone resistance,q, (MPa)

Figure 1.33: Cone resistance vs.

Young's and shear moduli for sand
(after Robertson and Campanella
(198311

Clays
The measurement of clay properties using the CPT is highly dependent on the rate
of cone penetration, because of the build up of excess pore water pressures. The
CPT in clays is principally used to estimate the undrained shear strength, S,,, and
the relationship with g, is expressed as:

qc = Nk
(1.24)
where: G,,,
is the total overburden stress;
Nk is the "one factor' analogous to the bearing capacity factor N,, see
Chapter 6.
Nk is established using empirical correlations, with shear strength measured
using other techniques. Lunne and Kleven (1981) found that for normally
consolidated clays N, was independentof plasticity index and suggested an average
value of 15, with the majority of data being between 11 and 19.

30 / Finite element analysis in geotechnical engineering: Application

For overconsolidated clays Marsland and Quarterman (1 982) presented cone

factors based on the undrained shear strength from plate loading tests and showed
that for stiff fissured marine clays the cone factor was as high as 30, with an
average of about 27.

1.4.4 Pressuremeter testing

The pressuremeter is a device that applies a
uniform pressure to the walls of a borehole
and measures stress and deformation of the
wall, which is then used for interpretation of
soil properties. Although there are three types
of pressuremeter device (i.e. Menard (MPM),
self-boring (SBP) and cone (CPM)) the basic
concept consists of a defomable, cylindrical,
rubber membrane which inflates laterally to
radially deform the soil (see Figure 1.34).
Interpretation ofpressuremeter test results
is based on cavity expansion theory. It
consists of: (i) an elastic phase, during which
soil deformation is considered to be isotropic
elastic, and (ii) a plastic phase, during which
soil behaviour is assumed to be perfectly
plastic. From measurements taken during the
elastic phase estimates of the shear modulus
can be made, while during the plastic phase
the measurements can be used to evaluate the
shear strength.

P - cavity pressure

Figure 7.34: Schematic

presentation of Menard
pressuremeter

Elastic expansion
The main parameters and variables for membrane expansion are shown in Figure
1.34. The radius of the cavity, when the cavity pressure P is equal to the initial
, , is marked as p,. When P>o,,,,, the cavity will
horizontal stress in the soil, a
deform, having a new radius p. The cavity strain, E,, is then given by:

The pressuremeter tangent shear modulus of the soil is then calculated as:

For most practical purposes pip,,= 1, which simplifies Equation (1.26).

However, G,, is not the secant modulus of the soil, G,$,which is the parameter
normally used for analysis in geotechnics. It can, however, be related to it through

Obtaining geotechnical parameters / 3 1

the following expression:

Estimates of a single (average) value of G,vcan be made from the initial portion
ofthe pressuremeter P-&,curve or from unload-reload loops.
Jardine (1992) compared triaxial G,v-~s
curves, where 6, is the shear strain
(=2/3(~,,-E,)), with pressuremeter Gp-c, curves and derived an empirical
relationship between E, and by comparing the strain values at identical shear
moduli levels:

With the above expression pressuremeter G,-&, curves can be transformed into
triaxial G,,-< curves, which are normally used in geotechnical design. For example,
such curves can be used to obtain parameters for the small strain models described
in Chapter 5 of Volume l .
When using the pressuremeter in sands it should be noted that the shear
modulus is likely to be stress level dependent. As the stress level in the sand
adjacent to the pressuremeter decreases with distance, the value determined from
the pressuremeter test represents an average value.
It is not possible to obtain estimates ofthe bulk modulus, K, or Poisson's ratio,
p, from pressuremeter tests.

Has tic expansion

If the soil behaves as a Tresca material, yielding will occur when the shear stress
in the cavity wall reaches the undrained shear strength of the soil:
(1.29)
P = oh,+ S,,
If the cavity pressure increases above this yield value, an annulus of plastic soil
will develop around the cavity, while the soil further away from the cavity will
continue to behave elastically. The cavity pressure can then be expressed in the
following form:

where V is the volume of the cavity occupied by the inflatable membrane.

With further loading a limit condition will be reached where the cavity expands
with no change in pressure, at which stage AV/V=I and the limiting cavity
pressure, P,, becomes:

32 / Finite element analysis in geotechnical engineering: Application

The current cavity pressure, P, given by Equation 1.30 can therefore be reexpressed as:

If the cavity pressure P is plotted

against In(AV1V) for the duration of
the pressuremeter test, it should give
a straight line with a gradient equal to
the undrained strength S,, during the
later stages of the test, see Figure
1.35. Undrained strengths determined
from pressuremeter tests are often
significantly larger than values a,
obtained by other means and therefore
should be used with caution.
e,
E
While it is theoretically possible to
obtain estimates of the angles of
shearing resistance, p', and dilation,
v, from a pressuremeter test in sand, it Figure 1-35:Cavity Pressure vs.
volume change during a
is difficult in practice due to the
pressuremeter test
disturbance involved in installing the
instrument in the ground, see Mair
and Wood (1 987).

Coefficient of earth pressure at rest

As noted above, under ideal circumstances, the radius ofthe cavity will only begin
to deform when the pressure P equals the in-situ horizontal stress a,,. If this
pressure is noted along with the in-situ pore water pressure, then it is possible to
estimate the coefficient of earth pressure at rest, K,, in the ground. In practice such
measurements are only usually possible with the selfboring types ofpressuremeter.

1.4.5 The plate loading test

The plate loading test provides a direct measurement of the cornpressibility and
bearing capacity of soils. A plate (0.3m to 1.0m in diameter) is loaded, via a
reaction system (see Figure 1.36), and a loading test is performed in one of two
possible ways:

Maintained load test, where load is applied in increments of about 115th the
design load and held at that level until the rate of settlement reduces to less than
0.004 mmlmin over a one hour period. The test continues until either the soil
fails in shear, or the bearing pressure reaches 2 to 3 times the design pressure.
The results are usually plotted as time vs. settlement and load vs. settlement
curves (see Figure 1.37).

Obtaining geotechnical parameters 1 33

- Constant rate of penetration test, where load is applied in acontinuous manner,

to give a plate penetration of2.5 mmlmin. The ultimate bearing capacity is then
defined as the load required to produce a settlement equal to 15% if the plate
diameter. This test is less common.
The interpretation of deformation properties is usually made using isotropic elastic
theory or empirical correlations.

Figure 1.36: Schematic /a yout f ~ ar

plate loading test
Bearing pressure beneath plate (Ha)

Time from commencement of test (h)

Figure 1.37: Typical presentation of

results from a plate loading test

34 / Finite element analysis in geotechnical engineering: Application

Sand test
Using elastic theory the settlement of a rigid surface plate of diameter B, with a
uniform load q applied on a semi-infinite isotropic soil characterised by Young's
modulus E and Poisson's ratio p, is given by:

Terzaghi and Peck (1 948) observed that elastic theory tends to overestimate
settlements on sands and proposed that the settlementp, of a footing ofwith B was
better represented by the following correction:

where p, is the elastic settlement (i.e. Equation (1.33)) of a one foot plate (i.e.
B=0.3m).

Clay test - settlement

Marsland and Easom (1973) highlighted the .
fact that the settlement of a loaded plate on
clay deviates more from the isotropic elastic
theory than the equivalent plate on sand.
The test can be performed either on the
surface of the clay or in a borehole. The
borehole test, where D is the borehole
diameter, is shown in Figure 1.38. Elastic
theory predicts that the settlement of a fully
buried rigid plate is equal to half of .the
settlement of the equivalent surface plate.
However, in a borehole the plate is not fully
buried and Burland (1969) showed that
correction factor of 0.7 should be applied. figure 1-38:Plate loading rest
When BID<0.6, where B is the plate
in a borehole
diameter, the surface analysis applies.

Clay test - undrained strength

The undrained strength of clays can be calculated using bearing capacity theory:

where:

q, is load at failure, measured in a test;

p, is the overburden pressure;
N, is a bearing capacity factor, see Chapter 6.

Obtaining geotechnical parameters / 35

1.4.6

Pumping tests

Because of the limited size of the samples that can be tested in the laboratory, they
may not provide a representative value of the in-situ permeability. For example,
the sample may not contain a thin seam of more porous material that may exist insitu. In addition, samples of granular soil have to be reconstituted in the laboratory.
For these reasons in-situ measurements of permeability are generally more reliable
than laboratory tests. However, they are usually considerably more expensive.
In the field the permeability of a stratum of soil is most commonly determined
by measuring the discharge from a well. A well (borehole) is sunk into the stratum,
and water is pumped from it at a constant rate. This pumping lowers the
piezometric level in the vicinity of the well, and the resulting hydraulic gradient
causes water to flow towards the well. The piezometric level may be determined
by sinking observation boreholes at various distances from the well, and by
measuring the height at which the water stands in each, once a steady state has
been established. From this data and using Darcy's law an estimate of the average
permeability can be determined.
While pumping tests are applicable to relatively permeable soils, they are not
necessarily appropriate for some clays. In these cases a borehole can be sunk into
the clay and sealed over its length in contact with overlying soils. If left water will
flow from the clay into the borehole. By measuring the rate of increase in the
height of the water level in the borehole and again using Darcy's law it is possible
to estimate the permeability. Alternatively, the borehole can be filled to the ground
surface with water and the rate of reduction in level noted. Coupled with Darcy's
law this allows an estimate of k to be made.

1.5
1.

2.

3.

4.

Summary
Both laboratory and field tests are available from which soil parameters can
be derived. Both types of testing have their strengths and weaknesses and
ideally most soil investigations should contain a combination of tests.
The aim of laboratory tests is to measure as accurately as possible and in a
controlled manner the response of a soil element to changes in stresses, strains
andlor pore water pressures.
Oedometer tests can be performed to provide information on
overconsolidation ratio, OCR, the compression index, C,, and the swelling
index, C,. With an appropriate assumption for K, it is also possible to estimate
the gradients of the virgin consolidation line,I, and swelling lines, K, in v-lnp'
space. It is also possible to obtain estimates of the constrained modulus, E,,
and the coefficient of permeability, k.
The conventional triaxial apparatus can be used to estimate K, values, strength
values c', p' or S,, and stiffness values E and p (or K and G). It can also be
used to estimate permeability. However, it is restricted to testing samples
under triaxial compression (b=O) or triaxial extension (b=l) conditions and
therefore cannot conveniently be used to investigate the effect of varying b or
for investigating anisotropic behaviour.

36 1 Finite element analysis in geotechnical engineering: Application

The true triaxial apparatus is an. improvement over the conventional

apparatus, enabling stiffness to be determined in three independent directions
and strength values under various b values. It is, however, not possible to vary
the orientation of the principal stresses in a controlled manner. It is also
cumbersome to use and therefore few exist.
Direct shear box tests can be used to estimate strength values. Due to the non uniformities induced in the sample they are not normally used to estimate
stiffness parameters.
Simple shear tests, while potentially an improvement over direct shear box
tests, also induce non-uniformities in the sample and consequently are usually
only used to estimate strength parameters.
Ring shear tests suffer from all the weaknesses associated with the direct and
simple shear tests, but allow samples to be sheared to very large
displacements. They are used to estimate residual strength.
The hollow cylinder apparatus overcomes many of the shortcomings of the
above apparatus, allowing control over both the magnitude and direction of
the principal stresses. Both stiffness and strength parameters can be estimated
from tests performed in this apparatus. Such tests are ideal for investigating
the anisotropic behaviour of soil.
The directional shear cell can also be used to investigate anisotropic soil
behaviour. However, tests are usually restricted to plane strain conditions. As
very few apparatus exist such tests are not common.
incorporation of bender elements into the soil samples in the above apparatus
enables geophysical techniques to be used to estimate stiffness parameters.
Their use therefore provides a welcome addition to standard laboratory tests.
Permeameters, both constant and falling head types, are used to measure soil
permeability in the laboratory. However, for very low permeabilities, k<lO-'
mlsec, indirect methods (i.e. consolidation andlor dissipation tests) are often
employed to determine k.
Field tests come in many forms, but usually involve testing soil under nonuniform conditions (i.e. many tests are essentially boundary value problems).
In addition, many ofthe geotechnical parameters are estimated from empirical
correlations, which have often been calibrated against laboratory tests.
The standard penetration test (SPT) is probably the most common form of insitu test. It is simple to perform, but all geotechnical parameters must be
estimated from empirical correlations. Such correlations exist between the
measured SPT blow count and strength parameters c', p' and S,,,and Young's
modulus E.
The cone penetration test, while an improvement over the SPT, still relies on
empirical correlations to obtain geotechnical parameters. Again, it is possible
to estimate values of S,,, p' and E. Such tests also provide good profiling
tools.
Pressuremeter tests can be used to estimate strength parameters S,,and p', the
coefficient of earth pressure at rest K, and the elastic shear stiffness G. The

Obtaining geotechnical parameters / 37

interpretation is based on theory related to the expansion of an infinitely long

cylinder. As such, the estimated parameters are very sensitive to the
disturbance caused during installation of the pressuremeter. This is
particularly so for tests performed in sandy soils.
17. Plate loading tests can be used to provide estimates of both soil stiffness and
strength. To obtain stiffness parameters the interpretation is based on the
assumption that the soil is isotropic linear elastic, whereas the soil is assumed
to be linear elastic perfectly plastic when estimating strength parameters.
18. Pumping tests can be performed to obtain estimates of the average in-situ
permeability.

Tunnels

2.1

Synopsis

This chapter presents the application of numerical methods to the analysis of tunnel
construction, with emphasis on soft ground tunnelling. First, different methods of
tunnel excavation are presented, some of which have application in rock and soil.
The simulation of this excavation process is then discussed, with particular
reference to plane strain analysis, followed by the ways of modelling tunnel
linings. The chapter then focuses on the modelling of time dependent behaviour
and on constitutive models appropriate to the modelling of soft ground tunnelling.
The final section presents examples of analyses of tunnelling in stiff clay. These
examples include the analyses of case studies and parametric studies, covering
greenfield tunnelling, tunnelling beneath buildings, and tunnelling adjacent to
existing tunnels.

2.2

Introduction

In this chapter some ofthe numerical techniques presented in Volume 1 are applied
to the analysis of tunnel construction. Attention is restricted to bored tunnels.
Tunnels formed by the cut and cover constructiontechnique are covered in Chapter
3. The Authors' experience and expertise is in the area of tunnelling in soils and
soft rocks modelled as continuum materials, where structural features, such as
faults and joints, are not represented. This chapter therefore deals with soft ground
tunnelling.
The first question to be asked of anyone advocating the application of
numerical methods to tunnelling is surely '"hy use numerical methods to analyse
tunnel construction?" There are non-numerical ways of obtaining good predictions
of the likely ground response to tunnelling, and the likely loads in a tunnel lining.
These conventional design tools are arguably cheaper and quicker to use. But they
are characteristically uncoupled, i.e. the loads are determined by one technique
(usually an elastic solution), and movements by another (usually empirical), the
two not being linked together. Furthermore, the information gained from
conventional analysis is often limited. For example, empirical predictions are
limited to greenfield situations, where there is no existing surface or sub surface
structure to influence the pattern, or magnitude of ground displacement. In a real
tunnel, however, the different facets are clearly coupled and the problem is

Tunnels 1 39

complex, involving pore pressure changes, plasticity, lining deformations and

existing structures. Numerical procedures, such as the finite element technique,
lend themselves to the analysis of such complex problems. The finite element
method can:
.. simulate the construction sequence;
-

deal with complex ground conditions;

model realistic soil behaviour;
handle complex hydraulic conditions;
deal with ground treatment (e.g. compensation grouting);
account for adjacent services and structures;
simulate intermediate and long term conditions;
deal with multiple tunnels.

This establishes a strong case for using numerical methods. It also focuses the
mind on the type of problem for which one might consider the use of numerical
methods. This chapter deals with all aspects of the above list: the construction
sequence and its simulation (excavation and lining); the choice of an appropriate
soil model (stress-strain-strength and permeability); the modelling of hydraulic
boundaries; and examples of analyses of tunnelling beneath buildings and
alongside existing tunnels.

2.3
Tunnel construction
2.3.1 Introduction
Historically soft ground tunnels were excavated by hand, using spades for cutting
clayey soils and picks for breaking up granular soils and weak rocks. Openings
would be supported by timbering and subsequently lined in brick. The first use of
a tunnelling shield was for construction ofBruneI's Thames Tunnel between 18251843 (Skempton and Chrimes (1 994)). The men excavating the face were protected
from falls by a large shield within which they worked. The advent of pneumatic
spades and picks clearly enabled more rapid hand excavation. Hand excavation is
still in use today on short lengths of tunnel, or where access for mechanical means
is restricted.
Civil engineering construction starts the 21" Century with many alternative
mechanical methods for creating underground openings. The engineer can design
the most remarkable tunnel excavations, cathedral like in proportion and complex
in shape. The decision as to which method to adopt is based on knowledge of the
geology and the size and shape of the opening required. This section briefly
introduces three of the most common tunnel excavation techniques, highlighting
some of the key features which have an implication for numerical modelling.
Temporary support is today provided by a number of means - shield support,
compressed air, slurry machine, earth pressure balance machine (EPB), or
shotcrete. Permanent support is also provided by various means, including wedgeblock precast concrete and cast iron, bolted cast iron lining, reinforced shotcrete,

40 / Finite element analysis in geotechnical engineering: Application

or cast in-situ reinforced concrete. Section 2.4.5 of the chapter deals with
modelling the tunnel lining.

2.3.2 O p e n faced shield tunnelling

Tunnel shields are designed to protect
against falls ofground, see Figure 2.1.
They do not provide permanent
support, and a lining is usually
constructed within or just behind the
shield. The shield itself is jacked
forwards through reaction against the
completed lining. The -face is
Figure 2. 1: Schematic vieW of
unsupported, but can be partially
shielding during tunnel excavation
supported by face jacks, face plates,
or compressed air working. Compressed air can be used to control squeezing of
soft clays or flow of water and loss of fines in sandy soils. During down time the
face can be fully supported by timbering or shotcrete.
The method of excavation of the open face within the shield can vary. Hand
excavation has already been mentionecl. Mechanical alternatives are hydraulic
backhoe excavators, or roadheaders mounted within the shield.

2.3.3 T u n n e l Boring M a c h i n e s (TBM), including s l u r r y s h i e l d a n d

E a r t h P r e s s u r e B a l a n c e (IEPB) tunnelling
TBMs are closed face systems which
advance through the ground by
rotation of a cutting head, propelled
by reaction against the constructed
lining, see Figure 2.2 (in rock
reaction against lining
applications the TBM can propel
itself by reaction against the rock.)
Such tunnelling has traditionally been
restricted to single bore circular
section tunnels. Recent technological Figure 2.2: Schematic view of the
developments in Japan have led to the
TBM
creation of quite remarkable multiple
faced TBMs (e.g. Kuzuno et al. (1996)). Face support is provided through
controlling the applied thrust and the rate of removal of excavated material.
Tunnel headings may be additionally supported through application of slurry
or EPB machines. These TBMs pressurize the face. A slurry machine uses a
bentonite slurry pressurised within the cutting bulkhead. An EPB machine controls
the pressure of the excavated material itself against the face within the bulkhead,
so balancing the earth pressures in the ground.

Tunnels / 41

2.3.4 The sprayed concrete lining (SCL) method

The SCL method is a soft ground application of the New Austrian Tunnelling
Method (NATM), see Figure 2.3. As well as standard circular section tunnelling,
SCL can be used in competent ground to create large non-circular openings. The
method of excavation is usually by independent track or wheel mounted hydraulic
excavators. Support is provided as soon as possible by the application of sprayed
concrete (shotcrete). This is often reinforced by a steel mesh or a series of steel
hoops or arches installed before concreting. A permanent reinforced lining is
usually created at a later date, either by the application of further shotcrete, or insitu concreting. The current trend is, however, towards using a single shotcrete
lining, adequately reinforced.

Figure 2.3: Schematic view of SCL method

For large openings using SCL it is always the case that the tunnel is created by
the method of advanced headings. This can involve excavation of the crown first,
leaving a temporary invert, or the use of left and right side drifts, or a combination
depending on the ground quality and the size of opening. In all cases the advanced
heading is fully lined by shotcrete before the following drift commences.

2.3.5

Ground response t o tunnel construction

Excavation in the ground induces
stress relief which causes soil
movement towards the opening. If
there were no closure of the ground
direction of drive
into the opening, the volume of soil
excavated would exactly equal the
volume of tunnel created. Because
ground does however close into the
E
I D Iclel4
A cuftin beadand hood
opening, extra soil over and above the
B slueld
volume of final tunnel created must
D ungrouted lining
E grouted lining
be excavated. This excess soil is
termed the ground loss, and is often
quoted as a percentage of the Figure 2.4: Sources of volume loss
theoretical tunnel volume, the
(example of the TBM)

m{

42 / Finite element analysis in geotechnical engineering: Application

percentage volume loss. During the driving of a tunnel there are various distinct
causes of loss of ground into the excavation. Figure 2.4 shows, as an example, a
conventional shield tunnelling machine geometry and highlights the sources of
short-term volume loss discussed here:

Face loss: Loss first occurs into the face of the excavation, causing ground
settlements to appear at the surface ahead of the tunnel. These losses can be
reduced through the use of a closed face TBM, compressed air, slurry shield or
EPB machine.
Shieldloss: Soil will relax radially towards the shield after the cutting bead has
passed, filling the bead width void if the deformations are large enough. Any
over-excavation will contribute to increase this radial loss. The shield drives
itself forward by pushing off the last ring of lining segments to be erected. If
the soil deformations are large enough to fill the bead width void, then the tail
to the shield provides support to the cut perimeter of the soil. Another source
of shield loss occurs if the shield
does not move forward with its
axis coincident with the axis of
see Figure 2.5. In such
the
a situation the size of the
excavated hole is larger than the
final lining. This behaviour is
often referred to as pitching and
Figure 2.5: Schematic view of
occurs due to bad steering, or
pitching
when tunnelling on a curved
alignment.
Post shield loss: If the lining is erected behind, rather than within the shield,
then there will be a length of unsupported soil between the back of the tail to
the shield and the erected liner, over which radial deformation can occur. Once
the lining is in place, whether constructed within or behind the tail to the shield,
ground will continue to squeeze into the excavation if a void exists behind the
lining. It is common practice therefore to grout up any voids between liners and
the exposed soil surface. Expanded concrete liners are theoretically expanded
against the soil leaving no void.
These three aspects form the major contribution to the short term volume loss.
If excavation is taking place in a low permeability clay soil, then rapid construction
progress may result in a no volume change, undrained ground response. Thus the
volume loss at the tunnel will be exactly reflected at the surface in the settlement
trough. This statement will only hold rigorously true if there are no other
boundaries which can deform in response to the volume of soil moving into the
new excavation, such as open excavations. In a clay soil, after the tunnel has been
constructed and the lining installed, ground displacements are likely to be seen to
continue at the surface. This is due to the consolidation movements of the clay as
pore water pressures equilibrate to the new steady state regime, dictated by the

Tunnels 1 43

presence of the tunnel. The tunnel lining may also deform as the earth pressures act
on it, causing ground displacements.

2.4

Simulation of the construction process

2.4.1

Introduction
Tunnel excavation is a three
,
dimensional engineering process.
Whilst recognising that three
dimensional analysis is becoming
possible in the work place, it is still
two dimensional modelling that
dominates. This is because there are
practical limits on cost and computer
resource which, when performing
analyses sufficiently sophisticated to
handle all the complexities outlined in
Section 2.2, restrict us to two
dimensional modelling. If multiple
shallow tunnels are to be analysed, or
if the ground surface response is key
Figure 2.6: Plane strain geometry
to the analysis, then a plane strain
representation of the transverse
section is required (e.g. to study
effects on structures, Figure 2.6). If a
single deep tunnel is to be
investigated, and surface effects are f i a m ~ ~ O ~1
~
not of prime interest, then an axially boundaries
symmetric approximation may be
appropriate and heading advance can
be studied, all be it within a simplified Figure 2.7: Axi-symmetric geometry
stress regime (Figure 2.7).
More recently three dimensional analyses have been carried out using the
Fourier series aided finite element method (FSAFEM), applied to an axially
symmetric geometry to analyse the advance of a single tunnel heading (Shin
(2000)). FSAFEM is covered in detail in Chapter 12 of Volume 1. Such an
approach allows the material properties, initial stresses and boundary conditions
to vary with 8 (Figure 2.8), so introducing the ground surface whilst maintaining
the 2 dimensional simplification for the finite element mesh. This still demands
substantial computer time and restricts the analyst to considering one tunnel, but
is a significant advance for research into tunnel heading behaviour.
When modelling in plane strain at least one assumption must be made, i.e.
something must be prescribed, rather than predicted. There are various accepted
assumptions which are addressed in Section 2.4.4 of this chapter. The reader might

44 1 Finite element analysis in geotechnical engineering: Application

also find useful the excellent

summary by Clough and Leca (1989)
of work done, up to 1989, using the
finite element method as a means to
analyse soft ground tunnels.

2.4.2

Setting up the initial

conditions
Following the generation of a suitable
finite element mesh, before the effects
of tunnelling activities can be
analysed, the initial conditions in the
ground must be established. This can
be achieved by modelling the
complete geological history of the
site, if it is known. If it is not known,
then the initial conditions can be
achieved in two stages. First, the
160m
conditions appropriate to a greenfield
b) Fvvtc clement mesh (shaded area e gmmd at H)'
site are input into the analysis.
Commonly, the engineer directly
Figure 2.8: FSAFEM geometry
specifies the distributions of vertical
and horizontal effective stress in the ground prior to any construction activity (by
means of the material unit weight, the pore water pressure profile and the
coefficient of earth pressure at rest, K,,). Depending on the constitutive models to
be used, it might also be necessary to specify the initial void ratio and any
hardening parameters required to model the soil behaviour. The second stage
involves simulating any previous construction activities that have occurred at the
site. These could consist of demolition or construction of buildings, excavation of
deep basements, construction of services, or any other tunnels. This usually
involves several increments of analysis. For tunnels constructed in a greenfield site,
or where previous construction activities have caused minimal disturbance to the
in-situ soil conditions, stage two may be omitted.
It is noteworthy that if stage two is to be included, then the elements
representing the soil to be excavated for the tunnel must be included within the
mesh, and removed at the relevant construction stage. This is also true if staged
multiple tunnel analysis is to be modelled.
Having established the initial conditions, the analysis goes on to model tunnel
excavation, for which there are various accepted methods. A recommended way
of dealing with excavation during an analysis is to remove all the elements to be
excavated at the beginning of the relevant construction stage. In doing this the
mesh boundary is redefined excluding the excavated elements, so permitting the
specification of necessary boundary conditions (loadldisplacement and pore

Tunnels / 45

pressurelflow conditions). In addition, computer resources are saved because when

an element is excavated it makes no contribution to the global stiffness matrix.

2.4.3

Important boundary conditions

A number of boundary conditions are required to model tunnel construction. They

include the boundary displacement conditions, required to represent the far field
conditions or any symmetry of the problem; any surface traction; the excavation
of solid soil elements; the construction of structural shell elements; and the
hydraulic conditions at the far field boundaries, the soil strata interfaces (if an
interface is between consolidating and non-consolidating elements) and the tunnel
lining itself.
Loading and flow boundary conditions affect the right hand side of the global
equilibrium equations, whilst displacement and pore water pressure boundary
conditions affect the vectors of nodal displacements and pore pressures on the left
hand side of these equations, see Section 2.8 of Volume 1. Sufficient displacement
conditions must be prescribed in order to retain any rigid body modes, such as
rotations or translations of the complete mesh.
Analysis of tunnelling in drained granular materials will require careful
consideration of the hydraulic boundary condition, both during and after
excavation. In contrast, excavation in clay is usually rapid enough to be treated as
an undrained process, so the tunnel perimeter may remain impermeable until after
excavation is complete. Section 2.5.3 considers the hydraulic boundary condition
in more detail.

2.4.4 Modelling tunnel excavation

The 'gap' method

This method was introduced by Rowe et

al. (1983). A predefined void is
introduced into the finite element mesh
which represents the total ground loss
expected. In this way the out of plane and
in plane ground losses are incorporated
together with additional losses to allow
for miss-alignment of the shield, the
quality of workmanship, and the volume
change due to soil remoulding. It is clear
therefore how one can account for the
different tunnel construction methods ~;!f$;==&;~~~~"Z",,"''''~e
outlined above by varying the size of the
void. For example, if modelling an EPB
machine, the out of plane component of Figure 2.9: The gap method for
excavation
the total ground loss could be reduced.

46 / Finite element analysis in geotechnical engineering: Application

The void is placed around the final tunnel position and so locates the soil boundary
prior to excavation (Figure 2.9). This is achieved by resting the invert ofthe tunnel
on the underlying soil and prescribing the gap parameter at the crown. The gap
parameter is the vertical distance between the crown of the tunnel and the initial
position before tunnelling. The analysis proceeds by removing boundary tractions
at the perimeter of the opening and monitoring the resulting nodal displacements.
When the displacement of a node indicates that the void has been closed and the
soil is in contact with the predefined lining position, soilllining interaction is
activated at that node. The soil and the lining are actually treated as separate
bodies, related only by nodal forces (Rowe et al. (1978)).

The 'convergence-confinement' method

Another approach to modelling
excavation1 lining installation1
excavation is the 1 or convergenceinitial
unloading
unloading
confinement method (Panet and Guenot
{Ool
(1982)), in which the proportion of /.,""......
unloading before lining construction is \._ ..j
.. ...
""""
prescribed, so volume loss is a predicted
value. An internal force vector, { ~ , I,= { O J {vdl=hd ( 0 " ) I O ~ ' J = ( ~ - ~ { O , I
l
( I - l ) { F , , } ,is applied at the nodes on the
I
I
I
tunnel boundary ({F,,}being equivalent
:
to the initial soil stresses {G,,)). 1 is
initially equal to 0 and is then
progressively increased to 1 to model the
Figure 2.10: Convergenceexcavation process. At a prescribed value
confinement method
Ad the lining is installed, at which point
the stress reduction at the boundary is
Ad{o-,,}.The remainder of the stress reduction is applied to create the lining stress.
The stress reduction with the lining in place is then (l-&){a,,), see Figure 2.10.
'_

5
l
t

The 'progressive softening' method

A method termedprogressivesojiening was developed for the modelling ofNATM
(or sprayed concrete) tunnelling, Swoboda (1979). The soil within the heading is
softened by multiplying the soil stiffness by a reduction factorb. The effects of the
softening are evident when excavation forces are applied to the boundary of the
future tunnel. As with the convergence-confinement method, the lining is installed
before the modelled excavation is complete, see Figure 2.1 1. If the tunnel is
constructed with a bench and heading, then the above procedure can be applied to
each of them sequentially. The same methodology could be applied to an analysis
with side drifts.
The 'volume loss control' method
This method is similar to the convergence-confinement method, but instead of
prescribing the proportion of unloading prior to lining construction, the analyst

Tunnels I 4 7

prescribes the volume loss that will result on completion of excavation, see Figure
2.12. This method is therefore applicable to predictive analyses of excavation in
soil types for which the expected volume loss can be contidently (and
conservatively) determined for the given tunnelling method. It is also invaluable
for worthwhile back analysis of excavations for which measurements of volume
loss have been made.
initial

E, = E ,

sti&ess reduction

and unloading

E,'

excavation, lining
installation and unloading

= p E,

Figure 2.1 7 :Progressive softening method

With the A. method, the outward

support pressure on what is to be the
tunnel boundary is progressively
reduced. An alternative is for the
{F,,),thetoequivalent
calculate the
nodalvalue
forces
of
program

V, = volume o f settlement trough

vL= v,- v,

-jJ+([

V, = excavated tunnel volume

which represent the pressure exerted,
V, = final tunnel volume
on what is to be the tunnel boundary,
by the soil to be excavated. This is
linearly apportioned to the number of Fig,,re 2.
Volume loss method
increments, n, over which the
excavation is to take place, to give
Forces ~mposedon
fAF>= f F<,)/n.The equal and opposite
tunnel boundiuy due to
Forces imposed
force vector { - AF) is then applied at by soil to be excavated
rxcavabon
the excavation boundary for each of
the n increments of excavation
(Figure 2.13). The volume loss
induced by each increment of
boundary loading can be monitored
(given by displacement norrnal to the
tunnel perimeter, or by the surface
settlement profile in an undrained Figure 2. 13: Modelling
greenfield excavation) and the tunnel
of solid elements
lining constructed on the increment at
which the desired volume loss is achieved. After lining construction, the loading
boundary condition, { - A F ) , is still applied to the excavation perimeter for the
remainder of the n increments, thus introducing an initial stresses into the lining.
S

48 1 Finite element analysis in geotechnical engineering: Application

Depending on the stiffness of the lining further volume loss can occur during the
latter process. It may therefore be necessary to install the lining at an increment
which has a smaller volume loss than that desired, so that after full excavation the
desired volume loss is achieved.

Problems associated with soft clay

When modelling tunnelling in soft clay, problems may occur. When a lining is
constructed during excavation, the subsequent unloading within the lining results
in a desire for the complete tunnel to move upwards, with a corresponding
reduction in ground surface settlement. In a soft clay, this upward movement Inay
be so significant as to cancel out all the ground surface settlement occurring before
lining construction. Analysis of tunnelling in soft clay may therefore require the
application of support pressures within the tunnel during excavation. Tunnelling
in Singapore has been analysed at Imperial College (Ong (1996) and Ong (1 997)).
The very soft marine clay was modelled using a modified Cam clay model. The
incremental nodal forces, {-AF), were applied at the tunnel boundary, and
simultaneously an isotropic radial support force (akin to a compressed air working
pressure) was prescribed at each node in the opposite direction to the excavation
boundary force. The support force was adjusted to give the desired volume loss on
completion of excavation. The lining was not constructed until excavation was
complete. The sum of the incremental support forces accumulated during the
excavation gave the magnitude of the final support force. The removal of this final
support force required the application of an equal and opposite radial force at each
boundary node. This introduces initial stresses into the lining.
The application of an isotropic outward pressure on the tunnel boundary can be
used to represent working under compressed air conditions. A similar approach
was adopted recently in association with the R method (Bemat (1 996)). Bernat was
reproducing the effect of grout injection behind the tunnel lining.

2.4.5 Modelling t h e tunnel lining

In the previous section it was apparent that insertion of a lining is required to
control movements. It might be possible in a soil with sufficient undrained strength
to complete excavation of a plane strain tunnel with no lining support. The ground
movements predicted will however be significant, as plane strain analysis fails to
recognise the support from the lining already constructed behind the face, into
which the stresses arch.
Construction of an element can only usually take place if it existed within the
initial mesh. Elements which are to be constructed at some point during an analysis
(such as a tunnel lining) must therefore have been generated within the mesh at the
outset. Many software packages have a special facility permitting initial excavation
of these elements without the application of any loads at the commencement of any
analysis. Construction can be executed over a number of increments, or in a single
increment. With the volume loss control and convergence-confinement approaches

Tunnels / 49

to modelling tunnel excavation, the lining is constructed in a single specified

increment.

The use of solid elements

Using solid elements to represent a tunnel lining allows the analyst a very wide
range of constitutive models. A significant drawback is the need to maintain an
acceptable element shape (defined by the aspect ratio of length to width). A tunnel
lining is likely to be very thin relative to the tunnel diameter and boundary
distance. The consequence of maintaining an acceptable aspect ratio is the need for
a large number of elements.

The use of she4 elements

Using zero thickness curved shell elements to model a tunnel lining removes the
problem of aspect ratio control mentioned above. (Mindlin beam elements used in
plane strain or axial symmetry are effectively shell elements, and such elements are
covered in Chapter 3 of Volume 1.) Their use therefore allows more flexibility in
the mesh definition, but such elements do introduce numerical problems which can
be overcome by using selective reduced integration. It is additionally advantageous
that the solution for shell elements is of a structural nature, i.e. shear force, hoop
force, and bending moment. Special constitutive models can be coded for shell
elements to model cracking or crushing.
Modelling the connec rims between lining segrnents
Many tunnel linings are jointed. This
introduces particular modelling
requirements. Probably the simplest way
to model these tunnel linings is to use
shell elements and leave a gap at the
b) Solid elements
positions of the joints between lining a) Shell elements
modelling tunnel
modelling tunnel
segments, see Figure 2.14a. Tied freedom
lining
boundary conditions, see Section 3.7.4 of
Volume 1, can then be applied between
lhing
the two nodes representing the ends of
14:
segment
adjacent segments (i.e. nodes A and B on Figure
Connection
as
a
Figure 2.14a). The two nodes can be tied
joint
so that their displacements are the same,
but their rotations left untied. This will result in a joint that is free to rotate (i.e.
cannot sustain a moment), but will be able to transmit axial thrusts and shear
forces. Clearly, this is a simplification as most segmented linings will be able to
transmit some moments across the their joints.
A similar procedure could be applied if solid elements were used to represent
the tunnel segments, see Figure 2.14b. In this case the displacements of the midside nodes C and D would be tied.

>\

I 3

50 / Finite element analysis in geotechnical engineering: Application

.=?:
The drawback with this approach is
:*.,'
that problems can arise with the elements
representing the soil adjacent to the joint.
For example, consider the situation
shown in Figure 2.15 where shell L
elements have been used to represent the
lining segments. Although the lining is
discontinuous at the joint, the soil is not,
consequently at least one soil element
must span the joint region. In Figure 2.15
the mesh generation has resulted in two
soil elements which span the joint and the
nodes E, F and G are not supported by the
lining. If the soil is modelled with a finite Figure 2. 15: Problems associated
with the m~~ment-free
joint
strength (i.e. not linear elastic), then
during the analysis the two soil elements
adjacent to the joint are likely to fail due to the lack of radial support and the soil
will be squeezed through the gap between the lining elements. At the very least,
this is likely to give rise to numerical instability, but it could also represent a
legitimate failure mechanism, albeit an unrealistic one. This problem also arises if
solid elements are used to represent the lining.
To overcome this problem one
solution would be to tie the displacements
of nodes E, F and G to those of nodes A
and B. However, experience indicates
that such an approach is likely to result in
luung segment
numerical instability due to the very large
gradients of stress and strain that Such
Small shell element
constraints place in the soil elements
modelllogjotnt between
adjacent I W ~ segments
adjacent to the joint. Another solution,
but not one that is recommended, is to
make the two soil elements adjacent to
the joint linear elastic.
An alternative approach to modelling
the joints between tunnel segments is to
place a small shell element between those
representing the segments (Figure 2.16). Figure 2.16: Modelling segment
A special constitutive model which
connection using smaN shell
simulates rotation at the joints between
elements
tunnel segments can be used for these
elements. In this way more accurate bending moment distributions can be predicted
in situations of significant distortion. The model requires the specification of the
maximum compressive stress permissible across the joint, 9 ,and the position of
the neutral axis within the lining section (see Figure 2.17a). The bending moment,
,S.

$:ients

Tunnels 1 51

M, and hoop force, N, developed at each joint between two lining segments can be
represented by the normal force, N, acting at an eccentricity, e, such that eN=M
(Figure 2.17b). As the induced bending moment increases during lining
deformation, the normal force translates across the joint, see Figure 2.17~.The two
limiting moment values, M, and Mi , above which the joint will open on the
intrados or extrados, are set by the model based on the defined position of the
neutral axis and the maximum compressive stress:

If M, or Mi is exceeded at any joint,

cast iron liner
the joint will rotate and open, ( ~ segmental
h ~ discounts bolt
expandedconcretc lber
allowing the extra moments to
redistribute through the lining,
connectors :
maintaining the limiting value at the
a
joint. Two further conditions are that B
g
if N is tensile the joint will open, and
if N is such that the maximum
neu~ai
neutrafi
aXlS ;
axis j
compressive stress is violated across
the entirejoint, failure due to crushing
a) Examples of llning joints
will occur.
In some circumstances it may also
be necessary to apply restrictions to
the shear forces that can be
developed. For example, asimple, but
appropriate assumption, is that the
h) Bending moment and hoop force at a joint
maximum shear force, S,,,,,, is related
between two lining segments
to the normal force, N, via the angle
of friction, 6, between the lining
segments (i.e. S,, = N .tans).
To use this model in an analysis it
is necessary to construct a finite
element mesh in which the lining is
represented by a continuous series of
shell elements. The small elements at
c) Distribution of normal stress at a joint:
initial and final condition
the positions ofthe joints are assigned
the special constitutive model
described above, while those Figure 2.17: Modelling tunnel lining
joints
representing the tunnel segments are
assigned appropriate structural
properties.
"

52 / Finite element analysis in geotechnical engineering: Application

In a plane strain analysis the

idealisation shown in Figure 2.16
represents a tunnel with an unrolled
lining, as shown in Figure 2.18a. If in
reality the lining segments are rolled,
see Figure 2.18b, the above model
may not be appropriate.

Modelling sprayed concrete

linings
Figure 2.18: a) Unrolled and bl
The lining model presented in the
rolled tunnel lining
previous section can also be applied
to the modelling of sprayed concrete linings. In such linings there are no segments
and joints, but the entire lining uses this special model, so cracking is permitted to
develop at any location around the lining. If solid elements are being used to
represent a sprayed concrete lining, then a limiting tensile strength yield criterion
would be an appropriate plastic model to incorporate in conjunction with a
crushing model. A suitable double yield surface model is presented in Section 8.3
of Volume 1.
The behaviour of fresh sprayed concrete is also time dependent. Both its
strength and stiffness increase rapidly during the construction period. This
behaviour should be accounted for in any analysis. This problem is recently
discussed in depth by Shin (2000) and Shin and Potts (2001).

2.5

Modelling time dependent behaviour

2.5.1 Introduction
It is often important to model the consolidation behaviour of a soil during and after
excavation of a tunnel. To reliably study the time variant behaviour of the ground
and the tunnel, coupled consolidation analysis must be carried out. For coupled
consolidation analyses the hydraulic boundary conditions must clearly be correct
and representative of the field, as must the permeability of the continuum. A
crucial, and often ill considered, hydraulic boundary condition is that controlling
flow around the newly constructed tunnel. This section discusses such
considerations.

2.5.2

Setting up the initial conditions

It is useful when modelling time dependent behaviour to use software with the
flexibility to handle consolidating and non-consolidating elements in the same
mesh. A consolidating element will have both displacement and pore pressure
nodal degrees of freedom. A discourse on such elements and the choice of the
function to define pore pressure variation across an element can be found in
Section 10.4 of Volume l . It is clear that having both displacement and pore

Tunnels 1 53

pressure degrees of freedom at every node in amesh will significantly increase the
computer resource and time demand on that analysis. In a mixed situation the
engineer benefits therefore from being able to treat granular strata as free draining,
whilst only considering the consolidation behaviour of silty or clayey strata.
Any consolidating elements will require a model to define the permeability (see
Section 2.5.4). It is important to prescribe an initial pore water pressure distribution
which is in agreement with field measurements from the site. Reliable
measurements of permeability are less likely to have been made, but reasonable
judgement should be used when defining the permeability profile. The initial
stresses which prescribe the pore pressure profile at the outset of the analysis must
be consistent with the material permeability distribution. In the simple case of a
linear initial pore pressure profile, combined with a homogeneous permeability,
there is no problem (these will always be compatible). Modelling a nonlinear initial
pore pressure distribution, in combination with a more realistic inhomogeneous
permeability model, needs more care. Such a situation exists in London, and the
following example serves as a useful guide to approaching such situations, see also
Chapter 9.

Under-drained pore pressures in London

Historically water has been pumped from
Pore pressure (kPa)
a deep aquifer beneath London. Although
o
loo
200
now ceased, this action has induced
under-drained pore pressure profiles
through the low permeability London
-lo
Clay. Recent pore water pressure
measurements at Waterloo (Hight et al.
(1992)) clearly show the sub-hydrostatic
-20
profile plotted in Figure 2.19. Also shown $
is a data point of Skempton and Henkel g
field data
(1 957). Such a steady state profile can be
-30
achieved numerically with a model in 9
model
which permeability reduces with
increasing depth, or reduces with
-40
increasing mean effective stress (see
Section 2.5.4 below). The numerically
bottom of
-50
London Clay
predicted profile shown through the data
in Figure 2.19 was achieved using the
Figure 2.19: Under-drained pore
latter.
A useful approach to establishing
water pressure profile for
compatible pore water pressures and
Waterloo, London
permeability in such circumstances is a
unit width column of consolidating elements. Displacement boundary conditions
appropriate to a one dimensional analysis are specified and an initial hydrostatic
pore pressure profile prescribed. Figure220 shows the column in comparison with

-a
8

IT-J

54 / Finite element analysis in geotechnical engineering: Application

p ~ m b c p r rprolurs

,opbounduy
the actual boundary value
problem to be analysed. The
elements in the column are
given the stress-strain and
permeability constitutive
models that will be used in the
full analysis and the desired
steady state pore pressures at
the top and bottom are
prescribed as boundary
prrwnb; pm p-ur
conditions. Taking the analysis
Figure 2.20: Column method for
to the long term (through the
predicting
steady state pore water
use of sufficient time step
pressure
profile
increments) gives a prediction
of the steady state pore
pressure profile. This prediction can be compared with the data, and if the match
is not acceptable the parameters in the permeability model can be adjusted and the
analysis repeated. Using a unit width column for this exercise, rather than the full
mesh developed for the boundary value problem, is quicker. Once an acceptable
pore pressure profile has been obtained, this can be included in the full analysis in
the initial stresses, in the knowledge that it is compatible with the soil permeability.
The example analyses in Section 2.5.5 demonstrate the influence of an underdrained pore pressure profile on long tern ground movement predictions, through
comparison with a hydrostatic scenario.
,U

t o b m m ~

2.5.3 Hydraulic boundary conditions

Subsequently, during an analysis, representative flow or pore pressure conditions
must be stipulated at the borders of consolidating layers, be they mesh boundaries
or material interfaces. Many hydraulic boundary conditions are introduced in
Section 10.6 of Volume 1. Typically for tunnel analysis free draining and no-flow
boundaries are necessary: a boundary which is free draining is dictated by no
change from the initial pore water pressure condition; a mesh boundary which
represents a line of symmetry must be a no flow boundary, and the pore water
pressures must be permitted to change accordingly. The tunnel boundary itself
poses more of aproblem. From the outset oftunnel excavation the tunnel perimeter
is a mesh boundary. During rapid excavation in low permeability clay the tunnel
boundary can be considered to be a no flow boundary. During consolidation
however the tunnel lining can be treated as either free draining or impermeable.
This decision must be based on knowledge of whether the completed and fully
lined tunnel will act as a drain or not. The example analyses in Section 2.5.5 show
how significant this decision can be with respect to ground movement predictions.
.If it is decided that the tunnel will act as drain, the pore pressure condition to
be applied on the tunnel boundary is logically zero (or equal to whatever positive

Tunnels 1 55

outward pressure is being applied to represent compressed air working, etc.). Soil
adjacent to a tunnel excavation might develop tensile pore pressures (suctions)
during excavation. This will arise if the reduction in pore water pressure due to
unloading is greater than the initial pore water pressure around the tunnel. In this
case, if a zero (or positive) pore pressure boundary condition is prescribed, the soil
will draw water across the tunnel boundary. This is clearly unrepresentative of
reality, as a new tunnel is not a source of water. A more sophisticated boundary
condition is therefore required.
The application of a precipitation boundary condition to a tunnel lining is
presented in Section 10.6.6 of Volume 1. When using such a condition, the
program monitors the pore water pressures at the boundary nodes. If, at any time,
at any node, the soil pore water pressure is tensile with respect to the boundary
pressure, then a no flow condition is maintained at that node to prevent flow from
the tunnel into the soil; if however it is compressive, then the boundary pressure
is prescribed and free flow at that boundary node is permitted. In this way, in the
intermediate term, the swelling soil is forced to take water from surrounding soil
and not across the boundary, but the long term steady state is a flow regime
towards the new tunnel. As the condition is applied node by node it is possible at
some stage in an analysis to have flow across the tunnel boundary (into the tunnel)
in places, whilst still forcing a no flow condition in other places.

2.5.4 Permeability models

Coupled consolidation analyses require the material permeability, k, to be defined.
Most software packages permit simple linear isotropic permeability to be defined,
or anisotropic permeability, both having the option to vary k spatially
(inhomogeneity). Alternatively, nonlinear permeability can be defined. With a
nonlinear model the permeability varies during an analysis,the variation depending
on void ratio or mean effective stress. A number of alternatives are presented in
Section 10.7 of Volume l .
Pore water pressure (kPa)
As noted above, if initial pore
100
200
oo
//A\\
'
v
'
pressure profiles are nonlinear then an
taw
inhomogeneous model must be
Thames Gravel
10adopted. It is also the case that
different permeability models will 3
lead to different long term pore 2
20pressure conditions after completion
oftunnelling. Figure 2.2 1 exemplifies
this through presentation of the
3o
numrically generated long term pore
4 . 1 4 6diameter
~
tunnel
pressure profiles down the centre line
with permeable l i g
above a 34m deep, 4m diameter
tunnel in London Clay, which acts as Figure 2.2 Long term pore
a drain. The initial pore water

..

. ..

, - m

pressure profiles

56 / Finite element analysis in geotechnical engineering: Application

pressure profile was hydrostatic. Two permeability models were used and the
analyses taken to the long term. The first model was a linear, isotropic,
inhomogeneous permeability (reducing with depth). The second model was a
nonlinear model of the form presented below in Equation (2.2), in which
permeability varies with mean effective stress and is continually updated during
the analysis. The hydraulic boundaries in the long term are hydrostatic pressure at
the top of the London Clay (defined by the water table in the gravel) and zero pore
pressure at the permeable tunnel boundary. Figure 2.21 shows that the nonlinear
permeability model predicts a near hydrostatic profile down to within 10 m of the
crown, below which a rapid fall off of pore pressure is evident. The linear
permeability model however predicts a pore pressure profile significantly below
hydrostatic. If a back analysis is to be carried out, and piezometer readings have
shown the pore pressures to be near hydrostatic above the tunnel in spite ofthe fact
that the tunnel is acting as a drain, then clearly the nonlinear model is more
appropriate.
If the decision is made to use a nonlinear model for a tunnel analysis, then it is
noteworthy that the necessary data on which to base the parameters for such a
model are unlikely to have been obtained during a standard site investigation.
Hence the need to have reliable pore water pressure measurements which can be
matched, as outlined in Section 2.5.2. It is also noteworthy that a model in which
k depends on mean effective stress implicitly assumes something about the volume
change behaviour of the soil. Observations of seepage pressures in fills and in-situ
soils have shown that measured pore water pressures are generally very different
from those predicted using conventional theory, hence the development of more
sophisticated models for analysis (Vaughan (1989)). It is known that permeability
varies with void ratio, often by several orders of magnitude over the range of
possible void ratios for a given soil. Void ratio itself varies with effective stress,
and effective stress is dependent on pore pressures, so the problem is nonlinear. To
model London Clay a logarithmic law, relating permeability to mean effective
stress, has been adopted:

where k,, is the permeability at zero effective stress, a is a constant incorporating

the initial void ratio at zero effective stress and the coefficient of volume
compressibility, m,, and p' is the mean effective stress. The derivation of this
logarithmic law therefore requires the assumption that m,, is constant. This is a
useful assumption for many soil types, including London Clay (Vaughan (1989)).
An alternative model assumes that the compression index, C,., remains constant,
as for one dimensional normal consolidation of wet clays, and leads to a power law
relating permeability and mean effective stress. If soil yields during an analysis,
there is clearly a question over the validity of the constant m,, or C,. assumptions.
Additionally, if the plasticity model adopted allows for dilation on yielding, the
mean effective stress can increase dramatically, so reducing the permeability of the
yielding soil to unreasonably low values.

Tunnels / 57

2.5.5 A parametric study of the effect of permeable and

impermeable tunnel linings
The analyses presented
Porrwatrr pwurs(wa)
pennwbility
here, carried out by the
O"
Geotechnical Consulting
Group
(1993),
,
considered many factors
which influence the
,
intermediate and long d
term behaviour of
,,
tunnelled ground. All the
a
analyses employed the
coupled consolidation
solution technique to
Figure 2.22: Soil profile and pore water
analyse the time
pressure profiles for GC6 parametric study
dependent consolidation
settlement above railway tunnels excavated in London Clay. The study considered
the three different tunnels in the profile shown in Figure 2.22. Only single tunnels
were modelled in any one analysis. Two 6.5m diameter running tunnels (R) were
analysed for axis depths below ground level of z = 15m and 25m. One 121n
diameter station tunnel (S) was analysed at z = 25m. The water table was
maintained at the top of the London Clay ( z = 5m), but two alternative initial pore
water pressure profiles were considered, under-drained (U) and hydrostatic (H),
which are also indicated in Figure 2.22. The initial pore water pressure profile was
in equilibrium with the spatially varying linear anisotropic permeability profile
specified. The final far field equilibrium pore water pressure profile was the same
as that prescribed initially, i.e. U or H. The pore water pressures in the Thames
Gravel (a non-consolidating material) was maintained as zero. On the remote
vertical boundaries, at some distance from the centre line, and at the interface
between the Lambeth Group Clay and Sand, it was assumed that a source of water
maintained the pore water pressure at its initial value throughout the analysis. On
any boundary representing a line of symmetry, a no flow boundary condition was
prescribed. The near field long term pore water pressure profile was governed by
the drainage boundary condition prescribed at the excavated tunnel perimeter
during consolidation. The two alternatives investigated were: a fully permeable
lining (P) using the special control boundary condition introduced in Section 2.5.3;
or a completely impermeable lining (I) modelled as a no flow boundary.
The volume loss control method was used with a target value of between 1.7
and 2.0% (see Section 2.4.4). The soil was represented by combining the pre-yield
small strain stiffness model (see Section 5.7.5 of Volume 1) with a Mohr-Coulomb
plasticity model (see Sec.tion 7.5 of Volume 1). The permeability data was based
on field measurements reported by Burland and Hancock (1977) in the
Westminster area of London. The Thames Gravel and Lambeth Group Sand were
modelled as fully drained throughout the analyses. The profiles of permeability

58 / Finite element analysis in geotechnieal engineering: Application

Time (years)
(reducing with depth) in the clay are
0.2
1.0
5.0
20.0
100.0
depicted on Figure 2.22.
Figure 2.23 shows the influence of
tunnel diameter (R or S) on the time 3
dependent surface settlements above g
a tunnel at a fixed depth (z = 25m)
with a permeable lining. Both H and
U analyses are presented. The
diameter has little effect on the .$
hydrostatic analyses, with RH and SH E:
- - _ _ - _ SH
_-__
showing trends o f increasing
loo
maximum settlement, S,,,, over a 5 to
Figure 2.23: Effect of tunnel
10 year period in the order of 30 to
35mm. This is clearly a greater
diameter
percentage of the short term
settlement for RH than for SH. The
Tune (years)
trend is reversed with the underoo
0.2
1.0
5.0
20.0
100.0
drained profile, as S,, is seen to
I
I
I
I
I
25U
reduce over a 5 to 10 year period. The 'fJ reduction for RU is only 3mm, whilst g i = - - -W
15U
that for SU is 20mm.
5H
- * - - I = - - --- ---- - 1
Figure 2.24 isolates the influence
25H
H = hydrostatic
of depth (15 or 25m) on the time
dependent surface settlements above
a running tunnel with a permeable
lining. During consolidation, S,,,,, Figure 2.24: Effect of tunnel depth
increases by 18mm for 15H and
30mm for 25H. Similarly there is an
Time (years)
increase of 8mm for 15U. In contrast
S,,, reduces by 3mm for 25U. This
opposite trend between 25U and 15U 3
is a result of the pore water pressure 3
t:
values at those depths relative to e
hydrostatic. From Figure 2.22 it can
be noted that a tunnel located at 15m
is at a depth where the under-drained .-5
pore water pressures are near
'0
hydrostatic. The surface response in
the intermediate period is therefore
Figure 2.25: Effect of lining
the same as for the hydrostatic
analyses. All movements essentially
permeability
cease within 10 years.
Figure 2.25 presents the influence ofthe lining permeability (P or I) on the time
dependent surface settlements above a large diameter station tunnel at a depth of

----

'-

3
3

-_--__--:.

Tunnels / 59

25m. There is an increase in S,,,,, of 35mm for PH, but a decrease of 20mm for PU.
These movements are complete within 5 to 10 years. With an impermeable lining
the surface heaves by 45mm for IU, and by 40mm for IH. The drainage path
lengths in the consolidating clay are longer if the lining is impermeable, and this
is reflected in the increase in time taken for the movements to cease from within
10 years to 20 years.
During the consolidation period after construction (the intermediateperiod) the
predicted ground surface response depends on many factors: the equilibrium pore
pressure profile and the tunnel depth within this profile, the tunnel lining drainage
condition, and the tunnel diameter.

2.6

Choice of soil model

2.6.1

Introduction
The task of choosing an appropriate soil model for the analysis of tunnelling is a
specific, not a general one. Volume 1 deals with many ofthe available constitutive
models, and the particularities of them. This section aims, through example, to
show that the choice of soil model can be of great importance in tunnel modelling.
Some studies recently completed at Imperial College tested a number of pre-yield
constitutive models against each other through the comparison with field
measurements of surface and subsurface ground movements induced by tunnelling
(Addenbrooke (1996), Addenbrooke et al. (1997)). Results from these studies are
presented in Section2.6.2. When modelling tunnelling in an overconsolidatedclay,
adopting a valid soil model is not necessarily going to give good predictions of
ground movements when plane strain geometry is used. Section 2.6.3 discusses a
couple of devices for improving predictions of ground movement.

2.6.2 Results from a parametric study

This section presents the results from analyses oftunnelling 34m beneath Regent's
Park, 29m beneath Green Park and 30m beneath St. James's Park, in London Clay
overlain by Thames Gravel and made ground. All three sites are on the Jubilee
Line, the first two from stage I (constructed in the 1970s) and the last one from
stage I1 (Jubilee Line Extension opened in 1999). Being in parks, the ground
movements at these sites were not influenced by surface structures, and in each
case the first tunnel to be
excavated was analysed so
4
there could be no question of
sub-surface interaction effects
B
(covered in Section 2.7.3).
4
Each site was analysed in
M
M
M
M
plane strain, the soil was
elastic perfectly plastic, and Figure 2.26: Finite element mesh for St.
the excavation was modelled
James's Park analysis

60 / Finite element analysis in geotechnical engineering: Application

as undrained. The mesh for St. James's Park is presented in Figure 2.26 as an
example (only the deeper tunnel is considered here). Volume loss controlled
excavation was used, with a target of between 1.3% and 1.5% for Regents Park and
Green Park and 3.3% for St. James's Park (based on field measurement). The
initial stresses prescribed a pore pressure profile considered to be representative at
each site, and employed an initial stress ratio, K,, of 1.5 in the London Clay. In
each case linear isotropic, linear anisotropic, and nonlinear elastic constitutive
models were alternatively adopted for the pre-yield behaviour. All three cases had
stiffness increasing with depth. Plastic behaviour was modelled using the nonassociated Mohr-Coulomb model.
600Linear isotropic, linear anisotropic
and nonlinear elastic constitutive G,e,
- - . CAU compression test
models are introduced in Sections 5.5, -7
CAU extension test
5.6 and 5.7 of Volume 1 respectively.
The use of a nonlinear pre-yield
model recognises the nonlinear nature
of real soil behaviour in the elastic
region. One of the models which
reproduces nonlinear behaviour at
small strains is outlined in Section
I
I
0
5.7.5 of Volume l. Figure 2.27 shows
0 001
o 01
0.1
10
Shear straln (%)
the secant shear modulus (normalised
by mean effective stress p') decay
with shear strain, predicted by the Figure 2.27: Small strain Stiffness
curves from triaxial tests
model with appropriate parameters
for London Clay. This is compared
with laboratory data for undrained triaxial compression and extension (samples
consolidated anisotropically).
The surface settlement profiles presented in Figures 2.28a to c reveal a number
of interesting points. First, it is noteworthy how much more precise the settlement
monitoring was in the 1990s (Figure 2 . 2 8 ~ )compared with the 1970s (Figures
2.28a and b). A full description ofthe state-of-the-art instrumentation used and the
data obtained on the more recent project is presented in Nyren (1 998). Secondly,
linear elastic models, both isotropic and anisotropic, are wholly ineffective. In all
three cases the settlement profiles are too shallow and too wide when compared
with the field data, and the isotropic analyses predict profiles of the wrong shape.
Thirdly, the introduction of a more representative nonlinear elastic model
significantly improves the predictions, see also Addenbrooke et al. (1997). The
same improvement in prediction with the nonlinear elastic model can be
demonstrated for sub-surface ground movements.

,,,

2.6.3 Devices for improving the surface settlement prediction

The analyses in Section 2.6.2 were plane strain and the initial stresses prescribed
prior to tunnel excavation were the at rest stresses (based on K,, > 1). Axially

Tunnels / 61
Distance from centre line (m)
symmetric and three dimensional
analyses have revealed that, as a
tunnel heading approaches and
passes a plane in space, the
effective stress ratio reduces to the
4
side and increases above and
below the tunnel. The introduction
a) Regent's Park, 34m deep tunnel
(data from Barratt and 51er (1976))
of a stress scenario such as this in
advance of the plane strain
excavation may be considered a
Distance from centre line (m)
logical assumption to represent
stress changes ahead of the tunnel
face. In this section the results of
PRE YIELD MODEL
analyses with a local zone of
--..-..-..- - - - anisotropic
isotropic linear elastic
linear elastic
reduced K,, (equal to 0.5) within
4
the London Clay with a much
$v1
b) Green Park, 29m deep tunnel
higher global K, of 1.5 are
(data fiom Attewell and Fanner (1974))
presented. This zone extended
vertically between the crown and
Distance from centre line (m)
the invert of the tunnel, and
horizontally a distance, a, of
approximately three timer the
radius ofexcavation (Figure 2.29). ,
The Jubilee Line tunnels were reanalysed with the reduced K, 3
approach. Figures 2.30a to C show
C)St. James's park, 30m deep tunnel
(data 6omNyren (1998))
the improvement in the predicted 20 *
surface settlement profiles.
Another device for obtaining Figure 2.28: Effect of a pre-yield model
improved surface settlement
on predicted settlement profile
profiles is to employ an
anisotropic nonlinear pre-yield
soil model. If the soil is cross
anisotropic then five parameters
must be defined: the vertical
Young's modulus, E,', the
horizontal Young's modulus, E,',
the Poisson's ratio for the
influence of increments of vertical
effective stress on horizontal
strain, p,,,',the Poisson's ratio for
the influence of increments of
horizontal effective stress on the
horizontal strain in the orthogonal Figure 2.29: Local zone of reduced K,

l!
g

62 / Finite element analysis in geotechnical engineering: Application

Distance 6om centre line (m)
horizontal direction, ,U,', and the
shear modulus in the vertical
plane, G,,,, (the independent shear
modulus). In a linear anisotropic
model these parameters are
defined directly. The Authors
have adapted the nonlinear model
from section 5.7.5 of Volume 1 to
a) Regent's Park,34m deep tunnel
investigate the advantages given
by combining nonlinearity and
anisotropy. The model defines the
Distance 6om centre line (m)
variation of shear and bulk moduli
G and K, from which the modulus
E' and a Poisson's ratio p ' can be
- - - - - nonlinear isotropic
high K,
calculated and so the isotropic
reduced K,
elastic stiffness matrix is defined.
In the adapted model the
anisotropic elastic stiffness matrix
is obtained from constant, user
defined ratios of pyh'lpl, phhfIp',
E,,'IE, n' (=E,,'IEAf) and m'
Distance from centre line (m)
(=G,,,)E,.'). The most influential of
centre line
25
SO
these ratios was found to be m'.
There are two analyses presented
here, for both of which n'=0.625.
For the first analysis mt=0.444
(from Burland and Kalra (1986)).
C)St. James's Park,30m deep tunnel
Based on laboratory and field data
for London Clay, the initial G,,,,,,
has been determined by Simpson
al. (1996) to be 0.65G,,, (where Figure 2.30: Effect of Zone Of reduced
K,, on predicted settlement profiles
jn a
G,,, is the shear
horizontal plane, dependent on
E,'). With p1=0.2 and ni=0.625,
Distance fiom tunnel cenhe line (m)
this gives m1=0.433. This is very
similar to the value adopted here,
m1=0.444. It is not known how C,,
decays with strain, so it is an 3
PREYIELD MODEL:
assumption that it varies along the 3 l0
- - - . noolinear anisotropic
m' = 0.44
same function as E'. For the
-nonlinear anisotropic
second analysis a very soft
field data
independent shear modulus was
defined by m1=0.2. The initial

stresses

maintain the

Figure 2.37: Soft shear modulus effect

Tunnels / 6 3

condition in the London Clay. Figure 2.3 1 shows the predicted surface settlement
above the St. James's Park tunnel in comparison with the field data. The
introduction of an independent shear modulus based on field and laboratory data
gives little improvement over the isotropic model. A desirable improvement is
however achieved through the use of a very soft independent shear modulus.
The reason for using the term 'devices' for these adjustments which improve
the surface settlement profiles predicted by the finite element analysis, is that they
cannot at present be supported as representative of field conditions. Using a
reduced K,, on the premise that there have been stress changes ahead of the tunnel
is inconsistent with adopting a nonlinear small strain stiffness which assumes that
no straining has taken place ahead of the tunnel. The effect of reducing K,, to the
sides of the tunnel is to cause plastic zones to develop in these areas early on in the
incremental excavation procedure. This is in contrast to the high K,, analyses in
which plastic zones develop above and below the tunnel.
The effect of softening the anisotropic shear modulus in a nonlinear model is
to modify the pattern of ground movement close to the tunnel. It also increases the
horizontal component ofdisplacement across the whole finite element mesh. There
is no laboratory or field evidence for such a soft initial shear modulus G,,,,,,for
London Clay. It may be that softening the soil in this way represents softening of
the soil due to straining ahead of the tunnel face. A major drawback with a
nonlinear anisotropic model is that there is no information on how G,,,,varies with
strain level.
With both devices the percentage unloading to achieve the controlled volume
loss is reduced from the analyses in Section 2.6.2. This has an effect on the
subsequent soil and lining behaviour. If a lining is constructed early in an analysis,
then the growth of any zones of developing soil plasticity will be arrested, which
in a dilatant soil will reduce the magnitude of any pore water pressure reductions
around the tunnel. Also, the sooner the lining is constructed the greater the loads
transferred into the lining by the end of excavation.

Interaction analyses
2.7.1 The influence of building stiffness on tunnel-induced
ground movements

2.7

As mentioned in Section 2.2 above, empirical methods for the determination of

tunnelling-induced settlements are based on measurements at greenfield sites. This
is considered inappropriate to the urban situation where tunnels pass beneath
existing buildings. Potts and Addenbrooke (1997) applied the finite element
method to the analysis ofthis soil/structure interaction problem. They presented the
results ofover 100 finite element analyses oftunnel excavation in stiff clay beneath
an existing building. Figure 2.6 shows the geometry represented by their analyses.
D, z, B and eccentricity e were all varied, as were the bending stiffness ( E l ) and
axial stiffness (EA) of the buildings. The soil was again modelled as nonlinear
elastic perfectly plastic. The buildings were represented by shell elements (Mindlin

64 / Finite element analysis in geotechnical engineering: Application

beam elements in plane strain). The results were synthesised into simple building
distortion assessment charts. These new charts allow the engineer to take account
of building stiffness when assessing the likely distortion and damage expected
from tunneliing-induced ground movements in an urban environment. Two new
soil/structure relative stiffness parameters were defined. The relative bending
stiffness, p* (with units m-'), and relative axial stiffness, a*:
P* =-

EI
E,, 'v4

a" =- E A
E,v H

where E, is a representative soil stiffness, being the secant stiffness at 0.1% axial
strain from a triaxial compression test of a sample taken from a depth 212, and H
is the building half width, Bl2.
Each analysis gave a building settlement profile and building horizontal
displacement profile. Examples for 60m wide buildings with zero eccentricity
above a 20m deep tunnel are shown in Figures 2.32 and 2.33 respectively. These
were interpreted in terms of building distortion parameters: a sagging deflection
ratio (DR,,,) combined with a compressive horizontal strain (E,,) , and a hogging
deflection ratio (DR,,,) combined with a tensile horizontal strain (c,,) for the
building. These parameters are often related to limiting tensile strain in a building
in order to predict the expected damage due to such a combination of distortions
(Burland and Wroth (1 974)).
centre line

E12 = 30m

centre line

Distance from centrc line (m)

~ i g u r e2.32: The effect of p' on

surface settlement profile

Distance from centre line (m)

' on
Figure 2.33: The effect of a
horizontal surface displacement

Tunnels / 65

In association with these analyses the greenfield situations were analysed for
comparison. Greenfield settlement and horizontal displacement are commonly used
for damage assessment, and it is relatively cheap and simple to acquire such
predictions empirically. For each building size and location (B and e respectively),
the greenfield ground surface movements were used to obtain greenfield values for
deflection ratio and horizontal strain. These represent the distortion parameters for
a completely flexible building. Potts and Addenbrooke (1997) therefore
investigated the degree of modification that buildings with different relative
stiffness (p' and a*) and relative position (elB) make to the greenfield values of
distortion. Modification factors, M, were defined which gave a quantitative
measure of the degree of modification for a given building. The sagging deflection
ratio for an analysis with a building was divided by the equivalent distortion
parameter from the greenfield analyses, giving the modification factor MDR,,,.
Similarly, the horizontal compressive strain for an analysis with a building was
divided by the equivalent distortion parameter from the greenfield analysis, giving
modification factor M,,. For those buildings which lay in part or in total in the
hogging region the modification factors for hogging deflection ratio, MDR,,, ,and
tensile horizontal strain, M,,, vvere obtained in the same way.
It was determined that in the likely range of true building stiffness the relative
bending stiffness controlled the degree of modification to the deflection ratio (i.e.
vertical settlement profile); and the relative axial stiffness controlled the degree of
modification to the horizontal strain (i.e. the horizontal displacement). Potts and
Addenbrooke (1 997) therefore plotted the modification factors for deflection ratio
in both sagging and hogging forms, MDR,,, and MDR,,,, againstp*for each elB. The
modification factors for horizontal strain (E,) in compression and tension, M",,, and
M,,, were plotted against a' for each elB. Empirical design curves were fitted
through the data and these are reproduced in Figure 2.34.

l
a) Deslgn curves to deterrmne deflect~onrat10

06

b) Design curves to deterrmne horizontal smln

Figure 2.34: Design curves to determine a/ deflection

ratio and 6) horizontal strain

66 / Finite element analysis in geotechnical engineering: Application

The new modified empirical damage assessment procedure for a given building
therefore requires:

- The relative bending and axial stiffness for the building in question. As a first

estimate the engineer could consider the contribution to stiffness of the

foundation alone, before considering the independent or coupled contributions
to bending and axial stiffness of slabs, beams, columns and load-bearing walls.
The modification factors read off the design chart in Figure 2.34. These clearly
depend on the relative stiffness and the eccentricity ratio.
The greenfield distortion parameters for the ground at foundation level,
assuming no building were present.
The multiplication of the greenfield parameters by the relevant modification
factors.

These modified parameters give the likely distortion to the building, taking account
of the building stiffness.

2.7.2 The Treasury building

- a case study

The parametric study presented in Section 2.7.1 dealt with arbitrary buildings
above typical tunnels. This section presents the results from the finite element
analysis of one particularly sensitive building affected by construction of the
Jubilee Line Extension in London - the Treasury building at Westminster. The
building had been monitored during the project. Results are given here from those
analyses that were performed to model the tunnelling-induced response of the
building. To mitigate the effects of tunnel construction an extensive programme
of compensation grouting was implemented at the site and this too has been
analysed. Only a summarised description of the analyses and the results are
included here, further details are given in Standing- et al. (1998).
The Treasury is a massive stoneclad brick-masonry structure,
approximately 2 10 m long and l00 m
wide with four storeys above ground
and two basement levels. The
foundations consist of strips and pads
connected by an unreinforced
concrete slab founded in the Terrace
Gravels which overlie London Clay.
The top of the foundation is
approximately 6 m below ground
level. As Pa* of the Jubilee Line Figure 2.35: Plan of Treasury site
Extension project, two running
tunnels were excavated under one corner of the building (shown in plan on Figure
2.35). The westbound tunnel was the first to be excavated. Following this there was
a rest period before compensation grouting and excavation ofthe eastbound tunnel.
Compensation grouting was implemented after driving the westbound tunnel and

Tunnels 1 67

during construction of the eastbound

tunnel. The level of the tubes-&
Upper London Clay
manchette (TAh4s) used for this work
Middle London Clay
was roughly 16m below ground level,
Lower London Clay
see Figure 2.36, extending beneath the
basement between the tunnels and the
]Cm
foundation slab. The analysis followed
the construction activities and
associated durations recorded during
the works. Both tunnels are 4.95m in
Figure 2.36: Cross-section AA
diameter, the westbound is 34m below
and the eastbound 24m below ground
level, both in the London Clay.
Concrete segments form the linings
and these were placed and expanded
immediately behind the shield.
Plane strain conditions were
assumed and the cross-section AA
shown on Figures 2.35 and 2.36 was
analysed. A section of the finite
element mesh is shown in Figure 2.37.
For clarity only the detailed part of the
mesh in the vicinity of the tunnels is
shown. The soil stratigraphy was
obtained from the site investigation
Figure 2.37: Section of finite
and different strata were modelled as
element mesh
nonlinear elastic perfectly plastic. The
small strain stiffness model described in Section 5.7.5 of Volume 1 was used to
represent pre-yield behaviour and a Mohr-Coulombyield and plastic potential were
used to model the plastic behaviour. The permeability was modelled as linear
anisotropic inhomogeneous in order to reproduce the under-drained pore pressure
profile at the site. Table 2.1 summariesthe coefficients for the small strain stiffness
model.
An equivalent elastic raft of 3m thickness was used to model the Treasury
foundation (Young's modulus E = 28x106 kPa, Poisson's ratio p = 0.15). The
tunnel linings were modelled with elastic shell elements (Mindlin beam elements
in plane strain), with Young's modulus E = 15x106kPa, Poisson's ratio, p = 0.15,
-~
thickness of 0.2m2/mand second moment of area of 6 . 6 7 ~ 1 0m4/m.
Compensation grouting was modelled by applying a pressure within a
horizontal slot in the finite element mesh, at the location ofthe TAM's, see Figures
2.37 and 2.38. This slot was created by including interface elements in the mesh
generation, but excavating them prior to the beginning of the analysis. When
compensation grouting was not taking place during the analysis, the displacements
(both vertical and horizontal) of adjacent nodes on opposite sides of the slit were

68 / Finite element analysis in geotechnical engineering: Application

tied, see Figure 2.38a and Section

3.7.4 of Volume 1. During an
increment of the analysis in which
compensation grouting was being
modelled, these ties were removed and
a pressure was applied to the boundary
a) Displacements of opposite nodes are tied when
of the slot, see Figure 2.38b. For all
compensationgroutkg is not active
increments the pore pressure degrees
Interface elements
of freedom of adjacent nodes on
opposite sides of the slit were tied.
The volume loss control method
was used for excavation, with the
target volume loss based on field
measurements from a green field
monitoring station at St. James's park. b) Application of grout pressure during compensation groutmg
The results presented here focus on
movements after construction of the
Figure 2.38: Modelling
westbound tunnel and movements 18
cornpensation grouting
weeks after the compensation grouting
works and the construction of the
, & , &,,b
x eoordmts (m)
eastbound tunnel. Figure 2.39 shows
the settlement troughs after excavation
of the westbound tunnel (i.e. the first
tunnel to be constructed). Field
measurements at ground surface from - FE
prcd#sflan
Msasursd dau (T~a6w-j)
St. James's Park (greenfield) and at
Ssnlcmmt (mm)
foundation level for the Treasury are
presented.
the
Figure 2.39: Settlement trough
observations from these two sites
after WB tunnel
indicates the influence of the building.
It reduces the maximum settlement,
,PTreajurybuilding*.*" J
but increases the width of the L
coordsnatc (m)
settlement trough. The maximum
settlement beneath the Treasury is
beneath its corner, slightly offset from
the tunnel centre-line. Also shown is
the finite element prediction. The
20 Settlement (m)
agreement between the prediction and
the field observations is excellent.
Figure 2.40 presents settlement Figure 2-40;
Sett1Wnent trough 18
troughs at foundation level for the weeks after E5 tunnel construction

Treasury, 18 weeks after construction

of the eastbound tunnel. Both observations and finite element predictions are
presented. The agreement between prediction and observation is again excellent.

Tunnels / 69

Table 2. 1: Coefficients for elastic shear modulus

Table 2.1 /cont.): Coefficients for elastic bulk modulus

70 / Finite element analysis in geotechnical engineering: Application

Table 2.1 (contl: Other properties

2.7.3 Twin tunnel interaction

Modern metro construction in congested urban environments often involves the
excavation of new tunnels in close proximity to each other. In addition,
improvement to existing networks requires new construction adjacent to existing
tunnels. The finite element method is an appropriate analysis tool for considering
more than one tunnel excavation in a given project. At the design stage it is
important to address the interaction of the tunnels and their effects on overall
ground response. Current design practice, which is based on zero interaction
assumptions, rules of thumb, and linear analyses gives little guidance in this
respect.
Addenbrooke and Potts (200 1) have used the plane strain finite element method
to produce interaction diagrams that establish the likely degree of interaction
between two tunnels constructed in stiff clay. They have assessed the surface
settlement effects and the subsurface lining interactions under various conditions.
At each stage any available field data has been used to validate the finite element
predictions. They used a coupled consolidation formulation and a nonlinear elastic
perfectly plastic soil model. The nonlinear pre-yield model was that of Puzrin and
Burland (1998). As published the model was restricted to conventional triaxial
stress space. Addenbrooke et al. (1997) used an extended version of the model
valid for use in general stress space. The extended formulation is presented in
Section 5.7.6 of Volume 1. As well as reproducing the expected decay in stiffness
with strain level, the model reproduces the kinematic nature of soil behaviour by

Tunnels 1 71

responding to changes in stress path

direction. The non-associated MohrCoulomb model was used to describe
post-yield behaviour. The tunnels
were lined with elastic shell elements
(Mindlin beam elements in plane
strain). There were two sets of
analyses, the first modelled two
tunnels running parallel side-by-side,
and the second one above the other
(piggy-back), see Figure 2.41. The
spacing between the tunnels was
varied so as to ascertain the zones of
2.41: Side b y
influence, and the time between the ~i~~~~
and p;ggy
construction of the two tunnels was
back geometry
varied to establish the influence of
prolonged consolidation or swelling.
Consider first two tunnels
constructed side-by-side, the first
Ward (1969)
+ honzootslbamner o ven~caldiameter
completed 3 weeks before the second. $9
Klmmancc er nl (1 996)
Figure 2.42 shows the influence ofthe
0.1
excavation of a second tunnel on the
lining to the first. The existing lining
is pulled toward the new excavation
and a squatting deformation is
-0.2
sholteolng
induced. The figure shows the
,,
I
lengthening ofthe horizontal diameter
0
2
4
6
8
1
0
P~llarwrdtb (number of &ameters)
and shortening of the vertical
diameter (as a percentage of the
Figure 2.42: Deformation of Is'
design diameter) and how these
reduce with increasing tunnel in response to zndtunnel for

1;
8; -,,

side by side geometry

pillar width (extrados to extrados
tunnel spacing divided by diameter). Field data is included for comparison: the
average deformation from three sets of closely spaced twin tunnels on the Victoria
Line in London (Ward (1969)); and recent data from three sites of the Jubilee Line
Extension in London (Kimmance et al. (1996)). The numerical analyses are in
good agreement with the data. The analyses indicate that sub-surface tunnel
interaction can be considered negligible for pillar widths greater than 7 diameters.
When considering piggy-back tunnels, there are two construction sequence
options - upper tunnel first and lower tunnel first. Figure 2.43 shows the
deformations induced in the existing lining in response to the second excavation
for both construction sequences. The change in diameter, as a percentage of the
design diameter, is plotted against pillar depth (extrados to extrados spacing
divided by diameter). Again the existing lining is pulled towards the new

120 1 Finite element analysis in geote~chnicalengineering: Application

Figure 3.18 shows a cross section through the site of the new Jubilee Line
Extension (JLE) station at Westminster which was completed in 2000. Very close
to the station excavation are two underground tunnels and the Big Ben clock tower.
There was obvious concern over the potential movements of the clock tower.
Although this is a rather special case, it is riot an uncommon situation in a crowded
city such as London.
Quite often developments are proposed which are close to other buildings
andtor services. It is therefore necessary to make an assessment of potential
damage or distortion. The distortion of a tunnel may prevent an underground train
from running, or at least mean that speed restrictions have to be imposed, since
tolerances are generally quite tight. In some instances they may be no more than
a few millimetres. In addition, there may be machinery, such as an escalator, close
by which is very sensitive to movement. If distortions become too large certain
parts can fracture in a brittle manner and suddenly jam the escalator, potentially
causing loss of life.
There are a number of methods of assessing building damage (e.g. Boscardin
and Cording (1989), Burland (1995)), but to make an assessment it is necessary to
know how a structure will distort. Therefore in any analysis only soil models which
are capable of making reasonable predictions of the magnitude and pattern of
movement should be used. This normally rules out any constitutive model which
incorporates linear elasticity (including linear elastic perfectly plastic models), see
Section 3.4.5.
To make an assessment of the
A
movement of an adjacent structure,
this structure has to be modelled in an
appropriate way and in sufficient
detail. For example, if the object of
the analysis is to estimate the
potential distortion of an adjacent
segmental tunnel, it is pointless
modelling the tunnel as a stiff ring.
Equally, if the tunnel is modelled as
fully flexible then the predicted E'
(10storey)
distortions may be excessive and
YO*
unrealistic. In order to be able to
make reasonable predictions, special Figure 3.60: Plan of the site for the
techniques have been developed to
Waterloo International Terminal
allow rotation ofjoints between lining
segments once limiting conditions are exceeded, see Section 2.4.5. This may be a
limiting stress or some other criteria.
Hight et al. (1993) present a good case history of where accurate predictions
of tunnel distortions were necessary. Figure 3.60 shows in plan the site of the
Waterloo International Terminal in London. Beneath the basement are two
Bakerloo Line Underground tunnels and to the side of the excavation are two
R

Earth retaining structures 1 1 2 1

Northern Line Underground
tunnels. Also at the side of the
excavation is an existing
structure, Elizabeth House,
and to complicate matters
further an additional two JLE
tunnels were planned (now
constructed) beneath the site,
see Figure 3.61. The scheme
that was originally proposed

PI-ad

Jubilee

had a deeper basement than Figure 3.61: Cross-section of the site for
the scheme that was
the Waterloo International Terminal
built, which made the base of
the excavation come to within less than 2m of the crown of one of the tunnels.
Plane strain finite element analyses were used to develop methods of construction
which limited the distortion of the tunnels and to prove to third parties that there
were unlikely to be any adverse effects. If movements had been excessive, at the
very least restrictions would have been imposed on trains running in the tunnels.
As an example of the results from this study, predicted deformations of the
station and running tunnels of the Bakerloo Line are shown in Figure 3.62. These
predictions were made before construction. However, during construction the
deformations of the tunnels were monitored, the recorded movements are also
shown on Figure 3.62. Overall, there is excellent agreement between the
predictions and the measurements.
There are a number of other examples, mostly unpublished, were such detailed
modelling has been necessary. These are truly soil-structure interaction analyses.

a) Deformations of station tunnels

b) Deformations of running tunnels

Figure 3.62: Predicted and measured deformations of

122 1 Finite element analysis in geotechnical engineering: Application

3.8
1.

2.

3.
4.

5.

6.

7.

8.

9.

Summary

The design ofretaining walls involves the assessment of forces imposed in the
wall and other structural members and the likely movements of both the
retaining wall and retained soil.
The design of retaining walls has traditionally been carried out using
simplified methods of analysis (limit equilibrium, stress fields) or empirical
approaches. However, such approaches cannot, and do not, provide all the
desired design information.
Numerical methods provide a viable: alternative and have the advantage that
they provide all of the desired design information in a consistent manner.
In reality, all geotechnical problems involving retaining structures are three
dimensional. However, at present, due to insufficient computing resources,
simplifications usually have to be made and two dimensional plane strain or
axi-symmetric analyses undertaken.
Care must be taken when assuming planes of symmetry. it is often tempting
to only analyse a 'half section'. This is only valid if the geometry, soil
conditions, construction sequence and loading conditions are all symmetric.
In practice this is rarely the case.
The lateral and vertical extent of a finite element mesh can have a significant
effect on predictions, if they are not large enough. Consequently, it is
important to insure that these boundaries are placed at a sufficient distance to
have a negligible influence. Unfortunately, it is not possible to give general
recommendations for the appropriate lateral and vertical extent of a mesh
because they will depend on the problem being analysed, the constitutive
models used for the soil and the facet of behaviour under investigation. In
practice it is therefore sensible to experiment with meshes of different size,
unless experience has already been gained of analysing a similar problem in
the past. It is always sensible to add some large elements to the sides and
bottom of a mesh. Relatively few of these elements are needed to expand the
mesh considerably and consequently they do not add a large overhead to the
analysis.
The methods of supporting walls have to be addressed (e.g. details of props,
ties, anchors, berms, etc.), as they can have a significant influence on
behaviour. In particular the type ofconnection between support and wall (e.g.
simple, pin-jointed or full moment) can have a large effect on displacements
and structural forces.
It is important to use realistic constitutive models to represent soil behaviour.
In this respect it is usually necessary to use models that can account for both
nonlinearity at small strains and soil plasticity.
The initial stresses within the ground before construction of the retaining wall
can have a significant influence on wall behaviour. The presence of adjacent
structures (e.g. existing tunnels, deep basements, etc.) will inevitably modify
the state of stress within the ground and these effects must be accounted for
in any analysis.

Earth retaining structures 1 1 2 3

Consideration must be given to how the retaining wall is to be constructed and
the time taken to do so. Different sequences of construction could result in the
soil experiencing different stress paths. As a consequence additional forces
might be imposed on the wall.
Any measures taken to control groundwater are an important consideration
and must be accurately simulated in any analysis.
Gravity walls are not usually used where there are adjacent structures and /or
services which are sensitive to ground movements. Consequently,
displacements are not explicitly calculated and simple design methods are
adequate. However, if movements are of concern, or in situations where the
soil conditions are unusual, or where the structure has dimensions outside
current practice, numerical analysis provides an ideal design tool.
A key feature of any numerical analysis of a gravity wall is the modelling of
the interface between the soil and the structure. Consequently, interface
elements combined with an appropriate constitutive model for the interface
should be used.
The behaviour of reinforced earth walls is extremely complex. Numerical
analysis provides an ideal tool to analyse this behaviour. However, such
analysis is not simple and involves the application of complex boundary
conditions.
Embedded retaining walls come in many forms and sizes. They are usually
used in situations where ground movements are critical. They also often have
several rows of props or anchors and are therefore difficult to analyse using
simple structural methods. Consequently, numerical analysis is widely used
to investigate the behaviour of such structures.
Embedded walls can be modelled using either solid or beam (or shell)
elements. However, the use of the latter type of element which has a zero
thickness implies further approximations. For example, for embedded
cantilever walls the use of beam elements results in larger lateral wall
displacements and higher bending moments.
If the wall is impermeable or has a finite permeability and a coupled analysis
is being undertaken, then the use of beam (or shell) elements to model the
wall can be problematic and special steps must be taken.
Anchors have three dimensional geometries and are therefore difficult to
model in plane strain analyses.

Appendix 111. '9 : Stress level

The stress level S is defined as (see Figure 111. I):

S=- J
Jf

where:

J is the current value of the deviatoric stress;

124 / Finite element analysis in geotechnical engineering: Application

Jf is the value of J o n the failure surface at the current value ofp' and 8.

For the Mohr-Coulomb constitutive model, see Section 7.5 of Volume 1 :

c'

-current

(I11.2)

stress state

Figure //l. I : Definition of the stress /eve/, S

4.

Cut slopes

4.1

Synopsis

This chapter describes the application of the finite element method to the analysis
of 'man-made' cut slopes (slopes which are formed by geological processes are not
considered here). To perform any meaningful finite element analysis of a cutting,
it is necessary to model the process of soil excavation (see Section 3.7.10 of
Volume 1). This is true even when any other geotechnical structure (tunnel,
embankment, foundation, etc.) in the vicinity of a slope is analysed. In such cases,
to establish initial stresses in the ground, the modelling of slope formation by
excavation is essential.
Cut slopes can be formed in both granular and clayey soils. Granular soils,
which have a high permeability, behave in a drained manner, both during
excavation and subsequently. Consequently, they are much easier to analyse. An
exception is, however, when they are subjected to earthquake loading, when the
granular material may respond in an undrained manner. The superficial stability
andlor modelling of the various drainage measures are often the only problems
associated with slopes cut in granular soils. Clay slopes have proved to be more
elusive, and therefore most of this chapter is devoted to the problems encountered
with the finite element analyses of slopes cut in these materials.

introduction
An excavation in clays of low permeability is likely to be undrained. However, in
me normally or lightly overconsolidated soft clays, excavation may almost be
ained (Chandler (1984a)). These clays are often of moderate thickness and are
ely to have sandy or silty drainage layers which accelerate pore pressure
uilisation, despite their relatively low permeability.
Generally, the excavation process unloads the soil and pore water pressures
become depressed, both beneath and adjacent to the excavation. With time the soil
ells as the pore water pressures equilibrate to the long term steady state seepage
gime, imposed by the 'new' hydraulic boundary conditions on the excavation
. With swelling and equilibration, mean effective stresses reduce and the
state approaches failure. The stability of the slope therefore reduces with
e. A detailed discussion of this process is given by Bishop and Bjerrum (1960).
One of the fundamental distinctions made in limit equilibrium slope stability

126 / Finite element analysis in geotechnical engineering: Application

analysis is that between short and long term conditions. It is usually assumed that
undrained conditions prevail in the short term. Given that it is not easy to predict
the pore water pressure changes immediately after slope excavation, stability
analysis in terms of total stresses (v, = 0" analysis) is often undertaken, in which
the undrained soil strength is assumed to be that existing prior to the formation of
the slope. in the long-term all excess pore water pressures generated during
excavation have dissipated, and the pore water pressures assumed to be in
equilibrium with the newly establishled hydraulic boundary conditions can be
determined. Consequently, an effective stress analysis can be undertaken.
Intermediate term situations cover the period between short and long term.
Unfortunately, the pore water pressure distribution during this transition period
cannot easily be calculated, and it is therefore difficult to carry out effective stress
stability analysis by the limit equilibrium method. However, finite element
analyses, which incorporate coupled consolidation/swelling, readily predict these
pore water pressures, and this is one reason (among others) why the finite element
method is particularly suited for this type of boundary value problem.
When analysing clay slopes, it is important to distinguish between the different
types of clay in which a slope is cut. The choice of constitutive model will largely
depend on the clay type. For example, stiff plastic clays are brittle and prone to
progressive failure. The mechanism of progressive failure is complex and
impossible to analyse without the incorporation ofstrain-softening in the analysis.
On the other hand, low plasticity tills do not usually soften to a significant extent,
and 'simpler' elastic-perfectly plastic constitutive models may well be employed
in analyses of slopes cut in this type of material. Soft normally to lightly overconsolidated clays usually undergo a substantial amount ofplastic straining during
the process of excavation. Then a constitutive model which incorporates plastic
behaviour pre-peak may be essential.
In this chapter examples of finite element analyses of slopes cut in the various
clay types are presented. Being more straightforward, the analyses of 'nonsoftening' clays are discussed first. Softening' analyses require a substantial
amount of computing effort, and they will be outlined later, after the mechanism
of progressive failure has been discussed in some detail.

4.3

'Non-softening' analyses

4.3.1
Introduction
The finite element method has only recently become a powerful alternative to the
limit equilibrium method of slope stability analysis. The recent papers by Griffiths
and Lane (1999) and Naylor (1999) outline the various techniques involved in the
analyses of slopes, and summarise many advantages of the finite element method
in conjunction with an elastic-perfectly plastic soil model of the Mohr-Coulomb
type.
Finite element analyses of this kind have been used here to shed further light
on the mechanisms taking place inside a slope since its formation. As it will be

Cut slopes 1 1 2 7

seen later (Section 4.5), these mechanisms can be extremely complex in cut slopes
formed in strain-softening materials. Thus 'non-softening' analyses provide a
useful reference point against which more complex slope behaviour can be judged.

Cut slope in stiff 'non-softening' clay

4.3.2
4.3.2.1 Introduction
Finite element analyses of a hypothetical cut slope in a stiff 'non-softening' clay
were carried out to verify the method of analysis subsequently employed (Section
4.5) in the analyses of slopes cut in stiff plastic 'strain-softening' clays.

4.3.2.2 Soil parameters

A nonlinear elastic-perfectly plastic soil model incorporating a Mohr-Coulomb
yield criterion given by the cohesion intercept, c', the angle of shearing resistance,
v', and the angle of dilation, v, was used in the analyses. The model requires the
specification of a Young's modulus, E, a Poisson's ratio, p, the (peak) strength
parameters, 9,' and c,', and the angle of dilation, v. The analyses were of a coupled
nature, and therefore it was also necessary to specify the coefficient of
permeability, k.
Detailed discussion of the derivation of the material properties is given in
Section 4.5.4.2. The model parameters are listed in Table 4.1.
4.3.2.3 Finite element analyses
'Non-softening' analyses were performed on a 10m high, 2: 1 slope. Note that
analyses of 3: 1 and 2.5: 1 slopes, 10m high, were also performed with a nonsoftening soil, but collapse was not predicted. The finite element mesh was similar
to the mesh shown in Figure 4.33, where a 3: 1 slope is modelled. No horizontal
displacement was allowed on the vertical boundaries, whereas the bottom boundary
was completely fixed in both the vertical and horizontal direction. The initial pore
pressure was assumed to be hydrostatic, with a surface boundary value of
pj,,=-1OkPa, this being the average surface value indicated by field measurements
in the U.K. (Vaughan (1994)).
Excavation was simulated by removing layers ofelements at a uniform rate. For
a 10m high cutting, excavation was complete in an overall time of 3 months. There
was no significant swelling during excavation.
After excavation, swelling was allowed with a surface boundary suction of
pfi=-1 OkPa maintained at the excavated slope, and on the original ground surface.
Seasonal fluctuations in surface boundary conditions were not modelled. The
vertical and bottom boundaries of the mesh were assumed to be impermeable.
Swelling was simulated in the finite element analyses by applying increments of
time, At, with no change to the other boundary conditions. The first time step after
excavation was At=0.25years, the second At=O.Syears. The subsequent steps were
At-lyear until collapse was approached, when the time step had to be reduced
substantially.

128 / Finite element analysis in geotechnical engineering: Application

Two sets of analyses were performed. In the first set, the angle of dilation was
assumed to be v=OO,and the initial in-situ stresses were assumed to be given by the
coefficient of earth pressure at rest K,,=1 .O, 1.5 and 2.0. The second set of analyses
was similar to the first set, except that the angle of dilation was v = q'.
Table 4.1: Properties of London Clay assumed in the analyses
l

Property

In-situ clay

Compacted f i l l

Bulk unit weight, y (kN/m3)

cPr=14kPa,(p,'=20

Peak strength (intact)

Discontinuity strength

c,'=OkPa, q,,'=20

no discontinuities

t
Bulk peak strength

cP1=7kPa,q,,'=20

Residual strength

c, '=2kPa7 v,'= 13"

Plastic strain at peak

(E;),

=0

- 5%

as peak intact

as in-situ clay

1OOO.(p'+ 1 OO)/ 100

(min. 2000)

(mean effective stress p' in

Permeability, k (mls)

Coeff. of earth pressure at

rest, K,,

varies

not relevant

4.3.2.4 Results of analyses

In the first set of analyses collapse was predicted during the swelling process, with
a well defined rupture zone of the same shape in each analysis. Figure 4.1a shows
the change with time of the predicted average pore water pressure on this rupture

Cut slopes 1 129

surface, expressed as the pore

pressure ratio, r,,' = p,lo,, , where p, is
pore water pressure, and g,, is the
predicted vertical stress on the rupture
surface. This definition differs
slightly from the conventional one, r,,
= p,l(yh), where yh is an overburden
pressure. r,,' is the average value
along the rupture surface. All three
-*
analyses gave r , = 0.235 at collapse.
Also shown on this figure is the
equilibrium value of 6"
= 0.3. This is
the value that would have been
obtained after a very long time, if
collapse had not occurred. It was
calculated by performing a separate
long-term steady state seepage
analysis. The associated change with
time of the mid-slope horizontal
displacement, 6, , is given in Figure
4.lb. It is evident from these plots
that collapse occurs suddenly, after a
delay which depends on the value of

o,50i
i.:

l_" - - - - - _ - - - _"LsY~!?!!u~~.:Io.?o..

- --- - - --

0.25~

P
Time since excavation, t (years)

Or-=
a)

Time since excavation, t (years)

3 ,,

10

5
g 0.50

Kpl.0

B
a

IS

20

25

30

35

40

45

50

K71.5

Ko=2.0

1.00

Figure 4.1: Non-softening clay with

v= 0"; development of a) 7,' on the
rupture surface and b) mid-slope
displacement, 6,,, with time
0.50

:;i
f 0.25

K,, .

The results from the second set of 2

g 0.00
analyses are shown in Figure 4.2.
Well defined rupture surfaces were
again predicted, of the same shape in O
each analysis, but of different shape 9 -o.50
to that predicted with V = 0. This is
Time sincc excavation, t (years)
why there is a slight difference 8 o,ooo
between the steady state equilibrium 5
values of
in Figures 4. l a and 4.2a.
o.50
The final values of the average pore :
water pressure at collapse were all
1.00
= 0.25, which is slightly higher than
for the first set of analyses.
Figure 4.2: Non-softening clay with
Figure 4.2b shows the change of
v= I$> development of a) P,' on the
mid-slope horizontal displacement
rupture surface and b) mid-slope
with time. When rupture develops and
displacement, 6 , with time
the average pore water pressures
become constant, the displacements do not accelerate as in Figure 4. l b. Instead,
there is a gradual increase in displacement with time. This is due to two
compensating effects. During swelling pore water pressures are generally

c*

c*8

130 / Finite element analysis in geotechnical engineering: Application

increasing with time, whereas in the thin rupture zone the soil is dilating and the
pore water pressures are reducing. At collapse the rate of decrease in pore water
pressure in the rupture zone is exactly matched by the increase due to swelling. The
generation of negative pore water pressure is controlled by the permeability of the
clay and the thickness of the rupture zone. This accounts for the gradual increase
in displacement, which is predicted with v = p' once a rupture surface has formed,
and for the higher average pore water pressure at collapse. To check the validity
of the latter statement, an attempt was made to restart these analyses but with the
angle of dilation, v, reduced to zero. In each case it was not possible to maintain
equilibrium of the solution and very large displacements occurred. The pore water
pressures generated by dilation essentially act as a partial brake on post rupture
movements. This effect is of great practical significance, since it helps to prevent
rapid post-collapse movement of slides in strain-softening clays (Vaughan (1 994)).
In reality, dilation would be restricted in the field to a much thinner rupture
zone than that predicted by the analyses, where the rupture surface is
approximately half an element thick (typically 0.5m for the current mesh).
Equilibration of pore water pressure would occur more rapidly in the field. Thus,
if it is assumed that dilation occurs post-peak and only on thin shear surfaces, as
occurs in laboratory tests (Sandroni (1977)), the analyses in which v=Oo are likely
to give a better representation of undrained effects in clay in which discontinuities
develop post-peak, unless an unrealistically large number of elements are used to
allow a thin rupture zone to develop. Consequently, as it will be seen later, for
analyses involving strain-softening, an angle of dilation v=OOwas adopted.
The angle of dilation also affects the position of the rupture surface predicted
and the stresses acting on it. The angles which the predicted rupture surfaces
(velocity characteristics) make with the plane on which the major principal stress
acts are &(45"+v/2). The shear stresses acting on these rupture surfaces are given
by:
c'cosp'cosv+ Q,', sinp'cosv
Zf =
1 - sinp'sin v
When v=pl, Equation (4. l ) reduces to the Coulomb failure criterion, namely:
r, = c' + Q; tan p'

(4.2)

If v#y~',the strength on the rupture surface from Equation (4.1) does not reduce to
the Coulomb criterion. For instance, if v=OO,Equation (4.1) reduces to:

r, = c' cssq' +Q; sinp'

(4.3)

For clays, the'differences are small, but they need to be taken into account
when comparisons are made between the results of finite element analyses and
limit equilibrium calculations in which the Coulomb equation is adopted (see Potts
et al. (1990)). If the rupture surface is kinematically constrained, then the
differences are likely to be mainly in the strength on the rupture surface. If the
rupture surface is not constrained, the differences are likely to be mainly in the
location of the rupture surface itself.

Cut slopes 1 1 31
v=cp'V=OAs noted previously, the rupture
f
surfaces predicted by the nonsoftening finite element analyses
depend on v. The two surfaces
predicted for v=@ and v=OOare shown
in Figure 4.3. Also shown is the
critical circular slip surface obtained
from a limit equilibrium calculation Figure 4.3: Comparison of predicted
using the Bishop rigorous method of
rupture surfaces from 'nonslices (Bishop (1955)). In the cut
softening' analyses
slope problem there is little kinematic
restraint on the position of the rupture surface, and it is therefore not surprising that
the rupture surfaces from the finite element analyses are dependent on v. It may be
noted that at the soil surface the major principal effective stress is vertical and the
rupture surfaces should be inclined at an angle of (45" - v12) to the vertical.
Inspection of Figure 4.3 shows this to be so.
From the limit equilibrium ,
analysis the average pore water
25
l*
pressure
ratio
on
the
slip
surface
was
g
v,,' = 0.233 at collapse. This is in $ 2 0
good agreement with the finite
element results (K*=0.235 for v=OO
Q
and r," = 0.25 for v=#). The average a 1s
L I ~ eqlhbnum
I ~
0 FEM - w ~ t hdllat~on
values of the shear and normal stress P
D FEM - no d~latlon
acting on the rupture surfaces at 3
E
30
35
40
45
50
failure are plotted in Figure 4.4 along P
Average normal effectrve stress, c' (Ha)
with Equations 4.2 and 4.3. As would
be expected, the results from the finite
Figure 4.4: Average shear stress at
element analyses with v=OOagree with
collapse predicted by 'nonEquation 4.3, whereas the analyses
softening' FE analyses and limit
with v=cpl and the limit equilibrium
equilibrium analysis
analysis agree with Equation 4.2.
In the softening analyses reported later, Equation 4.3 is used. Since there is
little apparent kinematic constraint on the rupture surfaces, and v=OOis assumed,
a small difference between the predicted rupture surfaces and those in the field is
likely.

4.3.3 Cut slope in soft clay

4.3.3.1 Introduction
In spite of the limited evidence for depressed pore water pressures following
excavation, it is generally accepted that an excavation in stiff, heavily overconsolidated clay with low in-situ permeability is undrained. The situation is less
clear with soft, normally or lightly-overconsolidatedclays. As mentioned above,

132 / Finite element analysis in geotechnical engineering: Application

slopes cut in these materials may achieve long-term pore water pressures either
during excavation or within a few months after excavation. The time required for
the pore water pressures to equilibrate to the steady-state condition is important
when designing a cut slope. Given that stability of a slope reduces with time, it
may be wise to either measure the pore water pressures during excavation, or
design slopes in soft clays on the basis of long-term conditions, even when the
slope is only temporary.
It has been appreciated for some time (Tavenas and Leroueil(1980)) that partial
drainage can play a significant role in short term problems involving embankments
on soft clays, where the measured pore water pressures during the early stages of
relatively rapid loading fell short of those expected for undrained conditions.
However, while Chandler (1984a) has argued that soft clays are often of moderate
thickness and are likely to have sandy or silty drainage layers, which accelerate
pore water pressure equalization despite their relative low permeability, Tavenas
and Leroueil (1980) concluded that these phenomena may be due to 'nonlinear
consolidation'. Namely, soft clays have high initial shear and bulkstiffnesses when
loaded (or, in the case of cuttings, unloaded) from their in-situ conditions.
Although these stiffnesses reduce rapidly as strains grow, the operational 'linear'
equivalent consolidation (swelling) coefficients would have been very high initially
as construction (excavation) proceeds, and when a significant factor of safety
operates in soft clays (further away from yield').
To investigate the possible effects of a high soil stiffness at small strains on the
time required for the pore water pressure equilibration, a series of finite element
analyses of a hypothetical slope cut in a soft clay was undertaken.

4.3.3.2 Soil parameters

pressure (kPa)
A form of the modified Cam clay model (see -50 0 Pore50water100
150 200 250
\L'"'""""~'"'
Chapter 7 of Volume l), which had a MohrWeathered crust
Coulomb hexagon and a circle for the shapes
2
k
of the yield and plastic potential surface in
the deviatoric plane respectively, was used to
represent the soft clay (and its weathered
crust). The elastic behaviour was modelled in
two ways. In the first approach, the basic
critical state formulation, in which both bulk
and shear moduli are derived from the slope
of swelling lines, K, assuming a constant
value of Poisson's ratio, p, was utilised. In
the second approach the nonlinear elastic
behaviour proposed by Jardine et al. (1986)
Sand
(see Section 5.7.5 ofvolume 1) was used. As
Figure 4.5: Observed and
this model is able to capture the high soil
stiffness at small strains for stress states adopted pore water pressure
profile
inside the yield surface, it is more reakistic.
" " " I

Cut slopes / 133

The soil profile used in the analyses is presented in Figure 4.5, together with the
ground water conditions. It consists of four distinctive layers: (i) a weathered crust
extending to 4m depth, (ii) a 10m thick layer of soft clay, (iii) stiff clay extending
from 14 to 22m, overlying (iv) a sand layer (not modelled in the analyses).
Model parameters for the soft clay
'OoO
were derived on the basis of a
comprehensive set of both field and m
laboratory investigations on the soft
Bangkok Clay (Shibuya (1999)).
60
They are listed in Tables 4.2 and 4.3
for the soft and stiff clays 2
respectively. The derived small strain
200
stiffness parameters are presented in
0.1
1
10
O.OO'
O."
Table 4.4 and they generate the small
Axial strain, E~ (%)
strain stiffness curves shown in
Figure 4.6.
so
Single element tests have been $ 40
performed to predict the behaviour of
the soft Bangkok Clay in triaxial 4
compression and extension (Figure
,,,,
0.1
I
10
4.7) and oedometer loading and
Volumetric strain, E, (%)
unloading (Figures 4.8). For the latter
test a prediction is also made using
~i~~~~
4.6:~
d non-l;near
~
~
the standard critical state formulation
elastic s m a strain
~
secant
for the elastic component of the
and b, bulk stiffness
a,
model (Figure 4.9j. The observed
response in laboratory oedometer tests on three samples from different depths is
superimposed in plots of the vertical effective stress, o,,',versus volumetric strain,
E,, . The match with the finite element predictions is reasonable.

33 ,

-a-Triaxial compression
-c Triaxial extension

- - - Critical State Line

compression

.....--.Critical State Line

extension

Figure 4.7: Predicted response of soft Bangkok

Clay in triaxial compression and extension tests
fsmaN strain stiffness used)

134 1 Finite element analysis in geotechnical engineering: Application

FE prediction:
--c Loading
-Critical
State L i e
compression
---- Critical State L i e
extension
-+-Unloading

I
W

Lab tests:
---Depth 5.5m
......,.,...Depth 8.5m
Depth 12.5~11

..

-100

Average effcctivc stress, S' (kPa;

100

200

-.
300

Vmical effective stress, U,' (kPa)

10
100
IOW

50
l0

.S-

20

'

40

Figure 4.8: Predicted and observed response of

soft Bangkok Clay in oedometer test
(small strain stiffness used)

FE prediction:
-+Loading
Critical State L i e
compression
- - - - Critical State Line
extension
-c Unloading

150

-100

-.

Lab tests:
--- Depth 5.5m
............Dep(h8,Sm
- .- .- Depth 12.5~11
Venical effective stress, a,' (kPa)
10
100
1000

Figure 4.9: Predicted and observed response of

soft Bangkok Clay in oedometer test
('standard' modified Cam clay model usedl

The most critical issue when dealing with the modelling of soft clays is usually
their undrained shear strength, S,, . While the undrained shear strength is not an
input parameter to the modified Cam clay model, it can be calculated from the
input parameters and the initial stress conditions (see Section 7.9.3 of Volume 1).
The model parameters listed in Tables 4.2 and4.4, in combination with the profiles

Cut slopes / 135

of the overconsolidation ratio, OCR, (Figure 4.10) and the coefficient of earth
pressure at rest, K,, ,(Figure 4. l l) have generated the undrained strength profiles
given in Figure 4.12. It can be seen that predictions depend not only on the type of
the test modelled (triaxial compression or plane strain), but also on the elastic
component of the model used in the analysis. However, the trend observed in both
field (vane tests) and laboratory (triaxial compression tests) is reasonably captured.
Table 4.2: Soil properties assumed for soft clay (modified Cam clay)

Poisson's ratio, p

used, see Table 4.4

136 / Finite element analysis in geotechnical engineering: Application

Overconsolidation ratio, OCR

Coefficient of earth pressure at rest, K,

10 15 20 25

30 35 40

Figure 4. 7 0: Adopted OCR

profile in soft clay

0.5

1.5

2.5

Figure 4.1 I: Adopted K,

profile in soft clay

4.3.3.3 Finite element analyses

The finite element analyses were of the coupled consolidation/swelling type, and
therefore it was necessary to specify the permeability of the soils involved. It can
be seen from Figure 4.5 that the observed pore water pressure profile is
underdrained. To model underdrainage, it is necessary to adopt a permeability
model in which permeabilities decrease with depth (see Chapter 9). The non-linear
model used here, which relates the permeability, k, to the mean effective stress,pf,
is given by the following relationship (Section 10.7 of Volume 1):

where k,, is the coefficient of permeability at pl=O and a is a material parameter.

The value of a=0.0125m2/kPa was deduced in order to match the observed
underdrained pore water pressure profile (see Figure 4.5). The permeabilities
obtained from site investigations at three different depths are presented in Figure
4.13. Also shown is the permeability profile calculated using the above equation
adopting the value of k,, = 1.2 X 10-gmls . Better matching is possible, but then the
observed underdrained permeability profile could no longer be recovered. Note
that the same nonlinear permeability model parameters, k , and a, were used for
both soft and stiff clays. The sudden drop in permeability predicted when entering
the stiff clay layer is due to the higher adopted value of K,, in this material.
The finite element mesh used to model a 1 :1 cut in the soft Bangkok Clay is
shown in Figure 4.14, together with the displacement and pore water pressure
boundary conditions. Excavation has b.een modelled by removing soil in Im thick

Cut slopes 1 137

horizontal layers, with short time steps to ensure an undrained soil response. One
analysis increment per layer was typically employed initially when a significant
factor of safety operated in the slope. When collapse was approaching it was
necessary to use more increments (typically ten) per layer in order to catch the
critical depth of 'undrained' excavation.
Undrained shear strength, S, (kpa)

Coefficient of permeability, k (mls)

1o ' ~

0 10-'O

- o

-wiaxial compression
- - - plane s w m
Q

16t

vane shear test 2

vane shear test 3

Weath ed Clay

Stiff Clay

adopted

22

Figure 4.12: Measured and

predicted S, profiles in soft
clay

Figure 4.13: Measured and

adopted permeability
profiles

0 Scale 10m

Figure 4.14: Finite element mesh and boundary

conditions

Undrained excavation in soft clay slopes causes a substantial amount of

shearing, particularly in the later stages of excavation when failure conditions are
approached. Consequently, the influence of the stiffness at small strains on
subsequent times for equilisation of pore water pressures depressed by excavation
may well be diminished. To check the influence of small strain stiffness on
swelling times after 'undrained' excavation, it was necessary to ensure a sinall

138 1 Finite element analysis in geotechnical engineering: Application

'disturbance' of the ground due to excavation, i.e. a shallow depth of excavation.

Then the swelling was modelled in ever-increasing time increments until the long
term steady-state seepage condition was established. This condition was imposed
by a surface boundary suction of pfi=- lOkPa at the excavated slope and on the
original ground surface. Underdrainage at the bottom of the mesh was maintained
throughout the analyses.

4.3.3.4 Results of analyses

Figures 4.15a and 4.15b show the
development of horizontal and vertical
movements of the crest of the slope during
150
excavation. A critical depth of 7m was
predicted by the analysis employing the small
strain stiffness model. The 'standard'
modified Cam clay model, in which the
elastic stiffness at small strains was not
included, recovered a slightly shallower
Depth of excavation (m)
critical depth (dc=6.7m). The reason for this
zso
can be found in Figure 4.12, where it can be
seen that the introduction of the greater
stiffness at small strains in the modified Cam
clay model leads to the prediction of a higher
undrained shear strength, S,,, at least in the 2
first 5m of the profile. Also, it is of interest to
,,
note that due to this higher stiffness smaller $
horizontal displacements of the crest are
O0 1 2 3 4 ~ 6 7 8 9 1
~ e p t hof excavation(m)
predicted, particularly during the early stages
of excavation, when no significant yielding
occurs in the slope. Surprisingly, the crest Figure 4. 15: a) Horizontal and
movements in the vertical direction are
b) vertical movements of
almost the same.
slope crest against depth of

1 ,,

excavation

Figure 4.16: Critical slip surface and vectors of

incrernental displacem.ents just prior to collapse

Cut slopes / 139

The position of the critical slip surface, predicted by the finite element analysis
in which the elastic small strain stiffness model had been used, is shown in Figure
4.15, together with incremental displacement vectors at the 'end' of excavation,
just before collapse. The vectors show the current mechanism of movement, and
their absolute values are of no syignificance.
Table 4.4a: Coefficients and limits for nonlinear elastic secant shear
modulus expression

Table 4.46: Coefficients and limits for nonlinear elastic secant bulk
modulus expression

c',

Development of the average pore water pressure ratio,

along this slip
surface (the critical slip surface after 7m of excavation) with time is shown in
Figure 4.17, for an analysis in which
excavation stopped after only 3m of I"'
2
em at smau smm
the soft clay had been excavated. 5
Figure 4.18 shows the contours of
0.35
stress level, S, at this stage of the
analysis. The stress level, S, is a
measure of the current strength g
mobilised, and its reciprocal value can 4
O2
be considered as a local factor of
safety, see Appendix 111.1. A
T~mesmce excavahon (years)
reasonable (global) factor of safety
still operates in the slope. For the Figure 4. 17: Average pore pressure
permeability profile adopted in Figure
ratio against time

,,

8
2,

140 / Finite element analysis in geotechnical engineering: Application

4.13, a time period in excess

of 20 years is necessary for
nearly complete equilisation of
the depressed pore water
pressures caused by a 3m deep
excavation. Also shown on
Figure 4.17 are results from a
similar analysis, but with the
modified camclay
elastic parameters. It is
apparent that, although the
small strain stiffness model
speeds up the process of
swelling for some time after
excavation (when strains are
small), the time for complete
equilisation is largely
unaffected.

Sllp surface from Flgure 4 16

-A

-A

L
-A-OS

-B-07

Figure ,4.18: Contours of stress level, S

Slrp surfice ram F~gute4 16

To investigate this matter Figure 4- 19: Contours of deviatoric strain,

further, an additional analysis
E d , at Zm and 3m of excavation
has been carried out in which
the high soil stiffness at small strains has been re-invoked (the shear strains have
been 'zeroed') immediately after excavation. Because small time steps have been
used during excavation to ensure an undrained soil response, the volumetric strains
are already zero. It can be argued that at the start of swelling (after undrained
excavation has been carried out) the stress path (or strain path) changes direction,
at least in a large part of the mesh. This is associated with an increased soil
stiffness response (see e.g. Jardine (1994)). Then, the assumption of zeroing strains
in order to account for an increased soil stiffness may well be justified. Also, a 3m
deep excavation in soft Bangkok Clay causes a reasonable amount of straining (see
Figure 4.19), and therefore partially eliminates the influence of the small strain
stiffness on the time for swelling. Thus, the analysis with 'zeroing' the strains
immediately after excavation may also indicate the response that may be expected
after excavation to a shallower depth, when the amount of straining is smaller
(Figure 4.19) and consequently soil stiffness higher.
Results from such an analysis are also presented in Figure 4.17. It can be seen
that in this case a much quicker pore water pressure response has been predicted
for all times after undrained excavation. Again this is particularly so during the
initial stages of swelling when smaller strains operate in the slope. It seems that
partial drainage can occur in the short term when excavating in soft clays.
However, for excavations in soft Bangkok Clay at least, the soil response is not
likely to be completely drained. For this to happen, a soft clay must be of higher
permeability andtor possess sandy or silty drainage layers which accelerate pore
water pressure equalization.

Cut slopes 1 141

4.4

Progressive failure

The term progressive failure is best

explained by considering the simple
example of a beam of elastic material
10m
sliding on a rough surface with a
=m, tan11
brittle shear stress displacement
characteristic (see Figure 4.20). The
'
normal stress acting on the rough ,
surface due to the weight of the beam
0
10
20
"O
20
is G,,, the friction angle between the
6 (m)
6 (mm)
beam and the rough surface is p and
the relative displacement between the
Figure 4.20: Progressive failure of a
beam and the surface is 6.
beam
As noted on Figure
4.20, the
friction angle drops with further displacement, after a peak value has been
mobilised, to a constant lower residual value. The relationship between the shear
stress, r, generated between beam and surface, and the relative displacement, 6, is
also shown on Figure 4.20. It can be seen that initially the shear stress increases
with increase in relative displacement until a peak value is mobilised. With further
displacement the shear stress drops until a residual value is reached. It then remains
constant at this residual value. This is brittle behaviour.
The beam is loaded at its left hand end by a force F, giving rise to an associated
displacement, A. Two scenarios will now be considered. In the first the beam is
assumed to be rigid, whereas in the second it is assumed to have a finite stiffness.
Consider first the case when the beam is rigid. The beam is loaded at its left
hand end by progressively increasing the displacement A. The resulting behaviour
is shown in Figures 4.21a to 4.21f. Each of these figures refer to a different stage
of loading (i.e. to a different value of A). Several diagrams are shown on each
figure. The top diagram indicates the distribution of relative displacement, 6, along
the beam. The dots on this diagram represent the displacement, 6, at equi-spaced
positions along the beam. In a similar manner the bottom diagram shows the
distribution of interface shear stress, r, along the interface. Also marked on this
diagram are the peak, r,,, and residual, r,, shear stresses. These are evaluated from
G,,,which remains constant along the beam, and p,, and p, respectively. The three
small diagrams in the centre of each figure are used to track the shear stressrelative displacement relationship as it is mobilised at three positions along the
beamlsurface interface. The centre diagram is for the middle of the beam. The two
end diagrams represent positions just in from the ends of the beam. The thin line
in these diagrams indicate the shear stress-relative displacement curves that will be
followed as the beam is loaded (i.e displaced).
Initially (see Figure 4.21a), the beam is not Ioaded and both the relative
displacement and shear stress are zero. The situation after the left hand side of the
beam has been displaced by 2mm is shown in Figure 4.21b. Not surprisingly, as
the beam is rigid, behaviour along the beam is everywhere the same. This uniform
7

"";p

142 / Finite element analysis in geotechnical engineering: Application

behaviour continues as the beam is loaded (i.e. displaced) further. Figure 4.2 1c
shows the situation when peak shear stress is mobilised everywhere along the
beam. With further displacement (see Figure 4.21d) the shear stress drops from
peak towards residual. Once enough displacement is applied to reach residual
conditions (see Figure 4.2 le), the stresses remain at this value with further loading
(see Figure 4.2 1f).

Peak

Oo

ti

p]

,
6

,
8

Distance
io@)

,
2

Dtstance
IO(~)

, DtsIanee

IO("')

, Dislance
IO(~)

Residual
L

, Distance "
to("')

0]

d)

c)

]
0

i " ~ - - - - - -

'0

e)

, Distance
10@)

O
0

Distance v
10(~)

Figure 4.2 1: Progressive faiiure of a rigid beam

--Residual
Distance
10("')

Cut slopes 1 143

The relationship between the force
F and the displacement A of the
150
loaded end of the beam (i.e. the left
hand end) is shown in Figure 4.22.
loo
The behaviour mirrors the shear
stress, t,versus relative displacement,
Rigid beam
6, behaviour at the interface. The
maximum resistance, F,,, , is simply
the peak shear stress multiplied by the
0
area of contact.
o
S
10
IS
20
D~splacement,A (m)
If the same loading exercise is
now repeated with a beam of finite
stiffness, the results are as shown in Figure 4.22: F-A relationship for
rigid beam
Figures 4.23a to 4.23f. These figures
are in a similar format to those shown
in Figure 4.21 and discussed above. Before loading is applied, conditions are the
same as for the rigid beam shown in Figure 4.21a. The situation after the left hand
end of the beam has been displaced by 2mm is shown in Figure 4.23a, and can be
directly compared with the results for the rigid beam given in Figure 4.21 b. The
displacement and shear stress are no longer uniform along the beam. They are both
larger at the left hand end of the beam and decrease with distance along the beam.
As loading continues, non-uniformity develops further (see Figure 4.23b).
Figure 4 . 2 3 ~shows the situation when the shear stress at the left hand end of the
beam has reached peak and has begun to soften towards residual. In the lower
diagram of Figure 4 . 2 3 ~it can be seen that the two dots nearest the left hand of the
beam are on the softening branch of the shear stress-relative displacement curve.
The third dot has just reached peak shear stress and the dots remaining are still on
the initial (hardening) branch ofthe shear stress-relative displacement curve. With
further loading the resistance at the left hand end of the beam reduces, whereas it
increases along the rest of the beam (see Figure 4.23d). Eventually (see Figure
4.23e), all of the beam-surface interface is operating on the softening branch of the
shear stress - relative displacement curve. After large displacements all positions
along the beam reach and maintain a residual condition (see Figure 4.230.
At no time is there more than one position along the beam which is mobilising
peak shear stress at the beam-surface interface. Hence the term progressive failure.
In contrast to the rigid beam, the maximum total resistance, F,,,,,,must be less than
the peak shear stress multiplied by the contact area. The total resistance, F, is
plotted against displacement, A, of the left hand end of the beam in Figure 4.24.
Also shown in this figure is the result for the rigid beam. Compared to the rigid
beam it may be noted that: (i) the behaviour has become nonlinear, (ii) the
maximum resistance, F,, , has been reduced, and (iii) more displacement is
required to mobilise the maximum resistance.

144 1 Finite element analysis in geotechnical engineering: Application

;. . .,..
2

.,6

Distance
IO(~)

Peak
Residual

10

"

'

"

Peak

Distance

Iso~

Peak

Residual

Distance
10(~)

Distance
IO("')

d)

c)

-- Peak

...............

- -

Residual

Figure 4.23: Progressive failure of a compressible beam

The data in Figure 4.24 is replotted in Figure 4.25 with the force F divided by
the contact area between the beam and surface to give a plot of average shear
resistance against displacement A. Also shown on this figure are two further
results. One is for the same compressible beam as above, but loaded at its centre
instead of at its end (see inset in Fjgure 4.25a). The other is for a similar

Cut slopes 1 145

compressible beam of twice the
length of the one considered above,
150
but still loaded at its left hand end
(see inset in Figure 4.25b). For the
.Compressible beam
beam loaded at its centre the
maximum force, F,,, , is greater and 4
8
the response stiffer than for the case $ ,,
discussed above. Clearly, the way in
which the beam is loaded affects the
o
5
10
15
20
amount of progressive failure, and
Displacement, A (mm)
consequently the maximum
resistance. For the longer beam the
response is softer, and the maximum Figure 4.24: F-A relationship for
compressible beam
average shear stress, tmaXav
is lower.
Thus, increasing the size of the beam
increases the amount of progressive failure and reduces the maximum resistance.
This simple example shows that for progressive failure to occur: (i) the material
must be brittle, (ii) a stress concentration must exist (i.e. the interface in the above
example), and (iii) the loading must be non-uniform. The example also shows that
the amount of progressive failure is influenced by: (i) the relative stiffness of the
materials involved, (ii) the nature of the loading, and (iii) the scale of the shearing
surface. Consequently, it is not surprising that progressive failure cannot be
analysed using simple methods.

gm-

Displacement, A (mm)

Displacement, A (mm)

Figure 4.25a: %,-A relationship

for a single length beam

Figure 4.256: r,,-A relationship

for a double length beam

4.5

'Softeningpanalyses

4.5.1
Introduction
The delayed collapse of old British railway cutting slopes in stiff plastic clays, by
the formation of deep-seated slides, has been studied extensively (see e.g.

146 / Finite element analysis in geotechnical engineering: Application

Skempton (1977)). Collapse is delayed primarily by slow swelling and pore
pressure equilibration (Vaughan and Walbancke (1973)). However, the average
drained strength operating at collapse is considerably less than the peak strength
measured in the laboratory. The clays involved are brittle, and the discrepancy
between the strengths has often been attributed to progressive failure.
Recent enhancement to the finite element code at Imperial College has enabled
both strain-softening and coupled consolidation/swelling to be included in an
analysis. Using these developments Potts et al. (1997) were able to disclose the
role of progressive failure and explain most of the phenomena associated with the
delayed failures of slopes composed of strain-softening clays. These analyses,
together with parametric studies into the effects of the initial in-situ stresses, the
surface hydraulic boundary conditions and the slope geometry, will be presented
first and the results compared with field data. In an attempt to model the commonly
observed sliding in the superficial soil layers of a slope, a higher permeability layer
due to deterioration of the surface clay was introduced and the results of some of
these analyses are presented. Finally, the effects of subsequent changes to slope
geometry, such as those frequently employed to widen the existing motorway
network in Britain, are examined in some detail and the possible consequences
discussed.

4.5.2

Choice of constitutive model

For progressive failure to play a role in the delayed collapse of cut slopes, the soil
must exhibit brittle properties and nonlinear strains must be mobilised along any
potential rupture surface (see Section 4.4). Simple methods of analysis, such as
limit equilibrium, are incapable of dealing which such complexities, and therefore
little progress has been made theoretically in quantifying the possible effects of
progressive failure. The use of the finite elemenr method to analyse the recent endof-construction slip in the Carsington embankment (see Section 5.4.4 of this
volume) has shown that progressive failure can be accurately reproduced and
quantified. However, those studies involved large embankments which were
subjected to loading and consolidation. Here the same techniques have been
applied to cuttings which involve unloading and swelling. In addition, the analyses
include coupled consolidation which
enable the history of swelling and, in -g (p,
particular, the time to collapse to be
v.' , c,'
predicted.
The constitutive model employed
to model soil strain-softening
(E$
(E>,
E,'
behaviour has been described in
Section 7.8 of Volume 1. This is an
elasto-plastic model of the Mohr- Figure 4.26: Variation of 4' and c '
Coulomb type in which softening
with the deviatoric plastic strain
behaviour is accounted for by invariant, E/ (POtts et al. (1990))

p\r
,C;

Cut slopes 1 147

allowing the angle of shearing resistance, rp', andlor the cohesion intercept, c', to
vary with the deviatoric plastic strain invariant, E l , as shown in Figure 4.26. The
incremental deviatoric plastic strain invariant, AEJ, is defined as:

The model requires the specification of peak (q,,', c,') and residual (v,', c,')
strength, the angle of dilation, v, pre-peak stiffness, E, stiffness in unloading, E,,,
and the rate at which strength is lost with strain ((EJ), , (EJ'),).As the analysis
involves swelling, it is also necessary to specify a permeability, k.
The stiffness prior to failure is given by a simple elastic model in which
Poisson's ratio, p, is constant, and Young's modulus, E, varies with mean effective
stress,pl, but not with shear stress level. No distinction is made between unloading
and loading. Consolidation and swelling behaviour indicate approximately the
same modulus in stiffclays, and therefore the Young's modulus used in the present
analysis is based on the appropriate swelling modulus. The parameters used are
listed in Table 4.1.

4.5.3
Implications for convergence
Monitoring of convergence is difficult for analyses involving strain-softening. The
norms (see Section 9.6 of Volume 1) of the iterative nodal displacements, loads,
flows and pore water pressures were all kept less than one percent of the norms of
the associated incremental values. However, this procedure alone was not adequate
and, in addition, the residual stresses at all integration points were monitored and
kept small. For the early stages of an analysis, residual stresses were kept below 2
kPa. However, when collapse was approached more stringent conditions were
introduced and stresses were kept below 0.1 kPa. To achieve this level of accuracy,
small time steps (typically 0.0025 years) and a large number of iterations (typically
200) were required towards the end of an analysis. The time at collapse was
deduced by plotting horizontal mid-slope displacement against time. Once the
collapse time was exceeded, the solution became unstable, since the slope could
not then be in equilibrium. This was always confirmed by running the final
increment for a small time step of 0.0025 years and a large number of iterations
(typically 400). Instability was indicated when deformations increased according
to the number of iterations, without an improvement in convergence (KovaCevid
( 1 994)).

Cut slopes in London Clay

4.5.4
4.5.4. 1 Introduction
The delayed sliding of slopes cut in stiff plastic clays has been the subject of
considerable soil mechanics interest for a number of years (see e.g. Terzaghi and
Peck (1948)). In Britain, most of the data concerns slopes and soil properties for

148 / Finite element analysis in geotechnical engineering: Application

Brown London Clay (see e.g. Skempton (1977)), and that is why slopes in this
material have been chosen for this study.

4.5.4.2 Soil parameters

As analyses of progressive
failure are complex, it is
essential to pay great attention
to the derivation of the model
parameters for the materials
involved. As an example, the
determination ofthe properties
for Brown London Clay is
now briefly presented.

100

3
6

'
.I-

Oo

100

200

300

(U,'+U,,)I~(Wa)

Figure 4.27: Peak strength of fissured

Brown London Clay in terms of effective
Figure 4.27 shows the data
stress (Sandroni ( 19 7 711

Peak strength

from undrained triaxial

compression tests on soil
50
samples of different diameter.
The tests on larger samples
give a peak strength envelope,
according to Mohr-Coulomb,
of cPf=7kPa and G'= 20,
50
100
150
Normal effective stress, a,' (Wa)
which is considered to be
representative of the fissured Figure 4.28: Drained residual strength of
clay structure.
Brown London Clay

Residual strength
Measurements of residual strength are
summarised in Figure 4.28. Both ring
shear and field data are shown. The L
strength back-calculated from field slips
is slightly higher, probably due to the less
planar shear surfaces usually formed in +
the field. The field strength of cr1=2kPa
and qrl=l3" is adopted.

8
5

test (Apted (1977))

8
-9

StifJess
The clay in a slope swells with time, and
the strains and strain-softening are
Strongly influenced by the Strain energy
released by Swelling. Very little
experimental data is available for

100
200
Vertical effective stress, cr,' (kPa)

Figure 4-29; Observed and

predicted swelling of Brown
London Clay

Cut slopes 1 149

swelling along different stress paths. Some swelling data from oedometer tests
(Apted (1977)) are shown in Figure 4.29. Also shown is the swelling curve
obtained by plotting depth-water content profiles (Skempton and Henkel(1957)).
The swelling lines predicted by the simple elasto-plastic constitutive model
adopted are also shown.

Permeability
The coefficient of permeability
Coefficientof permeability, k (tn/s)
was assumed to vary with depth as
shown in Figure 4.30, where field F:$'$
data for both the Upper Lias Clay
and the London Clay are also
shown. This permeability,
together with the swelling
modulus from Figure 4.29 for
K,,=1.5, gives a coefficient of
swelling c,,=2.7m21year at 5 m
depth. Walbancke (1976) quotes
field rates of swelling equivalent Figure 4.30: Variation of permeability
to c.v=3.2m2/~ear
for the Brown with depth - assumed in the analyses
London Clay.
and laboratory data

Rate of softening
The rate at which a clay strain-softens post-peak is difficult to establish. A
reasonable assumption was made based on precedent from other analyses.
Parametric studies, performed as part ofthe present investigation, have shown that
this property, when varied over a realistic range, does not have a major influence
on the results obtained (KovaEevid
(1994)). One or more shear surfaces
75 are likely to form as peak strength is
mobilised in real plastic clays, and .;:
loss of strength post-peak occurs as a d 50 consequence of sliding on these g
surfaces. Such thin discontinuities are
not reproduced in the finite element
0
analysis, in which the minimum
o
5
10
15
20
25
30
Shear strain, y (%)
thickness of a rupture surface is
approximately half the thickness of an
0
0.05
0.1
0.15
element. This must be taken into
Displacement across a 0.5m thick layer, A (m)
(112 the thickness, T, of a typical element)
account when specifying the rate at
which strength is lost post-peak. The
shear stress-displacement plot for a Figure 4.3 1: Assumed stress-strain
shear zone of thickness T in simple and stress-displacement relationship
shear is shown in Figure 4.3 1.
in simple shear

150 / Finite element analysis in geotechnical engineering: Application

g
C-\

Typically, in the present investigation,

half an element is 0.5m thick. A
displacement scale equivalent to this 2
I
is also shown in the figure.
bFigure 4.32 shows a comparison g
between the stress-strain behaviour 4
observed in two unconsolidated 3
B
undrained triaxial tests on 265 mm
diameter samples of Brown London
Clay (Sandroni (1977)) and the
predictions made from the model
adopted. The comparison is
reasonable, although the generation of
pore water pressure during shear is
underestimated.

75
50
37 kPa

12 kPa

25

10

15

Axial strain, E, (Oh)

- Io

10

32s

4.5.4.3 Finite element

analyses
Figure 4.32: Measured and
A typical finite element mesh for a
predicted behaviour of Brown
10m high, 3:1 slope is shown in
London Clay in triaxial tests
Figure 4.33. The boundary conditions
and method of analysis were the same
as in the 'non-softening' analyses presented in Section 4.3.2. However, greater care
had to be exercised when collapse was approached in the 'softening' analyses, as
discussed in Section 4.5.3.

Fixed boundary: impermeable

0

Scale

20m

Figure 4.33: Typical finite element mesh

4.5.4.4 Results of a typical analysis

Results are now shown from a typical analysis of a 10m high, 3:I slope, with a
value of the initial coefficient of earth pressure at rest, K,,=1.5. This slope has a
conventional long-term factor of safety of 1.4, based on a limit equilibrium
calculation with peak soil strengths (cP1=7kPaand yl,'=2Oo).
Figure 4.34 shows predicted vectors of current (incremental) displacement 9
years after excavation, and 14.5 years after excavation, just before collapse is
predicted. The vectors show the current mechanism of movement and indicate the
eventual collapse mechanism. The absblute values of the displacements are of no

Cut slopes / 1 5 1
significance. The figures show that the outer part ofthe slip surface has developed
after 9 years, and that the complete slip has developed after 14.5 years.
Strain wfbiing starts when E:=5%.
It is complete when E,!=20%.

'

a) 9 years ailer excavation

---

Rupture surface at peak

Rupture surface between peak and residual
Rupture surface at resi

scale

Scale

I
h) 14.5 years after excavation just before collapse
0

a) 9 years after excavation

;5m

25m

b) 14.5 years after excavation just before collapse

Note: Accumulated deviatoric plastic strain during
excavation is less than 5%

Figure 4.35: Contours of

accumulal'ed deviatoric plastic
strain, E/

Figure 4.34: Vectors of

incremental displacemen ts

j
a) 9 years after excavation 0

a) 9 years after excavation

Scale

2:m

h) 14.5 years after excavation -just before collapse

8
water content
Note: In a saturated clay of ~ 1 8 . kN1m1and
32%, a volumehic strain of 1.0% is equivalent to a water
content change of 0.7%.

Figure 4.36: Contours of

volumetric strain, 8 ,

Scale
0

25m

Pressure scale 2OOkPa

b) 14.5 years after excavation just before collapse

Figure 4.37: Contours of pore

water pressure

Contours of deviatoric plastic strain, EJ' (see Equation 4.5) are given in Figure
4.35. The 5% and 20% contours represent the start and end of strain softening. A
horizontal shear zone propagates from just below the toe of the slope, in a manner
similar to that observed in the field by Burland et al. (1977). The strength acting
on this base shear rapidly drops from peak to residual as the rupture zone develops.

152 / Finite element analysis in geotechnical engineering: Application

Figure 4.36 shows the volumetric strasns which develop due to swelling. The
strains within the slip are quite large, and are equivalent to a change in water
content at the toe of the slope before collapse of the order of 1-2%. The analysis
predicts an abrupt change in volumetric strains across the rupture surface.
Pore water pressure development is shown by contours in Figure 4.37, with
some vertical profiles on which both changes towards equilibrium and equiIibrium
pressures are shown. In the middle of the slope, pore water pressure changes due
to excavation are almost constant with depth. For this value of K, they change by
an amount almost equal to the weight of soil removed from the slope above. At the
crest of the slope the reduction is nearly linear with depth. It varies from 20kPa to
30kPa due to lateral stress relief, although there is no removal of weight at this
section. Below the toe of the slope, high Lateral stresses remain after excavation.
The reduction in mean total stress is less, and the reduction in pore water pressure
is significantly less than the weight of soil removed. The interaction between
changes in total stress and pore water pressure is clear. After 9 years, the pore
water pressures at depth have reduced due to undrained unloading, as the slope
deforms outwards and releases the locked-in lateral stresses, rather than increased
due to swelling.
Residual strength has been reached along the complete horizontal part of the
rupture zone (see Figure 4.35). The amount of progressive failure is substantial,
with the residual factor R=0.623, i.e. the average strength has dropped 62.3% of
the way from peak to residual. The residual factor R is a measure of the difference
between the average field strength at collapse and the strengths measured in the
laboratory: R=(<- ?)l(?,-F),where F is the average shear stress at collapse and
r,, and T, are the average peak and iL=
residual strengths at the current p
normal effective stress along the 2
eventual rupture surface (Skempton g
(1964)). The average operational
025.
strength along the eventual rupture
surface at collapse is given by G
QK1=0.328, which is equivalent to g
cl=O, q1'=18.2". The average pore 4 0
S
10
IS
20
water pressure ratio on the final
Time since excavation, t (years)
rupture
surface has increased from .g.
O O
r,,*=0.06 after excavation, to
r ,,*=0.247at collapse. The long-term
equilibrium pore water pressure is
0.4
equivalent to c8=0.375.Thus, on the .g
n
rupture surface, the average pore
Figure 4-38:Softening clay;
water pressure has changed 60% of
development of al F,' on the
the way towards final equilibrium.
Figures 4.38a and 4.38b show the rupture surface and bl mid-slope
displacement, 6,, with time
change with time of the predicted.

,,

Cut slopes / 153

c*,

average pore water pressure ratio on the rupture surface,

and the horizontal
displacement at mid-height of the slope, 6,. The average pore water pressure has
got close to the value at collapse, before the rupture surface at the base of the slide
starts to form. The mid-height displacement then starts to accelerate. Swelling then
involves reducing lateral total stress, rather than increasing pore water pressure.
The coupled effect of the reducing lateral total stress causes a small decrease in
average pore water pressure, although the clay is continuing to swell. There is only
a relatively small increase in pore water pressure in the five years preceding
collapse. Thus the analysis implies that the measurement of pore pressure in the
field is a poor way of monitoring risk of collapse. Final collapse is quite abrupt,
without significant warning from accelerating displacement.
The predicted deformations are consistent with the limited precedents from
field observation of London Clay slopes. The best recorded data are of the Kensal
Green retaining wall (Skempton (1977)), which moved nearly 0.5 m before failing
some 29 years after its construction. The movements were about 100mm after 26
years.

4.5.4.5 Effect o f coefficient o f earth pressure a t rest

In order to study this effect, 6
analyses were performed for a 10m i~:
high, 3: 1 slope, varying the value of
K,,(=1.0, 1.25,1.5, 1.75,2.0and2.15,
025
the last value is approximately equal
to K,,, the passive coefficient of earth C
O
pressure). The variations of average
pore water pressure ratio and midheight horizontal displacement with 4 -0.25
time from these analyses are plotted
in Figure 4.39. It may be noted that 3
Time since excavation, t (years)
the 'ultimate' equilibrium pore water ,- o O --..!0-----_20
.~.=1=10
pressure ratios shown on this plot are
from the same steady state seepage
0.5
solution, but differ because the 3
rupture surface for each analysis is
K.=Z 15
different. This plot shows that the
time to collapse is strongly dependent
Figure 4.39: Effect of K,;
on K,, and reaches a minimum of
development of a) 7', on the
about 13 years when K(, is about 1.25. rupture surface and bl mid-slope
Maximum pore water pressure ratios
displacement, 4, with time
at-collapse are similar, ranging from
r ,,*=0.225to Ca=0.275. However, two types of behaviour can be seen. First, when
K,,< 1.25, pore water pressures reach close to their maximum values soon after
excavation. Second, when K,? 1.5, there is a rapid equilibration of pore water
pressures over about 5 years, followed by a much slower increase until collapse.

-9
2

'

154 / Finite element analysis in geotechnicai engineering: Application

/-

This slow increase is accompanied by an 4=1.0 (RZ0.436)

initially increasing rate of displacement,
which then slows down. Rates of
displacement only accelerate again a few &= 1.25 (R = 0.583)
months before collapse. There is little
warning of collapse from observations of
either pore water pressure or movement.
Collapse is predicted when pore water
pressures are significantly lower than the
final long term equilibrium values.
(R =
The shapes of the rupture surface and
the residual factor, R, for each analysis
are shown in Figure 4.40. The rupture
surfaces are strongly influenced by the
value of K,, . They become deeper as K ,
increases from 1.0 to 1.75. As K ,
-+-.,
increases further, the back of the slip
moves out towards the slope. In all cases
only a short length of the rupture surface
+*
is at a strength between peak and
residual. The rupture surface at the crest
of the slip forms after the collapse event 1:" Rupture surface not formed
Rupture surface at peak
when K,, is low. The rupture surfaces are ---Rupture surface between peak
and residual
compared with each other and with field
Rupture surface at residual
surfaces in Figure 4.4 1. The field surfaces
have been geometrically scaled to the
same slope and height as the analyses. Figure 4.40: Rupture surfaces
predicted for varying K,
The predicted and observed rupture
surfaces are in broad agreement.
The amounts of progressive failure, as
expressed by the residual factor R,
predicted from all the analyses are
similar. However, the influence of
progressive failure changes as K , exceeds
l
1.75, as shown by the values of R in
Figure 4.42. Initially the influence
b)
increases with K,,, then it decreases,
finally it increases again. The reason for
this is shown in Figure 4.40.The length
Range of predictions 60m (a)
of the horizontal rupture surface which
develops from the toe of the slope
increases with K,,. With high values of K, Figure 4.4 1: a) Predicted rupture
it extends beyond the crest of the slope. surfaces; bl comparison with
Then the back of the final rupture surface field observations in London Clay
--L-

Cut slopes 1 155

develops subsequently closer to the
slope, and the inner part of the horizontal
rupture surface is not incorporated in
the final rupture surface. As a
consequence, part of the horizontal
stress in the zone in which the back of
the slip finally develops is relieved
before collapse, and the amount of
progressive failure is reduced.

23

1
0.5

1.5

2.5

Coefficient.of earth pressure at rest, K,

Figure 4.42: Change in residual

4.5.4.6 Effect of surface

factor with the initial K,
boundary suction
Analysis of a 3: 1 slope, 10m high, .
at equilibrium:
with a K,,=2 and with the surface '"1:
'i;'=0.38
suction raised to pf,=-2OkPa was 2
steady seepage aft2250 years
r 4.30
performed. Such an increase in
union 20Wa
_ _- no_softening
_ _ - - -:- - - - -.
S
&
suction 2OkPa
suction might be produced by the g
management of slope vegetation. The
6b
sb
160
results may be compared directly with 2 O
Time since excavation, r (years)
the analysis in which p#=- lOkPa. 2
:
Figure 4.43 shows how the average 4 -0.25- a)
%me since excavation, t (years)
pore water pressure ratio on the
rupture surface and the horizontal
o
---- ...........................
displacement at mid-height develop in
Sudan 2OkPs - no oa&Zig
8
the two analyses. Collapse with the 9 0.5
urfaec sunion 2 0 ~ a
increased boundary suction occurs 4
Surface suetion lO@a
after 100 years. Also shown are
results from an analysis in which no
strain softening was modelled. No
~i~~~~
4-43; ~
fof surface
i
~
collapse occurred, and the changes in suction;deve,opment of a, 7,* on
behaviour as the base shear develops the rupture surface and b) mid-slope
about 5 years after excavation were
d;sp/acement, 6, with time
eliminated.
The rupture surface for the analysis withpf,=-20 kPa was slightly deeper than
that for the analysis withpfh=- 10 kPa. Also, the average pore water pressure ratio
at collapse on the rupture surface was lower and, to compensate, the amount of
progressive failure was higher. However, in spite of more progressive failure, the
small increase in boundary suction (from p@=- lOkPa to pfi=-20 kPa) improves
stability substantially. Extrapolation indicates that a further increase in boundary
suction to pfi=-30 kPa would be just sufficient to prevent collapse.

,,,

8'

4.5.4.7 Effect of slope geometry

Two analyses were performed to examine the stability of2: 1 and 2.5: 1 slopes, 10m
high, with K,,=2.0 and p,,=-lOkPa. Results may be compared with the

156 1 Finite element analysis in geotechnical engineering: Application

corresponding analysis of the 3:l i$ O5 slope. The development of pore water .$

slope3 I
pressure and deformation with time
are shown in Figure 4.44. Both slopes
fail in less than 10 years, with
relatively low average pore water
pressure ratios on their rupture
surfaces. There was slightly more 4 -025- a)
progressive failure than for the 3:l
Time since excavation, t (years)
slope.
Analyses of a 4: 1 slope with high
K,, showed similar behaviour to the
3: 1 slope. They indicated that a 4:l 4
slope 3 I
slope was nearly stable when 10m
high, even in the very long-term, but
a 15m high slope failed with pore
Figure 4.441 Effect of
water pressures some way from the inclination; development of a) i,' on
predicted equilibrium values. While
rupture sumce and 61
predicted times to collapse are very
displscement,
with time
long (i.e. greater than l00 years), such
that the eventual collapse may be considered rather academic, such slopes may be
vulnerable if geological features, such as sand or silt layers, within the slope lead
to more rapid swelling.
Analyses of 6m high slopes were more stable. While the 3: 1 slope was just
stable in the long-term, the behaviour of the 2:5 and 2:1 slopes was similar to that
of the 3: 1 slope, IOm high. Equilibrium long-term pore pressures were decreased
and the amount of progressive failure was slightly reduced.

4.5.4.8 Effect o f surface cracking

Shallow slides have been commonly observed on both road cuttings and
embankment slopes (Peny (1989)). No such slides were reproduced by analyses
of cutting slopes with constant surface boundary conditions and soil properties.
Surface effects can involve cyclic shrinkage and swelling with down-slope
movement due to non-recoverable strain, cracking and, possibly, loss of strength.
Calculations made by limit equilibrium methods of analysis for embankment
slopes which include high pore water pressures, such as those generated in a
saturated cracked zone (Crabb and Atkinson (1991)), have shown that slipping can
only be explained if the operational strength is much lower than the peak strength
of the fill. Analyses are presented here which examine whether this strength
reduction could occur due to progressive failure. The analyses were similar to those
described above, with the exception that a higher permeability zone due to
cracking was incorporated on the slope surface.
Analyses were performed on 3:l cutting slopes, 10m high, with a high
permeability zone approximately 2m thick (three rows of finite elements, see

Cut slopes / 157

Figure 4.33). Both high and low values of K, (2.0 and l .O) were examined, with
surface boundary suctions of p,,,=- l0kPa and zero.
The permeability adopted for the 'cracked' layer was k=lOgmls. This is the
same as the permeability of the 'uncracked7 clay at the top of the slope, and is
seven times higher than the 'uncracked' clay at the bottom of the slope. The layer
only produces a discontinuity in the pore water pressure distribution during
swelling.
Nearly all analyses showed deep-seated sliding, similar to that shown by the
equivalent analyses without the 'cracked' zone. Collapse was significantly
accelerated by the permeable surface layer, which promoted swelling. A slight
increase in progressive failure due to the less uniform seepage pressure distribution
induced by the permeable layer was also observed.
Only the analysis with Ko=l.0 and zero boundary suction developed a shallow
slip shortly after excavation. Figures 4.45 and 4.46 show the development of the
slip with time. It can be seen that a thin superficial layer swells quite quickly and
fails by progressive failure from the bottom up. It seems that this mechanism is
inhibited by a high initial value of K,, apparently because of the early formation
of a deep-seated rupture surface and the stress relief associated with it. It reduces
subsequent swelling and the potential for progressive failure in a superficial layer.

a) 0.5 years after excavation

d) 1.5 years after excavation

b) 1.0 year after excavation

e) 1.55 years after excavation

c) 1.25 years after excavation

e) 1.6 years after excavation

0
l

Scale

25m
l

Figure 4.45: hcremental displacement vectors

showing the development of the superficial slip

158 / Finite element analysis in geotechnical engineering: Application

a) 0.5 years after excavation d) 1.5 years after excavation

b) 1.0 year after excavation

e) 1.55 years after excavation

c) 1.25 years after excavation f ) 1.6 years after excavation

Scale

2qm

Figure 4.46: Contours of accumulated deviatoric

plastic strain, E / , showing the development of
the superficial slip

The predicted time-scale for the shallow slip to develop (i.e. 1.6 years) is
shorter than is usually observed in the field. However, the prediction involves a
monotonic swelling process with zero boundary pore pressure (equivalent to winter
conditions), with cracks already developed. The time scale in the field may be
controlled by the time taken for cracks to develop, and by the delay before the
slope is subjected to a long 'wet' winter.

4.5.4.9 Effect of subsequent changes to slope geometry

Increases in motorway traffic often require widening of existing motorways. In the
case of a motorway cutting, this is usually achieved by removing its toes and thus
forming steeper overall slopes. Such an approach is attractive because no extra land
is required. The local stability at the toe is often provided by a form of gravity
retaining wall. Walls formed of reinforced earth have become increasingly popular.
The effect ofwidening a motorway cutting in Brown London Clay by removing
the toe of a 4: 1, 10m high slope, to form a 3: 1 slope overall, at different times after
initial construction, has been examined (see Figure 4.47). The typical finite
element mesh shown in Figure 4.33.has been modified in order to account for

Cut slopes I 159

excavation of the toe and formation of
a concrete gravity wall. No attempt
has been made to model formation of
the more widely used reinforced earth
type retaining wall. Such modelling is
complex and in this case notjustified, Figure 4-47:Scheme for motorway
because it would not have a
widening
significant impact on the results of the
analyses. An initial in-situ value of K,=1.5 was assumed in all analyses. A surface
boundary suction ofp,*=- l0 kPa was assumed initially, although the influence of
an increasing surface suction due to the management of slope vegetation was also
investigated. An efficient drainage system behind and beneath the retaining wall
is needed, particularly when constructing clay filled reinforced soil slopes. In the
analyses described here it was assumed that the concrete retaining wall will drain
the surrounding soil. This was achieved by specifying a zero pore water pressure
condition at the wall-soil interface.
The results of the analyses are
equilibriumvaluer of
~'-0.415
presented as before, in terms of the iz 0.5
wauiosta~~ed
50 yean
change with time of the predicted 'g
afin sxcavation
Wall insfalled
0.25
average pore water pressure ratio,
afier steady
rfatc condition
and the associated mid-slope
Wall insralled immediately
in reschcd
ow8'Cc:P"""
horizontal displacement, 6, (see
50
75
loo
125
Figure 4.48). Wall installation was g
Tie since excavation, t (years)
modelled at four different times after G
a)
slope formation: (i) immediately after 4
excavation, (ii) 15 years after
m
Time since excavation, t (years)
excavation, (iii) 50 years afier g
excavation, and (iv) after a steady
O ----.
WaU insl~lled
Wall instsllsd l5 ycan
sRcr steady
O 2
state condition is reached. Widening
swc condition
-. '.-.
---.-._____
j:rcachcd...
h
greatly accelerates delayed failure (a
0.5 ~22;;
4: 1, 10m high slope without widening .g
eR.rcxcavation
Wa11'nsfa'1ed50years
a&
excavation
n
b)
is nearly stable, see Section 4.5.4.7)
and may lead to immediate failure, if
done 50 years or more after initial Figure 4.48: Effect of widening;
development of a) 7," on the
excavation. However, the amount of
progressive failure as quantified by rupture surface and 6 ) mid-slope
displacement, 6, with time
the residual factor, R, remains about
the same as before.
If widening is done a short time after excavation (e.g. 15 years), stability may
improve to some extent as a result of the additional depression in pore water
pressures due to toe excavation (the immediate increase in pore water pressure due
to wall construction is localised and negligible). However, eventual failure is
inevitable. It can be suppressed if the surface suction produced by, for example,
management of slope vegetation is increased (see Figure 4.49). The increased

""7

c*,2

,:

,:

160 / Finite element analysis in geotechnical engineering: Application

surface suction at the slope behind the

wall will also eliminate delayed
failure when the wall is installed
immediately after slope excavation
(see Figure 4.50). Note that no
increase in suction in front of the wall
has been modelled in the analyses.
The increasing surface boundary
suction reduces the equilibrium values
of the average pore water pressure
ratio in the slope and the amount of
slope movement during swelling.
Although difficult to achieve in
practice, high values of the average
surface boundary suction (in excess of
30kPa) may
the base shear
formation, if not completely, then for
a substantial period of time (see
Figure 4.50a).

ii

.g
"

-3

Time since excavation, t (years)

wall iortalled IS yea,x

0.75

Figure 4-49: Wall installed after 15

years; development of a) 7,' on the
rupture surface and b) mid-slope
displacement, 6, with time

4.5.4.10 Further discussion

The delay predicted for collapse and
the
- average pore water pressure ratio,
r ,,',on the rupture surface at collapse,
for the various analyses performed,
are shown in Figure 4.5 1. Also shown
on this figure are field observations
(Chandler (1984a), ( 1 984b)). The
agreement between the finite element
predictions and field measurements is
good. Considering the difficulties in
Time since excavation, r (years)
estimating the coefficient of '3
permeability, k, on which the time to
collapse directly depends, the times to
collapse predicted by the analyses are
consistent with field observations,
particularly when higher K,, values
(which are likely to be more
Figure 4.50: Wall installed
representative of stiff clay) are
immediately;
development of a) 7,'
assumed.
on
the
rupture
surface and bl midThe average strength at collapse
predicted by the analyses is shown in slope displacement, 6, with time
Figure 4.52, together with the peak and residual strengths. Also shown are the
operational strengths deduced from back analysis (using limit equilibrium method)

Cut slopes 1 161

of slips by Chandler and Skempton

(1974). The strengths from the finite
eEement analyses are generally lower
than those of Chandler and Skempton
(1974). However, the finite element
analyses typically predict deeper
rupture surfaces than the critical
surfaces predicted by limit
equilibrium analyses. As the rupture
surface becomes deeper, the amount
of progressive failure increases, and
the average strength, expressed as a
stress ratio, decreases. At low values
of K,, the relatively shallow rupture
surfaces approximate to the critical
surfaces predicted by limit
equilibrium analyses. Then the
average strengths predicted by the
finite element analyses agree with
those obtained by limit equilibrium
back-analysis.
Figure 4.53 shows values of r,'
on the eventual rupture surface, after
excavation, at collapse and after full
equilibration (steady state), for all the
analyses, plotted against slope
inclination. The pore water pressures
required for collapse increase with
decreasing slope angle, but SO do the
equilibrium pore water pressures. The
analyses show that collapse is
sensitive to the magnitude oft-he pore
water pressures in the slope, and thus
to the equilibrium pore water
pressures to which these pressures
tend. This plot could be used to assess
the effectiveness of any remedial
drainage measures. For example, the
effect of increasing the surface
boundary suction frompfi=- 1OkPa to
pf,,=-20 kPa for a 3:l slope, 10m
high, is to reduce the long term steady
state value of
to just above the
'collapse' range (see Figure 4.53).

c*

;i

,,,

g , ,.S

0.3

"

O o 0 a0 p 0 s

20.2

gj

0;

Slope 2.0:L - predicted

o
a slope
Slope 23.0:1
. 5 : ~-predicted
-predicted

O$

[ O.'-do
g

50

Field
slope obse~ations
4.0: I -predicted

IW

150

200

Time to failure, $(years)

%.o.l.0.2-*
Y

Figure 4.5 1: Predicted average pore

pressure ratio on the rupture
surface vs. time to collapse, and
field observations
Q100

.$

C!

slope Z.O:I-predicted

o
A

sslop
l o p 2.21
3 . 0 : ~-predicted
predicted
Slope 4.0:L -predicted

3
*

50

Q
3

4
O0

50

100

I50

Average normal effective stress, 2"

(kPa)

Figure 4.52: Average operational

strength at collapse predicted by FE
analyses and limit equilibrium
analysis

i ~0.5=0.41

8
E
k
"

F
4

0.3
o.2-

0.1 '-1

-0.1 -

a.2.

-O".0.4 L

after excavatron

steady state

Figure 4.53: Variation of average

pore pressure ratio on the rupture
surface with slope inclination

162 / Finite element analysis in geotechnical engineering: Application

This implies that, although the time to collapse would increase, collapse would still
occur (see Section 4.5.4.6). For collapse to be prevented, the surface suction would
have to be increased still further, so that the long-term value of r,' fell below the
'collapse' range indicated in Figure 4.53.
Figure 4.36 shows a sharp discontinuity in volumetric strain across the
horizontal rupture surface of 2-3%,equivalent to a change of water content of 12%. Such a discontinuity could be detected by site investigation. This would be a
potential way of examining the stability of old slopes. As can be seen from Figure
4.35a7the basal part of the rupture surface forms quite early on. Its formation does
not indicate that collapse will eventually occur, but its presence with continuing
movement on it (which could be
monitored) indicates decreasing
stability. The absence of movement
would indicate stability.
BIOWII London Clay
Figure 4.54 shows a section
through a slip on the M11 near
Loughton in Essex, in a slope 18
Water
Liquid
Plastic Remoulded
years old. Results from tests on
content l i t
limit quick undraiied
shear strngth
samples taken from just above and
W % WC%
W,%
S, kPa
below the rupture surface are given
SampleA 33.2
70
33
26
on the figure. There is nearly a 4%
sampleB 29,4
75
32
62
change in water content and a two
fold change in remoulded undrained
strength across the rupture surface. Figure 4.54: Slip on the east side of
the M I I near Loughton, Essex
This observation is consistentwith the
( 19931
numerical analysis.

4.6

Construction of cut slopes under water

Excavation of cut slopes underwater can be problematic. For example, consider the
situation shown in Figure 4.55a, where a trench is to be excavated in the sea bed.
At the beginning of the analysis, the initial stresses and pore water pressures in the
soil are in equilibrium with the sea level (the mean sea level is often assumed, but
its fluctuation can, in principle, be accounted for). Excavation is then simulated by
sequentially removing rows (or blocks) of elements. However, care must be taken
to properly account for the water pressures applied to the newly excavated soil
surface.
When excavating the first row (row 1) of elements shown in Figure 4.55b, the
process essentially also removes the water pressure acting on the original ground
surface between points A and H. It is therefore necessary to simultaneously (i.e.
over the same increments as excavation occurs) apply a boundary stress over the
newly excavated soil surface ABGH, to represent the water pressure acting on it.
This process must be repeated as each row of elements is removed.
Although the process described above is logical, it is cumbersome and often
r
This arises as many programs
difficult to achieve with some c o m p ~ e software.

Cut slopes / 163

V
do not correctly account for the stresses
applied along the previous excavated soil
A
U
surface when further elements are
removed. For example, when excavating
row 2 in Figure 4.55c, the software must
a)
specifically remove the nodal forces
associated with points B and G, which are
v
due to the boundary stresses representing
the water acting over the surface BG, as
well as simulating excavation of the
elements in row 2 and the boundary stress
associated with the water acting over the
newly excavated surface BCFG. Note
that it is not necessary to specifically
v
remove the nodal forces due to the
boundary stress representing the water
acting over the old excavation surface BG
at any other nodes between B and G, as
these are implicitly accounted for in the
c)
simulation of the excavation of the
Figure 4.55: Scheme far
elements in row 2. The reason why the
under
nodes at points B and G require special
consideration is that part of the nodal force due to the water pressure is applied to
soil which is not to be removed. This process requires complex book keeping and
consequently many software packages neglect to account for this. The error
involved increases with the depth of water above the original ground surface.

4,7
1.

2.

Summary

When analysing cut slopes, it is important to distinguish between different

types of materials in which they are cut. Granular soils behave in a drained
manner. Clays of low permeability are largely undrained during excavation.
Coupled consolidation/swelling is necessary for intermediate and long term
situations. The choice of constitutive model depends on the material type.
The analyses of2: 1 cut slopes, 1Om high in stiff 'non-softening' clay all show
well defined rupture surfaces at collapse. The average pore water pressures
were independent of K, . They are
on the rupture surface at collapse,
influenced slightly by any dilation that occurs when the soil fails. The
depression in pore water pressures due to dilation acts as a partial brake on the
rate of post-collapse movement on the slip surface.
The values of r," at collapse predicted by the analyses with zero dilation are
in good agreement with those obtained from limit equilibrium analysis
assuming a circular rupture surface. However, the position of the rupture
surface predicted by the numerical analysis differs from the critical slip circle.

c',

3.

164 / Finite element analysis in geotechnical engineering: Application

The time to collapse increases with KO.

Partial drainage can occur in the short term when excavating in sofi clays. For
shallow excavations this may well be due to high initial stiffness. Partial
drainage can also operate in deeper excavations, when soft clays are of
relatively higher permeability andlor of moderate thickness, due to the
presence of sandy or silty layers which accelerate pore pressure equilisation.
Progressive failure refers to the non-uniform mobilisation of shear strength
along a potential rupture surface. If a brittle soil is loaded non-uniformly,
some elements of soil will reach peak strength before others and a rupture
surface will begin to develop. With further loading, the post-peak strains
along the rupture surface increase, and the strength reduces from peak
towards residual. At collapse, part of the rupture surface has formed and lost
strength post-peak, and a part has not yet formed. Thus the average strength
of the soil mass at failure must be smaller than the peak strength and greater
than the residual strength.
The parametric studies ofthe 3: 1 slope, 10m high in stiff 'softening' soil, such
as London Clay, show that the probability of collapse is relatively insensitive
to the value of K,, assumed. The value of K,, strongly influences the location
of the shear surface and the time to collapse. There is a slight increase in the
amount of progressive failure with increasing K, up to a critical value, after
which the base rupture surface extends beyond the final inclined rupture
surface, which relieves the stress in the slip prior to its final formation and
reduces the amount of progressive failure.
The rupture surface has four parts, namely: (i) the horizontal base shear,
which forms early and reaches residual well before collapse; (ii) a short
transition length between peak and residual; (iii) an inclined back surface, in
which rupture has occurred but where strains are not sufficient for strainsoftening to start; and (iv), when K, is low, an inclined section near the
original ground surface on which rupture only occurs post-collapse.
The deep-seated nature of slips predicted at the high values of K,, , usually
present in in-situ London Clay, is typical of field experience. The analyses
show that, at the higher values of K,, , the rupture surface is developed by
progressive failure at a greater depth than the critical surface which would be
determined by limit equilibrium analysis. As a consequence, the predicted
operational strength on the rupture surface is less than that which would be
determined for the same slope with the same pore water pressures by limit
equilibrium analysis, using a search technique to find the critical surface.
There is considerable variability in the predicted time to collapse ofthe slopes
analysed, bearing in mind that a constant permeability is assumed. The
parametric studies on the 3: 1 slope predict time to failure between 1 1 and 45
years. The steeper slopes fail in less than a decade. The flatter ones may fail
after more than 100 years. The field experience also shows considerable
variability in time to collapse. There will be seasonal variation in superficial
pore water pressures, not represented in the analyses, superimposed on the

Cut slopes I 165

general trend. Thus collapse will tend to occur in winter, when surface suction
is zero (see e.g. Chandler (1984b)).
The surface hydraulic boundary condition has a strong effect on stability. The
analyses show that an increase in surface suction from l0kPa to 20kPa goes
more than half way to stabilise the 3: 1 slope. Surface pore water pressures can
be reduced by increasing evapo-transpiration through the controlled use of
vegetation, or by surface drainage which reduces pore water pressures below
the depth of the drains.
The analyses show, beyond reasonable doubt, that the mechanism causing
deep-seated delayed slips in these slopes is progressive failure promoted by
swelling. They enable the importance of the controlling variables to be
established, and the effect of possible stabilising measures to be evaluated.
The new and sophisticated techniques of numerical analysis have an
encouraging ability to reproduce observed behaviour, provided that realistic
input data is used. No assumptions need then be made about behaviour
mechanisms, as these are established by the analysis.
No shallow slips were predicted in the analysis in which uniform soil
properties were assumed. However, shallow slides can be reproduced by
differential strains and progressive failure at the base of a superficial layer of
clay of higher permeability due to shrinkage cracks, although this is not the
only possible reason for such slides. This mechanism was inhibited by high
initial values of K,, , when early formation of a horizontal rupture surface
probably reduced the strain energy stored in the superficial 'cracked' layer.
Care should be exercised when widening amotonvay cutting by removing the
toe. The analyses show that widening greatly accelerates delayed failure,
although the amount of progressive failure remains about the same. The
increasing surface boundary suction produced by management of slope
vegetation, or surface drainage which reduces pore water pressures below the
depth of drains, may halt development of slips in the longer term.

Embankments

5.1

Synopsis

Many attempts have been made to predict the behaviour of various types of
embankments. The finite element method is a powerful tool for such predictions.
This chapter deals with applications of the finite element method in the analyses
of this type of structure. Examples of modelling are presented and discussed.

5.2

Introduction

Geometrically, embankments are very simple structures. However, the materials

involved may vary widely, involving construction on foundations ranging from
soft clays to hard rocks and fills ranging from compacted clays to rockfills.
They are often used to retain water, varying from simple flood protection
embankments to large embankment dams, hundreds of metres high and involving
tens of millions of cubic metres of fill. They also may be used to support structures,
like roads and railways.
Embankments are usually placed by earth-moving plant, in layers of varying
degrees of mechanical compaction, although they may be loose dumped or even
placed hydraulically. Quite often embankments of compacted fill are used to retain
hydraulic fills as in land reclamation and waste disposal.
The mechanisms involved in the embankment behaviour can be quite simple,
as in a non-water retaining embankment of permeable granular fill on a "strong'
foundation. On the other hand, they can also be much more complicated, as in the
case of low permeability fills being placed on a 'weak' foundation, where pore
pressure changes are generated by total stress changes (short-term undrained
events) and by equilibration during the processes of consolidation and seepage in
the longer term.
It is apparent that the finite element analysis of stresses and deformations in
embankments is an exceedingly complex problem. Obviously, all types of
embankments cannot be covered here, neither can all the various mechanisms
involved. Thus a selection has been made to present examples of finite element
analysis of various types of rockfill dams, earth embankments and embankments
on soft ground.

Embankments 1 167

5.3

Finite element analysis of rockfill dams

5.3.1 Introduction
According to the nature and position of the watertight element, there are two main
types of rockfill dam: (i) that containing a relatively impervious internal earth core,
either wide or narrow, central or inclined, and (ii) that with an upstream impervious
membrane, made either of concrete or asphalt. Usually rockfill and clayey fill are
involved in the former and just rockfill in the latter.
Being derived from rockfill only, dams with an upstream impervious membrane
are simpler to analyse and therefore an example of finite element analysis of this
type of dam will be presented first. This membrane is usually thin and flexible and
has little bearing on rockfill deformation during first reservoir impounding and
subsequent operation. First, some general features involved in embankment dam
modelling will be briefly discussed.
5.3.2 Typical s t r e s s p a t h s
A great variety of stress paths occurs in embankment dams (and their foundations).
They are often accompanied by a large rotation of principal stresses and may be
particularly complex during first reservoir impounding and subsequent operation.
It is difficult to reproduce all these real stress paths, even using sophisticated
laboratory equipment such as the hollow cylinder or true triaxial apparatus.
Nevertheless, it is common to
describe the behaviour of fills using
standard laboratory tests. This is
particularly SO for rockfills. Because
?for oned~menstonal
")::;Edopug
compresston
of their coarse nature, the tests have to
be performed on large samples and
200
4w
600
P'
are therefore expensive and
labourious. Consequently, their
number and types are limited.
Sometimes both large triaxial and
oedometer tests are performed for
rockfills, but quite often only results
from one of these two tests are
available. Therefore, it is necessary to
establish which of these is more
representative to simulate rockfill 4
behaviour in an embankment dam.
It can be argued that the
200
400
600
oedometer test, providing a constant
P'
stress ratio path, R = aI1/a,'
= lIK,,
can better capture fill behaviour, at Figure 5.1: Stress paths for three
least during the construction stage.
types of rockfill dam
Although the principal stress ratio, R,
(after Charles 119 76))

168 1 Finite element analysis in geotechnical engineering: Application

is not ideally constant during dam

construction, in most cases its value
changes within a relatively narrow
limit and usually around the assumed
R = 1/K, value (Figure 5.1). That is
why oedometer tests, with R typically
between 2.0 and 3.0, give a
reasonable prediction of overall 4 4w
rockfill dam behaviour during F
P 200
construction. However, in spite of L
this, triaxial tests should not be
0,
,,,,,, ,, , ,, ,,
(a,'+o,')n(kPa)
neglected. Apart from defining the
rockfill strength envelope, these tests
provide a useful insight into the Figure 5.2: Stress paths for three
stress-strain properties of rockfills points along a line normal to the
too. Also, due to lateral yield, the
concrete face of a hypothetical
stress paths in the dam shoulders
rockfill dam
deviate from the K, line towards the
(after Mori and Pinto (1988))
strength envelope and can approach
those imposed by standard triaxial testing figure 5.2). It seems that, if rotation of
principal stresses is ignored, the typical stress path during dam construction is
somewhere between those imposed by oedometer and standard triaxial testing.
Ideally, any constitutive model should predict the observed rockfill behaviour in
both oedometer and standard triaxial tests (if both types oftest are available). Then,
it is reasonable to assume that the rockfill behaviour may be captured for any stress
path in between. How well the available constitutive models can achieve this is,
however, another question!

5.3.3
Choice of constitutive models
It has been repeatedly shown (see e.g. Duncan (1992)) that the most influential
factor in the finite element analysis of embankment dams is the modelling of the
stress-strain behaviour of the fill by an appropriate constitutive law. However, in
spite of the diversity of the stress-strain relationships being used, reasonable
agreement has usually been found when the results of finite element analyses
(typically movements) have been compared with field observations. This is not
surprising bearing in mind the fact that most of the analyses were done after the
field measurements had been made, resulting in after-the-event or so called 'Class
Cl' predictions (according to Lambe (1973)).
A review of different constitutive laws used in the numerical analysis of
embankment dams can be found in publications by Naylor (1991a) and Duncan
(1992). Both elastic and elasto-plastic formulations are available. In the following
the most representative models will be reviewed briefly, with emphasis on their
validity and parameter derivation. Some of them have already been described in
Chapters 5 and 7 of Volume 1 of this-book.

Embankments 1 l69

5.3.3.1 Linear elastic analysis

Soils are far from being either linear or elastic (see Chapter 4 of Volume I).
Nevertheless, because of simplicity, geotechnical engineers have often
characterized the behaviour of real soil using idealized models of linear isotropic
elasticity. Reasonable results can only be obtained for conditions far away from
failure, when a significant factor of safety operates. These conditions usually
prevail in rockfill dams, and thus it is not surprising that linear elastic analyses
have been successful in a number of cases.
Only two elastic constants are needed to characterise the stress-strain behaviour
of isotropic linear elastic materials. They are usually Young's modulus, E, and
Poisson's ratio, p. In order to obtain reasonable values of stresses and particularly
displacements, it is essential to find the most suitable values of the above elastic
constants. A 'constant equivalent compressibility' method for determination of
Young's modulus using data from oedometer tests, was proposed by Penman et al.
(1 97 1). They showed that the internal distribution of vertical displacement during
construction of a thick, broad layer (one dimensional condition) possessing self
weight, can be predicted with little error by the use of constant Young's modulus
determined to give the correct final displacement of a point half-way up the
complete layer.

5.3.3.2 'POwer la W' models

These models were developed in an attempt to fit the data from the oedometer test,
which is sometimes the only available test for fills (e.g. rockfill). For granular
materials, the relationship between volumetric strain, E,,, and axial stress, g',',in a
one dimensional compression test can be described by a power law of the form
(Rowe (1971)):
E,,

1 pPalD

(5.1)

where p,, is atmospheric pressure in the same units as G',C and D are
dimensionless parameters. Skinner (1 975) used the above relationship to derive the
closed form solution for the settlement profile on the centre line of an embankment
of height H constructed of rockfill of bulk unit weight y:

where S is the settlement of a marker placed at a height h above the foundation and
C , = ClpuD.Settlements in this case are treated as one dimensional and vertical
effective stress is presumed to be equal to the overburden pressure (pore pressure
is zero). The former results in an underestimate of settlements, the latter in an
overestimate. The two effects are likely to be relatively small and compensating.
By differentiating Equation (5. l), the expression for the tangential constrained
modulus, E,', can be obtained. The tangential Young's modulus, E,, can be derived
using the value of Poisson's ratio, p, estimated from the K,, value (i.e.
p=K,,/(I+K,,)) from elastic theory, where K, is related to the angle of shearing

170 / Finite element analysis in geotechnical engineering: Application

resistance q' according to K , = 1 -sinpl, or determined directly from the oedometer

test as K,l=ur'lcu', by measuring the radial stress U,'. If a constant K,, value is
assumed (a frequently made assumption iFor soils), at least three model parameters
are required to represent loading conditions. Additional parameters are needed to
model unloading (see e.g. Naylor (1991~1)).

5.3.3.3 Hyperbolic model

As described in Section 5.7.4 of Volume I, this model is based on a hyperbolic
function which approximates the stress-strain curve in a conventional triaxial
compression test. Various modifications have been made to the original
formulation. Using a Mohr-Coulomb failure criterion, the following expression for
the drained tangential Young's modulus, E,, can be derived:

where R, is the stress ratio at failure with a value always less than unity, p,, is
atmospheric pressure, M, is a modulus number, and n is an exponent determining
the rate of variation of the initial tangential modulus, E,, with confining stress, G,'.
The curved nature of the failure envelope for granular materials (ci=O) can be
accounted for by using the following logarithmic expression for p':

where q,,' is the angle of shearing resistance for a confining stress u,'=p,, and Aq'
is the reduction in p' for a tenfold increase in IS,'.
Instead of modelling the tangential Poisson's ratio, p,, Duncan et al. (1980)
preferred to model the volume change behaviour of soils by an exponentional
relationship between the bulk tangent modulus, K,, and confining pressure, c,':
(5.5)
K, = M , P , ( ~ 1 P,)"
where M, and m are dimensionless mode%parameters with a similar function as M,
and n in Equation (5.3).
It can be noted that seven parameters are required to characterize the model. If
a stiffer response on unloading is to be modelled, the additional parameter M,, (an
increased value of M,) can be employed. All parameters can be evaluated using
data from conventional triaxial tests.
The hyperbolic model has been used extensively in the finite element analyses
of different geotechnical problems, particularly embankments. Consequently,
considerable experience with the model has been accumulated and this has enabled
Duncan et al. (1980) to compile parameter values for different fills taking into
account relative density (compaction), grading, particle shape and mineral
composition. This has been the main advantage of the model, especially when
applied to the analyses of rockfill dams, given the difficulties associated with the
laboratory testing of rockfills.

Embankments / 17 1
A nonlinear elastic model, similar in formulation to the above hyperbolic
model, has been used at Imperial College for numerical analyses of a number of
embankment dams (Wamza (1976), Dounias (1987), Kovaeevid (1994)). The
drained elastic parameters vary according to the following expressions:

E,, = H E,

(5.9)

where E and E,, are the drained Young's moduli on first loading and
unloading/reloading respectively, p is the Poisson's ratio, p, is atmospheric
pressure, and E,, , A, B, C, H, I, L are model parameters. The stress level, S,
represents the proportion of the shear strength mobilised at the current mean
effective stress, p', and varies from zero, when the state of stress is on the
hydrostatic axis, to unity, when it is on the failure envelope given by cohesion, c',
and angle of shearing resistance, p' (see Appendix 111.l).
Problems can be experienced with this type of model when deriving the model
parameters to satisfy both oedometer and triaxial test data. It is often not possible
to accurately model results from both types of test. Bearing in mind the above
discussion about typical stress paths, emphasis is often placed on fitting the
oedometer stress-strain curve. When results of analyses are compared against field
measurements, it emerges that the predicted settlements are in reasonable
agreement with the measured ones, but horizontal displacements are overpredicted
(see e.g. Potts et al. (1990), Kovatevid (1994)).

5.3.3.4 K-G model

It was recognized a long time ago that modelling of soils in terms of the bulk
modulus, K, and the shear modulus, G, had some advantages over the use of
Young's modulus, E, and Poisson's ratio, p. A particular form of this model has
been described in some detail in Section 5.7.3 of Volume 1.
In comparison with the hyperbolic model this model is simpler. Only five
parameters are required. For unloading an additional parameter is needed.
However, their derivation is somewhat 'messy' (Naylor (1991a)). As with the
hyperbolic model discussed above, it seems that a good fit to both oedometer and
triaxial test data cannot be obtained. This is in agreement with the previous
experience gained at Imperial College using Equations 5.6 to 5.9.
5.3.3.5 Elasto-plastic models
It is well known that elasto-plastic stress-strain relationships are capable of
modelling the behaviour of real soils more closely. Taking into account the

1 7 2 l Finite element analysis in geotachnical engineering: Application

influence of stress path on soil behaviour, elasto-plastic finite element analyses

should provide a better prediction of stresses and especially movements in
embankment dams. Unfortunately, in spite of the existence of different elastoplastic constitutive models, these have not been widely used in practice. One
reason might be the considerable complexity of such analyses. Another can be
found in the difficulties connected with the testing of fill materials, especially
rockfills.
Nevertheless, there have been attempts in the past to characterize the behaviour
of rockfill using elasto-plastic stress-strain relationships. The use of various elastic
perfectly plastic models gave, essentially, an elastic solution as no (or only small)
zones of plastic yielding were discovered. Significant factors of safety usually
operate in rockfill dams and the modelling of plastic behaviour at peak and postpeak, even with the inclusion of hardening or softening behaviour, is unnecessary.
Modelling such behaviour is only of importance in special circumstances when a
shear failure surface develops (see e.g. Potts et al. (1 990)).
The inclusion of plasticity pre-peak has usually been associated with the
modelling of dilatancy, i.e. the tendency of well compacted granular fills to
increase, or soft clayey fills to decrease, their volume on shearing. There is both
laboratory and field evidence that suggests dilatancy is unlikely to occur widely in
rockfill dams. Thus, the use of a complex elasto-plastic model to account for
dilatancy alone is not warranted. However, the modelling of the plastic behaviour
pre-peak has another, more subtle advantage which concerns the pattern of
deformation during yielding. Namely, in contrast to the elastic strains which are
dependent on the stress increment only, the plastic strains depend on the
accumulated stress as well. The implications of this was clearly demonstrated by
Naylor (1975) who analysed the behaviour during construction of a clay core
embankment dam using two different constitutive models: 'variable' elastic and
critical state elasto-plastic. He showed thatthe deformation behaviour was sensitive
to the type of model being used even though both models produce similar triaxial
test stress-strain curves.
There exist a large variety of elasto-plastic models that have been proposed in
order to characterize the stress-strain and strength behaviour of soils. Most of them
have been conceived from the concept of critical state soil mechanics. The Cam
clay model and its various modifications have been increasingly used in the finite
element analysis of various dams and embankments. Unfortunately, most of these
applications were associated with embankments on soft ground where the overall
behaviour is usually governed by the sof3 materials present in the foundation (see
Duncan (1992), (1994)). Very few examples of the application ofthe critical state
models to predict the behaviour of rockfill dams have been reported.
However, the suitability of critical state models to predict the behaviour of fill
materials is questionable. These models were originally developed for sedimentary
clays. Compacted fills have a behaviour pattern similar to that of sands. Thus, the
elasto-plastic constitutive models originally proposed for sands can better capture
the behaviour of compacted fills, even of clays. Rockfill can be considered as an
extreme granular material.

Embankments / l 7 3
Elasto-plastic models for sands usually separate the effects of consolidation and
shearing. During consolidation deformations are governed mainly by crushing and
yielding of interparticle contacts. During shearing to high stress ratios deformations
develop also due to sliding and rolling. To account for different plastic
deformations during consolidation and shearing, so-called 'double hardening'
models have been proposed. A double hardening model of the form proposed by
Lade (1977) has been described in Section 8.5 of Volume 1.

5.3.4

Layered analysis, stiffness of the simulated layer and

compaction stresses
Embankments are built up in
'Ghost'elemcnts
New laycr at lcasl
relatively thin horizontal layers.
,-\;-----.
one elemenl thick
Consequently, there will be a large
number of layers during the
construction of a large dam. The
Elements in foundation
limitations of computer modelling
require relatively thick layers to be
Figure 5.3: Finite element modelling
used in the idealization (Figure 5.3).
of fill construction
In the past, to provide an insight
into the errors involved, the closed form solution of the incremental analysis was
usually compared to the finite element 'layered' analysis using a one dimensional
model which represents either a soil column or a fill of large lateral extent. It was
concluded that as few as five layers is usually sufficient in embankment dam
analyses if a nonlinear elastic constitutive model is employed (see e.g. Naylor
(1991b)). However, the implication of this finding if an elasto-plastic model is
employed has not been considered. It will be shown below that significant errors
can occur if the layers are too thick when the effects of compaction are modelled,
even in the case of elastic constitutive models (linear or nonlinear).
The effect of compaction is to increase horizontal stresses in a layer as further
fill is placed. This effect typically becomes insignificant when 3 to 5m of fill has
been placed. To model this, the technique described in Section 3.7.9 of Volume 1
has been used. Construction of a new layer is essentially performed by a gravity
'turn-on' analysis. For a broad fill, the vertical stresses are equal to the overburden
pressure, whereas the horizontal stresses are determined by the value of K,, input
by the user.
To provide an insight into the effect of modelling compaction stresses in a
'newly' constructed layer on the settlement profile during dam construction, a
finite element 'layered' analysis using a one dimensional model (no lateral strain)
can been performed. A 50m high soil column is constructed in four, ten and twenty
layers respectively. The fill is modelled using the Lade's double hardening model
with the parameter values derived for the Winscar sandstone rockfill (KovaEevic
(1994)). In the first set of analyses the horizontal stresses in a 'newly' constructed
layer were determined by the value of Poisson's ratio p = 0.45 leading to K,, =

174 / Finite element analysis in geotechnical engineering: Application

50
$(l-p)=0.80. In the second set of
analyses the horizontal stresses were
reduced to be equal to K,, = 0.25 times
the calculated vertical stresses, where
40
K,,=0.25 is a typical K, value
predicted by the idealised, single
element oedometer test on the ,
sandstone rockfill using the Lade's
double hardening model (KovaCevid
(1 994)).
The results of this exercise are 8 20
presented in Figure 5.4. It can be seen
that the second set of analyses
(broken lines with open symbols)
l.
predicts settlement profiles which are
nearly independent of the number of
construction layers used. This is not
o
loo
200
300
400
500
so when the higher horizontal stresses
due to compaction were modelled as
Settlement, s (mm)
in the first set of analyses (full lines
&'ith shaded
the
Figure 5.4: Effect of number of
larger the number layers e m ~ l O ~ e d/a, yers and ,-ompact;on stresses on
the more the settlement profile
settlement of a
approaches to that predicted by the
second set of analyses.
Although the one dimensional model (no lateral strain) is agood approximation
of dam construction (at least in the central portion of the dam), and therefore is
very useful in determining the optimum number of layers in the finite element
analysis, it is important to recognise the particular influence of the stiffness of the
new layer in two dimensional analysis. This is due to the difference in bending
between one 'thick' simulated layer in the finite element analysis when gravity is
suddenly applied, and progressively occurring bending of several 'thin' layers in
the field. It was recognised that some stiffness between zero and the full in-place
stiffness would make the finite element layer equivalent to the real situation, and
some work has been done in this respect for both linear and nonlinear elastic
materials (see Naylor ( l 99 l b)). Bearing in mind the complexity associated with
modelling of compaction stresses (and, in the case of clayey fills, initial suctions),
no attempt has been made to include these considerations in the examples reported
here. Consequently, a negligible stiffness assigned to the layer of elements during
construction (see Section 3.7.9 of Volume 1) results in an overestimate of
deformation in the underlying layers. The effect of this when fills are modelled as
elasto-plastic materials is unknown, but is likely to be small.

Embankments / l 75

Example: Analysis of Roadford dam

5.3.5
5.3.5.1 Introduction
Roadford dam is a 41m high rockfill dam with an upstream membrane of asphaltic
concrete, constructed in the South West of the United Kingdom. A typical cross
section of the embankment is shown in Figure 5.5a. The rockfill is composed of
mudstone, siltstone and sandstone. Details of the design of the dam and of the
properties of the rockfill are given by
Virilson and Evans (1 990).
An important design detail of the
dam is the junction of the asphaltic
membrane and the concrete
inspection gallery at the upstream toe.
To minimise differential movements
in this region a zone of stiffer fill
(sand waste) was used immediately
against the inspection gallery (see
Figure 5.5b). A grout curtain
approximately 40m deep is located
below the inspection gallery.
Sand waste
Instrumentation to measure
vertical settlements and lateral
foundat~on
In
e tatio :
displacements of the dam, and to ?:S,:
~ettle~e~plates
measure the movement of the
asphaltic membrane in the vicinity of
the inspection gallery, was installed
Figure 5.5: Roadford dam:
during construction of the dam. The
a t i o n of some of t h i s a) Cross-section showing location of
instrumentation is also shown in instrumentation and
b) Detail of the upstream toe
Figure 5.5.

~;~~~$;;t;;~''~ter

5.3.5.2 Material parameters

Analyses of Roadford dam have been performed using two different constitutive
models. One of these was a 'simple' nonlinear elastic perfectly plastic model. It
incorporated a Mohr-Coulomb failure criterion (c' and v') as a yield surface. Nonassociated plasticity was assumed with a value of the angle of dilation v = ~ ' 1 4 .
However, the actual value adapted was of little significance since none of the fill
in the dam reached failure. Thus essentially only the nonlinear elastic part of the
model is active and the model is referred to as non-linear elastic. The elastic
parameters are defined in Equations (5.6) to (5.9). The other model was a
'complex' work hardeninglsoftening elasto-plastic model of the Lade's type, with
two intersecting yield surfaces (see Section 8.5 of Volume 1).
Large scale laboratory tests were carried out at the Building Research
Establishment (BRE) to determine the relevant properties of the low grade rockfill

176 / Finite element analysis in gecitechnical engineering: Application

and sand waste material (Charles and Watts (1985)). These included investigations
of compressibility in Im diameter oedometer tests and strength (and
compressibility) in 0.23m diameter triaxial compression tests. The data from these
tests have been used to derive the parameters for the two constitutive models (see
Tables 5.1 and 5.2) to give the best fit.
Comparison of predictions from the two constitutive models with results from
triaxial and oedometer laboratory tests for the rockfill are shown in Figure 5.6.
Inspection of this figure indicates that the fit from the Lade's model is good for
both the triaxial and oedometer tests. The fit for the nonlinear elastic perfectly
plastic model is not so good but is still satisfactory, especially at low stress levels.

Lade's elasto-plastic

.-.,..
-.- ....-.
-.-,
'4: - - -.-?,--,.,-C-.-.-

.L.

A'"

C.*

----

.r

r r r r

C o a r , peravc
V 2SkPn -Modcl
%+Pm -Modcl
697kP.-Modcl

-697kPa-TcW

.-.%.-'

'
I
f-

'

,*-,
.-m

Axial strain (%)

_.-'

_ ...

"

0 1 2 3 4 5 6 7 8 9 1 0
Axialstrain (%)

a) Thiaxial tests

Axial stress (kPa)

Axial stress (kPa)

b) Oedometer tests

Figure 5.6: Comparison of predicted and

measured behaviour in laboratory tests

Embankments 1 1 77
Table 5.7: Nonlinear elastic model parameters for Roadford rockfill
and sand waste fill

5.3.5.3 Finite er'ement analyses

The finite element mesh is shown in Figure 5.7a. The mesh was refined at the
upstream embankment toe, since particular attention was paid to the deflections of
the asphaltic concrete membrane during reservoir impounding.
The
foundation
0
Scale
25
b)
material and the concrete
of the inspection gallery
were modelled as linear
elastic materials with
Young's moduli of
225MPa and 30GPa and
Poisson's ratios of 0.2 and
0.15 respectively. No
specific Iaboratoj test
data or field observations Figure 5.7: Roadford dam: a) Finite element
were available for the mesh and boundary conditions and b) Mesh
drainage layer located
detail at the upstream embankment toe
below the asphaltic
membrane and the foundation drainage layer. These materials have only a small
influence on deformation and, for convenience, their properties were assumed to
be the same as those of the sand waste fill. Only the lower part of the asphaltic
S

178 1 Finite element analysis in geotechnical engineering: Application

concrete membrane was modelled as a separate zone (see Figure 5.7b). Because of
the very high viscosity of bitumen, the asphaltic concrete of the membrane was
modelled as an undrained granular soil in terms of effective stresses, assuming the
properties of the sand waste fill and the drainage materials (K, = 100 K,,,, , see
Section 3.4 of Volume 1).
Table 5.2: Double hardening model parameters for Roadford rockfill
and sand waste fill

Construction of the dam was simuiated by building in nine layers, each

horizontal layer of elements being treated as a construction layer. The foundation
drainage layer and sand waste fill at the upstream toe of the embankment were
constructed together with the first layer of rockfill material. The layer of drainage
material on the upstream face and asphaltic concrete membrane were placed after
the embankment had been built.
Reservoir impounding was simulated by applying water pressures to both the
upstream slope and upstream foundation (see Figure 5.8). During this process

Embankments 1 179
drained conditions were
assumed in the fill material.
V4hile a static water pressure,
consistent with a water table,
was maintained at foundation
Initial and f d wat
level behind the grout curtain,
an increase in water pressure
in front of the grout curtain by ,jW
upstream ~ f g r oC~& t
due to impounding was
Figure 5.8: Simulation of reservoir loading
modelled.
during impounding

5,3.5.4 Comparison with observations

Vertical movement of the embankment fill was measured with an array of
settlement plates at two locations within the dam, S1 and S2, and the lateral
,,
displacement with an
extensometer at location H
-O b w d sntlcmcnt
(see Figure 5.5a). A
.
I
'b.... ....._
comparison of the predicted a
and measured movements tit
<...
these locations after dam
C..
,>.
construction is shown in 4
,, ..
,
,
B
,
,<*
Figures 5.9 and 5.10.
P l0
,'......
<.C..."'
inspection of these figures
,...
,'
indicates that the analysis
0,
,, ,, ,, , , ,, ,,
Sntlrmstrr (mm)
SetUemcnt (mm)
using the Lade's model is in
b) at settlementplates 'S2'
a) at settlementplates 'Sl'
much better agreement with
the
data. The
Figure 5.9: Comparison of predicted and
with the
measured settlements after construction
model predicts settlements
'OO
reasonably, but overpredicts
h o r i z o n t a l movements
severely. These predictions g ,
could be improved by 5 %
adjusting the nonlinear elastic 4
model parameters, but then the
4
laboratory test data could no
longer be recovered.
40
During impounding of the
.loo
-IW -80
4 -20
0
20
40
64
80
LOO
reservoir the deflections of the
U ~ S ~ (m)
~ I U
Downstream (m)
lower part of the upstream
slope of the embankment were Figure 5-10:Comparison of predicted and
measured lateral movements at
measured using a series of
extensometer 'H'
electro levels. The location of

.,

"

>;.

\\

<:
,
': .

.X

-W

BI

-1
g,

-M)

I !

180 1 Finite element analysis in geotechnical engineering: Application

these gauges is marked as E in Figure

5.5a. A comparison ofthe predicted and
measured deflections are given in
Figure 5.1 1 for three different levels
of reservoir impounding. Both models 'f!
predicted the magnitude of the
movements normal to the membrane
reasonably well.

5.3.6

Distance item toe (m)

60

- - Double-hardeningmodel
-Observed deflection
Imvoundinp,level 14.5111above toe

Example: Analysis of
old puddle clay core
dams
Impounding level 25.0111above toe
5.3.6.1 in troduc tion
100
o
A large number of embankment dams
with central cores of puddle clay were
built in Great Britain in the 19'h
century. As the puddle clay was
60;
expensive, the cores were made as "
thin as possible. This made the cores
- Impounhng level 30.5111above toe
loo
vulnerable to cracking by hydraulic
fracture if they were subjected to full
Figure 5-1 1: Comparison
reservoir pressure on their upstream
boundary, where the seepage pressure predicted and t r ~ a s u r e d
may exceed the normal total stress by of as~halticconcrete membrane
an amount greater than the tensile
strength (which is usually negligible).
It has been deduced that hydraulic fracture of the core was the most probable
cause of the failure of the Dale Dyke dam at the end of first impounding (Binnie
(1978)). Dounias et al. (1996) have performed finite element analyses of the
construction and first impounding of this dam to see whether this mode of failure
could be recovered. Their findings will be briefly presented in the first part of this
section.
There is little information of how these old dams were built. In particular,
shoulder fill was often variable and placed in rather thick layers with little
compaction. This, in combination with extremely soft puddle cores, may give rise
to significant crest settlements as a result of operational cycles of drawdown and
re-filling (Tedd et al. (1994)). KovaEeviC et al. (1997) have analysed the
movements of four embankment sections to examine whether these measured
movements could be recovered, and their implications for long term stability.
Results for one section (Ramsden dam) will be briefly discussed.

Embankments / l 8 1

5.3.6.2 Dale Dyke dam

The collapse of the Dale Dyke embankment in 1864 provoked considerable
professional discussion and caused changes in the design of old British dams. The
changes typically included the use of concrete-filled rather than puddle-filled cutoff trenches, and the adoption of slightly wider cores and zones of compacted
her-grained fill on either side of the core.
The cross-section analysed,
together with the finite
element mesh and assumed
boundary conditions, is shown
in Figure 5.12. The puddle
clay core, the fill and the top
3m of the foundation were
modelled in terms of effective
stresses using the elasto-plastic
b,
constitutive model described
in Section 4.5. The foundation
below 3m was assumed rigid.
~h~ undrained strength of Figure 5. 12: Assumed section, FE mesh
and boundary conditions (Dale Dyke)
puddle clay was of the order
of 12 kN/m2. A small amount
of strain softening (as a result of dilation) was modelled only in the fill and the
foundation, though the strength properties of the fill and the foundation had little
influence on the results.
The fill and the foundation were assumed to be fully drained, i.e. no
construction pore pressures were generated.
,oAnalyses were not 'coupled' so the effects of
100consolidation in the core were considered by
Pore water
pressure (kpa)
changing the pore water pressure at the various
200 - so
----- 100
stages of the analysis according to a simplified
150
........ 200
300
one dimensional consolidation analysis carried
our 'externally'. For example Figure 5.13 shows
400
contours of pore water pressure in the core, S
, . ,,
4
;,i
;.;
immediately after construction, based on these 3 500 $
simple one dimensional calculations.
0 600 During impounding the water load was
700simulated as an external pressure on the
upstream slope of the dam plus an increase in
800 pore water pressure throughout the upstream
shoulder. Collapse settlement has not been
modelled in the analyses. It was considered that Figure 5. 13: Pore water
it would have only a minor influence on the pressures in the core at
behaviour of old British dams built in wet upland the end of construction
areas. This comes from ,the consideration that
(Dale D yke)

8# , \

8 I, 0

I,

$1

182 1 Finite element analysis in geotechnical engineering: Application

collapse settlement primarily occurs

150
in p o o r l y c o m p a c t e d a n d
'unsaturated' rockfill when particle
strength is reduced by wetting due to
inundation in the first reservoir
impounding, or rain infiltration,
leading to additional crushing at &
particle contacts. The full reservoir
pressure was assumed to act at the
50
upstream core boundary, since the fill
was very permeable. The water level
was raised in increments and one
0
dimensional consolidation
calculations were performed after
-1.5 -1.0 -0.5 0 0.5 1.0
1.5
each increment to define the pore
9
Distance (m)
B
U
U
water pressure response in the dam
Start of impounding
core. An example of the changing
x- - Water level reaches section
e - - " - -Reservoir
pore pressures in the puddle clay core
'
pressure = 24 kPa
8
Reservoir pressure = 74 kPa
at one particular level is shown in
S
~eservokpressure = 147 ~a Figure 5.14.
end of impounding
The hydraulic fracture potential at
the end of impounding is examined in
Figure 5.15. Minimum total stresses Figure 5. 14: Pore water pressure
distribution along a horizontal
are plotted at the upstream core
boundary and at a small distance
Section of the
17m
crest level# during
inside the boundary. A comparison of
these minimum stresses with the
(Dale Dyke)
reservoir pressure shows that in the
cut-off trench cracking by hydraulic
Pressure (w/m2)
fracture of the puddle clay is almost
oO
100
200
300
400
inevitable. In reality, it is likely that
failure is prevented by the nonerodible nature of a rock foundation,
which prevents the enlargement of an
erosion channel. In the core, the
reservoir pressure is lower than the
minimum stresses at the upstream
boundary, but is almost equal to the
minimum stresses acting a small
distance inside the boundary.
In the analyses presented above it Figure 5.15: Minimum total stress
was assumed that the impounding was near the upstream boundary of the
relatively slow and carried out at a
core after drained impounding
constant rate. The Dale Dyke dam
(Dale D ykel

2
g

;::1

Embankments 1 183
Pressure (m/ml)
collapsed during a heavy storm,
and it is quite likely that the last
few metres were filled rapidly.
This scenario was investigated in
a further analysis where the last
7.5m was filled quickly, assuming
that the core behaved in an
undrained manner. The results are
presented in Figure 5.16. It can be
seen that minimum total stresses
at the upstream core boundary are
equal to the reservoir Pressure at a Figure 5.1 6: Minimum total stress near
between 7 and 12m
the upstream boundary of the core,
the crest. At a small distance
effect of undrained impounding
inside the boundary they are lower
(Dale D ykel
than the reservoir pressure almost
along the whole length of the core. It seems that the rapid undrained final
impounding may induce hydraulic fracture around a depth of 10m from the crest.
Once formed, it will propagate rapidly downstream, where the total stresses are
lower, as an undrained fracture.

5.3.6.3 Ramsden dam

Ramsden dam in Derbyshire,
England, built between 1879 and
1883, has a central puddle clay
core and a concrete filled cut-off
trench (not tmdelled in the
Figure 5.17: Ramsden dam: Finite
analysis, see Figure 5.17). With
element mesh
time it has shown signs of
significant settlements, and the
crest had to be reconstructed at
least twice prior to 1988, when the
reservoir was almost emptied
g ,so
-( l 6.5 m drawdown). It was drawn
down by 6 m in 1989. The
ongoing settlements have given
Axial smh, s. [%)
rise to considerable concern, and
therefore detailed observations
have been made during two
drawdowns in 1988 and 1989
(Tedd et al. (1990)).
Figure 5.18: Material properties:
The material properties a) assumed 1D compression,
adopted for the shoulder fill and b) assumed triaxial compression,
the puddle core are shown, as c) cyclic1D loading (Tedd et al., 1994)

184 / Finite element analysis in geotecl~nicalengineering: Application

predictions for conventional laboratory soil tests, on Figures 5.18a and b. The core
was simulated using a model based on modified Cam clay (see Section 7.9 of
Volume l ) and the fill by Lade's single haridening model (Lade and Kim (1988)).
This model is different from the Lade's double hardening model used in the
analysis of Roadford dam and briefly described in Chapter 8 of Volume1 . Plastic
yield before failure is still accounted for with this model, but this time utilising
only one yield surface. The results of a cyclic oedometer test on a typical fill from
a mudrock are shown on Figure 5 . 1 8 ~(Tedd et al. (1994)).
The load-unload properties adopted, which dominate predictions during
reservoir operation, were based on this test. The irrecoverable deformations during
operational cycles of drawdown and re-filling are due to the different elastic
stiffnesses adopted during unloading and reloading, not due to plastic straining.
The analyses were performed in terms of effective stress. The shoulder fill and
the foundation were assumed to be free-draining. Coupled consolidation/swelling
was adopted for the puddle clay core, with either a constant permeability or one
varying with the mean effective stress, p' (see Section 10.7.5 of Volume 1):

where k,, = 10v8m/s is the coefficient of permeability at zero mean effective stress
and a = 0.03 m2/kN is a material constant. The analysis using Equation 5.10 gave
a nearly perfect fit with the observations (see Figure 5.19). The degree of
consolidation in the clay core controlled by coupling of volumetric strain and
permeability enabled the time dependent response to load changes to be modelled
in real time. No separate approximate consolidation calculations were needed for
these analyses.
The purpose of the analyses was to examine
Wet boundary
behaviour during operation of the reservoir only.
Thus a simplified construction history was
simulated, involving construction in layers, full Predicted
consolidation of the core, impounding and the pressure
establishment of steady seepage, prior to the
impositions ofthe drawdown and re-impounding
cycles.
The observations and predictions of crest
movement are shown in Figure 5.20. Small
adjustments to the elastic properties shown in
Figure 5.18 were made to recover the difference
in movement observed between the core and the
shoulders. There is no doubt that better
agreement between prediction and observation
Figure 5-19: Seepage
could have been obtained by further adjustments
to the assumptions made. However, the reality is pressure through a narrow
core (lVughan (1994))
unknown and there is a limitation on what can be
achieved without a detailed site investigation.

Embankments 1 185
Predicted
Although the analyses cannot
displacements
predict the extent of the observed
irrecoverable deformation towards
downstream of the whole
embankment crest due to a
drawdown cycle, they show that the 4'
-16.5m
-16.51~1
large and irrecoverable deformations
observed can be reproduced by
a) 1988 drawdown
assuming
- soil parameters which are
consistent with the actual materials.
The average local factor of safety
,
(the ratio of strength mobilised to
b) lggg drawdown
strength available at the current stress
Movement scale (mm)
level) at the end of major drawdown
0
20
40
60
is quite high (about 1.8). Thus large
irrecoverable deformations occur
Stages: 0 - Reservoir full
despite a substantial factor of safety
1 - 1988 dmwdown (16.5m)
2 - Held at drawdo& levei
against overall slope stability, and
3 -Raised to top water level
such large movements do not
4 -Held at top water level
5 - 1989 drawdown (6.0m)
necessarily indicate incipient
6 - Raised to top water level
instability. In the analyses the
irrecoverability arises through the
Figure 5.20: Ramsden dam:
difference in elastic properties
Observed and predicted
assumed in unloading and reloading,
displacements during two
not through plastic strains. Thus
reservoir dra wdo wns
irrecoverability. is predicted
for the
.
second smaller drawdown. Large drawdowns do not eliminate irrecoverable
movements during subsequent smaller drawdowns.
I

5.4

Finite element analysis of earth embankments

5.4.1
Introduction
Finite element analysis of earth embankments is potentially more difficult than that
of rockfill dams, as an earth embankment is usually derived from clayey fill only
and may sit on a foundation of varying strength and permeability which will affect
its performance. The extreme condition of an embankment placed on a soft clay
will be presented in Section 5.5. Here, only embankments sitting on a relatively
'stronger' foundation will be considered. The first example is a typical road
embankment derived from a stiff clay on a stiff clay foundation, where behaviour
of the embankment is mostly governed by the undrained shear strength of the
foundation material due to changes in pore water pressure near the surface. The
second example is of a large embankment dam founded on a competent foundation
covered by a plastic clay deposit which was left in place below the embankment,
enabling the embankment to fail progressively during its construction.

186 / Finite element analysis in geotechnical engineering: Application

5.4.2
Modelling of earthfill
During construction it is usually assumed that the earthfill is saturated and will
behave largely undrained. Because of simplicity, analyses of undrained saturated
fill in terms oftotal stresses is quite common (see e.g. Section 5.4.4). However, the
pore pressures could then only be estimated from the total stresses.
Analyses in terms of effective stresses are more tempting because they predict
undrained pore water pressures directly. Besides, in clay fills it is often necessary
to model initial suctions (negative tensile pore pressures). To do this, after the
element has been constructed, the mean effective stressp' can be reset to be equal
to an appropriate initial suction pjo, and the pore water pressure is reset to p, =
pfi,+p,', where p,' is the mean effective stress (equal to the mean total stress, p,)
after construction of the element. By essentially resetting the mean effective stress
p', it is possible to control the undrained shear strength, S,,,ofthe clay fill material.
It is worth noting that, by changing both mean effective stress and pore water
pressure, total stresses remain constant and equilibrium of forces is maintained.
In spite of their low permeability, real earthfills are partly drained at the end of
construction, at least at the outer boundaries of the dam. To model this time
dependent behaviour in terms of dissipation of construction generated excess pore
water pressures, 'coupled' approaches which consider the interaction between the
soil skeleton and the pore water are needed. The problem is solved on the basis of
equilibrium and the constitutive relationships ofthe soil skeleton together with the
water continuity equation (see Section 10.3 of Volume 1). This approach has been
used in some of the examples presented here (see Sections 5.3.6 and 5.4.3).
In the above approach it is assumed that (i) water completely fills all soil voids
(the so called 'coupled saturated' approach) and (ii) water is incompressible.
Nevertheless, these are reasonable assumptions for fine grained fills placed in a wet
compaction state, such as a typical British clay fill derived from a saturated borrow
pit.
In order to model the actual soil compressibility in unsaturated conditions, it is
necessary to use the 'coupled unsaturated' approaches (see, for example, Pagano
(1998)) which consider the actual multiphase character of earthfills (soil skeleton,
pore air, pore water). The problem is then solved on the basis of (i) equilibrium
equations, (ii) water continuity equation, Qiii) air continuity equation, (iv) soil
skeleton constitutive equations, and (v) water retention characteristic curve. By
assuming the air pressure to be equal to the atmospheric value, the air continuity
equation can be eliminated from the set of governing equations.
5.4.3
Example: Road embankments on London Clay
5.4.3.1 Introduction
Foundations for a typical road embankment may vary from an undrained clay to
a drained granular material. During embankment construction on a saturated low
permeability clay, such as stiff plastic London Clay, excess positive pore pressures
are generated. As these pore pressures dissipate, stability should then improve with

Embankments 1 l87

time. Undrained stability in the short-term may be of a concern, and clearly it will
depend on the undrained shear strength of the clay in the foundation. Regarding the
fill material, placement and compaction of clay fills may generate quite high pore
water suctions, particularly in more plastic clays. Thus in plastic clay fills of
modest height, typical of road works, long-term equilibrium pore pressures are
likely to be higher than those at the end of construction. The strength of the fill
may then decrease with time as the pore pressures equilibrate and there is a risk of
delayed slides.
Undrained shear strength
Probable avenge surface value
of stiff plastic clays depends
more on effective stress rather
than water content. There is a
seasonal variation of pore
water pressure in the surface
of a typical British clay
stratum as shown in Figure
5.21, and therefore the
undrained shear strength will
also vary seasonally. Figure 5.2 1: Pore water pressures near
Examples of short-ferm the surface of the ground: slopes in clay
in the UK (Walbancke ( 1 976),
behaviour of embankments
Vaughan ( 1994))
derived from London Clay and
constructed on London Clay foundations in either summer or winter weather
conditions will be now presented (Vaughan (1994)).

5.4.3.2

Material properties
Soil properties for the in-situ weathered London Clay were the same as those used
in the analyses of slopes cut in this material (see Chapter 4). The assumed
properties for the fill derived from London Clay are also listed in Table 4.1. The
higher value of cr=12kPa,
compared to ci=7kPa adopted S, max~al(kPa)
for the in-situ Brown London
Clay, reflects the absence of
seasonal
fissures in the fill. Also, a
reduced pre-peak stiffness and
an increased post-peak
straining to reduce strength
from peak to residual have
p'sm0'+c'cos~'
S.=
been assumed for the fill. The
l-sui+'/3
fill was simulated as a
Figure 5.22: Changes in undrained
saturated clay with an initial
suction as placed to give an
strength of stiff plastic clay due to
changes in pore water pressure near
undrained shear strength of
S,,=60 kPa, typical of a slightly
to the surface of the ground

188 / Finite element analysis in geotechnical engineering: Application

wet fill. The permeability of plastic clay fill is low, and a constant value of the
coefficient of permeability of k=I .5xlO-'O m/s has been adopted to match, with the
stiffness assumed, typical field deduced values for the coefficient of
consolidation/swelling in the range of c, = 0.5-2.0 m2/year.
The predicted peak undrained strengths (the soil model described in Section 4.5
has been used in these predictions and subsequent analyses), as would be measured
in a triaxial compression test on a saturated sample obtained from the ground
without a change in mean effective stress, are shown in Figure 5.22. It can be seen
that there is a significant difference between the winter and summer strengths, the
former being much lower. Also shown on this figure are strength profiles based on
empirical relations proposed by Sandroni (1977). The implication of this with
respect to stability will be briefly discussed.

5.4.3.3 Finite element analyses

A typical finite element mesh,
Embankment constrwted m
3
with appropriate boundary
conditions, used in the
analyses is shown in Figure
5.23. It was intended to build a
10 m high embankment with
Fired boundary - impamcable
an external slope 2.5 to 1 8 '7
X
(horizontal to vertical). The
o
~cak
zs m
embankment was constructed
in layers, with the weight of Figure 5.23: Mesh for analysis of 1Om
each layer increased in several
high embankment
increments so that the height
at which the embankment collapses can be determined accurately. Analyses were
of the coupled consolidation/swelling type, although the London Clay response,
with the permeability assumed and time steps used, was nearly undrained.
Figure 5.24 shows the displacement vectors just before collapse when the
embankment was built on the clay foundation with undrained strengths based on
pore water pressures at the end of winter. A rupture surface developed in the
weakest clay just below the level to which consolidation has penetrated. The
strength along this rupture surface is between peak and residual. The final rupture
surface through the fill has not formed when the embankment collapses. This is a
consequence of progressive failure. Collapse is predicted when the embankment
reaches a height of 6.2m.
Figure 5.25 shows results from a similar analysis with the foundation strength
profile at the end of summer. A deeper rupture surface is predicted this time. It
penetrates to just below the level where the strength has been increased by the
suctions induced by summer weather. Collapse is now predicted when the
embankment reaches a height of 10.4 m. The importance of the initial pore water
pressures assumed in the analyses (and thus the time of the year when an
embankment is constructed) is clearly demonstrated.

Embankments / 189

Figure 5.24: Collapse with pore pressures

at the end of winter; critical height 6.2m

'YT

--m

Ru~turesurface between peak and residual

Rupture surface at residual

Figure 5.25: Collapse with pore pressures

at the end of summer; critical height
10.4m

Example: Failure of Carsington embankment

5.4.4
5.4.4. 1 In troduction
The failure of the upstream side of Carsington embankment just prior to its
completion in 1984 has been reported in the literature (Skempton and Coats
(1985), Skempton (1 985)). The detailed investigationsinto the failure showed that
an analysis by conventional limit equilibrium methods gave a safety factor at the
time of the collapse significantly greater than one. Since the soils involved were
brittle, this discrepancy could be attributed to progressive failure. As an attempt to
provide an independent demonstration of this, methods of nonlinear finite element
analysis involvingthe modelling of strain softening and brittleness were developed
(see Section 4.5 of this volume) and analyses of the failure were then performed
(Dounias (1987), Potts et al. (1990)). The failure of the embankment was
recovered realistically, assuming soil properties consistentwith laboratory and field
test data, and the significant role of progressive failure in the collapse was
confirmed.
This example briefly considers the various material parameters used in the
analyses, describes the finite element analyses performed on the original
Carsington section, and discuses some findings deduced from the analyses carried

190 1 Finite element analysis in geoteci-~nicalengineering: Application

out subsequently on embankment geometries which are variants of the original

Carsington section.

5.4.4.2 Material parameters and soil model used

The cross section of the Carsington embankment used for analyses, with the
various materials involved, is depicted in Figure 5.26. Although some analyses
were performed as part of a parametric study in which material properties were
varied, the most probable material properties are summarised in Table 5.3. Most
parameters have been derived from test data, others largely assumed. Their
derivation has been discussed in detail by Potts et al. (1990). The core has been
assumed to be undrained and was modelled using total stress parameters. The
foundation and the shoulder materials were assumed to be fully drained.
A generalised nonlinear elastic strain-softening plastic Mohr-Coulomb model
of the form already described in Section 4.5 of this volume was adopted for all
materials other than the core. The undrained and saturated core reached yield early
on in the analyses and consequently was assumed to be linear elastic before peak
undrained shear strength was reached. Strain-softening was modelled as a
reduction of undrained strength with increasing plastic strain.
Table 5.3: Material properties assumed in the Carsington dam
analysis

Embankments 1 191

5.4.4.3 Finite element analyses

The finite element mesh used
40 m
in the analyses is shown in , - ?
El. 201
Figure 5.27. The dam was
built in stages as shown in
Figure 5.27. It was constructed
until it failed and could not be
raised any further. As failure
was approached, smaller load
increments were adopted, Figure 5.26: Carsington embankment:
Cross-section analysed
typically equivalent to 0.1 m
of fill. The failure height was
deduced by plotting typical
lateral deformations against
height. If the embankment had
not failed when the last layer
o
had been placed, the crest was
- - 5'am
surcharged with a vertical
pressure, in a manner shown
Figure 5.27: Mesh and boundary
on Figure 5.27, until failure
conditions used in analysis
occurred.
Once the failure height was exceeded the solution became unstable, since the
embankment could not then be in equilibrium. However, it was necessary to be
sure that this was not due to a failure of the numerical techniques employed. As in
the analyses of progressive failure in cut slopes (Chapter 4), the failure was
confirmed by running the final solution for a given number of iterations. Instability
was indicated when the deformations increased according to the number of
iterations without an improvement in convergence.

5.4.4.4 Original Carsing ton section

The results obtained at the section through the initial failure are shown in Figure
5.28. The failure (Figure 5.28a) occurred through the core and core extension (the
'boot'), which were of relatively weak clay and which were undrained during
construction, and along a layer of clay on top of the foundation (the 'yellow' clay),
which was fully consolidated throughout construction. Figures 5.28b and 5 . 2 8 ~
show the computed situation at the moment of collapse.
The analysis showed that at collapse (the recovered crest elevation was 200.9m,
as compared with actual collapse elevation of 201.0 m) loss of strength had
occurred through strain softening in the core (A), and in the inner part of the clay
foundation (B) as the failure surface propagated along it. This propagation started
when the constructed level was some 5 m lower than at collapse, yet the measured
and predicted movements of the dam remained small and gave no warning of
collapse (see Figure 5.31). Near the toe of the dam (C), where the final rupture
surface developed post-coilapse, the strength was not fully mobilised. This can also

192 1 Finite element analysis in geotechnical engineering: Application

be seen from Figure 5.29 which

shows the proportion of the ")
available strength mobilised, S, for
the whole embankment just prior to
failure. Corresponding deviatoric
strains are shown in Figure 5.30. A
concentrated shear zone has
developed in the core and boot and b)
in the yellow clay near the boot.
Figure 5.3 1 shows the horizontal
displacements for two characteristic
points, one on the upstream slope
just above the small added berm and
the other at the toe of the boot. The
measured displacements at a peg
near point B are also shown. The
0
-20
-40
agreement between measurements
and predictions is reasonably good,
Figure 5.28: The failure of
although the creep during the winter
Carsington embankment
shutdown is not modelled in the
analysis.
The average mobilised shear
stress along the yellow clay
upstream at failure was 73kN/m2
- - - - 0.52
------ 0.67
and the residual factor, R, was 0.52
0.82
---0.97
mymarn.0.97
(see Section 4.5.4.4). The equivalent
values for the fill in the boot and
core were 37kN/m2 and 0.42. The Figure 5.29: Representative analysis:
ratio s,lr,,,,,
was 1.32 and 1 . l 4 proportion, S, of curren t strength
respectively for the two materials
mobilised just before failure
through which failure occurred. The
factor of safety in terms of peak
strength would have been
approximately equal to 1.25.
According to the analysis,
progressive failure was therefore
responsible for lowering the factor Figure 5.30: Representative analysis:
of safety by 20%.
strains in the embankment just

~~~~~

before failure

5.4.4.5 Effect of the core

geometry on progressive failure
There is no doubt that the stress concentration at the transition from the weak
undrained boot to the strong drained foundation, which can be seen in Figure
5.28c, promoted progressive failure-. The nature of this stress concentration was

Embankments / 193
examined by performing analyses of
Redicled - A A
Predicted- PIB
the Carsington section with the same
- IM)
9
soil properties, but with different core
m
configurations (Dounias et al.
- 195 '3
(1989)). The geometries consisted of
- I90 5
a wide core, a narrow core and an
inclined core. These analyses are ,
l
I
l
I
!
185
summarised on Figures 5.32a to
'0
'O0
O
5.32d.
The collapse heights and the stress
Figure 5.3 1 :Observed and
concentration effects are virtually the
predicted pre-failure movements
same for the wide and inclined cores.
With the narrow core the embankment could not be built high enough to cause
failure through the foundation. However, Figure 5.32e shows that a stress
concentration was developing in the foundation near the core at the highest crest
height which could be reached (local instability near the crest due to applying
surcharge loading did not allow an eventual failure height to be determined).

790,,m~~disp~'ement7mm~

Core height at eollapne, H,:

as built 24.9m

inelied 27.5m

wdc 26.5m

Figure 5.32: Geometries of the Carsington

section considered in analyses

It seems that, given the same width, the influence of the core geometry (core
and boot, wide core, inclined core) is small, although the core and boot of the
original section give slightly worse stability than the other sections. While the
section with the narrow central core shows signs of eventual progressive failure,
the overall stability along the rupture surface through the core and the yellow clay
is much higher than for the other sections.

5.4.4.6 Effect of a beam in improving the stability

An obvious way to improve the stability of a slope such as that at Carsington
would be to add a berm at the toe. To investigate the significance of this, simple

194 / Finite element analysis in geotechnical engineering: Application

analyses of the toe of the

embankment were made (Dounias et
al. (1 989), Vaughan (l 994)).
In order to save computer time a
wedge was analysed, modelling the
slope of the embankment. The
geometry of the upstream slope of the
original Carsington section upstream
from the toe of the boot was initially
considered (see Figure 5.33a). Two
furthergeometries were analysed: one
Analysis
with a berm added as in Figure 5.33b,
and another where a volume equal to
the added berm was used to create a
A. Original
B. Fit berm
1.02
uniform slope (see Figure 5.33~).The
1.15
height of the vertical face was the
same for all three analyses.
The material properties of the Figure 5.33: Effect on stability of
foundation, the weak yellow clay adding a berm on Carsington slope
layer and the shoulder fill were the
same as those assumed for the initial analyses (see Table 5.3). The wedges were
constructed in stages with the vertical face restrained horizontally. The wedge was
then loaded by triangular horizontal and shear stress distributions on the vertical
face. The resultant thrust had an inclination of 10"to the horizontal. The thrust was
applied in increments with decreasing size as failure was approached. All three
sections showed similar failure patterns. A yield zone initiated at the vertical face
and propagated along the weak zone.
According to limit equilibrium calculations, the resistance of sections (b) and
(c) should be equal, and 30 % greater than the resistance of section (a). However,
when the effects of progressive failure are included, section (b) has a resistance 2%
greater then section (a), and section (c) has a resistance 14 %greater. It is clear that
the addition of a berm to stabilise a slope has a reduced effect when the slope is
subject to progressive failure. Indeed, the degree ofprogressive failure can increase
so much that the effect of the berm may be negligible. The best result is obtained
if the additional fill is used to make the wedge stiffer as well as heavier.

5.5

Finite element analysis of embankments on soft

clay

5.5.1
Introduction
Earth embankments are often built on soft clay foundations. Such soils usually
occur adjacent to river estuaries and, typically, the embankmentsare used for flood
potection or for carrying roads or railways. As the soft clays are weak,
embankments can usually only be built to a height of approximately 3m to 4m.

Embankments l l 9 5
However, higher embankments are often needed, for example for grade separated
junctions and flood protection, and consequently special construction measures are
required. Examples of such measures include the use of light weight embankment
fill, the provision of reinforcement in the bottom of the embankment andlor
stagged construction.
In the latter approach the embankment is constructed in stages. After each stage
a rest period is allowed for dissipation of excess pore water pressures in the soft
clay. Such dissipation is accompanied by consolidation and a gain in soil strength.
This in turn enables the embankment to be raised to a greater height in the next
stage of construction. Clearly, the amount of gain in soil strength during the rest
periods governs the height to which the embankment can be raised in the next
construction stage. Consequently, staged construction is often accompanied by the
provision of drainage in the soft clay to increase the amount of pore water pressure
dissipation during the rest periods.
Soft clays are often anisotropic, having both stiffness and strength dependent
on the orientation of the major principal stress. This complicates any analysis and
as sufficient data is not often available to define the magnitude of the anisotropy,
empirical factors are introduced into conventional design procedures (Hight et al.
(1 987)).

5.5.2
Typical soil conditions
Soft clays have not usually been subjected to
much erosion and therefore have not
O
experienced greater vertical stresses in the
past. Consequently, they are normally or very
2
lightly overconsolidated and in principle they
should therefore have an undrained strength
profile of the form shown in Figure 5.34. g
This distribution is based on a ground water
level 2m below the ground surface and an
a
overconsolidation ratio of 1.O. Note that pore
water suctions (hydrostatic) are assumed to
8
occur in the clay above the water table.
However, in most, but not all, locations
the surface layer of the clay has been
101
i
I
\ I
0
10
20
30
subjected to wetting and drying due to
Undrained strength, S, (Ha)
fluctuations in the ground water level. This
(in triaxial compression)
process results in the clay having a surface
crust of a higher strength than that indicated Figure 5.34:
of
in Figure 5.34. A typical strength profile for
strength with depth
a soft clay site is shown in Figure 5.35 (Mair for normally
clay
et al. (1992)). The surface crust extends
down to a depth of approximately 2m and below this level the strength is similar
to that indicated in Figure 5.34.

196 1 Finite element analysis in geotechnical engineering: Application

0
The surface crust can have higher
undrained strengths than would be associated
with a lightly overconsolidated soil. In
2
addition, the stiffness of the crust can be
larger than that of the material below it. This
strength and stiffness and their distributions 'a 4
with depth can have a significant effect on a
F
the stability and deformation of arry Cl 6
embankment.
Most embankments built on soft clay
8
foundations are constructed from fill material
that is much stronger than the soft clay. A
granular fill is often used. Consequentiy,
10
0
10
20
30
stability of the embankment fill is not usually
Undrained strength, S, (kPa)
the main concern.
(in triaxial compression)
V

5.5.3

40

Choice of constitutive
Figure 5.35: Typical
model
distribution of undrained
strength with depth for
The most critical issue with embankments on
a soft clay
soft clay is usually the strength of the clay.
As the embankments are constructed
relatively quickly compared to the time
0
required for any significant consolidation, it
is the undrained strength of the clay that is
2
important. As noted above, this strength c m
vary significantly with depth and any crust
can have a marked effect on the behaviour of
24
an embankment.
-5
As much of the clay is lightly 3
overconsolidated and as critical state models
are reasonably good at reproducing the
behaviour of such soils, this type of model is
often employed to represent the behaviour of
the foundation. These models also have the
ability to predict changes in undrained
l0 o
2
4
6
strength with consolidation. It should be
OCR
noted that a Tresca model cannot reproduce
this behaviour. However, there are two Figure 5-36; Distribution of
potential difficulties when using conventional
OCR to give undrained
critical state models.
strength in Figure 5.35
The first difficulty is that the undrained
strength is not an input parameter to the model. It can, however, be calculated from
the input parameters and the initial stress conditions as described in Section 7.9.3

Embankments / l 9 7
of Volume 1. The required undrained strength profile can then be obtained by
adjusting the overconsolidation ratio (OCR).For example, to obtain the distribution
of undrained strength shown in Figure 5.35, using a form ofthe modified Cam clay
model which has a Mohr-Coulomb hexagon and a circle for the shapes of the yield
and plastic potential surfaces in the deviatoric plane respectively, and the
parameters given in Table 5.4, requires the distribution of OCR shown in Figure
5.36. Clearly, inputting such a distribution of initial OCR and its associated K,,
distribution may be problematic with some computer programs. This problem is
discussed further in Chapter 9.
Table 5.4: Properties for soft clay

It should also be noted that the undrained strength depends on the shapes
adopted for both the yield and plastic potential surfaces in the deviatoric plane. For
example, the undrained strengths shown in Figures 5.34 and 5.35 are appropriate
to triaxial compression conditions. For plane strain conditions the constitutive
model, using exactly the same input parameters, will produce smaller undrained
strengths. In this case the plane strain strengths will be 0.82 times the triaxial
compression strengths.
The second difficulty is that the majority of critical state models assume the soil
to be isotropic. For soft clays this is not necessarily correct and both strength and
stiffness may be anisotropic and depend on the orientation of the major principal
stress. If a clay does exhibit strong anisotropy, then either properties representing
some average strength and stiffness must be input into an isotropic model or,
preferably, a constitutive model such as MIT-E3 (see Section 8.6 of Volume l),
which can reproduce anisotropic behaviour, should be used. An example of the use
of such a model is given in Section 5.5.8.
As noted above, the embankment fill material is usually much stronger and

198 / Finite element analysis in geotechnicai engineering: Application

stiffer than the soft clay and is not therefore critical. However, if stability is to be
modelled in the analysis, the constitutive model used to represent fill behaviour
must simulate yield and ultimate strength. Consequently, a simple Mohr-Coulomb
model is often used. Sometimes the elastic properties are assumed to be stress level
dependent. More recently models of the Lade type (see Section 8.5 of Volume 1)
have been used.

5.5.4
Modelling soil reinforcement
As noted in Section 5.5.1,
reinforcement layers, in the form of
sheets of geotextile, are sometimes
added to the base of an embankment,
see Figure 5.37. The effect of such
reinforcement is discussed further in
Section 5.5.6. To model the
/

reinforcement, membrane elements j

RCSOrccmau/
(or beam elements with low bending j
and shear stiffness), see Section 3.5.5 j
of Volume 1, can be used to represent i
the geotextile. These elements must '
Figure 5.37: Example of an
then be sandwiched between interface
embankment on soft c/ay
elements (see Figure 5.4 l), so that the
interaction between the soil and the geotextile can be realistically modelled. While
the stiffness and strength of the geotextile can be obtained from the supplier or
manufacturer, obtaining appropriate interface properties is more difficult.
Another complication arises as the behaviour of geotextiles is often time
dependent and consequently it may be necessary to include the effects of creep.
Whereas, in principle, it is not difficult to include such effects into a constitutive
model, it is not so easy to obtain the necessary parameters.

Example: Effect of a surface crust

5.5.5
5.5.5.1 ln troduc tion
To illustrate the effect of a surface crust, two plane strain finite element analyses
have been performed for the embankment geometry shown in Figure 5.37. In one
example the distribution of soil strength with depth shown in Figure 5.34 was
simulated, whereas in the second example the strength profile shown in Figure 5.35
was matched.

5.5.5.2 Soil conditions

A form of the modified Cam clay model, which had a Mohr-Coulomb hexagon and
a circle for the shapes of the yield and plastic potential surfaces in the deviatoric
plane respectively, and the parameters given in Table 5.4, was used to represent the

Embankments / l 9 9

soft clay in both analyses. To represent the strength profile shown in Figure 5.34
the OCR was assumed to remain constant with depth and have a value of 1.O. To
obtain the undrained strength profile shown in Figure 5.35 the distribution of OCR
with depth shown in Figure 5.36 was used. It should be noted that in this analysis
the K,, value also varied with depth in accordance with the equation given in Table
5.4. Soil suctions (hydrostatic) were assumed to exist in the clay above the water
table. Consequently, there were finite vertical effective stresses at the soil surface
and hence non zero undrained strengths, see Chapter 9.
In both analyses the embankment was unreinforced and assumed to be
constructed of a granular fill. The fill was modelled using a simple linear elastic
Mohr-Coulomb model with the properties given in Table 5.5. During the analyses
the embankment behaved in a drained manner (i.e. K, = 0), whereas the soft clay
behaved undrained (i.e. K, = 1600K,,,).
Table 5.5:Properties for embankment fill

5.5.5.3 Finite element analysis

The finite element mesh used for both analyses is shown in Figure 5.38 and
consists of 360 eight noded quadrilateral elements. The embankment was
constructed in layers until failure occurred.
Obtaining the failure height was
not straightforward. As failure is F,m.. ,,,approached, each layer of elements
must be constructed over several
increments. Failure is then likely to
occur during the construction of a
,**
mm
layer of elements. The increment at
which this occurs indicates the
proportion of the layer that causes the Figure 5.38: Finite element mesh
for embankment on soft clay
onset of instability. The equivalent

200 / Finite element analysis in geotechnical engineering: Application

O3
height of constructed material can
then be obtained. Failure is identified
when the solution fails to converge
and displacements continue to
develop without further addition of 2 ,
Rll. At this stage equilibrium cannot
Ol
be maintained. To obtain an accurate 8 "
estimate ofthe collapse height several
0
0.5
1
IS
2
25
3
3.5
restarts, each with progressively more
Embankment he~ght(m)
increments to construct the last
construction layer, are necessary.
Figure 5.39: Horizontal
Alternatively, an automatic
displacement of embankment toe
incrementation algorithm can be used
with increase in embankment height
(see for example Abbo and Sloan

ito,

1i

5.5.5.4 Results
The variation ofthe lateral movement
1.;.
of the toe of the embankment (point A I.;
....
on Figure 5.37) with height of the
embankment for both analyses is '
m
shown in Figure 5.39. The surface
crust has the effect of increasing the
embankment height at failure from
I
I
b) Wttb a surfacec w
2m to 3.5m.
Vectors of incremental
displacement during the last Figure 5.40: Vectors of incremental
increment of each analysis are
displacement just before failure
presented in Figure 5.40. These
vectors represent the magnitude and direction of the incremental displacements. It
is the relative magnitudes of the displacements and their directions that are of
interest and not their absolute values. In both analyses these vectors clearly identify
the failure surface. For the analysis with a surface crust the failure surface is
deeper and wider, passing through the centre line of the embankment.

,
b

S)

NO surface

5.5.6
Example: Effect of reinforcement
5.5.6, 1 Introduction
To examine the effect of reinforcement in the bottom of the embankment, a small
parametric study was performed on the example embankment geometry shown in
Figure 5.37. In this study both the stiffness and the strength of the reinforcing layer
were varied.
The reinforcement was assumed to be a layer of geotextile placed on the surface
of the soft clay prior to embankment construction. This was modelled by a layer

Embankments / 201
of membrane elements sandwiched
between two layers of interface
elements, see Figure 5.41. These
elements were added to the mesh
shown in Figure 5.38. Plane strain
finite element analyses were
performed.

! I

Figure 5.4 1: Connection between

5.5.6.2 Soil conditions
membrane, interface and solid
The embankment fill and the soft clay
elements
were modelled in the same manner as
for the previous example. The soft clay had a surface crust obtained by adopting
the OCR profile given in Figure 5.36. The interface elements were assigned the
properties given in Table 5.6. The membrane elements had a Poisson's ratio of 0. l ,
a thickness of 0.8mm and were assumed incapable of sustaining compressive
forces. The Young's modulus and tensile strength were varied in the study.
Table 5.6: Properties of interface between reinforcement and soil

5.5.6.3 Results
As shown in the previous section, with no reinforcement the embankment can be
built to a height of 3.5m. When reinforcement is added, the embankment can be
constructed to a greater height. This can be seen in Figures 5.42 and 5.43 which
show the variation of the maximum embankment height with increase in
reinforcement strength and stiffness respectively.
Figure 5.42 shows results from a series of analyses in which the stiffness of the
reinforcement was 1 . 5l~O6 kN/mZ/m(a typical value for a geotextile) and its tensile
strength was varied. Increasing the tensile strength to 75 kN/m increased the height
of the embankment from 3.5m to 4m. Any further increase in the tensile strength
of the reinforcement produced an insignificant increase in the embankment height.
The results therefore indicate that for areinforcement stiffness of 1 . 51~O6 kN/mZ/m
the maximum height of embankment is 4m and, to achieve this, the reinforcement
must have a tensile strength greater than 75kN/m. If greater tensile strengths are

202 / Finite element analysis in geotechnical engineering: Application

Mobilised tensile force in a membrane at failure ( M m )
used the embankment still fails at 4m
20 50 75 75
75
75
75
and, although the greater tensile
4.0
strength will enhance the stability, the
3,9
extra force available is so small
5
compared to the disturbing force that 2 3.8
it does not result in any significant
increase in embankment height.
0
Figure 5.43 shows results from a 6 3.6 /nnbanklnat
series of analyses in which the tensile
3 . 5 A . . - ' . .
'
.
' . .
0
I00
200
300
400
strength of the reinforcement was
Tensile capacity of a membrane (kNIm),
200kN/m (a typical value for a
constant stiffness E = ~ . S X I OkN~rn~/rn
~
geotextile) and its stiffness was
varied. As the reinforcement stiffness
Figure 5.42: Variation of
increases so does the height to which
embankment height with
the embankment can be constructed.
strength of reinforcement
For example, when the stiffness is
107kN/m2/mthe embankment can be
Moblllsed tensde force m a membrane at farlure ( k ~ / m )
constructed to a height of 4.75m.
However, for this analysis the tensile
L5
75
170 195 200
200
capacity of the reinforcement is fully
mobilised. The results also indicate d
that any further increase in the
stiffness of the reinforcement fails to
3
increase the embankment height as
the full tensile capacity of the
36
reinforcement is mobilised at a height
of 4.75m.
10
10
Io7
IO~
These results clearly show that it
Membrane shffness (kN/m2/m)
is a combination of both the stiffness
and strength of the reinforcement that
Figure 5.43: Variation of
enables the embankment height to be
embankment height with
increased. It is probably impossible to
stiffness of reinforcement
include both of these effects in any
simplified analyses and therefore this example demonstrates the advantages of
numerical analysis. The results also indicate that it is possible to increase the height
of an embankment by 35% by adding reinforcement. However, it should be noted
that the analysis presented above do not account for large displacement effects.

,,
,,

5.5.7
Example: Staged construction
5.5.7. 1 Introduction
As an example of an embankment on soft clay constructed in stages consider the
example shown in Figure 5.44. This road embankment is approximately 8m high
and was constructed in the west of the -United Kingdom. The foundation consists

Embankments 1 203
of ss thick deposit of soft clay
overlying sand. The depth of
the soft clay is not constant but
varies across the site. Because
it is not possible to construct
Figure 5.44: Example of a stage
an embankment 8m high on
constructed embankment
the soft clay in a single
- lift, a
staged construction approach is required.
At the design stage it was therefore necessary to decide on the number and
timing of the construction stages and the number and position of the drains that
must be provided in the foundation. As part of the design process, a series of plane
strain finite element analyses were carried out. One of these analysis will briefly
be described here.
For calculation purposes the sand below the soft clay was assumed to be rigid
and therefore the bottom of the soft clay provided a lower boundary to the finite
element mesh. As the depth of the soft clay varied and as the geometry of the
embankment was not symmetric (see Figure 5.44), the complete cross section had
to be modelled in the analysis.

5.5.7.2 Soil conditions

The soft clay was again modelled using a form of modified Cam clay which used
a Mohr-Coulomb hexagon and a circle for the shapes of the yield surface and the
plastic potential in the deviatoric plane respectively. A Hvorslev surface was
employed for stress states dry of critical, see Section 7.10 in Volume 1. The
material properties are given in Table 5.7. The Hvorslev surface had a slope half
that of the critical state line. The water table was O.5m below the ground surface.
Table 5.7: Properties for soft clay for staged construction example

204 / Finite element analysis in geotechnical engineering: Application

The initial distribution of OCR is shown

in Figure 5.45. Combining this with the
material properties given in Table 5.7 enables
the initial distribution of undrained strength
with depth to be calculated. The distribution "

5-

is compared with the design profile in Figure

5.46. The values plotted are for triaxial
compression. To obtain values appropriate to
plane strain conditions these values should be
reduced by 0.86. Clearly, the undrained
strength will change during the consolidation
stages of embankment construction.
The drainage blanket was modelled as a
simple elasto-plastic Mohr-Coulomb
material. The following parameters were
adopted: y,,,=20kNlm3, E=25MNlm2,p=0.3,
c'=O, rp'=35" and v=OO.The embankment fill
was also modelled using the same model and
the same parameters, except that the Young's
modulus was different for material in each
construction lift. For lifts 2, 3 and 4 the
Young's moduli were 21.8, 16.5 and
7MNlm2respectively.

5.5.7.3 Finite element analysis

34

OCR

l4

OCR = 40

.
6 -

10

3,2 4 14
l6

1.24

Figure 5.45: Variation of OCR

with depth
Undra~nedstrength, S, (kPa)
l0

20

30

19 ,

$ ,o
The finite element mesh is shown in Figure P
5.47. By today's standards this mesh is rather a 12
-Assumed m FE analys~s
14
coarse. However, the analyses were carried
out in 1983 when computer resources were
limited. The calculation process involved
four stages of construction, each separated by Figure 5.46: Distribution Of
strength prior to
a period of consolidation. The material
embankment
involved in each of the construction lifts is
indicated in Figure 5.48 and the main
calculation steps are summarised below:

a) Set initial stresses in the foundation soil;

b) Construct drainage blanket (lift 1);
c) Apply a traffic surcharge of 1OkPa to surface of newly constructed material;
d) Remove surcharge of IOkPa;
e) Dissipate all excess pore water pressures;
f) Construct lift 2;
.g) Apply a traffic surcharge of IOkPa to surface ofnewly constructed material;
h) Remove surcharge of I OkPa;

Embankments / 205

i) Dissipate 90% of accumulated excess pore water pressures;

j) Construct lift 3;
k) Apply a traffic surcharge of 1OkPa to surface of newly constructed material;
I) Remove surcharge of 10kPa;
m) Dissipate 90% of accumulated excess pore water pressures;
n) Construct lift 4;
o) Apply a surcharge equivalent to 0.5m and 0.75m of fill to side berms and top
of embankment respectively;
p) Apply a traffic surcharge of 1OkPa to surface of newly constructed material;
q) Remove surcharge of l OkPa;
r) Dissipate all accumulated excess pore water pressures.
Throughout the analysis the embankment and drainage blanket were assumed
to behave in a drained manner. The soft clay was only assumed to behave in a
drained manner during the
dissipation stages of the
analyses. During construction
and application
and removal
-.
of surcharge loading
undrained conditions were
Figure 5.4 7: Finite element mesh for
assumed. A coupled
stage constructed embankment
consolidation analysis was not
undertaken. It was assumed
that enough drains and enough
time was available for the
dissipation stages to be fully
effective. If the analysis were
to be carried out today, a ,
Second l ~ f f
coupled analysis would
m ~hvM
d
Bill F O IIR ~
probably be undertaken and
the drains modelled using the
approach of Hird et al. (1990).
Figure 5.48: Construction sequence

5.5.7.4 Results
Lateral movements ofthe vertical line AA through the toe of the embankment (see
Figure 5.48) are presented in Figure 5.49. Settlements of the original ground
surface below the left hand side of the embankment are presented in Figure 5.50.
On both figures results are presented for calculation steps 'q' and 'r'.
From Figure 5.50 it can be seen that after construction (step 'q') the original
ground surface has settled nearly 2m. This increases to 2.5m in the long term after
dissipation of all excess pore water pressures (step 'r'). Such large settlements are
often associated with embankments on soft ground.

206 / Finite element analysis in geotechnical engineering: Application

Lateral movement (mm)

Kg
// 7
(i.e. steptern
'I,)

afier
applying 4th lift +
surcharge loading
(1.e step .q')

Figure 5.49: Lateral

movement of vertical line AA
through toe of embankment

70

-50

Distance born centre line of embankment(m)

40
30
20
10

- -0.5
0

o.o

-+

Immediately after
applying 4th lit? +
g
(i.e. step 'q')

In long term
(i.e. step '1')

Figure 5.50: Settlement of original

ground surface

Example: Effect of anisotropic soil behaviour

5.5.8
5.5.8.1 In troduc tion
The Champlain clay covers a large part of a densely populated region of Eastern
Canada. Construction of the road infrastructure requires numerous overpasses
containing embankments approximately 5 m in height. As the clay is only lightly
overconsolidated, it is soft and weak and it is therefore difficult to design and
construct embankments to the required height.
For this reason the geotechnical group of Eaval University decided in 1972 to
undertake a research programme involving the construction of test embankments
on soft Champlain deposits. The objective of this project was to investigate the
failure conditions of fills built on soft, sensitive foundations and also to study the
magnitude and rate of settlement of such structures. Four embankments were
constructed on instrumented clay foundations at the Saint-Alban site in the
Province of Quebec: the first one, which is the subject of this study, was built until
failure occurred; three others were built to a smaller height and with different side
slopes in order to study settlement behaviour.
Test embankment A, described by La Rochelle et al. (1974), was the subject
of finite element analyses (Zdravkovic et al., 2002), with the objective of
investigating the effect of anisotropic soil behaviour on embankment height.

5.5.8.2 Geometry
The geometry of this boundary value problem is presented in Figure 5.5 1. One side
of the embankment has a slope of 2:l (horizontal to vertical) inclination. The

Embankments / 207
stability of this slope is
enhanced by the presence of a
1.5 m high berm, so that any
k
237m
.,.,. 4 6 m
son Champlu slay
failure is forced onto the other - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - side of the embankment, where
the slope is steeper, with 1.5: 1
inclination. The predicted Figure 5.5 1: Geometry of the designed
embankment at St. Alban
height of the embankment,
based on conventional analysis, was 4.6 m.
The construction sequence was such that the first O.6m of embankment was
lifted in a day, followed by 0.3m per day until a height of 1.5m was reached. After
that two 0.3m layers were lifted per day, until failure occurred at about an
embankment height of 4 m, which was less than the predicted 4.6m.

5.5.8.3 Soil conditions

A detailed geotechnical investigation was
Undrained strength, S., (kPa)
carried out at the Saint-Alban site, with a
,
, 2o ,, ,, ,,
O
substantial number of boreholes and both insitu and laboratory experiments on .I
Champlain clay samples. Details of the -2
results of these tests are given by Tavenas
OCR=2.2
and Chapeau (1973), Sarrailh and Tavenas 3 -3
S
TXC - theoretical
(1972).
R4
In general, the soil profile on the site has
PSE
0
CIU
"
a weathered clay crust extending down to -5
CIU-experimental
Range of vane tests
-2.0m below the ground surface, followed by -6
a soft silty marine clay deposit down to
-'
-13.7m. Beneath this clay layer there is a
layer of a dense, fine to medium sand, -8
followed by bedrock. The ground water level
is at 0.7m below the ground surface.
Figure 5.52: Distribution of
Clay below the crust is in an undrained strength profiles
overconsolidated state, with OCR = 2.2. The
with depth
permeability of the clay is estimated to be
about 10-' mlsec, which implies that with the construction rate applied, failure
occurred under undrained conditions. Stress-strain curves from triaxial tests
showed substantial strain softening after the clay had reached peak strength
(Leroueil(1977)). A summary of undrained strength profiles obtained from vane
tests and CIU tests is shown in Figure 5.52 (after Leroueil(1977)).
The embankment fill is granular, with 9' = 44".
v

5.5.8.4 Finite element analyses

For the finite element analyses, these soil conditions were modelled using both
MIT-E3 soil model (to simulate anisotropic soil strength) and the modified Cam

208 1 Finite element analysis in geotechnical engineering: Application

clay soil model, MCC (to simulate isotropic soil strength). The parameters
necessary to describe both models are summarised in Tables 5.8 and 5.9
respectively. It should be noted that for the MIT-E3 model the effect of the
bounding surface plasticity within the state boundary surface was switched off.
Consequently, no values for the parameters y and h are given in Table 5.8.
The undrained strength variation with depth for different loading conditions,
simulated by the MIT-E3 model, is presented in Figure 5.52. It can be seen from
this figure that the predicted direct simple shear strength (DSS) matches, on
average, the strength obtained from vane tests. Also, the simulated CIU strengths
match closely the experimental CIU strengths. It can therefore be concluded that
the observed anisotropic strength has been modelled very well with the MIT-E3
model. Also presented in Figure 5.52 are the profiles of plane strain compression
(PSG) and plane strain extension (PSE) undrained strength.
For the isotropic analyses, the undrained strength in triaxial compression was
adopted and the modified Cam clay model was used to simulate the triaxial
compression strength profile (TXC) in plane strain, as predicted by the MIT-E3
model in Figure 5.52.

5.5.8.5 Results
The finite element mesh for
the embankment analysis is
presented in Figure 5.53 and
the problem was analysed as
plane strain. The same
construction sequence and rate
were simulated as those
implemented on site, such that
failure occurred under
undrained conditions.
Figure 5.54 shows the
development of horizontal
displacement, U, of the
embankment toe with the
embankment height. This
horizontal displacement is
normalised by the maximum
horizontal displacement, U,, ,
predicted just before the slope
failed.

130m

Figure 5.53: Finite element mesh for

embankment analysis

a 2.0

$
*

%
.g

MCC
1.5

- 1.25

DSS

-2
.-

oo
1

Embankment height (m)

The
Figure 5.54: Maximum embankment
using the
soil
height from different soil models
which simulated the
anisotropic soil strength as depicted in Figure 5.52, predicted failure at an
embankment height of 3.9m, which is very close to the 4m observed on site. For

Embankments / 209
the same embankment, but now analysed with the MCC model with an isotropic
undrained strength corresponding to the triaxial compression profile in Figure 5.52,
the failure height is grossly overestimated at 4.9m. Back analysis using the MCC
model to achieve a failure height of 3.9m with an isotropic strength resulted in the
necessary strength being 1.25 times the DSS strength. Analysis with the isotropic
strength being equal to the DSS profile, which is what some design practice
suggests, is largely conservative, predicting failure at only 3.3m embankment
height.
Table 5.8: MIT-E3 parameters for soft Champlain clay

A further set of analyses was performed, where the embankment geometry was
altered to have berms on both sides of the embankment, see Figure 5.55. Again
both anisotropic and isotropic analyses were performed and the results are
presented in Figure 5.56, as normalised horizontal displacement of the
embankment toe versus embankment height.

21 0 I Finite element analysis in geotechnical engineering: Application

Table 5.9: Modified Cam clay parameters for soft Champlain clay

An analysis using the MITE3 model predicted that, with

the help of a berm, the ,
SOm
237m
46m
:
embankment could now be j
,mc,ay
erected to 4.4m. The same - :- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - .- - - - - -:e m b a n k m e n t was then
Figure 5.55: Geometry of the
analysed with the MCC

' soeA

with the
strength back-calculated from
the previous set of analyses
(i.e. 1.25 times the DSS
strength, see Figure 5.54).
This analysis predicted an
embankment height at failure
of 4.9m and therefore
overestimated that obtained by
the analysis with the MIT-E3
model. To obtain the same
failure height as the MIT-E3
analysis, the MCC analysis
needed a strength of only 1.15
times the DSS strength.
Several conclusions can be
drawn from this embankment
study:

p
-

embankment at St. Alban with new berm

2.0

1
B
g

1.0

.2
0.5

,
O

Embankment height (m)

Figure 5.56: Predicted height of the new

embankment with different soil models

- An analysis which assumes isotropic soil behaviour and takes either the
triaxial compression strength profile or the DSS strength profile will either
overestimate or underestimate the embankment height, respectively;

Embankments 1 2 1 1

- Taking account of soil anisotropy, observed through

laboratory and field

experiments, enables a more accurate prediction of embankment height;
- If an isotropic strength profile is back-calculated from the observed behaviour
of a certain embankment geometry, it is not necessarily appropriate if the
embankment geometry becomes slightly altered.

5.6
1.

2.

3.

4.

5.

6.

7.

8.

Summary
'Variable' elastic models were widely used in the past to predict behaviour of
embankment dams, primarily because of their simplicity and ability to model
nonlinear and stress dependent behaviour of fill materials.
A great variety of stress paths, accompanied by a rotation of principal stresses,
occurs in embankment dams during construction, and subsequent impounding
in particular. 'Variable' elastic constitutive models cannot account for stress
path dependency which is caused by nonelastic components of soil behaviour.
Compacted fills have a behaviour pattern similar to that of granular materials.
The double hardening elasto-plastic constitutive model of the Lade's type
accounts for pre-peak plasticity and is able to capture several important
characteristics of the stress-strain and strength behaviour of fill materials,
especially rockfills.
The limitations of computer modelling require relatively thick layers to be
used in the idealization of dam construction. However, significant errors can
occur if the layers are too thick, particularly when the effects of compaction
are to be modelled.
Comparison of the predictions with the observed behaviour during
construction of Roadford dam indicated that the predictions using the Lade's
double hardening model were in reasonable agreement with the field data.
The analysis using the nonlinear elastic model severely overpredicted lateral
movements. This prediction could be improved by adjusting the nonlinear
elastic model parameters, but then the laboratory test data could no longer be
recovered.
The analyses of old puddle clay core dams, which involved construction in
layers, impounding, development of seepage pressures, and cyclic
drawdowns, with soils behaving drained, undrained and partly consolidated,
demonstrated that the behaviour of these water retaining structures can be
fully modelled using the finite element method.
Cracking by hydraulic fracture ofpuddle clay in a cut-off trench seems almost
inevitable. Narrow puddle clay cores have stresses after construction and
during impounding which are close to those allowing hydraulic fracture. The
failure of Dale Dyke dam was recovered when rapid undrained final
impounding was modelled.
The analysis of Ramsden dam has shown that quite large and non-recoverable
movements can occur during operational cycles of drawdown and reimpounding, even though the factor of safety against an overall shear failure

21 2 1 Finite element analysis in geotechnical engineering: Application

is rather high. The combination of accurate field measurements with

sophisticated numerical analyses provides a useful tool for assessing the
safety of old embankment dams.
Finite element analyses of earth embankments are more complex because they
are usually derived from clayey fill only and may sit on a foundation of
varying strength and permeability. The short-term behaviour of a typical road
embankment derived from London Clay and constructed on a London Clay
foundation depends on the initial pore pressures assumed in the analyses, and
thus the time of the year when an embankment is built.
Finite element analyses ofthe original Carsington Dam section reproduced the
progressive development of the actual failure surface and were in reasonable
agreement with the field observations. The role of progressive failure in the
collapse was confirmed independently and the method of analysis was shown
to be effective in predicting it.
Progressive failure develops when the centre ofan embankment is ofweak fill
and imposes a substantial increase in active force on the shoulder fill as the
embankment is completed. The exact geometry ofthe core (narrow with wide
base, wide or inclined) has little influence, provided the total trust is the same.
The addition of berms under drained conditions to the above embankment
slope increases the amount of progressive failure, and the improvement in
stability is much less than that indicated by limit equilibrium calculations.
Embankments constructed on soft clay can usually be built to a height of
approximately 3m to 4m without special measures. For higher embankments
special construction techniques are required, such as providing reinforcement
andfor staged construction.
Soft clays are usually lightly overconsolidated but often have a surface crust
of higher strength and stiffness.
Critical state constitutive models, such as modified Cam clay, are often used
to model soft clay. However, care must be taken to ensure that the model
gives a realistic undrained strength profile.
The surface crust can have a significant influence on the height to which an
embankment can be constructed and on the shape and location of the final
failure mechanism. In the example considered the effect of the crust was to
increase the maximum embankment height by 75%.
The provision of reinforcement in the base of an embankment can
significantly increase the height to which the embankment can be constructed.
Both the stiffness and strength of the reinforcement affect the maximum
embankment height.
Staged construction allows embankments to be constructed to significant
heights. However, any analyses must use a constitutive model for the soft clay
that can account for changes in undrained strength during the consolidation
periods.
Large settlements, greater then lm, are to be expected when constructing
embankments on soft clay.

Embankments 1 2 13

20. Anisotropic soil behaviour has a significant effect on embankment behaviour.

Accounting for anisotropy observed in laboratory and field experiments
enables a more accurate prediction of embankment behaviour. Complex
constitutive models, such as MIT-E3, must be used to represent anisotropic
behaviour.
21. If an isotropic strength is back calculated from the observed behaviour of a
certain embankment geometry, it is not necessarily appropriate if the
embankment geometry becomes slightly altered.

Shallow foundations

6. l

Synopsis

This chapter considers the analysis of shallow foundations. It begins by defining

shallow foundations and categorising them into two groups, those that are founded
at the ground surface and those that are founded at shallow depth. Each category
is then considered separately and the implications for numerical analysis discussed.
Issues such as the bearing capacity of pre-loaded foundations, the effect of
anisotropic strength on bearing capacity and instability due to insufficient soil
stiffness are discussed. Examples are given, where ever possible, to help clarify
some of the issues raised. The chapter ends by describing some of the analyses
performed for the stabilisation of the leaning Tower of Pisa.

6.2

Introduction

In terms of complexity, shallow foundations are probably the simplest of

geotechnical structures. The loads applied to them are often well defined and their
purpose is to transfer these loads to the soil. This is in contrast to slopes and /or
retaining wall problems, where the soil provides both the activating and resisting
forces and there is therefore a complex interaction between the two.
Shallow foundations come in many shapes and sizes and this chapter begins by
categorising them as either surface or shallow foundations. Each category is then
considered in turn and the implications for finite element analysis are discussed.
Examples are given, where possible, to help clarify some of the issues raised.
These include the analysis ofdifferent shaped footings, the bearing capacity of preloaded foundations, the effect of anisotropic strength and instability due to
insufficient soil stiffness. Many of these examples are yet to be published and have
been selected to demonstrate the ability of the finite element method to provide
accurate solutions to conventional problems and also insights into novel problems.
For all analyses an accelerated modified Newton-Raphson scheme, with a substepping stress point algorithm, was employed to solve the nonlinear finite element
equations, see Chapter 9 of Volume 1. For the plane strain, axi-symmetric and
Fourier Series Aided finite element analysis, 8 noded isoparametric elements were
used, with reduced (2x2) integration. For full three dimensional analysis 20 noded
isoparametric elements were used, with reduced ( 2 x 2 ~ 2 )integration.

Shallow foundations / 21 5

6.3

Foundation types

Shallow foundations may be broadly categorised into one of the following two
types.

6.3.1
Surface foundations
These are foundations which are placed
on, or very near to, the soil surface. They
can come in many shapes and sizes, but
are often categorized as either strip,
circular or rectangular footings, see
Figure 6.1.

/
Strip

6.3.2
Shallow foundations
Shallow foundations are similar to
surface foundations, except that they are
embedded below the soil surface, see
Figure 6.2. For economic reasons their
depth below the surface is usually not
great, hence the term shallow. Again,
they are often categorized as either strip,
circular or rectangular.

C~rcle

Rectan6le

Figure 6.1: Types of surface

foundations

6.4
Choice of soil model Figure 6.2: Shallow foundations
Conventional foundation design
considers bearing capacity and
deformations separately. Using these 3
approaches it is not possible to establish
Ultimate
the complete load-displacements curve. It
load
is only possible to provide estimates of
the ultimate foundation load and of the
initial gradient of the load-displacement
Displacement
curve, see Figure 6.3.
The calculation of bearing capacity is
Figure
Typical loadbased on stress field solutions andlor
limit analysis, combined with some
displacement curve
empirical correlations. The soil is
essentially assumed to behave as an elastic Tresca material, if undrained bearing
capacity is being considered, and as an elastic Mohr-Coulomb material, if drained
bearing capacity is under investigation. Such simple constitutive models can also
be used in numerical analysis and it can be shown, see below, that numerical
analysis can recover the analytical bearing capacity solutions where they are
available. However, while the conventional bearing capacity solutions are only
"3

21 6 / Finite element analysis in geotechnical engineering: Application

available for simple foundations resting on one soil type, with either a constant
strength or a strength that varies linearly with depth, numerical analysis can
consider complex foundation geometries resting on inhomogeneous foundations.
Numerical analysis can also cope with complex loading combinations (e.g. vertical,
horizontal and moment loading).
Conventional bearing capacity solutions are restricted to either undrained soil
or fully drained soil. They do not therefore provide information on the variation of
bearing capacity with time. This may be important for foundations constructed on
clay soils. Using a coupled numerical approach, see Chapter 10 of Volume 1, it is
possible to simulate such time dependent behaviour, as long as a realistic
constitutive model is used to represent the soil. In this respect the use of a Tresca
model is not appropriate, as the strength of the soil remains constant. Although, in
principle, a Mohr-Coulomb model could beused, this is not advisable as unrealistic
predictions of bearing capacity can be obtained under undrained and partially
drained conditions, see Chapter 9. To obtain realistic predictions it is necessary to
use a strain hardening/softening model of at least the complexity of a simple
critical state model (e.g. modified Cam Clay).
Conventionally, foundation displacements are predicted using elastic theory.
Several approaches are available and they differ in the manner in which the
stiffness parameters are derived. Some assume the soil is linear elastic, others
obtain the stiffness from an oedometer test. While these approaches appear to give
reasonable predictions of average foundation settlement, they do not provide
accurate predictions of differential settlements, deformations under combined
loading, or movements in the soil adjacent to the foundation. This is perhaps not
surprising as the stiffness of the foundation is usually ignored, with the foundation
assumed to apply a uniform pressure to the soil. Numerical analyses have the
ability to deal with a range of constitutive models and can therefore produce more
realistic predictions.
In principle, a soil model that accurately simulates the behaviour of the soils
supporting the foundation under investigation should be used. If the foundation
consists of layered soils, then several soil models might be used. If bearing
capacity is of interest, then clearly the soil model should accurately simulate the
soil strength. If deformations are of concern, and in particular those adjacent to the
foundation, then it is advisable to use a constitutive model that can accurately
represent the nonlinear behaviour of the soil under small strains.

6.5

Finite element analysis of surface foundations

6.5.1
Introduction
If the foundations are subjected to vertical loading only, then a strip footing can be
idealised as plane strain, a circular footing as axi-symmetric, but a rectangular
footing remains a three dimensional problem. Consequently, while it is relatively
straight forward to analyse strip and circular footings under vertical loading, as
they can be reduced to two dimensions, considerably more computer resources are

Shallow foundations I 21 7
required to analyse rectangular footings
which involve a full three dimensional
analysis, although account can often be taken
of the two vertical planes of symmetry to
reduce the complexities of the analysis, see
Figure 6.4 (i.e. in this case only a quarter of
the full geometry needs to be considered).
If the loading is not vertical but inclined,
and the inclination is in the plane
I
r
/
Plane of
perpendicular to the length of the footing,
dJ
symmetry
plane strain analysis can still be performed
for a strip footing. Otherwise, and for both
v
the circular and rectangular footings, a three
Figure 6.4: Vertical planes of
dimensional analysis is required. For the
symmetry in 30 analysis of a
circular footing this analysis can be
rectangular footing
performed using either a full three
dimensional, or a Fourier series aided approach. A conventional three dimensional
analysis will be required for the rectangular footing.
In general, it is necessary to discretise both the soil and the foundation into
finite elements, see Figure 6.5. This enables the correct stiffness of the foundation
to be included in the analyses. Interface elements can also be positioned between
the underside of the foundation and the soil, so that the correct interface behaviour
can be modelled. However, as the foundation is often very stiff compared to the
soil, numerical problems can arise due to the large difference in stiffness between
adjacent elements on each side of the interface. Although these numerical
instabilities are not so much of a problem with modern computers, where it is
possible to use increased numerical precision (e.g. double or quadruple precision),
surface foundations are often analysed using one of two extreme assumptions (i.e.
a perfectly flexible or rigid foundation). Such assumptions are also used in
conventional soil mechanics practice. If one of these assumptions is made, then
there is no need to include the foundation itself in the analysis.

Foundation

Interface elements

Figure 6.5: Finite element discretisation of

both soil and foundation

21 8 / Finite element analysis in geotechnical engineering: Application

6.5.2
Flexible foundations
If the footing is assumed to be flexible, then it is possible to simulate its behaviour
in a finite element analysis as shown in Figure 6.6. As the footing is flexible, it is
assumed that any loading is uniform and can be represented by a surface surcharge
pressure applied to the surface of the soil immediately below the position of the
footing. As noted in Chapter 3 of Volume I , such a uniform surcharge load can be
converted to give equivalent nodal forces, AFy . In fact, most computer programs
do this automatically for the user.
If the footing is smooth, Flexible foundation:
then the other boundary
Load control
Displacement control
rough
condition that must be applied ""00"
to the nodes under the footing
is that the horizontal nodal
NOT POSSIBLE
forces are zero (i.e. AF,=O).
Most computer programs
assume that AF,=O if no
b o u n d a r y condition is
Figure 6.6: Boundary condition options for
specified in the X direction for
flexible foo ting
a particular node.
Alternatively, if the footing is rough, then the horizontal displacement (Au) of
the nodes on the soil surface below the position of the footing must be restricted
and therefore set to zero (Au =0), see Figure 6.6.
It is not possible to apply displacements to the footing and perform the analysis
in displacement control. Consequently, care must be taken when approaching the
collapse load of the foundation, as the application of too much load will cause
unlimited displacements. This will probably manifest itself as a convergence
problem in the nonlinear solution algorithm. Some finite element programs have
facilities for automatically controlling the size of the incremental loads. This is
particularly useful for load control problems and can result in accurate estimates
of limit loads.
6.5.3

Rigid foundations
If the footing is rigid, analyses can be performed under either loador displacement
control. The various alternatives for the boundary conditions to be applied to the
soil surface below the position of the footing are given in Figure 6.7. As noted in
this figure, for load control it is necessary to tie the vertical displacements of the
nodes below the position of the footing, see Section 3.7.4 of Volume 1 . This will
ensure that all the nodes move vertically by the same amount. As the vertical
displacements are tied, the load can be applied as auniform pressure over the width
of the footing, or as a single point load at the node on the centre Iine of the footing.
In addition, as with flexible foundations, care must be taken when approaching the
collapse load of the foundation if the analysis is being performed under load
control.

Shallow foundations 1 21 9
Rigid footing:
Load conk01
smooth

Displacement conIr01
smooth

m
rough

AV a pli

Figure 6.7: Boundary condition options for rigid

footing

If the analysis 'is performed under displacement control, with vertical

displacements applied to the soil surface below the position of the footing, then the
footing load is obtained from summing the vertical reactions of the nodes which
have been subjected to this displacement. In a nonlinear finite element analysis,
these nodal reactions are best calculated from the total stresses in the elements
connected to the displaced nodes, using:

where {R) is the vector of reactions in the coordinate directions, [B] is the strain
matrix (see Section 2.6 of Volume l), {c} is the vector of current total stresses,
{B,} is the vector of total stresses existing prior to displacing the footing, and the
volume integration is performed for all soil elements connected to the displaced
nodes. Most finite element software have facilities for obtaining such values.

6.5.4
Examples of vertical loading
6.5.4. 1 Introduction
As a simple example of the application of numerical analysis, the problem of a
vertically loaded surface footing resting on either an undrained clay or a drained
soil (clay or sand) is considered. To be consistent with conventional foundation
design, the undrained clay and drained soil are modelled using a linear elastic
Tresca and a linear elastic Mohr-Coulombmodel respectively. The results of these
analyses are compared with the conventional bearing capacity solutions.

6.5.4.2 Strip footings on undrained clay

Analyses have been performed to obtain the load-displacement curves for rigid
strip surface footings. The soil was assumed to be elasto-plastic, with a Tresca
yield surface and properties: E=100MPa, p = 0.49 and S,,= 100kPa. The mesh
shown in Figure 6.8 was used. The two vertical sides of the mesh have been
restrained in the horizontal direction, while the base of the mesh was not allowed
to move in either the vertical or horizontal direction. Loading was simulated by
applying increments of vertical displacement to the soil surface below the position
of the footing, as described above.

220 / Finite element analysis in geotechnical engineering: Application

Load-displacement curves for the

strip footing are shown in Figure 6.9.
In conventional design the bearing
capacity of a vertically loaded strip
footing on undrained clay is
expressed as:

Q,,,

(6.2)
where Q,,, is the maximum vertical
load applied to the footing, A its area Figure 6.8: Finite element mesh for
and NLthe bearing
In strip and circular footing analyses
Figure 6.9 the load is therefore
expressed in terms of the mobilised
P 6
bearing capacity factor NCmob
(= Ql(A 2"
S,,)), where Q is the load on the a?
2
footing. The displacement is $ 4
normalised by B, where B is the half $
width of the footing. Results from
2
three analyses are given. One of these
I
modelled a smooth footing and the 7
rl
other two a rough footing. The
difference between the two analyses
001
002
003
OM
00s
006
Normal~sedsettlement, &/B
modelling a rough footing is that in
one analysis a row of interface
Figure 6.9:Load-dis~lacement
elements was added between the soil
curves for strip footing
surface and the underside of the
footing, see Figure 6.10. In the other
Foundation width
analysis and in the analysis modelling .1
Interface elements1
a smooth footing, these interface
elements were not present. The
interface elements were given a shear
strength of IOOkPa and shear and
normal stiffness values of
Kf=K,,=105kNlm3 (note that these
stiffness values do not have a major
of mesh with
Figure 6. 10:
influence on the results). It can be
interface elements
seen that the three analyses give
slightly different load-displacement curves and slightly different N, values at
failure. These values are quoted on Figure 6.9.
Conventional bearing capacity theory indicates that, for a strip footing resting
on undrained clay with a constant strength, the bearing capacity factor N, should
be 5.1416 (= 2+x), for both smooth and rough footings. This solution can be
obtained from limit analysis (i.e. both upper (unsafe) and lower (safe) bound
solutions) and from a closed form plasticity solution (i.e. combination of stress and
=4
' Nc S u

Shallow foundations 1 221

velocity field solutions). It is therefore
theoretically exact. The finite element
analyses give values of 5.19,5.39 and
5.29 for the analyses with smooth,
rough and rough with interface
elements footings respectively.
Compared to the analytical solution,
the numerical analyses are in error by
0.94%, 4.6% and 2.8% respectively.
Clearly, the error is larger for the
rough footing analyses, and in
particular is larger for the analysis
with no interface elements. The
reason for this is partly explained in
Figure 6.1 1, which shows the vectors
of incremental displacement for the
last increment of each analysis (i.e. at
failure). The orientation of these
vectors indicates the direction of
movement and their length the
magnitude of movement. As it is the
orientation of the vectors and their Figure 6.1 I : Vectors of incremental
relative magnitude that indicates the displacements at failure for three
failure mechanism, the absolute
different soil-footing interfaces
magnitude of the incremental
displacements is irrelevant and consequently no magnitude scale is given on the
figure.
Inspection of these failure mechanisms indicates that some horizontal
movement occurs at the soil surface immediately below the smooth footing. For
the rough footing this is not possible due to the boundary conditions.
Consequently, the failure mechanism for the rough footing is slightly deeper and
wider than that for the smooth footing. This in turn implies a larger area to the slip
surface which, when combined with a constant S,,, leads to a greater ultimate
footing load and hence N, .
The error for all three analyses arises because ofthe nature of the finite element
mesh immediately below and adjacent to the footing. As shown in the insets on
Figure 6.1 1, the vectors of displacement adjacent to the edge of the rough footing
change direction very abruptly, from vertically downwards under the corner of the
footing to an angle of 45" to the horizontal in an upward direction, on the soil
surface adjacent to the footing. This change occurs over the single element
positioned at the corner. The accuracy of the solution will therefore depend on the
ability of this single element to accommodate this rapid change in displacement.
The rapid change in displacement at the edge of the footing is less pronounced in
the smooth footing analyses and in the analysis where interface elements have been

222 1 Finite element analysis in geotechnical engineering: Application

used. It can be seen from the inset on

Figure 6.1 1 that for the latter analysis
some horizontal movement of the soil
surface immediately under the edge of
the footing occurs. This is possible
because the interface elements in this
vicinity are plastic and although the
tops of these elements are constrained
to move vertically downward, the
bottoms can have a horizontal
component of movement (note the
displacements of the tops of the
interface elements are not shown on
Figure 6.1 1).

Normal~sedverhcai displacement, 61B

Figure 6.12: Effect of K, on

load-displacement behaviour

Figure 6.13: Effect of K, on failure mechanism and

plastic zone beneath the footing

Shallow foundations / 223

As shown in Chapter 9 and by Day and Potts (2000), more accurate solutions
can be obtained by using smaller elements around the edge of the footing.
However, these solutions involve larger computational times because of the
singularity at the edge of the footing. Hence, for the work described in this chapter
the mesh shown in Figure 6.8 has been retained. The qualitative comparisons that
are made are therefore likely to be correct, but it should be remarked that more
accurate quantitative results could have been obtained by using a more refined
mesh.
The effects of different initial stress conditions in the ground are shown in
Figures 6.12 and 6.13. Results from analyses with K, values of 0.5, 1 .O and 2.0 are
shown. All analyses had the same soil properties (i.e. same stiffness and strength),
and had a saturated bulk unit weight of 20kN/m3 and a ground water table at the
soil surface. The load-displacement curves shown in Figure 6.12 indicate that K,,
has only a minor influence on the behaviour of the foundation.
Figure 6.13 shows vectors of incremental displacement at three stages, dlB =
0,006,0.02 and 0.05, for each analysis, where 6 is vertical displacement and B is
the half width of the footing. Also shown on these plots are the zones of soil which
have become plastic. All three analyses indicate that at 6/B=0.02 the vectors of
incremental displacement show significant relative movements at depth below the
footing. This is accompanied by an extensive zone of plastic soil. However, at
failure the pattern of displacements have changed and the final failure mechanism
involves quite shallow movements. There is some indication that the zones of
yielded soil extended further laterally and less vertically as K,, increases.

6.5.4.3 Effect of footing shape on the bearing capacity of

undrained clay
Further analyses have been
7
performed to obtain the loaddisplacement curves for rigid P
\; ; ;~*, .
B
circular and square surface
a
foundations. The same soil
properties, as used for the strip
foundations, were adopted. D
For the circular footing, the
mesh of 8 noded elements
66
6
shown in Figure 6.8 was used,
while for the square footing
Figure 6.14a: Horizontal cross-section of
the mesh of 20 noded
a 30 mesh for square footing analysis
shown in Figure 6.14 was
used. As can be seen from Figure 6.14b, the aspect ratios of the elements at depth
below the edge of the footing are not ideal. However, as they are a considerable
distance from the footing, they did not appear to adversely affect the results.
Horizontal movements on the vertical boundaries of the mesh were restrained,
while the base of the mesh was not allowed to move in either horizontal or vertical
,

97;

';
~

I1

i\Tmm

224 1 Finite element analysis in geotechnical engineering: Application

directions. Loading ofthe footing was

again simulated by imposing vertical
displacements to the soil surface
below the position of the footing.
A comparison of the loaddisplacement curves for strip, circular
and square footings is shown in
Figures 6.15 and 6.16 for smooth and
rough footings respectively. Again,
the load is expressed in terms of the
mobilised bearing capacity factor
NCmob(=Q/(AS,,)),
and the displacement
is normalised by B, where B is the
half width of the footing, for the strip
and square foundations, and the
radius for the circular foundation. It
should be noted that, while there is an
analytical solution for the ultimate
bearing capacity factor for the strip
foundation (i.e. N, = 5.14) and a semianalytical/numerical solution for the
circular foundation (i.e. N, = 5.69 and
6.2 for smooth and rough footing
respectively),there is no such solution
for the square foundation. However, it
is commonly assumed that circular
and square footings have the same
bearing capacity, although there is no
theoretical justification for this
assumption. The finite element results
indicate that the bearing capacity of a
square foundation is, in fact, less than
that of a circular foundation.
Analyses have also been
performed for rectangular footings
with length to breadth ratios, LIB, of 2
and 4. These were performed with
meshes similar to that used for the
square footing, but with additional
elements added to account for the
increase in length of the footing. The
resulting load-displacement curves
are'presented in the form of mobilised
bearing capacity factor, NCmob,
versus

Figure 6.146: Vertical cross-section

of a 3D mesh for square footing
analysis

{. *
g

Oo

0.01 0.02 0.03

0.04 0.05 0.06

0.07 0.08 0.09

Normalised displacement, SIB

Figure 6,15: Load-displacement

curves for smooth strip, circular and
square foo rings

.g
2

O,

,,, ,,, ,
,

,,,

Normalised settlement, SIB

Figure 6. 161 Load-disPlacement

curves for rough strip, circular and
square footings

Shallow foundations 1 225

vertical normalised displacement of
the footing, d/B, in Figures 6.17 and
6.18, for smooth and rough footings
respectively.
Conventionally, the undrained
bearing capacity, Q,,, , of a
rectangular surface footing resting on
a clay soil is expressed in terms of the
Ultimatebearing capacity factor Of a
strip footing, NtUiP,and a shape
factor, S,, by the following equation:

..............................................

Smoatb:

............ UB = 111 (5.72)

- - - - - LIB U1 (5.63)

- LIB = 411 (5.45)

0.h

103

0.b5

0.k

0.h7

0.68

Normalised settlement, &/B

Figure 6.1 7: Load-displacement

Q,,, = A S, NCstnP
S,,

(6'3)
curves for smooth rectangular
The shape factor S, depends on the
footings
LIB ratio of the footing and is usually
determined from the following fi
............................................................
2"
empirical relationship (Skempton g 6 1951):
. . . . . . . . . . . . . . . . . . . . . . . .

Rough:

S,

............ U B = Ill
- - -U
-B. U1

= 1+0.2 B / L

This relationship does not disti

between rough and smooth fo
The results presented in Figu
and 6.18 can be used to
values of S, for rectangular footings,
with LIB values of 1 , 2 and 4. These
are compared with Equation (6.4) in
Figure 6.19.

(6.37)
(6.02)

UB = 411 (5.72)

o.& 0.b5 0.b6

Normalised settlement. &/B

0.b7

0.08

Figure 6,78: Load-displacement

curves for rough rectangular
footings

6.5.4.4 Strip footings on

weightless drained

1.3

SO;/

1.25

In conventional design the bearing

l,2 capacity of a vertically loaded strip g footing on a drained soil (i.e. sand) is $ ''l5
expressed as:
3 1.1 V)

QInax
1 A = c'N,

--

Skemptun's empirical cxpnssion

+ q 1 N y+ y ' B N y
1

(6.5)
where B is the half width of
footing, q' is the magnitude of
surcharge pressure existing on
ground surface adjacent to

the
any
the
the

I
0

Ratio of footing dimensions, LIB

Figure 6. 79: Variation of shape

factor with L/B

226 1 Finite element analysis in geotechnical engineering: Application

footing, see Figure 6.20, and N,, N,, and N,

are bearing capacity factors. The factors N,
and N, are derived theoretically assuming the
soil is weightless, while N, is found from an
approximate calculation assuming the soil
has weight, but no cohesion or surcharge.
Figure 6.20: Strip footing
Superposition is then assumed to give
Equation (6.5).
with surface surcharge
Analytical stress field solutions have been
used to obtain the following expressions for N, and N, (Prandtl(1920)):

Nq =

1+ sin p' entanps

l - sin p'

For the situation where the angle of dilation equals the angle of shearing resistance
(i.e. v = p'), it is possible to show that a compatible displacement mechanism is
associated with these stress fields. It is also possible to extend the stress fields
throughout the soil mass, satisfying equilibrium and without violating the yield
condition. The solutions expressed by Equations (6.6) and (6.7) are therefore
theoretically exact, see Chapter 1 of Volume 1. The solutions can also be shown
to be applicable to both smooth and rough footings. If the angle of dilation, v, does
not equal the angle of shearing resistance, p', then the above solutions are only
approximate.
To assess the ability ofnumerical analysis to predict these results, finite element
analyses have been performed using the geometry and finite element mesh shown
in Figure 6.8. The Mohr-Coulomb model was used to model soil behaviour, see
Section 7.5 of Volume 1. The soil was assumed to be weightless (y = 0) and
cohesionless (c1= 0), with an angle of shearing resistance, p', of 25", a Young's
modulus, E', of IOOMPa and a Poisson's ratio, p, of 0.3. Four analyses were
performed with a surcharge, g', of 100kPa and one analysis with g' = IOkPa. The
four analyses with g' = 1OOkPa differed in that different combinations of the angle
of dilation and roughness of the footing were assumed. For the analysis with g' =
1OkPa the footing was smooth and the angle of dilation was zero. For the analyses
with rough footings interface elements were positioned between the footing and the
soil, see Figure 6.10. These were given the same values of p' and v as the soil, and
stiffness values K,=K,,= 10' kN/m3.
The predicted load-displacement curves are shown in Figure 6.2 1. The load is
expressed as the mobilised bearing capacity coefficient, N F b (=Q/(Aqr)). This
figure shows that there is little effect of the footing roughness, but that the angle
of dilation affects both the shape of the load-displacement curve and the ultimate
value of N,. The magnitude of the surcharge load, g', affects the amount of
displacement required to reach failure. However, it does not affect the ultimate
value of N,.

Shallow foundations / 227

For y1=250Equation (6.6) gives a fa*
value of N, = 10.7. The ultimate g ,, __.".,-..Ei..-.<.:E
*_-.
.--(failure) values ofN, for each analysis a 9 are listed in Table 6.1. As noted
above, Equation (6.6) is theoretically o, 6
enact for a soil with an angle of
smaath, v = 0"
dilation equal to the angle of shearing
resistance and is valid for both $ 2
smooth and rough footings. The 2 ' 0
0.05
0.1
0.15
0.2
0.25
Normalised settlement, &/B
analyses in which v = 25" can
therefore be directly compared with
Figure 6.2 1: Load-displacement
the analytical value and the errors are
curves for strip footings on
0.9% and 3%, for the smooth and
rough footing respectively. As with weightless soil (effect of roughness,
dilation and surcharge)
the analyses for undrained clay, the
numerical predictions are improved if
a finer mesh is used in the vicinity of the edge of the footing. The three analyses
with a dilation angle of zero give a value of N, of approximately 10.0. This is
some 7% lower than value given by Equation (6.6).
Vectors of incremental displacement at failure for the four analyses with
ql=lOOkPa are shown in Figure 6.22. These show that for all four analyses the
failure mechanism is deep seated and has a wide lateral extent.
..El

smooth, v = 25"

smooth, v = 0"

rough, v = 0'

. .

rough, v = 25'

Figure 6.22: Effect of roughness and dilation on

the failure mechanism for a strip footing on
weightless soil

6.5.4.5 Strip footings o n a drained soil

If the soil is cohesionless, has no surcharge but has weight, then only the third
term in Equation (6.5) is non-zero. As noted in Section 1.9.2 of Volume 1, it is
rarely possible to obtain analytical stress field solutions for a frictional soil (i.e.

228 1 Finite element analysis in geotechnical engineering: Application

q'>O) which has weight, due to the complexity ofthe governing hyperbolic partial
differential equations. Consequently, there is no theoretically exact expression for
N,. In fact, several alternative expressions can be found in the literature. One of the
most popular is that by Hansen (1970):

This equation, as with many of the alternatives, is often used for both rough and
smooth footings. However recent work by Bolton and Lau (1993) has shown that
footing roughness affects the magnitude ofNy.Their results are based on numerical
integration of the stress field equations and are therefore, again, approximate.
Results from four finite element B ,
analyses, using the mesh shown in 5 ,Figure 6.8 and with the properties
given above for the N, analyses, are
shown in Figure 6.23. Dry conditions
-smwth, v = 0"
were assumed and the soil had a bulk
------ smwth. v = 25"
unit weight of 18kN/m3and K,,=0.577
(i.e. 1 -sinvl). A different combination
of footing roughness and angle of soil
dilation was used in each

The results are expressed as graphs of

mobilised NYb against normalised
footing displacement, JIB. In contrast
to the analyses presented above for
N,,, the final value of N, is dependent
on footing roughness, but not on the
magnitude of the angle of dilation.
The final values of N, are tabulated in
Table 6.2, where they are also
compared with values from Bolton
and Lau (1993) and from Equation
(6.8). It can be noted that for smooth
footings the finite element results are
in agreement with those of Bolton and
Lau (1993). However, for the rough
footings the numerical analyses are in
much better agreement with Equation
(6.8).
The vectors of incremental
displacement at failure are shown in
Figure 6.24. The smooth footings
have much shallower failure
mechanisms and the difference
between these and those for the rough

Normalised settlement, 6/B

Figure 6.23; Load-displacement

curves for a strip footing on
cohesionless soil
(effect of roughness and dilation)

r71
.

... .

,... .

smooth strip, v = 0"

smooth stnp, v = 25'

K, = 0.577, 4'=2s0

.::::-z::,:-

. . ... ., ..-,.... .. .
., . - .. .

..S,..

rough~tnp.V = W

rough sap, V = 25"

Figure 6.24: Failure mechanism for

a strip footing on cohesionless soil
(effect of roughness and dilation)

Shallow foundations / 229

footings explains the difference in the N, values.
Vectors of incremental displacement at three different stages of loading for the
analysis with a smooth footing and with a zero dilation angle are shown in Figure
6.25. Also shown on these figures are the zones of plasticity. As for the undrained
analyses described above, the mode of deformation changes as loading proceeds.
The zone of plasticity is much more extensive than the final failure mechanism.

Figure 6.25: Development of failure beneath a smooth

strip footing on cohesionless soil

In all of the N, analyses presented above K, = 0.577. To investigate the effect

of K,, the analysis with a smooth footing and zero dilation was repeated with K,,
values of l .O and 2.0. The resulting
,_...,_ -------1.5 load-displacement curves are shown
s
in Figure 6.26. The value of K, has an 3 3 effect on the load-displacementcurve .$ 2.5
- K*=0.577
- - .K,= l
prior to failure, but does not affect the $ 2
.....-K p 2
ultimate value of N;. The vectors of .f 1.5
incremental displacement at failure 3 I '3
0.5
for each analysis are compared in
Figure 6.27. Also shown on this 3 0
0,;6
O,&B
O,;l
0,012
figure are the plastic zones at failure.
Normalised settlement, SIB
While the failure mechanism is
independent of KO the size of the Figure 6.26: Effect of K, on loadplastic zone is not.
displacement behaviour

?.

I_

El

C!
Plastic

smwth ship:
,$'=25'
v=@

K.

2.0

Figure 6.27: Effect of K, on failure mechanism and

development of plastic zone

230 / Finite element analysis in geotechnical engineering: Application

If a surcharge loading exists on the soil surface adjacent to the foundation and
the soil has weight, then both the N , and N, terms of Equation (6.5) are non-zero.
As noted above, this equation assumes that the effects of soil weight, y, and
surcharge, g', can be superimposed. To investigate this hypothesis the analysis
presented above for a rough footing on sand with an angle of dilation v = 25" was
repeated with a surcharge g' = 1OkPa. This analysis gave an ultimate load on the
footing Q,,, = 556 kN/m. Based on N, = 11.03 (see Table 6.1) and N, = 6.72 (see
Table 6.2), Equation (6.5) predicts an ultimate load of463 kN/m. Comparing these
values indicates that the superposition assumption implied in Equation (6.5) is
conservative. In this particular case by 17%.
One reason why the superposition assumption is not valid is that the failure
mechanisms associated with N, and N, differ. This can be seen by comparing the
failure mechanisms shown in Figures 6.22 and 6.24. The failure mechanism for N,
is deeper and wider than that for N,.
It should also be noted that the magnitudes of the footing settlements required
to mobilise full N, and N, values differ (see Figures 6.2 1 and 6.23). In addition, the
settlement required to mobilise the ultimate N, value depends on the magnitude of
the surcharge, g'. Conventional design procedures do not explicitly account for
this. However, in practice this could have significant implications, especially in
situations where the soil strength degrades (i.e. v' reduces) with straining. For
example, at ultimate load the average strength associated with the first mechanism
to form (i.e. N,) is likely to have decreased from its peak value, while that
associated with the second mechanism (i.e. N,) is unlikely to have reached its peak
value. In such a case Equation (6.5) may not be conservative (see Chapter 4 for a
detailed discussion on progressive failure).

6.5.4.6 Circular footings on a weightless drained soil

To evaluate the bearing capacity for a vertically loaded circular footing on a
drained soil Equation (6.5) is again used in conventional design. However, the
bearing capacity coefficients N,, N, and N, are modified to account for the circular
shape of the footing. This can be done by using different values for N,/, N, and N,,
or by maintaining the values associated with a strip footing, but introducing shape
factors S,/,S, and S , in a similar manner as shown in Equation (6.3) for undrained
clay.
In contrast to the strip footing, where the expressions given by Equations (6.6)
and (6.7) for N, and N, are theoretically exact, for a circular footing all of the
bearing capacity factors used in conventional design are approximate. To quantify
the effect of footing shape on N, the finite element analyses, presented above in
Section 6.5.4.4 for weightless sand with ql=lOOkPa, were repeated with a circular
footing.
The predicted load-displacement curves are shown in Figure 6.28 and these can
be compared with those for the strip footing given in Figure 6.2 1. In contrast to the
strip footing analyses, which indicate that the footing roughness does not affect the
ultimate value of N , , the circular footing analyses indicate that the rougher the

Shallow foundations / 231

rough, = 25'
footing the larger the ultimate N, $
value. In addition, and again in
contrast to the results for the strip
footing, the angle of dilation has only
a minor influence on the ultimate
value of N , .
One of the reasons for the
differences outlined above can be
seen in Figure 6.29, which shows
Normalised settlement, SIB
vectors of incremental displacement
at failure. These vectors indicate that
the failure mechanisms for the smooth Figure 6.28: Load-displacement
curves for a circular footing on
ciscnlar footings are shallower and of
weightless soil
a smaller lateral extent than those for
the rough circular footings. (effect of roughness and dilation)
Comparison of these failure
mechanisms with those for the strip footings shown in Figure 6.22 indicates that
even for the rough circular footings the failure mechanisms are smaller both
laterally and vertically than for the strip footings.

rough, v = 0"

rough, v = 25'

Figure 6.29: Failure mechanism for a circular

footing on weightless soil
(effect of roughness and dilation)

The ultimate values ofN, are listed in Table 6. l, where they are compared with
the equivalent values from the strip footing analyses. The values for the circular
footings are substantially larger than those for the equivalent strip footings. This
is in agreement with conventional design practice which, depending on the actual
design code or advice manual, suggests a shape factor of between 1.2 and 1.5 to
be applied to the N , value for a strip footing.

232 / Finite element analysis in geotechnical engineering: Application

Table 6.1: Bearing capacity factor N,

Rough footing, g'

100kPa, v = 0"

9.92

18

Rough footing, g'

1OOkPa, v = 25"

11.03

19.42

IOkPa, v = 0"

10

Smooth footing, g'

6.5.4.7 Circular footings on a drained soil

To investigatethe effect of foundation
shape on the N,term in Equation (6.5) $- '
the analyses presented in Section
66.5.4.5 were repeated using an axi- 4 s symmetric, as opposed to a plane a 4
strain, idealisation. The resulting g
-smoolh, v = 0"
load-displacement curves are shown
- - - - - - mmooth,v=25"
in Figure 6.30 and can be compared
to those from the equivalent strip
footing analyses presented in Figure
Normalised settlement, SIB
6.23. The two sets of analyses show
similar behaviour, with the exception
that the ultimate N,values are smaller Figure 6.30: Load-displacement
curves for a circular footing on
for the circular footings. This is in
cohesionless soil
contrast to the N, analyses, which
showed that higher values were (effect of roughness and dilation)
obtained for circular footings.
The ultimate values of N,are listed in Table 6.2, where they are compared to
those from the strip footing analyses and also to those obtained by Bolton and Lau
(1 993). Comparing the values obtained from the finite element analyses, the ratios
of circular to strip footing N, values are 0.88 and 0.98, for the smooth and rough
footings respectively. Compared to the values given by Bolton and Lau (1993), the
smooth footing results are in reasonable agreement, whereas the rough footing
results are not. In fact, Bolton and Lau give values ofN, for rough circular footings
which are substantially larger than for the equivalent strip footing. This is not
predicted by the finite element analyses and is also not in agreement with
conventional design practice, where shape factors of between 0.6 to 0.9 are
recommended (Sieffert and Bay-Gress (2000)).

Shallow foundations 1 233

Table 6.2: Bearing capacity factor N,

The vectors of incremental displacements at failure are shown in Figure 6.3 1 .

Comparing these with those for the equivalent strip footings given in Figure 6.24,
indicates that for circular footings the failure mechanisms are much shallower and
of smaller lateral extent.

smooth circle, v = 25"

Figure 6-31 :Failure mechanism for a

circular footing on cohesionless soil
(effect of roughness and dilation)

6.5.5.
Undrained bearing capacity of non-homogeneous clay
6.5.5.1 Introduction
In Sections 6.5.4.2and 6.5.4.3discussing the undrained bearing capacity of clay,
the strength was assumed to remain constant with depth. In many field situations
this is not a realistic assumption. For some simple linear and bi-linear strength
variations with depth, approximate analytical solutions are available (Davis and
Booker (1973)), however for more general situations such solutions do not exist.

234 / Finite element analysis in geotechnical engineering: Application

6.5.5.2 Constitutive model

Accounting for a variation in undrained strength with depth is relatively easy in a
finite element analysis. A Trescamodel could be employed, in which the undrained
strength is allowed to vary spatially across the finite element mesh. The anaiysis
could then be carried out as described above.
However, in anticipation of the content of Section 6.5.6 of this chapter, and to
illustrate the use of a more complex constitutive model, a form of the modified
Cam Clay model will be used here. In this model the shapes of the yield and plastic
potential surfaces are given by a Mohr-Coulomb hexagon and a circle respectively
(see Sections 7.9.2 and 7.12 of Volume 1). Both stiff and soft clay soils are
considered and the parameters, listed in Table 6.3, have been selected to represent
real soil types. It may be noted that these parameters are essentially effective stress
parameters. It is not possible to input directly the undrained strength or its change
with effective stress. However, as shown in Appendix VII.4 of Volume l , the
undrained strength can be derived from the basic parameters and is given by the
following equation:
OCR
-- - g(0)~~sO-(l

where:

+2~:')[1+

D,:,

g(0) =

2(1+ 2 K
)':

BI]

(1

+ 2 K~')OCR[I + B']

]:

(6.9)

sin p'
sinOsinpl

cos@+

B is the Lode's angle; OCR is the overconsolidation ratio; a,,,' is the

initial vertical effective stress.
:(K have been assumed to be given by:
In the present analyses, K(:' and'

Table 6.3: Soil properties for pre-loaded strip footings

Shallow foundations 1 235

O
The stiff clay parameters are based on the
behaviour of London Clay. An OCR of 6 has
been used and the water table has been
-2
assumed to be 2.5m below the ground
Dafa km maxiat
surface. Above this level the clay remains
saturated and can sustain tensile pore water
pressures. The undrained strength profile is E
shown in Figure 6.32. Also shown in the 33
figure is field data for London Clay from n -6
Carswell et al. (1993).
The soft clay parameters are based on site
-8
investigation and laboratory data from a site
in Grimsby, Yorkshire (Mair et al. (1 992)).
At this site the clay is normally consolidated
-10
below a depth of approximately 2m, with a
o
so
100
150
200
Undrained strength, S, (Ha)
stronger crust at the surface. The water table
has been assumed to be at a depth of 2m and,
as with the stiff clay, the soil above this
Figure 6.32: Undrained
surface is assumed saturated and able to strength profile fitted through
sustain tensile pore water pressures. To
London Clay data
obtain a realistic distribution of undrained
strength with depth, the OCR has been varied as shown in Figure 6.33. Also shown
in this figure is the variation of K,OC,according to Equation (6.10). The resulting
distribution of undrained strength S,, together with the field data, is show in Figure
6.34. The values of S,, plotted are those appropriate to rriaxial compression.

-,

OCR, K,

Figure 6.33: OCR and KO

profiles for soft clay

Undrained strength, S, (kPa)

Figure 6.34: Undrained strength

profile fitted through soft clay
data

236 / Finite element analysis in geoteclhnical engineering: Application

6.5.5.3 Geometry and boundary conditions

Analyses have been performed of 2m and 10m wide rigid strip footings. As
undrained conditions are required, two alternative types of analysis are available.
A fully coupled analysis, combining mechanical deformation and pore water flow
could be performed, see Chapter 10 of Volume 1. Alternatively, only the
mechanical behaviour could be considered, using a large value for the bulk
compressibility of the pore fluid, see Section 3.4 of Volume 1. Although the latter
approach is the simplest and probably the best option, the first approach is used
here. This is because the analyses are to be further extended to drained conditions
in Section 6.5.6 of this chapter.
The finite element mesh for the 2m wide footing is that shown previously in
Figure 6.8. Eight noded plane strain isoparametric elements have been used, with
the four corner nodes having pore water pressure, as well as displacement, degrees
of freedom. The two vertical sides of the mesh have been restrained in the
horizontal direction, while the base of the mesh was not allowed to move either in
the vertical or horizontal direction. Loading of the footing was modelled by
applying increments of equal vertical displacement to nodes located at the soil
surface underneath the position of the footing. Both smooth and rough footings
have been simulated. For the rough footings the horizontal movement of the nodes
immediately below the footing has been restrained.
Throughout the analyses no flow of water has been allowed through the base
of the mesh, immediately beneath the footing, across the ground surface adjacent
to the footing, nor through the left hand boundary which forms the vertical plane
of symmetry through the footing. On the right hand boundary the pore water
pressures have been maintained equal to their original values determined by the
position of the water table. As the footing was loaded rapidly to failure in 2 1 days
and the permeability of the soil was low (see Table 6.3), undrained conditions
prevailed in the soil and therefore the pore pressure boundary conditions on the
right hand side of the mesh had little influence on the results. Similar results would
have been obtained if it had been assumed to be a no flow boundary, however such
a boundary condition was not appropriate for the intended analysis discussed in
Section 6.5.6 of this chapter.
For the analyses performed with a 10m wide strip footing, the mesh dimensions
have been increased by a factor of 5. The soil properties and initial conditions
have, however, been left unchanged.
6.5.5.4 Failure mechanisms
For a smooth strip footing on a clay soil, having a constant undrained shear
strength, two classes of failure mechanisms are theoretically possible: the Hill (Hill
(1 950)) and the Prandtl (Prandtl(1920)) mechanisms.These are shown graphically
in Figure 6.35. For a rough strip footing the Hill mechanism is not appropriate as
it implies horizontal soil movements at the soil-footing interface. Consequently,
only the Prandtl mechanism is valid for a rough strip footing. Interestingly, for a
soil with a constant undrained shear strength, both the Hill and Prandtl solutions

Shallow foundations / 237

give the same value of the bearing capacity
factor N, (i.e. Nc=2+.n). As the Prandtl
solution is valid for both smooth and rough
footings, this also implies that in this case the
a) Hill mechanism of failure
bearing capacity is independent of footing
roughness.
While the above is true for a soil with a r
*I
constant undrained shear strength, S,,,it does
not hold if S,, varies spatially (Davis and
Booker (1973)). The effects of a spatial
variation of S,,on the failure mechanism and
bearing capacity can be seen from the results
b) Prandtl mechanism of failure
of the present analyses. Vectors of
incremental displacement at failure, for a
Figure 6.35: Possible
selection ofthe analyses, are shown in Figure
mechanisms of footing failure
6.36. The vectors clearly indicate the nature
and extent of the failure mechanism for each analysis.
Results from 2m wide smooth and rough footings on stiff clay are shown in
Figures 6.36a and 6.36b respectively. The failure mechanism for the smooth
footing is essentially of the Hill type, with soil immediately under the footing
having both vertical and horizontal components of displacement. In contrast, the
failure mechanism for the rough footing is more like the Prandtl type, with the soil
immediately below the footing moving predominantly vertically. Clearly, the
failure mechanism for the rough footing penetrates deeper below the footing and
involves a considerably larger volume of soil than that for the smooth footing. This
occurs because the horizontal restraint imposed at the soil-footing interface causes
the failure mechanism to penetrate deeper into the soil for the rough footing case.
As the undrained shear strength increases with depth, see Figure 6.32, then it is
perhaps not surprising that the bearing capacity for the rough footing is higher than
for the smooth footing (note: the bearing capacities are noted on Figure 6.36 as
Qmax

1'4).

Because the initial distribution of S,, with depth is linear for the stiff clay, it is
possible to obtain an estimate ofthe bearing capacity using results from Davis and
Booker (1973). This gives a value for the smooth footing of Q,,, lA=145 kPa. To
obtain this value interpolation and scaling from the figures supplied in the above
paper are necessary. Davis and Booker's figures are based on calculations which
involve some finite difference approximations and have been drawn to provide
conservative estimates for design purposes. The above result is therefore subject
to error. Nevertheless, this value of ultimate bearing capacity is within 5% of the
finite element result.
Similar results are also shown in Figures 6 . 3 6 ~and 6.36d, which are from
analyses of a 10m wide footing on soft clay. These results imply that the surface
crust, see Figure 6.34, does not have a major influence on the failure mechanism.
This is not so for the 2m wide footing on soft clay, as can be seen in Figures 6.36e

238 / Finite element analysis in geotechnical engineering: Application

and 6.36f. In this case a deep Prandtl type mechanism is predicted for both the
smooth and rough footing. Here the failure mechanism, the extent of which
depends partly on the footing width, is controlled by the 2m deep surface crust
which has a strength reducing with depth. The failure mechanisms for both the
smooth and rough footings are forced into the weaker soil, as this provides a failure
mechanism involving the least resistance. As the failure mechanism is the same for
both smooth and rough footings, then the bearing capacity is also the same, which
is evident from the results shown on Figures 6.36e and 6.36f.

b) 2m rough footing on stiff clay

a) 2m smooth footing on stiff clay

. . . . . . . . . .
. . . . . . . . . . . . . . 22.5m . . . . .
. . . . . . . . . . .
........................
. . . . . . . . . . .
.........................
. . . . . . . . .
.
. ... . .. . .. . .. . .. . .. . ..

............

. .. ... ... .. ... ... .. .... ........... . . . . .

L

.--a..

.........................
. . . . . . . . . . .
........................
. . . . . . . . . .
.........................
. . . . . . . . .

c) 10m smooth footing on soft clay

d) 10m rough footing on soft clay

e) 2m smooth footing on soft clay

f) 2m rough footing on soft clay

Figure 6.36: Failure mechanisms for smooth and

rough footings on stiff and soft clay

6.5.6

Undrained bearing capacity of pre-loaded strip

foundations on clay
6.5.6.1 Introduction
The bearing capacity of a soil is essential for the design of near surface
foundations. It controls the load which can be supported by the ground and will

Shallow foundations 1 239

influence the size and type of foundation to be used. The problem is well
understood for both undrained and drained loading situations, where the soil
strength parameters are either constant or vary in a simple manner with depth, and
there are several different ways, theoretical and empirical, of determining the
bearing capacity (see above). The estimation of the bearing capacity of new
foundations is therefore relatively straight forward and in this respect it is the short
term (undrained) bearing capacity that usually governs the design of foundations
on clay soils.
The situation is not so clear, however, if extra load is to be added to an existing
foundation on clay soil, some time after initial construction was completed. This
could occur, for example, ifnew machinery is to be installed, or if additional floors
are to be added. It can also occur if old foundations are to be reused for a new
building, a situation that is becoming more common as the civil engineering
profession is becoming more sensitive to issues related to sustainability. Due to
consolidation of the soil after initial loading there is a change in undrained shear
strength. The undrained bearing capacity at the time the additional load is applied
will therefore differ from that during initial construction. At present, there are no
guidelines for determining the magnitude of this change. This is perhaps not
surprising as there will be a complex distribution of undrained strength below the
foundation, which will render simple analytical solutions almost impossible.
This problem is therefore an ideal candidate for finite element analysis and has
been investigated first by Jackson et al. (1997) and then by Zdravkovic et al.
(2001), who extended the range of soil conditions considered. They performed a
series of coupled finite element analyses of strip foundations. A strip footing was
first placed on a clay soil, with known initial conditions, and rapidly loaded to
failure. This provided the initial short term (undrained) bearing capacity. A series
of further analyses were then performed in which the footing was first rapidly
loaded (undrained) to a percentage ofthe original short term bearing capacity. Preload values of 20,40,60, 80 and 100% were used. The load was then held at this
value while all excess pore water pressures in the clay dissipated (i.e. full
consolidation). The footing was then subjected to further rapid loading (undrained)
to failure, and the new undrained bearing capacity determined.

6.5.6.2 Constitutive model

An important facet of this problem is the change of undrained strength which
occurs during consolidation, as this governs the undrained bearing capacity under
any subsequent rapid increase in loading. Consequently, it is necessary to employ
a constitutive model that can accurately predict such behaviour and therefore the
soil was modelled using a form of modified Cam clay. In fact, the same model and
soil parameters as described in the previous section were used. The undrained
analyses for the stiff clay (OCR=6) and the soft clay described previously, were
used to determine the initial undrained bearing capacities. Additional analyses have
been performed with the stiff clay properties with OCRs of l , 2 , 4 , 9 and 25.

240 1 Finite element analysis in geoteichnical engineering: Application

6.5.6.3 Geometry and boundary conditions

Analyses have been performed on 2m and 10m wide rigid strip footings and the
finite element mesh for the 2m wide footing is shown in Figure 6.8.
The boundary conditions for the loading stages of the analyses were the same
as those described in Section 6.5.5. However, during the consolidation stages ofthe
analyses the nodes located at the soil surface underneath the position ofthe footing
(i.e. those displaced during the undrained Roading) were free to displace vertically,
but by the same amount, with the total load on the foundation remaining constant.
This was achieved by using the tied degrees of freedom concept, see Chapter 3 of
Volume 1. The magnitude of the movement that occurred during these
consolidation stages of the analysis was therefore not prescribed, but was a result
of the analysis.
The same hydraulic boundary conditions (i.e. prescribed flows and pore water
pressures), as described in Section 6.5.5, were also maintained throughout the
consolidation stages, except that during these stages of the analyses the pore water
pressures along the ground surface adjacent to the footing have been set and
maintained at the initial values existing prior to footing construction.
Small time steps were used to maintain undrained conditions during the loading
stages. For example, the initial loading to failure was achieved in 21 days.
Sufficient time was allowed during the consolidation stages for all excess pore
water pressures to fully dissipate. Typically this was achieved in 80 years and 30
years for the soft and stiff clay analyses respectively.
6.5.6.4 Results of the analyses
Detailed results from the investigation are given in Zdravkovid et al. (2001).
Consequently only the main conclusion concerning the effect of the pre-load will
be discussed here.
As an example of the results, the load-displacement curves for the 2m wide
smooth footing are shown in Figure 6.37a and 6.37b, for the soft and the stiff
(OCR=6) soils respectively. The curve labelled 'initial' corresponds to the situation
where the footing is loaded rapidly from the initial conditions and therefore
produces the initial short term bearing capacity. These results are from the analysis
discussed in Section 6.5.5. The other curves are the results from the analyses in
which the footing has been pre-loaded to a percentage of this initial bearing
capacity and then allowed to consolidate with no change in load before being
subjected to rapid loading to failure. It can be seen that the higher the pre-load, the
higher the bearing capacity.
When the footing is initially loaded rapidly, excess pore water pressures are
established in the soil. The excess pore water pressures can be considered to consist
of two components. The first (positive) being due to an increase in mean total
stress, and the second (positive or negative) due to an increase in the deviatoric
(shear) stress. The first component will be compressive, with its magnitude
dependent on the footing load. The magnitude of the second component will also
depend on the footing load, but in addition it will depend on the type of clay and

Shallow foundations 1 241

in particular on its overconsolidation
ratio. For a clay with a low OCR the
second component will also be
csmpressive. However, as the OCR
increases, the magnitude of this
compressive pore water pressure
component will decrease, becoming
tensile at high OCRs. Consequently,
the magnitude of the excess pore
water pressure that is generated
during the initial undrained loading
stages of each analysis will depend on
the amount of load applied and on the
initial OCR of the clay.
The changes in the undrained
shear strength of the clay during the
consolidation stages of the analyses
depend directly on the magnitude of
the excess pore water pressures and
the soil properties. For a given soil,
the greater the compressive
magnitude of the excess pore
pressure, the greater the gain in
strength. If the excess pore water
pressures are negative, then the
undrained strength will reduce as
water will be sucked into the soil.
Results from all the rough footing
analyses on stiff clay (in which the
undrained strength increased linearly
with depth) are plotted in Figure 6.38,
as undrained bearing capacity after
pre-loading normalised by the initial
undrained bearing capacity, against
amount of pre-load for different OCR
values. The analyses with OCRs less
than 6 show an increasing bearing
capacity with the amount of pre-load.
For the OCR = 9 analyses the bearing
capacity actually decreases when the
pre-load increases above 80%. A
similar trend of reducing bearing
capacity is also predicted for the
OCR=25 analyses when the pre-load

'0

60

--

40

80%
IW%pre-load
p-e-load
60% prc-load
-40%
pn-load
-o- 20% pre-load

zo
Oo

0.2

0.4

0.6

0.8

1.2

Displacement (m)

Figure 6.37a: Load-displacement

curves for pre-loading on soft clay
180

,,,
120

g ~w

1 so

--

60

40
20

O0

0.1

0.2

100%pm-load
80% PE-load
60% pm-load
40% pn-load
20% pn-load

0.3

Displacement (m)

0.4

0.5

Figure 6.3 76: Load-displacement

Curves for pre-loading on stiff clay
,M

,,

2 g ,,

4 ,,

+OCR 9

,,

$Tj 130

1;

. a

IIO
W

Im,

,, ,

7o

, ,,

Percentage of p~e-loading(%)

Figure 6-38,
(Jndrained bearing
capacity after pre-loading vs.
percentage of pre-load

242 / Finite element analysis in geotechnical engineering: Application

exceeds 60%. In fact for this soil, when an analysis with a 100% pre-load was
attempted, the footing failed during the consolidation stage before it could be
loaded any further.
This behaviour can be explained by considering the excess pore water pressures
developed during pre-loading. As noted above, this consists of two components.
For soil with a high OCR the relative magnitude of these two components changes
with the amount ofpre-load. For the OCR=9 and OCR=25 analyses the component
due to a change in total mean stress is compressive, whereas that due to deviatoric
shearing is tensile. As the amount of pre-load increases above 80% for the OCR=9
and 60% for the OCR=25 analyses, the tensile component increases in magnitude
relative to the compressive component, and consequently the total magnitude of
the excess pore water pressure drops and results in a smaller increase in undrained
strength during the consolidation stage. For the OCR=25 soil with a 100% preload, the tensile deviatoric component of the excess pore water pressure exceeds
that component due to the increase in mean stress. This results in tensile excess
pore water pressures which on consolidation cause swelling, a drop in undrained
strength and therefore a reduction in bearing capacity. Further investigations
indicated that for pre-loads in excess of 9396, footings on a stiff OCR=25 soil failed
during the consolidation stages, before any further load could be added.
To show more clearly the effect of m
180
OCR, the results presented in Figure 1
O h
I7O
6.38 have been re-plotted in Figure
6.39, as normalised undrained bearing 3
* 150
capacity against OCR (plotted on a g.8
8
140
logarithmic scale) for different g"
ta.3 130
amounts of pre-load. For pre-load
s
values less than 50% the gain in
undrained bearing capacity first
,
1
10
25
reduces with OCR, but later recovers 3
OCR
slowly as OCR is increased still
further. This may at first seem to
,cigure 6-39;undrained bearing
contradict the arguments Put forward capacity after pre-loading vs. OCR
above, which suggest that as the OCR
increases then the relative magnitude of the excess pore water pressures which are
available to dissipate is likely to reduce and therefore lead to smaller gains in
undrained bearing capacity. However, as the OCR increases, so does the S,,/aV1
ratio, see Equation (6.9). Consequently, at the higher OCRs there will be a greater
increase in S,,for a given change in G'.This effect therefore acts in the opposite
sense to that associated with the magnitude of the excess pore water pressure, and
explains the behaviour observed in Figure 6.39. For a 100% pre-load the excess
pore pressure effect dominates and the gain in bearing capacity falls continuously
with increase in OCR, see Figure 6.39. In contrast, for the 80% pre-load the gain
in bearing capacity first reduces (excess pore water pressure effect dominating),
then increases slightly (S,jaV1effect dominating) before reducing again (excess
pore water pressure effect dominating) as the OCR increases.

::
::
3-

Shallow foundations 1 243

It is also of interest to note that for the stiff clay analyses, in which the initial
undrained shear strength increases linearly with depth, the gain in bearing capacity
due to pre-load is independent of footing width. Figures 6.38 and 6.39 are therefore
applicable to a footing of any width and may be useful for design purposes.

6.5.6.5 Concluding remarks

The results show that increases of bearing capacity of up to 74% are possible for
pre-loaded foundations on normally consolidated clays, whilst for moderately
overconsolidated clays the maximum increase is only 15%. In the former case the
bearing capacity increases with the amount of pre-load applied. It will also depend
on the width of the footing in relation to the depth of any surface crust in the clay.
However, for the moderately overconsolidated clays the maximum bearing
capacity may occur at pre-loads less than 100%. For heavily overconsolidated
clays pre-loading may cause a reduction in undrained bearing capacity.
In practice most foundations will be designed with a factor of safety on load of
at least 2. This means that the maximum pre-load is therefore likely to be only
50%. The results of the present investigation indicate that such a pre-load will
cause at most an increase in undrained bearing capacity for subsequent loading of
35% for a normally consolidated soil, but only 12% for a soil with an OCR greater
than 4. Consequently, for the majority of real situations it is unlikely that preloading will give rise to a substantial improvement in undrained bearing capacity.

Effect of anisotropic strength on bearing capacity

6.5.7
6.5.7.1 Introduction
In reality most soils behave in an anisotropic manner. This anisotropic behaviour
is often assumed to consist of two components, inherent and induced anisotropy.
Inherent anisotropy is that component which is controlled by the fabric of the soil
before it is loaded. Induced anisotropy is that component which arises due to
further loading.
In principle, if anisotropic behaviour is to be modelled, the constitutive model
should be able to simulate both components. However, if the inherent component
dominates behaviour, then a constitutive model that accounts for this component,
but not the induced component, might be acceptable. In such a situation a model
of the form described in Section 7.10.2 ofvolume 1 might be appropriate. Because
such models do not represent induced anisotropy, they will not predict any effect
of rotating the direction of the principal stresses.
If both components of anisotropy are to be accounted for, then a model which
is formulated in terms of all six components of stress and strain, and not in terms
of invariants, is required. The MIT models described in Chapter 8 of Volume 1 are
such models. MIT-E3 has been used to investigatethe influence of anisotropic soil
behaviour on the bearing capacity of both strip and circular rough, rigid surface
footings, subjected to inclined loading.

244 I Finite element analysis in geotechnical engineering: Application

6.5.7.2 Soil behaviour

Zdravkovic (1996) performed an
150
investigation into the anisotropic
125
behaviour of a silt, using the Imperial
100
College large Hollow Cylinder
2,
Apparatus (HCA). Results from one
of the series of tests (the M-series) are 4
6- so
shown in Figure 6.40. In these tests "
all samples were initially K,,
25
consolidated to point A, with aLode's
o
0 20 40 M) 80 100 120 l40 160 180 200 :
angle 8=-30, (i.e. triaxial
P' (Wa)
compression) and the major principal
stress vertical, i.e. a=OO,where a is
Figure 6.40: Effective stress paths
the angle of inclination of the
from hollow cylinder tests on silt
direction of the major principal stress
to the vertical. At Point A the Lode's angle was changed to &OO, maintaining the
major principal stress in the vertical direction. i.e. a=OO.The test labelled MO was
then sheared undrained to failure, keeping &OO,by increasing the major principal
stress, again at a=OO.The test labelled M90 was sheared undrained to failure,
keeping &P, by reducing the vertical stress so that when failure occurred this was
the minor principal stress and hence the major principal stress was now horizontal,
i.e. a=90. Test M1 5 was initially unloaded to point B1 5 with 8-0' and a=OO.The
major principal stress was then rotated to be inclined to the vertical by 15" (i.e
a=15"). This was done maintaining a constant J, p and B and with the soil
undrained. The stress path moved to point C15. The sample was then sheared
undrained keeping a=15" and &OO. A similar procedure was followed for tests
M30, M45 and M70, except that the direction of the major principal stress was
rotated to be at 30, 45" and 70 to the vertical respectively. The details of testing
are presented in ZdravkoviC (1996) and ZdravkoviC and Jardine (2000).
The results given in Figure 6.40 therefore indicate the effect of changing the
direction of the major principal stress on the behaviour of the soil. Clearly this
behaviour is anisotropic, with both the drained angle of shearing resistance and
equivalent undrained strength reducing with an increase in the angle of rotation of
the direction of the major principal stress.
Parameters were obtained for the MIT-E3 constitutive model, see Table 6.4.
Because the model was to be used'to simulate both isotropic and anisotropic soils,
the effect of the bounding surface plasticity below the state boundary surface had
to be switched off. If this was not done, it was impossible to make the model
behave in an isotropic manner. Consequently, no values for the parameters y and
h are given in Table 6.4. In any case, as the soils were normally consolidated, the
effect of the bounding surface plasticity would not be significant. The model was
then used to reproduce the test series given in Figure 6.40. The results are shown
in Figure 6.41. Overall, the analyses followed the same steps as imposed in the
experiments, except that all samples were unloaded by the same amount, i.e. A to

Shallow foundations 1 245

B. Comparison of Figures 6.40 and 6.41 indicates that the model is able to
reproduce the key aspects of the experimental results.
Table 6.4: MIT-E3 material properties

Parameter description

MIT-E3 value

Slope of the VCL in v-lnp' space

Parameter affecting the hysteretic elasticity

However, it is deficient in one important aspect: while the silt soil showed
dilatant behaviour after reaching phase transformation, MIT-E3 indicates ductile
failure. This is not a serious deficiency, as the analyses to be described here were
intended to model foundations on a soft clay, which shows similar behaviour to the
silt up to phase transformation, without dilating at large strains (Porovic (1995),
Leroueil ( l 977)).
The undrained triaxial compression strength profile for this soft clay is shown
in Figure 6.42. It varies linearly with depth, starting from a finite value at the soil
surface, giving S,,/aV1=0.36.This would be the profile normally adopted in any
analysis using an isotropic soil model. However, MIT-E3 allows the undrained

246 1 Finite element analysis in geotechnical engineering: Application

strength profile to change, depending on the direction ofthe major principal stress,
as shown in Figure 6.42.

20

40

60

80

100

l20

Undrained strength, S,, (kPa)

Figure 6.4 1:Hollo W cylinder

effective stress paths simulated
with MIT model

Figure 6.42: Aniso tropic

undrained strength profiles
from MIT model

6.5.7.3 Behaviour o f S trip footings

Two series of strip footing analyses have been performed, in the first the properties
given in Table 6.4 were adopted, whereas in the second the values of p,,' and p,,'
were reassigned to be pTCf=35"
and pTE1=63".
This had the effect of making the
model isotropic and consequently, when the experimental results given in Figure
6.40 were simulated, each test was predicted to give the same result which was
very similar to the MO test in Figure 6.41. Comparison of the analyses with these
different soil properties therefore allows the effects of anisotropy to be quantified.
For each set of material properties five analyses were performed, each having
a different inclination of the applied
load. This inclination was kept
constant in each analysis, and the
footing was loaded to failure under
load control. To achieve this,
components of vertical (V) and
horizontal (H) loads were applied as
point loads, see Section 3.7.7 of
Volume 1 , to the node at the centre of
the top ofthe footing. The magnitudes
of the load increments were reduced Figure 6.43: Finite element mesh
for analysis of strip footing
as failure was approached. The finite

Shallow foundations 1 247

element mesh is shown in Figure
6.43. As the analyses were performed
in load control, the footing itself had
to be included in the analysis.
Load-displacement curves from a
typical analyses (H/V=0.3) are shown
in Figure 6.44. It can be seen that
failure is well defined, with both the
horizontal and vertical loads reaching
limiting values.
The results of both sets of analysis
are shown in Figure 6.45 in the form
of an interaction diagram showing
values of the vertical (V,,,) and
horizontal (H,,,) loads at failure. The
results clearly show the adverse effect
of the anisotropic behaviour for all
inclinations of loading.
It should be noted that for the
analyses with horizontal loading only,
H,,,=2BSl,,where S, is the value of the
undrained strength mobilised at the
soil surface below the footing. In the
present analysis this will be the
strength appropriate to plane strain
conditions and will therefore be
different to the values plotted in
Figure 6.42, which are for triaxial
compression conditions.

I2O

loo

I
60

Vortical (v) and horizontal (U) displacement (m)

Typical loadcurves for combined

loading analyses

Figure

g,

HIV- 1.0
f#V= 0.3

d 30

__.-F

20
to
Q

lror ic M ~ Tmodel

____---------b l k__-model
HIV-0.1
______------

,,,, F-*m

.,.--.:----------------

l'

"

60

no

I4O
loo

M~

v* (W

Figure 6-45:Envelopes of ultimate

load for strip footing

6.5.7.4 Behaviour o f circular

f o o tings
The above exercise was repeated for
a circular footing. The Fourier series
aided finite element approach was
adopted using the parallel symmetry
option, see Section 12.3.2 of Volume
l . By performing a small parametric
study it was found that only 5
harmonic coefficients were required
to obtain accurate results. The finite
element mesh is shown in Figure
6.46. Vertical and horizontal loading
were applied as line loads (in the

Figure 6-46;
Finite element mesh
for analysis of circular footing

248 1 Finite element analysis in geotechnical engineering: Application

circumferential direction) at the node

1 YI
representing the top right hand corner
of the footing, point A, in Figure
6.46. For the vertical load ( 0only the g g
zero harmonic coefficient was
specified as non zero (i.e.
AV=AV'+O ...).To obtain a horizontal
o
SO
loo
~SO
200
~ S O
3~
load, both radial (H,) and
v* (W
circumferential (H,) line loads were
specified. For the radial load only the
first cosine harmonic was specified as Figure 6.47:Envelopes of ultimate
load for circular footing
non zero (i.e. AH,=O+A
cos9+0...),
whereas for the circumferential load the first sine harmonic coefficient was set (i.e.

2c

g:

AH,,=O+A
sine+O...). The latter was of opposite sign but equal magnitude to that
set for the radial load, see Figure 12.2 and Appendix XII.2 of Volume I .
Again predictions using both sets of soil parameters were obtained. The results
are shown as an interaction diagram in Figure 6.47. The adverse effect of the
anisotropic behaviour is evident again. Comparisonwith Figure 6.45 indicates that
this effect is larger for the circular footing than for the strip footing.

6.6

Finite element analysis of shallow foundations

6.6.1
Introduction
Most of the comments given above for surface foundations also apply to shallow
foundations. However, in this case it is necessary to account for the interface
between the soil and the sides of the foundation, see Figure 6.48. It is not usual to
model the construction of the foundation itself, but to simply assume it to be
'wished in place'.

6.6.2

Effect of foundation depth on undrained bearing

capacity
To account for the depth of a foundation below the ground surface, depth factors
are usually introduced into the general bearing capacity equation in a similar way
to shape factors. For undrained bearing capacity the general equation becomes:
-Qma

- dcsc N:'~~S,

+ p(,

where d, is the depth factor andp, is the total overburden stress at foundation level,
see Figure 6.49. Exact theoretical solutions are not available to account for
foundation depth and therefore the magnitude of the depth factor is often based on
semi-empirical correlations. One of the most popular correlations currently in use
is that proposed by Skempton (1951) and shown in Figure 6.50. This correlation

Shallow foundations I 249

-------.------.------

is largely based on th,e results of

!
:
Shallow
foundat~on
/-----model footing experiments. It shows
that d, increases with foundation
depth until reaching a normalised
depth, DIB, of 10, after which it
remains constant. It is worth noting
that the maximum value of d, was
partly determined by the results from
cavity expansion analysis. The curve
shown in Figure 6.50 does not Figure 6.48: General finite element
distinguish between 'different footing modelling of shallo W foundations
shapes or different footing roughness.
It is therefore commonly applied to
all footing types and in this respect B
refers to the half width of strip and
rectangular footings and the radius of
To investigatethe effect of footing
depth a series of finite element
analyses have been performed with
DIB = 0, 2 , 4, 6, 8 and 10. Both
smooth and rough strip and circular
footings have been analysed. A
typical finite element mesh, for
D/B=8, is shown in Figure 6.5 1. Due
to symmetry only halfthe geometry is
modelled. As with the undrained
analyses presented in Section 6.5.4.2,
the clay was modelled as an elastic
with E=lOOMPa,
kPa, K,=l and

Figure 6-49:General loading

scheme for shallow foundations
\.S

l.4-

1
10

10

The foundation itself was not

Ratio of foundation depth to w~dth,DIB
modelled and therefore displacement
boundary conditions were applied to
Figure 6.50: Skempton 'S
that part of the mesh boundary
relationship for depth factor
adjacent to the footing, see Figure
6.5 1. For a smooth footing zero horizontal displacementswere applied along 'ab',
while increments of downward vertical displacement were applied along 'bc'. For
a rough footing zero horizontal displacements and increments of vertical
displacement were applied along 'abc'.
If the mesh shown in Figure 6.5 1 is used to analyse a rough circular footing,
then the ultimate footing load is 16500kN. Assuming that the full undrained shear
trength of the soil is mobilised in shear along the side of the footing, and noting

250 1 Finite element analysis in getotechnical engineering: Application

41*B~'m
that the diameter of the footing is 2m
and its depth below the ground surface
is Sm, gives a shear force on the side
of the footing of 2~8x?cx100=5026kN.
Taking this value away from the total
footing load at failure gives the
contribution from the base of the1 - 1
footing as 11474kN. Comparing this
with the ultimate load from a similar
analysis performed with the footing at
the surface implies a depth factor, d, of
5.6. This value is considerably larger
than that given in Figure 6.50.
Closer inspection of the analysis
indicates the shear stress mobilised
along the side of the footing is too
large. Summing the vertical nodal Figure 6.5 1: Finite element mesh
reactions along the side of the footing used for analysis of depth factor
gives a force of 12686kN. This is
t
considerably larger than the maximum;
value of 5026kN calculated above,
assuming that the full undrained soiU
3
strength is mobilised in shear along the
side of the footing.
At first, such a result might
Interface elements
(zero tbiclmess)
indicate some problem with the finite
element program. However closer
horizontal
inspection of the analysis shows that at (Note Lffermt
and v a c n l scalu)
all integration points in the finite
element mesh the stresses nowhere
exceed the Tresca failure condition.
The program is therefore functioning
correctly. The problem arises because
in the finite element method the
stresses are only sampled at integration
points. While they may obey the
failure condition at these discrete Figure 6.
Detail of finite elemen
points, the shape functions associated
mesh adjacent to footing
with the elements used might imply a
variation of stresses over the element, which is not representative of those
occurring in the real problem. In the present case eight noded elements have been
used with 2x2 Gauss integration. This implies a linear variation of stresses,
whereas in the axi-symmetric circular footing problem the shear stresses are likely
to vary exponentially away from the side of the footing.

Shallow foundations l 251

To overcome this problem thinner elements could be used to represent the soil
immediately adjacent to the side of the footing. Alternatively, interface elements
could be used, as shown in Figure 6 . 5 2 . n e s e are given normal and shear
stiffnesses compatible with that of the soil (i.e. K,,=K,=10SkNlm3)and a shear
strength of 100kPa (i.e. the same as the soil). (Note that the vertical and horizontal
scales are different in Figure 6.52 and that, for clarity, the zero thickness interface
elements have been drawn with a finite thickness). Repeating the analysis now
gives an ultimate footing load of 9307kN. The contribution from the shear stresses
on the sides of the footing is 4945kN, which is now less than the maximum
possible value of 5026kN and therefore acceptable. (Note that for the maximum
possible value to be mobilised, the major principal stress would have to make
everywhere an angle of 45" to the vertical sides of the footing). Comparing the
contribution from the base of the footing with that from a similar footing located
at the ground surface gives a depth factor d, oE2.2.
The above problem is less severe
in the analyses of a rough strip 3 1
footing and clearly does not occur in Y 2
8
the analyses of both smooth strip and
circular footing, as the shear stress
9
along the side of the footing is zero. $
While the analyses of the rough
circular footing with interface
elements is an improvement over the
0
200
400
600
original analysis, it is still flawed as
Horizontal stress on side of footlog, kPa
tensile horizontal total stresses are
predicted between the sides of the
footing and the adjacent soil. This is Figure 6.53: Horizontal total stress
on side of foothg with tension
shown in Figure 6 . 5 3 , which presents
a110 wed in interface elements
the distribution of total horizontal
stress down the side of the footing at
failure. These stresses are taken from
the interface elements. If the soil is
unable to sustain such tensile stresses $
then the analysis is unrealistic. In 3 '
reality the soil is likely to separate
from the footing, leaving a vertical $
8-6
crack.
n
To overcome this problem, the
o
200
400
600
interface elements positioned down
Horizontal
stress
on
side
of
footing,
kPa
the side of the footing can be given a
zero tensile strength. This means that
they can sustain compressive normal Figure 6.54: Horizontal total stress
stresses, but will open when the stress on side of footing with no tension
becomes zero, essentially modelling a
allowed in interface elements
W

2
g

,
,
,

250 / Finite element analysis in geotechnical engineering: Application

that the diameter of the footing is 2m

and its depth below the ground surface
is 8m, gives a shear force on the side
of the footing of 2~8x~x100=5026kN.
Taking this value away from the total
footing load at failure gives the
contribution from the base of the
footing as 1 1474kN. Comparing this
with the ultimate load from a similar
analysis performed with the footing at
the surface implies a depth factor, d, of
5.6. This value is considerably larger
than that given in Figure 6.50.
Closer inspection of the analysis
indicates the shear stress mobilised
along the side of the footing is too
large. Summing the vertical nodal Figure 6.5 1: Finite element mesh
reactions along the side of the footing used for analysis of depth factor
gives a force of 12686kN. This is
W
considerably larger than the maximum
3
value of 5026kN calculated above,
assuming that the full undrained soil
strength is mobilised in shear along the
side of the footing.
At first, such a result might
Interface elements
(zero tll~ckness) 3
indicate some problem with the finite
element program. However closer
honmtal
inspection ofthe analysis shows that at (hate as-c
and va+xe.l aulca)
all integration points in the finite
element mesh the stresses nowhere
exceed the Tresca failure condition.
The program is therefore functioning
correctly. The problem arises because
in the finite element method the
stresses are only sampled at integration
points. While they may obey the
failure condition at these discrete
6.5Zr Detail of finite element
points, the shape functions associated
mesh adjacent to footing
with the elements used might imply a
variation of stresses over the element, which is not representative of those
occurring in the real problem. In the present case eight noded elements have been
used with 2x2 Gauss integration. This implies a linear variation of stresses,
whereas in the axi-symmetriccircular footing problem the shear stresses are likely
to vary exponentially away from the side of the footing.

Shallow foundations / 251

To overcome this problem thinner elements could be used to represent the soil
immediately adjacent to the side of the footing. Alternatively, interface elements
could be used, as shown in Figure 6.52. These are given normal and shear
stiffnesses compatible with that of the soil (i.e. K,,=K,=10SkNlm3)and a shear
strength of 100kPa (i.e. the same as the soil). (Note that the vertical and horizontal
scales are different in Figure 6.52 and that, for clarity, the zero thickness interface
elements have been drawn with a finite thickness). Repeating the analysis now
gives an ultimate footing load of 9307kN. The contribution from the shear stresses
on the sides of the footing is 4945kN, which is now less than the maximum
possible value of 5026kN and therefore acceptable. (Note that for the maximum
possible value to be mobilised, the major principal stress would have to make
everywhere an angle of 45" to the vertical sides of the footing). Comparing the
contribution from the base of the footing with that from a similar footing located
at the ground surface gives a depth factor d, of 2.2.
The above problem is less severe
in the analyses of a rough strip 2 I
footing and clearly does not occur in 2 2
the analyses of both smooth strip and "
circular footings, as the shear stress $
8
along the side of the footing is zero. 3
While the analyses of the rough 3 a
circular footing with interface A
elements is an improvement over the
-200
0
200
400
600
original analysis, it is still flawed as
Horizontal stress on side of footrog, kPa
tensile horizontal total stresses are
predicted between the sides of the
footing and the adjacent soil. This is Figure 6.53: Horizontal total stress
on side of footing with tension
shown in Figure 6.53, which presents
allowed in interface elements
the distribution of total horizontal
stress down the side of the footing at
failure. These stresses are taken from
the interface elements. If the soil is 2
.;;'
unable to sustain such tensile stresses
then the analysis is unrealistic. In
reality the soil is likely to separate 2
'
from the footing, leaving a vertical
crack.
o& ' 6
To overcome this problem, the '
interface elements positioned down
o
200
400
60
norizo~talstress on side of footing, kPa
the side of the footing can be given a
zero tensile strength. This means that
they can sustain compressive normal Figure 6.54: Horizontal total stress
stresses, but will open when the stress on side of footing with no tension
becomes zero, essentially modelling a
a110 wed in interface elements

,
,

252 1 Finite element analysis in geotechnical engineering: Application

vertical crack. The distribution of

24
22
horizontal total stress down the side
of the footing at failure, from an
2.0
analysis performed with such 2 1 8
interface elements, is shown in Figure 2
6.54. Zero tensile stresses are % I
indicated to a depth of 2.5m below
l
the ground surface and consequently
l
separation occurs over this depth. The
0
2
4
6
8
10
Ratio of foundattondepth to width, DIB
ultimate footing load from this
analysis was 8613kN and the depth
Figure 6.55: Depth factor from FE
factor d, = 2.14.
study compared to Skem~ton's
Potential separation of soil from
relationship
the side of the footing also occurs for
smooth footings. Analyses for such
footings should therefore also include interface elements as shown in Figure 6.52.
To model the smooth nature of the interface between the soil and footing, these
elements are given very low shear stiffness and strength values (e.g. ideally zero
values should be given). To model separation of the soil from the footing these
elements are again given a bulk stiffness consistent with that of the soil and a zero
tensile stress capacity.
Interface elements were used in all the analyses for the parametric study on the
influence of foundation depth. The results of this study in terms of the variation of
depth factor, d, , against normalised foundation depth, DIB, are shown in Figure
6.55. Also shown on this figure for comparison is Skempton's curve. The results
for the strip footing are in reasonable agreement with Skempton's curve, while
those for the circular footing give higher values of d,. This agrees with results from
pile tests and the recent work of Randolph et al. (2000), which indicate higher
values than implied by Skempton's curve.

Example: The leaning Tower of Pisa

6.6.3
6.6.3. 1 Introduction
In 1989 the civic tower of Pavia, in Italy, collapsed without warning, killing four
people. This prompted the Italian Minister of Public Buildings and Works to
appoint a Commission to advise on the stability of the Pisa Tower, which was
perceived to have a high risk of collapse. The Commission recommended closure
of the Tower to the general public and this was instituted at the beginning of 1990.
There was an immediate outcry by the Major and citizens of Pisa who, correctly,
foresaw the damage that the closure would inflict on the economy of Pisa, heavily
dependent on tourism as it is. In March 1990 the Prime Minister of Italy set up a
new Commission to develop and implement measures for stabilising the Tower.
One of the early decisions of the Commission was to develop a numerical model
of the Tower and the underlying ground that could be used to assess the
effectiveness of various possible remedial measures. The purpose of this section

Shallow foundations 1 253

is to describe this model, the work that went into its calibration and how it was
used to investigateboth the temporary lead counterweight solution and some ofthe
various permanent solution options.
The only practical means of calibrating the model was to attempt to get it to
match the history of inclination of the Tower during and subsequent to its
construction. Therefore, the fnrst part of this section is devoted to a brief
description of the history of the Tower. The differences between bearing capacity
failure and leaning instability are then discussed before the numerical model is
described and some of the results presented. Further details of the analyses can be
found in Burland and Potts (1994) and PoMs and Burland (2000).

6.6.3.2 Details of the Tower and ground profile

Figure 6.56 shows a cross-section
through the Tower. It is nearly 60m high
and the foundations are 19.6m in
diameter. The weight of the Tower is
14500t. At present the foundations are
inclined due south at 5.5' to the
horizontal. The average inclination ofthe
axis of the Tower is somewhat less, due
its slight curvature. The seventh cornice
overhangs the first cornice by about
4. lm.
Construction is in the form of a
I.,
hollow cylinder. The inner and outer
FIO
surfaces are faced with marble, and the 'n"men
annulus between these facings is filled
with rubble and mortar within which
extensive voids have been found. A spiral
staircase winds up within the annulus.
Figure 6.56 clearly shows that this Figure 6.56:Schematic crossstaircase forms a large opening on the
section through Pisa Tower
south side, just above the level of the first
cornice where the cross section of the masonry reduces. The high stresses within
this region are a major cause of concern and could give rise to an instantaneous
buckling failure of the masonly without warning. In the summer of 1992 this
masonry was stabilised by applying lightly prestressed steel strands around the
Tower in the vicinity of the first cornice. At present, the masonry is being
consolidated by grouting and the temporary steel strands will soon be reduced in
number.
Figure 6.57 shows the ground profile underlying the Tower. It consists of three
distinct horizons, the properties ofwhich are described in detail later in this chapter
and in AGI (1991). Horizon A is about 10m thick and primarily consists of
estuarine deposits laid down under tidal conditions. As a consequence, the soil

254 1 Finite element analysis in geoteehnical engineering: Application

+IStypes consist of rather variable sandy
and clayey silts. At the bottom of
o
Horizon A is a 2m thick medium
S.ady .adclay
dense fine sand layer (the upper g
Upper clay (Pauwne)
sand). Based on sample descriptions .g
and piezocone tests the material to the
F3 =:$:%
3
south of the Tower appears to be
-30Lower clay
more silty and clayey than to the
north and the sand layer is locally
hwer,,
3
thinner.
Horizon B consists of marine clay
which extends to a depth of about
Figure 6.57: Soil profile beneath
40m. It is subdivided into four distinct
Pisa Tower
layers. The upper layer is a soft
sensitive clay known as the Pancone. It is underlain by a layer of stiffer clay (the
intermediate clay), which in turn overlies a sand layer (the intermediate sand). The
bottom of Horizon B is a normally consolidated clay known as the lower clay.
Horizon B is laterally very uniform in the vicinity of the Tower.
Horizon C is a dense sand which extends to a considerable depth (the lower
sand). The water table in Horizon A is between l m and 2m below the ground
surface. Pumping from the lower sand has resulted in downward seepage from
Horizon A, with a vertical pore water pressure distribution through Horizon B
slightly below hydrostatic.
The many borings beneath and around the Tower show that the surface of the
Pancone clay is dished beneath the Tower, from which it can be deduced that the
average settlement is approximately 3m.

6.6.3.3 History of construction

The Tower is a campanile for the Cathedral, construction of which began in the
century. Work on the Tower began on 9'hAugust 1 173 by the
latter half of the l Ith
modern calendar. By about 1 178 construction had progressed to about one quarter
of the way up the fourth storey when work stopped. The reason for the stoppage
is not known, but had it continued much further the foundations would have
experienced an undrained bearing capacity failure. The work recommenced in
about 1272, after a pause of nearly l00 years, by which time the strength of the
ground had increased due to con'solidation under the weight of the Tower. By
about 1278 construction had reached the 7" cornice, when work again stopped due
to military action. Once again there can be no doubt that, had work continued, the
Tower would have fallen over. In about 1360 work on the bell chamber was
commenced and was completed in about 1370 - nearly 200 years after
commencement of the work.
It is known that the Tower must have been tilting to the south when work on
the'bell chamber began, as it is noticeably more vertical then the remainder of the
Tower. Indeed, on the north side there are four steps from the seventh cornice up

Shallow foundations 1 255

to the floor of the bell chamber, while on the south side there are six steps. Another
important detail of the history of the Tower is that in 1838 a walk-way was
excavated around the foundations. This is known as the catino and its purpose was
to expose the column plinths and foundation steps for all to see, as was originally
intended. This activity resulted in an inrush of water on the south side, since here
the excavation is below the water table, and there is evidence to suggest that the
inclination of the Tower increased by as much as a half of a degree as a result.

6.6.3.4 History o f tilting

As mentioned in the Introduction, the only possible means of calibrating a model
of the Tower is to attempt to simulate the history of tilting of the Tower during and
subsequent to its construction. Hence it was necessary to learn as much as possible
about the history of the tilt of the Tower. The only reliable clues on the history of
tilt lie in the adjustments made to the masonry layers during construction and in the
shape of the axis of the Tower.
Based on the measured thickness
+
of each masonry layer and a
m
F
W
o
S 5W
m
E
hypothesis on the manner in which
l5
r!
?
-------.L-a
the masons corrected for the
progressive lean of the Tower,
Burland (see Burland and Potts
(1994)) deduced the history of
inclination of the foundations of the
Tower shown in Figure 6.58. In this
figure the weight of the Tower is "
plotted against the deduced
- Loading
----. Consolidation
inclination. During the first phase of S
construction to just above the third
*
Plane strain
3D analysis
cornice (1 173 to 1178), the Tower
0
Historical
inclined slightly to the north. The
northward inclination increased
slightly during the rest period of
0
1
2
3
4
5
6
nearly 100 years to about 0.2". When
Inclination of foundations (deg)
construction recommenced in about
1272, the Tower began to move
Figure 6.58: Predicted and historical
towards the south and accelerated
inclination of the Tower during and
shortly before construction reached
after construction
the seventh cornice in about 1278
when work again ceased, at which stage the inclination was about 0.6" towards the
south. During the next 90 years the inclination increased to about 1.6". After the
completion of the bell tower in about 1370, the inclination of the Tower increased
significantly. In 1817, when Cressy and Taylor made the first recorded
measurement with a plumb line, the inclination of the Tower was about 4.9". The
excavation of the catino in 1834 appears to have caused an increase in inclination
A
,

256 1 Finite element analysis in geotechnical engineering: Application

of approximately 0.5" and the present day inclination of the foundations is about
5.5". It can be seen from Figure 6.58 that significant inclination ofthe Tower only
began once the height exceeded the sixth cornice. If the inclination had been due
to much more compressible ground beneath one side than the other, it would have
developed much earlier. Therefore, another explanation for the rapid onset of
inclination is required and will be discussed later. It is the history of inclination
depicted in Figure 6.58 which was used to calibrate the numerical models
described later in this section.
For most ofthis century the inclination ofthe Tower has been increasing. These
changes in inclination are extremely small compared with those that occurred
during and immediately following construction. The rate of inclination of the
Tower in 1990 was about 6 seconds per annum. The cause of the continuing
movement is believed to be due to fluctuations ofthe water table in Horizon A. No
attempt has been made to model these small movements.

6.6.3.5 The motion of the Tower foundations

Previously, studies have concentrated
I
on the changes of inclination of the
I
Tower. Little attention has been
devoted to the complete motion of the
foundations relative to the Centreo
surrounding ground. The theodolite 'Oation
and precision levelling measurements
made in the last century help to
clarifLthis. These observationscan be
used to define the rigid-body motion
of the Tower during steady-state
rotation, as shown in Figure 6.59 (see
Burland and Potts (1 994)).
I
It can be seen that the Tower is
rotating about a point a ~ ~ r o x i m a t e Figure
l~
6-59:Motion of the Tower
located level with point V, and
vertically above the centre of the foundation. The direction of motion of points F,
and F, are shown by vectors and it is clear that the foundations are moving
northwards with F, rising and F, sinking. It can therefore be concluded that the
seat of the continuing long term tilting of the Tower lies in Horizon A and not
within the underlying Pancone clay, as has widely been assumed in the past.

6.6.3.6 Stability of tall towers

Before considering the numerical analysis of the Pisa Tower, it is illuminating to
consider the possible mechanisms of failure associated with the stability of the
foundations of a tall tower. There are two possible mechanisms that could account
for failure of such a tower: (i) bearing capacity failure due to insufficient soil
strength, and (ii) leaning instability due to insufficient soil stiffness. Bearing

Shallow foundations / 257

capacity failure is the more common type of instability and the one covered in most
text books and codes of practice. All the analyses presented so far in this chapter
showed this type of failure. Leaning instability is not so common and is only
relevant to tall structures. It occurs at a critical inclination when the overturning
moment, generated by a small increase in inclination, is equal to or greater than,
the resisting moment of the foundations generated by the same rotation. In all but
the simplest of cases it is difficult, probably impossible, to analyse without using
numerical analysis.
These two alternative failure
mechanisms are best demonstratedby
a simple example. Figure 6.60 shows
a simple tower resting on a uniform
deposit of undrained clay. The clay is
modelled as a linear elastic Tresca
material, with an undrained strength
Sl,=80kPa. The dimensions of the
tower are similar to those of the Pisa
Tower. To trigger a rotation failure
some initial defect (imperfection)
must be present. In this example the
tower was given an initial tilt of 0.5"
(i.e. the initial geometry of the tower
had a tilt). The self weight of the
tower was then increased gradually in
a plane strain large displacement
Figure 6.60: Geometry of simple
finite element analysis.
tower
Three analyses were performed,
each with a different value of shear
4
stiffness, G, of the soil, and the results
are presented in Figure 6.6 1. Here the
increase in rotation of the tower
8
above the initial 0.5" imperfection is .g 2
plotted against weight of the tower, S
1
for analyses with GIS,, values of 10,
100, 1000. Real soils are likely to
have properties such that they lie
Weight of Tower (MN)
between the two extreme values. It
should be noted that in all cases the
Figure 6.6 1: Rotation of simple
strength of the soil S,(was SOkPa. The
results show that failure occurs very
tower for varying soil stiffness
abruptly, with little warning and that
the weight ofthe tower at failure is dependent on the shear stiffness of the soil. The
weight at failure for the analysis with the softer soil, GISlt=IO,is about half of that
for the analysis with the stiffest soil, G/S,=1000.

258 / Finite element analysis in geotechnical engineering: Application

It is of interest to examine the

analyses with the two extreme values
of G/S,in more detail. In particular, it
is instructive to consider what is
happening in the soil at failure. Figure
6.62 shows vectors of incremental
displacements for the soft soil. The
vectors represent the magnitude and
direction of the displacements from
the last increment of the analysis.
This figure shows that the movements
are located in a zone below the Figure 6.62: Pattern of mOVement
a* failure for
with a low
foundation and indicate a rotational
stiffness
type of failure. At first sight this looks
like a plastic type collapse
mechanism. However, examination of
the zone in which the scil has gone
plastic (also shown on Figure 6.62),
indicates that it is very small and not
2 *.,.,,..
.
.*',,"""
consistent with a plastic failure
. .
mechanism. Consequently, this figure
. .
,,, ...--.:.:::
, . . .. . .
. . . , ' :, , ,.::.:--:
- .: ,: :. .
indicates a mechanism of failure
.. . .. , .. +. . , . . _ + . . .
consistent with a leaning instability.
. . . . . . . . . .
In view of the temporary
counterweight scheme, which
involved adding lead weights to the
Figure 6.63: Effect of
north side of the Pisa Tower and
counterweight
for soit with a 10 W
which will be discussed in more detail
stiffness
later in this chapter, it is of interest to
examine the response of the simple
tower in the above example if, at the
point of collapse, weight is added to
the higher side of the foundation. The
effect of a 1.5MNIm load is shown in
Figure 6.63. Again vectors of
incremental displacement are shown.
These indicate the nature of the
movements due only to this additional
load. It is noted that under this load
the sense of movement is reversed
and the tower rotates back and
Figure 6.64: Pattern of movement
collapse is arrested.
at failure for soil with a high
Considering the results from the
stiffness
analysis performed with the stiffer
"

_ * _ * . . I . .

: : h

.__O_.a

( . , . . . _ . . . . I . . .

Shallow foundations 1 259

soil, vectors of incremental
displacement just before collapse are
shown in Figure 6.64. The mechanism
of failure indicated by these vectors is
vely different to the one shown in
Figure 6.62 for the softer soil. Instead
of the soil rotating as a block with the
foundation, the vectors indicate a
more traditional bearing capacity type
mechanism, with the soil being
pushed outwards on both sides. The
Figure 6.65: Effect of
plastic zone, also indicated on Figure
counterweight
for soil with a high
6.64, is very large and therefore the
results clearly indicate a plastic
stiffness
bearing capacity type mechanism of
failure.
If, as before, load is applied to the higher side of the foundation at the point of
collapse, the vectors of incremental displacement given in Figure 6.65 are
obtained. These show that, in contrast to the softer soil analysis, the tower
continues to increase its inclination. In fact, it was not possible to obtain a
converged solution when the weight was added. The addition of the load initiates
collapse even though the load acts to reduce the overturning moment.
As well as demonstrating the difference between the two types of instability,
these analysis also indicate that a counterweight type scheme will only be
beneficial to the Pisa Tower if it is suffering a predominately leaning instability.
To complicate matters further, real soils are likely to have stiffness values between
the two extremes considered above and therefore both mechanisms of behaviour
are likely to be active to some degree. In this respect the motion of the Pisa Tower
discussed in Section 6.6.3.5 and shown in Figure 6.59 is more consistent with the
movements shown in Figure 6.63 than those shown in Figure 6.65, indicating that
the Tower is probably suffering predominately from a leaning instability.

6.6.3.7 Soil properties

The constitutive model chosen for the clay strata was a form of modified Cam clay
(Roscoe and Burland (1968)), in which the shapes ofthe yield and plastic potential
surfaces in the deviatoric plane are given by a Mohr-Coulomb hexagon and a circle
respectively (see Section 7.9 of Volumel). For the sand layers a Mohr-Coulomb
model was used. Fully coupled consolidation for all the soil layers was
incorporated into the analyses. In order to implement these models the following
soil parameters are required:
Ysat

(Pc.$'

C,

saturated unit weight,

critical state angle of shearing resistance,
compression index (note C, = 2.3025;1),

260 / Finite element analysis in geotechnical engineering: Application

c,
Glp,,'

eI
k
K,,
OCR
v

swelling index - taken as O.lCc (note C, = 2.3025~),

elastic shear modulus,
void ratio on the VCL whenpl=l kPa (note e, = v,- 1.O) ,
permeability,
coefficient of earth pressures at rest,
overconsolidation ratio,
angle of dilation (sands).

Given the above parameters, the value of the undrained strength S,, can be
determined from Equation (6.9). In Horizon A (see below) S,, turned out to be a
more reliable parameter than OCR and was therefore specified in its place.
The soil profile at Pisa was characterised in detail by the Polvani Commission
(Minister0 dei Lavori Pubblici 1971). Horizons A and B were divided into a
number of sub-layers, the descriptions of which are as follows:
Horizon A:
MG Top soil and made ground,
A1
Loose to very loose yellow sandy silt to clayey silt without
stratification,
Uniform grey sand with interbedded clay layers, broken fossils - Upper
A2
sand,
Horizon B:
Highly plastic grey clay with fossils,
B1
Medium plastic grey clay with fossils,
B2
Highly plastic grey clay with fossils,
B3
B4
Dark grey organic clay,
Blue grey to yellow silty clay with calcareous nodules,
B5
Grey, sometimes yellow, sand and silty sand - Intermediate sand,
B6
Medium to highly plastic clay, with fossils and thin sand layers in the
B7
upper part,
Grey clay with frequent thin sand lenses,
B8
Blue grey silty clay with yellow zones; calcareous nodules; some dark
B9
organic clay at centre,
B10 Grey clay with yellow zones; fossils in the lower part.
Laterally Horizon B is very uniform. However, there is much evidence to show that
in Horizon A, layer A I changes from predominantly silty sands and sandy silts
north of the Tower to clayey silts south of the Tower. There is also some evidence
from piezocones that the Upper sand layer A2 thins just south of the Tower. A
careful study of the detailed sample descriptions given in the Polvani Report
suggests that beneath the Tower, in Horizon A, there exists a lense of clayey silt
which thins from south to north.
It will become evident that it is the compressibility of the underlying soils
which has played the dominant role in the historical behaviour of the Tower (i.e.
leaning instability), so that emphasis is placed in this section on these properties.
Two major programmes of thin wall sampling and testing have been carried out,

Shallow foundations / 26 1
one in 1971 (at the instigation of the Polvani Commission) and another in 1986.
The results of these studies have been summarised by Calabresi et al. (1993) and
Lancellotta and Pepe (1990) respectively. Figures 6.66a, b and c show the
experimental values of C,', C, and OCR respectively, where C,' is the
compressibility of the reconstituted material - defined by Burland (1990) as the
intrinsic compressibility.The values of C,' were derived from the correlation with
the water content at the liquid limit, W,, established by Burland. Also shown on
Figure 6.66 are the sub-layers established by the Polvani Commission and
described above. It can be seen from Figure 6.66a that the values of C,' are
reasonably well defined for each sub-layer and the average values are shown by the
vertical lines. It is of interest to note that sub-layer B7 is made up of two distinct
soil types and should perhaps be considered as two separate sub-layers.
o 0,

c,'
0;s

1.:

cc
9.5

1.0 0I

OCR
f

Figure 6.66: Experimental values of Cc*, Cc and OCR;

values used in the analysis are shown as full lhes

It can be seen from Figure 6.66b that the experimental values of C,, determined
from high quality samples of the natural clay, show considerable scatter and it is
not easy to decide on suitable representative values for analysis. The use of the
intrinsic compressibility C', has been particularly useful in this respect (Burland
and Potts (1994)). It is well known that measured values of C, are particularly
sensitive to sample disturbance and the scatter in Figure 6.66b is a reflection of
this. Nash et al. (1992) presented the results of oedometer tests on a sensitive
marine clay from the Bothkennar test bed site and showed that the values of C,
obtained from high quality block samples were significantly higher than for other
sampling methods including thin wall sampling. Nash7sresults showed that the
measured values of C, for the best samples were between 1.9 and 2.3 times larger
than C,". In Figure 6.66b the vertical lines were obtained by multiplying the
average values of C,' by a factor depending on the plasticity of the material. For
the Pancone clay (B1 to B3) the factor was 2, for the Lower clay (B7 to B10) the

262 l Finite element analysis in geotechnical engineering: Application

factor was 1.5 and for Horizon A and the Intermediate clay (B4 and B5) the factor
was unity. It is these values that have been used in the finite element analyses and
it can be seen that they tend to lie at, or a little beyond, the upper limits of the
experimental values, as is to be expected.
Figure 6 . 6 6 ~shows experimental values of OCR. The vertical lines are the
values used in the finite element analyses. Values were chosen near to the upper
limit of the experimental values, as the effect of sample disturbance is to reduce the
value somewhat.
Table 6.5 gives the values of all the soil parameters that were used in the
analyses. The mean values of q,' for each sub-layer were obtained from undrained
was taken as 0.1 for all the soil layers. The values of permeability
triaxial tests. 1~12
were derived from the oedometer tests, taking values in the upper quartile of the
range of results for each layer. The values of Glp,' were chosen on the basis of
experience with small strain testing of a wide variety of materials. The choice of
values was not important for modelling the history of inclination ofthe Tower, but
proved to be more crucial when predicting its response to small perturbations of
load in its present condition. The water table was assumed to be Im below ground
level and in hydrostatic equilibrium - a condition that must have existed at the time
of construction.
Table 6.5: Soil parameters used in the analysis

Shallow foundations 1 263

6.6.3.8 Finite element analysis

Geometry
It must be emphasised that a prime objective of the analysis was to develop an
understanding of the mechanisms controlling the behaviour of the Tower. It was
felt that until these had been clarified, it would be unhelpful to attempt highly
sophisticated and time consumingthree dimensional analysis.Accordingly, a plane
strain approach was used initially, recognising that the interpretation of the results
would require some care. Some of the earlier analyses were carried out using a
large strain formulation, but it was found that this introduced time consuming
complications in relation to the excavation of the catino and also relative ground
water level changes (i.e. settlements exceeded the depth of the original ground
water table). The results did not differ significantly from traditional infinitesimal
strain analysis, which was therefore adopted for the work described here. Later a
limited number of three dimensional analyses were carried out using the Fourier
series aided finite element method (see Chapter 12 of Volume I). The results of
both the plane strain and three dimensional analyses are reported here.
The layers of the finite element mesh matched the soil sub-layering discussed
above. The lower sand was assumed to be rigid but permeable. The full mesh, for
the plane strain analyses, is shown in Figure 6.67 and extends laterally 100m either
side of the axis of the Tower. In Horizon B the soil was assumed to be laterally
homogeneous. However, a tapered layer of slightly more compressible material
was incorporated into the mesh for layer A1 as shown in Figure 6.68, which shows
a detail of the mesh in the vicinity of the Tower. The shaded elements beneath and
to the south have the properties listed in Table 6.5 for layer A l ', with C,=0.25 and
k=lO-gmls.The remaining elements in layer A1 have the properties listed for layer
A1 ",with C,=0.15 and k-10Smls.In applied mechanics terms the insertion of this
slightly more compressible tapered layer beneath the south side may be considered
to be an 'imperfection'.

Figure 6.67: Finite element mesh

For the three dimensionalFourier series aided analysis the geometry is assumed
to be axi-symmetric. However, the spatial distribution of soil properties and
loading can be three dimensional. A mesh similar to that shown in Figure 6.67, but
only considering the geometry to the right of the centre line of the Tower, was
used. However, no tapered layer was incorporated to provide an 'imperfection'.
Instead, the soil properties were assumed to vary linearly beneath the Tower in
sub-layer A 1. To the north of the Tower the properties were as listed for layer A 1"

264 / Finite element analysis in geotechnical engineering: Application

in Table 6.5, to the south as listed for

layer Al', and beneath the Tower
they varied linearly from north to
south from those listed for layer A l "
to those listed for layer A l
Note that, in modelling the
foundation of the Tower, the central
hole was neglected for the plane strain
analysis, since it only represents a
small proportion of the total area of
the foundation.
l.

Figure 6-68..it^ element mesh

close to Tower foundation

Foundation overturning moment

If any inclination 0 of the Tower takes place, the centre of gravity moves
horizontally and its weight W generates an overturning moment M. If the height of
the centre of gravity above foundation level is h,, then:

I, is a correction factor which takes account of the ratio between the second
moments of area of a rectangular and a circular foundation. For a rectangular
foundation of width 19.6m and the same area as the foundation of the Pisa Tower,
the value of I, for rotation about the centre is 1.266. For the three dimensional
analyses in which the circular shape of the foundation is modelled I, should
theoretically be unity. Equation (6.12) was incorporated into the analysis such that
any inclination of the foundations during or subsequent to construction
automatically resulted in the application of an appropriate overturning moment to
the foundations.

Analyses
All analyses involved coupled consolidation.The calibration analyses were carried
out in a series of time increments in which loads were applied to the foundation to
simulate the construction history of the Tower together with the rest periods, as
summarised Section 6.6.3.4. During a construction period it was assumed that the
load was applied at a uniform rate. The excavation ofthe catino was also simulated
in the plane strain analysis.
The only factor that was adjusted to calibrate the model was the factor I, in
Equation (6.12). For the first run, the value of l, was set equal to unity. At the end
of the run the final inclination of the Tower was found to be less than the present
value of 5.5". A number ofruns were carried out with successive adjustments being
made to the value of l, until good agreement was obtained between the actual and
predicted value of the final inclination. It was found that, for the plane strain
analysis with a value of I, = 1.27, the final calculated inclination of the Tower was
5.44". Any further increase in l, resulted in instability of the Tower. It is therefore
clear from this analysis that the Tower must have been very close to falling over.

Shallow foundations / 265

The final value of I, is very close to the theoretical value for rotation about the
centroid, but this is probably coincidental.
For the three dimensional analysis a final inclination of 5.48" was obtained with
a calibration factor I, of 1.7. As noted above, theoretically I, should be unity for
these analyses. However, in these three dimensional analyses the excavation of the
catino was not simulated.

"7

6.6.3.9 Simulation of the history of inclination

Figure 6.69 shows a plot, from the
60
1272 1360
plane strain analysis, of the predicted
0
b
, , :
Predictions changes in inclination and sett1ement g 4
of the Tower with time since the start
2,
Observations
of construction in 1173. It is
,o,
Date
important to appreciate that the only
point that has been pre-determined on
this plot is the final inclination of
Predictions
about 5.5". All other displacements $
f
were generated by the analysis. It can
be seen that completion of the Tower
up to the seventh cornice (-1278)
Figure 6.69: Relationship between
results in quite small inclinations. It is
time, inclination and settlement for
only when the bell chamber is added
plane strain simulation of the
in 1360 that the inclinations increase
history of the Pisa Tower
dramatically. Also of considerable
interest is the fact that simulation of the excavation of the catino results in a
significant increase in inclination of about %". The final settlement of the
foundations is seen to be about 3.8m, which is larger than the value deduced from
the depression in the surface of the Pancone clay (Figure 6.57). The reason for this
is thought to be due to the plane strain analysis. The depth of influence of a circular
foundation would be less and would result in a smaller settlement.
The results of the analysis are also plotted on Figure 6.58 as a graph of load
against inclination and can be compared with the deduced history. Again it must
be emphasised that it is only the present inclination that has been fitted. The
agreement between the simulated and historical behaviour is remarkable and gives
considerable confidence in the reliability of the computer model. A striking
difference is that the model does not predict the initial northerly inclination of the
Tower. This is not felt to be of importance for the intended application of the
model. It was found that, during the early stages of loading, the model did show
a small inclination to the north. This was due to the fact that consolidation of the
thin northern end ofthe tapered layer of compressible soil took place more rapidly
than the thicker southern end. It should be possible to devise a soil profile in
Horizon A that more accurately simulates the early history of inclination of the
Tower, but this was outside the scope of the project.
Also shown on Figure 6.58 are the results from the three dimensional analysis.

'3

,,

,,

,,,, ,,,,

266 1 Finite element analysis in geotechnical engineering: Application

It can be seen that the agreement

between the plane strain and three
dimensional analysis is very
satisfactory. The final settlement for
the three dimensional model is 3. lm,
which is much closer to the deduced
value than the lane strain model.
Figure 6.70 shows zones of fully
Fully mobilised
mobilised strength at the end of the
strength
plane strain analysis following
complete pore water pressure
dissipation after excavation of the
Figure 6.70: Zones of fully
catino. As mentioned previously, at mobilised strength at the end of the
this stage the Tower is in a state of
analysis
unstable equilibrium. Surprisingly,
there are no zones of contained failure within the Pancone clay, but there are
extensive zones within Horizon A. The lowest zones are in the upper sand layer
and result from the lateral extension ofthis layer. There is a large zone beneath and
outside the southern edge of the foundation and a smaller zone underneath the
northern side. It is evident from this figure that the impending instability of the
Tower foundations is not due to the onset of a bearing capacity failure within the
Pancone clay. The cause of the instability can be attributed to the high
compressibility of the clay which results in leaning instability.
Figure 6.71 shows the distribution
metxes
of vertical effective contact pressure
acting on the foundation, g O corresponding to complete pore water "^
g
pressure dissipation after excavation
of the catino in the plane strain P m?!
analysis. The distribution is far from y
linear. Beneath the southern edge of
the foundation the effective contact
'm- B e f o r e installation of countenueight
----Atter installationof countenve~ght
pressure is about 850kPa. At the
northern edge the effective contact
Figure 6.71: Distribution of
pressure is zero over approximately
the first metre, but further south it effective foundation contact stress
increases rapidly.
In summary, the finite element models described above give remarkable
agreement with the deduced historical behaviour of the Tower. It is important to
emphasise that the history of foundation inclinations and overturning moments
were self generated and were not imposed externally in a pre-determined way. The
only parameter that was used to calibrate the model was the present inclination of
the Tower. The analyses have demonstrated that the lean of the Tower results from
the phenomenon of leaning instability due to the high compressibility of the
+

Shallow foundations / 267

Pancone clay. The role of the tapered layer of slightly increased compressibility
beneath the south side ofthe foundations in the plane strain analysis and the linear
variation of properties assigned to sub-layer A1 in the three dimensional analysis,
is to act as an 'imperfection'. Its principal effect is to determine the direction of
lean rather than its magnitude. The model provides important insights into the basic
mechanisms of behaviour and has proved valuable in assessing the effectiveness
of various proposed stabilisation measures. Its role in evaluating the effectiveness
of the temporary counterweight solution will now be described.

6.6.3.10 Temporary counterweight

Fully aware that the selection, design and implementation of permanent
stabilisation measures for both the foundations and the masonry of the Tower
would take a long time, the Commission took an early resolution to implement
short-term temporary and fully reversible measures to increase slightly the stability
of these elements (Burland et al. (1993)). The use of lightly prestressed steel
tendons to stabilise the masonry was described earlier.
The similarity between the motion of the
Tower depicted in Figure 6.59 and the
movements associated with leaning instability
shown in Figure 6.62 suggested a possible
temporary means of increasing the stability of
the foundation. The observationthat the northern
side of the foundation had been steadily rising
led to the suggestion that application of load to
the foundation masonry on the north side could
be beneficial in reducing the overturning
b)
moment. Clearly, such a solution would not have
been considered if it had not been recognised
that leaning instability, rather than bearing
capacity failure, was controllingthe behaviour of
initial >\.,
the Tower. Before implementing such a solution
yield
surface
it was obviously essential that a detailed analysis
should be carried out. The purpose of such an
P,' PoY1P'
analysis was two-fold: firstly to ensure that the
in
proposal was safe and did not lead to any ,cigure6-72..a) increase
undesirable effects and secondly to provide a yield stress due to aging;
class A prediction which could be used to assess b)
shift in
the observed response of the Tower as the load
surface due to aging
was being applied.
Both the plane strain and three dimensional analyses were therefore extended
to simulate the addition of a counterweight to the north side of the Tower. As
explained by Potts and Burland (1994), before simulating the addition of the
counterweight it was necessary to account for ageing of the clay layers over the
past 100 years or so, since the end of construction. There is much evidence to show

1 *.y

268 1 Finite element analysis in geotechnical engineering: Application

that ageing significantly increases the stiffness of clays. For one dimensional
compression Leonards and Ramiah (1959) showed that the process of ageing
,
' as illustrated in Figure 6.72a. More
results in an increase in the yield stress a
generally a shift in the yield surface takes place as illustrated in Figure 6.72b. The
~ ' the intersections with the p' axis of the yield surfaces
parameters p,' a n d ~ , ,are
of a 'young' and the 'aged' clay respectively. The ratiopV1lp,,'is a measure of the
degree of ageing and is defined as the yield stress ratio. The effects of ageing can
be introduced into the finite element analysis by increasing the current value ofp,,'
for each integration point to give a prescribed value of the yield stress ratio.
In the laboratory, significant ageing effects have been observed over periods
of a few days. Leonards and Ramiah found that the yield stress ratio for a
reconstituted clay that was allowed to age for about 90 days was as high as 1.3. In
the present analysis the effect of introducingvarious values of yield stress ratio was
studied. The results presented here are for a value of 1.05, which was felt to be
conservative. It should be noted that ageing of the Pancone clay increases the
stability of the Tower foundations significantly.
The full line in Figure 6.73 shows
the predicted response of the Tower Ae(deg)
due the application of a counterweight 0,08
to the foundation masonry, at an
eccentricity of 6.4m to the north, after
allowing ageing of the Pancone and
Lower clays to give a yield stress
ratio of 1.05. No ageing of Horizon A
o
was assumed. The soil parameters are
e=6.4m
those given in Table 6.5. At the
design load of about 690t the
---- e = 7 . 8 m
inclination of the Tower is predicted
to reduce by about 27.5 seconds of Senlement(mm)
arc with a settlement of 2.4mm. More
importantly, the overturning moment
Figure 6.73: Predicted response o f
is reduced by about 14%. In Figure
the lower due to application o f
6.71 the broken line represents the
counterweight
predicted effective contact pressure
beneath the foundation after application of the counterweight.
Also shown in Figure 6.73 are the effects of increasing the load above the
planned level and the results are of considerable interest. It can be seen from the
full lines that, as the load is increased, the rate of increase of inclination to the
north reduces, becoming zero at about 1400t. With hrther increase in load the
movement reverses and the Tower begins to move towards the south. The
settlement rate also begins to increase once the load exceeds 1400t. Figure 6.73
also shows the results of increasing the eccentricity of the counterweight. At an
eccentricity of 9.4m (dotted lines) the Tower continues to rotate northwards as the
counterweight is increased. At an intermediate eccentricity of 7.8m (broken lines)
a curious response is obtained in which the Tower first moves northwards, then it

,,

Shallow foundations 1 269

reverses and then reverses again. It can be seen that the settlement response is
similar for each eccentricity. It would be wise to ensure that (irrespective of the
eccentricity) a significant rate of increase of settlement is avoided as this implies
the onset of yield.

6.6.3. 1 1 Observed behaviour during application of the

counterweight

A description of the counterweight

scheme is given by Burland et al. 800(1993). It consists of a temporary
prestressed concrete ring cast around 3the base of the Tower at plinth level.
400This ring acts as a base for supporting <
specially cast lead ingots which were
o
1/6/93
1/9/93
1/12/93
1/2/94
Date
placed one at a time at suitable time
spans. The movements experienced
by the Tower are measured with a
Figure 6.74: History of
highly redundant monitoring system.
counterweight loading
Burland el al. (1994) describe the
response of the Tower to the
application of the counterweight.
Construction of the concrete ring
30 commenced on 31dMay 1993 and the
first lead ingot was placed on 14thJuly 3
1993. Figure 6.74 shows the sequence
of load application. It can be seen
that, after construction ofthe concrete
ring, the load was applied in four
phases, with a pause between each
Observations
phase to give time to observe the
response of the Tower. The final
phase was split in two either side of
the Christmas break. The last ingot
was placed on 20"' January 1994.
Applied load (t)
Figure 6.75 shows a comparison
of the class A predictions from the
plane strain analysis and
measurements of (a) the changes in 9
inclination, and (b) the average
3settlements of the Tower relative to
the surrounding ground during the Figure 6.75: Plane strain prediction
and observed response due to
application of the lead ingots. The
application of counterweight
points in Figure 6.75 represent the
(class A prediction)
measurements at the end of each
A,

270 1 Finite element analysis in geotechnical engineering: Application

phase of loading and the vertical lines
extending from them show the
amount of creep movement between
each phase. For the final phase the
creep after one month is shown. It can
be seen that the class A predictions of
the computer model give changes in
inclination which are about 80% of
the measured values. However, the
settlements are in excellent agreement
with the measurements.
It is perhaps worth emphasising
that the purpose of the model was to
clarify some of the basic mechanisms
of behaviour and it was calibrated
against inclinations measured in
degrees. The use of the model in
studying the effects of the
counterweight was to check that
undesirable and unexpected responses 5;
e
4 3
of the Tower did not occur. In this
respect the model has proved to be
very useful. It has led to a Figure 6.76: Comparison of the 3D

and plane strain

for the
consideration of the effects of ageing
a
~
~
l
i
c
a
t
i
o
of
n
the
counterweight
and it has drawn attention to the
(class C prediction)
importance of limiting the magnitude
of the load so as to avoid yield in the underlying Pancone clay. Also the beneficial
effects of increasing the eccentricity have been highlighted.
It is perhaps expecting too much of the model for it to make accurate
quantitative predictions of movements which are two to three orders of magnitude
less than those against which it was calibrated, and the fact that it has done as well
as it has is remarkable. However, the observed movements due to the
counterweight may be used to further refme the model. It has been found that the
difference between the predicted and measured inclinations are due largely to the
values of Glp,,' in Horizon A used in the analysis and given in Table 6.5.
Considerably improved agreement can be obtained if the values of Glp,,' in
Horizon A are reduced. By trial and error it was found that by reducing these
values by a factor of 0.65, the northward rotation due to a counterweight of 6.9MN
was increased to 37.5 arc seconds, which corresponds to the observed value one
month after the application ofthe final lead weight. The results are shown in Figure
6.76 and the agreement with the observed rotations is much improved compared
with Figure 6.75. It is of interest to note that the reduction in shear modulus in
Horizon A only slightly increases the predicted settlement. The re-calibrated model
is used in the remainder of this section.

Shallow foundations / 271

The three dimensional analyses were carried out after the application of the
counter-weight and the revised values of Glp,' were incorporated in it. The
predictions ofthe response ofthe three dimensional model to the application of the
counterweight and comparison with the plane strain model is shown on Figure
6.76. Excellent agreement is achieved for the changes in inclination. The three
dimensional model gives slightly smaller average settlements than the plane strain
model.

6-6.3.12 Permanent stabilisation of the Tower

Several alternative options for permanent stabilisation of the Tower were
considered and some of these were analysed using the numerical models described
above. One of these options involved the installation of ten ground anchors on the
north side of the Tower. Initially this was suggested as a replacement for the
counterweight solution, however, its use as a permanent solution was also
considered. Analysis of this option indicated that as the anchor loads increased, the
Tower initially rotated towards the north, but that after a relatively small rotation
it reverted to rapidly rotating towards the south. The associated settlements also
increased rapidly at this point. In light of the results for the counterweight solution
presented in Figure 6.73 such a result is perhaps not surprising.
Another permanent solution that was considered was the provision of a north
pressing slab. This involved casting a concrete slab on the ground surface to the
north of the Tower and loading it with ground anchors extending into the Lower
sand. Numerical analysis of this option showed that large loads were required to
cause modest rotations of the Tower and that under some loading scenarios the
Tower could revert to rotating towards the south.
The numerical analysis therefore showed that neither of the above options were
likely to be viable and consequently they were not pursued further. The permanent
solution that was finally adopted involved soil extraction from under the north side
of the Tower. This was extensively analysed and some of the results are presented
below.

6.6.3.1 3 Soil ex traction

Introduction
Under-excavation is the process whereby small quantities of soil are excavated
locally by means of a specially designed drill, so as to leave a cavity which closes
under the overburden pressure. Closure ofthe cavity results in localised subsidence
ofthe overlying ground surface. The technique can be used for inducing controlled
subsidence of a building.
The Commission considered the use of this method as a means of controlled
reduction of the inclination of the Tower and the arrangement is shown
schematically in Figure 6.77. There was considerable uncertainty about the
response of the Tower to the process of soil extraction from beneath the north side
of the foundation, as the Tower is very close to leaning instability. The purpose of

272 / Finite element analysis in geotechnical engineering: Application

the numerical analyses presented here
was to explore the response of the
computer model of the Tower to Dril,ingrig
-22.5
under-excavation and to examine the
\ P
stress changes that the process
induces beneath the foundations and -------------within the soil beneath the south side.
It should be emphasised that the
finite element meshes had not been
developed with a view to modelling
under excavation. The individual
elements are rather large
for Figure 6-77:Schematic presentation
representing regions of extraction. To
of the soil extraction process
have repeated the whole analysis
using a new finer mesh would have been both expensive and very time consuming.
In any case, the purpose of the modelling was to throw light on the mechanisms of
behaviour rather than attempt a somewhat illusory 'precise' analysis.
The objective ofthe simulation of the soil extraction process was to reduce the
volume of any chosen element of ground incrementally, so as to achieve a predetermined reduction in volume of that element. Localised soil extraction was
modelled by selecting a chosen element in the finite element mesh, eliminating its
stiffness and then progressively reducing its volume by the application ofequai and
opposite vertical nodal forces to its upper and lower faces so as to compress the
element. The nodal forces were increased progressively until the desired volume
reduction had taken place. The stiffness of the element was then restored. It is
worth commenting on why only the vertical stresses were reduced. The actual
operation of soil extraction is from a near horizontal element about l m long and
0.2m diameter. With such a long thin element the boundary displacements will be
primarily vertical. As mentioned above, the finite element mesh was not designed
to simulate the process of soil extraction and the length to depth ratios of the
elements are much smaller. So as to encourage vertical displacements ofthe upper
and lower faces, only the vertical stresses were reduced. To date only plane strain
analyses have been performed to simulate soil extraction.

-+@

Critical line
Simple studies carried out on one-g models on sand at Imperial College pointed to
the existence of a Critical line. Soil extraction from any location north of this
Critical line gave rise to a reduction in inclination, whereas extraction from south
of the line gave rise to an increase in inclination. The first objective of the
numerical analysis was to check whether the concept of a Critical line was valid.
Figure 6.78 shows the finite element mesh in the vicinity of the Tower.
Elements numbered 1,2,3,4 and 5 are shown extending southwards from beneath
the north edge of the foundations. Five analyses were carried out in which each of
the elements was individually excavated to give full cavity closure, and the

Shallow foundations 1 273

response of the Tower computed. For
excavation of elements l, 2 and 3 the
inclination of the Tower reduced so
that the response was positive. For
element 4 the response was
approximately neutral with an initial
slight reduction in inclination which,
with further excavation,was reversed.
For element 5 the inclination of the
Tower increased as a result of Figure 6.78: Excavated elements
used to investigate the existence of
excavation.
the Critical line
The above analyses confirm the
concept of a Critical line separating a
positive response from a negative one. For the plane strain computer model the
location of the Critical line is towards the south end of element 4, which is at a
distance of4.8m beneath the foundation ofthe Tower, i.e. about one half the radius
of the foundations.
It was noted that, as the location of excavation moved further and further south
beneath the foundation, the settlement of the south side steadily increased as a
proportion of the settlement ofthe north side. Excavation of elements 1 and 2 gave
a proportion of less than one quarter.
It is concluded from this shdy that, provided soil extraction takes place north
of a Critical line the response of the Tower is positive, even though the foundations
are close to leaning instability. For the plane strain model the Critical line is
located about half the radius of the foundations south of the northern edge of the
foundations. It is further concluded that it should be possible to keep the
settlements of the south edge of the foundations to less than one quarter of the
settlements at the north, provided soil extraction does not extend southwards by
more than about 2m beneath the foundations.

Simulation of under-excavation process

Having demonstratedthat localised soil extraction gives rise to a positive response,
the next stage was to model a complete under-excavation intervention aimed at
safely reducing the inclination of the Tower by at least 0.25'. A preliminary study
of extraction was carried out, using a shallow inclined drill hole beneath the
foundations. Although the response of the Tower was favourable, the stress
changes beneath the foundations were large. Consequently, a deeper inclined
extraction hole was investigated.
The inset in Figure 6.79 shows the finite element mesh in the vicinity of the
foundations on the north side. The elements numbered 6 to 12 were used for
carrying out the intervention and are intended to model an inclined drill hole. It
should be noted that element 12 lies south of the Critical line established by
localised soil extraction as described above. The procedure for simulating the
under-excavation intervention was as follows:

274 / Finite element analysis in geotechnical engineering: Application

a)
b)

c)

d)

e)

The stiffness of element 6 was reduced to zero.

Equal and opposite vertical nodal forces were applied progressivelyto the
upper and lower faces until the volume of the element had reduced by
about 5%. The stiffness of the element was then restored.
The same procedure was then applied successively to elements 7,8,9, 10
and 1l , thereby modelling the progressive insertion of the extraction drill
probe. For each step the inclination of the Tower reduced.
When element 12 was excavated the inclination of the Tower increased,
confirming that excavation south of the Critical line gave a negative
response. The analysis was therefore re-started after excavating element
11.
The retraction of the drill probe was then modelled by excavating
elements 10,9,8,7 and 6 successively. For each step the response of the
Tower was positive.
The whole process of insertion and extraction of the drill probe was then
repeated. Once again excavation of element 12 gave a negative response.

The computed displacements of

the Tower are plotted in Figure 6.79.
The sequence of excavation of the
elements is given on the horizontal G 0.4
axis. The upper diagram shows the 3
change of inclination of the Tower g 0.3
during under-excavation and the 3
lower diagram shows the settlements .3 0.2
of the north and south sides of the ci; o.l
foundation.
%
It can be seen that as under0
U
1
excavation progresses from elements
Element number excavated
6 through to 1I , the rate of change of
northward inclination increases, as do
the settlements. As the drill is
100
retracted the rate decreases. At the
end of the first cycle of insertion and g 200
B
extraction the inclination of the
300
Tower is decreased by 0.1". The *
settlement of the south side is rather
400
more than one half of the north side.
For the second cycle of insertion and
Figure 6.79: Response of Tower to
extraction of the drill a similar
soil extraction
response is obtained, but the change
of inclination is somewhat larger. After the third insertion of the drill the resultant
noGhward rotation was 0.36". The corresponding settlementsofthe north and south
sides of the foundation were 260mm and 140mm respectively.
.W

-g
g

Shallow foundations 1 275

Figure 6.80 shows the contact stress distributions at various stages ofthe underexcavation intervention. The process results in a small reduction of stress beneath
the south side. Beneath the north side fluctuations in contact stress take place as is
to be expected, but the stress changes are small.
Distance (m)
0

12
l

16

20

Figure 6.80: Effective contact

stress distribution beneath the
Tower foundation

6.6.3.14 The response of the Tower to soil extraction

The results of the modelling work were sufficiently encouraging to undertake a
large-scale development trial of the drilling equipment. For this purpose a 7m
diameter eccentrically loaded instrumented footing was constructed in the Piazza
north of the Baptistry. Drilling was carried out using a hollow-stemmedcontinuous
flight auger inside a contra-rotating casing. When the drill was withdrawn to form
the cavity, an instrumented probe located in the hollow stem was left in place to
monitor its closure. The trials showed that cavities formed in the Horizon A
material closed gently and that continued extraction from the same location could

276 1 Finite element analysis in geotechnical engineering: Application

be achieved. The trial footing was successfully rotated by about 0.25" and
directional control was maintained even though the ground conditions were
somewhat non-uniform. The measured contact stress changes were very small.
In view of the encouraging results from the numerical analysis and the largescale trial, the decision was taken by the Commission to carry out preliminary soil
extraction beneath the north side of the Tower itself, with the objective of
observing the response of the Tower to a limited and localised intervention.
Preliminary soil extraction was carried out over a width of Gm, using twelve
200mm diameter boreholes. A target of a minimum of20 arc seconds reduction in
inclination was set as being large enough to demonstrate unequivocally the
effectiveness of the method.
On 9IhFebruary 1999, in an atmosphere of great tension, the first soil extraction
took place. For the first week the Tower showed no discernable response, but
during the following tense days it began very gradually to rotate northwards. As
confidence grew, the rate ofsoil extraction was increased. At the beginning of June
1999, when the operation ceased, the northward rotation was 90 arc seconds and
by mid-September it had increased to 130 arc seconds. At that time three of the 97
lead ingots (weighing about lot each) were removed and movement ceased.
During preliminary under-excavation soil extraction mainly took place outside
the footprint of the foundation and locally only extended beneath the north edge
of the foundation by about 1.5m. It is of interest that the southern edge of the
foundation was observed to rise by about one tenth of the settlement at the north.
This may be contrasted with the numerical model which predicted small
settlements at the south. The reason for this difference may be due to the fact that
a plane strain model was used, whereas the soil extraction process is highly three
dimensional. It is hoped to study this in more detail. Whatever the reason, the uplift
at the south is highly beneficial, as the volume of soil to be extracted is reduced
and it seems likely that reduction of stress is taking place in this critical region.
Having demonstrated that soil extraction produced a positive response, the
Commission formally approved the application of the method for permanent
stabilisation. Using 4 1 extraction tubes, work on the full intervention commenced
on 21" February 2000. It is estimated that it will take about eighteen months of
careful soil extraction to reduce the inclination of the Tower by about half a
degree, which will be barely visible. At the time of writing (July 2000) a reduction
of inclination of 800 arc seconds has been achieved and the pattern of uplift at the
southern edge has been maintained. There is still a long, tense journey ahead but
without the positive results of the numerical analysis it is doubtful that this very
sensitive operation on a Tower that is on the point of leaning instabilitywould have
been undertaken.

6.6.3.15 Comments
This section describes the development and calibration of two finite element
computer models of the Pisa Tower and underlying ground using the modified
Cam clay model with coupled consol-idation.One ofthe models is plane strain and

Shallow foundations 1 277

the other is three dimensional. The choice of compressibility parameters turned out
to be crucial and the use of the intrinsic compressibility C,' has proved useful in
interpreting the results of oedometer tests. The main purpose of the model was to
aid the understanding of the basic mechanisms of behaviour of the groundstructure interaction. Unlike most other attempts to simulate the behaviour of the
Pisa Tower, the model is self consistent in that it generates its own overturning
moment in response to any change of inclination.
Calibration ofthe model was achieved by ensuring that it gave the correct final
inclination of the actual Tower. It is shown that the model is in remarkable
agreement with the deduced historical inclinations of the Tower during and
subsequent to construction. The results of the analysis confirm beyond reasonable
doubt that the lean of the Tower is caused by a mechanism known as 'leaning
instability', which results from the high compressibility ofthe underlying clays and
is not related to strength. At an inclination of 5.44" the Tower is predicted to be in
a state of unstable equilibrium, confirming that the actual Tower is very close to
falling over.
The computer model has proved valuable in assessing the effectiveness of the
north counterweight solution for temporally increasing the stability of the Tower
foundations. Although the model was not developed with a view to making precise
predictions of very small movements, nevertheless it predicted the response to the
application of the counterweight with remarkable accuracy. The predicted
reductions in inclination were about 80% of the observed ones and the settlements
were almost exactly as predicted. The observed response of the Tower has been
used to slightly refine the model for future use in assessing the effectiveness of
various possible permanent solutions.
The technique of soil extraction is currently being used as means of inducing
controlled subsidence on the north side of the Tower as a permanent solution. In
view of the uncertainty about the response to soil extraction of a Tower close to
leaning instability, it was essential to study the mechanisms of behaviour with the
numerical model. A method of simulating local soil extraction was developed.
Initial studies have demonstrated the existence of a Critical line north of which soil
extraction leads to a positive response. The location of this Critical line appears to
be about !4 a radius in from the north side.
The process of under excavation by means of an inclined drill was then
simulated. A significant reduction in inclination ofthe model Tower was achieved.
Moreover, it was shown that the changes in foundation contact stress distribution
were very small. By confining soil extraction to a short distance beneath the north
side, settlements beneath the south side can be kept to less than % of those on the
north side.
It can be concluded from the numerical studies that the technique of soil
extraction offers a very positive method of permanently reducing the inclination
of the Tower by as much as a !h0.This encouraging result led to a successful large
scale field trial ofthe soil extraction technique. The technique is now being applied
to permanently stabilise the Tower. At present (July 2000) a reduction of 800 arc

278 / Finite element analysis in geotechnical engineering: Application

seconds has been achieved and the observed behaviour has been consistent with
that predicted by the analyses. It is doubtful that this very sensitive operation on
a Tower that is on the point of leaning instability would have been undertaken
without the positive results of the numerical analysis.

6.7
1.

2.

3.
4.

5.

6.

7.

8.

Summary

Shallow foundations can have different shapes in plane. If they are long in one
dimension, they are classified as strip foundations and can be analysed
assuming plane strain conditions. If they are circular, they can be analysed
either assuming axi-symmetric conditions, if the loading is vertical, or using
the Fourier series aided finite element method for general loading. For other
shapes a full 3D analysis is required.
When modelling surface foundations both the soil and the foundation should,
in general, be discretised into finite elements. However, if the loading is
vertical and the footing is assumed to be either very flexible or very stiff
compared to the soil, further approximations can be made in which it is no
longer necessary to include the foundation in the mesh.
The bearing capacity for a square surface footing is smaller than that of a
circular footing.
Finite element analyses enable theoretical shape factors to be determined for
surface foundations. These differ depending on the roughness of the footing,
but in general agree with the empirical formulae used in most design manuals.
For drained soils there are three coefficients in the general bearing capacity
equation, namely N, , N , and N, . For strip footings exact theoretical
expressions are available for N, and N, . Only approximate solutions (e.g.
stress fields) exist for N , . It has been shown that finite element analyses can
recover the theoretical correct values of N, , but more importantly such
analysis provide considerable insight in to the values of N , . They have also
indicated some surprising differences between the behaviour of strip and
circular footings.
No design guidance is available for the undrained bearing capacity of preloaded strip foundations on clay. It has been shown how finite element
analyses can be used to tackle this problem. It was concluded that for the
majority of real situations it is unlikely that pre-loading will give rise to a
substantial improvement in undrained bearing capacity.
Results of finite element analyses using the sophisticated MIT-E3 have been
presented to show how the effects of observed anisotropic soil behaviour can
be reproduced in finite element analyses. The effect of this anisotropic
behaviour on the bearing capacity of both strip and circular surface
foundations has been quantified by finite element analysis.
For shallow footings founded below the ground surface it is important to
correctly model the interface between the sides of the footing and the soil.
Interface elements are useful for this purpose, but must be given a zero tensile
stress capacity, as well as the appropriate shear strength.

Shallow foundations / 279

9. Finite element analyses indicate that the depth factors for circular footings are
higher than for strip footings.
10. The example of the leaning Tower of Pisa has been used to demonstrate finite
element analyses of shallow foundations. The difference between bearing
capacity failure and leaning instability has been described. It was shown how
finite element analyses were used in the decision making process for the
temporary North weighting and permanent under-excavation schemes at Pisa.

Deep foundations

7.1

Synopsis

This chapter considers the analysis of deep foundations. It begins by considering

axial and lateral loading of a single pile. The pros and cons of using interface
elements at the pile shaft are considered, as are the effects of soil dilation. Analysis
of pile groups are then discussed. A method of analysis and a design procedure,
both based on the analysis of a single pile combined withe the use of superposition,
are presented. The chapter finishes by considering the behaviour of bucket
foundations subjected to pull-out loads. The effects of geometry, interface strength,
anisotropic strength and position of load application are discussed.

7.2

Introduction

In the previous chapter surface and shallow foundations have been considered.
While such foundations are usually adequate where the soils below the foundations
are both strong and stiff andlor the foundation loading is light, they may not be
appropriate in weak soils or in situations where the imposed loads are high. In such
situations it is common to extend the foundations deeper into the soil. This has two
beneficial effects, firstly soils usually become stronger and stiffer with depth such
that bearing capacity increases, and secondly shear and lateral stresses can be
mobilised on the sides of the foundation which provide additional resistance to
support the applied loads.
Another reason for increasing the depth of a foundation is where pull-out forces
have to be resisted. For example, the foundations for some offshore structures often
have to resist upwards forces. This can be resisted by the shear and lateral stresses
mobilised on the sides of a deep foundation.
The most common form of deep foundation is the pile. This can either be
constructed first and then driven into the ground, or a hole can be bored and the
pile cast in-situ. Driven piles can be constructed from steel, wood or reinforced
concrete and come in many different cross-sectional shapes (e.g. circular,
rectangular, H-shaped, etc.). Bored piles on the other hand are usually constructed
from reinforced concrete. Sometimes an undersized hole is bored, a pile is then
driven and grouted afterwards. There are clearly many other combinations of
installation that are possible.
Piles are rarely used in isolation but are installed in groups. They are usually

Deep foundations I 281

connected at the soil surface by a pile

cap, see Figure 7.1. There are therefore
many parameters involved in the design
of piled foundations, namely the crosssectional size and the length of each pile,
wether the piles are solid or just a tube,
how they are to be installed, the spacing
of the piles in the group and the nature of
the connection of the piles to the pile cap.
Caissons sometimes offer an
Figure 7. 7 : Piled foundation
alternative to a pile group. They are
essentially large hollow piles, see Figure
7.2. They can be precast or constructed
in-situ. If they are precast, they are first
placed in position on the soil surface. Due
to their self weight some penetration
occurs. Soil is then excavated from inside
the caisson and, as this is continued, the
caisson sinks into the ground. If they are
built in-situ, the walls are constructed as
excavation proceeds. They are often
circular in plane, but not always, and
usually have a much larger cross
sectional area and a lower height to
breadth ratio than a pile. Single caissons
are often used as a foundation, for
example to support bridge piers.
Bucket foundations are similar to
caissons and, as their name suggests, they Figure 7.2: Caisson foundation
resemble inverted buckets, see Figure 7.3.
They are a recent development and are
used offshore where the structures they
support impose a range of loading
conditions including pull-out forces.
They are sometimes used solely to resist
pull-out forces and, when used for this
purpose, are often called suction caissons.
Installation into the sea bed usually
involves the use of suction forces. For
example, the bucket would be
prefabricated on land, towed to site and
placed on the sea bed. Due to its self
weight some embedment would occur. Figure 7.3: Bucket foundation
The pressure in the water enclosed inside

282 / Finite element analysis in geotechnical engineering: Application

the bucket is then lowered by pumping, to cause a positive differential water

pressure across the top of the bucket which forces the bucket into the soil. This
process continues until the required embedment is obtained.
The analysis of single piles will be considered first in this chapter. This will be
followed by the analysis of pile groups and then bucket foundations. As the
analysis of a caisson has similarities with both that of a single pile and that of a
bucket foundation it will not be specifically considered.
All analyses were performed using the Imperial College Finite Element
Program (ICFEP). An accelerated modified Newton-Raphson scheme, with a substepping stress point algorithm, was employed to solve the nonlinear finite element
equations, see Chapter 9 of Volume 1. FOPthe axi-symmetric and Fourier Series
Aided Finite Element analyses (FSAFEM, see Chapter 12 of Volume 1) 8-noded
isoparametric elements were used, with reduced (2x2) integration.

7.3
7.3.1

Single piles
Introduction

Although single piles are rarely installed in isolation, it is useful to consider their
analysis, as this is similar to, but less complex than, the full analysis of a pile
group. As will be discussed subsequently, the latter analyses are currently beyond
the capacity of present computing resources for all but some very simple cases.
If the pile is circular in cross section, installed vertically in the ground, which
has a horizontal lithology and horizontal water table, then the geometry to be
analysed is axi-symmetric. If only vertical loading is applied to the pile then an axisymmetric analysis is appropriate. However, if either inclined andlor moment
loading is applied to the pile then a three dimensional analysis is required. As the
geometry is axi-symmetric this can be performed with the computationally
efficient FSAFEM. If any ofthe above conditions are not satisfied, for example the
pile is not circular, or is not installed verticalfy in the ground, or the soil lithology
is not horizontal, then a full 3D analysis using three dimensional finite elements
must be used even if axial loads are applied to the pile.
For the remainder of this section the geometry will be assumed to be axisymmetric. Consequently, for vertical loading axi-symmetric analysis, and for
lateral loading three dimensional analysis using the FSAFEM, will be considered.

7.3.2

Vertical loading

One of the important issues when analysing a pile subject to vertical loading is the
modelling of the interface between the pile and the soil adjacent to the pile shaft.
As noted in Section 6.6.2, problems can arise if the elements representing the soil
are too large and no interface elements are used. To overcome the problem it is
important to either use thin solid elements in the soil adjacent to the pile shaft
andlor use interface elements. However, for the pile problem care must be
exercised when assigning material properties to the interface elements.

Deep foundations I 283

50m
4
As an example consider the
t.
+cO.Srn
problem of lm diameter and 20m
long pile, installed in clay and loaded
vertically under undrained conditions.
The pile is assumed to be very stiff
and the clay is represented by the
T r e s ~ amodel, with E1,=105kN/m2,
p=0.49 and Sl,=100kN/m2.The finite
element mesh used for the analyses is
shown in Figure 7.4 and consists of
1024 eight noded elements. It should
be noted that relatively thin (0.05m
wide) solid elements, with an aspect
ratio of approximately 5 vertically to 1
horizontally, are used adjacent to the
pile shaft. The pile itself is modelled Figure 7.4:Finite element mesh for
with solid elements and behaves
single pile analysis
linear elastically, with a Young's
modulus E=20x106kNlm2and a Poisson's ratio p=0.15. The loading of the pile is
modelled by applying increments of vertical displacement to its top. The other
boundary conditions consist of restricting both vertical and horizontal movements
on the base of the mesh and the horizontal movements on the right hand side of the
mesh and along the axis of symmetry (i.e. left hand side of the mesh). Initial
stresses in the clay were calculated assuming a saturated bulk unit weight
y,,,=l 8kNlm3and coefficient of earth pressure at rest K,=1 .O.
In the first analysis no interface
8000
elements were used. The pile head
load-displacement curve obtained
from this analysis is shown in Figure
7.5. This figure also shows the 4 4000
mobilisation of shaft and base 3
capacity with displacement. These
2000
capacities have been calculated from
oo
0.01
0.02
0.03
0.04
the nodal forces arising from the
stress changes (from the beginning of
Displacement (m)
the analysis) in the soil elements
adjacent to the pile. In agreementwith ~i~~~~
7.5: undrained loading of a
field observations the shaft capacity is
single pile without interface
mobilised much earlier than the base
elements along its shaft
capacity. The analysis predicts an
ultimate shaft capacity of 6345kN, which is not dissimilarto that obtained from the
simple calculation in which the undrained strength is multiplied by the shaft area
giving 6283kN. It is slightly on the high side due to the inherent problem of only
sampling the stresses at integration points, which are not located on the pile-soil

284 1 Finite element analysis in geotechnical engineering: Application

interface. This problem has been discussed in Section 6.6.2. The ultimate base
capacity suggests a bearing capacity factor Nc=9.
Three further analyses were then performed, all of which used the finite
element mesh shown in Figure 7.4, but with zero thickness interface elements
positioned adjacent to the pile shaft and its base. The same soil properties as above
were used and the interface elements were assigned an undrained strength
S,,=100kN/m2 (i.e. the same as the soil) and equal shear, K, , and normal, K,, ,
stiffnesses. The only difference between the three analyses was the values assigned
to K, and K,, ,these being 103kN/m3, 105kN/m3 and 107 kN/m3 for each analysis
respectively. As noted in Chapter 9 the choice of stiffness values for interface
elements is problematic, especially because they have different units (kN/m3) to
those of the soil (kN/m2).
The resulting total pile head load8ow
displacement curves for the analyses
with interface stiffnesses of 103kN/m3
and 105kN/m3are compared with each X
other and with that obtained from the
4000
analysis with no interface elements in .B
Figure 7.6. The analyses with an 3t: 7.000
interface stiffness of 105kN/m3andno
interface elements agree quite well,
oo 0.02 0.04 0.06 008 o 1 0.12 0.14
the former producing a slightly less
Displacement (m)
stiff response and having a slightly
smaller limit load. Closer inspection
,=igure7.6: Effect of interface
of the results indicates that the element stiffnesson pile behaviour
ultimate shaft capacity is now
6279kN, which is near to the value obtained from the simple calculation in which
the undrained strength is multiplied by the shaft area (i.e. 6283kN). As the
integration points for the interface elements are located on the pile-soil interface,
the predicted shaft capacity is likely to be more accurate than that obtained from
the analysis with no interface elements.
The results from the analysis with an interface stiffness of 107kN/m3 are not
shown on Figure 7.6 for clarity. During the early stages of loading the predicted
load-displacement curve is indistinguishable from that of the analysis without
interface elements. The ultimate load, however, is slightly smaller than that
obtained from the latter analysis, but is in agreement with that from the analysis
with an interface stiffness of 105kN/m3.
The analysis with the softer interface stiffness of 103 kN/m3 predicts a much
softer load-displacement curve, see Figure 7.6. It also does not appear to reach the
same ultimate load. The reason for this can be seen in Figure 7.7, which shows the
separate mobilisation of shaft and base resistance for this analysis. While the shaft
resistance is fully mobilised at a displacement of 0.105m, the base resistance is
extremely small and requires considerably more displacement before it becomes
fully mobilised.

g ,,
d

Deep foundations / 285

Clearly, the load-displacement

6000
behaviour is dominated by the
5000
stiffness of the interface elements. As v
it is difficult to measure interface
4000
stiffness, it might be concluded that 5 3000
the most sensible option is to use high 'E
9 2000
'000
stiffness values. However, as noted in
o
Chapter 9, the use of too high a
stiffness can lead to numerical illDisplacement (m)
conditioning. A better alternative is
probably not to use interface elements Figure 7.7: Undrained loading of a
at all. As the comparison shown in pile with soft interface elements
Figure 7.6 indicates,the analyses with
no interface elements produces adequate results. While this is acceptable when
analysing piles subjected to only vertical loading, it may be necessary to include
interface elements for other reasons, when apile is subjected to lateral or combined
loading. This will be discussed further in Section 7.3.3 of this chapter.
Drained analyses of this pile problem have also been performed. For these
analyses the soil was modelled using the Moh- Coulomb model with E=105kN/m2,
p=0.3 and q1=25".Initial stresses in the soil were based on yh=l 8kN/m3and K,=l
and on the assumption that the soil was dry.
,,
Load-displacement curves from
two analyses, both without interface
elements between the pile and the
soil, are shown in Figure 7.8. The g 6000
only difference between these two
analyses is that in one the angle of .2 4000
dilation of the soil is v=25", whereas 3 2000
in the other v=OO.As can be seen from
0
this figure, the analysis with ~ 2 5 "
0
0.02
0.04
0 06
0.08
01
Displacement (m)
predicts much higher pile loads for a
given displacement than the analyses
with v = ~ o~h~
. results are shown in Figure 7.8: Brained loading of a
single pile without interface
more detail in Figures 7.9a and 7.9b,
elements along its shaft
where the resistance from the pile
shaft and the pile base have been separated. For the analysis with v=25" both the
shaft and base components show no sign of reaching a limiting value, see Figure
7.9a. Even if the pile is subject to further displacement, there is no indication of
either the shaft or the base resistance reaching a limiting value. In contrast, for the
analyses with v=Oo the shaft resistance has reached a limiting value at a
displacement of 0.01m (6/D=1%), see Figure 7.9b. The base capacity is, however,
still increasing even at a displacement of 0. lm (d/D=lO% which is normally taken
as a measure of pile failure in practice) and only reaches a limiting value (not
shown in Figure 7.9b) when the pile displacement is 2.0m.

5
g

g
g

286 / Finite element analysis in geotechnical engineering: Application

Displacement (m)

Displacement (m)

Figure 7.9a: Behaviour of a pile

in drained soil with full dilation

Figure 7.96: Behaviour of a pile

in drained soil with no dilation

The failure of the analysis with v=25O to predict a limiting pile load is a
consequence of the kinematically constrained nature of the problem, combined
with the continued plastic dilation predicted by the Mohr-Coulomb model. The
problem is not confined to this model but will occur with any constitutive model
that predicts finite plastic dilation indefinitely, without reaching a critical state
condition. This potential pitfall is discussed in greater detail in Chapter 9.
The use of interface elements
6000
between the soil and the pile can, in
some respects, suppress the above ,
5000
shortcomings of the constitutive g 4000
model used to represent the soil. For
o,,
example, the drained analyses have
.B 20,
been repeated with zero thickness
1000
interface elements adjacent to the pile.
These elements were given a stiffness
0
0.01 0.02 0.03 0.04 0.05 0.06 0.07
K,,=K,=lOS kN/m3, an angle of
Displacement (m)
shearing resistance 9'=2S0 and an
angle of dilation v=OO. The loadFigure 7.10: Drained loading of a
displacement curves are shown in
single pile with interface elements
Figure 7.10. As with the undrained
along its shaft
analysis discussed above, the
interface elements dominate the behaviour and the results of the analyses are much
closer. Figures 7.1 1a and 7.1 1b show the mobilisation of the shaft and base
resistance for the two analyses. There are differences in the base resistance. If the
pile is subject to further displacement a limiting base resistance is predicted by the
analysis with v=OOin the soil, while no such limiting value is predicted for the
analysis with v=2So in the soil. There are also small differences between the
mobilised shaft resistance from the two analyses. This arises due to the different
normal stresses mobilised along the pile shaft, especially in the vicinity of the pile
base. It can therefore be concluded that while the behaviour, and therefore
properties of, the interface elements dominate the mobilisation of pile shaft
resistance, they have only a small influence on the base resistance.

l-

Deep foundations 1 287

Displacement (m)

Displacement (m)

Figure 7.1 lb: Behaviour of a pile

with interface elements full dilation in a drained soil

Figure 7. 1 la: Behaviour of a pile

with interface elements no dilation in a drained soil

7.3.3

Lateral loading

As noted above, if a vertical pile is subjected to lateral andor moment loading a

three dimensional analysis is required. If the geometry is axi-symmetric the
FSAFEM can be used. As an example, the pile considered above has been
subjected to lateral loading.
The finite element mesh shown in Figure 7.4 was employed. The top of the pile
was displaced horizontally by specifying the following incremental radial, Au,., and
circumferential, Au,, displacements around its out edge:
Au, = Adcos6
Au, = -M sin 8

This results in an incremental lateral displacement of the top of the pile of Ad. The
remaining boundary conditions consisted of restricting all three components of
displacement (i.e. U,, U , and U,) along
4000
the base of the mesh and the radial
displacement (U,) on the right hand
IOW
side of the mesh. Unlike the vertical
analyses described above, no $
displacements were restricted on the 4 2000
left hand side of the mesh which
looo
represents the axis of symmetry of the 8
geometry. Under lateral loading this
o
0.02
0.04
0.06
0.08
0.1
boundarv must be free to move. The
Displacement (m)
analyses were performed using
parallel symmetry and ten harmonics,
Figure 7.12: Undrained lateral
see Chapter 12 of Volume 1.
loading
of a single pile with and
Initially the soil was assumed to
without
interface
elements along
be undrained with E,=1O5 kN/m2,
its
shaft;
constant
S, with depth
,u=0.49andS,,=lOOkN/mZ.Theresults

.g

288 / Finite element analysis in g~eotechnicalengineering: Application

of two analyses, one without interface elements and the other with, are shown in
Figure 7.12 in the form of pile head lateral load-displacement curves. The lateral
load has been determined from the reactions at the nodes at the top of the pile
subjected to the prescribed lateral displacement. The interface elements were
assigned normal and tangential stiffhesses of Kn=K,=1O5kN/m3and an undrained
strength S,,=100kN/m2.In addition, these elements were unable to sustain tensile
normal total stresses. If they experienced such stresses they were allowed to open
freely. In such conditionsthese elements could not transfer shear stress between the
soil and the pile shaft.

Horizontal stress (kPa)

Figure 7. 13: Effect of interface

elements on horizontal total
stress down the back of a pile

Horizontal displacement
of pile (m)

Figure 7. 14: Effect of 'gapping'

on horizontal displacement along
the pile

The horizontal load-displacement curves shown in Figure 7.12 are very

different. For a given horizontal load the analysis without interface elements
predicts much smaller displacements. The reason for this is apparent from Figure
7.13 which shows the distribution of horizontal total stress acting down the back
(i.e. the side opposite to the direction of loading) of the pile shaft at a horizontal
pile head displacement of O.lm. For the analyses performed without interface
elements large tensile stresses are predicted over the top 5.5m of the pile. In
contrast, because of the no-tension capacity assigned to the interface elements, the
analysis with interface elements has zero horizontal stress over this region of the
pile shaft. This analysis has therefore modelled a crack forming behind the pile as
it moves forward. In practice such behaviour is referred to as 'gapping'. The effect
of gapping on the displaced profile of the pile is shown in Figure 7.14, where the

Deep foundations I 289

results from the two analysis are
compared for a pile head load of 2000kN.
A further analysis was performed in
which the soil strength and stiffness
increased with depth, Sf,=2.7z kN/m2
(equivalent to S,,=0.3~,,') and E,,=1350z
kN/m2(equivalent to Ef,=500Sf,),where z
is the depth below the soil surface. Such
an undrained strength distribution with
depth is more representative of lightly
overconsolidated clay. No interface
elements were used adjacent to the pile
shaft. The resulting pile head horizontal
load- displacement curve is shown in
Figure 7.15 and the distribution of
horizontal total stress down the back of
the pile, at a pile head displacement of
O.lm, in Figure 7.16. For this analysis no
tensile horizontal stresses are predicted
down the back of the pile and
consequently an analysis using interface
l the pile
elements between the s ~ i and
would have produced similar results.
It can therefore be concluded that if
gapping is likely to occur, then interface
elements should be installed along the
pile soil interface. Wether or not gapping
is likely to occur depends on the soil
strength and in particular its distribution
with depth.

g
E

500
400

3oo

'3 '0

100

.,

Figure 7. 5:Undrained lateral

loading of a single pile without
a,ong its shaft
- linear variation of
with depth

g 10
-

.$
4
PI

20
0

7.4

Pile group behaviour

7.4.1

introduction

Displacement(m)

100

200

300

400

Horizontal stress (Ha)

Figure 7. 16: Horizontal total

srress down the back of a pile
for
linear with depth

The analysis of a pile group is three

dimensional. Consequently current
methods of analysis of pile groups subjected to combined axial loads, lateral loads
and turning moments are either wholly based on linear elastic soil behaviour
(Randolph and Poulos (1982)), or combine nonlinear single pile response with
linear elastic forms of interaction (O'Neil et al. (1977)). The former methods are
too simplistic to give accurate predictions of pile group response. The latter hybrid
methods suffer from both the inconsistent nature in which they treat soil response
(both linearly and nonlinearly) and the unrealistic assumptions made in obtaining
the nonlinear single pile response. These hybrid methods account for nonlinear

290 / Finite element analysis in geotechnical engineering: Application

single pile behaviour by either using a n~on-rigorousmodified form of elastic

response (Poulos and Davis (1968)), or by ignoring the continuum nature of soil
behaviour, e.g. the t-z method (Coyle and Reese (1966)) and thep-y method (Reese
(1 977)).
While in principle a complete pile
..
group could be analysed by a three
e 6i B'
......
dimensional finite element analysis, .............
.@ ........, . : ..........
. : ..
@.
.. . . .. . Analysed
the number of elements required for
. . .. ..
(3
an accurate analysis would be
Analysed
@Q.
: .B
6
excessive and too large for current
..~...~...~...~...~.
computer hardware to handle. In the ..&.........g
.........6... .. .. .. .
special case where loading on the pile ;
;
..&...*...&...@...&.
..
.. ....
[email protected]#
group is vertical (i.e. no horizontal or 6.........'
moment loading), then axes of j
...g$...&...@..
.. ..-Q.
symmetry can be accounted for and
... ......@.............
..@...&,...@...@...
. .
. . . . .
only a segment of the geometry
analysed. Some examples are given in
Figure 7.17 where it can be seen that Figlure 7.1 7: Examples of pile group
this significantly reduces the
symmetry
complexity of the geometry to be
analysed. The comments made earlier for the analysis of a single pile to vertical
loading also apply to the analysis of such pile groups. In addition, it should be
noted that because of the three dimensional nature of the geometry a full three
dimensional analysis must be performed.
If the loading is not vertical and includes a lateral andor moment component
at the pile cap, then there are fewer axes of symmetry and more of the piles must
be included in the analysis. Again, full three dimensional finite element analysis
must be performed. Because the geometry is no longer axi-symmetric it is not
possible to use the Fourier Series Aided Finite Element Method (FSAFEM).
With current computer hardware it is not really a practical proposition to
analyse a pile group. Consequently for practical problems simplifications must be
made.
A new method for considering pile groups has recently been proposed by
Ganendra (1993), Ganendra and Potts (2002a, b). This method is based on results
from nonlinear three dimensional analyses of a single pile and superposition
techniques which use the same results to account for pile group effects. This
method for considering pile groups is thought to be an improvementon the existing
methods because:

I;-

-6

'

<

y ; :F

- A more sophisticatedand rigorous method of analysis of single piles is used

which is able to incorporate complex soil and pile-soil interaction

behaviour.
A more coherent method for accounting for pile group interaction is used
which is based on the results of the single pile analyses.

Deep foundations / 291

This new method is divided into two parts, the analysis of a pile group
(Ganendra and Potts (2002a)) and the design of a pile group (Ganendra and Potts
(2002b)). The proposed method of analysis of a pile group is able to give
predictions of pile group rotation, lateral deflection and vertical settlement for a
particular applied load, which may consist of any combination of axial load, lateral
load and turning moment. This method is essentially an extension ofthat described
by Jardine and Potts (1988), (1992) which was for vertically loaded pile groups
only.
The proposed method of design for a pile group provides 'design charts' for
any particular pile group such that predictions of pile group rotation, lateral
deflection and vertical settlement can be obtained for any combination of axial
loads, lateral loads and turning moments. Note, however, that the design method
does not give any recommendations regarding the number, size or configuration
of piles in a pile group.
As this approach is based on nonlinear finite element analysis of a single pile
and as it appears to provide many advantages over current design approaches, it
will be described in subsequent sections of this chapter. The application of the
proposed analysis and design methods to the foundations of the Magnus oil
platform will then be discussed.

Analysis of a pile group

7.4.2
Both the FSAFEM and a full three dimensional finite element analysis are able to
analyse the behaviour of a single (vertical and cylindrical) pile subjected to any
combination of lateral load, vertical load and turning moment. However, as the
FSAFEM is more efficient all the three dimensional analysis referred to in
subsequent sections have been performed using this approach. It should be noted
however that it is perfectly valid to perform a conventional three dimensional
analysis in its place.
To extend the analysis to consider the behaviour of a pile group subjected to
such loads two assumptions have to be made:

The manner in which the loads applied to one of the piles in a group affects
the displacement of the other piles.
The manner in which the loads applied to the group are distributed between
the piles in the group.

The proposed method of analysis uses superposition techniques for the first
assumption and a rigid pile cap criterion for the second assumption, both of which
are described below.

7.4.3

Superposition
The displacements of any pile are affected by the loads imposed on the
surrounding piles. This effect is accounted for in the proposed method of analysis

292 / Finite element analysis in geotechnical engineering: Application

using a superposition technique. As superposition is theoretically invalid for a

nonlinear system, approximations must be made and Ganendra and Potts suggested
three alternative superposition techniques for the proposed method of analysis:

- Simple superposition;
- Nonlinear superposition;

Virtual pile.

Of these the simple superposition has been used in this chapter and it is described
below. The other techniques are presented in Ganendra (1993).

7.4.3.1 Simple superposition

This method assumes that the additional displacements at any pile head, due to the
loads imposed on surrounding piles, can be evaluated from the ground surface
movements obtained from a single pile analysis. Consider a pile group with Npiles
which are labelled from 1 to N. Let the ground surface displacements, U,', at any
point due to the loads F, on the thpile be given by the function:

where r, and 8, are the coordinates expressed in terms of the local cylindrical
coordinate system for the thpile, which has its axis about the centre of the pile, see
Figure 7.18. The function C,(r,,8,) is obtained from a 3D analysis (i.e. using the
FSAFEM) of an isolated pile.
The simple superposition method assumes that the additional displacement U/
of the jth
pile, induced by the load F, on the thpile, is equal to C,(r:, B:). r,' and 8,'
are the coordinates of the/lh pile expressed in terms of the local coordinate system
for the ithpile. Thus the displacements U,;/of the jth
pile in a pile group can be
obtained using simple superposition, by adding the displacement contributions
from all the piles in the pile group:

where:

U/ = C,(r,',

0,') is the displacement of thefh pile due to the load on the

jthpile;

(r,', 0,') = (0,O) are the coordinates for the centre of the jlh pile;
U,! is the additional displacements induced at the jThpile due to the
loads imposed on the other piles in the group.
The vectors of displacements U,,' and u j are expressed in the pile group global
Cartesian coordinate directions X, y and z (where z is the depth from ground
surface), see Figure 7.18. The origin for the x and y coordinates is located at the
plan view centroid of the pile group and shall be referred to as the group datum.
Thus the ground surface displacements (obtained from a FSAFEM analyses)
associated with the loads on each pile have to be resolved from the local axisymmetric coordinate system of the pile into the global pile group Cartesian
coordinate system.

Deep foundations / 293

Only pile head and ground surface displacements are considered using this
superposition technique. Methods of extending these superposition techniques to
predict pile displacements with depth are described subsequently. An important
feature of this method is that the ultimate load of a pile is unaffected by the loads
on the other piles, however the displacements at which this load is mobilised will
be different.
This approach is a generalisation of the method used by Jardine and Potts
(1988) and is schematically illustrated in Figure 7.19.

Load

Displacement

Figure 7.18: Local and global pile

group coordinate systems

Figure 7.19; Simple

superposition

7.4.3.2 Pile displacemen ts with depth

The superposition technique should only be used to give estimates of pile head
displacements. If the displacements along the length of the pile are also modified
using this technique, the resulting deformed shape of the pile may result in
unrealistic stress distributions within the pile. Two options exist for obtaining such
displacements from the superposition of pile head displacements and the pile
displacements (along the length of the pile) obtained from an isolated single pile
FSAFEM analysis:

The ratio between the superposition pile head displacements and the
isolated pile head displacements is used to factor the isolated pile
displacements (along the length of the pile). The stresses in the pile
associated with these factored displacementswould no longer be consistent
with the loads applied to the pile.
The difference between the superposition pile head displacements and the
isolated pile head displacementsare added to the isolated pile displacements
(along the length of the pile). The resulting pile stress distribution is the
same as the isolated pile stress distribution and thus is consistent with the

294 / Finite element analysis in geotechnical engineering: Application

applied pile loads. However, this displacement distribution is considered to

be unrealistic, since it predicts significant pile displacements even at great
depths.
To obtain reasonable estimates of pile displacements and stresses using the
superposition technique, the former option can be used for displacements and the
latter option for stresses.

7.4.4

Load distribution within a pile group

The manner in which the loads are distributed between the piles in a pile group
depends on the nature of their pile cap connectivity and the stiffness ofthe pile cap.
In the Ganendra and Potts approach a rigid pile cap criterion is used, which
assumes that the pile cap is fully rigid and there is full connectivity between piles
and pile cap at ground level. In this case the pile head displacement and rotations
of any pile can be evaluated from their location within the pile group and the pile
group displacements and rotations. These pile group displacements can be
expressed as U,, U, and U:, which are the displacement of the pile group datum
in the X, y and z (where z is depth) global coordinate directions respectively. The
pile group datum and the origin for the global coordinate axes are taken to be the
plan view centroid of the pile group at ground level, see Figure 7.18. The group
rotations can be expressed as 0, and B,, which are the rotations of the group about
the X and y axes respectively. As a simplification, torsions applied to the pile group
and accordingly rotations about the z axis are not considered. U,, U,, U,, Or and
O, are schematically illustrated in Figure 7.20.
Due to the rigid pile cap criterion, the displacements U,', U,,' and U,' (in the X, y
and z directions respectively) of any thpile in a pile group can be related to the pile
group displacements via the equations:

where X' and y' are the x and y coordinates respectively of the thpile.
Due to the assumption of full connectivity between pile and pile cap, the pile
head rotations 0; and 0; (about the X and y axes respectively) of the any thpile
in a pile group should equal the global group rotations:

Note that the global group displacements and rotations (U,, U,, U:, 0, and 0,) can
be evaluated from the pile head displacements, rotations and locations (U,', U,,',U,',
O,'? Q,,' X' and y') from any one pile in the pile group.
The loads that are applied to a pile group can be expressed as forces F,, F, and
F, and turning moments M, and M,. F,, F, and F, are the forces in the X, y and z

Deep foundations 1 295

directions respectively, and M, and M, are the turning moments about the X and y
axes respectively. They are schematically illustrated in Figure 7.2 1. These loads
are distributed among the piles in the group in such a way that the rigid pile cap
criterion is satisfied, i.e. the same values of global group displacements and
rotations (U,, U,, U,, Oxand 0,) are evaluated from the pile head displacements
and rotations (U:, U,', U:, 0:' and 0)'; of each pile in the group. An iterative
pra'cedure can be used to evaluate these individual pile loads. The steps in the
procedure are:

Obtain an initial trial division ofthese applied loads between the piles in the
group;
Evaluate, using a superposition technique, the displacements of each
individual pile due to the loads applied to it and the loads applied to its
surrounding piles;
These pile displacements are used to check if the rigid pile cap criterion is
satisfied:
- If it is satisfied (to within a small tolerance) this trial division of loads
is assumed correct and the resulting solution for the global group
displacements is accepted;
- If the rigid pile cap criterion is not satisfied, a new trial division of loads
is obtained and the procedure repeated until the criterion is satisfied.

A detailed descriptions of these steps is presented below.

Figure 7.20: Schematic

i/lustration of global pile group
displacements

Figure 7.2 7: Schematic

illustration of global pile group
loads

296 l Finite element analysis in geotechnical engineering: Application

7.4.4. 1 Obtaining an initial trial division of the applied loads

The initial trial division of the applied forces F,, F, and F: can be obtained by
assuming that they are divided equally between all piles in the pile group.
Obtaining a trial division of the turning moments M, and My is not so
straightforward. The restoring moments Rom the piles for the applied turning
moment My can be divided into two parts:

M;p, the push-pull moment, which is the moment resulting from the
difference in pile axial loads in the x direction;
M?, the pile head bending moment, which is the sum of the bending
moments about they axis which are applied to the pile cap from each pile
in the pile group.

The division of My between M,PP and Myhbis illustrated in Figure 7.22. To

determine what proportion of M, is M,PP and what proportion is M?, estimates are
obtained of the push-pull rotational stiffness K,PP (=MyPPIOy),and the pile head
bending rotational stiffness K t b (=MyhblOy).A procedure for estimating K,PP and
K,hb is described in Section 7.4.5.5. Trial values of M,Pp and Myhb can then be
obtained using the following equations:

My"P=M

KY""

K,PP
and M; = M y
K~+K?
K ~ P+P~

yhb

(7.6)

To determine the increaselreduction in individual pile axial loads which would

yield a push-pull bending moment equal to MiP, it is assumed that there is a linear
variation of pile axial load with the global x coordinate. The trial division of the
moment M? is obtained by assuming it is shared equally between all the piles in
the pile group.
The above procedure is repeated for the division of M, into M,PP and M,hb and
the division of these into individual pile loads.

Figure 7.22: Division of M, into M,PP and M?

Deep foundations 1 297

7.4.4.2 Evaluating pile head displacements

The pile head displacement of each pile, due to the loads applied to it and the loads
applied to its surrounding piles, is evaluated using FSAFEM analyses of single
piles and the superposition technique mentioned above (or an alternative
technique). It is of interest to note that the FSAFEM analyses used would in
general involve simultaneously loading a pile with latera! loads in both the X and
y d.irections, axial loads, and turning moments about both the X and y axes.

7.4.4.3 Checking the rigid pile cap criterion

The rigid pile criterion can be divided into nine separate conditions. These
conditions and the appropriate corrections that should be made if any one of them
is not satisfied are described below:
l . Check that the lateral deflections in the X direction for all the piles are the

2.

3.

4.

5.
6.

7.

same, if so a value of U, is obtained. If not, the lateral loads in the X

direction of the piles with larger displacements (in the X direction) are
reduced and those with smaller displacements are increased. To evaluate
these corrective loads, the average displacementof all the piles is calculated
and an appropriate tangential value of lateral pile stiffness is evaluated. The
corrective load that should be applied to each pile is the product of this
stiffness and the difference between the average displacement and the pile
displacement. A similar approach can be used for the other conditions
discussed below.
Check that the lateral deflections in they direction for all the piles are the
same, if so a value of U, is obtained. If not, the lateral loads in the y
direction of the piles with larger displacements (in the y direction) are
reduced and those with smaller displacements are increased.
Check that the variation of individual pile settlements with x coordinate is
linear, if so a value of 63, is obtained. If not, the axial loads are reduced for
piles with settlements that are too large and are increased for piles with
settlements that are too small.
Check that the variation of individual pile settlements with y coordinate is
linear, if so a value of is obtained. If not, the axial loads are reduced for
piles with settlements that are too large and are increased for piles with
settlements that are too small.
If a linear variation of axial displacements exist in both the X and y
direction, a value of U, is obtained.
Check that the rotations about they axis for all the pile heads are the same,
if so a value of O, is obtained. If not, the applied pile head moments about
they axis of the piles with larger rotations (about they axis) are reduced
and those with smaller rotations are increased.
Check that the rotations about the X axis for all the pile heads are the same,
if so a value of Oxis obtained. If not, the applied pile head moments about
the X axis of the piles with larger rotations (about the X axis) are reduced
and those with smaller rotations are increased.

298 / Finite element analysis in geotechnical engineering: Application

8. Check that the values of O, obtained from checks 3 and 6 are the same. If
not, the ratio between M,PP and Myhbis adjusted accordingly.
9. Check that the values of Oxobtained from checks 4 and 7 are the same. If
not, the ratio between M,PP and Mxhbis adjusted accordingly.
If any one of these conditions is not satisfied to within a prescribed tolerance, the
appropriate corrections are made and a new set of pile head displacements is
evaluated and all nine checks repeated. If all nine conditions are satisfied
concurrently, the solutions for U,, U,, U,, Oxand O, are accepted.
The above procedure seems very cumbersome. However, the trial division of
loads described in Section 7.4.4.1 should give solutions that satisfy (to within a
reasonable tolerance) the majority of the checks described. For most cases it is
anticipated that only checks 8 and 9 may not be fulfilled by the trial division and
that relatively few iterations would be required to obtain a solution that satisfies all
nine conditions. The solutions obtained from this proposed method ofanalysis only
make two basic assumptions:

- Superpositiontechniques can be used to determine the manner in which the

-

loads applied to one of the piles in a group affect the displacements of the
other piles;
The rigid pile cap criterion can be used to determine the manner in which
the loads applied to the group are distributedbetween the piles in the group.

7.4.5
Pile group design
The design of an offshore platform is a complex interdisciplinaryprocess of which
foundation design is a component. The method of analysis of a pile group
described in Sections 7.4.2 to 7.4.4 is of limited use in such a procedure, since it
is an iterative process for obtaining solutions for a particular set of applied loads
and it requires that 3D finite element analyses (i.e. using the FSAFEM) of each pile
be undertaken for each iteration. Predictions of this type are more suitable for back
analyses than for design. Thus to provide a more flexible tool for the design of an
offshore structure a 'design chart' procedure was devised which gives predictions
of pile group displacements and rotations for any set of design loadings (without
recourse to additional 3D finite element analysis). The manner in which these
design charts are obtained and how they should be used is described below. The
assumptions that are made in addition to those made for the analysis of a pile group
are stated. The procedure used for obtaining these design charts extends the linear
elastic procedure described by Randolph and Poulos (1982) to incorporate
nonlinear soil behaviour.

7.4.5. 1 Matrix formulation of the pile group response

The response of a pile group to a general set of loading conditions can be
expressed in matrix form by representing both the pile group displacements and
rotations, U,, U,, U:, Or and O, (as defined in Section 7.4.4), and the applied loads

Deep foundations 1 299

to the pile group, F,, F,, F,, M, and M, (also defined in Section 7.4.4), as vectors
Uand Frespectively. These vectors are related via a pile group stiffness matrix K:

In the linear elastic formulation described by Randolph and Poulos (1982), the
components ofthe matrix [Kj are constants. Thus if all the components of [Kj were
known, displacement U can be calculated for any applied load F by premultiplying Fwith the inverse of [K]. Evaluating these values of U for an applied
load F is more complex for nonlinear soil behaviour. The components of [W are
no longer constant and vary as a hnction of the applied loads and displacements.

7.4.5.2 Superposition of loads

To consider the nonlinear problem it is assumed that the loads required to cause
displacements U can be obtained by summing the loads required to cause each
component of U (superposition of loads). This can be expressed vectorially by
dividing the vector of displacements U into sub-vectors U,, U,, , U,,, U,, and U,!,
where:

Let the loads required to cause displacements U,, U, U,,,, UOxand U,, be F,, F,
F,, ,F,,, and F,,,, respectively. The assumption of superposition of loads means that
the applied load F, required to produce group displacement U, is equal to the sum
of the of the loads F,, .Fm,F,,, , F,, and F(,:

The components of the load vector F, are given by the reaction forces that have
to be applied to the pile group in order that it displaces a distance U.vertically,
while all other components of displacement and rotation are fixed (at zero). The
same principle can be used to relate the other load vectors, F,, F,, ,F(,,and F,, ,
with their associated displacements, U,, U,, B, and B, respectively. The manner in
which these load vectors can be obtained from single pile FSAFEM analyses is
described below.

Evaluating load vector F,

The load vector F, consists of five terms: three forces, F;, F;" and F: (in the z,X
and y directions respectively), and two turning moments, M;' and M; (about the X

300 / Finite element analysis in geotechnical engineering: Application

and y axes respectively). These are the forces and turning moments required for a
pure settlement of the pile group. To simplify the procedure it is assumed that F;",
F;, M; and M: are zero. F: is obtained from an (axi-symmetric or FSAFEM)
analysis of an axially loaded pile and a superposition technique (i.e. the simple
superposition method described in Section 7.4.3.1. This is illustrated in Figure
7.23.

Figure 7.23: Evaluation oJ load vector F,

Evaluating load vectors F, and F,,

The load vector F, consists of five terms: three forces, F,,:, F,,: and F,,: (in the z,
M,,; (about the
X and y axes respectively). These are the forces and tuming moments required for
a pure displacement of the pile group in thex direction. To simplify the procedure
it is assumed that,'F
,, F,,; and M', are zero. F,,' and M,,; are obtained from a
FSAFEM analysis of a fixed head (i.e. not allowed to rotate) laterally loaded pile
and a superpositiontechnique. F,," is the force required to give lateral displacement
Uxand M,! is the turning moment required to maintain the fixed head condition.
This is illustrated in Figure 7.24.
The components of the load vector F,,,,are evaluated in a similar manner.
X and y directions respectively), and two turning moments, M,#+nd

Figure 7.24: Evaluation of load vector F,,

Deep foundations / 301

Evaluating load vectors F,,, and F,,

The load vector F,,,, consists of five terms: three forces, F g , F,,' and F,$ (in the z,
and MII; (about the
X and y axes respectively). These are the forces and turning moments required for
a pure rotation of the pile group about they axis. To simplify the procedure it is
assumed that F@', F,$' and M@' are zero. M4;Y consists of two components (see
Figure 7.22):

X and y directions respectively), and two turning moments, M($

- (Ml$')pp, the push-pull component, which is the moment resulting from the

differential in pile axial loads in thex direction. (M,J)pp is obtained from an

(axi-symmetric or FSAFEM) analysis of an axially loaded pile and a
superposition technique. This is illustrated in Figure 7.25.
the pile head bending component, which is the sum of the bending
moments about they axis which are applied to the pile cap from each pile
in the pile group. (M&y)hbis obtained from a FSAFEM analysis of a pile
where the pile head is rotated and lateral deflections are restrained.
is the turning moment required to give rotation G,. F,," is the lateral force
required to prevent lateral deflections of the pile head. This is illustrated in
Figure 7.26.

The value of M(,,? for a particular Oy value is found by performing the above two
calculations and then summing the values of (M@Y)PPand (M,$)hb obtained.
The components of the load vector F,, are evaluated in a similar manner. Note
that these values of (Ml?;)PP and (M,,?')hb can be used to give the estimates of K,PP
and K,hb required in Section 7.4.4.1 :
(M&IPP and ~ , h =
~ v p p,b(M;)hb
@Y

@Y

Similarly, estimates for K,PP and Kxhbcan be obtained.

Figure 7.25: Evaluation of (MwY)PP

302 / Finite element analysis in geotechnical engineering: Application

Figure 7.26: Evaluation of / M + ~and

) ~ F+x
~

7.4.5.3 Evaluating the solution displacements and rotations

Vectors K,, Km, K,,,,,K,, and K,, are obtained by dividing the load vectors Fz9
P;,,,
F,, , F,, and F,, by their associated scalar displacements U:,U,,U,,0,and 0,
respectively. These vectors are measures of the secant stiffness of the pile group
and can be combined to give the pile group stiffness matrix [K]:
[ K ] = [ K z K, K , K , K,]

(7.1 1)

The various terms in [W and the method in which each term is derived is
depicted in Equation (7.12):

where:
K:

KZ,;
K,,,"Y
K,,?

K,,"

K,,;
K,,"'"

is obtained from the vertical force applied in an axi-symmetric analysis

of an axially loaded pile;
is obtained from the lateral force applied in a FSAFEM analysis of a
laterally loaded fixed head pile (in the x direction);
is obtained from the restoring moment applied in a FSAFEM analysis
of a laterally loaded fixed head pile (in the X direction);
is obtained from the moment applied in a FSAFEM analysis of a pile
subjected to turning moments and restrained from horizontal deflection
(about y axis);
is obtained from the restraining lateral force in a FSAFEM analysis of
a pile subjected to turning moments and restrained from horizontal
deflection (about they axis);
is obtained from the lateral force applied in a FSAFEM analysis of a
laterally loaded fixed head pile (in they direction);
is obtained from the restoring moment applied in a FSAFEM analysis
of a laterally loaded fixed head pile (in they direction);

Deep foundations / 303

K,,""

K,,?

is obtained from the moment applied in a FSAFEM analysis of a pile

subjected to turning moment and restrained from horizontal deflection
(about the x axis);
is obtained from the restraining lateral force in a FSAFEM analysis of
a pile subjected to turning moments and restrained from horizontal
deflection (about the X axis).

Using [K] the displacement U can be calculated for any applied load F.
However, the nonlinear nature of soil response means that [Kl is not a constant but
varies with U. An iterative technique is described below which ensures that the
value of U due to an applied load F is evaluated from a [K] which is consistent
with U.
The procedures described above allow the variation of F,, F,, F,, F,, and F,,
with respect to their associated scalar displacements U : , U, , U , , O, and O,
respectively, to be evaluated. These can be used to relate the vectors of secant
stiffness K,, K,, K,, , K,, and K,, with their associated scalar displacements U:,
U,, U,, Ox and O, respectively. These relationships can be expressed as design
charts which can be used to obtain the pile group displacements, U, due to any
applied pile group load, 8;: A simple iterative procedure for obtaining U is as
follows:

A trial value of pile group displacement, U', is used to evaluate from the
design charts the initial trial values of K,, K,, K,, ,K,,, and K,,;
These vectors are combined to obtain a trial value of pile group stiffness

EK!;
A trial solution for the pile group displacement, U""',is obtained based on
this value of [K] and the: applied loads F;
The validity of this solution is evaluated:
- If F = U""'(to within a specified tolerance) this solution is accepted;
- Otherwise this value of U"'is used as the new value trial displacement,
U',and the procedure repeated.

To start the procedure an initial trial displacement has to be obtained; usually a

small arbitrary displacement is used.
The solutions obtained from this proposed method of design only make three
basic assumptions:
- Superpositiontechniques can be used to determine the manner in which the
loads applied to one of the piles in a group affect the displacements of the
other piles;
- The rigid pile cap criterion can be used to determine the manner in which
the loads applied to the group are distributed between the piles in the group;
- The loads required to cause displacements U can be obtained by summing
the loads required to cause each component of U (superposition of loads).
To assess the validity of the last assumption for any applied load, the solution
displacements using this design procedure could be compared with the solution

304 1 Finite element analysis in geotechnical engineering: Application

displacements obtained using the analysis procedure described in Section 7.4.4.

This latter solution only makes the first two assumptions above.

7.4.6
Magnus
7.4.6.1 Introduction
The design and analysis procedures
Deckload 330 MN
for pile groups described above were
applied to the foundations of the
Magnus platform. Magnus is located
in the UK sector of the North Sea and
was installed in 186m of water in
1982. The structure is shown in
Figure 7.27. When the structure was NO
being designed, in 1979, it was to be
the deepest and most northern
structure in the North Sea. This
resulted in particularly severe
environmentaldesign criteria. This, in
with the very high deck Figure 7.27: Schematic presentation
of the Magnus structure
loads, results in the structure being
one of the heaviest steel structures
installed off-shore (at float out it weighed 340,000 kN). To verify the existing
foundation design methods the foundations for one ofthe legs (A4 in Figure 7.27)
were instrumented. Jardine and Potts ( 1 992) studied the vertical response of this
foundation using an axi-symmetric finite element analysis. The predictions
obtained were in good agreement with the field measurements.
In this section the work carried out by Jardine and Potts (1992) is extended to
consider the Magnus foundation response to general loading (axial and lateral
forces and turning moments). Design charts ofthe type described in Section 7.4.5.3
are derived. These design charts and the method of analysis described in Section
7.4.4 are used to analyse the performance ofthe foundations when subjected to the
loads measured from the largest wave of the stom on the 22ndJanuary 1984.

7.4.6.2 Soil properties and initial conditions

Rigden and Semple (1983) summarized the geotechnical surveys performed at
Magnus, which identified a thick and relatively consistent sequence of Quartenary
soils. Jardine and Potts (1 992) divided the soil profile into five main geotechnical
units (labelled I to V) which are identified in Figure 7.28. Stratum I may have been
deposited as a lodgement till, but strata 11-Vwere considered to be of glaciomarine
origin.
The constitutive model adopted for all these strata was a variant ofthe modified
Cam clay model, but different nonlinear expressions were substitutedto govern the
elastic behaviour and a Hvorslev surface was used on the dry side, see Chapter 7

Deep foundations 1 305

of Volume 1. The modified Cam clay
material properties used for each layer
and the depths of each layer are
shown in Table 7.1. The shape of the
yield surface in the deviatoric plane
was assumed to be a Mohr-Coulomb
hexagon, and a non-associated flow
rule was adopted, with the plastic
potential forming a surface of
revolution in stress space around the
hydrostatic axis. An elliptical form of
the plastic potential in J-p' space was
used on both the wet and dry side.
The critical state angle of shearing
resistance, v', was 30" for all layers.
The slope of the Hvorslev surface
with respect to the hydrostatic axis at
any Lode's angle was 0.75 times the
slope of the critical state MohrCoulomb hexagon.

DESCRIPTlON

GEOTECHNICAL.

a
.!m

1stiff to hardverv silw CLAY

Very stiff to hard very srlty CLAY

10

20

Hard to very stiffsilly CLAY with gravel.

rounded chalk, shell fragments ...
Pockets and lenses of fine silty SAND

....becoming weaker, fewer laminae and

less gravel. Stiff to very st S...

....becoming weakly fssured with oblique

1
'I1

and sub-vertical diicontinuities

Lenses of h e silty sand

Hard silty CLAY, with particles of chalk,
gravel and shell

80

Hard silty CLAY with cobble sized

enatics of sandy clay

t
U

Figure 7.28: Soil profile at Magnus

Table 7. l: Modified Cam clay material properties for Magnus soil

profile

Elastic behaviour is governed using the small strain stiffness model described
in Section 5.7.5 of Volume 1. Minimum values of E,and E, are specified such that
at strains below these values the shear and bulk moduli respectively are assumed
to be constant. Similarly, maximum values of E, and E,, are specified. The same
small strain parameters were used for all the stratum and are shown in Table 7 . 2 .

306 / Finite element analysis in geotechnical engineering: Application

Table 7.2: SmaN strain stiffness parameters

Interface elements were used to model the pile-soil interface behaviour. A

frictional yield surface was specified with an angle of shearing resistance of 22.5"
in strata I and 11, and of 17.7" in strata 111, IV and V. The elastic normal and shear
stiffness were both 500MN/m3.No excess pore pressures were allowed to generate
within the interface. A non-associative flow rule was used which allowed for no
dilation. The interface elements were allowed to open when the normal effective
stress reduced to zero. The amount of separation was recorded and the element
would not close until the value of separation reduced back to zero.
The pile was modelled as an elastic solid cylinder of diameter 2.134m. Elastic
stiffness parameters were specified such that the bending stiffness of this solid
model pile was the same as the bending stiffness of the steel pipe pile used for the
Magnus foundations (a Young's modulus of 45.3 GN/m2and a Poisson's ratio of
0.3). Note however that there is slight error in the resulting axial stiffness of the
pile (if axial stiffnesses were equated, a Young's modulus of 24 GN/m2would be
required).
The initial stress regime specified was that proposed by Jardine (1985) for the
stress state in the soil six months after the installation of the piles. In the far field
(greater than 20 radii from the pile) the stress regime was defined by the effective
bulk unit weight of the soil (from which vertical effective stress, o:,,',is calculated)
and the coefficient of earth pressure, K,,, see Table 7.3. The stresses adjacent to the
pile were defined as ratios of a,,', these ratios are also presented in Table 7.3. A
semi-logarithmicvariation in stress with radial coordinate was assumed for stresses
in the soil less than twenty pile radii away from the pile:

+ O g -3O a P

0-= Gap

where:

0 is

r
lnro

any one of the principal stresses in the soil at a radial coordinate r

and r,><r<20r,,;

Deep foundations 1 307

,a is the associated stress adjacent to the pile;
of is the associated stress in the far field;
r , is the radius of the pile.
A hydrostatic pore water pressure regime was assumed.

The overconsolidation ratios, OCR,specified for the soil were such that the
same undrained shear strength (in triaxial compression) profile with depth, as
shown in Figure 7.29, was obtained from the soil at all radial locations. The
resulting OCR variation with depth adjacent to the pile and in the far field is plotted
in Figure 7.30. To obtain the required undrained shear strength profile, a semi-log
variation, similar to that given by Equation (7.13), was used for the OCR of the soil
less than twenty radii away from the pile. The details of how these soil parameters
and the initial stress regime were derived are described fully in Jardine (1985).

OCR

Figure 7.29: Undrained shear

strength profile at Magnus

Figure 7.30: Variation of OCR

with depth at Magnus

308 / Finite element analysis in geotechnical engineering: Application

Table 7.3: Specified effective stress regime at pile face and far-field

7.4.6.3 Finite element analyses

The Magnus structure is supported on four pile groups of nine 2.134m outside
diameter open ended pipe piles, one pile group for each platform leg. The piles
were provided with an externally flush 1400mm long driving shoe having a wall
thickness of 80mm. Apart from this shoe the piles have a uniform wall thickness
of 63.5mm. This differential in pile thickness is ignored in the finite element
presentation. The pitch circle diameter of each group is 13.8m. Details of the
design and installation ofthe Magnus foundations are given by Rigden and Semple
(1983).
The finite element mesh used for an
isolated pile analysis of a pile in the Magnus
pile group is shown in Figure 7.3 1. To extend
the results from such an analysis to consider G
the behaviour of the pile group, the simple 3
interaction method described in Section
7.4.3.1 was used. The piles are assumed to be
uniformly spaced around the pitch circle. To
obtain the design charts presented in Section
7.4.5.3 and to perform the analyses of the
group as described in Section 7.4.4, the
lateral loads are assumed to act in the
direction of the x axis (i.e. through the centre
of pile 1) and the moments are assumed to
act about they axis, see Figure 7.32.
The initial stress regime described in
Section 7.4.6.2 is the predicted stress regime Figure 7-3 Finite
mesh for a sing1e pile
after the installation of the pile and just prior

Deep foundations / 309

to installation of the top side modules. A

number of processes had to be simulated
,,....@.''.-...'e-a~~'
.''
prior to considering the performance of the
a
pile group to environmentalloading. The first
stage of the finite element analyses was to ;
apply the deadweight associated with the '
installation of the topside modules.
',.. ..,
Accordingly, an axial load of 13.5MN was
applied to the isolated single pile. This
' 7@
..
......'
........ loading was undrained and a pore water bulk
modulus of one hundred times the effective
Figure 7.32: Magnus pile
bulk modulus of the soil skeleton was
layout
specified. After loading, the excess pore
water pressures generated were allowed to
dissipate to the initial hydrostatic pore water pressures. This is consistent with the
load transfer measured from the mudmat (pile cap) to its connecting piles (Sharp
(1992)). The rate of this load transfer decayed with time and was found to be
negligible after 20 months. After complete dissipationofthese pore water pressures
the accumulated soil strains were zeroed to retrieve the large initial stiffness
obtained at first load from the small strain stiffness model. This stage is taken to
be the initial condition from which environmental loads are applied. All subsequent
loading was undrained and a pore water bulk modulus of one hundred times the
effective bulk modulus of the soil skeleton was again specified. The design charts
presented in Section 7.4.5.3 and the analysis of the group presented in Section
7.4.4 consider loads and displacements from this stage. All FSAFEM analyses
undertaken assumed parallel symmetry, see Chapter 12 of Volume 1, and used 10
harmonics.

B.

7.4.6.4 Design of Magmus foundations

Design charts for the foundations of
40
400
the Magnus platform are obtained
_--------_________
using the method of design described
in Section 7.4.5. Figure 7.33 is a plot
300of the group axial load, F:, versus the 2:
group settlemenf U,. These results are
obtained from an axi-symmetric 3
analysis of an isolated pile axially
loaded (and the simple superposition 3 loo c3
procedure). Also shown is the group
load-displacement curve if the group
e f f e c t s were ignored ( n o
o
4
8
12
Group axial displacement, Uz(mm)
superposition). The effect of
superposition is to increase Figure 7.33: Group axial load and
displacements by approximately20%
stiffness vs. settlement

5
9

'5

31 0 / Finite element analysis in geotechnical engineering: Application

to 30%. The near linear response of the pile group to axial loading is illustrated by
the curve representing secant stiffness, K;, with settlement. For design purposes
a constant value of K; equal to 29.8GNlm can be assumed to apply for loads up to
365MN, the ultimate load for design.
6o
The ultimate capacity of the pile
5
group from these analyses is 8 10MN
m
(not shown in Figure 7.33). This
compares favourably with the
ultimate capacity of 837MN predicted
by Jardine and Potts ( l 992) for the ten
year initial stress regime. This is an
indication of how well the ten year
stress regime proposed by Jardine and
Potts (1 992) reproduces the combined
effects of the additional loading due
o
5
10
15
20
to the installation of the top side
Group lateral displacement, U, (mm)
modules and the subsequent
dissipation of the excess pore water
pressures generated. However, both Figure 7.34: Group lateral load and
these predictions are significantly
stiffness vs. lateral deflection
larger than the conventionally
15
250
determined ultimate capacity,
-,____---------___
549MN, and the ultimate capacity a
predicted by Jardine and Pans for the
200
six month initial stress regime, 585
MN (which are in agreement with
themselves). This difference arises
because these latter two analyses do
not account for the increase in
capacity associated with the
dissipation of the excess pore water
O0
S
10
15
20
pressures generated by the installation
of the top side modules.
Group lateral displacement, U,(mm)
The variation of group lateral
load, FII:~with lateral displacement, Figure 7-35; Group turning moment
is presented in Figure 7-34. These and stiffness vs. /a teral deflee tion
results are obtained by considering
the lateral loads and lateral deflections from a FSAFEM analysis of a laterally
loaded fixed head pile. The effect of superposition is similar to the effect of
superposition in the axial response and there is an increase in displacements of
approximately 20% to 30% from the response if group effects were ignored.
However, in contrast to the axial response, there is significant nonlinearity in the
lateral response. The secant modulus, K,: is also plotted and it varies from an
initial value of 4.3GNlm to a value of 3.3GNlm at a load of 35MN (the ultimate
load for design).

Deep foundations / 31 1

The variation of group turning

moments, M,,:, with lateral displacement,
U,, is presented in Figure 7.35. These 3 ,20
results are obtained by considering the
restraining moments and lateral ; ?
deflections from the same FSAFEM 8
analysis. There is considerably less
nonlinearity in this curve than there is g
E 40
from the curve of F,,,"versus U,. This can a
be seen from the variation of L e secant
stiffness,
also plotted on Figure
0
0
0.2
0.4
06
08
1
7.35. K,? varies from an initial value of
Rotation, B, (rad X 1000)
-13.2GN to a value of -13.8 GN at
10.4mm displacement (the displacement
Figure 7-36:Group turning
corresponding to a lateral load of 35MN).
There is also the same increase in
due to pile head
bending
vs. rotation
displacements due to superposition, as
obtained with the lateral loads.
25
To obtain the curves based on simple
superposition shown in Figures 7.34 and
7.35, it was assumed that the lateral load
was equally distributed between all the
piles in the pile group. This equal load 3
assumption results in satisfaction of the j
rigid pile cap criterion to within a I
reasonable tolerance. To illustrate this,
the lateral deflections of all the piles in
0
the pile group are evaluated, on the basis
0
02
04
06
08
1
of the equal load assumption, for an
Rotation, 0, (rad X 1000)
applied load equal to the maximum
design load (35MN to the group). These
Figure 7.37: Group lateral load
results are presented in Figure 7.40,
due to pile head bending vs.
which is a plot of the lateral deflections
rotation
of the piles against
the x-coordinate of
their centres. The variation in the magnitudes of these lateral deflections is small
and it is inferred that the rigid pile cap criterion is satisfied. Thus for the curves
shown in Figures 7.34 and 7.35 (which are for a range of lateral loads), only the
displacements of pile 1 are evaluated on the basis of the equal loads assumption
and the displacements of the pile group are equated to them. Note that the
displacements of pile 1 are the easiest to calculate since the additional
displacements due to piles 2 to 5 are equal to the additional displacements due to
piles 6 to 9.
The variation of group turning moments due to pile head bending,
with
group rotation, 0,, is presented in Figure 7.36. These results are obtained by

3 12 1 Finite element analysis in geotechnical engineering: Application

500
considering the turning moments and
rotations from a FSAFEM analysis of a
pile subjected to rotation and restrained
from lateral pile head deflection. No
superposition is required in this approach 5 300
since rotations rather than displacernents
200 are applied. The response in moderately
linear. The same FSAFEM analysis is 8
used to obtain the plot of lateral load,
F", and secant stiffness, K@',with group
O
rotation, Oy shown in Figure 7.37. The
0
0.1
0.2
0.3
0.4
0.5
lateral loads are obtained from the
Rotation, 6, (rad X 1000)
restraining loads required in the analysis.
This response is very linear and a
Figure 7.38: Group turning
constant value of K,," equal to -23.7GN
moment due to push-pull effects
is assumed for design purposes.
vs. rotation
The variation of group turning
5~
moments due to the puih-p;ll effect,
(M(,,,Y)PP,with group rotation, Oy, is ,
presented in Figure 7.38. The response is
nearly linear and is much stiffer than the
corresponding pile head bending
response, see Figure 7.36. These results
are obtained from an axi-symmetric a
loo analysis of an axially loaded pile and the
simple superposition technique. In
o
0.1
02
03
04
OS
addition, a linear variation of axial pile
Rotation, 0, (rad X 1000)
loads with x-coordinate was assumed.
This linear load assumption results in
satisfaction of the rigid pile cap criterion Figure 7.39: Total group turning
to within a reasonable tolerance. To
n-tOment and stiffness vs.
rotation
illustrate this, the settlements of all the
piles in the pile group are evaluated, on
the basis ofthe linear loads assumption, for an applied turning moment equal to the
maximum design turning moment. These results are presented in Figure 7.4 1 which
is a plot of the settlements of the piles against the x-coordinate of their centres. A
near linear variation in the magnitudes of these settlements with x coordinate is
obtained and it is inferred that the rigid pile cap criterion is satisfied. Thus to derive
the curve shown in Figure 7.38 only the displacements of pile 1 are evaluated on
the basis of the linear load assumption. The rotation of the pile group is evaluated
from this displacement and the assumption that the group datum does not displace
vertically, see Figure 7.41.
The push pull moments, (M,$)pp, and the pile head bending moments,
are added to give the variation of the total group turning moment, M,,,,y, with group

Q
Q

Deep foundations / 3 1 3

rotation, 0,. This is presented in Figure 7.39. Due to the linear nature of its two
constituent parts this response is also strongly linear, as can be seen from the
variation of secant stiffness, K@my,shown in Figure 7.39. K,,? is assigned a
constant value of 1 159GNm for design purposes.
evaluated above
The values of secant stiffnesses K,', K,,' KGy, Kwx and KhnJY
can be combined to give the pile group stiffhess matrix

[a:

where 3.3 <K,,"

4.3 and - 14.1 < KIamY< - 13.2. In the above equation the forces
are in terms of GN, displacements in terms of m, and the moments in terms of
GNm.
Both K,,>nd K,$y vary with displacement U, and the iterative procedure
suggested in Section 7.4.5.3 should be used in conjunction with Figures 7.34 and
7.35 to find the appropriate value of U, and hence K,!; and K,,"Y for any applied
load.

pile 3&8 A

pile 2&9

-0.2

- coordinate (m)

Figure 7.40: Pile head deflection

for maximum design lateral load

- coordinate (m)

Figure 7.4 1:Pile head settlement

for maximum design turning
moment

Horsnell et al. (1992) evaluated an average pile group stiffness matrix on the
basis of the measured response of the Magnus platform:

31 4 / Finite element analysis in geotechnical engineering: Application

The group stiffness matrix derived from the new method of design (Equation
(7.14)) compares very well with the stiffness matrix inferred from field
measurements. The new design value for K; is only slightly smaller than that
measured. Additionally, the measured value falls within the range of secant
stiffness shown in Figure 7.33. The new design method moderately overpredicts
the values of lateral and rotational stiffness K,," and Kwmy.Although not explicitly
stated by them, the Authors believe that Horsnell et al. (1992) assumed that the
foundation stiffness matrix was symmetrical. These off diagonal terms in the
measured stiffness matrix are in good agreement with the new design value of K,,",
but are significantly larger than the new design value of K1,Fy.It is interesting to
note that due to the nonlinear approach used to derive the terms in the new design
stiffness matrix, the matrix is no longer symmetrical (i.e. K,,"Y#KwX).
Horsnell et al. (1992) also presented the group stiffness matrix derived from
conventional design methods:

This conventional group stiffness matrix is significantly softer than the stiffness
matrix inferred from field measurements. The axial stiffness, K;, is over 4.5 times
softer than the measured axial stiffness and the other terms in the stiffness matrix
are approximately twice as soft as the corresponding measured terms.

7.4.6.5 Environmental loading

The method of design and method of analysis were used to predict the performance
of the instrumented pile group (supporting leg A4) on the Magnus platform when
subjected to the largest wave of the storm on the 22" January 1984. Horsnell et al.
(1992) state that the measured loads on leg A4 for this wave are: an axial load
equal to 69. lMN, a lateral load equal to 11.7MN and a turning moment equal to
146.7MNm. It is assumed that these lateral loads act in the x direction and these
turning moments act about they axis. Thus the applied loads can be expressed as
a load vector F:

The predicted displacements due to this imposed load obtained from the new
design method, Equation (7.14), and from the conventional design stiffness matrix,
Equation (7.16), are presented in Table 7.4. The new design method required four
iterations to obtain the solution value of U,. Also shown on Table 7.4 are the
measured displacements of the pile group for this wave (Horsnell et al. (1992)).
These measurements are based on the readings from accelerometers at the base of

Deep foundations / 31 5

the leg. The new design predictions are in good agreement with the measured
displacements, underpredicting lateral displacements and rotations and
overpredicting settlements. The conventional design predictions are significantly
larger than the measured displacements,with settlements over four and a halftimes
larger and lateral displacements and rotations approximately twice the size of
measured lateral displacements and rotations.
Table 7.4: Predicted and measured pile group displacements

- - - - New method of design

A New method of analysis

--- Conventional design

-Measured

displacements

9
B

- coordinate (m)

Figure 7.$2: Pile head deflection

for storm of 22/1/84

-Measured

displacemen&

-10

- coordinate (m)

Figure 7.43: Pile head settlement

for storm of 22/1/84

The method of analysis described in Section 7.4.4 was also used to give
predictions of pile group displacements and rotations. The initial trial division of
loads gave predictions that satisfied all but one of the rigid pile cap conditions
given in Section 7.4.4.3, the values of 69, from pile head bending and push-pull
loading were not equal. Five iterations were required to obtain a solution that
satisfied all the conditions. Note that additional FSAFEM analyses were required
for each iteration. The predictions for lateral displacement of each pile from the
method of analysis are plotted against X coordinate in Figure 7.42. Also presented
are the measured displacementsand the predicted displacementsfrom both the new
and conventional design methods. There is good agreement between the new

31 6 / Finite element analysis in geotechnical engineering: Application

design and new analysis predictions. This implies that the assumption of
superposition of loads described in Section 7.4.5.2 is reasonable.
Figure 7.42 also shows the good agreement of these predictions with the
measured displacements, and the large overprediction in displacements obtained
from the conventional method. The validity of the assumption of superposition of
loads and the accuracy of the predictions from the new methods of design and
analysis are also illustrated in Figure 7.43, which is a plot of vertical settlement
against X coordinate for the same imposed load.
Kenley and Sharp (1992) provide distributions for bending moment with depth
for a pile in the pile group subjected to the above wave loading. Distributions
obtained from both conventional design methods and measurements were
presented. The measured bending moments were derived from strain gauges at
different levels in different piles to give a typical bending moment distribution
within a pile. These bending moments are shown in Figure 7.44 with the predicted
bending moments from the new method of analysis. The new method of analysis
gives a better agreement with the measured bending moments than the
conventional method. In particular, the new method does not overpredict the depth
of the point at which the maximum bending moment occurs by as much as the
conventional method.

.............. New method

Bending moment ( m m )
Figure 7.44: Distribution of
bending moment with depth

Displacement (mm)
Figure 7.45: Distribution of
lateral displacement with depth

Kenley and Sharp (1992) obtained an estimate of the distribution of lateral

displacement with depth within a pile by double integrating the bending moment
profile shown in Figure 7.44. This is presented in Figure 7.45 with the predictions
obtained from the new method of analysis. The predicted displacements have a

Deep foundations 1 31 7

similar distribution as the measured displacements, but overpredict them. This is

in contrast to the results shown in Figure 7.42, where the measured pile head
displacements are greater than the predicted pile head displacements. Horsnell et
al. (1992) state that the group displacements presented in Figure 7.42, obtained
from the accelerometers at the base of the leg, may not necessarily apply directly
to the piles at mudline, since the leg-pile connection provided by the grout in the
pile sleeve is not necessarily rigid at this depth (thus actual pile displacements may
be smaller). The displacementprofile provided by Kenley and Sharp (1 992) is also
subject to inaccuracies due to the approximate manner in which the bending
moment distribution is obtained and difficulties associated with the double
integration of bending moments. However, the predicted values of displacements
from the new method of analysis and the new method of design fall within the
range of possible values given by these two measured estimates of pile
displacement.

7.5

Bucket foundation

7.5.1
Introduction
Suction caissons (or 'bucket' foundations)
are becoming extensively used in the
offshore industry as deep water anchors for
floating structures or foundations for oil
platforms. The caissons are hollow
cylindrical structures which have a top cap
and a relatively thin wall, the so called skirt.
One example of floating tension leg oil
platform on bucket foundations is the Snorre
platform, illustrated in Figure 7.46, which
has a cluster of three buckets connected to
each of its four foundation legs
(Christophersen et al. (1992), Jonsrud and
Finnesand (1992)).
Installing a bucket foundation involves
initial penetration into the sea-bedunder selfweight. The pressure in the water trapped
inside the bucket, between the soil surface
and the top cap, is then lowed by pumping, to Figure 7.46: Schematic view
cause a positive differential water pressure of Snorre platform foundation
template
across the top of the bucket, thus forcing the
bucket further into the soil until its final
position is reached.
In the case of the Snorre structure, dead weights were added to ensure that
compressive loading was acting on the bucket foundations. However, oil
exploration is moving more into progressively deeper waters, and there is

31 8 / Finite element analysis in geotechnical engineering: Application

considerable interest in utilising the tensile capacity of bucket foundations.

Environmental loading produces both vertical and horizontal components of force,
such that the resultant force is inclined to the vertical. One of the main design
considerations for such foundations becomes the assessment of their pull-out
capacity.
Zdravkovid et al. (1998), (2001) present an extensive numerical study of the
short-term pull-out capacity of bucket foundations in soft clay. An initial
parametric study (Zdravkovid et al. (1998)), involving three dimensional finite
element analyses of bucket foundations, in which the diameter, skirt length, soilstructure adhesion and inclination of loading were varied, considered the soil to be
an isotropic soft clay, with an undrained shear strength increasing linearly with
depth. This study produced a general picture of the influence of each of these
parameters on the pull-out capacity of bucket foundations.
Further studies (Zdravkovid et al. (2001)) were then carried out to investigate
the effects of soil anisotropy on the pull-out capacity of bucket foundations. For
this purpose the soil was modelled using the MIT-E3 model, see Section 8.7 of
Volume 1. As bucket foundations are becoming popular as mooring anchors, more
recent research at Imperial college has considered the effect of varying the
position of the location of the anchor tether down the side of the bucket. Some of
the main findings of this research will be discussed below.

7.5.2
Geometry
The reference bucket foundation
geometry adopted for the research
study is shown in Figure 7.47 and is
based on the geometry of the bucket
foundations installed at the Snorre
platform. A single concrete cylinder is
analysed, with a.skirt 0.4m thick and
a top cap 1.Om thick. The diameter of
the cylinder, D, is 17.0m, while the
skirt penetrates the sea-bed to a depth Figure 7-47:Idealised geometry for
finite element analysis
L=12.0m, giving a D1L ratio of 1.4.
The depth of water is also based on that at Snorre and is assumed to be 3 10m.
Undrained uplift loading is applied at the centre ofthe top cap, at an inclination
/3 to the vertical, see Figure 7.47. Analyses were performed varying the inclination
p from 0 to 90'. The scenario considered is typical of that for a tension leg
platform (TLP). Other types of floating structure may apply their loads in different
ways.
I

7.5.3

Finite element analysis

Although the bucket foundation has an axi-symmetric geometry, loading at a
general inclination /3 produces a full three dimensional state of stress and strain in

Deep foundations / 31 9

the bucket and in the soil. This implies that any realistic numerical analysis of this
problem must be three dimensional. Only in the special case of vertical loading
(i.e. p = 0") can the problem be assumed to be axi-symmetric.
The FSAFEM method was D,2=8,5m
used in this study for the analyses
ofbucket foundations subjected to
inclined loading, adopting the
finite element mesh presented in
Figure 7.48. As indicated in this
figure, prescribed displacements
were imposed on both the bottom
and right hand side mesh
boundary. No such displacement
restrictions are imposed on the left
hand side boundary, because it
represents only the geometrical Figure 7.48: Finite element mesh for
axis of symmetry. The parallel
bucket foundation analysis
symmetry option was adopted and
ten harmonics were used to represent the variation of all quantities in the
circumferential direction. A small parametric study, varying the number of
harmonics, was performed to ensure that this was adequate. For a vertically loaded
foundation (i.e. P=O0) a conventional axi-symmetric finite element analysis was
performed, which further reduced the computational effort required.
The bucket foundation was modelled as concrete material, assuming linear
elastic behaviour (with Young's modulus E = 30* 106kN/m2 and Poisson's ratio
p=0.15). This implies that the concrete has sufficient strength and reinforcement
to sustain the applied loads. The installation of the foundation was not modelled
in the analyses, the 'bucket' was assumed to be 'wished in place'.
For the conventional axi-symmetric analyses, when P=O0,the foundation was
loaded to failure by applying increments of vertical uplift displacement, v, at the
centre of the top cap. The reaction force at this location represents the applied load.
For the foundation subjected to inclined loading, when P+O0,the analyses had to
be performed under load control, with increments of vertical and horizontal force
being applied as line loads (in the circumferential direction) at the node
representing the top right hand corner of th: top cap, point A in Figure 7.48. As
with the analyses described in Section 6.5.7.4 for circular footings, for the vertical
load (V) only the zero harmonic coefficient was specified as non-zero. To obtain
the horizontal load, both radial (H,) and circumferential (H,,) line loads were
specified. For the radial load only the first cosine harmonic was specified as nonzero, whereas for the circumferential load only the first sine harmonic coefficient
was set. The latter was of opposite sign but equal in magnitude to that set for the
radial load. The magnitudes of these increments were kept in the proportions
required to give the desired loading inclinations.As failure was approached, it was
necessary to reduce the size of the force increments to obtain an accurate estimate

320 1 Finite element analysis in geotechnical engineering: Application

of the ultimate loads. The resultant load from these boundary conditions acts at the
centre of the top cap as desired.
The finite element mesh shown in Figure 7.48 was used for both axi-symmetric
and FSAFEM analyses. It consisted of 350 eight noded isoparametric elements.
Undrained conditions in the soil were simulated by assigning a high bulk stiffness
(=1000K,,,,, where K,,, is the effective bulk stiffness of the soil skeleton) to the
pore water (see Section 3.4 of Volume l).

7.5.4
Modelling of the interface between top cap and soil
To model the interface between soil and the top cap, zero thickness interface
elements have been used (see section 3.6 of Volume 1). At the outset of this
research programme, three different conditions at the interface between the soil and
the top cap were identified. These are shown diagrammatically in Figure 7.49. In
Case 1 the bucket foundation has been perfectly installed such that there is no
water layer between the soil surface and the underside ofthe top cap. It is assumed
that the bucket foundation is water tight. Case 2 represents the other extreme where
a water layer exists between the soil surface and the underside of the top cap and
the pressure in the water remains equal to that at the sea bed outside the foundation
(i.e. there is a hole in the top cap). Case 3 models the intermediate situation in
which a water layer exists, but the foundation is water tight.
All three situations were
hole
modelled in a conventional
axi-symmetric finite element
analyses in which the
foundation (D= 17m,DIL= 1.4)
was subject to a vertical pullout load. The isotropic~soil
Case 1: NOinterface
Case 2: Interface elements
properties described in the
with no tension capacity
next section were used for
these analyses. Case 1 was
modelled by omitting the top
cap interface elements and
Case 3: Interface elements
thereby simulating perfect
with tension capacity
contact between the soil and
the underside of the top cap.
Case 2 was mdelled by Figure 7.49: Possible scenarios for the
letting the interface ekments
interface between the top cap and
open freely under any vertical
undersoil
loading and Case 3 was
modelled by giving the interface elements a very low shear stiffness, but a high
normal stiffness. In addition, if the water pressure was to reduce to an absolute
value of - IOOkPa, a facility was provided to reduce the normal stiffness to a very
small value, thus simulating cavitation. For all analyses full wall adhesion was
assumed between the soil and the skirts of the bucket foundation.

Deep foundations 1 321

The results of the three analyses

are presented in Figure 7.50 in the
form of vertical load-displacement E 40
curves. As expected, the analysis for $
Case 2 produces a much lower limit 3
20
load than for the other two cases. This
arises as resistance is only provided .?
by the adhesion between the skirts of $
the foundation and the soil. In Cases
o
0.2
0.4
0.6
0.8
I
Vertical displacement (lifting) (m)
1 and 3 there is an additional
contribution from the Soil below the
Figure 7-50;
Load-displacement
top cap, increasing the ~ltirnatepull curves for different interface cases
out by approximately 180%. The
result for Case 3 shows a softer response to that of Case 1. However, both Cases
1 and 3 give the same ultimate load. The differences between Cases 1 and 3 arise
because in Case 1 horizontal shear stresses are mobilised between the soil and the
top cap, whereas in Case 3 these are zero (i.e. the water layer trapped between the
top cap and the soil cannot transfer shear stresses).
As Case 3 was thought to better represent the field situation, all further analyses
were performed with this condition. It may be noted that for the Case 3 analysis
presented in Figure 7.50 and for all further analyses presented in this chapter, the
water trapped between the soil surface and the underside of the top cap did not
cavitate. This is, perhaps, not surprising as initially, before loading, this water had
a compressive pressure equivalent to 3 10m head of water.

7.5.5 isotropic study

7.5.5.1 Soil conditions
As mentioned in the Introduction, the initial
Undrained strength, S, (Ha)
study on the pull-out capacity of bucket
foundations in soft clay was performed by
?Oo
assuming the soil to be isotropic. A lightly
overconsolidated clay (OCR= l . l ) was assumed,
and a form of the modified Cam clay (MCC)
constitutive model was used to simulate its
behaviour. The yield and plastic potential
surfaces are given by a Mohr-Coulomb hexagon
and a circle respectively in the deviatoric plane
(see Chapter 7 of Volume 1). The material
properties used in the present investigation are
summarised in Table 7.5. The triaxial
compressionundrained shear strength,which can
be derived from the basic parameters as indicated Figure 7.5 1: Undrained
in Appendix VII.4 of Volume 1, varies linearly strength for isotropic soil

322 / Finite element analysis in geotechnical engineering: Application

with depth (see Figure 7.5 l), giving Su/a,,'=0.33, which is typical of soft clays in
triaxial compression (Hight et al. (1987)).
Table 7.5: Modified Cam clay parameters for soft clay

7.5.5.2 Parametric studies

The reference geometry, identified in Figure 7.47, gives a DIL ratio of 1.4. Three
parametric studies were performed to investigate the pull-out capacity of bucket
foundations:

To investigatethe effect ofembedment, the skirt length was varied, keeping

the same diameter(D=17.0 m), to L=8.0 m (DIL=2.1), L=12.0 m (DIL=1.4),
L=17.0 m (DIL=l .O) and L=24.0 m QDIL=0.7);
To investigate the effect of the bucket diameter, analyses were performed
with similar DIL ratios, but with a diameter D=8.5 m;
Most of the analyses were performed assuming full adhesion, c,(=aSu),
between the soil and the skirt, which means that the full undrained strength
is mobilised between the soil and the skirt at failure (i.e. a=l). To
investigatethe effect of a reduced skirt adhesion, some of the analyses were
repeated with a skirt adhesion equal to 50% the undrained shear strength of
the adjacent soil.

In each of these studies the inclination, p, of the pull-out force, T, was varied
between 0" and 90".

7.5.5.3 Results
Results from a typical analysis are shown in Figure 7.52. This analysis is for a
,4
bucket foundation with D=17.0m
and L=12.0m (i.e. DIL =1.4), 8
12
subjected to a pull-out force, T,
10
inclined at 70" to the vertical (i.e. 3
P=70). The vertical component, V,
of this force is plotted against the 8 6
vertical displacement and the
4
horizontal component, H, of this .g 2
force is plotted against the X
'0
0.2
0.4
0.6
0.8
1
1.2
1.4
horizontal displacement of the
Horizontal and vertical displacement (mm)
foundation. Figure 7.52 shows that
clear limit values are predicted for
Figure 7.52: Typical load-displacement
both the horizontal and vertical
curves for inclined loading
load components at failure.

!g

Deep foundations / 323

Figure 7.53 shows, for this

particular analysis, the vectors of
incremental displacements at failure.
While the absolute magnitudes of
vectors are not significant, the
directions and relative magnitudes
indicate the nature of the failure
mechanism.

Full skirl adhesion

. . . . . . . . . , . . . . . . . .. . . . . . . . .
: . . . . ._. .. , .
. . .. .
. / I .

l
I

The combinationsofultimate vertical, ~i~~~~

7-53..vectors
of incremental
VU!t and horizontal, Huit , loads at
djsplacements at the end of an
which failure is predicted, considering
analysis
a range of bucket skirt lengths and
diameters and full skirt adhesion, are plotted on the interaction diagram in Figure
7.54. As expected, the magnitudes of the ultimate loads are dependent on the
diameter and depth of embedment. However, for a particular foundation geometry
the finite element predictions indicate that there is an approximately elliptical
relationship between the ultimate vertical and horizontal loads, developed under
, varying from 0' to 90'. The ellipses fitted to the numerical
inclined loading with B
predictions are plotted in Figure 7.54. The maximum error between these curves
and the numerical predictions is =k (0.6 to 1.2)%. Three main conclusions can be
derived from this figure:
9

- For a particular geometry,the largest pull-out capacity occurs under vertical

loading and it continually reduces as the inclination of loading increases

(i.e. becomes more horizontal);
The capacities of bucket foundations with constant diameter are nearly
directly proportionalto their skirt length (e.g. halving the length of the skirt,
and hence its contact area with the soil, practically halves the limit loads);
Changing the bucket diameter, while keeping the same skirt length, has a
larger impact than varying the skirt length while keeping the same diameter
(e.g. halving the diameter results in a 3.5 times reduction in ultimate loads,
while the contact area between the soil and skirt reduces by a factor of 2).

The results plotted in Figure 7.54 are replotted in a normalised form in Figure
7.55. The data from a particular foundation geometry and for a particular load
inclination has been normalised by dividing the ultimate horizontal force, H,,,, and
the ultimate vertical force, V,, , by the maximum vertical force, V, . V, is the
maximum vertical load found from the analysis involving purely vertical loading
for that particular foundation geometry. The values of H,,/V,, and V,,,/V,,, found
for all the geometries and pull-out inclinations indicate a unique relationship which
can be represented by an ellipse of the following form:

324 1 Finite element analysis in geotechnical engineering: Application

All data fit this ellipse with a maximum error of only (1.5 to 3.0)%.
Clearly, Equation (7.17) cannot hold for large DIL ratios because in the limit,
when the embedment depth L approaches zero, the ultimate horizontal force also
becomes zero (i.e. the water interface between the underside of the top cap and the
soil cannot sustain shear stress). Further analyses have been performed to
determine the limiting value of DIL over which Equation (7.17) does not apply.
These analyses indicate that Equation (7.17) is valid for DIL < 4.
e D/L=17/12

Fitted ellipses
90"

E
A

DIL=17/17
DIL=17/8
D/L=17/24
DIL=8.5/8.5

Ultimate vertical tensile force, V, (MN)

Figure 7.54: Envelopes of ultimate horizontal and
vertical load - full skirt adhesion

Figure 7.55: Normalised ultimate loads full skirt adhesion

Reduced skirt adhesion

In the analyses described above, where the full adhesion between the foundation
skirt and the adjacent soil was assumed, there was no need to incorporate interface
elements in this locality. However, zero thickness interface elements were
introduced between the skirt and the soil for the set of analyses in which the skirt
adhesion was assumed to be 50% of the undrained strength ofthe adjacent soil (i.e.
c,=0.5Su). These elements were assigned normal and shear stiffnesses of 1OOG and

Deep foundations 1 325

10G respectively, where G is the elastic shear modulus of the adjacent soil which
varies with depth.
Two sets of analyses were repeated with the reduced skirt adhesion, considering
DIL=17/12 and DIL=17/24. The resulting interaction diagram is presented in Figure
7.55. It can be noted that:

- The finite element predictions for a particular geometry can be fitted with

an elliptical curve, similar to the analyses performed with full skirt

adhesion, shown as full lines in Figure 7.56.
The ultimate loads obtained for 50% reduced skirt adhesion are
approximately 10 to 15% smaller than those for the full skirt adhesion
(marked as dashed lines in Figure 7.56), indicating that the reduction in
ultimate loads is not directly proportional to the reduction in skirt adhesion.

The results are also plotted in normalised form in Figure 7.57, together with the
ellipse from Equation (7.17).They agree with this curve with a maximum error of
6%.

Fitted ellipses

D/L=I7/12
D/L=17/24

Ultimate vertical force, V , ,(MN)

Figure 7.56: Envelopes of ultimate horizontal and

vertical load - reduced skirt adhesion

Figure 7.57: Normalised ultimate loads reduced skirt adhesion

326 1 Finite element analysis in geotechnical engineering: Application

7.5.6

Anisotropic study
7.5.6.1 Introduction
To model the anisotropic behaviour of the soil the MIT-E3 model has been used.
Selection of appropriate parameters for a soft sea bed clay have been discussed in
some detail in Section 6.5.7. The parameters are listed in Table 6.4 and these have
been adopted for the analyses presented here.
The undrained triaxial compression strength profile for this soft clay is shown
in Figure 6.42. It varies linearly with depth, starting from a finite value at the soil
surface, giving S,,/cv'=0.36. This triaxial compression profile differs slightly from
that used in the previous isotropic analyses, but is a consequence oftrying to model
the anisotropy experienced in the Hollow Cylinder Apparatus (see Section 6.4.1).
The difference has no bearing on the conclusions that come.

7.5.6.2 Results
In order to quantify the effect of soil anisotropy on the pull-out capacity of bucket
foundations two sets of analyses were performed, assuming the foundation to have
a reference geometry (DIL=17/12) and full skirt adhesion:
Analyses where the bucket foundation was loaded with a tensile force
whose inclination, p, varied from 0" to 90" and in which the soil was
assumed to be anisotropic, i.e. to possess the variation of undrained strength
presented in Figures 6.4 1 and 6.42.
Analyses where the bucket foundation was loaded in the same way, but the
soil was assumed to be isotropic. The soil properties for the MIT-E3 model
were chosen such that, for any inclination a of the major principal stress,
the undrained strength was the same as that obtained from the anisotropic
soil when sheared with a=OO, see Section 6.5.7.3. This means that the
effective stress paths for any a match the effective stress path for test MO
in Figure 6.4 1 .
Again, an interaction diagram of ultimate vertical and horizontal forces for each
particular inclination of loading is produced, as shown in Figure 7.58. The upper
envelope represents the results from the isotropic analyses, while the lower
envelope is for the anisotropic analyses. Similar to the previous analyses, both
isotropic and anisotropic numerical predictions can be fitted with a simple elliptical
curve, with a maximum error of k (0.6 to 1.2)%. However, the main conclusion
from these analyses is that taking account of soil anisotropy reduces the pull-out
capacity of the bucket foundation. For this particular soil the anisotropic analyses
lead to capacities that are about 22% lower than those for isotropic cases. Clearly,
the reduction of pull-out capacity due to anisotropic soil behaviour will depend on
the amount of strength anisotropy that each particular soil possesses.
The resulting normalised interaction diagram is presented in Figure 7.59,
together with the ellipse from Equation (7.17). The analytical data points conform
to the curve with a maximum error of 6%.

Deep foundations / 327

Ultimate vertical force, V,,, (MN)

Figure 7.58: Envelopes of ultimate horizontal and

vertical load from isotropic and anisotropic analyses
using the MIT-E3 model

Figure 7.59: Normalised ultimate loads from isotropic

and anisotropic analyses using the MIT-E3 model

7.5.7
Suction anchors
7.5.7. 1 Introduction
In the previous sections the behaviour of bucket foundations loaded at the centre
of the top cap has been considered. Such a loading situation is typical of that of a
tension leg platform, but may not mobilise the maximum possible capacity of the
anchor. Recent results from large scale field test have shown that by lowering the
attachment point of the loading tether from the top to half way down the skirt, the
load capacity increases (Anderson (1998)). To illustrate this effect the results of
some numerical analysis in which the position of the loading tether is varied will
now be described.

7.5.7.2 Geometry
The bucket is now assumed to be made of steel and have a diameter of D=3m and
a skirt length of L=9m, see Figure 7.60. The loading tether is assumed to be
attached to a point some distance l down the skirt.

328 / Finite element analysis in geotechnical engineering: Application

The finite element mesh is shown in

+
D
e
Figure 7.61. As the software that was used
did not have the facility to use shell elements
in Fourier Series Finite Element Analyses,
the steel bucket was modelled using solid
elements. In reality the steel of the bucket is
likely to be approximately lOmm thick,
however if this had been modelled in the
analyses then a large number of elements
would have to be used to avoid elements with
large aspect ratios. Consequently the steel
was modelled as 100m thick, but its Young's
modulus, E, was reduced from 2 10x106
kN/m2 to 30x106 kN/m2 to compensate. The
Poisson's ratio p=0.15. The clay was
assumed to be isotropic and was therefore
modelled using the modifiedcam clay model Figure 7.60: Geometry of the
and the parameters given in Table 7.5.
suction anchor used in
analysis
L-15m

Figure 7.6 1: Finite element mesh

for suction anchor analysis

Figure 7.62: Anchor loading

modelled in FSAFEM

As the attachment point of the loading tether is positioned on one side of the
bucket and at some distance down the skirt, care must be exercised when
specifying the loading conditions. In the present analysis a single line load was
applied at the node on the outside of the bucket at the required depth down the
skirt. In addition a local axis (see Section 3.7.2 of Volume 1) was specified at this
node so that one of the axis aligned in the direction of loading (i.e. P to the
vertical). The line load was then oriented in this direction, but its magnitude varied

Deep foundations 1 329

in the circumferential direction as shown in Figure 7.62, which shows the line load
distribution for an increment ofthe analysis. The load has an intensity of IOOkNlm
over a 40" sector of the skirt and is zero over the remaining part. This gives a
resultant load in the direction of the tether of lOOkN per increment.
This loading condition was specified using the fitted method described in
Appendix XII.7 of Volume 1. In this approach a series a line load intensities and
the values of B at which they act are input.

7.5.7.3 Results
Analysis have been performed with
the attachment point at four positions,
namely at the top of the skirt, one
third of the way down the skirt, half
way down the skirt and at the bottom
of the skirt. For each attachment point
analyses have been performed with
the direction of loading P equal to 0"
(i.e. vertical), 45", 70" and 90" (i.e.
horizontal), see Figure 7.60.
Angle of anchor inclination to the vertical c)
The ultimate capacities predicted
by these analyses are summarised in Figure 7.63: ultimate anchor pullFigure 7.63. Interestingly
the out force for different anchoring
analysis with vertical loading predict
points
approximatelythe same load capacity,
independent of the depth of attachment of the tether. However, this is not so if the
loading is inclined, where the analyses indicate that the greatest capacity is
predicted for the analysis with the attachment point located half way down the
skirt. The analyses are therefore in agreement with the field observations.

7.6
1.

2.

3.

4.

Summary

When analysing a single pile subject to axial loading either thin solid elements
or special interface elements should be placed adjacent to the pile shaft. If
solid elements are used and they are not sufficiently thin, the analysis will
over estimate the pile shaft capacity.
If interface elements are positioned adjacent to the pile shaft care should be
taken when selecting their normal and shear stiffness. These values, if not
sufficiently large, can dominate pile behaviour. However, if the values are too
large numerical ill conditioning can occur.
Soil dilation can have a dominant effect on pile behaviour and consequently
care must be exercised when selecting an appropriate constitutive model and
its parameters.
When analysing axially loaded piles using an effective stress constitutive
model it is unwise to use a model which predicts finite plastic dilation

330 1 Finite element analysis in geotechnical engineering: Application

5.

6.

indefinitely, without reaching a critical state condition (e.g. the MohrCoulomb model with v>Oo). Such analyses will not predict an ultimate pile
capacity.
When analysing a single pile subject to lateral loading the possibility of a
crack forming down the back of the pile (i.e. gapping) should be considered.
If this is likely to occur interface elements should be installed along the pile
soil interface, which cannot sustain tensile normal stresses. Wether or not
gapping is likely to occur depends on the soil strength and in particular its
distribution with depth.
A new method of design and analysis for pile groups is proposed. This
method is thought to be an improvement on the existing methods because:

- A more sophisticated and rigorous method of analysis of single piles is

7.

used which is able to incorporate complex soil and pile-soil interaction

behaviour;
A more coherent method for accounting for pile group interaction is used
which is based on the results of single pile analyses.

The proposed method of analysis predicts pile group rotations and

displacements for any applied load. The method is an iterative procedure that
requires FSAFEM analyses be carried out for each iteration. The basic
assumptions in this method are:

- Superposition techniques can be used to determine the manner in which

-

the loads applied to one of the piles in a group affects the displacement of
the other piles;
The rigid pile cap criterion can be used to determine the manner in which
the loads applied to the group are distributed between the piles in the
group.

The design method produces design charts from which the pile group
displacements and rotations can be evaluated without recourse to additional
FSAFEM analysis. The basic assumption made by this method, in addition to
the two method of analysis assumptions, is that the loads required to cause
displacements U can be obtained by summing the loads required to cause each
component of U (superposition of loads).
9. The proposed analysis and design methods for a pile group were applied to
the foundations for the Magnus platform. The pile group stiffness obtained
from the new method of design was in good agreement with the measured
foundation stiffness.
10. The methods of design and analysis were used to predict the pile group
displacements and rotations for the loads imposed by the largest wave of a
storm on the 22ndJanuary 1984. The predictions obtained were shown to be
in good agreement with each other and with the measured displacements and
rotations. The predictions from the new methods are shown to be much more
accurate than the predictions from conventional methods.

8.

Deep foundations 1 331

1 1. A parametric finite element study has quantified the effects on the pull-out
capacity of 'bucket' foundations loaded at their top of (i) foundation
geometry, (ii) skirt adhesion and (iii) load inclination. The main conclusions
are as follows:
- The largest pull-out capacity occurs under vertical loading;
- Pull-out capacity continually reduces as the inclination of loading rotates
from the vertical;
- Foundation diameter affects pull-out capacity more significantly than
embedment depth;
- Ultimate pull-out capacity is not directly proportional to the value of skirt
adhesion.
12. Finite element analyses using the MIT-E3 model have shown how account
may be taken of the anisotropic soil strength observed in the laboratory.
Anisotropy in S, was seen to reduce the pull-out capacity by approximately
22%, compared to equivalent analyses assuming isotropic behaviour.
13. It has also been shown that normalising the ultimate horizontal and vertical
pull-out forces with respect to the ultimate vertical force, obtained from an
analysis involving only vertical pull-out, results in a unique interaction curve,
at least for the soil profiles considered and for DIL < 4, that specifies a nondimensional failure envelope that is independent of the foundation
dimensions, skirt adhesion or soil anisotropy. This curve is of a simple
elliptical form with ratio ofminor to major axis of 0.235, and once the vertical
pull-out capacity has been evaluated, designers may use such a curve to
determine the capacity under any inclined loading conditions, without
performing additional complex three dimensional analyses.
14. It should be noted that the above conclusions are valid for tension leg
platform type bucket foundations that are loaded at the centre of the top cap.
If such foundations are used as suction anchors, for mooring other vessel
types, the load is often applied at a point someway down the skirt. In such
cases it has been shown that the maximum capacity is obtained when the point
of load application is approximately half way down the skirt and the load is
inclined at 45Oto the vertical.

8. Benchmarking
8.1

Synopsis

This chapter discusses the benchmarking of finite elements programs, but the
comments made are equally applicable to the testing of other computer software.
The aim ofthis chapter is to demonstratethat problems can arise when undertaking
finite element analysis, which, unless they are detected, can have serious
implications.
A number of geotechnically orientated benchmark problems are presented and
discussed, together with the results of a benchmarking exercise carried out to
assess the reliability of a number of finite element programs.

8.2

Definitions

Throughout this chapter the following terns are used:

Benchmark: A problem formulated to check the performance of a computer
program.
Benchmarking: The processes of testing, validating, verifying or checking the
performance or operation of software.
Closed form solution: An exact and unique theoretical solution to a well
defined problem, which usually involves the analytical solution of a series of
equations without the use of iterative techniques.
Computer Hardware: Physical components of a computer, such as hard discs,
floppy discs, disc drives, printers, processors, etc.
Computer Software: The computer code that provides the computer with the
instructions necessary to perform a set of calculations. This includes the
operating system used by the computer (e.g. Windows@ 98, Windows@ NT,
Unix, etc), a compiler and any manuals that explain how the software operates
and the method of calculation.
Developer(s): The perssn(s) or organisation(s) responsible for writing and
developing the computer software.
User: The person or organisation who simply uses the developer's software,
but does not produce code or modify the operation of the software in any way.
Validation: Ratifying or confirming that a particular operation is being
performed correctly. The process of evaluating software at the end of the

Benchmarking / 333
software development process to ensure compliance with software
requirements.
VerlJication: Establishing the truth or correctness by examination or
demonstration. The process of determining whether or not the products of a
given phase of the software development cycle fulfil the requirements
established during the previous phase.
When applied to the testing of computer software, and its operation, the terms
'validation' and 'verification' are often confused, since to most people they have
a very similar meaning. Although in some technical literature a distinction is made
between the two, in this chapter they are taken to mean the process of testing
computer software to evaluate its performance, assessing its accuracy and correct
operation by an individual and on specified hardware.

8.3

Introduction

Computers have for many years been available to engineers to aid design. In recent
years engineers have had the possibility to routinely use advanced numerical
methods (finite element analysis, boundary element analysis and finite difference
techniques) in design. The possibility to do so is largely due to the increased
availability of such programs, combined with the rapid pace of hardware
development. It is apparent that the use of these methods is becoming more
widespread, as more engineers appreciate the benefit of using such techniques to
examine complex problems and the sensitivity of these problems to various
influences.
When undertaking any calculation, an engineer has a responsibility to check
that the calculation is appropriate and numerically correct. A check should
therefore be made to ensure that the tools used are working correctly and that the
results obtained are reliable. It is important to remember that computers and
computer programs do not have a mind of their own, they simply execute a series
of pre-defined instructions. The question is, do these instructions correctly follow
the desired method of analysis and produce the intended result, andtor does the
user understand what the result is.
There is a number of reasons why errors may occur in calculations performed
using computer programs. These include errors or bugs in the software through
mistakes in coding or interpretation of the method of analysis, the use of the
software, or, in exceptional circumstances, there may be hardware problems. A
technique for checking the computer program and the use of it is known as
benchmarking. Essentially, it is a technique which forms part of the validation or
verification process.
This chapter starts by discussing, in general terms, the issue of verification and
validation of computer software and then goes on to discuss the benchmarking of
finite element programs, with particular reference to geotechnical engineering.
Most of the principles and techniques are equally applicable to the finite element
analysis of other problems (structural, thermal, hydraulic, etc.) and to the use of
computer programs in general.

334 / Finite element analysis in geotechnical engineering: Application

The principles of benchmarking (what is benchmarking, why is it necessary and

what is to be gained from it) are outlined and some examples of benchmarking
exercises of geotechnical problems are discussed.

Causes of errors in computer calculations

Problems may arise with computer calculations for a number ofreasons. The most

8.4

common causes are as follows:

- An error in the interpretation of the underlying theory by the developer. The

-

analytical technique or theory has not been understood correctly by the

programmer and therefore has not been coded correctly.
There are errors, mistakes or 'bugs' in the code, which mean that the program
does the wrong calculation and not the calculation that the developer or user
believe the program is doing (refer to the problems associated with the Aston
Park multi-storey car park, Section 8.5). These 'bugs' may extend to the
operating system or compiler, which cause problems to occur even when a
previously tested version of a program is run. Such problems are not
uncommon even now.
The numerical techniques used are not appropriate or are prone to error. There
may be some numerical instability which results in mis-leading answers. In
finite element analyses nonlinear problems are solved using iterative methods.
Under certain circumstances convergence may be difficult to attain and may
require several attempts using different increment sizes (refer to Chapter 9 of
Volume 1).
Interface conflicts between software and hardware or between software and
software. The same program gives different answers when run on different
types of computer, or under a different operating system.
The user does not understand the theory or the method of calculation used by
the program, and therefore cannot formulate the problem in the correct way.
The program is not being used in an appropriate way, i.e. using it to analyse a
problem for which it was not designed.
Mistakes in manuals or the method used to input data (pre-processing) that
mislead the user and mean that data is not input correctly. Likewise, there may
be errors in the post-processing such that, although the calculations are correct
and accurate, incorrect results are presented to the user.
Faults in the hardware. When Pentium processors were first introduced there
was a flaw which gave inaccurate results in division calculations which lead to
a 'reduction in precision'. This obviously had serious implications for many
users who had to undertake comparative calculations using other hardware
(NCE ( 1 994)).

The computer programs and hardware need to be checked before they are used,
in order to ensure that none of these problems exist. As the three examples of
failures resulting from errors demonstrate (see Section SS), the consequences of

Benchmarking / 335
an error being made could be severe and could have extremely serious
implications. It is obviously important that any of the potential problems with the
computer program, or the use of it, are identified and dealt with.

8.5

Consequences of errors

The consequences of errors occurring in any calculation can vary enormously.

There may be small errors which, in practical terms, are of little significance for
most situations, but under certain conditions they become much more critical. This
may result in the reduction of a factor of safety, or, in exceptional circumstances,
a structure may fail. Alternatively, the error may lead to over design, incurring
unnecessary construction costs, or it may falsely highlight an apparent problem
which has to be resolved incurring additional cost or economic loss. In the
Authors' experience the latter is not uncommon. For example, tunnel linings have
incorrectly been shown to be overstressed and movements of structures have been
erroneously overestimated. These erroneous results have thrown into question the
viability of quite significant schemes, or delayed construction while the problems
have been resolved.
If failure occurs, the consequences may not be just economic, but lives may be
put at risk. Failures that were attributed to errors in computer programs, and their
use, include the following cases:
The Aston Park multi-storey car park (constructed circa 1983). After
construction cracks were noticed in the slabs of the reinforced concrete
structure on the upper two levels. Eventually, cracks spread to the lower floors
and it became clear that there was a serious problem. Apparently faulty
programming resulted in reinforcement being placed 90" out of orientation.
Essentially, 'the slabs were light of steel in the direction of span in which they
should have been heavy'. The report on the failure concluded that an 'off the
shelf computer program was to blame (it was subsequently acknowledged by
the developer that there was a bug in the program) and went on to severely
criticise the designers, claiming that there was no evidence of any checks being
carried out. To make matters worse, the program indicated that there was a
problem with the computation but, due to a failure of understanding between
the users and the developers, the full implication of the fault was not
appreciated. The designer apparently mistakenly believed that he fully
understood the output and the significance of the warning signal was missed
(NCE (1983), (1984a) and (1984b)). This failure caused some alarm at the
time, since the program was commonly used by engineers to design structures
of this form.
Trial wind turbine (circa 1990). The failure of a blade on an advanced wind
turbine was caused by a wrongly written computer program. The tension side
of the blade split and progressive failure caused the other half to break off. The
finite element program used for the analysis had apparently been adapted to suit

336 / Finite element analysis in geotechnical engineering: Application

the project (NCE (1 99 l)). At the time concerns were raised about the safety of
other turbine blades produced by the same manufacturer.
Sleipner offshore platform (1993). In August 1993 the concrete gravity base of
the Sleipner A platform sank during ballasting trials. An explosion was heard
and 15 minutes later the final 10m ofthie structure disappeared below the water
surface and the structure was lost. Twenty two people were on or around the
structure at the time and fortunately all were evacuated safely. The cost of the
structure was approximately $1 80m, but the economic loss was about $700m.
Problems arose when one of the buoyancy cells imploded under load. In the
design of the structure extensive use was made of finite element analysis, but
for whatever reason, problems with the use of the software or problems
associated with interpretation of the results, a critical design condition was not
identified.

In all three cases the consequences for the Client and the designers were
serious, although potentially in two ofthe cases matters could have been far worse.
At the time all three cases received unwanted publicity and were an unpleasant
reminder to all engineers that computer programs and calculations need to be
checked. Had adequate checks on the software been undertaken, before they were
used for design, errors in the computer code, or problems with the use of it, could
have been detected and the failures avoided.

8.6

Developers and users

8.6.1 Developers
Traditionally, developers of software have tended to be either organisations or
academics who program specific analytical andlor numerical techniques. The
software they produce may be made available to users commercially by selling or
leasing it.
Some software developers expend enormous resources on checking their
programs and ensuring that they are robust. Limitations may even be placed on the
use of the program, to prevent misuse, by introducing default xalues for certain
control features. They may submit their software for independent assessment or
type approval (fitness for purpose), which is undertaken by certain organisations
(e.g. Lloyds Register). This is a check on the 'product' rather than the procedures
used for development. Other developers cany out relatively little testing beyond
fairly basic checks on the code and its operation. They tend to adopt the principle
of 'user beware'. Such effort, or lack of it, is usually reflected in the cost.
Obviously, cheaper programs are likely to be less robust than more expensive
programs, which are likely to have been subjected to more testing. A potential user
should take this into account when considering the purchase of any program.
Although the developer may be responsible for the code, and should carry out
sufficient checks to ensure that the program is working correctly and that the
relevant theory has been correctly applied, the developer cannot be responsible for
the way the program is used, or the interpretation of the user on the way it is used.

Benchmarking 1 337

Further, the developer cannot trace all the bugs in a program, which may consist
of many thousands of lines of code, and some of these bugs may only become
apparent if the program is used in a certain way. Therefore, the user has to be
aware that such problems may exist.
In the past the user was rarely given the opportunity to modify code. If changes
had to be made this was usually done by the developer. However, in recent times
the situation has become more complicated. Some computer programs that are sold
commercially allow the introduction of 'User Defined Subroutines'(UDS) to allow
special features to be added, or new constitutive models to be introduced into the
program. Under these circumstances the traditional user or purchaser assumes the
role of the developer. Not only must the person writing these features ensure that
the subroutine is working correctly, but the interface with the rest of the program
needs to be checked. The subroutine may function perfectly adequately by itself,
but it may cause errors in other parts of the program which were otherwise
functioning correctly. This requires a very extensive checking procedure,
particularly with the more complex codes. The UDS will have to include all the
error control routines, which further complicates the issue. There are obviously still
further complications if there is limited access to the source code. Essentially, if
changes are made the program has to be re-validated (consider the example of the
wind turbine that failed).

8.6.2 Users
Users should expect that developers have carried out sufficient checks to ensure
that the program is working correctly, but this cannot, and should not, be taken for
granted. A number of issues arise: does the user or engineer have a responsibility
to carry out sufficient validation of any computer applications used in design; if so,
what procedures have to be adopted; is the implication of this that if the engineer
does not adopt such procedures, then there has been a breach of professional
responsibility?
Some people are of the opinion that users should be given greater access to
computer source codes. This would allow users to better understand how the
program was operating and might, under certain circumstances, allow them to
customise it. This is obviously desirable when testing the program by
benchmarking or other means because, if problems occur, it might be possible to
identify the source of any errors.
There is, however, a number of drawbacks. In the case of finite element
programs, and other programs based on advanced numerical methods, there are
very few people capable of interpreting the code and even fewer with sufficient
knowledge to successfully customise it. Even after a relatively minor modification
a sufficiently rigorous testing programme needs to be initiated to ensure that no
unforeseen errors have been introduced, or interface problems have developed with
other sections of the code.
Does the fact that the user has access to the source code, and has had the
opportunity to inspect it, shift the onus of responsibility for checking the program

338 1 Finite element analysis in geotechnical engineering: Application

further away from the developer onto the user? Is there even more need for the user
to comprehensively test or benchmark the program? If problems did arise, even in
an unmodified code, would the developer be able to use the fact that the user could
have identified these problems if it had been correctly tested, or benchmarked, to
minimise or even evade any responsibility?
The professional responsibility of an engineer to adequately check computer
programs has long been recognised by the Institution of Civil Engineers. The
following question was for sometime included in the list of essay questions
forming part of their professional examination (Q.24, The place of the engineer in
the community and management topics; questions applicable to Section B of the
Essay for the 1981 April and October Professional Examination):
"Discuss the impact of increasing computer applications on the civil engineer's
design responsibilities and the procedures that must be adopted to ensure the
proper use of such applications."

In the early 1980's there were numerous articles in New Civil Engineer, the
Institution of Civil Engineers' weekly journal, discussing these issues. Much was
made of a comparative study undertaken by the Construction Industry Research
and Information Association (CIRIA). The exercise compared the results obtained
from seven different programs that were based on code of practice for the design
of reinforced concrete beams. No two programs gave the same answers.
In 1990 the UK's National Economic Development Office warned the users of
computer hardware and software that, under current UK legislation, they were very
unlikely to be able to sue suppliers or developers if faults were found in hardware
or software. Commenting on this statement, the Association of Consulting
Engineers reminded engineers that, irrespective of any legal redress, they had a
responsibility to ensure that calculations were accurate, whether or not they were
done using a computer.
Identifying that there was potentially a problem with the use of computer
programs, in 1994 the Association of Geoenvironmental and Geotechnical
specialists (AGS (1994)) published a guide for the validation and use of
geotechnical software. This document was not aimed specifically at users of
advanced numerical methods, but it did draw on the work of NAFEMS (National
Agency for Finite Element Methods and Standards) for setting out strategies for
testing software. When published, the guide stimulated a number of debates as to
how these issues should be dealt with. A number of suggestions were put forward,
including formation of an industry sponsored testing organisation for software.
However, to date this has not materialised.
In recent times many engineers work in an environment where quality
assurance (QA) procedures are applied. Although at times these procedures may
appear to be unnecessarily bureaucratic, they lay down management principles
which are measured against a standard, most commonly I S 0 9000. There is a
requirement to ensure that the 'tools' being used to produce the product (i.e.
design) are of an adequate quality and that checks are made to ensure that this

Benchmarking 1 339
quality is maintained. Obviously, the results of computer analyses could have a
major impact on a design, or a product, and therefore sufficient checks need to be
made on the program and the results obtained to ensure their adequacy.

8.7

Techniques used to check computer calculations

There are several techniques that may be used by the developer or user to check
programs. These include the following approaches:

- A check carried out on the code by reading it, to check if there are no
inadvertent errors.

- Sample calculations of specific algorithms.

- Checks against closed form solutions.
- Checks against other calculations made by either another program, another
-

technique or an earlier version of the program, if that version has been tested
and found to be correct.
Checks made by running the same problem with the same computer program
can different hardware, to ensure there are no interface problems between
different software packages and the hardware or operating system.

In most instances, because of the complexity of the code or because there is

limited access to it, the checker is not in a position to check the code line by line.
Therefore the checker must rely on checking the program against other
calculations. This technique is known as benchmarking.
The process of benchmarking can be used to establish not only whether or not
the program in question is working correctly, but also to establish how accurate the
program is. Under some circumstances specific techniques may need to be applied
to improve accuracy.

$.8

Benchmarking
General

8.8.1
There may be a number of reasons for wanting to undertake a benchmarking
exercise. These include:

- Testing the fundamental operation of the software. As many of the features of

the program as possible, both as individual components and in combination
with each other, should be tested. Therefore problems need to be formulated
that adequately check these features. Obviously, the more complex programs
require a greater range of test problems. Ideally checks against closed form
theoretical solutions should be made. However, the results of one analysis
might be compared with the results obtained from an analysis using another
software package, but there is potentially a problem. What if both codes are in
error? If both codes use the same solution strategy and this is not accurate, is
this a fair test (this will be illustrated by one of the examples of benchmark
problems which will be presented later)?

340 1 Finite element analysis in geotechnical engineering: Application

Checking that there are no hardware-software interface problems, or no

interface problems when the operating system is changed. Then, apart from
checking against closed form solutions, it may be acceptable to compare results
of the same version ofthe program run in a different environment. The Authors
have had some experience of this problem. When upgrading an operating
system from one version to another (this was not a change in the operating
system), a problem occurred with a program that was in daily use. Meshes that
could be generated with the program when using the earlier version of the
operating system could not be generated when the same program was used with
the newer operating system. It was fortunate that the error was so obvious since
it was quickly found and rectified, but it involved a series of rather tortuous and
time wasting discussions with the supplier of the operating system.
To familiarise a user with the software or an updated version of it, or to test the
competence of a user to use a program. The benchmarking process therefore
becomes a test of the user and not just a test of the software. To illustrate the
importance of such an approach two benchmarking exercises, the
INTERCLAY I1 project (EUR 15285 (1993)) and a project initiated by the
German Society for Geotechnics, are discussed in Sections 8.10 and 8.1 1 .

Depending on the structure and/or complexity of the program the

benchmarking exercise may be, of necessity, a time consuming and extensive task.
Standard benchmarks do exist but other tests, non-standard benchmarks, may need
to be developed, particularly for geotechnical software.

8.8.2 Standard benchmarks

NAFEMS is an organisation based in the U.K., which was formed by users and
software developers, the majority of whom are associated with the aeronautical
industry or process engineering. Over a number of years NAFEMS has published
a number of standard benchmarks for finite element programs. The background to
the development ofthese benchmarks is explained in the publication "The standard
NAFEMS benchmarks" (NAFEMS (1990)), from which the following extract is
taken:
"In 1986 NA FEMS attempted to produce the first set of independent 'standard'
tests which could be applied to anyfinite element system. The tests were intended
for the use of code developers, users and others who were interested in the
performance of their speclfic codes. They were designed to be a little more
searching than those publicised or quoted by the software developers, since these
were naturally intended to demonstrate the favourable performance of a system,
rather than the deficiencies".

These benchmarks are actually quite demanding in the way the problem is
formulated. The specification for the problem is quite detailed and very rigid. Not
only are the essential features, such as geometry, loading, boundary conditions and
material properties defined, but so are the element types and the dimensions of
each element.

Benchmarking 1 341

The 1990 NAFEMS publication provided standard benchmarks, covering

structural problems (plates and shells), heat transfer and vibration problems. More
recently some other topics have been covered, but as yet not geotechnical
engineering.
The INTERCLAY I1 project, sponsored by the European Commission, was a
coordinated benchmark exercise on the rheology of clays. The report on the project
(EUR 15285 (1 993)) provides four examples of problems which might be used for
benchmarking purposes. This exercise will be discussed in some detail
subsequently (refer also to Jefferies and Knowles (1994)).

8.8.3

Non-standard benchmarks
For the reasons NAFEMS explained, the geotechnical engineer cannot rely on
those problems produced by the developer to properly check the program.
However, these may be useful when first installing the program, as they allow any
obvious problems with the hardware and operating system to be identified at an
early stage. The user of a geotechnical software package therefore has a serious
problem, since standard geotechnical benchmarks do not exist.
Users therefore need to develop sufficient and exacting benchmarks to
adequately test the software. Problems need to be formulated which have a known
solution that can be compared with the results of the program being tested.

8.9

The INTERCLAY I! project

The INTERCLAY I1 project was sponsored by the European Commission within

the 'European Atomic Energy Community 4'h Research and Development
Programme'. This was a collaborative exercise involving 10 organisations,
although not all organisations participated in every aspect of the project (see EUR
15285 (1993) for full details). Two of the objectives of the project were to:
"assess the usability and capability of the Cfinite element) codes available within
Europe, with respect to the geomechanical behaviour of clay" and to "assess the
accuracy and reliability of the numerical predictions of geomechanical
behaviour".

A number of test problems were specified by a steering group which were

analysed by the participants and then individual's results were compared with each
other.
When considering the results of the INTERCLAY I1 exercise, it is important
to remember that ail the participants were very experienced analysts, some of
whom were using finite element codes that they had developed and which had been
in use for some years. Others used commercially available codes, although these
were very highly developed programs. If it were possible to categorise such
programs, those used for this exercise would most certainly fall towards the upper
end of the market in terms of their development and the effort put into testing
them. As such, they are costly to purchase, lease or acquire the use of.

342 1 Finite element analysis in geotechnical engineering: Application

There were three stages to the project. The first stage involved the solution of
hypothetical problems with a defined geometry (not element type or size as for the
NAFEMS standard benchmarks), loading conditions and material properties. Stage
2 involved the back-analysis of two laboratory tests to predict results for a third
experiment. The third stage involved the analysis of in-situ tests at the Mol
(Belgium) underground facility. In the context of benchmarking only the first stage
of the project is of interest.
The results of the analyses undertaken during Stage 1 are discussed in Section
8.10 of this chapter, in which the Stage 1 INTERCLAY I1 problems are used as
examples of benchmarking problems. However, when assessing the results of the
project as a whole Jefferies and Knowles (1994) commented as follows:
"The project highlights the need to recognise, more formally, the opportunity for
human error and to devise a means of controlling it. The experience fvom this
exercise is that in most cases 'errors' were not discovered until the results were
compared with the experimental results and those of other participants. Errors
were primarily due to misinterpretations of the benchmark specijkation as a result
ofeither incomplete definition of the requiredparameters or misinterpretation of
their meaning".

Despite these comments some participants did discover that there were
fundamental problems with their software, or that it was not ideally suited to the
analysis of some problems. The exercise proved to be a beneficial experience for
all concerned and highlighted the need for caution when using computer programs
of any form.

8.10 Examples of benchmark problems

8.10.1 General

- Part I

In this section examples of benchmarking problems are presented and discussed.

They do not form a complete suite oftest problems since they do not cover all the
boundary conditions and load cases that a general purpose geotechnical based finite
element or finite difference package might be expected to cover. The problems
described by Schweiger (1998) could be added to this list if more details of the
comparisons made could be obtained for reference.
In Chapter 9 of Volume 1 four simple problems were discussed to demonstrate
the accuracy of different solution techniques. These were an idealised triaxial test,
a footing problem, an excavation problem and a pile problem. Any one of these
problems would be ideal for benchmarking a program, but the first of these, the
idealised triaxial test, is perhaps better than the others, because there is an
analytical solution to the problem. This is the first of the 5 examples reviewed.
The other four examples of benchmarking problems that are presented are taken
from the INTERCLAY I1 project. Selected results are presented, purely to illustrate
certain features of each problem. A full set of results and specifications for the
problems can be obtained by reference to the INTERCLAY I1 report (EUR 15285

Benchmarking 1 343
(1993)), although specifications for these problems are provided in Appendices
VIII. l to VITI.5. The specifications are less detailed than the NAFEMS standard
benchmarks, because there are no restrictions on the mesh details or the solution
technique. However, they are no less demanding.
Example 1

Example 2

Example 3
Example 4
Example 5

Analyses of an idealised triaxial test: two analyses, one drained and

one undrained, of a triaxial sample, isotropically normally
consolidated and subjected to compression.
Analysis of a thick cylinder: a thick cylinder of clay subjected to a
differential pressure between the inner and outer surfaces. This
problem has a closed form soiution.
Analysis of an advancing tunnel heading: simulation of the
progressive excavation and construction of a cylindrical tunnel.
Analysis of a shallow waste disposal: analysis of construction of a
shallow repository and emplacement of waste and capping material.
A one dimensional problem based on Example 4.

8.10.2 Example l: Analyses of an idealised triaxial rest

Definition o f the problem
The geometry of the problem is not specified, but as there are no 'end effects' a
single element could be used. Constant increments ofaxial strain, up to 20%, under
undrained and drained conditions, are applied to represent compression of the
sample, see Appendix VIII. 1.

Results
Analytical solutions for ideal (no end
effects) drained and undrained triaxial
tests on modified Cam clay are given in
Chapter 9 of Volume 1. Figure 8.1 shows
the results for the drained triaxial test,
using the modified Newton-Ralphson .F
technique with a substepping stress point 4
algorithm for solving the nonlinear finite
element equations, while Figure 8.2 8
shows similar results for the same test
performed using the tangent stiffness
approach. In each figure results from 3
finite element analyses are given. These
differ in the size of the increment of axial
strain applied, i.e. OS%, 1.0% and 2.0%.

Comment
There is an analytical solution available
which can be used for comparison with

-Analytical solution
"
Axial strain (%)

Figure 8.1: Modified NewtonRa~hsonprediction of drained

triaxial test

344 1 Finite element analysis in geotechnical engineering: Application

the results obtained from any numerical

analysis, but if such a solution was not
available then there is potentially a
problem. What if an analyst was using a
code that used one of the two apparently
Analytical solution
less robust solution techniques (e.g.
tangent stiffness or visco-plasticity, see
Chapter 9 of Volume 1) and chose to
compare it with the results from a
different code using a similar approach?
It is apparent from the discussion in
Chapter 9 of Volume 1 that there is
potentially a problem with software that
uses certain approaches for the solution
of the nonlinear governing equations. If
an analysis was run using the tangent Figure 8.2: Tangent stiffness
stiffness approach, then, even if there prediction of drained triaxial test
were no errors in the code or other
problems with the use ofthe program, the results obtained might not be particularly
accurate. Simply comparing these results with a program using a similar approach
and similar loading increments would not be a sufficiently exacting test. Equally
inaccurate results could be obtained.
From this problem the user is able to gauge the accuracy ofthe software and the
measures necessary to obtain sufficiently accurate results.

8.10.3 Example 2: Analysis of a thick cylinder

Definition of the problem
Using a total stress analysis a pressure, ai,
was applied to the internal boundary of a
thick cylinder (see Figure 8.3), so as to
unload it. The analysis started with an
initial internal pressure within the
cylinder of 5000kPa. This pressure was ,
then reduced in stages to give a final
pressure of 1500kPa at the end of the
analysis, see Appendix VIII.2.
Results
Figure 8.4 shows the pressuredeformation behaviour and Figure 8.5
shows the position of the elastic-plastic
interface.

8.3: Geometry of a thick

cylinder

Benchmarking / 345

Cornrnent
Like Example 1, this is an ideal benchmark because there is an analytical solution
available which can be used to compare with the results.
The results obtained from the INTERCLAY I1 exercise all showed apparently
good agreement between all participants and the analytical solution. All of the
participants were aware of the analytical solution when undertaking their analyses
and could have made adjustments to the increments of loading to obtain
sufficiently accurate results. However, given that all of the participants were very
experienced in the use of their own software, it was reasonable to assume that they
were able to use these programs well enough to obtain accurate results. The same
may not be true if an inexperienced user or a user of some less robust software was
to analyse the problem.

0.01
0.00

0.05

0.10

0.15

0.20

0.25

Radial convergence (m)

Figure 8.4: Pressure-deforma tion

behaviour

0.0
0.0

2.0

4.0

6.0

8.0

Radius of elasto-plastic interface (m)

Figure 8.5: Position of elasto-plastic

interface vs. pressure

10.0

346 / Finite element analysis in geotechnical engineering: Application

8.10.4 Example 3: Analyses of an advancing tunnel heading

Definition of the problem
Figure 8.6 shows the geometry
of the problem which was axisymmetric. The analyses
modelled progressive tunnel
construction by advancing the
tunnel face and then inserting a
rigid lining to provide support,
see Appendix VIII.3.
Results
Figure 8.7 shows the range of
predicted radial displacements
towards the tunnel, using the
total stress approach, at three
offsets from the tunnel along
section A-A (refer to Figure
8.6), while Figure 8.8 shows
comparable data for the
effective stress approach (the
effective and total stress
parameters are not compatible
and so comparisons between the
magnitude of the displacements
from the two analyses should
not be made). Shown on Figure
8.9 are the displacements of
point B (refer to Figure 8.6)
versus increasing length of
excavation, for both total and
effective stress approaches.

t
Not to scale

Figure 8.6: Geometry for analysis of

advancing tunnel heading
O.lO1

TOU~
stress analysis

Range of values
;at 2.5m oRsct

10

l5

20

25

Length of excavation (m)

The
On this work
Figure 8.7: Radial convergence with
(EUR 15285 (1993)) has a
progression of excavation
series of plots which show
(total stress analysis)
stresses at three specified offsets
from the tunnel at different stages in the analyses. Figure 8.10 summarises some
of these results for the radial stress at the end of excavation.
Figure 8.1 11 shows the distribution of the radial total stress on the tunnel's
periphery (2.5m offset) along the first 12m of the tunnel (from the start of the
tunnel up to section A-A, refer to Figure &.6),when the tunnel has reached point
B. Both this and the previous figure relate to the total stress analysis.

Benchmarking 1 347
0 l0

Cornment
This is not an ideal benchmark.
Unlike the previous two examples,
there is no closed form solution 8 006
available. Unlike the standard
OM
NAFEMS problems, no
restrictions were placed on the
oo2
geometry of the finite element
ow
mesh.
0
5
10
IS
20
25
There was significantly more
Length of excavahon (m)
scatter in the results obtained from
the analysis of this problem by the
INTERCLAY I1 participants. The Figure 8.8: Radial convergence with
progression of excavation
plots of displacement (Figures 8.7
(effective stress analysis)
and Figure 8.8) show more scatter
close to the tunnel than remote
030
from it. For the total stress 3 ozo
approach there is a difference of
almost 20mm between
0.10
participant's results for a point on
,
the edge of the tunnel, 2 . 5 1 ~ g
offset. This is probably a
-O.'O
lbo results hfferen
consequence of an ill-conditioned 2
problem, because of the boundary 2
-0 30
condition applied to the sides of
0
S
10
IS
20
25
Lengtb ofexcavation (m)
the excavation to represent the
tunnel lining (zero radial
Figure 8.9: Displacement of B vs.
displacement), rather than the
length of excavation
accuracy of individual codes. It is
important therefore, when
formulating benchmarks, that care Q
a Mmmum range
is taken to ensure that the
A Maximum range ~gnonng
one analysis
boundary conditions are
reasonable and realistic.
.g l0 However, close to the tunnel
all the participants obtained some
A
4 5oscillations in radial stresses, as 2
illustrated by Figure 8.1 1. There
a
a
was considerable scatter, not only 3 (I
I
I
between participants, but some of
o
2s
5o
75
10
Radial distance fiom tunnel centre lrne (1.e rad~aloffset)
the predictions showed
considerable fluctuation along the
Figure 8.10: Range of radial stress
length of the tunnel for the 2.5m
variation at different offsets
offset (along the boundary of the

-5
1

2
3

.g

. /

348 / Finite element analysis in geotechnical engineering: Application

excavation). The variation of

I2.O
radial stress was much less along
the profiles offset 5.0m and 7.5m
from the centre line of the tunnel.
In discussions following
completion of this example it was 2
considered that the problem of 2
oscillation of stresses, and the
Radial ofiet 2.5~1
R ~ S U I ~from
S
~ I I radial
I
offset
differences in the predicted
-4.0
I
I
I
I
2
4
6
8
10
12
displacements, were caused by the
Distance kom tunnel mouth (m)
high stress gradients close to the
lining and a singularity at the
advancing face, although there Figure 8.11: Radial stress along tunnel
may have been some mesh
boundary
dependancy which influenced
results close to the edge of the tunnel (if results are influenced by the mesh, then
this problem is just as much a test of the user's ability to formulate the analysis, as
it is a test of the s o h a r e ) . However, as two participants demonstrated, much
smoother results could be obtained by looking at stresses along an offset 0.5m from
the edge of the tunnel, rather than along the edge of the tunnel itself, see Figure
8.1 1. For purposes of future comparisons these results are probably more useful.
On profiles further away from the tunnel agreement was much better.
Despite the apparent problems close to the tunnel, it was generally considered
that it had been a worthwhile exercise, since it posed a more demanding test than
the previous two problems described. It is a two dimensional problem unlike the
previous examples which were essentially one dimensional.

,,

8.10.5 Example 4: Analysis of a shallow waste disposal

Definition of the problem
This was a complex problem which was intended to model the construction of a
shallow nuclear waste disposal. The geometry is shown in Figures 8.12 and 8.13.
After setting the initial stress conditions the analyses modelled the excavation
for the disposal and construction of the cell walls and base slab. Pumping beneath
the base slab was then activated (i.e. zero pore water pressure under the base slab).
Filling with the waste material followed and then construction of the capping
layers. Long term pore water pressure conditions were then established and
collapse of the waste was modelled (see Appendix VIII.4).
There was some discussion about how the collapse of the waste should be
modelled. The method adopted may not have been particularly realistic, but for the
purposes of this exercise it was considered to be sufficient, since all participants
adopted the same procedure.

Benchmarking 1 349

Results
Again the INTERCLAY 11 report gives detailed results (EUR 15285 (1993)).
Figures 8.14 to 8.18 show vertical displacements along the underside of the base
slab, at each of the five stages in the analyses (see Appendix VIII.4). Figure 8.19
compares d,isplacementsof point B (refer to Figure 8.12), which is on the centre
line at the base of the sand capping layer.

25m

1-

25m

1 1-.

25m

(Not to scale)

Figure 8.12: Geometry of shallow waste disposal

Rockfill capping
material ,

material' '

material

Wall and base slab;

E = 560000 MPa
u=O.IS

B d'

0.3m

Figure 8.13: Detail of perimeter wall

and base slab

Cornment
This is the most demanding of the five examples and, like Example 3, there is no
analytical solution available and therefore there is no direct check on the results.
With this problem some of the deficiencies with some of the codes used by
INTERCLAY I1 participants became apparent. There were also some problems
with interpretation of the specification.
With the exception oftwo participants, similar results for the displacernents of
the underside of the base slab were obtained for the first stage of the analysis
(Figure 8.14). For stages 2,3 and 4 there was some scatter in the results, although

350 / Finite element analysis in geotechnical engineering: Application

4 or 5 of the participants obtained reasonable agreement with each other (Figures
8.15 to 8.17). The final stage of the analysis modelled the collapse of the waste.
There were obviously some problems with the original specification, since all but
one participant did not interpret it correctly.

2?:

0.30

0.30

0.20

'2

020

O.I0

O.W

c.

-Bg

3;

4.10

O'rO
0.w

4.10

f 4.20

"20

.-

4.30
25

75

50

Horizontaldistaoce from cmke Lint (m)

Figure 8.14:Stage l: Excavation

completed

Horizontal distance fram centre line (m)

Figure 8.16: Stage 3: Repository

filled and capped

>

4.30

25

50

Horizontal distance kom centre Line (m)

Figure 8.15: Stage 2: Pore

pressure dissipation

Horizontal distance from cenee line (m)

Figure 8.1 7: Stage 4: Pore

pressure dissipation

I
0

29
M
HorizontalLstancc from c c n a line (m)

Figure 8 . 1 8 : Stage 5: Waste

collapsed

75

75

Benchmarking 1 351

The analysis of this problem was

3
considered to be a fairly exacting test ;;; o
of any program. It involved the .;ij
g -0.1
analysis of a two dimensional %
problem with construction, t; -0.2
excavation, prescribed displacements
-o,3
and drainage.
3
Drainage was tackled in slightly 2 5 . 4
different ways (coupled and noncoupled analyses) and later work 9
identified this as a potential source of
-0.6
3
4
5
minor differences between codes.
Stage of analysis
Further differences were attributable
to the method of modelling the walls Figure 8.19: Vertical displacement
and base slab. Some participants
of point B on centre line
chose to model these components
with zero thickness elements (e.g. beam elements) to which they gave appropriate
properties, whereas others used solid elements.
Discussions following completion of this work did expose some problems with
some of the codes, which had not been identified before undertaking this exercise.
This reinforced the point made by NAFEMS concerning the nature of 'supplied'
problems for benchmarking. It also reinforced the need for very clear and exact
specifications to be drawn up before undertaking any analysis.
In the discussions between participants following completion of this exercise
it was apparent that there had been some subtle differences in the way some of the
participants had approached the problem. It was therefore resolved that a much
simpler analysis should be undertaken, that could be more clearly defined and
therefore less ambiguous. This is the fifth example of benchmarking problems that
is presented.

.g ,,,

8.1 0.6 Example 5: Simplified analysis of a shallow waste

disposal
Definition of the problem
Example 4 raised certain issues regarding the interpretation of the specification for
the problem and the boundary conditions applied. Therefore, to remove any
ambiguity regarding these difficulties, two further problems were analysed. Both
problems (5a and 5b) were intended to reproduce the main features of Example 4.
In case 5a the boundary conditions were explicitly defined (see Appendix
VIIIS), whereas in the secorid case (5b) participants were asked to use the
boundary conditions they had used for the main analysis (Example 4), if these
differed from those used for 5a.
Figure 8.20 shows the geometry used for the analyses. When the analyses
started, none of the soil above existing ground level was present. With the

352 / Finite element analysis in geotechnical engineering: Application

exception of construction of the cell Smof analysis
EndOf
3n1 R O C ~ ~ capping
~II
walls and the collapse of the waste,
1.smClay within capping
the sequence used for the analyses
followed the sequence used for
Example 4.
7m Waste material
The report (EUR 15285 (1993))
presents results for both cases
analysed. However, the second case 2
(5b) was intended merely to highlight
any inconsistences in the
interpretation of the specification for
Example 4. As such it cannot be
regarded as a benchmark, whereas the
.-+
first case, in which all the boundary
l element
conditions were explicitly defined, is
an excellent example of a benchmark. Figure 8.20: Simplified geometry for
Although there is not a theoretical
waste disposal analysis
solution to this particular problem, the
~.loy
results can be checked by a series of
I
B
fairly simple hand calculations.
005
1

,ezm,

Results
Figure 8.21 shows the displacement
of a point B at base slab level at
different stages in the analysis, for
case 5a. Similarly Figure 8.22 shows
the vertical total stress beneath the
base slab.

CI

OW

005

.a
3

.,,

1
p

g
015
3

Slage

Figure 8 . 2 1: Vertical displacement

Comment
at point B
Figure 8.21 shows that, with the
exception of one of the codes, there _.I,
was quite good agreement between $
those who chose to study this
loo
problem. The agreement might have
been even better, but at least one of f 2w
the participants was using fully 3
8 3w
coupled consolidation to model an .;d
undrained event. The time steps used
400
may not have been sufficiently short
,
and some drainage may have occurred
'
2
Stage
3
4
during this stage, resulting in some
settlement. However, the final Figure 8.22: Vertical total stress at
displacement obtained from this
point B

Benchmarking / 353

analysis was very close to that obtained from the other analyses. Again, with the
exception of one participant, the predicted total stresses beneath the base slab were
very close to each other (Figure 8.22).
Careful examination of the results of the analyses of this example (5a) and the
additional analysis (5b) of the same problem, using alternative methodologies,
allows conclusions to be drawn as to which analyses of Example 4 were the most
reliable (the INTERCLAY I1 report provides these comparisons, but the figures
presented in this chapter are not sufficient to do so).
This is another useful example of a benchmark problem, the specification is
very clear and the problem is simple, but it involves the use of a number of
different features of a program.

8.11

Examples of benchmark problems - Part I I

(German Society for Geotechnics benchmarking
exercise)

8. l 1 . l Background
Schweiger (1998) (see also Carter et al. (2000)) describes a benchmarking exercise
sponsored by a working group of the German Society for Geotechnics. The aim of
the exercise was to develop a series of problems which could serve the following
functions:

- Provide a check on commercial codes (and presumably on bespoke software).

- Provide a learning aid for young geotechnical engineers to become familiar

with numerical analysis.

Provide verification examples for proving competence in numerical analysis of
geotechnical problems.

The aims of this project were therefore broadly similar to the aims of the
INTERCLAY I1 project. This study aimed to use actual practical problems,
simplified in such a way that the solution could be obtained with reasonable
computation effort.
Two examples are given by Schweiger (1998). For the first example (Example
6) construction of a tunnel was modelled, and for the second example (Example
7) the formation of a deep excavation was modelled. Both are problems that are
commonly analysed by geotechnical engineers.
These problems were analysed by a number of participants and the results
compared. It should be noted that for both these problem no closed form solution
exists and so some reliance has to be placed on a consensus being obtained (this
was also true for the more complex INTERCLAY I1 problems).

$ . l 1.2 Example 6: Construction of a tunnel

Definition of the problem
The first of the two examples involved the construction of a non-circular tunnel,

354 / Finite element analysis in geotechnical engineering: Application

as shown in Figure 8.23. Two analyses
were requested for the exercise: a singlestep excavation and a two-step
excavation, both assuming an elasticperfectly plastic Mohr-Coulomb type
soil. Material parameters are given in
Appendix VIII.6, while a detailed
specification is provided by Schweiger
(1998).

;;?

Results
Figure 8.24 shows the predicted surface
settlements for one step tunnel
l
excavation, from 10 groups that
undertook the exercise, while Figure 8.25
-I
summarises the predicted bending
moments and forces in the tunnel lining. Figure 8-23:Geometry for tunnel

{J

excavation

Comment
Although half of the predictions gave
,
maximum surface settlements between
52mm to 55mm, overall there were B "
I
significant differences in the predicted 1
settlements. Two groups used a different
,
method of tunnel construction to that m
specified, because of restrictions of their
codes. It was difficult to match the
S
10
l5
20
different approaches. This probably
D~stancefmm tunnel axis (m)
accounts for the differences in the results. Figure 8-24; Surface settlement

profiles

3-"

5000
4000
3000

C1

2000

l000
0

Participants

108 to 200

(IcNmlm)

Figure 8.25: Maximum normal forces and bending

moments in lining

Benchmarking / 355

There were also discrepancies in the predicted bending moments and forces in
the tunnel lining and the positions of the maximum bending moment, as shown on
Figure 8.25. Even if the results of the groups that did not follow the specification
are discounted, there are still significant differences in the predicted magnitudes
and positions of maximum bending moment.

8.1 1.3 Example 7: Deep excavation

Definition of the problem
Figure 8.26 shows the geometry of the problem. The analysis was intended to
model a staged excavation 12m deep and 30m wide, assuming a Mohr-Coulomb
soil. At two intermediate stages struts were installed, but there was no strut at the
top of the wall. Therefore, before the first prop was installed 3m below ground
level the wall acted as a cantilever. Material properties and construction stages are
listed in Appendix VIII.7. Schweiger (1998) provides further details.

!
!
!

Layer 3

Figure 8.26: Geometry for deep excavation

Results
Figure 8.27 shows the predicted 8 0.03
vertical surface displacements behind 2
the wall for the first stage of 3 0.02
X
construction. Horizontal displacements 8
0.01
at the top of the wall at the same stage
in construction are shown on Figure 3
0.00
8.28. Figure 8.29 shows the vertical
displacements behind the wall upon g
completion of the excavation g 6.01,
(construction stage 3).

10

IS

20

Distance from W ~ I (m)

I

Figure 8.2 7: Surface settlement

behind the wall - stage I

25

356 1 Finite element analysis in geotechnical engineering: Application

Comment
A wide scatter of the results is
apparent, even at the first stage in
construction when the wall acted as a
cantilever (see Figures 8.27 and 8.28).
Some participants used beam
elements to model the diaphragm
wall, others used solid elements, and
Panicipants
some groups used interface elements
Figure 8.28: Horizontal
between the ground and the wall.
These differences in modelling displacement of the top of the wall
- stage 7
technique might explain some of the
variations in the results, but they do
not account for all of them. As far as
horizontal wall displacements are
concerned at this stage, only half the
participants show positive
displacements of the top of the
cantilever wall, i.e. movements
towards the excavation. Curiously,
the remaining groups showed
movements away from the excavation % -0.03
5I
10
l
15
I
20
I
25l
(negative) at this stage. At the final
Distance from wall (m)
stage in the construction process there
is significant variation in the results, Figure 8.29:Surface settlement
with the maximum surface
behind the waN - stage 3
settlements predicted being between
l Omm and 50mm.

8.1 1.4 General comments

Although, quite rightly, Schweiger questions the suitability of the linear elastic
perfectly plastic constitutive models to analyse these problems, all the calculations
have been done on the same basis and should be reasonably consistent. There were
some differences in the approaches which may go someway to explaining the
discrepancies in the results. For the tunnelling problem there were differences in
the method of tunnel construction and for the deep excavation different methods
of modelling the diaphragm wall were used. However, differences such as the
horizontal movement of the cantilever wall are hard to explain.
It is worth mentioning that 5 of the 12 participants used the same computer
program. In this case some consistency between users of the same program would
be expected, even if the results obtained were unrealistic. The comments made by
Jefferies and Knowles (1994), referred to in Section 8.9, may be relevant. As
Schweiger recognised, interpretation of the specification may have been a factor

Benchmarking 1 357

that led to some of the differences that were observed. It is however surprising that
experienced analysts would have submitted results with obvious errors, such as a
cantilever wall moving backwards into the ground, without comment.

8.12

Summary

1. This chapter has reviewed the role of benchmarks in the validation and
verification of computer software and discussed the responsibilities of the
developers and users of a code in this respect. Before a computer program is
used, it should be checked by both the developer and user to ensure that it is
accurate, it is working as intended and it is being used correctly. Any checks
made by a user should be independent of the developer, it is not sufficient
simply to use supplied problems.
2. Despite the developer having tested the program before releasing it for use, it
is unlikely that all the potential 'bugs' will have been identified. Even if this
programme of testing is rigorous, it is difficult to anticipate some of the ways
the program may be used and there may be interface problems with hardware
or operating systems. In commercial terms, the degree of testing by the
developer is likely to be reflected in the cost of the program, Users must be
aware that potentially problems may occur and must take sufficient measures
to ensure that these problems do not affect any analysis that may be undertaken.
The program has to be benchmarked.
3. Ideally a program should be checked against a closed form solution, or an
independent means of calculation, to validate it. A series of simple hand
calculations checking stages of an analysis may be sufficient.
4. Benchmark problems of this form tend to be fairly simple and may not be
sufficiently demanding to expose any potential problems. It may therefore be
necessary to devise more demanding problems which can be checked against
the results of the same analysis being run using at least one other program. For
most users this is not a very satisfactory or practical situation, since they may
not have access to more than one code. There is therefore a need to devise and
publish 'standard' benchmarks for geotechnical problems.
5. The consequences of errors or 'bugs' in computer programs can be serious and
can result in significant economic loss.
6 . A series of examples of benchmark problems have been reviewed. The
INTERCLAY I1 exercise has not only provided a number of valuable examples
of benchmark problems that could be used to validate geotechnical software,
but it also demonstrated the need for programs to be thoroughly tested. During
the course of the exercise potential problems with some of the codes were
exposed. The exercise also highlighted the need for detailed and unambiguous
specifications to be produced before any analysis is run.

358 / Finite element analysis in geotechnical engineering: Application

Appendix VIII.l:

Specification for Example 1 : Analyses of

an idealised triaxial test

VIII. 1 . l
Geometry
Not specific, single element may be used.
V111.1.2
Material properties and initial stress conditions
Modified Cam clay model for triaxial sample.
Table VIII. 1:Modified Cam clay parameters

Initial stress conditions: Vertical effective stress, q,':

Horizontal effective stress, c+,':
Pore water pressure, p/:

200 kPa
200 kPa
0 kPa

V111.1.3
Loading conditions
Unspecified, but constant, increments of compressive vertical strain up to 20%
under undrained and drained conditions. No change in horizontal total stress.

Appendix V111.2:

Specification for Example 2: Analysis of

a thick cylinder

V111.2.1
Geometry
The geometry is shown in Figure 8.3.
V111.2.2
Material properties
Total stress analysis using a Tresca soil model.
Table V///.2: Tresca parameters

Initial stress conditions: Vertical effective stress, a,':

Horizontal effective stress, a,,':

5000 kPa
5000 kPa

Benchmarking 1 359

VR11.2.3

Loading conditions

Using a total stress analysis a pressure was applied to the internal boundary so as
to unload it. The maximum pressure applied was 3500kPa, giving a residual
pressure of 1500kPa. Incremental load steps were not specified.

Appendix V111.3:
V111.3.1

Specification for Example 3: Analysis of

an advancing tunnel heading

Geometry

The geometry is shown on Figure 8.6 and the analysis was run as an axi-symmetric
problem.

V111.3.2
A)

Material properties

Total stress analysis using a Tresca soil model.

Table L/111,3: Tresca parameters

B) Effective stress analysis using a Mohr-Coulomb failure criteria with a fully

associated flow rule.
Table VllI.4: Mohr-Coulomb parameters

Note:

A porosity of 0.4 and permeability of 4 . 0 ~ 1 0 -m/s

' ~ were required for
some codes.

Initial stress conditions: Vertical effective stress, U,,':

Horizontal effective stress, U,':
Pore water pressure, pi

V111.3.3

2500 kPa
2500 kPa
2500 kPa

Loading conditions

The total stress analysis used stages 1 to 3 only, the effective stress analysis used
all stages:
1. Advance tunnel face instantaneously by 2m, assuming that the internal pressure
on the excavated face is zero.
2. Insert rigid lining to provide radial support.

360 / Finite element analysis in geotechnical engineering: Application

3. Repeat stages i and 2 until tunnel heading has reached its full length (i.e. 24m).
4. Allow excess pore water pressures to dissipate with the tunnel lining and face
fully permeable and rigid (around the outer boundary pore water pressure
~ ~ 2 MPa,
. 5 around the boundary of the tunnel p,=O and no longitudinal
displacement of the face).

Appendix V111.4

Specification for Example 4: Analysis of

a shallow waste disposal

V111.4.1
Geometry
The geometry is shown in Figures 8.12 and 8.13.
V111.4.2
Material properties
The Mohr-Coulomb parameters for the different materials involved in this analysis
are shown in Table VIII.5.

Benchmarking / 361

For the waste material a constant value of Poissons ratio of 0.2 was specified
for stages 1 to 4 in the analysis, but in stage 5 this was set to zero. All the materials
were drained, apart from the clay. Initially pore water pressures below ground
water level (GWL) were hydrostatic. Above GWL pore water pressures were zero.

V111.4.3
Loading conditions
Stages of analysis:
l.* Construct perimeter wall (wished in place), excavate repository, place

drainage layer (zero thickness), construct repository base slab and construct
partition wall.
2. Allow pore water pressures to dissipate to a steady state condition assuming
that pore water pressures in the drainage layer under the base slab are zero
and that the Upper and Lower sands and gravels are fully recharged.
3.* Fill with waste and place capping layers.
4. Allow excess pore water pressures to dissipate to the same conditions as 2.
5.* Reduce Poisson's ratio of waste to zero. Apply a vertical stress of 125MPa to
upper and lower boundaries of waste to model collapse.
(* No drainage)

Appendix V111.5:

Specification for Example 5: Simplified

analysis of a shallow waste disposal

V111.5.1
Geometry
The geometry is shown on Figure
8.20.

Stress (kPa)

50

100

150

200

250

I effective stress

V111.5.2
Material properties
The material properties and initial
stresses are the same as for Example 4.
The initial stresses are as shown on
Figure VIII. l .

Horizontal and vertical

V111.5.3
Loading conditions
Stages of analysis:
1.* Excavate to 7.8m below existing

Figure V///,1: /nitja/ stresses

ground level and construct 0.3m
thick concrete slab.
2. Allow pore water pressures to dissipate to the steady state.
3." Place the waste material to 0.5m below original ground level and then place
4.0m of sand capping, 1.5m of clay capping and 3m of rockfill capping.

362 1 Finite element analysis in geotechnical engineering: Application

4.

Allow excess pore water pressures to dissipate as in Stage 2.

(* No drainage in clays)

V111.5.4
Additional boundary conditions
1. All vertical boundaries are to have zero horizontal displacement.
2. The base of the mesh (the interface between the sandstone and the Lower sand
and gravel layer) is to have zero vertical and horizontal displacement.
3. All vertical boundaries to have no drainage.
4. The base of the concrete slab is to be undrained in Stages l & 3 and drained
in Stages 2 & 4.
5 . The base of the mesh is to be drained.

Appendix Q111.6:

Specification for Example 6:

Construction of a tunnel

V111.6.1
Geometry
The geometry of the problem is specified in Figure 8.23.
V111.6.2
Material properties
The Properties for the soil and shotcrete are listed in Table VIII.6.
Table Vlll.6: Material parameters for tunnel construction

* Shotcrete thickness is

Appendix Q111.7:

25cm

Specification of Example 7:
Deep excavation

V111.7.1
Geometry
The geometry of the problem is given in Figure 8.26.
V111.7.2
Material properties
The properties of materials used in this example are given in Table VIII.7.

Benchmarking 1 363

Table V///.7: Material parameters for deep excavation analysis

'Diaphragm wall thickness is 80cm

VE11.7.3
Construction stages
The following construction stages are specified:
1.
2.
3.

Excavation step 1 (to level -4.0m).

Excavation step 2 (to level -8.0m), with strut 1 (at -3.0m) active.
Full excavation (to level - 12.0m), struts 1 and 2 (at -3.0m and -7.0m
respectively) active.

Restrictions and pitfalls

9.1.

Synopsis

This chapter considers some of the common pitfalls that can arise when performing
finite element analysis of geotechnical problems. Some of the restrictions that may
occur when using this method are also discussed. Most of these arise when trying
to model what is clearly a three dimensional component of a problem in a two
dimensional plane strain analysis. For example, modelling ground anchors to tie
back a retaining wall. Problems associated with the use of the Mohr-Coulomb
model for undrained analysis are also discussed. In particular, it is shown that it is
possible to obtain a failure condition only if the angle of dilation is set to zero. As
noted in Chapter 7 of Volume 1 of this book, the shape of the plastic potential in
the deviatoric plane controls the stress conditions at failure in plane strain analyses.
The implications of this are examined by considering boundary value problems
involving modified Cam clay. Some of the pitfalls in using critical state
constitutive models for undrained analyses, as well as problems associated with
modelling underdrainage and the use of zero thickness interface elements are also
discussed.

9.2.

introduction

From the preceding chapters, both in Volume 1 and this volume of the book, it is
clear that the finite element method, when applied to geotechnical problems, can
be extremely complex. While in principal the method can be used to provide a
solution to most of the problems that we may wish to analyse, there are
approximations which can lead to errors. These approximations can be classified
into two groups. Firstly, there are approximations in the finite element method
itself and secondly, there are approximations arising from the idealisations made
by the user when reducing the real problem to a form which can be analysed.
The two main sources of error involved in the finite element method itself are
associated with the integration of the constitutive equation and the discretisation
of the problem geometry into finite elements. The errors associated with the
integration of the constitutive equations have already been considered in Chapter
9 of Volume 1 of this book, where it has been shown that the errors can be
minimised to an acceptable level by the use of an appropriate solution strategy.
However, the errors due to discretisation have not yet been considered and

Restrictions and pitfalls / 365

consequently, in the first section of this chapter, the likely magnitude of these
errors is investigated by using an example of the surface footing problem. Some
of the problems associated with the use of zero thickness interface elements are
also discussed.
The rest of the chapter then concentrates on errors that can arise due to the
manner in which the user idealises the real problem. First of all, some of the
approximations that may arise in plane strain analyses are considered. Although the
dominant feature of a geotechnical problem (e.g. a long excavation) may render
itself suitable to a plane strain idealisation, some of the smaller components of the
problem (e.g. ground anchors, piles, etc.) do not. Consequently, some restrictions
are imposed on any plane strain analysis ofthe problem. Some ofthese restrictions
are discussed and comments are made on how they may be accounted for in an
approximate manner.
Another potential source of error can be associated with a users lack of 'in
depth' understanding of the constitutive model employed to represent soil
behaviour. This is a common source of error, due to the complexities of many of
the constitutive models currently available. Two examples are considered in the
chapter. The first example involves the use of the Mohr-Coulomb model in
undrained analysis. It is shown that if the model is asigned any plastic dilation, no
failure can be predicted. The second example considers the effect of the shape of
the plastic potential on soil strength. This is demonstrated by using the modified
Cam clay model to show that what seems like sensible input parameters can result
in unrealistic predictions. Our experience indicates that this is a very common
pitfall that many users unknowingly fall into, with the result that their analyses
predict erroneous collapse loads which are usually unconservative.
As the undrained strength is not one ofthe input parameters to the conventional
critical state constitutive models, use of such models to analyse undrained
problems can be problematic. This is particularly so if the site investigation has
been designed towards obtaining undrained strength values. This problem is
considered in some detail in this chapter, where it is shown that realistic undrained
strength profiles can be obtained by a suitable selection of the model input
parameters and initial stress conditions. However, such conditions can only be
achieved if the finite element program has flexible input facilities. The chapter
concludes by discussing some of the problems associated with the simulation of
construction and underdrainage.

9.3

Discretisation errors

Errors arising from the discretisation of the problem geometry into finite elements
is demonstrated by considering the behaviour of smooth rigid strip and circular
surface foundations under vertical loading, see Figure 9.1. The soil is assumed to
be linear elastic perfectly plastic, with a Tresca yield surface (see Chapter 7,
Volume l ) , having a Young's modulus E=10000 kPa, Poisson's ratio p=0.45 and
undrained strength S,,=100 kPa.

366 / Finite element analysis in geotechnical engineering: Application

Two alternative finite element

meshes are shown in Figures 9.2a and
9.2b. The mesh shown in Figure 9.2a
has 110 eight noded isoparametric
elements, whereas the mesh shown in
Strip
Circle
Figure 9.2b has only 35 elements.
Both meshes have been used to
Figure 9.1: Strip and circular
perform plane strain (for strip footing) surface foundations under vertical
and axi-symmetric (for circular
loading
footing) analyses, in which the
foundation is displaced vertically downward to obtain the complete loaddisplacement curve. From a cursory glance at the two meshes, most users would
probably conclude that the 110 element mesh was likely to produce the more
accurate predictions due its larger number of elements and their more even spatial
distribution.

Figure 9.2a:Smooth footing l l0 element mesh

Figure 9.26: Smooth footing 35 element mesh

Although there is no analytical solution for the complete load displacement

curve for either the strip or circular surface footings, there is an analytical solution
for the ultimate load, Q, for the strip footing which gives Q,,, = 2BNcS,,per metre
length, where N, = (2+a) and 2B is the width of the footing. Often text books quote
an equivalent analytical solution for the circular footing of Q = N, S,, ?r R', where
N, = 5.69 and R is the radius of the footing. However, this solution is not strictly
an analytical solution, as it involves the numerical integration of some stress field
equations. It is however thought to be reasonably accurate.
Results from finite element analysis using the meshes shown in Figures 9.2a
and 9.2b are presented in Figure 9.3. Also marked on this figure are the theoretical
limit loads discussed above. Assuming that the theoretical limit loads are correct,
then the analyses with the 110 element mesh overestimate the collapse loads by

Restrictions and pitfalls / 367

3.8% for the strip footing (i.e.
7
Nc=5.33) and 8.8% for the circular
6
footing (i.e. Nc=6.19) . However, the g 5
errors are much smaller, 0.5% for the
4
l 10 element mesh
strip (Nc=5.17) and 2% for the
circular (Nc=5.8) footing, for the
1
analyses performed using the 35
0
element mesh. The reason for the
o 0.1 0.2 0.3 0.4 0.5 0.6
&/B(%)
relatively bad performance of the 1 10
mesh can be seen from
Figure 9.3: Load-displacement
Figure 9.4, which shows vectors of
curves for smooth footing
incremental nodal displacements for
the last increment of the 110 element
strip footing analysis. Each vector
indicatesthe magnitude (by its length)
and the direction (by its orientation)
of the associated incremental nodal
displacement. While the absolute
magnitudes of the incremental
displacements are not important, the
relative magnitude of each of the
vectors and their orientation clearly
indicate the mechanism of failure.
Figure 9.4: Vectors of incremental
Attention should, however, be
displacements at failure f l l0
focussed on the corner of the footing,
element mesh)
where a rapid change in the direction
of the incremental vectors is
indicated. The direction changes by
approximately 120' between the
vector immediately below the corner
of the footing and the adjacent node
on the soil surface. Such behaviour
clearly indicates that under the corner
of the footing there are large stress
and strain gradients. In the 110
element mesh the elements in the
vicinity of this corner are too large
and of insufficient number t o
a c c u r a t e l y r e p r o d u c e ,this Figure 9.5: Detail from 35 element
concentration of behaviour. In
mesh
comparison, the 35 element mesh,
which has smaller elements and a graded mesh under the corner of the footing, see
Figure 9.5, can accommodate this behaviour more easily and therefore produces
more accurate solutions.
0

368 / Finite element analysis in geotechnical engineering: Application

Clearly it is important that smaller and therefore more elements are placed
where there are rapid changes in stresses and strains and therefore directions of
movement. Some computer programs will automatically re-generate the finite
element mesh to improve the accuracy of the solution. They work by first
performing a solution with a relatively even mesh. Based on the results of this
analysis a new mesh is generated with more elements in the zones of greatest stress
and strain changes. The analysis is then repeated. In principal, this process of
refinement could be repeated until the results were unaffected by further mesh
refinement.

9.4

Numerical stability of zero thickness interface

elements

9.4.1 I n t r o d u c t i o n
As noted in Section 3.6 of Volume 1, there are several different methods available
to model the behaviour at soil-structure interfaces. Of these alternatives, the use of
zero thickness elements is probably the most popular and consequently details of
their finite element implementation were given in Volume 1. Examples of their use
in finite element analyses of boundary value problems have been discussed in
earlier chapters of this volume.
While these elements are extremely useful, they have limitations. For example,
numerical problems, such as ill-conditioning, poor convergence of the solution and
unstable integration point stresses, have been experienced by several authors in the
past (Wilson (1977), Desai et al. (1984), Bay and Potts (1994)). Some of these
pitfalls/restrictions will be discussed in this section.

9.4.2 Basic t h e o r y
Although a detailed description of the theory behind zero thickness interface
elements is given in Section 3.6.2 of Volume 1, it is instructive to restate here the
main equations for 2D analyses.
The interface stress consists of a normal and a shear component. The normal
stress, D, and the shear stress ,t,are related by the constitutive model to the normal
and tangential element 'strains', E and y:

The interface element 'strain' is defined as the relative displacement of the top and
bottom of the interface element:

Restrictions and pitfalls 1 369

where:

ul

= vsina+ucosa

vI = vcosa-usina
and u and v are the global displacements in the X, and y, directions respectively.
Hence:

It is important to note that the element 'strains' are not dimensionless, but have the
same dimensions as the displacements (i.e. length).
The elastic constitutive matrix [D] takes the form:

where K, and K , are the elastic shear stiffness and normal stiffness respectively.
to strains (in
Noting that these stiffnessesrelate stresses (in units of for~e/(length)~)
units of length) via Equation (9.1), implies that they must have units of
force/(length)'). They therefore have different units to the Young's modulus, E, of
the soil andfor structure adjacent to them (i.e. E has the same units as stress for~e/(length)~).
As it is difficult to undertake laboratory tests to determine K,, and
K,, selecting appropriate values for an analysis is therefore difficult.
If a pore fluid exists in the interface, undrained behaviour can be modelled by
in
including the effective bulk stiffness of the pore fluid (again in f~rce/(length)~)
the stiffness matrix, in the same manner as described for solid elements in Section
3.4 of Volumel. It is also possible to allow the interface elements to consolidate
by implementing one dimensional consolidation along their length.
The interface stresses can also be limited by imposing a failure criterion. This
is conveniently achieved by using an elasto-plastic constitutive model. For
example, the Mohr-Coulomb failure criterion can be used to define the yield
surface, F:

F = I zl+oltan p' - cf

(9.7)

along with the gradient of the plastic potential function, P:

where y' is the maximum angle of shearing resistance, c' is the cohesion (see
Figure 9.6) and v is the dilation angle. If the interface moves such that the
maximum normal tensile strength is exceeded (c'/tanyf), the interface is allowed
to subsequently open and close and in a finite element analysis the residual tensile
stress is redistributed via the nonlinear solution algorithm. When the interface is
open, the normal stress remains equal to c'ltanq' and the shear stress remains equal

370 / Finite elernent analysis in geotechnical engineering: Application

"

to zero. The amount of opening of the

interface is recorded. When the interface recloses and reforms contact, the constitutive
model again defines the interface behaviour.
The facility for opening and closing of
the interface element is particularly useful
when analysing problems where tensile
cracks may form, for example behind a
laterally loaded pile (see Section 7.3.3) or
Figure 9.6: Mohr-Coulomb
behind a retaining wall.
yield function
The formulation for 3D interface
elements is similar to that described above, except that there are three interface
stresses (a normal stress, a, and two mutually perpendicular shear stresses, X,, and
r,) and strains (E, y, and y,), and three displacements (U,,v, and W,).

9.4.3 Ill-conditioning
Ill-conditioning of the global stifhess matrix can occur if the element stiffness
matrices of adjacent elements vary in magnitude by a significant amount. This
results in large off-diagonal terms in the global stiffness matrix and, consequently,
loss of accuracy. The analysis of the overturning of an elastic block will be used
to illustrate problems associated with ill-conditioning (Wilson (1997), Day and
Potts (1994)). Some of the results of Day and Potts (1994) will be presented here.
This problem is crudely representative of
a gravity wall. It consists of an elastic block,
Young's modulus E=106 kPa and Poisson's
ratio p=O, separated from a rigid foundation
along the base AB by interface elements, see
Figure 9.7. For clarity, the interface elements
are shown artificially expanded in Figure 9.7.
Ten 6 noded interface elements and fifty 8
noded 2D solid elements were used to model
the problem. A downwards vertical stress of
200kPa and a shear stress of 5OkPa in the
positive x direction were applied to the top
erface
men&
boundary CD. A large number of analyses
were undertaken and in these the interface
elements were assumed to be elastic, with
stiffnesses varying from K, and K,, =106 to K,
and K,, =lOiOkN/m3.Note that in general K,
Figure 9.7: Overturning of
was not equal to K,,.
elastic block
The integration point stresses in the
interface elements for the analysis with
K,=K,,=107 kN/m3 are plotted in Figure 9.8. The linear normal and shear stress

Restrictions and pitfalls 1 371

distributions shown in this figure are ;a A
B
0
the theoretical distributions for a rigid g
block on elastic springs. Since the 2
vertical faces of the block are stress
8
free, the shear stress within the base g
m 80
of the block must reduce to zero at the
400
corners (A and B), however the shear ,
stress applied by the interfaceelement
to the base does not. To satisfy g 0
continuum theory a singularity k!
therefore occurs in the shear stress
and the vertical stress distributions at g
the corners within the block. The
-800
theoretical shear stress distribution is
parabolic (Figure 9.8) for a long
Figure 9.8' Integration point
flexible block on a rigid foundation
stresses
in interface elements
(Young 989)). The shear stress
distribution in the interface elements lies between these two theoretical extremes.
Results from analyses with 2 (reduced) and 3 (full) point Gaussian integrationgave
similar results. These results are well conditioned and do not suffer from the effects
of ill-conditioning of the stiffness matrix.
Ill-conditioning of the stiffness matrix, which causes fluctuating stresses in the
interface elements, occurred when the interface element stiffhess was large with
respect to the stiffness of the 2D
600
elements. For example, the results of
the analysis with Ks=K,,=109kN/m3,
using 3 point Gaussian integration, 3
Linear distribution
are shown in Figure 9.9. Fluctuations 8
0
in the normal stress at the edges of the I
block and of the shear stress across
the whole base of the block are '$
clearly evident. The use of reduced 2
integration (i.e. 2 point) does not
2D elements
prevent the fluctuation in interface -1000
U
stress which is observed in Figure 9.9.
The results of the study by Day and
0
m
Potts (1994), which included b:
variations of both the stiffness of the B
elastic block and of the interface a 60
elements, showed that the effect of illconditioning was noticeable if either
K, or K,, was greater than 100E. Figure 9.9: Effect of ill-conditioning
However, it should be noted that, on interface element stresses (full
because of the difference in the
integration)

-5
2

372 / Finite element analysis in geotechnical engineering: Application

dimensions of K, and K,, with those of E, this simple relationship depends on the
units used. It is also likely to be problem dependent.
An analysis with a thin layer of 2D elements in place of the interface elements
was also performed by Day and Potts (1994). The elastic properties of the thin
layer are E, and p, and the thickness is t. The equivalent normal and shear
stiffnesses of the thin layer are:

It is noted that when K,,,=K, then (1 -.~,2)=2(1+p,).The solution of this equation

gives p,=- l, in which case the stifnesses, K,, and K,, are infinite. To avoid this
paradox and to be consistent with the overlying elements, the Poisson's ratio was
chosen to be zero (,u,=O). With the thickness chosen to be 0.02m and E,=2xlQ7kPa,
the normal stiffness of the thin elements is equivalent to that of the interface
elements (i.e. K,,,=109 kN/m3) and the shear stiffness is half that of the interface
elements (i.e. K1,=5*108 kN/m3). The analyses was also performed with full
(3~3point)Gaussian integration and the integration point stresses along the centre
line of the thin 2D elements is also shown in Figure 9.9. These show fluctuations
similar to those from the analysis using interface elements. This demonstrates that
the fluctuations are not due to the formulation of the interface elements, but due to
a difference in stiffness between the elements in the elastic block and those
forming the interface.
Reducing the size of the 2D elements in the elastic block adjacent to the
interface elements reduced the effect
600
of ill-conditioning, even though the
total number of elements remained
the same. This is illustrated in Figure
9.10, where the analysis presented in g
Figure 9.9 has been repeated with a
different finite element mesh, see
inserts in Figures 9.9 and 9.10. In this
case the contribution to the global
stiffness matrix from the 2D elements -looo
is increased in the locations
concerning the interface element
degrees of freedom. This arises as the
stiffness contribution of the 2D
---- - -elements, given by (Equation (2.28)
m.
m. m m
.
of Volume 1):
m

a
E

- - _ _ _ A

[K,]=

i i ~ [ B ] ~ [ D ] [ B ] ~ J ~ ~Figure
s ~ T 9. 70: Reduced ill-

-I -I

(9.10)

conditioning with smaller elements

(full integration)

Restrictions and pitfalls / 373

depends on the [D] and [B] matrices and the determinant of the Jacobian matrix
JI (note: t r l for a plane strain analysis). As the constitutive matrix [D] is elastic,
it is independent of the size of the element. However, the strain matrix [ B ] is
proportional to the inverse of JI (see Equations (2.11) and (2.15) of Volume 1).
Consequently, the stiffness contribution given by Equation (9.10) is also
proportional to the inverse of IJI and, as IJI reduces as the element size reduces
(see Equation (2.16) of Volumel), the contribution to the global stiffness increases.
The contributions from the interface elements remain unchanged. Thus there is less
difference between the stiffness of the interface elements and the stiffness of the
2D elements and hence less ill-conditioning. This example clearly demonstratesthe
need to carefully proportion the size of the elements within the finite element
mesh.

9.4.4 Steep stress gradients

Another problem where stress
fluctuations can occur is in the simple
pull out test. In view of the potential c
applications to soil reinforcement and S=
2
pile analysis, this example
1.0 m
demonstrates the interface element in
4
+
a problem controlled by the sliding
mode of deformation. The problem Figure 9. 1 1: Finite element mesh
for pull-out test
consists of a thin strip of2D elements,
representing a reinforcing membrane,
on a rigid base, see Figure 9.11. The contact between the membrane and the base
is modelled by interface elements. Five 8 noded 2D elements and five 6 noded
interface elements were used to model the test. Initially a uniformly distributed
load of IOkPa was applied to the top of the membrane. End A of the membrane
was then moved horizontally by the amount 6, in the direction shown, under
displacement control. This problem involves a sliding front that moves
progressively from the point of load application towards the other end of the
membrane.
To demonstrate this behaviour, an example where stress fluctuations did not
occur will be considered initially. The membrane was assumed elastic, with
Young's modulus E=106 kPa and Poisson's ratio p=O, while the elasto-plastic
Mohr-Coulomb model was used to describe the behaviour of the interface, with
properties K,=K,,=1Q6 kN/m3, qi=300, ci=O and v=OO. The membrane was
withdrawn in displacement increments of 0.3mm. Figure 9.12 shows the shear
stress distribution along the interface and the distribution of axial force in the
membrane at different stages of the test. It can be seen that a sliding front moves
from the left to the right as the membrane is pulled out. The maximum calculated
force needed to withdraw the membrane is 5.7736 kN, which agrees with the
theoretical value of 5.7735 kN.

374 / Finite element analysis in geotechnical engineering: Application

B
Two further analyses were then
performed in which the Young's 3
modulus of the membrane was i?
Q
reduced to E= 1OSkPaand E=104 kPa, 3
all other parameters remaining 2 -6
unchanged. Figure 9.13 compares the
a) Shear stress on interface
shear stress distributions in the
Increment
interface elements from all three g
analyses when the displacement, 6, of g
point A (see Figure 9.1 1) is 0.12mm.
As the stiffness of the membrane 3 o
A
reduces, the integration point stresses , , ) ,
force in membraoe
oscillate widely near the sliding front.
Gens et al. (1989) also found such
behaviour and concluded that the Figure 9.12: Results of pull-out test

2
-

analysis
cause of the large stress oscillations
near the sliding front was due to the
use of Gaussian integration (note the
N=WO~-cotes
lntcgranon
above results were obtained using 3
point Gaussian integration). They $,
reported that if the Newton-Cotes
integration scheme was adopted, the
results were satisfactory. The analyses 4
presented above were therefore
-6
repeated with this integration scheme
(again 3 point integration was used)
Figure 9.73: Sheaf stress in
and the results are also shown on
Figure 9.13.
interface elements
The difference between Gaussian
and Newton-Cotes integration is the location of the integration points and their
corresponding weights (Zienkiewicz (l 977)). The integration points for the 3 point
Newton-Cotes method correspond to the positions of the three nodal pairs (i.e. at
each end and the midpoint) of each interface element. The stress in the element, at
each integration point, is then given by the relative displacement between the pair
of nodes at the integration point. The displacement of the other nodes in the
element has no influence. The element is therefore essentially a linkage element
(Frank et al. (1982), Hermann (1978)). Three point Newton-Cotes integration is
equivalent to Simpson's rule for integration. For Gaussian integration the
integration points are located between the end and the midpoint ofthe element, and
at the midpoint of the element. The relative displacement of all nodes affects the
stress at each integration point.
On first inspection of the results given in Figure 9.13 it appears that the
Newton-Cotes integration scheme greatly improves the behaviour of the interface
elements, as the oscillation of the stresses given by the analyses using Gaussian

Restrictions and pitfalls 1 375

integration have been suppressed. However, further investigation of the pull-out
test with E=104kPashows that the stress distribution given by the Newton-Cotes
integration scheme is actually a poor approximation to the correct solution.
Figure 9.13 indicates that, as the stiffness of the membrane reduces, the
gradient of the shear stress distribution on the interface becomes steeper. When
E=906kPa, the stress reduces from the maximum value of 5.8kPa to nearly zero
over a distance of 0.4m. This reduction therefore occurs over two elements, or six
integration points. When E=105 kPa the stress reduces over a distance of 0.2m,
which is the size of a single element. When E=104 kPa the same stress change
probably occurs over a distance less than that represented by one element. This is
likely to cause the stress oscillation seen in Figure 9.13. Smaller elements (less than
0.2m) are clearly necessary for the analysis of this problem if E<105kPa.
Further analyses were performed
6
with E=104kPa, but with smaller
- - - - - - Newton-Cotes integration
elements at end A of the membrane ,
Gauss integration
and interface. Elements with lengths
of 0.05m and 0.025m were placed at ?
the end of the membrane and e3
interface. Figure 9.14 shows the R
results of these analyses when
6=0.12mm and also the results taken
-8L
from Figure 9.13 where the element
a) Shear stress on interface
size is 0.2m.
flAr
At end A of the interface the shear
""
stress is greater than the expected g
Element size (m)
value of 5.77kPa (IOtan30) and the 3
distribution of shear stress in the first
0.025
0.05m is rather surprising. The
0A
membrane is experiencing a shear
b) Axial force in membrane
stress acting on its underside only.
ox
B
Being thin, the bending stiffness of
the membrane is very small. The $ 0.1
'0.025
shear stress acting on the side of the
c) Displacement of membrane
membrane causes end A of the
membrane to bend downwards,
increasing the normal stress in the Figure 9.14: Effect of element size
interface. Thus the maximum
on pull-out results
permissible shear stress is increased.
At a short distance from the end of the membrane the curvature of the membrane
causes a reduction in normal stress in the interface element, resulting in a complex
distribution of shear stress. When the membrane was constrained so that it could
not bend, the maximum normal stress in the interface element remained at lOkPa
and the maximum shear stress at 5.77kPa. This confirms that bending of the
membrane causes the unusual distribution of shear stress, shown in Figure 9.14.

3,
-

376 1 Finite element analysis in geotechnical engineering: Application

The problem of oscillating stress is due to the use of elements too large to
model adequately the steep stress gradient that occurs in this problem. Stress
oscillation does not occur when small elements are used. The numerical problems
encountered here are not due to poor performance of the interface element, but are
due to inadequate modelling of the problem at hand. Only high-order elements that
allow for a complex distribution of stress across the element will be able to
accurately model the sliding front with the use of larger elements. Newton-Cotes
integration is unnecessary and perhaps undesirable, as it has the effect of 'glossing
over' or 'smoothing out' the steep stress gradient, thus hiding the real solution.
When sufficiently small elements are used to describe adequately the stress
gradient, Newton-Cotes and Gaussian integration give similar results.

9.5

Modelling of structural members in plane strain

analysis

With many geotechnical problems, while the main feature might be adequately
represented by a plane strain idealisation, other, smaller components of the same
problem, might not be. For example, atypical urban excavation problem, as shown
in Figure 9.1 5, is considered. Here an embedded retaining wall is supported by
ground anchors and a foundation slab is tied down by tension piles. This could
represent a cross section of an excavation for a road. The structural members are
now considered in turn below.

9.5.1 Walls
If the wall is of the concrete diaphragm type,
then it is best modelled using solid finite
elements with the appropriate geometry and
material properties. There are no serious
modelling problems here as the wall satisfies
the plane strain assumption. However, if the
wall is made from secant or contiguous
concrete piles, steel sheet piles, or some
combination of steel columns and sheeting,
then the properties and geometry of the wall
will vary in the out of plane direction and
therefore will not satisfy the requirements of
plane strain (see Section 1.6 in Volume 1).
Modelling of these components in a plane
strain analysis will therefore involve some
Figure 9.15: Typical urban
approximations. Usually it is possible to
exca va tion
estimate the average axial (EA) and bending
stiffness ( E 0 per metre length ofwall. If beam elements are used to model the wall,
then these parameters can be input directly as the material properties, see Section

Restrictions and pitfalls / 377

3.5 in Volume I. However, if solid elements are to be used, then the average axial
and bending stiffness must be converted into an equivalent thickness t of the solid
elements and an equivalent Young's modulus, E,,. This is done by solving the
following two simultaneous equations:

Axial stiffness:

cE,, = E A

(9.1 1)

Bending stiffness:
where E is the Young's modulus of the wall and A and I are cross sectional area
and moment of inertia per metre length, respectively. It may also be necessary to
calculate some form of average strength, but this will depend on the constitutive
model employed to represent the wall. Clearly, the above procedure only treats the
wall in an approximate manner and it will not be possible to accurately estimate the
details of the stress distribution within the separate components of the wall.

9.5,2 Piles
Modelling the piles below the base slab involves additional assumptions as the
piles are not continuous in the out ofplane direction, but are separated by relatively
large expanses of soil. While it is again possible to estimate average axial and
bending stiffnesses per unit length in the out of plane direction and calculate
equivalent parameters, as was shown for the wall above, modelling the piles with
solid or beam elements implies that the soil is not able to freely move between the
piles, as the piles are essentially modelled as a wall in the out of plane direction,
see Figure 9.16. As a consequence, the lateral movements of the soil below the
excavation will be restricted. It may therefore be more realistic to neglect the
bending stiffness of the piles s s that they provide no resistance to lateral soil
movement. A spring or a series of membrane elements (see Chapter 3 in Volume
1) can then be used to represent the piles.

Real structure

Modelled structure

Figure 9.7 6: Comparison between real

and simulated conditions for piles

378 / Finite element analysis in geotechnical engineering: Application

If a linear spring is used to represent the pile, only elastic behaviour of the pile
can be represented. In addition, if the spring is connected between a node on the
concrete slab and a node in the soil (e.g. nodes A and B in Figure 9.17), only forces
at these two nodes will be accounted for (i.e. the pile and soil are not connected
between A and B). It will then not be possible to account for shaft friction nor set
a limit to the end bearing resistance of the pile (i.e. at node B).
If a series of membrane (or beam) elements are used to model the pile, see
Figure 9.18, then the pile will be connected to the soil along its shaft and
consequently mobilisation of shaft resistance will be simulated to some limited
extent. However, the zero thickness ofthe elements in the plane of analysis implies
that base resistance will be mobilised through a single node and it is therefore
difficult to impose a limit to the end bearing resistance.

Figure 9.17:Springs used to

model piles

Figure 9.18:Membrane elements

used to model piles

9.5.3 Ground anchors

The ground anchors located behind the wall
can be conveniently divided into two parts:
theJixedandfree anchor lengths, as shown in
Figure 9.19. As with the piles, equivalent
axial and bending stiffnesses per unit length
in the out of plane direction can be calculated
for both components of a row of anchors.
Account must be taken of the spacing of the
anchors in the out of plane direction when
calculating these equivalent values. Again, it
is probably best to ignore the bending tzigure 9.19:~~~~~d anchor

Restrictions and pitfalls 1 379

stiffness and model the two anchor components with either springs, membrane
elements or a combination of both, see Figure 9.20. The situation shown in Figure
9.20c, where a spring is used to model the free anchor length, and a series of
membrane elements the fixed length, is probably the more realistic approach. It is
usual to neglect any shear stresses mobilised between the soil and the free anchor
length, consequently the springs are connected to the wall at one end (point A in
Figure 9.206) and the soil (and fixed anchor length) at the other end (point B). As
with the piles, modelling the anchors in this way is rather crude as it is not possible
to accurately account for the limiting end and shaft capacity of the fixed anchor
length.

Figure 9.20;Modelling ground anchors using a) springs, b)

membrane elements and c) a combination of spring and
membrane elements

An alternative way of modelling the fixed

anchor length is shown in Figure 9.21 and
involves the use of solid and interface elements.
It is now possible to simulate the mobilisation of
both shaft and end bearing resistance and to
apply limits to these quantities. However it is
necessary to assign equivalent properties to the
solid and interface elements to take account of
the spacing of the anchors in the out of plane
direction. This is often not straight forward. By
assigning an equivalent Young's modulus to the
solid elements, the fixed anchor length will have
bending stiffness. This will limit soil movements
perpendicular to the anchor. In the real problem
the soil will be able to deform between the
anchors.
Figure 9.2 1:Modelling a
The above discussion indicates that while it ground anchor with solid
is possible to approximate the wall behaviour in
elernents
a realistic manner, much greater approximations

380 / Finite element analysis in geotechnical engineering: Application

are involved with the modelling of both the piles and ground anchors. This arises
because these structural elements are not continuous in the out of plane direction.

9.5.4

Structural members in coupled analyses

A further pitfall can occur when using

membrane or beam elementsto model
structural members in coupled
consolidation analyses. The problem
is best explained by considering the
example of the retaining wall shown
in Figure 9.22. The wall is modelled
using beam elements and therefore
b)
shares a common set of nodes with
both the elements in front of and
behind it, see Figure 9.22b. As the
solid elements on either side of the
wall have common nodes at the wall,
Figure 9.22: Dealing with
the wall will be implicitly assumed to
impermeable line elements in
be permeable in a consolidation
consolidation problems
analysis. Water will therefore flow
(use of interface elements)
freely through the wall. This may not
be the desired outcome. If the wall is to be assumed impermeable, this can be
achieved by placing interface elements along one side of the wall as shown in
Figure 9 . 2 2 ~ .If these elements are non consolidating, they will provide an
impermeable break between the solid elements on either side of the wall.
C)

9.5.5

Structural connections

Potential problems can arise when modelling connections between different

structural components. As an example, the connection between a prop and a
retaining wall is considered. Such a connection can be modelled as either asimple,
pin-joinedorfullmoment connection. These conditions and the possible alternative
ways that can be used to model them in plane strain analyses are shown in Figures
9.23,9.24 and 9.25. It is often all to easy to use the options given in Figure 9.25
when in reality the connection is pin-jointed.

li

Shear = O
Moment = 0

71-

Tie nodes A and B in

horizontal direction

nodes A and B in
rizontal direction

sup,..

Axial thrust * 0

Figure 9.23: Simple connection between wall and prop

Restrictions and pitfalls 1 381

Wall

Moment-0

Axial thrust 0
Shear * 0
Tie nodes A and B in
both horizontal and
n i c a l direction

Figure 9.24: Pin-jointed connection between wall and

Prop
Wall

II

Axial thrust * 0
Shear * 0
Moment * 0

Figure 9.25: Full connection between wabl and prop

9.5.6 Segmental tunnel linings

Many tunnel linings are constructed from segments. Usually between 8 and 12
segments are required to form a complete circular ring, see Figure 9.26a.
Sometimes the segments are bolted to each other, whereas sometimes they are not.
In either event there is little bending resistance between the segments. It is often
convenient to model the lining using beam or solid elements. However, if these
elements are placed as a sequential ring, see Figure 9.26b, a full moment
connection between the segments will be implied. If this is not what is required,
then an alternative approach must be used. At Imperial we have developed special
beam elements which represent the interface between the segments (see Chapter
2). Each segment is modelled as a group of beam elements and these are then
joined together by these small special beam elements, see Figure 9.26~.

382 1 Finite element analysis in geotechnical engineering: Application

Segments of
lining

Modelled with
beam elements

Modelled with beam and

special beam elements

Figure 9.26: Modelling tunnel lining

9.6

Use of the Mohr-Coulomb model for undrained

analysis

As noted in Chapter 7 of Volume1 of this book, the Mohr-Coulomb mode1 can be

used with a dilation angle ranging from v = 0' to v = p'. This parameter controls
the magnitude of the plastic dilation (plastic volume expansion) and remains
constant once the soil is on the yield surface. This implies that the soil will
continue to dilate indefinitely if shearing continues. Clearly such behaviour is not
realistic as most soils will eventually reach a critical state condition, after which
they will deform at constant volume if sheared any firther. While such unrealistic
behaviour does not have a great influence on boundary value problems which are
unrestrained (e.g. the drained surface footing problem), it can have a major effect
on problems which are constrained (e.g. drained cavity expansion, drained axial
pile loading), due to the restrictions on volume change imposed by the boundary
conditions. In particular, unexpected results can be obtained in undrained analysis
in which there is a severe constraint imposed by the zero total volume change
restriction associated with undrained soil behaviour. To illustrate this problem two
examples are presented here.
The first example considers ideal (no end effects) undrained triaxial
compression (Aa, > 0, Au,, = 0) tests on a linear elastic Mohr-Coulomb plastic soil,
with parameters E1=lOOOOkPa, p=0.3, cl=O, p1=24". As there are no end effects,
a single finite element is used to model the triaxial test with the appropriate
boundary conditions. The samples were assumed to be initially isotopically
consolidated with p1=200kPa and zero pore water pressure. A series of finite
element analyses were then performed, each with a different angle of dilation, v,
in which the samples were sheared undrained. Undrained conditions were enforced
by setting the bulk modulus of the pore water to be 1000 times larger than the
effective elastic bulk modulus of the soil skeleton, K', see Chapter 3 of Volume 1.
The results are shown in Figure 9.27a and 9.27b in the form of J-p' and J-E:plots.
It can be seen that in terms of J-p' all analyses follow the same stress path.
However, the rate at which the stress state moves up the Mohr-Coulomb failure
line differs for each analysis. This can be seen from Figure 9.27b. The analysis
with zero plastic dilation, v=OO,remains at a constant Jandp' when it reaches the

Restrictions and pitfalls 1 383

failure line. However, all other analysis move up the failure line, those with the
larger dilation moving up more rapidly. They continue to move up this failure line
indefinitelywith continued shearing. Consequently, the only analysis that indicates
failure (i.e a limiting value ofJ) is the analysis performed with zero plastic dilation.

Figure 9.27: Prediction of a) stress paths and b) stress-strain

curves in undrained triaxial compression using the Mohr-Coulomb
model with different angles of dilation

The second example considers the

,o
undrained loading of a smooth rigid
60
strip footing. The soil was assumed to
have the same parameters as that used
for the triaxial tests above. The initial
stresses in the soil were calculated on j 30
the basis of a saturated bulk unit
20
weight of 20 kN/m3, a ground water
10
table at the soil surface and a
K,,=l -sinql. The footing was loaded
Vertical displacement, V (mm)
by applying increments of vertical
and undrained
Figure 9-28:,load-displacement
were again enforced
curves for a str,@footing, using the
sertingthe
the pore M0hr-Coulomb model with different
water to be 1000times K'. The results
angles of dilation
of two analyses, one with v = B)" and
the other with v = p', are shown in Figure 9.28. The difference is quite staggering:
while the analysis with v = 0" reaches a limit load, the analysis with v = q' shows
a continuing increase in load with displacement. As with the triaxial tests, a limit
load is only obtained if v = 0".
It can be concluded from these two examples that a limit load will only be
obtained if v = 0". Consequently, great care must be exercised when using the
Mohr-Coulomb model in undrained analysis. It could be argued that the model
should not be used with v > 0' for such analysis. However, reality is not that simple
and often a finite element analysis involves both an undrained and a drained phase
(e.g. undrained excavation followed by drained dissipation).Consequently, it may

384 1 Finite element analysis in geotecknical engineering: Application

be necessary to adjust the value of v between the two phases of the analysis.
Alternatively, a more complex constitutive model which better represents soil
behaviour may have to be employed. However, similar problems will occur with
any constitutive model which does not reproduce critical state conditions.
Consequently, great care must be exercised when selecting the model to use.

9.7

Influence of the shape of the yield and plastic

potential surfaces in the deviatoric plane

As noted in Section 7.9 of Volume 1, the shape of the plastic potential in the
deviatoric plane can affect the Lode's angle 8 at failure in plane strain analyses.
This implies that it will affect the value of the soil strength that can be mobilised.
In many commercial software packages, the user has little control over the shape
of the plastic potential and it is therefore important that its implications are
understood. This phenomenon is investigated here by considering the modified
Cam clay constitutive model.
Many software packages assume that both the yield and plastic potential
surfaces plot as circles in the deviatoric plane. This is defined by specifying a
constant value of the parameter M,. Such an assumption implies that the angle of
shearing resistance, (p', varies with the Lode's angle, 6. By equating M, to the
expression for g(6) given by Equation (7.41) in Volume 1 and re-arranging, gives
the following expression for (p' in terms of M, and 8:

From this equation it is possible to express M, in terms of the angle of shearing

in triaxiai compression, 8 = -30, see Equation (7.26)' Volume 1:
resistance, ,p(',

Figure 9.29 shows the variations of (p' with 8, given by Equation (9.13), for
three values of M,. The values of MJhave been determined from Equation (9.14)
using v', = 20, 25" and 30". If the plastic potential is circular in the deviatoric
plane, it can be shown, see Chapter 7 in Volume I,that plane strain failure occurs
when the Lode's angle 6 = 0". Figure 9.29 indicates that for all values of M, there
=25", then
is a large change in (p' with 8. For example, if M , is set to give 'v,
under plane strain conditions the mobilised (p' value is (p', =34.6". This difference
is considerable and much larger than indicated by laboratory testing. The
and p,' becomes greater the larger the value of M,.
difference between 'p,(,

Restrictions and pitfalls / 385

The effect of B on undrained
strength, S,,, for the constant Mj
formulation is shown in Figure 9.30.
The variation has been calculated
using Equation (7.50) (in Volume 1)
with:

'O-

45-

3 ,h

OCR = l
g(@ = M,
K<,= 1- sin&
K 1 /Z = 0.1

The equivalent variation based on the

formulation which assumes a constant
v', instead of a constant M,, is given
in Figure 9.3 1. This is also based on
Equation (7.50) in Volume 1 and the
above parameters, except that g(0) is
now given by Equation (7.41) in
Volume 1. The variation shown in
this figure is in much 'better
agreement with the available
experimental data than the trends
shown in Figure 9.30.
To investigate the effect of the
plastic potential in a boundary value
problem, two analyses of a rough
rigid strip footing, 2m wide, have
been performed. The finite element
mesh is shown in Figure 9.32. The
modified Cam clay model was used to
represent the soil which had the
following material parameters:

TE

Lode's angle, 8 (deg)

Figure 9.29: Variation of 4' with B

for constant MJ

,,
0.45

$,

,
0.35

0.25~~

-20

10

20

30

Figure 9-30: Effect of B on S, for

the constant M, formulation

,,,
04

OCR = 6
v, = 2.848

2.= 0.161
K

= 0.0322

-30

-20

-10

10

20

30

In one analysis the yield and plastic Figure 9.31: Effect of Bon S, for
the constant 4' formulation
potential surfaces were assumed to be
circular in the deviatoric plane. A
value of M,=0.5 187 was used for this analysis, which is equivalent to yT,'=230. in
the second analysis a constant value of y11=23" was used, giving a Mohr-Coulomb

386 1 Finite element analysis in geotechnical engineering: Application

hexagon for the yield surface in the
deviatoric plane. However, the plastic
potential still gave a circle in the
deviatoric plane and therefore plane
strain failure occurred at 8 = 0", as for
the first analysis. In both analyses the
initial stress conditions in the soil
were based on a saturated bulk unit
weight of 18kN/m3,a ground water
table at a depth of
and a Figure 9.32: Finite element mesh
K,,=1.227. Above the ground water
for footing analysis
table the soil was assumed to be
saturated and able to sustain pore
water suctions. Coupled consolidation
200
analyses were performed, but the
permeability and time steps were g
650
chosen such that undrained conditions 2
existed. Loading of the footing was
IM
simulated by imposing increments of >g
50
vertical displacement.
In summary, the input to both
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
analysis is identical, accept that in the
Vertical displacement, v (m)
first the strength parameter Mj is
specified, whereas in the second p' is
Figure 9.33: Load-displacement
input. In both analyses p,,' = 23" and
curves
therefore any analysis in triaxial
compression would give identical results. However, the strip footing problem is
plane strain and therefore differences are expected. The resulting load displacement
curves are given in Figure 9.33. The analysis with a constant M, gave a collapse
load some 58% larger than the analysis with a constant p'. The implications for
practice are clear, if a user is not aware of the plastic potential problem andlor is
not fully conversant with the constitutivemodel implemented in the software being
used, hetshe could easily base the input on p,,' = 23". If the model uses a constant
M, formulation, this would then imply a p,' = 3 1.2",which in turn leads to a large
error in the prediction of any collapse load.

,,,

9.8

Using critical state models in undrained analysis

The input parameters to most critical state models are based on drained soil
behaviour and do not involve the undrained shear strength, S,, . Consequently,
undrained analyses can be problematic. For example, if constructing an
embankment or foundation on soft clay, short term undrained conditions are likely
to be critical from a stability point of view. It is therefore important for any
analysis to accurately reproduce the undrained strength that is available. It is also

Restrictions and pitfalls / 387

likely that establishingthe undrained strength profile would be a priority of any site
investigation.
Although the undrained strength is
,
not one of the input parameters to the
constitutive model, it can be "
calculated from the input parameters ,
,
and the initial state of stress in the
ground, as shown in Appendix VII.4
in Volume 1 for the Cam clay and
modified Cam clay models.
Consequently, if the undrained "Oo
z
4
6
o
10
20
30
40
OCR
S. Wa)
strength profile is known, it is
possible to use Equations (7.49) or
(7.50) in Volume 1 (for Cam clay or Figure 9.34: Variation with depth of
a) overconsolidation ratio and
modified Cam clay respectively) to
6) undrained strength
back calculate one of either OCR,
K,"(' or ~ 1 1 .For example, if the
undrained strength profile at a site resembles that shown in Figure 9.34b, which is
typical of a soft clay deposit, it is possible to set all parameters, except the OCR,
and then to use either Equation (7.49) or (7.50) to calculate the distribution of OCR
which is consistent with the required S,,profile. Such a distribution of OCR, based
on Equation (7.50) (modified Cam clay) is given in Figure 9.34a. Clearly it will be
necessary for the finite element software to be flexible enough to allow the user to
input such a variation of OCR.
It should be noted that for both Cam clay and modified Cam clay the undrained
Consequently,
strength, S,,, is linearly related to the vertical effective stress, aVl1.
= 0,then so will the undrained strength. This explains why it is necessary for
if a,,,'
the OCR to increase rapidly near to the ground surface in Figure 9.34a. However,
even if a,,' = 0 at the ground surface (i.e. no pore water suctions present) it is still
possible to perform a finite element analysis which simulates a finite undrained
strength at the surface. This is possible because the constitutive model is only
evaluated at the integration points which lie at a finite distance below the ground
surface.

9.9

KK1

Construction problems

The procedure that is usually followed when constructing material in a finite

element analysis is described in Chapter 3 in Volume l. During construction the
elements representing the new material are given a constitutive model which is
consistent with its behaviour during construction. In practice this is usually
achieved by assuming the material is linear elastic with a very low stiffness. Once
constructed, the constitutive model is changed to represent the behavicur of the
material once in place. This usually involves an elasto-plastic constitutive model.
Clearly problems can arise if the stress states in the elements immediately after

388 / Finite element analysis in geotechnical engineering: Application

construction are inconsistent with the new

constitutive model.
When constructing material in a finhe
element analysis, problems can arise if the
elements are constructed to the side of
existing material. For example, consider the
situation shown in Figure 9.35 where
material is to be backfilled behind a retaining
wall. Typically each row of elements in the
backfill would be constructed sequentially.
During construction of row 1, element A
immediately adjacent to the retaining wall
will be restrained by the stiffness of the wall.
This could result in the element hanging Figure 9-35;Construction of
from the wall and the calculation of tensile material adjacent to existing
direct stresses at the integration points
material
adjacent to the wall. Ifthe constitutive model
to be used after construction cannot tolerate these tensile stresses, then an
incompatibility arises. A similar pattern of events could occur when constructing
the remaining rows of backfilled elements. There are several options to overcome
this problem, including:

- Adjust the stresses after construction such that they are consistent with the new

constitutive model. Clearly, there is no obvious way in which this should be

performed and therefore this process involves and ad-hoc approach. The
accumulated right hand side vector of the finite element equations must also be
adjusted to account for the changed stresses so that equilibrium is maintained.
Construct a new mesh with more rows of shorter elements behind the wall and
try again.
Use frictionless interface elements between the wall and soil during
construction. Once constructed, the adjacent interface element should be reset
to sustain friction.
Use the same constitutive model to represent the soil during and after
construction. Unfortunately, many ofthe current constitutive models are unable
to cope with the conditions arising during construction.

9.1 0 Removal of prescribed degrees of freedom

During a finite element analysis it is sometimes necessary to change the boundary
conditions so that a degree of freedom at a particular node changes from being
prescribed to being not prescribed. For example, consider the problem of undrained
loading followed by consolidation under constant load of a smooth rigid surface
strip footing. During the undrained stage the loading of the footing could be
simulated by applying increments of vertical displacement to the nodes on the
ground surface immediately below the footing (i.e. along AB in Figure 9.36).

Restrictions and pitfalls 1 389

During the subsequent consolidation
stage the load on the footing is
required to remain constant but the
footing is free to move vertically if it
wishes. This could be modelled in the
4
finite element analysis by replacing ?a
3
the prescribed displacements along 3
!
l
AB with tied freedom boundary
conditions which tie the subsequent
incremental vertical displacementsof
the nodes along AB.
What happens when the
AU=AV=O
prescribed displacement boundary
condition for the nodes along AB is Figure 9-36:Boundary conditions
removed is crucial. In the above
for footing problem
discussion it has been assumed that
the reactions associated with the prescribed displacements,and which represent the
load on the footing, remain for further increments of the analysis. This is the
preferred manner in which the software should deal with the situation. However,
it involves a considerable amount of book keeping and it is much simpler for the
software to automatically remove the reactions on the increment at which the
prescribed displacement condition is removed. This is not desirable and software
which works in this manner should be used with caution. Of course, there are
situations where the user may wish to remove some or all of the reactions created
by previous prescribed displacements. For example, at some stage after or during
consolidation it may be necessary to unload the strip footing in the example quoted
above. This is best simulated by having a facility in the software for removing a
proportion of the reactions. If this is not availablethen the analysismust be stopped
so that the reactions can be evaluated and then restarted with stress or point load
boundary conditions which impose nodal forces in the opposite direction and ofthe
required magnitude to cancel all, or part of, the reactions.

9.1 1 Modelling underdrainage

It is not uncommon, especially in urban areas, to find that the pore water pressure
distribution in the ground is controlled, in part, by underdrainage. For example, in
London, the situation shown in Figure 9.37 is not uncommon. Due to extraction of
water from the chalk aquifer during the Industrial Revolution in the early part of
the 19'h century, the pore water pressures in the chalk and in the overlying
Lamberth Group were depressed. They have remained depressed and currently, if
extrapolated upwards, indicate a phreatic surface (zero pore water pressure) some
distance into the London Clay. To complicate matters, a perched water table has
remained in the permeable Thames Gravel which lies on top of the London Clay.
The distribution of pore water pressure in the less permeable London Clay is

390 / Finite element analysis in geotechnical engineering: Application

Pore water pressure

therefore controlled by these two
----- -Lc;Y?
hydraulic boundary conditions. This has
Hydrostatic
T
resulted in the complex distribution of
gradient
Thames mavel
pore water pressure shown in Figure 9.37.
If the London Clay had a constant
, w u n d e r d r a i n e d profile
permeability which did not change with
depth and if sufficient time for full
consolidation had passed, then the pore
water pressure distribution should 2
layer
theoretically vary linearly with depth, fi
from the value at the bottom of the
London Clay
Thames Gravel to that at the top of the
Lambeth Group, as shown by the dotted
line in Figure 9.37. The fact that it does
Lambeth Group
not can be explained by one of, or the
\
Chalk
combination of, two scenarios. Firstly,
the permeability of the clay is so low that
full equilisation of excess pore water
Figure 9.37: Typical
pressures has not occurred and therefore
underdrained pore water
consolidation is not complete, or p,,
in London Clay
secondly, the permeability of the clay
varies with depth. Calculations based on
Pore water pressure (@a)
measured values of permeability indicate
-50
0
50
100 150 200 250
that it is likely that full consolidation is
more or less complete and therefore that
it is a variation ofpermeability with depth
that causes the nonlinear distribution of
pore water pressure that is observed in the
field. There is also considerable evidence
from laboratory and field tests that 3 10
permeability varies with voids ratio. As
voids ratio decreases with depth, so will
permeability.
l6
A similar situation occurs in the soils
l8
below Bangkok, as discussed in Chapter
4 and shown in Figure 4.5, which is
20
repeated here as Figure 9.38. A perched
22
water table exists in the more permeable
Sand
weathered surface crust and the pore
Water pressures have been reduced to Figure g, 38: Underdrained pore
zero at the top of the deep sand layer.
water pressure profile in
This results in a nonlinear distribution of
Bangkok Clay
pore water pressure with depth
throughout the soft and stiff clay layers.

I \

Restrictions and pitfalls 1 391

As noted on many occasions throughout this book, it is essential to establish or
input realistic and accurate distributions of the initial stresses and pore water
pressure in the ground prior to the analysis of any loading and unloading due to
construction or excavation. For example, if a tunnel was to be excavated in either
the lower levels of the London Clay or in the stiff clay below Bangkok, then it is
important to accurately model the initial pore water pressure conditions, as these
would influence the initial effectives stresses which in turn would control the
undrained strength of the clay.
It is desirable, and in principle acceptable, to input the stresses and pore water
pressures directly as initial conditions into a finite element analysis. In this way it
is quite easy to input the nonlinear distribution of pore water pressure. However,
if the analysis involves coupled consolidation, this must be done with great care
because the distribution of pore water pressure must be consistent with the
distribution of permeability that is input in the form of soil properties.
As an illustration of the problems that can
arise, the Bangkok situation shown in Figure
9.38 will be considered. The soil profile is
modelled by the simple column of elements
shown in Figure 9.39, which is constrained
vertically at the bottom and horizontally along
the vertical sides. The soil properties are
discussed in detail in Chapter 4 and are listed in
Tables 4.2, 4.3 and 4.4. Note that the standard
14
critical state formulation of the elastic
component of the model was used here. The
initial distributions of overconsolidation ratio
(OCR) and K,, are shown in Figures 4.10 and
Stiff clay
4.11.
,=a
I
The initial effective stress distribution based
22
on the bulk unit weight of the soil, the K,,
Z-X NO^ to scale)
d.%
distribution and the initial pore water pressure
distribution shown in Figure 9.38, were input as
initial values.
Figure 9.39: Finite element
Two coupled consolidation analyses were mesh for column analysis
then performed. In both analyses the pore water
pressures at the soil surface and at the bottom of the stiff clay layer were
maintained at their initial values and only time increments were imposed. The two
analyses differed in that in the first constant values of permeability (independent
was used for
of depth) were specified for the clay layers. A value of k=8~lO-'~m/s
and lx 10"Om/swere used
the weathered crust, whereas values of k of 5.5~10-'~m/s
for the soft and stiff clays respectively, see Figure 9.40. In the second analysis the
permeability was assumed to vary according to Equation (4.4), as given in Figure
4.13 and repeated in Figure 9.40. Ideally the permeability should vary with voids
ratio, however in practice there is seldom sufficient data available and, as an

392 1 Finite element analysis in geotechnical engineering: Application

Coefficient of permeability, k ( d s )
expedient, an expression in terms of
mean effective stress is often employed,
o 10.'~
ioe9
0
see Chapter 10 of Volume 1.
2
In both analyses sufficient time steps
were imposed to allow steady state
4
conditions to be reached. Of course, if the
permeability distribution is consistent
with the initial pore water pressure
8
distribution, steady state conditions 3 10
should exist from the beginning of the $. 12
analysis. In addition, no movements or
l4
stress changes within the soil should
occur. The results of the two analysis are
l6
presented in Figures 9.4 1,9.42 and 9.43.
Figure 9.4 1 compares the predicted
20
steady state pore water pressures with the
for each layer
initial values. While the analysis using a
22
e measured
permeability based Equation (4.4)
indicates that the steady state values are Figure 9.40: Permeability profiles
identical to the initial values, this is not so adopted for two column analyses
for the analysis with constant
Pore water pressure (kPa)
permeabilities, where a linear distribution
of pore water pressure is predicted within
'? Is''
2s!
the soft and stiff clay layers. The
Weathered crust
distribution of vertical movements within
the soil is shown in Figure 9.42. While
there are negligible movements from the
analyses using Equation (4.4), significant
8
Init~aIpore water pressure
profile, also steady state
movements are predicted for the analysis g l 0
with constant permeabilities. Figure 9.43 f3 12
shows the predicted changes in undrained
14
strength. Again significant changes are
predicted by the analysis with constant
permeabilities.
Steady state profile
The analysis based on permeabilities
given by Equation (4.4) indicates
22
compatibility between the distributions
of initial permeability and pore water
Figure 9-41: Comparison of
pressure. This is not so for the analysis
initial and steady state pore
based on constant permeabilities. The
water pressure profiles from
implications for finite element analyses
two column analyses
are clear. For example, if a tunnel was to
be excavated in the stiff clay, then if permeabilities inconsistent with the initial
pore water pressure profile were used the analysis would be in error. If sufficient

Restrictions and pitfalls 1 393

Vertical displacement (m)
time was allowed for equilisation of pore
water pressures before excavating the
I
tunnel, then the analyses would
I
essentially not be simulating construction
\
I
Variable permeability
in soil with the assumed initial
\
\
conditions. Alternatively, if sufficient
\
\
time was not allowed for initial
equilisation, the predicted movements
and stress changes would be a
combination of those arising from tunnel
excavation and equilisation due to the
initial incompatibility between
permeabilities and initial pore water
/
pressures.
Constant permeability \ \
Clearly, an incompatibility between
for each layer
the initial distributions of pore water
pressure and permeability is a potential
pitfall when performing analyses. In this
respect it should be noted that it is not the
Figure 9.42: Comparison of
absolute values of permeability that vertical displacements at steady
dictate the shapes of the pore water state from two column analyses
pressure distribution, but their relative
changes with depth. The shape is also
Undrained strength, S, (kPa)
dependent on the hydraulic boundary
conditions. If these conditions are such
that the initial pore water pressure profile
is hydrostatic, then this condition is not
affected by the distribution of
permeability, as such a profile is
independent of both the magnitude and
distribution of permeability.
As noted in Chapter 4, the parameters
defining the permeability variation given
by Equation (4.4) were chosen to be
compatible with the initial pore water
profile. It is for this reason that the
analysis based on this permeability
distribution was correct. In practice
information on the pore water pressure
profile is often available, but there is little
information on the distribution of
Figure 9.43: Comparison of
permeability. It is therefore useful to initial and steady state undrained
perform the simple column analysis
strength profiles from t w o
described above to experiment with
column analyses

394 / Finite element analysis in geotechnical engineering: Application

different distributions. Even if permeabilty data is available, it is still sensible to

perform the column analysis to ensure compatibility between the initial
permeability and pore water pressure distributions.
While in the above example the correct behaviour was obtained using a
permeability varying with mean effective stress, see Equation (4.4), initial
compatibility could have been obtained by allowing the permeability simply to
change with depth. In such a case the distribution shown in Figure 4.13 would have
to be adopted.

9.12 Summary
1.

2.

3.

4.

5.

6.

7.

The two main sources of error involved in the finite element method itself are
associated with the integration of the constitutive equations and with the
discretisation of the geometry into finite elements. The magnitude of the
errors involved in the former have been discussed in Chapter 9 of Volumel,
where it was shown how they could be minimised by the use of a robust stress
point algorithm. In this chapter the magnitude of the discretisation error has
been investigated. This can be significant, but can be reduced by a suitable
choice of finite element mesh. In particular, small elements need to be placed
where there are rapid changes in stresses and strains.
The zero thickness interface element is useful for modelling relative slip and
opening and closing on a pre-defined surface. Numerical problems can
however occur through ill-conditioning of the stiffness matrix and high stress
gradients in the interface elements. In many situations stress gradients are
likely to be high and are increased with increased interface stiffness. The
problem can therefore easily be confused with ill-conditioning.
Ill-conditioning was noticed in the problems analysed when the stiffness of
the interface element was greater then approximately 100 times the Young's
modulus of the surrounding soil. This however depends on the units used in
the problem and the size of the surrounding elements.
The normal and shear stiffness of zero thickness interface elements have
dimensions of force/(length)'. They are therefore different to the bulk and
shear stiffness of solid elements which have dimensions of for~eI(1ength)~.
Newton-Cotes integration tends to improve the numerical behaviour of
interface elements at, possibly, the expense of hiding the true solution. When
sufficiently small elements are used to model the interface behaviour,
Newton-Cotes and Gaussian integration give similar results.
The modelling of structural components in plane strain analysis can be
problematic. While the major component of the problem may render itself to
a plane strain idealisation, many of the structural components may not satisfy
such requirements. Approximations must then be made. This usually involves
some averaging of material properties and is likely to impose some
restrictions on the analysis.
Particular attention must be made where structural elements are modelled
using beam or membrane elements in coupled consolidation analyses. It is

Restrictions and pitfalls / 395

8.
9.

10.

1 1.

12.

13.

14.

very easy to make a mistake and model what are supposed to be impermeable
structural components as permeable in such situations.
Care must also be exercised when modelling the conriections between
structural members.
When using the Mohr-Coulomb model for undrained analysis, it must be
recognised that failure is unlikely to be predicted, unless the angle of dilation
is set to zero.
The shape of the plastic potential in the deviatoric plane can have a major
effect on the strengths mobilised in plane strain analysis. Particular care must
be exercised when using critical state models, as many finite element
programs assume that both the yield and plastic potential surfaces are circular
in the deviatoric plane. This can lead to unrealistic angles of shearing
resistance andlor undrained strengths being mobilised. It is more realistic to
use a model which assumes a Mohr-Coulomb hexagon for the shape of the
yield surface in the deviatoric plane, in conjunction with a plastic potential
which produces a circle.
Although the undrained strength, S,,, is not an input parameter to most critical
state models, it can be evaluated from the input data and initial stress
conditions. It is therefore possible to adjust one or some of these values to
give the desired undrained strength profile. One alternative is to keep all
parameters fixed, except the OCR, which is varied with depth to obtain the
desired S,,profile.
Problems can occur when constructing material in finite element analyses.
This arises due to an incompatibility between the stresses calculated during
construction and the constitutive model that is then assigned after
construction. Several alternatives are available for dealing with this problem.
There is a potential problem associated with changing the boundary condition
for a particular degree of freedom from prescribed to not prescribed during
an analysis. It is important the user knows how the software being used will
treat such a change.
If the initial pore water pressure distribution at the beginning of an analysis
is influenced by underdrainage and a coupled finite element analysis is to be
undertaken, then care must be exercised to ensure compatibility between this
profile and the distribution of permeability. Otherwise, significant errors will
be introduced into the analysis. In such cases a simple 1D column analysis is
recommended before undertaking the main analysis.

References

Abbo A.J. & Sloan S.W. (1996), "An automatic load stepping algorithm with error
control", Int. Jnl. Num. Meth. Engng, Vol. 39, pp 1737-1759
Addenbrooke T. I. (1996), "Numerical analysis of tunnelling in stiff clay", PhD
thesis, Imperial College, University of London
Addenbrooke T.l & Potts D.M. (2001), 'Twin tunnel interaction - surface and
subsurface effects", accepted for publication in the Int. Jnl. Geomech
Addenbrooke T.I., Potts D.M. & Puzrin A.M. (1997), "The influence ofpre-failure
soil stiffness on numerical analysis of tunnel construction",
Geotechnique, Vol. 47, No. 3, pp 693-712
A.G.I. (199 l), "The leaning Tower of Pisa: present situation", Proc. 1Oth Eur. Conf.
Soil Mech. & Found. Eng., Florence
A.G.S (1994), "Validation and use of geotechnical software", Association of
Geotechnical Specialists, London
Anderson K.H. (1998), "Skirted anchors - case histories in cost effectiveness",
NR. 199, Norwegian Geotechnical Institute
Apted J.P. (1977), "Effects of weathering on some geotechnical properties of
London Clay", PhD thesis, Imperial College, University of London
Arthur J.R.F. & Menzies B.K. (1972), "Inherent anisotropy in a sand",
Geotechnique, Vol. 22, No. 1, pp 1 15-128
Barton M.E., Cooper M.R. & Palmer S.N. (1988), "Diagenetic alteration and
micro-structural characteristics of sands: neglected factors in the
interpretationof penetration tests", Proc. Conf. Penetration Testing in the
UK, I.C.E., Birmingham
Bernat S. (1996), "Modelisation du creusement d'un tunnel en terrain meuble",
Qualification sur chantier experimental, PhD thesis, Ecole Central de
Lyon
Binnie G.M. (1978), "The collapse of the Dale Dyke dam in retrospect", Q.J.
Engng.Geol.,Vol. l l, pp 305-324
Bishop A. W. (1955), "The use of the slip circle in the stability analysis of slopes",
Geotechnique, Vol. 5, No. l , pp 7-1 7
Bishop A.W. & Bjerrum L. (1960), "The relevance of the triaxial test to the
solution of stability problems", ASCE Research Conf. Shear Strength of
Cohesive Soils, Boulder, Colorado, pp 437-45 1
Bishop A.W & Henkel D.J. (1962) "Test measurement of soil properties in the

References 1 397

triaxial test", Arnold, London

Bishop A.W. & Wesley L.D. (1975), "A hydraulic triaxial apparatus for controlled
stress path testing", Geotechnique, Vol. 25, No. 4, pp 657-670
Bolton M.D. (1986), "The strength and dilatancy of sands", Geotechnique, Vol.
36, NO. 1, pp 65-78
Bolton M.D. & Lau C.K. (1993), "Vertical bearing capacity factors for circular
and strip footings on Mohr-Coulomb soil", Can. Geotech. Jnl., Vol. 30,
pp 1024- 1033
Boscardin M.D. & Cording E.J. (1989), "Building response to excavation induced
settlement", Jnl. Geotechnical Eng, ASCE, Vol. 115, No. 1, pp 1-2 1
Burland J.B. (1969), Discussion. Proc. Conf. Insitu Investigations in Soil & Rock,
I.C.E., pp 62
Burland J.B. (1990), "On the compressibility and shear strength of natural clays",
Geotechnique, Vol. 40, No. 3, pp 327-378
Burland J.B. (1995), "Assessment of risk ofdamage to buildings due to tunnelling
and excavation", Proc. 1" Int. Conf. Earthquake Geotechnical Eng, ISTokyo
Burland J. B. & Hancock, R. J. R. (1977)," Underground car park at the House of
Commons, London: Geotechnical aspects", The Structural Engineer, Vol.
55, No. 2, pp 87-100
Burland J.B., Jamiolkowski M., Lancellotta,R., Leonards G.A. & Viggiani C.
(1 993), " Leaning Tower of Pisa - what is going on", ISSMFE News, 20,
2
Burland J.B., Jamiolkowski M., Lancellotta R., Leonards G.A. & Viggiani C.
(1994), "Pisa update - behaviour during counterweight application",
ISSMFE News, 2 1 , 2
Burland J. B. & Kalra J.C. (1986), "Queen Elizabeth I1 Conference Centre:
Geotechnical Aspects", Proc. Inst. Civ. Eng, Part I, Vol. 80, pp 14791503
Burland J.B., Longworth T.I. &Moore, J.F.A. (1977), "A study of ground
movement and progressive failure caused by a deep excavation in Oxford
Clay", Geotechnique, Vol. 27, No. 4, pp 557-591
Burland J.B. & Maswoswe J. (1982). Discussion on "In-situ measurement of
horizontal stress in overconsolidated clay using push-in spade-shaped
pressure cells" by P. Tedd & J.A. Charles, Geotechnique, Vol. 32, No. 3,
pp 285-286
Burland J.B & Potts D.M. (1994), "Development and application of a numerical
model for the leaning tower of Pisa", Pre-failure Deformation of
Geomaterials, Edt. Shibuya, Mitachi & Miure, Balkema, Rotterdam, pp
715-737
Burland J.B., & Wroth C.P. (1974), "Settlement of buildings and associated
damage", BGS conference 'Settlement of Structures', Cambridge, pp
61 1-651
BS8002 (1 994), "Code of practice for earth retaining structures", BSI publications,

398 / Finite element analysis in geotechnical engineering: Application

British Standards Institution, London

BS8006 (1995), "Code of practice for strengthenedlreinforced soils and other
fills", BSI publications, British Standards Institution, London
Calabresi G., Rampello S. & Callisto L. (1993), "The Leaning Tower of Pisa Geotechnical characterisation of the Tower's subsoil within the
framework of the Critical State Theory", Studi e Ricerche, Dipartimento
di Ingegneria Strutturale e Geotechnica, Universita di Roma 2 La
Sapienza
Carter J.P., Desai C.S., Potts D.M., Schweiger H.F. & Sloan S.W. (2000),
"Computing and computer modelling in geotechnical engineering",
GeoEng 2000, Melbourne, Technomic, Lancaster, Vol. 1, pp 1 157- 1252
Caswell I., Carder D.R. & Gent A.J.C. (1993), "Behaviour during construction of
a propped contiguous bored pile wall in stiff clay at Walthamstow",
TRRL Project Report 10 (E468A/BG), Transport and Road Research
Laboratory
Chandler R.J. (1974), "Lias C1ay:The long-term stability of cutting slopes",
Geotechnique, Vol. 24, No. 1, pp 24-38
Chandler R.J. (1984a), "Recent European experience of landslides in overconsolidated clays and soft rocks", Proc. 4'h Int. Symp. Landslides,
Toronto, Vol. 1, pp 61-81
Chandler R.J. (1984b), "Delayed failure and observed strengths of first-time slides
in stiff clay", Proc. 4Ih Int. Symp. Landslides, Toronto, Vol. 2, pp 19-25
Chandler R.J. & Skempton, A.W. (1974), 'The design of permanent cutting slopes
in stiff fissured clays", Geotechnique, Vol. 24, No. 4, pp 457-464
Charles J.A. (1976), "The use of one-dimensional compression tests and elastic
theory in predicting deformations of rockfill embankments9', Can.
Geotech. J., Vol. 13, No. 3, pp 189-200
Charles J.A. & Watts K.S. (1985), "Report on testing of rockfill for Roadford
Reservoir", Job No.W84/203, Building Research Establishment, UK
Christophersen H.P., Bysveen S. & Stove O.J.(1992), "Innovative foundation
systems selected for the Snorre field development9', Behaviour of
offshore structures (BOSS), Vol. 1, pp 81-94
Clough G. W. & Leca E. (1989), "Whit focus on use of finite element methods for
soft ground tunnelling", Tunnels et micro-tunnels en terrain Meuble du
chantier a la theome (Proc. Int. Conf. on tunnelling and micro-tunnelling
in soft ground, from field to theory). Presses de I'ecole national des pot
et chaussees, Paris, Vol. l , pp 53 1-573
Coyle H.M. & Reese L.C. (1966), "Load transfer for axially loaded piles in clay",
Jnl. Soil. Mech. Found. Div., ASCE, Vol. 92, SM2, pp 1-26
Crabb G.I. & Atkinson J.H. (1 991), "Determination of soil strength parameters for
the analysis of highway slope failures", Proc. Int. Conf. Slope Stability
Engineering, inst. Civ. Eng., London, pp 13-18
Davies R.V. & Henkel D.J. (1982), "Geotechnical problems associated with the
construction of Chater Station", The Arup Journal 17, No. 1, pp 4- 10

References / 399

Davis E.H. (1968), "Theories of plasticity and the failure of soil masses", Soil
Mechanics - Selected Topics, Edt. I.K. Lee, Butterworths, London, pp
34 1-380
Davis E.H. & Booker J.R. (1973), "The effect of increasing strength with depth on
the bearing capacity of clays", Geotechnique, Vol. 23, No. 4, pp 55 1-565
Day R.A. & Potts D.M. (19941, "Zero thickness interface elements - numerical
stability and application", Int. Jnl. Num. Anal. Geomech., Vol. 18, pp
698-708
Day R.A & Potts D.M. (1998), "The effect of interface properties on retaining wall
behaviour", Int. Jnl. Num. Anal. Meth. Geomech., Vol. 22, pp 1021-1 033
Day R.A. & Potts D.M. (2000), Discussion on "Observations on the computation
of the bearing capacity factor N, by finite elements" by Woodward &
Griffiths, Geotechnique, Vol. 50, No. 3, pp301-303
De Moor E.K. (1994), "An analysis of bored pileldiaphragm wall installation
effects", Geotechnique, Vol. 44, No. 2, pp 341-347
Desai C.S., Zaman M.M, Lightner J.G. & Siriwardane H.J. (1984), "Thin-layer
element for interfaces and joints", Int. Jnl. Num. Anal. Meth. Geomech.,
Vol. 8, pp 19-43
Dounias G.T. (1987), "Progressive failure in embankment dams", PhD thesis,
Imperial College, University of London
Dounias G.T., Potts D.M. & Vaughan P.R. (1996), "Analysis of progressive failure
and cracking in old British dams", Geotechnique, Vol. 46, No. 4, pp 62 1640
Dounias G.T., Potts D.M. & Vaughan, P.R. (1989), "Numerical stress analysis of
progressive failure and cracking in embankment dams", Report to the
Department of the Environment, Contract No.PECD 7171222, Building
Research Establishment
Duncan J.M. (1 992), "Static stability and deformation analysis", Proc. Spec. Conf.
Stability and Performance of Slopes and Embankments - 11, Geot. Eng.
Div., ASCE, Vol. l , (Edt. Seed and Boulanger), pp 222-266
Duncan J.M. (1994), "The role of advanced constitutive relations in practical
applications", Proc. 131h Int. Conf. Soil Mechs. Found. Eng., New Delhi,
Vol. 5, pp 3 1-48
Duncan J.M., Byrne P.M., Wang K.S. & Mabry, P. (1980), "Strength, stress-strain
and bulk modulus parameters for finite element analyses of stresses and
movements in soil masses", Geotechnical Engineering Research Report
No. UCBlGTI8O-0 1, {Jniversity of California, Berkeley
Duncan J.M. & Seed R.B. (1986), "Compaction induced earth pressures under K,
conditions", Jnl. Geotech. Eng., ASCE, Vol. 112, No. 1, pp 1-21
EUR 15285 (1 993), "INTERCLAY I1 project: A co-ordinated benchmark exercise
on the rheology of clays"", The European Commission, Brussels
Fernie R., Kingston P., St John H.D., Higgins K.G. & Potts D.M. (1996), "Case
history of a deep 'stepped box' excavation in soft ground at the seafront Langney Point, Eastl>ourne", Geotechnical Aspects of Underground

400 / Finite element analysis in geotechnical engineering: Application

Construction in Soft Clay, Edt. R.J. Mair & R. N. Taylor, Balkema,
Rotterdam, pp 123- 130
Finno R.J., Harahap I.S. & Sabatini P.I. (1991), "Analysis of braced excavations
with coupled finite element formulations", Computers & Geotechnics,
Vol. 12, pp 91-1 14
Frank R., Guenot A. & Humbert P. (1982), " Numerical analysis of contacts in
geomechanics", Proc. 41hInt. Conf. Num. Met. Geomech., Rotterdam, pp
37-42
Ganendra D. (1 993), "Finite element analysis of laterally loadedpiles", PhD thesis,
Imperial College, University of London
Ganendra D. & Potts D.M. (2002a), "A new approach to the analysis and design
of pile groups - Theory", In preparation
Ganendra D. & Potts D.M. (2002b), "A new approach to the analysis and design
of pile groups - Application", In preparation
Garga V.K. (1970), "Residual strength under large strains and the effect of sample
size on the consolidation of fissure clay", PhD thesis, Imperial College,
University of London
Gens A., Carol I. & Alonso E.E. (1989), "An interface element formulation for the
analysis of soil-reinforcement interaction", Comput. Geotech. Vol. 7
(1,2), pp 133-151
Gens A. & Potts D.M. (1988), "Critical state models in computational
geomechanics", Engineering and Mechanics, Vol. 15, pp1 78-1 97
Geotechnical Consulting Group (1993), "Finite element parametric study on long
term settlement caused by tunneliing", Report to 'Crossrail', May 1993.
Germaine J.T. (1 982), "Development of the Directional Shear Cell for measuring
cross anisotropic clay properties", DSc thesis, MIT, USA
Gourvenic S. (1998), "Three dimensional effects of diaphragm wall installation
and stagged construction sequences", PhD thesis, University of
Southampton
Gun M.J., Satkunananthan A. & Clayton C.R.I. (1993), "Finite element modelling
of installation effects", Retaining Structures, Edt. C.R.I. Clayton, Thomas
Telford, London, pp 46-55
Griffiths D.V. & Lane P.A. (1999), "Slope stability analysis by finite elements",
Geotechnique, Vol. 49, No. 3, pp 387-403
Hamza M.M.A.F. (1976), "The analysis of embankment dams by nonlinear finite
element method", PhD thesis, Imperial College, London University
Hansen J.B. (1970), "A revised and extended formula for bearing capacity",
Bulletin No. 28, Danish Geotechnical Institute, Copenhagen
Harris D.I., Potts D.M., Brady K.C., Highe D.W. & Higgins K.C. (1993), "Finite
element analysis of anchored earth retaining walls", Retaining Structures,
Edt. C.R.I. Clayton, Thomas Telford, London, pp 599-608
Head K.H. (1 994), "Manual of soil laboratory testing", Vol. 2, Wiley, New York
Hermann L.R. (1 978), "Finite element analysis ofcontact problems", Proc. ASCE,
Vol. 104, EM5, pp1043-1057

References 1 401

Hight D.W. (1993), "A review of sampling effects in clays and sands", Offshore
site investigation and foundation behaviour, Society for Underwater
Technology, Vol. 28, pp 115-146
Hight D.W. & Higgins K.G. (1994), "An approach to the prediction of ground
movements in engineering practice: background and application", Prefailure Deformation of Geomaterials, Edt. Shibuya, Mitachi & Miura,
Balkema, Rotterdam, pp 909-945
Hight D. W., Higgins K. G., Jardine R. J., Potts D. M., Pickles A. R., DeMoor E.
K., & Nyirende, Z. M. (1992), "Predicted and measured tunnel distortions
associated with construction of Waterloo International Terminal",
Predictive Soil Mechanics, Proc. Wroth Memorial Symp., pp 3 17-338
Hight D.W., Jardine R.J. & Gens A. (1987), "The behaviour of soft clays",
Embankments on soft clays, Bulletin of the Public Works Research
Centre, Athens, pp 33-158
Higgins K.G. (1983), "Diaphragm walling, associated ground movements and
analysis of construction", MSc thesis, Imperial College, London
University
Higgins K.G., Fernie R., Potts D.M. & Houston C. (1999), " The use of numerical
methods for the design of base propped retaining walls in a stiff fissured
clay", Geotechnical Aspects of Underground Construction in Soft
Ground, Edt. Kusakabe et al, Balkema, Rotterdam, pp 5 1 1-5 16
Higgins K. G., Potts D. M., and Mai, R. J. (1996), "Numerical modelling of the
influence of the Westminster Station excavation and tunnelling on Big
Ben Clock Tower", Geotechnical Aspects of Underground Construction
in Soft Clay, Edt. R.J. Mair & R. N. Taylor, Balkema, Rotterdam, pp
525-530
Higgins K.G., Potts D.M. & Symonds I.F. (1993), "The use of laboratory derived
soil parameters for the prediction of retaining wall behaviour", Retaining
Structures, Edt. C.R.I. Clayton, Thomas Telford, London, pp 92- 10 1
Hill R. ( 1 950), "The mathematical theory of plasticity", Clarendon Press, Oxford
Hird C.C., Pyrah I.C. & Russell D. (1990), "Finite element analysis of the collapse
of reinforced embankments on soft ground7',Geotechnique, Vol. 40, No.
4, pp 633-640
Horsnell M.R., Norris V.A. & Ims B. (1992), "Mudmat interaction and foundation
analysis", Proc int. Conf. Recent Large Scale Fully Instrumented Pile
tests in Clay, Thomas Telford, London
Huder J. (1 972), "Stability of bentonite slurry trenches with some experience of
Swiss practice", Proc 51h European Conf. Soil Mech. & Found. Eng,
Madrid, pp 5 17-522
Jackson C., Zdravkovic L. & Potts D.M. (1997), "Bearing capacity of pre-loaded
surface foundations on clay", Computer Methods & Advances in
Geomechanics, Edt. J.X. Yuan, Balkema, Rotterdam, Vol. 1, pp 745-750
Jamiolkowski M, Lancellotta K., & Lo Presti D.C.F. (1994), "Remarks on the
stiffness at small strains of six Italian clays", Pre-Failure Deformation of

402 / Finite element analysis in geotechnical engineering: Application

Geomaterials, Edt. S. Shibuya, T. Mitachi & S. Miura, Balkema,

Rotterdam, pp8 17-836
Jardine R.J. (1985), "Investigation of pile-soil behaviour with special reference to
the foundations of offshore structures", PhD thesis, Imperial College,
London University
Jardine R.J. (1992), "Non-linear stiffness parameters from undrained
pressuremeter tests", Canadian Geotechnical Journal, Vol. 29, pp 436-447
Jardine R. J. (1994), "One perspective of the pre-failure deformation characteristics
of some geomaterials", Pre-failure Deformation of Geomateriais, Edt.
Shibuya, Mitachi & Miura, Balkema, Rotterdam, pp 855-885
Jardine R.J. & Chow F.C., "New design methods for offshore piles", MTD
Publication 961103
Jardine R.J & Potts D.M. (1988), "Hutton tension leg platform foundations:
predictions of driven pile behaviour", Geotechnique, Vol. 38, No. 2,
pp23 1-252
Jardine R.J & Potts D.M. (1992), "Magnus foundations: soil properties and
predictions of field behaviour", Proc. Int. Conf. Recent Large Scale Fully
lnstrumented Pile tests in Clay, Thomas Telford, London
Jardine R. J., Potts D. M., Fourie A. B. & Burland J. B. (1986), "Studies of the
influence of nonlinear stress-strain characteristics in soil-structure
interaction", Geotechnique, Vol. 36, No. 3, pp 377-396
Jefferies R.M. & Knowles N.C. (1994), "Finite element benchmarks for clay - the
INTERCLAY project", Proc. 3rdEur. Conf. Num. Meth. Geotech. Eng.,
Edt. I.M. Smith, Balkema, Rotterdam, pp 165-170
Jonsrud R & Finnesand G. (1992), "Instrumentation for monitoring the installation
and performance of the concrete foundation templates for the Snorre
tension leg platform", Behaviour of offshore structures (BOSS), Vol. 1,
pp 690-703
Jovitic V. & Coop M.R. (1998), "The measurement of stiffness anisotropy in clays
with bender element tests in the triaxial apparatus" Geotechnical Testing
Journal, ASTM, Vol. 21, No. 1, pp 3-10
Kimmance J.P, Lawrence S., Hassan O., Purchase N.J. & Tollinger G. (1996),
"Observations of deformations created in existing tunnels by adjacent and
cross cutting excavations", Geotechnical Aspects of Underground
Construction in Soft Ground, Edt. R.J. Mair & R. N. Taylor, Balkema,
Rotterdam, pp 707-7 12
Kovatevic N. (1994), "Numerical analyses of rockfill dams, cut slopes and road
embankments", PhD thesis, Imperial College, University of London
Kovatevid N., Potts D.M., Vaughan P.R, Charles J.A. & Tedd P. (1997),
"Assessing the safety of old embankment dams by observing and
analysisng movement during reservoir operation", Proc. 19'h Int. Cong.
Large Dams (ICOLD), Florence, pp 55 1-566
Kenley R.M. & Sharp D.E. (1992), "Magnus foundation monitoring project instrumentation, data processing and measured results", Proc Int. Conf.

References / 403

Recent Large Scale Fully Instrumented Pile tests in Clay, Thomas

Telford, London
Kuwano R. (1999), "The stiffness and yielding anisotropy of sand", PhD thesis,
Imperial College, University of London
Kuzuno T., Takasaki H., Tanaka M. & Tamai, T. (1996), "Driving control and
ground behaviour of triple circular face shield machine", Geotechnical
Aspects of Underground Construction in Soft Ground, Edt. R.J. Mair &
R. W. Taylor, Balkema, Rotterdam, pp 283-288
La Rochelle P., Trak B., Tavenas F.A. & Roy M. (1974), "Failure of a test
embankment on a sensitive Champlain clay deposit"; Canadian
Geotechnical Journal, Vol. 11, pp 142-164.
Lade P.V. (1977), "Elasto-plastic stress-strain theory for cohesionless soil with
curved yield surfaces", Int. J. Solids Structures, Vol. 13, pp 1019-1035
Lade P.V. & Kim M.K. (1988), "Single hardening constitutive model for frictional
materials, 111. Comparison with experimental data", Computers and
Geotechnics, Vol. 6, pp 13-29
Lambe T.W. (1973), "Predictions in soil engineering", Geotechnique,Vol. 23, No.
2, pp 149-202
Lancellotta R. & Pepe C. (1990), "Pisa Tower - A preliminary report", Politecnico
di Turino, Dipartimento Ingegnaria Strutturale, Rapporto di Ricerca, No
2.1
Leonards G.A. & Ramiah B.K. (1959), " Time effects in the consolidation of
clay", ASTM, STP 254, pp 116-130
Leroueil S. (1977) Quelques considerations sur le comportement des argiles
sensibles; PhD thesis, University of Lava!, Quebec, Canada
Lo Presti D.C.F., Pallara O., Lancellotta R., Armandi M. & Maniscalco R. (1993),
"Monotonic and cyclic loading behaviour of two sands at small strains",
Geotechnical Testing Journal, Vol. 16, No. 4, pp 409-424
Lunne T. & Kleven A. (1981), "Role of CPT in North Sea foundation
engineering", Cone penetration testing, ASCE
Lupini J.F., Skinner A.E. & Vaughan P.R. (1981), "The drained residual strength
of cohesive soils", Geotechnique, Vol. 3 1, No. 2, pp 181-2 13
Mair R.J., Hight D.W. & Potts D.M. (1992), "Finite element analyses of
settlements above a tunnel in soft ground", Contractor Report 265,
Transport and Road Research Laboratory
Mair R.J. & Wood D.M. (1987), "Pressuremeter testing: Methods and
interpretation", CIRIA Ground Eng. Report: In-situ testing, CIRIA &
Butterworths, London & Sevenoaks
Marsland A. & Eason B.J. (1973), "Measurements of displacements in ground
below loaded plates in boreholes", Proc. Conf. Filed Instrumentation,
Butterworths, pp 304-3 17
Marsland A. & Quaterman R.S. (1982), "Factors affecting the measurements and
interpretation of quasi-static penetration tests in clays", Proc. 2"* Eur.
Sym. Penetration Testing, Amsterdam

404 1 Finite element analysis in geotechnical engineering: Application

Minister0 dei Lavori Pubblici (1971), "Ricerche e studi sulla Torre di Pisa ed i
fenomeni connessi alle condizione di ambiente, Vol. 3, I.G.M., Florence
Mori R.T. & Pinto N.L.deS. (1988), "Analysis of deformations in concrete face
rockfill dams to improve face movement prediction", Proc. 161hInt. Cong.
Large Dams (ICOLD), San Francisco, Vol. 2, pp 27-34
NAFEMS (1990), "The standard NAFEMS benchmarks", Publication P1 8,
NAFEMS
Nash D.F.T., Sills G.C. & Davison L.R. (1992), "One -dimensional consolidation
testing of soft clay from Bothkennar", Geotechnique, Vol. 42, No. 2, pp
24 1-256
Naylor D.J. (1975). "Numerical models gbr clay core dams, Proc. Int. Symp.
Criteria and Assumptions for Num. Analysis of Dams, Swansea, Edt.
Naylor, Stagg & Zienkiewicz, pp 489-5 14
Naylor D.J. (1991a), "Stress-strain laws and parameter values", Advances in
Rockfill Structures, Edt. Maranha das Neves, pp 269-290
Naylor D.J. (1991 b), "Finite element methods for fills and embankment dams",
Advances in Rockfill Structures, Edt. Maranha das Neves, pp 291-340
Naylor D.J., Tong S.L. & Shahkarami A.A. (1989), "Numerical modelling of
saturation shrinkage", Numerical Models in Geomechanics NUMOG 111,
Edt. Pietruszczak and Pande, pp 636-648
Naylor D.J. (1999), "On the use of the FEM for assessing the stability of cuts and
fills", Numerical Models in Geomechanics, NUMOG VII, Edt.
Pietruszczak & Pande, Balkema, Rotterdam, pp 553-560
NCE (1 983), New Civil Engineer, October Issue, Emap, London
NCE (1984a), New Civil Engineer, January Issue, Emap, London
NCE (1984b), New Civil Engineer, February Issue, Emap, London
NCE (1991), New Civil Engineer, March Issue, Emap, London
NCE (1994), New Civil Engineer, December Issue, Emap, London
Ng C.W.W. (1992), "An evaluation of soil-structure interaction associated with a
multi-propped excavation", PhD thesis, University of Bristol
Nyren R. (1998), "Field measurements above twin tunnels in London Clay", PhD
thesis, Imperial College, University of London
Ong J.C.W. (1997), "Compensation grouting for bored tunnelling in soft clay",
MSc dissertation, Imperial College, University of London
Ong H. L. (1 996), "Numerical analysis for tunnelling and grouting in soft ground",
MSc dissertation, Imperial College, University of London
O'Neil M.W., Ghazzaly 0.1. & Ha H.B. (I977), "Analysis of three dimensional
pile groups with non-linear soil response and pile-soil-pile interaction",
Offshore Technology Conf., Paper 2838, pp 245-256
Padfield C.J. & Mair R.J. (1984), " Design of retaining walls embedded in stiff
clay", CIRIA Report 104
Pagano L. (1998), "Interpretation of mechanical behaviour of earth dams by
numerical analysis", Proc. Symp. Prediction and Performance in
Geotechnical Engineering, Hevelius Edizioni, Napoli, pp 89- 150

References / 405

Panet M. & Guenot A. (1982), "Analysis of convergence behind the face of a

tunnel", Proc. Tunnelling '82, London, The Institution of Mining &
Metalluragy, pp 197-204
Peck R.B. (1969), " Deep excavations and tunnelling in soft ground", Proc. 7Ihint.
Conf. Soil Mech. & Found. Eng., Mexico, pp 225-290
Penman A.D.M., Burland J.B. & Charles J.A. (1971), "Observed and predicted
deformations in a large embankment dam during construction", Proc.
Instn. Civ. Eng., London, Vol. 49, pp 1-49
Penman A.D.M. & Charles J.A. (1985), "Behaviour ofrockfill dam with asphaltic
membrane", Proc. 11th Int. Conf. Soil Mechs. Found. Eng., San
Francisco, Vol. 4, pp 201 1-2014
Perry J. (1989), "A survey of slope condition on motorway earthworks in England
and Wales", Transport and Road Research Laboratory, Research Report
No. 199, pp 49
Porovic E. (1995), "Investigations of soil behaviour using a resonant column
torsional shear hollow cylinder apparatus", PhD thesis, Imperial College,
University of London
Potts D.M. & Addenbrooke T.I. (1997), "A structure's influence on tunnellinginduced ground movements", Proc. Inst. Civ. Eng. Geotech. Eng., Vol.
125, pp 109-125
Potts D.M., Addenbrooke T.I. & Day. R.A. (1993), "The use of soil berms for
temporary support of retaining walls", Retaining Structures, Edt. C.R.I.
Clayton, Thomas Telford, London, pp 440-447
Potts D.M. & Burland J.B. (2000), "Development and application of a numerical
model for simulating the stabilisation of the Leaning Tower of Pisa",
Developments in Theoretical Geomechanics, Booker Memorial Symp.,
Edt. D.W. Smith & J.P. Carter, Balkema, Rotterdam, pp 737-758
Potts D.M. & Day R.A. (1990), "The use of sheet pile retaining walls for deep
excavations in stiff clay", Proc. Inst. Civ. Engineers, Part 1, Vol. 88, pp
899-927
Potts D.M., Dounias G.T. & Vaughan P.R. (1987), "Finite element analysis of the
direct shear box test", Geotechnique, Vol. 37, No. 1, pp 1 1-23
Pons D.M., Dounias G.T. & Vaughan P.R. (1990), "Finite element analysis of
progressive failure of Carsington embankment", Geotechnique, Vol. 40,
NO. 1, pp 79-101.
Potts D.M. & Fourie A.B. (1986), "A numerical study of the effects of wall
deformation on earth pressures", Int. Jnl. Num. Anal. Geomech., Vol. 10,
No. 4, pp 383-405
Peas D.M. & Knights M.C. (1985), "Finite element techniques for the preliminary
assessment of a cut and cover tunnel", Tunnelling '85, Edt. M.J. Jones,
The Institution of Mining & Metallurgy, pp 83-92
Potts D.M., KovaEevid N. & Vaughan P.R. (1997), "Delayed collapse ofcut slopes
in stiff clay", Geotechnique, Vol. 47, No. 5, pp 953-982.
Potts D.M. & Zdravkovid L. (1999) "Finite element analysis in geotechnical

406 / Finite element analysis in geotechnical engineering: Application

engineering: Theory", Thomas Telford, London

Poulos H.G. & Davis E.H. (1968), "Pile foundation analysis and design", Wiley
Prandtl L. (1920), "Uber die Harte Plastischer Korper", Nachrichten von der
Koniglichen Gesellschaft der Wissenschaften,. Gottingen, Math. Phys.,
Klasse, pp 74-85
Puzrin A, & Burland J.B (1998), "Non-linear model of small-strain behaviour of
soils", Geotechnique, Vol. 48, No. 2, pp 217-233
Randolph M.F. & Poulos H.G. (1982), "Estimating the flexibility of offshore pile
groups", Proc. 2ndInt. Conf. Num. Meths, Offshore Piling, Austin Texas,
pp 3 13-328
Randolph M.F., Martin C.M. & Hu Y., "Limiting resistance of a spherical
penetrometer in cohesive material", Geotechnique, Vol. 50, No. 5, pp
573-582
Reese L.C. (1977), "Laterally loaded piles: program documentation", Jnl. Geot.
Eng. Div., ASCE, Vol. 103, No GT4, pp 287-305
Rigden W.J. & Semple R.M. (1983), "Design and installation of the Magnus
foundations; Predictions of pile behaviour", Conf. on design and
construction of offshore structures, Thomas Telford, London, pp 29-52
Robertson P.K. & Campanella R.G. (1983), "Interpretation of cone penetration
tests. Parts 1 & 2", Canadian Geotechnical Journal, Vol. 20, pp 7 18-745
Roscoe K.H. & Burland J.B. (1968), "On the generalised stress-strain behaviour
of 'wet' clay", Eng. plasticity, Cambridge Univ. Press, pp 535-609
Rowe R. K., Lo K. Y. & Kack G. J. (1983), "A method of estimating surface
settlement above tunnel constructed in soft ground", Canadian
Geotechnical Journal, Vol. 20, pp 1 1-22
Rowe P.W. (1971), "Theoretical meaning and observed values of deformation
parameters for soil; Stress-Strain Behaviour of Soils", Proc. Roscoe
Memorial Symp., Cambridge University, Foulis, pp 143- 194
Rowe R. K., Booker J. R. & Balaam N. P. (1978), "Application ofthe initial stress
method to soil - structure interaction", Int. Jnl. for Num. Meth. Eng., Vol.
12, pp 873-880
Sandroni S.S. (1977), "The strength of London Clay in total and effective stress
terms", PhD thesis, Imperial College, University of London
Sarrailh J. & Tavenas F.A. (1972), "Etude geotechnique preliminarie du site de
Saint-Alban", Internal report GCN-72-09-02 (MS-N2), University of
Laval, Quebec, Canada
Schultheiss P.J. (1981), "Simultaneous measurement of P & S wave velocities
during conventional laboratory soil testing procedures", Marine
Geotechnology, Vol. 4, No. 4, pp 343-367
Schweiger H.F. (1998), "Results from two geotechnical benchmark problems",
Proc. 4'h European Conf. Num. Meth. Geotech. Eng., Edt A. Cividini,
pp645-654
Schweiger H.F. & Freiseder M. (1 994), "Three dimensional finite element analysis
of diaphragm wall construction", Computer Methods and Advances in

References 1407

Geomechanics", Edt. H.J. Siriwardane & M.M. Zaman, Balkema,

Rotterdam, pp 2493-2498
Sharp D.E. (1992), "Magnus foundation monitoring - an overview", Proc Int.
Conf. Recent Large Scale Fully Instrumented Pile tests in Clay, Thomas
Telford, London
Shibuya S. (1999), Personal communication
Shin J.H. (2000), "Numerical analysis of tunnelling in decomposed granite soil",
PhD thesis, Imperial College, University of London
Shin D.M. & Potts D.M. (2001), "Time-based two dimensional modelling of
tunnelling", in preparation
Sieffert J.G. & Bay-Gress Ch. (2000), "Comparison of European bearing capacity
calculation methods for shallow foundations", Proc. Inst. Civ. Eng.
Geotech. Eng., Vol. 143, pp 65-74
Simpson B., Atkinson J. H. & Sovieid, V. (1996), "The influence of anisotropy on
calculations of ground settlementsabove tunnels", Geotechnical Aspects
of Underground Construction in Soft Ground, Edt. R.J. Mair & R. N.
Taylor, Balkema, Rotterdam, pp 5 1 -5 14
Skempton A.W. (1951), "The bearing capacity of clays", Proc. Build. Res.
Congress, Vol. 1, pp 180-189
Skempton A.W. (1964), "Long term stability of clay slopes", Geotechnique, Vol.
14, No. 2, pp 77-101
Skernpton A. W. (1977), "Slope stability ofcuttings in brown London Clay", Proc.
91hInt. Conf. Soil Mechs. Found. Eng., Tokyo, Vol. 3, pp 261-270
Skempton A.W. (1985), "Geotechnical aspects of the Carsington dam failure",
Proc. 161hInt. Conf. Soil Mech. Found. Eng., San Francisco, Vol. 5, pp
2581-2591
Skempton A.W. (1986), "Standard penetration test procedures and the effects in
sands of overburden pressure, relative density, particle size, ageing and
overconsolidation", Geotechnique, Vol. 36, No. 2, pp 425-447
Skempton A.W. & Chrimes M.M. (1994), "Thames tunnel: geology, site
investigation and geotechnical problems", Geotechnique, Vol. 44, No. 2,
pp 191-216
Skempton A.W. & Henkel D.J. (1957), "Tests on London Clay from deep borings
at Paddington, Victoriaand South Bank", Proc. 4thInt. Conf. Soil Mechs.
Found. Eng., London, Vol. 1, pp 100-106
Skinner A.E. (1975), "The effect of high pore water pressures on the mechanical
behaviour of sediments", PhD thesis, Imperial College, University of
London
Standing J.R., Farina M. & Potts D.M. (1998), "The prediction of tunnelling
induced building settlements a case studyV,Tunnels& Metropolises,
Edt. A. Negro &A.A. Ferreira, Balkema, Rotterdam, pp 1053-1058
St John H.D. (1975), "Field and theoretical studies of behaviour of ground around
deep excavations in London Clay", PhD thesis, University of Cambridge
St John H.D., Harris D.I., Pott!j D.M. & Fernie R. (1993), "Design study for a

List of symbols

This list contains definitions of symbols and an indication of the section in the
book where they first appear. Because of the large number of parameters that are
used, some symbols represent more than one quantity. To minimise any confusion
this may cause, all symbols are defined in the text, when they are first used. If
some symbols are missing from the list below, they are in the List of symbols in
Volume 1.

C'

c,,'
c,.'
C,

constant for the log permeability law

horizontal extent of reduced K, zone
relative magnitude of the intermediate
principal stress
cohesion
peak value of cohesion
residual value of cohesion
interface cohesion
void ratio
eccentricity of building axis relative to tunnel
axis
sleeve resistance
gradient of the yield function in J-p' plane,
as a function of Lode's angle
coefficient of permeability
coefficient of permeability at zero effective
stress
coefficient of permeability in vertical
direction
permeability in x and y direction
volume compressibility
ratio of E,,' to E,'
ratio of G,' to E,,'
atmospheric pressure
mean effective stress
inner cell pressure
outer cell pressure

List of symbols 1 41 1
total overburden pressure
pore fluid pressure
pore fluid pressure on boundary
in-situ mean effective stress
deviator stress in triaxial plane
cone resistance
ultimate load
pore pressure ratio
(o;,'+aaf)/2

S
S<' S,,'

Sy

bearing capacity shape factors

(a',' - ou')/2

element thickness
horizontal displacement
vertical displacement
specific volume
specific volume at unit mean effective stress
area
pore water pressure parameter
diameter of plate
building width
half width of the footing
compression index
swelling index
tunnel diameter
final tunnel diameter
initial tunnel diameter
relative density
hogging deflection ratio
sagging deflection ratio
Young's modulus
effective Young's modulus
undrained Young's modulus
constrained modulus
hammer energy in SPT
free-fall energy in SPT
secant Young's modulus
tangent Young's modulus
small strain Young's modulus
vertical effective Young's modulus
horizontal effective Young's modulus
bending stiffness
axial stiffness
rod energy ratio in SPT
axial force

v
v
VI

A
A
B
B
B

,\

cc
C,
D

D,
D",
D,DRhos

DRsag
E
E'
El,
E,
Er

Ex

E,
E,
E',,,
E,,'
Eh'

EI
EA
ER,
F',

(1.4.5)
(1.3.3)
(4.3.2)
(1.3.3)
(1.3.3)
(1.4.3)
(1.4.2)
(4.3.2)
(1.3.3)
(6.5)
(1.3.3)
(9.4.3)
(1.3.5)
(1.3.5)
(1 3 . 2 )
(5.5.3)
(3.4.5)
(1.3.3)
(1.4.5)
(2.4.1)
(6.5.4)
(1.3.2)
(1.3.2)
(2.4.1)
(2.4.4)
(2.4.4)
(1.4.2)
(2.7.1)
(2.7.1)
(1.3.2)
(1.3.3)
(1.3.3)
(1.3.2)
(1.4.2)
( 1.4.2)
(1.3.3)
(1.3.3)
(1.3.3)
(2.6.3)
(2.6.3)
(2.7.1)
(2.7.1)
(1.4.2)
(1.3.3)

41 2 / Finite element analysis in geotechnical engineering: Application

K,,')('
K11
K1
K,
Kt,
M
M,
M,
M/.
M,, My
N
160

Nk
N,, N,,, N,
OCR
P
P,.

Q
Qmax

R
R

Smax
S11
S

ur,uy,U:

vertical and horizontal force

forces on pile group in X, y and z direction
shear modulus
shear modulus in horizontal plane
shear modulus in horizontal-vertical plane
shear modulus in vertical-horizontal plane
building half-width
deviatoric stress invariant
value of J at failure
pile group stiffness matrix
bulk modulus
bulk modulus of pore fluid
bulk modulus of soil skeleton
coefficient of earth pressure at rest
K,, normally consolidated
K,, overconsolidated
passive coefficient of earth pressure
tangent bulk modulus
shear stiffness of interface element
normal stiffness of interface element
bending moment
limiting moment on joint extrados
limiting moment on joint intrados
torque load
moments on pile group about x and y axes
SPT blowcount
normalised SPT blowcount
cone factor
bearing capacity factors
overconsolidation ratio
cavity pressure
limiting cavity pressure
load on the footing
maximum load on footing
residual factor
principal stress ratio
maximum shear force
undrained strength
stress level
pull-out force of suction anchor
pile group displacement in X, y, and z
direction
volume of cavity
vertical and horizontal load

List of symbols / 41 3

excavated tunnel volume

volume loss
volume of settlement trough
final tunnel volume
vertical load
water content
water content at liquid limit
water content at plastic limit
inclination of the major principal stress to
the vertical
relative axial stiffness
soil stiffness reduction factor
bulk unit weight
effective bulk unit weight
Ydry
Ysat

YXY
Yr2

Y,o
Y20

Y
6
6
6

, ,b

6h

E,,

4.
E11
&,l

9
&hc
&h!

E/,

0
K

R
R
2,
P
P
Pc,rl

dry bulk unit weight

saturated bulk unit weight
shear strain in X-y plane
shear strain in r-z plane
shear strain in r-8 plane
shear strain in z-8 plane
unit weight of water
interface friction angle
relative displacement
vertical displacement of footing
horizontal displacement at mid-slope
axial strain
radiaI strain
circumferential strain
volumetric strain
cavity strain
compressive horizontal strain
tensile horizontal strain
horizontal strain
Lode's angle
inclination of swelling line in v-lnp' plane
inclination of VCL in v-lnp' plane
convergence-confinement parameter
value of l at which lining is installed
Poisson's ratio
coefficient of friction
Poisson's ratio for straining in the radial
direction due to changes in axial stress
Poisson's ratio for straining in the horizontal

41 4 1 Finite element analysis in geotechnical engineering: Application

V,,'

P
P
P'

02, G3
4"' 0:

01,
or,

0,
G',

(~,,'>b
0,

0,
G,.'

0,
~ 1 1

a;,
G/,<,

o;,,,

o/
5
51,

TV
Tmob

5,.
4
Tr:

6.0

v'
VC,,?

v,,'
E',
Q,

Qy

direction due to changes in vertical stress

( l .3.8)
Poisson's ratio for straining in the vertical
direction due to changes in horizontal stress (1.3.8)
Poisson's ratio for straining in the horizontal
direction due to changes in horizontal stress (I .3.8)
angle of dilation
(1.3.5)
angle of interface dilation
(5.5.6)
(1.3.10)
material density
settlement
(1.4.2)
relative bending stiffness
(2.7.1)
major, intermediate and minor principal stress (1.3.4)
direct stresses in X , y and z direction
(1.3.5)
axial total stress
(1.3.2)
axial effective stress
(1.3.2)
pre-consolidation pressure
(1.3.2)
cell pressure
(1.3.3)
radial total stress
(1.3.3)
radial effective stress
(1.3.3)
vertical total stress
(1.3.8)
(1.3.8)
circumferential total stress
normal stress
(1.3.6)
in-situ horizontal stress
(1.4.4)
total overburden pressure
(1.4.3)
maxim permissible compressive stress
across interface
(2.4.5)
(1.3.5)
shear stress
peak shear stress
(1.3.5)
residuaI shear stress
(1.3.5)
mobilised shear stress
(5.4.4)
shear stress in X-y plane
(1.3.5)
shear stress in z-8 plane
(1.3.8)
shear stress in z-r plane
(1.3.8)
shear stress in P-8 plane
(1.3.8)
angle of shearing resistance
( 1 -3.2)
critical state angle of shearing resistance
(1.4.2)
peak angle of shearing resistance
(1.3.3)
invariant deviatoric strain
(1.3.3)
(7.4.4)
pile group rotations about X and y axes

Index
A

Abbo, A.J., 200

Addenbrooke, T. I., 59,60,63,65,70,72
adjacent materials, 388
adjacent structures, 1 19-1 21
advancing tunnel heading, analyses, 346348,359-360
AGI, 253
AGS see Association of Geoenvironmental
and Geotechnical Specialists
alls see earth retaining structures
anchors
see also ground anchors
embedded walls, 115-1 16
reinforced earth walls, 97-98, 98, 103
Anderson, K.H., 327
angle of shearing resistance
oedometer test, 4
triaxial test, 8
true triaxial test, 11
anisotropic strength, effect on bearing
capacity, 243-248
anisotropy see soil anisotropy
approximations, problems, 364, 365
Apted, J.P., 149
Arthur, J.R.F., 20
Association of Geoenvironmental and
Geotechnical Specialists (AGS),
338
Aston Park multi-storey car park, 335
Atkinson, J.H., 156
axi-symmetric geometry, 43, 77
single piles, 282
B
backfill
compaction stresses, 94-95
cuttings, 96
reinforced earth walls, 96,98, 101
soil, 85
Bangkok Clay
parameters, 133, 133-134
pore water pressure profile, 390-391,
391
straining, 140
Bay-Gress, C., 232
beam, progressive failure, 141-145, 142,
144
bearing capacity

effect of anisotropic strength, 243-248

effect of footing shape, 223-227
factor, 233
failure, 256-257
non-homogeneous clay (undrained),
233-238
plate loading test, 32
pre-loaded strip foundations (undrained),
238-243
shallow foundations, 2 15-216
Bell Common tunnel, 104,109, 110,112
benchmarking, 332-363
definitions, 332-333
examples of problems, 342-357
advancing tunnel heading, 346-348
deep excavation, 355-357,362-363
idealised triaxial test, 343-344, 358
shallow waste disposal, 348-35 1,
360-361
simplified shallow waste disposal,
351-353,361-362
thick cylinder, 344-345, 358-359
tunnel construction, 353-355, 362
INTERCLAY I1 project, 340,34 1-342
non-standard benchmarks, 341
reasons for testing, 339-340
standard benchmarks, 340-341
bender element technique, 20-2 1
bending moments
pile groups, 3 12, 3 16-3 17
tunnel lining, 354,355
tunnel and wall construction, 92
berms
Carsington embankment, 193-194
Champlain clay embankment, 2 10
embedded wall support, 115
Bernat, S., 48
Binnie, G.M., 180
Bishop, A.W., 6
Bolton, M.D., 24,228,232,233
Booker, J.R., 233
bored tunnels, 38-72
Boscardin, M.D., 120
brittle behaviour, 141
BS 8006,75
bucket foundations, 3 17-329
analysis, 3 18-320
anisotropic study, 326-327

416 1 Finite element analysis in geotechniml engineering: Application

full skirt adhesion, 323-324
geometry, 3 18
installation, 3 17
interface modelling, 320-321, 320
isotropic study, 321-325
load displacement curves, 321
reduced skirt adhesion, 324-325
Snorre platform, 31 7
suction anchors, 327-329
Building Research Establishment (BRE),
175
building stiffness, influence on tunnelinduced settlement, 5 7 , 6 3 4 6
buildings, damage assessment procedure, 66
Burland, J.B., 10,70, 85, 86, 120, 151,255,
256,259,261,267,269,707
C
caissons, 281
see also bucket foundations
Calabresi, G., 261
Cam clay model (modified), 48, 132, 138
embankments, 172, 184, 198,203,208
lack of understanding, 365
Magnus soil profile, 304-305,305
soft Champlain clay parameters, 210
soft clay parameters, 322
tower stability, 259,276
Cambridge University, simple shear
apparatus, 14
Campanella, R.G., 28,29
cantilever wall see embedded cantilever
wall
car parks
Aston Park multi-storey, 335
House of Commons, 109, 110
Carder, D.R., 104
Carsington embankment, 189-194
core geometry and failure, 192-193
effect of berm on stability, 193-194
end-of-construction slip, 146
failure, 189, 191-192
finite element analyses, 191
material parameters and soil model, 190
original section, 191-192
Carswell, I., 235
Carter, J.P., 353
Champlain clay embankments, 206-21 1
MIT-E3 parameters, 209
modified Cam clay parameters, 210
Chandler, R.J., 125, 132, 160, 161

Chapeau, C., 207

Charles, J.A., 196
Chater Station, Hong Kong, 104, 106
Chow, F.C., 14
Christophersen, H.P., 3 17
circular footings
behaviour, 247-248
drained soil, 232-233
horizontal stress, 25 1-252
load-displacement curves, 230-23 1
weightless drained soil, 230-232
GIRIA see Construction Industry Research
and Information Association
clays
see also soft clay; stiff clay
coefficient of earth pressure at rest, 10
compression and swelling potential, 3-4
cone penetration test, 29-30
oedometer test results, 3-5
plate loading test, 34
residual strength, 14
standard penetration test, 23
stiffness test, 26-27
strength test, 26
Clough, G.W., 44
coefficient of earth pressure at rest, 4
effect on London Clay, 153-1 55
embedded walls, 107,107-108
pressuremeter test, 32
soft clay, 136
strip footings, 222
triaxial test, 10
cornpaction stresses
backfill, 94-95
rockfill dams, 173-1 74,174
compressibility, plate loading test, 32
compressible beam, progressive failure,
143-145,144
compression index, clays, 3
computer hardware, testing, 340
computer programs
causes of errors, 333,334-335
checking techniques, 339
consequences of errors, 335-336
developers, 336-337
responsibilities of users, 337-339
concrete piles, installation, 104
concrete walls, stiffness, 109
cone penetration test (CPT), 27-30
clays, 29-30
sand, 28-29
connections

Index 1 417
embedded walls, 113-1 14,114
pin-joint, 105, 106
plain strain analysis, 380-381, 381
support systems, 83, 84,84
tunnel linings, 49-52
constant head permeameter, 22
constitutive model
errors, 365
users' misunderstanding, 365
Construction Industry Research and
Information Association (CIRIA),
338
construction problems, 387-388
construction under water, cut slopes, 162163
convergence-confinementmethod, tunnel
excavation, 46
Coop, M.R., 20
Cording, E.J., 120
counterweight, Leaning Tower of Pisa,
267-27 1
coupled consolidation analyses, 380
Coyle, H.M., 290
CPT see cone penetration test
Crabb, G.I., 156
critical line, tower instability, 272-273, 274
critical state models, undrained analysis,
386387
cut and cover construction
see also earth retaining structures
George Green Tunnel, 91-92,92
cut slopes, 125-163
see also London Clay cut slopes
clay types, 126
construction under water, 162-163
drainage during excavation, 125-126
'non-softening' analyses, 126-140
progressive failure, 141-145
soft clay, 131-140
'softening' analyses, 145-1 63
stability, 125
stiff 'non-softening' clay, 127-131
cuttings
backfill, 96
collapse, 145-146
effect of widening, 158-160, 159, 160
slip, 162
cylinder see thick cylinder

D
Dale Dyke dam, 181-183
collapse, 181

hydraulic fracture, 180, 182

pore water pressures, 181
dams see rockfill dams
Davies, R.V., 104
Davis, E.H., 13,233,290
Day, R.A., 14,109, 11 1,223,368,370,371
De Moor, E.K., 105
de-watering, need for, 93
deep excavation, benchmarking exercise,
355-357,362-363
deep foundations, 280-329
bucket foundations, 28 1-282,3 17-329
caissons, 28 1
pile groups, 280-28 1,289-3 17
single piles, 280,281, 282-289
uses, 280
degrees of freedom, removal, 388-389
Desai, C.S., 368
design
Magnus platform foundations, 309-3 14
pile groups, 298-304
design failure, computer programs, 335-336
developers, computer programs, 336-337
deviatoric plane, yield and plastic potential
surfaces, 384-386
dilatancy, rockfill dams, 172
direct shear test, 12-14
results, 13-14
soil strength, 12
stress, 12
directional shear cell, 20
discretisation errors, 364, 365-368
Dounias, G.T., 171, 180, 189, 193, 194
drained soil
circular footings, 230-233
strip footings, 227-230
Duncan, J.M., 168, 170, 171

E
earth embankments, 185-1 94
Carsington embankment failure, 189-194
earthfill modelling, 186
London Clay road embankments, 186189
earth pressure at rest see coefficient of earth
pressure at rest
earth pressure balance (EPB) tunnelling, 40,
45
earth retaining structures, 74-121
construction and design, 74-75, 91-92
embedded walls, 76-77, 103-12 1
gravity walls, 75-76, 93-96

418 1Finite element analysis in geotechriical engineering: Application

'greenfield' conditions, 88-89
ground water control, 93
initial soil stresses (modified), 89-91, 90
model geometry, 79-82
model performance, 86-88
reinforced earth walls, 76,96-103
soil models, 85-88
structural component models, 84,85
support systems, 79,82-84,83
symmetry, 77-79
time related movements, 92-93
types, 75-77, 75, 76
earth walls see earth retaining structures
elastic bulk modulus, coefficients, 69-70
elastic expansion, pressuremeter testing,
30-31
elastic models, 171, 474, 478
elastic shear modulus, coefficients, 69
elasticity, soils, 9
elasto-plastic models
Mohr-Coulomb, 86, 146-147
rockfill dams, 171-173
'softening' analyses, 146-147
Elizabeth House, Waterloo, 120, 121
embankments, 166-21 1
behaviour, 166
construction and uses, 166
earth embankments, 185-194
rockfill dams, 167-185
soft clay embankments, 194-21 1
embedded cantilever wall, 86-87,86
analysis, 87
earth pressures, 8 7
element type, 107
embedded walls, 76-77, 103-121
active support systems, 114-1 15
adjacent structures, 119-121
analyses, 105-106
beam or shell elements, 107-109, 108
benchmarking exercise, 355-357, 362363
berms, 115
connection details, 113-1 14, 114
ground anchors, 115- 116
installation effects, 104-106
interface behaviour, 111
long term behaviour, 1 18-1 19
plain strain analysis, 376-377
relieving slabs, 116-1 17
support stiffness, 112-1 13
types and uses, 103,103
wall permeability, 11 1-1 12

wall stiffness, 109-1 1 1, 110, 111

environmental loading, Magnus platform,
314-317
EPB see earth pressure balance tunnelling
equation integration, errors, 362
errors in computer calculations
approximations, 364,365
causes, 334-335
consequences, 335-336
constitutive model, 365
discretisation, 364,365-368
equation integration, 364
European Commission, EUR 15285 report
(1993), 340,341,343, 346, 349,
352
Evans, J.D., 175
F
face loss, tunnelling, 42
facing membrane, reinforced earth walls,
96-97,97,100-102
failure
Carsington embankment, 189-1 94
compressible beam, 143-145, 144
computer programs, 333,334-336
multi-storey car park, 335
reinforced earth walls, 98-99, 99, 102
rigid beam, 141-143,142
Sleipner offshore platform, 336
trial wind turbine, 335-336
failure mechanisms, strip footing on clay,
236238,238
falling head permeameter, 22
finite element mesh
Dale Dyke dam, 181
earth retaining structures, 79-82
Leaning Tower of Pisa, 263,264
London Clay embankment, I88
London Clay slope, I50
Magnus pile, 308
reinforced earth wall, 100, 100
Roadford dam, 177
single pile analysis, 283,284, 287
smooth footing, 366,367
soIt clay embankments, 205,208
soft clay slopes, 136-1 37, 137
tunnelling, 59, 67
Finnesand, G., 317
flexible foundations, 21 8
footings see circular footings; strip footings
Fourier Series Aided finite element analysis
(FSAFEM), 43,44,2 14

bucket foundations, 3 19
Magnus platform, 3 10,3 12,315
pile groups, 290,291,297,298
single piles, 193, 282, 287, 291, 292
Frank, R., 374
Freiseder, M., 105
FSAFEM see Fourier Series Aided finite
element analysis

G
Ganendra, D., 290,29 1,292,294
gap method, tunnel excavation, 45-46
Gens, A., 374
geometry
axi-symmetric, 43, 77
earth-retaining structures, 79-82
tunnel, 91-92, 92
geophysical techniques, 20-21
George Green tunnel, 91-92,92, 109
Geotechnical Consulting Group, 57
geotextile
earth wall reinforcement, 101-102
facing membrane, 97
reinforcement, 200
Germaine, J.T., 20
German Society for Geotechnics
deep excavation exercise, 355-357
tunnel construction exercise, 353-355
Gouwenic, S., 105
granular soils, standard penetration test, 23
gravity walls, 75-76, 75,93-96
cuttings, 96
earth pressure (compaction), 94-95
finite element analysis, 95-96
interface elements, 95
limit states, 94
greenfieid conditions, earth retaining
structures, 88-89
greenfieid settlement, 65
Griffiths, D.V., 126
ground anchors, 83,115-1 16
plain strain analysis, 376, 378-3150
ground loss, tunnelling, 41-43
ground water control, earth retaining
structures, 93
Gun, M.J., 105
H
Hamza, M.M.A.F., 171
Hancock, R.J.R., 57
Head, K.H., 5
Henkel, D.J., 6, 53, 104, 149

Hermann, L.R., 374

Higgins, K.G., 105, 110, 120
Hight, D.W.,
2,53,110,113,120, 195,322
Hill mechanism, failure, 236
Hill, R., 236
hollow cylinder apparatus, 17,326
hollow cylinder test, 16-19
Poisson's ratio, 18-1 9
soil anisotropy, 16, 17, 18, 19
soil stiffness, 18
strains and stresses, 17
Horsnell, M.R., 3 13,3 14,3 17
House of Commons car park, 109, 110
Huder, J., 104
Hvorslev surface, 203,304,305
hydraulic boundary conditions, 54-55, 56
hydraulic fracture, puddle clay core dams,
180,182
hyperbolic model, rockfill dams, 170-171

I
ICFEP see Imperial College Finite Element
Program
ill-conditioning, interface elements, 370373
Imperial College Finite Element Program
(ICFEP), 282
impounding, rockfill dams, 178-179, 182183
in-situ tests, 23-35
limitations, 23
instability, reinforced earth walls, 98, 101
Institution of Civil Engineers, professional
responsibility, 338
interaction analysis (tunnels), 63-72
Treasury building case study, 66-70
twin tunnels, 70-72
INTERCLAY I1 project, 340,341-342
examples of problems, 342-35 1
objectives, 341
interface elements
bucket foundation, 320-32 1,320
reinforced earth walls, 96-97, 100-102
ring shear test, 16
single piles, 283-285
zero thickness
basic theory, 368-370
ill-conditioning, 370-373
numerical stability, 368-76
steep stress gradients, 373-376
IS0 9000 standard, 338
isotropy see soil isotropy

420 / Finite element analysis in geotechnical engineering: Application

Jackson, C., 239

Jamiolkowski, M., 20
Jardine, R.J., 17, 18, 19,31, 132, 140, 244,
291,293,304,306,307,310
Jefferies, R.M., 34 1,356
joints, tunnel linings, 49-52,51
Jonsrud, R., 3 17
Jovitit, V., 20
Jubilee Line see London Underground

K
K-G model, rockfill dams, 171
Kalra, J.C., 85, 86
Kenley, R.M., 3 16,3 17
Kim, M.K., 184
Kimmance, J.P., 71
Knights, M.C., 91
Knowles, N.C., 34 I, 356
'Kork' packing, 112
KovaceviC, N., 149, 171, 173, 174, 180
Kuwano, R., 20
Kuzuno, T., 40
L
La Rochelle, P,, 206
laboratory tests, 2-22
scope, 2
Lade, P.V., 173, 184
Lade's model, 85, 173, 175, 176, 179, 184
Lambe, T.W., 168
Lambeth Group Clay and Sands, 57,389
Lancellotta, R., 261
Lane, P.V., 126
Langley Point, Eastbourne, deep
excavation, 119
lateral loading, single piles, 287-289
Lau, C.K., 228,232,233
layered analysis, rockfill dams, 173-1 74
leaning instability, Pisa tower, 256-257
Leaning Tower of Pisa, 252-278
background, 252-253
construction and ground profile, 253254,254
counterweight application, 267-271
critical line, 272-273, 274
finite element analysis, 263-265
history of construction, 254-255
history of tilting, 255-256
model assessment, 277-278
motion of foundations, 256

permenent stabilisation, 27 1,276

simulation of inclination history, 265267
soil extraction and response, 271-276
soil parameters, 262
soil properties, 259-262
stability of tall towers, 256-259
Leca, E., 44
Leonards, G.A., 268
Leroueil, S., 132,207,245
linear elastic analysis, rockfill dams, 169
Lo Presti, D.C.F., 20
load distribution, pile groups, 294-298,
295,296
load vectors, pile groups, 299-302,300,
301-302
load-displacement curves, 215
bucket foundations, 321
circular footings, 230-23 1
pre-loading on stiff clay, 241
single piles, 283-287,284-285, 288,289
strip footings, 220,222,224-225,226227,228
London area, under-drained pore pressures,
53-54,389-390,390
London Clay
Jubilee Line tunnels, 59, 66,67, 67
permeability models, 56, 57
pore water pressure profile, 389-390,
390
stiff clay parameters, 235
swelling behaviour, 118, 118
tunnel linings permeability study, 57-59
underground tunnels, 62-63
London Clay cut slopes, 147-162
coefficient of earth pressure at rest, 153155
deviatoric plastic strain, 151, 158
displacement vectors, 150-1 5 1, 151
finite element analyses, 150-153, 151
pore water pressure, 151, 152, 153
properties, 128
rupture surfaces, 154-1 55, 154
slope geometry, 155-1 56,156,158-160,
159,160
soil parameters, 148-1 50
superficial slip, 157-158, 157, 158
surface boundary suction, 155, 155
surface cracking, 156-1 58
triaxial tests, 150
London Clay road embankments, 186-189
collapse with pore pressure, 189

Index 1 421
finite element analyses, 188
material properties, 187-188
London Underground
Bakerloo Line tunnels, 120, 121
Jubilee line tunnels, 59,61,66,67
tunnel distortion, 120
tunnels beneath parks, 59-60,61, 62,68
Westminster station construction, 89-91,
89, 120

M
Magnus platform, 304-3 17
analyses, 308-309
design of foundations, 309-3 14
environmental loading, 3 14-3 17
pile group displacements, 315
small strain stiffness parameters, 306
soil profile, 305
soil properties and initial conditions,
304-307
turning moments, 311,312,316
Mair, R.J., 30, 74, 195,235
man-made slopes see cut slopes
Marsland, A., 30
Masoswe, J., 10
MCC model, 209,210
mean effective stress, 5
membrane elements
modelling, 378,379
reinforced earth walls, 96-97, 100-102
rockfill dams, 167,180
Menard pressuremeter, 30
Menzies, B.K., 20
Mindlin beam elements, 100
Minister0 dei Lavori Pubblici, 260
MIT-E3 model, 246
anisotropic soil behaviour, 243,244
bucket foundation, 3 18, 326
Champlain clay parameters, 209
embankments, 197,207-208,2 10
material properties, 245
modelling
see also Cam clay model; Lade's model;
MIT-E3 model; Mohr-Coulomb
model
critical state models, 386-387
elastic models, 171,474,478
K-G model, 171
MCC model, 209,2 10
plain strain analysis, 376-381
soil models, 59-63,85-88,215-216
time dependent behaviour, 52-59

underdrainage, 389-394
Mohr-Coulomb model
elastic, 85
elasto-plastic, 86, 146-1 47
footings, 226
lack of understanding, 365
linear elastic, 199
piles, 285,286
plastic, 57,60
shallow waste disposal parameters, 360
soil stiffness, 80
undrained analysis, 365, 382-384
Mori, R.T., 168
motorway cuttings
effect of widening, 158-160, 159, 160
slip, 162
N
NAFEMS see National Agency for Finite
Element Methods and Standards
Nash, D.F.T., 261
National Agency for Finite Element
Methods and Standards
(NAFEMS), 338,340
standard benchmarks, 340-341
National Economic Development Office,
338
Naylor, D.J., 126, 168, 170, 171, 173, 174
NCE see New Civil Engineer
New Civil Engineer (NcE), 334,335,336,
338
Newton-Cotes integration scheme, 374375,376
Newton-Raphson technique, 214,282,343
Ng, C.W.W., 105
NGI see Norwegian Geotechnical Institute
non-homogeneous clay, undrained bearing
capacity, 233-238
'non-softening' analyses, clay, 127-13 1
Norwegian Geotechnical Institute (NGI),
simple shear apparatus, 14, 15
Nyren, R., 60

0
OCR see overconsolidation ratio
oedometer test, 3-6
Bangkok Clay prediction, 133, 134
clay soils, 3-4
rockfill dams, 167-168, 171,176
sands, 5-6
strain, 3
stress, 3

422 1 Finite element analysis in geotechnical engineering: Application

offshore platforms
design, 298
Magnus platform, 304-3 17
Sleipner platform failure, 336
Snorre platform, 31 7
oil platforms see offshore platforms
O'Neil, M.W., 289
Ong, H.L., 48
Ong, J.C.W., 48
open faced shield tunnelling, 40
overconsolidation ratio (OCR), 204, 2 4 6
242
Magnus soil profile, 307
soft clay, 136, 196, 197

P
Padfield, C.J., 74
Pagano, L., 186
Pancone clay, Pisa, 254,261,265,266,267,
268
peak strength, London Clay, 148
Peck, R.B., 24,34,75, 147
Penman, A.D.M., 169
Pepe, C., 261
permeability, 22
Bangkok Clay, 391
embedded walls, 11 1-1 12
London Clay, 149,389-390
modelling, 53, 55-57
profiles, 392-394
pumping tests, 35
triaxial test, 11
tunnel linings study, 57-59
void ratio, 56
permeameters, 22
Perry, J., 156
piggy-back tunnels, 71-72, 71, 72
pile groups, 280-281,281,289-3 17
analysis, 291
bending moments, 3 12,3 16-3 17
design, 298-304
displacements with depth, 293
load distribution, 294-298,295,296
load vector evaluation, 299-302,300,
301-302
Magnus platform, 304-3 17
matrix formulation, 298-299
rigid pile cap criterion, 294-295,297298
solution displacements and rotations,
302-304
superposition, 291-294,293,299-301

symmetry, 290
three-dimensional analysis, 289-291
turning moments, 3 1 1-3 13
piles, 280-281,281
see also bucket foundations; embedded
walls; pile groups
caissons, 28 1
plain strain analysis, 377-378
single piles, 282-289
interface elements, 283-285
lateral loading, 287-289
load displacement curves, 284-285,
284-287,288,289
vertical loading, 282-286
Pinto, 168
Pisa, Tower of see Leaning Tower of Pisa
pitching, tunnels, 42
plain strain analysis, 376-381
coupled analyses, 380
embedded walls, 106,376-377
ground anchors, 367,378-380
piles, 377-378
segmented tunnel linings, 381,382
structural connections, 380-38 1,381
plain strain geometry, 43
plastic expansion, pressuremeter testing,
3 1-32
plastic model, Mohr-Coulomb, 57,60
plastic potential surfaces, influence in
deviatoric plane, 384-386
plate loading test, 32-34,33
clay, 34
sand, 34
platforms see offshore platforms
Poisson's ratio, hollow cylinder test, 18-19
Polvani Commission (Pisa), 252,260, 261,
267,27 1,276
pore water pressures, 55-56,55,57
Bangkok Clay, 390491,391
clay slopes, 187
Dale Dyke dam, 181,182
London Clay, 151, 152, 153,389-390,
390
Ramsden dam, 183-185
tunnelling study, 57
under-drained, 53-54
~ o r o v iE.,
t 20,245
post shield loss, tunnelling, 42
Potts, D.M., 12, 14, 52, 63,65,70,72, 91,
1 1 1, 115, 146, 171, 172, 189, 223,
255,256,261,267,290,291,292,
293,294,304,3 10

Index / 423
Poulos, H.G., 289, 290
Prandtl, L., 226,236
Prandtl mechanism, failure, 236
pre-loaded strip foundations, undrained
bearing capacity, 238-243
pressuremeter testing, 30-32
coefficient of earth pressure at rest, 32
elastic expansion, 30-3 1
plastic expansion, 31-32
problems in analysis, 364-394
construction problems, 387-388
degrees of freedom removal, 388-389
dicretisation errors, 365-368
modelling underdrainage, 389-394
plain strain analysis, 376-381
undrained analysis, 382-384, 386-387
yield and plastic potential surfaces, 384386
zero thickness interface elements, 3683 76
progessive softening method, tunnel
excavation modelling, 46, 47
progressive failure
compressible beam, 143-145, 144
cut slopes, 141-145
rigid beam, 141-143, 142
props, 83,84,85
see also support systems
puddle clay core dams, 180-185
Dale Dyke dam, 181-183
hydraulic fracture, 180, 182
pumping tests, 35
Puzrin, A., 70

quality assurance procedures, 338

Quarterman, R.S., 30
R
railway cutting slopes, collapse, 145-146
Ramiah, B.K., 268
Ramsden dam, 183-1 85
drawdown displacements, 185
Randolph, M.F., 252,289
Reese, L.C., 290
reinforced earth walls, 76, 96-103
anchors, 97-98,98, 103
construction, 96-99,95, 97, 101
failure, 98-99, 99, 102
finite element analysis, 99-103
materials stability, lO@-101
membrane and interface elements, 96-

97,100-1 02
reinforcement materials, 101-1 02
soil nails, 98, 98, 103
stability, 100-101
reinforcement
interface properties. 201
soft clay embankments, 200-202
relieving slabs, 116-1 17
residual strength, London Clay, 148, 152
Ridden, J.W., 304, 308
rigid beam, progressive failure, 141-143,
142
rigid foundations, 21 8-2 19
rigid pile cap criterion, 294-295,297-298
ring shear test, 15-1 6
interface testing, 16
soil strength, 15
road tunnel
construction method, 91-92, 92
symmetry, 78, 78
Roadford dam, 175-1 80,175
comparison with observations, 179-180
construction, 178-179
double hardening model parameters, 178
finite element analyses, 177-1 79
material parameters, 175-1 76, 177
Robertson, P.K., 28,29
rockfill dams, 167-185
dilatancy, 172
elasto-plastic models, 171-1 73
hyperbolic model, 170-1 71
impervious earth core, 167
K-G model, 171
layered analysis, 173-1 74
linear elastic analysis, 169-1 70
non-linear elastic model, 171
old puddle clay core dams, 180-1 85
'power law' models, 169-1 70
Roadford dam analysis, 175-1 80, 175
stress paths, 167-168, 167, 168
upstream impervious membrane, 167
Roscoe, K.H., 259
Rowe, R.K., 45,46, 169
rupture surfaces, London Clay, 154-1 55,
1.54

S
St. Alban, Quebec, 206-207
St. John, H.D., 105, 112, 116
Sandroni, S.S., 148, 150
sands
cone penetration test, 28-29

422 1 Finite element analysis in geotechniwl engineering: Application

offshore platforms
design, 298
Magnus platform, 304-3 17
Sleipner platform failure, 336
Snorre platform, 317
oil platforms see offshore platforms
O'Neil, M.W., 289
Ong, H.L., 48
Ong, J.C.W., 48
open faced shield tunnelling, 40
overconsolidation ratio (OCR), 204,240242
Magnus soil profile, 307
soft clay, 136, 196, 197

P
Padfield, C.J., 74
Pagano, L., 186
Pancone clay, Pisa, 254, 261, 265,266,267,
268
peak strength, London Clay, 148
Peck, R.B., 24,34,75, 147
Penman, A.D.M., 169
Pepe, C., 261
permeability, 22
Bangkok Clay, 391
embedded walls, 1 1 1-1 12
London Clay, 149,389-390
modelling, 53, 55-57
profiles, 392-394
pumping tests, 35
triaxial test, 11
tunnel linings study, 57-59
void ratio, 56
permeameters, 22
Perry, J., 156
piggy-back tunnels, 71-72, 71, 72
pile groups, 280-28 1,281,289-3 17
analysis, 291
bending moments, 3 12,3 16-3 17
design, 298-304
displacements with depth, 293
load distribution, 294-298,295,296
load vector evaluation, 299-302,300,
301-302
Magnus platform, 304-3 17
matrix formulation, 298-299
rigid pile cap criterion, 294-295, 297298
solution displacements and rotations,
302-304
superposition, 291-294,293,299-301

symmetry, 290
three-dimensional analysis, 289-291
turning moments, 3 11-3 13
piles, 280-281,281
see also bucket foundations; embedded
walls; pile groups
caissons, 28 1
plain strain analysis, 377-378
single piles, 282-289
interface elements, 283-285
lateral loading, 287-289
load displacement curves, 284-285,
284-287,288,289
vertical loading, 282-286
Pinto, 168
Pisa, Tower of see Leaning Tower of Pisa
pitching, tunnels, 42
plain strain analysis, 376-381
coupled analyses, 380
embedded walls, 106,376-377
ground anchors, 367,378-380
piles, 377-378
segmented tunnel linings, 381,382
structural connections, 380-381,381
plain strain geometry, 43
plastic expansion, pressuremeter testing,
31-32
plastic model, Mohr-Coulomb, 57, 60
plastic potential surfaces, influence in
deviatoric plane, 384-386
plate loading test, 32-34,33
clay, 34
sand, 34
platforms see offshore platforms
Poisson's ratio, hollow cylinder test, 18-19
Polvani Commission (Pisa), 252,260,261,
267,27 1,276
pore water pressures, 55-56,55,57
Bangkok Clay, 390-391,391
clay slopes, 187
Dale Dyke dam, 181,182
London Clay, 151, 152, 153,389-390,
390
Ramsden dam, 183-1 85
tunnelling study, 57
under-drained, 53-54
~ o r o v iE.,
t 20,245
post shield loss, tunnelling, 42
Potts, D.M., 12, 14, 52,63,65,70,72, 91,
11 1, 115, 146, 171, 172, 189, 223,
255,256, 261, 267,290,291,292,
293,294,304,3 10

Index 1423

Poulos, H.G., 289,290

Prandtl, L., 226,236
Prandtl mechanism, failure, 236
pre-loaded strip foundations, undrained
bearing capacity, 238-243
pressuremeter testing, 30-32
coeMicient of earth pressure at rest, 32
elastic expansion, 30-3 1
plastic expansion, 3 1-32
problems in analysis, 364-394
construction problems, 387-388
degrees of freedom removal, 388-389
dicretisation errors, 365-368
modelling underdrainage, 389-394
plain strain analysis, 376-381
undrained analysis, 382-384,386-387
yield and plastic potential surfaces, 384386
zero thickness interface elements, 3683 76
progessive softening method, tunnel
excavation modelling, 46, 47
progressive failure
compressible beam, 143-145,144
cut slopes, 141-145
rigid beam, 141-143,142
props, 83, 84,85
see also support systems
puddle clay core dams, 180-185
Dale Dyke dam, 181-183
hydraulic fracture, 180, 182
pumping tests, 35
Puzrtn, A., 70

Q
quality assurance procedures, 338
Quarterman, R.S., 30
R
railway cutting slopes, collapse, 145-146
Ramiah, B.K., 268
Ramsden dam, 183-1 85
drawdown displacements, 185
Randolph, M.F., 252,289
Reese, L.C., 290
reinforced earth walls, 76, 96-103
anchors, 97-98,98,103
construction, 96-99,96,97, 101
failure, 98-99, 99, 102
finite element analysis, 99-103
materials stability, 100-1 01
membrane and interface elements, 96-

97,100-102
reinforcement materials, 101-1 02
soil nails, 98, 98, 103
stability, 100-101
reinforcement
interface properties. 201
soft clay embankments, 200-202
relieving slabs, 1 16-1 17
residual strength, London Clay, 148, 152
Ridden, J.W., 304, 308
rigid beam, progressive failure, 141-143,
142
rigid foundations, 2 18-2 19
rigid pile cap criterion, 294-295, 297-298
ring shear test, 15-1 6
interface testing, 16
soil strength, 15
road tunnel
construction method, 9 1-92, 92
symmetry, 78, 78
Roadford dam, 175-1 80, 175
comparison with observations, 179-180
construction, 178-1 79
double hardening model parameters, 178
finite element analyses, 177-179
material parameters, 175-1 76, 177
Robertson, P.K., 28, 29
rockfill dams, 167-185
dilatancy, 172
elasto-plastic models, 171-173
hyperbolic model, 170- 1 7 1
impervious earth core, 167
K-G model, 171
layered analysis, 173-1 74
linear elastic analysis, 169-1 70
non-linear elastic model, 171
old puddle clay core dams, 180-1 85
'power law' models, 169-1 70
Roadford dam analysis, 175-180, 175
stress paths, 167-168, 167, 168
upstream impervious membrane, 167
Roscoe, K.H., 259
Rowe, R.K., 45,46, 169
rupture surfaces, London Clay, 154-155,
154

S
St. Alban, Quebec, 206-207
St. John,H.D., 105, 112, 116
Sandroni, S.S., 148, 150
sands
cone penetration test, 28-29

424 1 Finite element analysis in geotechnical engineering: Application

oedometer test results, 5-6
plate loading test, 34
stiffness test, 25
strength test, 24-25
Sarrailh, J., 207
Schultheiss, P.J., 20
Schweiger, H.F., 105, 342, 353, 354, 355,
356
SCL see sprayed concrete lining method
Seed, R.B., 95
segmented tunnel linings, 49-52
joints, 49-52,51
moment-free joints, 49,50
plain strain analysis, 38 1, 382
Semple, R.M., 304, 308
settlement
rockfill dams, 174
Treasury building case study, 66-69
tunnel-induced, 6 3 4 6
shallow foundations, 2 14-278
see also surface foundations
foundation depth and undrained bearing
capacity, 248-252
Leaning Tower of Pisa case study, 252278
types, 215,215
shallow nuclear waste disposal
analysis, 348-35 1
simplified analysis, 351-353, 361-362
Sharp, D.E., 309,3 14, 3 17
shear box, direct shear test, 12-14
shear moduli, 21
shear stress-horizontal displacement
diagram, 13-14, 14
shearing resistance, angle of, 4, 8, l l
Shibuya, S., 133
shield loss, tunnelling, 42
Shin, D.M., 43, 52
side-by-side tunnels, 7 1-72, 71
Sieffert, J.G., 232
simple shear test, 14-1 5
Cambridge University device, 15
NGI device, 14, 15
soil strength, 15
Skempton, A.W., 24, 53, 146, 148, 149,
153, 161, 189,248,249,252
Skinner, A.E., 169
Sleipner offshore platform, failure, 336
Sloan, S.W., 200
slope see cut slopes
slope geometry, effect on London Clay,
155-156, 156, 158-160,159,160

small strain stiffness parameters, 81

Magnus platform, 306
Snorre platform, 31 7
soft clay, 131-1 40
anisotropy, 197
Cam clay model, 132, 138
critical slip surface, 138, 139
drainage in excavation, 13 1-132, 137138
finite element analyses, 136-140
permeability model, 136
properties, 197,203
soil parameters, 132-136
soil properties, 135
straining after excavation, 140, 140
strength profile, 196-1 97, 196
tunnel modelling problems, 48
undrained shear strength, 134-135
undrained strength parameters, 235
soft clay embankments, 194-21 1
Champlain clay, 206-21 1
constitutive model, 196-1 98
construction and uses, 194-195
displacement, 200
effect of anisotropic soil behaviour, 20621 1
effect of reinforcement, 200-202
effect of a surface crust, 198-200
embankment fill properties, 199
finite element analysis, 199-200, 204205,207-208
soil conditions, 198-199, 20 1,207
soil reinforcement modelling, 198
staged construction approach, 202-205,
203,205
typical soil conditions, 195-196
soft ground tunnelling, 38-72
'softening' analyses, 145-163
cut slopes in London Clay, 147-162
elasto-plastic model, 146-147
implications for convergence, 147
softening rate, London Clay, 149-150
soil, categories and uses, 85-88
soil anisotropy
bucket foundations, 326-327
directional shear cell, 20
hollow cylinder test, 16, 17, 18, 19
soft clay embankments, 206-21 1
soil elasticity, triaxial test, 9
soil extraction, Leaning Tower of Pisa,
271-275
soil isotropy, bucket foundations, 321-325

&
-

"r

r u * & a ; r a u i 4 -

".--&wAuA

Index 1 425

soil models, 59-63, 85-88

parametric study, 59-60
shallow foundations, 2 15-2 16
surface settlement prediction, 60-63
soil nails, 76, 98, 98, 103
soil profile, Magnus platform, 305
soil properties
Magnus platform, 304-307
stiff clay, 135
strip footings, 234
soil reinforcement, soft clay embankments,
198
soil stiffness
directional shear cell, 20
earth retaining structures, 79-80
foundation type, 280
hollow cylinder test, 18
stability of tall towers, 257-259,258,
259
triaxial test, 8-9,8
true triaxial test, 11-12
soil strength
direct shear test, 12
directional shear cell, 20
ring shear test, 15
simple shear test, 15
sprayed concrete lining (SCL) method, 41,
41.52
springs, modelling, 378,379
SPT see standard penetration test
stabilisation, Leaning Tower of Pisa, 271
stability
cut slopes, 125
reinforced earth walls, 100-101
tall towers, 256-259
standard penetration test (SPT), 23-27
standards
BS 8006,75
I S 0 9000,338
National Agency for Finite Element
Methods and Standards
(NAFEMS), 338,340
Standing, J.R., 66
steel sheet piles, installation, 104
stiff clay
soil properties, 135
undrained strength parameters, 235
stiff 'non-softening' clay, 127-13 1
analyses, 127-1 3 1
prediction of rupture, 128-113 1
soil parameters, 127
stiffness

buildings, 57, 63-66

London Clay, 148-149
walls, 809-1 1 1
stiffness test
clays, 26-27
sands, 25
storm loading, Magnus platform, 3 14-3 17
strength
soft clay, 196
stiff clay, 187
strength test
clays, 26
sands, 24-25
stress gradients, zero thickness interface
elements, 373-376
stress level, definition, 123-1 24
stress paths, rockfill dams, 167-168
strip footings
behaviour, 246-247
coefficient of earth pressure at rest, 222
drained soil, 227-230
failure, 22 1
footing shape and bearing capacity, 223227
load-displacement curves, 220,222,
224-225,226227,228
soil behaviour, 244-246
soil properties, 234
undrained clay, 21P-223
Stroud, M.A., 25,26, 27
structural connections, plain strain analysis,
386381,381
suction anchors, 327-329
suction caissons see bucket foundations
superposition, pile groups, 291-294,293,
299-301
support systems
connections, 83, 84,84
earth retaining structures, 79, 82-84, 8 3
embedded walls, 112-1 17
propped retaining wall, 79
surface boundary suction, London Clay,
155,155
surface cracking, London Clay, 156-1 58
surface crust, effect on soft clay
embankments, 198-200
surface foundations
analysis, 216-217
anisotropic strength and bearing
capacity, 243-248
bearing capacity, 2 15-2 16
choice of soil model, 215-216

426 1 Finite element analysis in geotechnic:a1 engineering: Application

circular footings, 230-233
flexible foundations, 2 18
footing shape and bearing capacity, 223227
rigid foundations, 2 18
strip footings, 2 19-223, 227-230
undrained bearing capacity of nonhomogeneous clay, 233-238
undrained bearing capacity of pre-loaded
strip foundations, 238-243
vertical loading, discretisation errors,
365-368
surface settlement
deep excavation, 355,356
tunnels, 60-63, 354
walls, 80-81
swelling index, clays, 3
swelling line, 3 , 4
Swoboda, G., 46
symmetry
earth retaining structures, 77-79
road tunnel, 78
Symons, I.F., 104

T
tall towers, stability, 256-259
Tavenas, F.A., 132,207
TBM see tunnel boring machines
Tedd, P., 104, 180, 184
Terrace Gravels, 66
Terzhagi, K., 24,34, 147
Thames Gravel, 57,59,389
thick cylinder, analysis, 344345,358-359
time dependent behaviour modelling, 52-59
hydraulic boundary conditions, 54-55
initial conditions, 52-54
permeability, 55-57
tunnel linings, 57-59
under-drained pore pressures, 53-54
time related movements, earth retaining
structures, 92-93
towers, stability, 256-259
Treasury building
settlement trough, 68
tunnel-induced response, 66-70
triaxial test, 6-1 1 , 6
angle of shearing resistance, 8
Bangkok Clay, 133
compressions and swelling, 7-8
idealised analyses, 343-344,358
London Clay, 150
permeability, 1 1

pore pressure, 7
rockfill dams, 168, 171,176
small strain stiffness, 60
soil elasticity, 9
soil stiffness, 8-9, 9
stress-strain diagrams, 8
true triaxial test, 11-1 2
angle of shearing resistance, 11
soil stiffness, 11-12
tunnel, 38-72
see also London Underground; road
tunnel
building stiffness and settlement, 57,6366,64
construction method, 91
modified soil stresses, 89-90, 90
numerical methods, 38-39
parametric study, 57-59
pitching, 42
time dependent behaviour modelling,
52-59
twin tunnel interaction, 70-72
tunnel boring machines (TBM), 40,41
tunnel construction, 39-43
benchmarking exercise, 346-348, 353355,359-360,362
ground response, 4 1-43
interaction analyses, 63-72
manual methods, 39
material parameters, 362
mechanical methods, 40-43
road tunnels, 91-92, 92
shielding, 40
sprayed concrete lining method, 41
support, 3 W O
tunnel construction simulation, 43-52
advancing tunnel heading, 346-348,
359-360
benchmarking exercise, 346-348
choice of soil model, 59-63
FSAFEM, 43,44
important boundary conditions, 45
initial conditions, 44-45
two dimensional modelling, 43
tunnel excavation modelling, 44, 45-48
convergence-confinement method, 46
gap method, 45-46
progessive softening method, 46, 47
soft clay problems, 48
volume loss control method, 46-48,57
tunnel lining modelling, 48-52
bending moments, 354,355

Index 1427
joints, 49-52, 51
moment-free joints, 49,50
permeability study, 57-59
segment connections, 49-52
shell elements, 49, 50
solid elements, 49
s,prayed concrete linings, 52
turning moments
Magnus platform, 311, 312,316
pile groups, 3 1 1-3 13
twin tunnels, interaction, 70-72, 71

U
UCL see University College
UDS see User Defined Subroutines
under-extraction, Leaning Tower of Pisa,
271-275
underdrainage
London area, 53-54,389-390,390
modelling, 389-394
underground tunnels see London
Underground
underpass, pin-joint connection, 105,106
undrained analysis
critical state models, 386-387
Mohr-Coulomb model, 365,382-384
undrained bearing capacity see bearing
capacity
undrained clay
strip footings, 219-223
tower instability, 257
University College (UCL), directional shear
cell, 20
User Defined Subroutines (UDS), 337
users
access to computer codes, 337-338
computer program responsibilities, 338339
constitutive model misunderstanding,
365

v
Vaughan, P.R., 56, 127,146, 184, 187, 194

vertical loading
shallow foundations, 2 19-233
single piles, 282-286
surface foundations, 2 19-233
Vesic, A.S., 25
Viggiani, G., 20
virgin compression line, 4, 5
virgin consolidation line, 5
void ratio, 56
volume loss control method (tunnel
excavation), 46-48,57
Treasury building case study, 68
W
Walbancke, H.J., 146, 187
wall see earth retaining structures
wall stiffness, embedded walls, 109-l l l
wall supports see support systems
Ward, W.H., 71
waste disposal see shallow nuclear waste
disposal
Waterloo International Terminal, 120-1 2 1,
120,121
Watts, K.S., 176
Wesley, L.D., 6
Westminster underground station, site plan,
89-90,89, 120
Wilson, A.C., 175
Wilson, E.L., 368, 370
wind turbine, failure, 335-336
Wood, D.M., 32

Y
yield surfaces, influence in deviatoric plane,
384-386
Young, W.C., 371

z
Zdravkovii, L., 17, 18, 19, 206, 239, 240,
244,318
zero thickness interface elements, 368-376
Zienkiewicz, O.C., 374