NUMERICAL ANALYSES OF ROCKFILL DAMS,

CUT SLOPES AND ROAD EMBANKMENTS

by

NEBOJSA KOVACEVIC
BSc, MSc, DIC

July 1994

A thesis submitted to the University of London

(Imperial College of Science, Technology and Medicine)
in partial fulfilment of the requirements for the degree of
Doctor of Philosophy in the Faculty of Engineering

BIBL
LONDON
UNIV ,j
 ABSTRACT

Large rockfill embankment dams are complex geotechnical structures and are often analyzed
using the finite element method. 'Variable elastic' type constitutive models are extensively
used to represent their behaviour. Problems have been experienced with this approach, both
in deriving model parameters from available test data and in producing predicted displacement
patterns which agree with field measurements.

In this thesis two elasto-plastic constitutive models, originally proposed by Lade, have been
incorporated into the Imperial College Finite Element Program (ICFEP). The behaviour of two
rockfill dams recently constructed in the UK have been investigated. For comparison purposes
analyses using a 'variable elastic' constitutive model have also been completed. It is shown
that while the elasto-plastic models are capable of reproducing the full range of available
rockfill test data for these dams this is not possible with 'variable elastic' model. Comparison
of predicted and observed behaviour of the two dams also highlights the superior performance
of the elasto-plastic models.

Many UK road cuttings and embankments constructed in plastic clay strata have suffered
delayed superficial sliding. Railway cuttings in similar strata have developed deep-seated slides
up to 100 years after construction. In both cases slides have occurred with an average
operational strength significantly 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 enhancements to ICFEP enable both strain-softening and coupled consolidation

(swelling) to be included in an analysis. Analyses of railway cuttings reproduce r the history of
deep-seated slides observed in the field and quantify the substantial effect of progressive
failure. Similar analyses of road embankment slopes show that shallow slides can be caused
by differential strains and progressive failure at the base of a superficial layer of clay which
has higher permeability due to shrinkage cracks. Delayed deep-seated slides may well occur
in the future in higher existing road cutting slopes where foundation drainage is poor. When
shallow slips are repaired, drainage to prevent possible subsequent deep-seated slides should
be considered.
 ACKNOWLEDGEMENTS

The research work presented in this thesis has been carried out under the joint supervision
of Prof P.R. Vaughan and Dr D.M. Potts. They have been a constant source of ideas, but
without Dr Potts' expertise and willingness to help in clarifying and transforming these ideas
into his remarkable finite element program, ICFEP, this research would not have been
possible. Their guidance, encouragement, and support (not only technical) is greatly
appreciated.

The following organisations provided financial support during the course of this research:
- Energoprojekt Company, Belgrade, Yugoslavia
- The Committee of Vice-Chancellors, UK, through the Overseas Research Student
Award
- The Scholarships Committee of the University of London, UK, through the Edmund
Davis Scholarship
- Geotechnical Consulting Group Ltd, London, UK
Their support is gratefully acknowledged.

I would like to express my gratitude to the staff of the Soil Mechanics Section of the Civil
Engineering Department, Imperial College, London for their help and support. Prof A.W.
Skempton's interest in part of this work is fully appreciated.

I would also like to thank to my research colleagues in the section who made my stay at
Imperial College so enjoyable and memorable.

Finally, my deepest gratitude goes to my parents for their support and care throughout my
life, and to my wife Zora for her patience and understanding.

II
 TABLE OF CONTENTS

ABSTRACT .
ACKNOWLEDGEMENTS............................................
CONTENTS.....................................................

CHAPTER 1 INTRODUCTION
1.1 GENERAL ..............................................1
1.2THESISLAYOUT .........................................4

CHAPTER 2 FINITE ELEMENT ANALYSIS

2.1 INTRODUCTION ..........................................6
2.2 FORMULATION OF FINITE ELEMENT METHOD ....................6
2.2.1 Element Discretization ...............................7
2.2.2 Displacement Approximation ..........................7
2.2.3 Element Equations .................................8
2.2.4 Global Equations ..................................8
2.2.5 Boundary Conditions ................................9
2.2.6 Numerical Integration ..............................10
2.2.7 Solution of Global Equilibrium Equations .................10
2.2.8 Interpretation of Results ............................10
2.3 NON-LINEAR ANALYSIS ...................................11
2.4 GEOTECHNICAL CONSIDERATIONS ...........................13
2.4.1 Initial Stresses ...................................13
2.4.2 Excavation ......................................14
2.4.3 Construction ....................................14
2.4.4 Pore Pressures ...................................15
2.5 COUPLED CONSOLIDATION ................................16
2.5.1 Formulation .....................................16
2.5.2 Time Integration ..................................19
2.5.3 Validation ......................................20

CHAPTER 3 NUMERICAL MODELLING OF EMBANKMENT DAMS

3.1 INTRODUCTION .........................................27
3.2 LAYERED ANALYSIS .....................................27
3.3 STRESS PATHS .........................................29
3.4 CONSTITUTIVE MODELS ..................................31

III
 3 .4.1 Linear Elastic Analysis ..............................32
3 .4.2 Non-linear Elastic Models ............................34
3 .4.2.1 'Power Law' Models ........................34
3 .4.2.2 Hyperbolic Model ..........................36
3.4.2.3 'KB-G' Model .............................38
3 .4.3 Elasto-plastic Models ..............................39
3.5 CONCLUSIONS ......................................... 42

CHAPTER 4 LADE'S ELASTO-PLASTIC MODELS

4 .1 INTRODUCTION .........................................46
4 .2 FORMULATION OF MODELS ................................46
4.2.1 Elastic Behaviour .................................47
4.2.2 Failure Criterion ..................................48
4.2.3 Double Hardening Model ............................49
4.2.4 Single Hardening Model .............................52
4 .3 IMPLEMENTATION OF MODELS ............................. 54
4.3.1 Single Hardening Model .............................55
4 .3.2 Double Hardening Model ............................58
4.4 VALIDATION OF MODELS ..................................63
4 .5 COMPARISON OF MODELS ................................. 64

CHAPTER 5 NUMERICAL ANALYSIS OF ROADFORD DAM

5.1 INTRODUCTION ......................................... 73
5.2 DETERMINATION OF MODEL PARAMETERS .....................74
5.2.1 Non-linear Elastic Perfectly Plastic Model .................74
5.2.2 Lade's Double Hardening Model .......................76
5 .2.2.1 Elastic Parameters ..........................76
5 .2.2.2 Failure Parameters ..........................76
5.2.2.3 Plastic 'Collapse' Parameters ..................77
5.2.2.4 Plastic 'Expansive' Parameters ................. 77
5.2.2.5 Summary of Double Hardening Model Parameters . . . . 79

5.2.3 Lade's Single Hardening Model ........................79

5.2.3.1 Elastic and Failure Parameters ..................79
5.2.3.2 Work-hardening Parameters ................... 80
5.2.3.3 Yield Parameters ........................... 80
5.2.3.4 Plastic Potential Parameters ...................81
5.2.3.5 Summary of Single Hardening Model Parameters .....81
5.3 MODEL PREDICTION FOR IDEALIZED, SINGLE ELEMENT TESTS .......82
5.4 FINITE ELEMENT ANALYSIS ................................ 85

iv
 5.4.1 Analysis During Construction . 86

5.4.2 Analysis During Impounding ..........................88

5.4.3 'Improved' Predictions ..............................89

5.5 CONCLUSIONS ......................................... 92

CHAPTER 6 NUMERICAL ANALYSIS OF WINSCAR DAM

6.1 INTRODUCTION ........................................ 134

6.2 DETERMINATION OF MODEL PARAMETERS ....................135

6.3 FINITE ELEMENT ANALYSIS ...............................135

6.4 CONCLUSIONS ........................................ 138

CHAPTER 7 FIRST-TIME SLIDES IN STIFF PLASTIC CLAYS

7.1 INTRODUCTION ........................................ 150

7.2 DELAYED SLIDES AND PROCESSES INVOLVED .................. 151

7 .2.1 Swelling ......................................151

7.2.2 Softening .....................................1 53
7.2.3 Progressive Failure ...............................1 55

7.3 METHODS OF ANALYSIS .................................157

7.3.1 Available Methods of Analysis .......................157

7.3.1.1 Limit Equilibrium Analysis ................... 1 158

7.3.1.2 Purpose Built Model .......................158

7.3.1 .3 Finite Element Analysis ..................... 158

7.3.2 Soil Model Used in the Thesis .......................1 60

7.4 RELEVANT AND ASSUMED SOIL PROPERTIES .................. 163

7 .4.1 Cut Slopes .....................................164

7.4.1.1 Peak Strength ............................ 164

7.4.1.2 Residual Strength .........................165

7 .4.1.3 Brittleness ..............................165

7.4.1.4 Pre-peak Stiffness .........................166

7.4.1.5 Coefficient of Earth Pressure at Rest ............ 166

7.4.1.6 Undrained Strength Profile ................... 167

7.4.1.7 Coefficient of Permeability ...................167

7.4.2 Fill Slopes ..................................... 169

7.4.2.1 Peak Strength ............................ 169

7.4.2.2 Residual Strength .........................170

7.4.2.3 Brittleness .............................. 170

7.4.2.4 Pro-peak Stiffness .........................170

7.4.2.5 Undrained Behaviour .......................171

7.4.2.6 Initial stresses due to Compaction ..............172

7.4.2.7 Coefficient of Permeability ...................172

7.5 HYDRAULIC BOUNDARY CONDITIONS AND SURFACE EFFECTS ......173

CHAPTER 8 ANALYSIS OF CUT SLOPES

8.1 INTRODUCTION ........................................187
8.2 TYPICAL ANALYSIS .....................................188
8.3 PARAMETRIC STUDIES ..................................192
8.3.1 Effect of Residual Strength .........................192
8.3.2 Effect of Angle of Dilation ..........................193
8.3.3 Effect of Coefficient of Earth Pressure at Rest ............194
8.3.4 Effect of Rate of Softening ..........................197
8.3.5 Effect of Surface Boundary Suction ....................197
8.3.6 Effect of Slope Geometry ...........................199
8.3.6.1 Flatter and Deeper Slopes ...................199
8.3.6.2 Steeper and Lower Slopes ...................200
8.3.7 Effect of Dra nage at Base Boundary ...................201
8.4 DETERIORATION OF SURFACE CLAY .........................202
8.5 FURTHER DISCUSSION AND CONCLUSIONS ....................204

CHAPTER 9 ANALYSIS OF ROAD EMBANKMENTS

9.1 INTRODUCTION ........................................254
9.2 EMBANKMENT ON CLAY FOUNDATION .......................255
9.3 EMBANKMENT ON DRAINED FOUNDATION

WITH AN IMPERMEABLE SURFACE BOUNDARY 257
258
9.4 EMBANKMENT ON DRAINED FOUNDATION ......
9.5 DISCUSSION AND CONCLUSIONS ............. 259

CHAPTER 10 CONCLUSIONS AND RECOMMENDATIONS

10.1 INTRODUCTION .......................................283
10.2 CONCLUSIONS .......................................283
10.2.1 Rockfill Dams ..................................283
10.2.2 Cut Slopes ....................................285
10.2.3 Road Embankments .............................287
10.3 SUGGESTIONS FOR FURTHER RESEARCH ....................288
10.3.1 Rockfill Dams ..................................288
10.3.2 Cut Slopes and Road Embankments ..................288

REFERENCES..................................................290
APPENDIX....................................................309

vi
 CHAPTER 1
INTRODUCTION

1.1 GENERAL

The origin of rockfill dams can be traced back to California in about the middle of the 1 9th
century. Since then this type of dam has had a long and turbulent history (Cooke, 1984). Two
main types of rockfill dams have emerged, according to the nature and position of the
watertight element, namely: (i) that containing a relatively impervious internal earth core,
either thick or thin, central or inclined, and (ii) that with an upstream impervious membrane,
either of concrete or asphalt. Nowadays rockfill dams are the most numerous and the most
popular choice when constructing a new dam 1 (Penman, 1986). In the search for
hydroelectric power, heights and volumes of the rockfill dams have been increasing, reaching
unprecedented heights (over 300m) in recent years.

In the beginning their design and construction was largely empirical and based on past
experience rather than on theory. With no intention of diminishing the important role of
experience, the advances made in the theory of soil mechanics started to change their
conception during the 1940's. In the past 30 years the major breakthrough in understanding
stress-strain-strength relations, particularly for rockfills, together with the development of
advanced numerical analyses has made it feasible to compute stresses, strains and
displacements in dams during construction, first impounding and subsequent operation. Today
finite element analysis using adequate constitutive laws is an essential tool to attain a good
and safe design of a rockfill dam.

Simulation of the rockfill behaviour in the field by laboratory tests is difficult and sometimes
impossible. This arises due to the coarse nature of rockfill materials. Laboratory tests require
large samples if representative behaviour is to be reproduced. The associated testing facilities
needed to incorporate such samples must also be large, and consequently such tests are
expansive and their number is usually limited. Because of this and the fact that a significant
factor of safety usually operates in rockfill dams, the stress-strn behaviour of rockfill
materials pre-peak has generally been modelled by a form of so-called the 'variable elasticity'

'If foundations allow and rockfill is available.

1
constitutive model in which elastic moduli change according to the magnitudes of the stresses
(and strains). Problems have been experienced with this approach, both in deriving model
parameters from available test data and predicting displacement patterns which agree with
field measurements. It has been suspected that part of the problem lies in the failure to model
the plastic behaviour of rockfiU materials pre-peak. Alternative constitutive models
incorporating pre-peak plasticity are available in the literature, but only those based on the
critical state concept have so far been used for the finite element analysis of rockfill dams.

The first objective of this thesis was to see how good these alternative elasto-plastic models
are at predicting the displacement patterns of existing rockfill dams using parameters derived
from available laboratory test data. Bearing in mind that modelling of an earth core can create
a source of uncertainty in an analysis, only rockfill dams with upstream asphaltic membrane2
have been considered. For comparison purposes analyses using the models based on elasticity
theory are also included.

The delayed sliding of slopes cut in stiff plastic clays has been a subject of considerable soil
mechanics interest for a number of years. In Britain, studies of the problem have focused on
London Clay for which there is a well established history of delayed deep-seated sliding in
railway cuttings formed in 19th and early 20th centuries (see e.g. Skempton, 1977). More
recently a major survey conducted by the Transport and Road Research Laboratory (Perry,
1989) revealed that both motorway embankments and cuttings in similar clay strata have
suffered delayed superficial sliding.

Deep-seated slips in cutting slopes have been studied extensively. It was recognised that the
very slow rate of equilibration of depressed pore pressures following excavation played an
important part in the explanation of delayed sliding (Vaughan and Walbancke, 1973).
However, analyses of the slips by conventional limit equilibrium methods have shown that the
average operational drained strength is significantly less than the peak strength as measured
in the laboratory (Skempton, 1964). Two explanations have often been offered for this
behaviour. The first is a reduction of the strength of stiff plastic clays to a 'fully softened'
value (Skempton, 1970), sometimes attributed to softening on fissures (Terzaghi, 1936). The
fully softened strength is often assumed to be that at critical state (Schofield and Wroth,
1968). This has lead to an empirical method of design in which the drained strength envelope
measured in the laboratory is adjusted by reducing the cohesion intercept to a nearly zero
value (Chandler and Skempton, 1974).

2 An upstream membrane of asphaltic concrete is thin and flexble, and has a little bearing
on rockfill deformation during first reservoir impounding and subsequent operation.

2
The second explanation involves progressive failure. When loaded, brittle soils generally reach
a peak strength which reduces to a residual strength with further straining. At rupture
involving progressive failure, strains along the f nal rupture surface are such that peak
strength is not mobilised uniformly. At collapse, some of the soil has already 'failed' and its
strength has dropped towards the residual strength, some of the soil has not yet reached peak
strength, and only a small part of the soil along the rupture surface is at peak strength. Thus,
the average operational strength mobilised along the eventual rupture surface at collapse is
less than peak strength, although it must be greater than residual strength.

Progressive failure has been suspected to play a part in slope failure for 25 years (Bjerrum,
1967; Turnbull and Hvorstev, 1967; Peck, 1967; Bishop 1967). The mechanism is complex
and analysis is difficult. It has been analyzed for simple idealised situations by the finite
element method at a surprisingly early date (Hoeg, 1972). Recent improvements in numerical
techniques arid computing power now enable such analyses to be performed in realistic
situations, as demonstrated by the analysis of the collapse of the Carsington embankment
(Potts eta!, 1990). These and other analyses (Dounias eta!, 1988, 1989) have shown that
the role of progressive failure can be significant. Thus progressive failure is the likely principal
cause of a discrepancy between the operational strength on the rupture surface of a slip in
a clay slope and the peak strength measured in the laboratory.

The stability of clay embankments has been studied less extensively. Construction of an
embankment typically generates excess pore water pressures in a saturated clay foundation,
and as these dissipate with time stability improves (Bishop and Bjerrum, 1960). However,
placement and compaction of plastic clays as fill can generate quite high pore water suctions
(Vaughan et a!, 1978), and in embankments of modest height, typical for road works, long-
term equilibrium pore water pressures are likely to be higher than those at the end of
construction. As swelling occurs there is potential for the development of deep-seated
delayed slides in road embankments of plastic clays similar to those occurring in cutting
slopes.

The behaviour of clay slopes is more complicated at shallow depth, where there are seasonal
changes in soil suction and cyclic shrinkage and swelling. This may lead to cracking, increased
permeability, softening and loss of strength in addition to that due to monotonic swelling. The
depth of the slips so far observed in motorway slopes is typically close to or within the zone
influenced by these effects.

The purpose of the research program described in the second part of this thesis was to review
the history of slides in both railway cutting and road embankment slopes formed in typical
British plastic clays, and to examine the mechanisms involved using the finite element

3
techniques developed for the investigation of the Carsington collapse. However, in the present
work improved simulation of a delayed collapse was modelled using coupled consolidation
(swelling). As there is a better history of deep-seated delayed sliding for cutting slopes they
were analyzed first, in order to calibrate the method of analysis against field experience.
Particular aims were to: (i) review relevant material properties, and hydraulic boundary
conditions, (ii) quantify the role of progressive failure in reducing the average operational
strength of in-situ stiff plastic clays and clay fills, derived from them, at collapse for both
railway and motorway slopes, (iii) examine the potential for the development of deeper
seated, longer term slides in road embankment slopes, including those where shallow slides
have already occurred and been repaired, and to reconcile the conclusions drawn with the
history of failure in railway cutting slopes, (iii) indicate how the trends determined by the
analyses might be monitored and controlled in the field. Representative slopes were analyzed
on a parametric basis to examine these questions.

1.2 THESIS LAYOUT

The work presented in this thesis is divided into the following chapters:

Chapter 2 briefly describes the finite element method of analysis which has been used
throughout the thesis. Some attention is given to those aspects of the method which are of
interest to the research work presented here, and, in particular, manner in which the Imperial
College Finite Element Program (ICFEP) deals with them.

Chapter 3 discusses some factors of importance in the finite element analysis of embankment
dams (layered analysis, stiffness of simulated layer and compaction stresses), considers
typical stress paths occurring during dam construction and reservoir impounding, and reviews
the constitutive models that have been used to describe the behaviour of fill materials.

Chapter 4 describes two of Lade's elasto-p astic constitutive models that have been used here
to characterize the behaviour of rockfills. The way they are implemented into ICFEP is
outlined, their proper implementation is validated, and a preliminary appraisal of their
capabilities to accurately capture the observed behaviour of granular material for various
stress paths is given.

Chapter 5 presents results from analyses of Roadford dam both during its construction and
first reservoir impounding using three dfferent constitutive models. One of these was a
'simple' non-linear elastic perfectly plastic model. The others were 'complex' Lade's work-
hardening (-softening) elasto-plastic models. Large diameter oedometer and triaxial tests for
Roadford rockfill were available, and the procedures employed for determnation of the model

4
parameters are presented in some detail. The models are compared in terms of their capability

of reproducing the full range of the observed rockfill behaviour in the laboratory and of

predicting displacement patterns in agreement with field observations.

Chapter 6 further checks the capabilities of both Lade's models by using them to analyze
Winscar dam. For this rockfill dam adequate laboratory test data to derive model parameters

were also available. Observed movements during dam construction and reservoir filling and

predictions of these movements based on elastic finite element analysis enabled useful

comparisons to be made.

Chapter 7 deals with the processes involved in delayed sI des of cut and embankment slopes

made in or of stiff plastic clays. It discusses the mechanism of progressive failure and the

various parameters affecting it. It briefly reviews the available methods of analysis for

progressive failure and outlines the soil model used during the course of this research. Finally,

relevant and assumed soil properties are presented together with a discussion of hydraulic

boundary conditions and surface effects.

Chapter 8 presents a parametric study of progressive failure in cut slopes. Effects of residual

strength, angle of dilation, coefficient of earth pressure at rest, rate of softening, surface

boundary pore water suction, slope geometry and drainage at the base boundary are

discussed in some detail. The chapter concludes with analyses in which a higher permeability

layer due to deterioration of the surface clay is simulated in an attempt to model superficial

sliding.

Chapter 9 analyzes road embankments built on different types of foundation. The emphasis

is on reproducing shallow slides by adopting a higher permeability zone near to the slope
surface and by invoking progressive failure. The risk of longer term deep-seated sliding, even

after a shallow slide has occurred and been repaired, is a so examined.

Chapter 10 summarises the main conclusions that are drawn from the thesis and provides

suggestions for further research.

5
 CHAPTER 2
FINITE ELEMENT ANALYSIS

2.1 INTRODUCTION

The finite element method (FEM) is a numerical technique which can provide approximate
solutions to a wide variety of real engineering problems including geotechnics. Its success,
when analysing any boundary value problem, is due to the satisfaction of the basic solution
requirements, namely: (i) equilibrium of forces, (ii) compatibility of displacements, (iii)
continuity of flow (coupled consolidation analyses), (iv) material constitutive laws, and (v) the
boundary conditions.

The Imperial College Finite Element Program (ICFEP) has been used for all analyses reported
in this thesis. ICFEP uses the displacement based FEM and is capable of handling two
dimensional problems, namely plane stress and strain, and problems involving axisymmetric
geometries. In the later case the material behaviour and/or boundary conditions need not be
axisymmetric. The program deals with material and geometric non-linearities and has been
developed specifically for geotechnical problems. In the current research only plane strain and
full axisymmetric problems (geometry, soil properties and boundary conditions all
axisymmetric) have been considered.

This chapter briefly describes the general principles of FEM, the basics of the technique used
in the present research for solution of the non-linear finite element equations and some
aspects of the method when applied to geotechnical problems. Finally, the chapter concludes
with the formulation and time integration of the finite element equations needed for a coupled
consolidation analysis together with the results from a validation exercise.

2.2 FORMULATION OF FINITE ELEMENT METHOD

Formulation of the FEM can be found in standard finite element textbooks (e.g. Zienkiewicz,
1977; Naylor eta!, 1981; Bathe, 1982). The basic steps in the FEM can be summarised as:
Ii) element discretization, Cii) displacement approximation, (iii) formulation of element
equations, (iv) assemblage of global equilibrium equations, Cv) boundary conditions, (vi)
solution of global equilibrium equations, and (vii) interpretation of the results.

6
2.2.1 Element Discretization

The first step in the analysis is the discretization of the problem domain into a number of
simple elements connected at their nodes to form a mesh. The simplest shape of an element
for plane strain or axisymmetric analyses is a triangular or quadrilateral with straight sides and
nodes located at the element corners (Zienkiewicz, 1977). For problems involving non-linear
material properties, curved boundaries and/or curved material interfaces higher order elements
with mid-side nodes should be considered. ICFEP uses quadrilateral elements with either four
or eight nodes.

2.2.2 Displacement Approximation

In the displacement based finite element method the primary unknown quantity is the
displacement field, d f , which varies over the problem domain. In plane strain and full
axisymmetric analyses the displacement field is characterized by the two global displacement
components, u and v, in the x (r) and y (z) coordinate directions respectively. Over each
element these displacement components are assumed to have a simple polynomial form which
can be expressed as a function of the nodal displacements, d

(2.1)
d 1 = Ndde

where d = [u,v] T , d e = EU i ,V iI U2 V2 ,...UnVnJ T and N d is the element displacement interpolation

matrix consisting of the shape functions, N d ,', whose polynomial order depends on the
number of nodes, n, in the element (see e.g. Naylor eta!, 1991).

The essential feature of the element approximation described above is that the variation of
the unknown displacement field within an element, d, is expressed as a simple function of
the displacements at the nodes, d 1 . The problem of determining the displacement field
throughout the finite element mesh is therefore reduced to determining the displacement
components at a finite number of nodes. These nodal displacements are oftenreferred to as
the unknown degrees of freedom.

The accuracy of a finite element analysis depends on the size of the elements and the nature
of the displacement approximation. For the accuracy to increase as the elements become

'In an isoparametric finite element formulation, which is used by ICFEP, the element
displacement field and element geometries are approximated using the same shape functions.
This is achieved by using shape functions based on the natural coordinate system. A
quadrilateral element in the global coordinate system (x,y) is mapped onto a square parent
element in the natural coordinate system (,i)) of dimensions 2x2 units (see e.g. Naylor et a!,
1981).

7
smaller, the displacement approximation must satisfy certain compatibility conditions. In order
to avoid gaps and overlaps occurring when the domain is loaded, the displacement
components must vary continuously within each element and across each element side. This
can be achieved by ensuring that the displacements of an element side depend only on the
displacements of the nodes situated on that side. In addition, the displacement approximation
must allow the element to undergo rigid body motions and constant straining if it so wishes.
Use of a simple polynomial approximation satisfies these requirements.

Stresses and strains are treated as secondary quantities and can be evaluated within each
element using the definition of strains (see e.g. Zienkiewicz, 1977) and applying the material
constitutive law2

e = Bd (2.2)

0 = D (2.3)

where B is the element strain-displacement matrix and 0 is the constitutive matrix relating the
stress vector, u, to the strain vector, €.

2.2.3 Element Equations

To satisfy equilibrium and the constitutive behaviour, the principle of virtual work is invoked
(see Zienkiewicz, 1977). For a single finite element of linear elastic material subjected to small
(infinitesimal) strains 3 this results in the following set of equations

Kd1 = R (2.4)

where K1 = J (BT DB)dV is the element stiffness matrix, d is the vector of the corresponding
nodal displacements (the primary unknown), and Re is the vector of the corresponding nodal
forces.

2.2.4 Global Equations

The next step in the formulation is the assembly of the separate element equilibrium equations
to give the global equilibrium equations for the whole body

2 To illustrate the method, a linear elastic material will be considered.

3 Large (finite) strain formulation is present in ICFEP (see Dounias, 1987), but it is beyond
the scope of this thesis.

8

Kd = R (2.5)

where
K is the global stiffness matrix assembled from the element stiffness matrices, K1

K = K, = I J (BT.D.B)dV (2.6)

d is the vector containing the unknown nodal displacements over the whole mesh
R is the global vector of applied nodal forces, or right hand side load vector

2.2.5 Boundary Conditions

The final stage in setting up the global system of equations is the application of the boundary
conditions. Ttiese are the loading and displacement conditions which fully define the boundary
value problem which is under investigation.

Loading conditions affect the right hand side of the global system of equations, R. These
include the effects of body forces, R b , surface tractions, R1 , initial stresses, R, and pore
pressure changes, R, given by

R = Rb+RI+R+Rf (2.7)
Rb = I f (NJT.fb)dV (2.8)
R1 = I f (Nf8)dS + F (2.9)
= I (B T.g) (2.10)
R1 = I (B T.u.m)dV (2.11)

In the above equations V is the volume of the element, S is the area of an element over which
surface tractions are applied, b is the vector of body forces, f1 is the vector of applied surface
forces, F is the vector of concentrated nodal loads, o, is the vector of initial stresses, u is
the change in pore pressure, and m is a column matrix as defined in equation (2.18).

Displacement boundary conditions affect the vector of nodal displacements, d. The equation
corresponding to the prescribed degree of freedom (displacement component) must be
effectively eliminated during the solution process. In all cases sufficient displacement
conditions must be prescribed in order to retain any rigid body modes of deformation, such
as rotations or translations, of the whole finite element mesh, If these conditions are not
satisfied the global stiffness matrix will be singular and the equations cannot be solved.

9
2.2.6 Numerical Integration

In order to evaluate the global stiffness matrix, K, and right hand side load vector, R, in
equation (2.5), integrations over element volumes, V. and surfaces, S. must be carried out.
The explicit evaluation of these element integrals cannot usually be performed, except for
some special cases, and therefore a numerical integration scheme is often used. In this
process the integral of a function f(x) is replaced by a weghted sum of the function f(x)
evaluated at 'm' integration points

f f(x)dx = ! f(x)w (2.12)

where wj is the weighting factor that corresponds to integration point 'i', (i = 1 ,2,.. .,m).

The values of w1 and the location of the integration points depends on the type of integration
scheme being used. The number of integration points determines the integration order and the
higher the order the more accurate the integration process. However, the computation effort
increases substantially with number of integration points. ICFEP employs the Gaussian
integration scheme with up to 16 (4x4) integration points (see Nyaoro, 1989). For the eight
noded quadrilateral elements used in this thesis exact integration can be achieved using 9
(3x3) integration points; however, this can result in the use of large computing resources.
Consequently, a 'reduced integration scheme' employing 4 (2x2) integration points has been
used. This, apart from reducing the storage requirements and saving computing resources,
can also lead to improved solution results in some cases (see Naylor, 1974; Bathe, 1982).

2.2.7 Solution of Global Equations

Once the global stiffness matrix has been established and the boundary conditions added, a
large system of simu taneous equations have to be solved to give values for the unknown
nodal displacements, d. There are several different mathematical techniques for solving large
systems of equations. ICFEP adopts a technique based on Gaussian elimination.

2.2.8 Interpretation of Results

After determining the global nodal displacement vector, d, from equation (2.5)

d = [Kr' • R (2.13)

secondary quantities, such as strains, , and stresses, o, can be evaluated using equations
(2.2) and (2.3). Strans and stresses can be calculated at any point within the element,

10
though it is a common practice to evaluate these at Gauss integration points using the B and
D matrices employed in the evaluation of the global stiffness matrix, K (see equation(2.6)).

2.3 NONLINEAR ANALYSIS

In a non-linear finite element analysis the change in boundary conditions is applied in a series
of Increments and the incremental form of the global equitibrium equation (2.5)

KAd = AR1 (2.14)

needs to be solved. Due to the non-linear constitutive behaviour the incremental stiffness
matrix, K 1, (see equation (2.6)) is not constant but varies over an increment 'i' and, therefore,
it is not straightforward to obtain the unknown incremental nodal displacement vector, Ad,
due to the applied incremental nodal load vector, AR. Different solution strategies exist and
three of them have been implemented into ICFEP, namely: (i) tangent stiffness, (ii) visco-
plastic, and (iii) modified Newton-Raphson sdhemes. Ganendra (1994) has shown that the
tangent stiffness and visco-plastic approaches are sensitive to increment size and can lead
to inaccurate predictions unless many small solution increments are adopted. On the other
hand, the modified Newton-Raphson method is both accurate and less sensitive to increment
size and appears to be the most efficient solution strategy; consequently, it has been used
throughout this thesis.

The modified Newton-Raphson method uses an iterative technique to solve equation (2.14).
An approximate stiffness matrix, K 0, and the incremental load vector, AR 1, are used in the first
iteration. However, it is recognised that the solution is likely to be in error and the predicted
incremental displacements are used to calculate the out of balance or residual load, (pi

qi = R-Z J(BT.u)dV (2.15)

which is a measure of the error in the analysis. Equation (2.14) is then solved again with this
residual load forming the right hand side vectoc

Ad = (2.16)

where the superscript 'j' refers to the iteration number, and qi°=AR. This process is repeated
until convergence is obtained. The approach illustrated in Figure 2.1 for a sing e degree
freedom system. The incremental displacements, Ad, is equal to the sum of the iterative
displacements, Ad.

11
In the analyses reported in this thesis convergence of the solution was determined by
checking the changes of both iterative applied loads (residual loads) and iterative
displacements. When coupled consolidation was included the changes in iterative nodal pore
pressures were also checked (see Section 2.5). Convergence was deemed to have occurred
when all these parameters satisfied the convergence tolerance which required the norm of the
iterative values of the parameters to be less than typically 1 % of the norm of the incremental
values4. For analyses involving strain-softening a special procedure for monitoring of
convergence was used (see Chapter 8).

To reduce the number of iterations required for convergence, the iterative displacements, Ad',
given in equation (2.16) can be increased before calculating the residual load, qi1 . An approach
employing a single acceleration parameter, a, (Thomas, 1984) was implemented into ICFEP
(see Figure 2.2). A maximum value of a=8.0 was used in the analyses reported here.

To reduce the amount of computation, the same stiffness matrix, K 0, is often used in equation
(2.16) for all iterations. The symmetric elastic stiffness matrix, K, evaluated at the beginning
of the increment, was used in the analyses reported in this thesis. To increase convergence,
it was updated for the first 2-3 iterations of each increment.

A key step in the modified Newton-Raphson method is determination of the residual load
vector, w' (see equation (2.15)). At the end of each iteration the current estimate of the
incremental displacements is used to evaluate the incremental strains. The constitutive model
is then integrated along the incremental strain paths to obtain an estimate of the stress
changes

Lo = J De9e , o)d E (2.17)

where D is the elasto-plastic stress-strain matrix given by equation (4.51), or (4.70). These
stress changes are added to the stresses at the beginning of the increment and used to
evaluate the integral in equation (2.15).

Since the constitutive behaviour varies over the increment, care must be taken when carrying
out the integral in equation (2.17) to obtain the stress changes. A method of performing this
integration is termed a 'stress point algorithm' and a number of different approaches exist.
The sub-stepping scheme using a modified Euler integration procedure with error control
(Sloan, 1987) was used in all analyses performed in this thesis.

4The norm, n, of both iterative and incremental values are defined as n =Vm.mT, where
m is either nodal load or displacement or pore pressure vector.

12
The sub-stepping technique divides the strain increment into a number of smaller sub-steps
(or flow steps). The modified Euler method uses the D matrix calculated at the beginning of
each sub-step to predict the stress change, at the end of the sub-step (see Figure 2.3).
A new prediction of the stress change, is made using the D8 calculated at the end of
the sub-step. The average of the two stress changes, Ao 1 and gives the assumed stress
change, Ao, for the sub-step. The error associated with this method can be controlled by
checking the difference between Ao and u2 . Here it was restricted to 0.01 % of the current
stress state, 0. At critical integration points several hundred sub-steps were typically required
to meet this criterion. However, for the analyses involving strain-softening, particularly when
collapse was approaching, more than this was required at the integration points along the
rupture surface, and consequently a number of sub-steps was increased to several thousand.

At the end of each sub-step, because of the approximation involved in the sub-stepping
techniques, it will be usually found that the stress state does not lie on the yield surface to
within a prescribed tolerance. To avoid error accumulation due to the drift of the stress state
from the yield surface and to satisfy the constitutive laws, a correction should be applied at
the end of the sub-step to return (project back) the stresses to the current yield surface.
Different methods of projecting back were studied by Potts and Gens (1985). The 'correct'
method which takes into account the changes in plastic strains as well as the associated
elastic strain changes was used here.

2.4 GEOTECHNICAL CONSIDERATIONS

Some aspects of the FEM of interest to geotechnical engineering problems and relevant for
the analyses reported in this thesis will be briefly discussed in this section. They include: (i)
initial stresses, (ii) excavation, (iii) construction, and (iv) pore pressures. Time dependent
consolidation will be treated separately in the next section.

2.4.1 Initial Stresses

The initial stresses at the beginning of the non-linear finite element analysis are needed for
two reasons: (i) material behaviour is stress dependent, and (ii) to calculate the unloading
forces in an analysis involving excavation, it is necessary to know the initial stresses, even
in a linear elastic analysis.

In ICFEP the initial stresses can be generated by performing a gravity 'turn-on' analysis, or
they can be specified directly through an user defined subroutine. The latter method is more
flexible and versatile (e.g. different and high K 0-values can be specified, non-level ground can
be easily dealt with) and was used here. In some cases it was necessary to establish the

13
specific equilibrium pore pressure dstribution, according to the hydraulic boundary conditions
and the variation of permeability with depth, by allowing several increments of consolidation
prior to performing the main analysis (see Section 8.4.7).

2.4.2 Excavation

Excavation is modelled by sequentially removing groups of elements. In ICFEP removal of

elements is simulated by first calculating the equivalent nodal forces arising from the stresses
acting within these elements and then applying those which acted on elements remaining in
the mesh in their opposite sense as 'new' boundary conditions in the analysis. Correct
account is taken of both the initial stresses and those stress changes which occurred during
the calculation process prior to excavation.

2.4.3 Construction

Construction is simulated by sequentially adding groups of elements. The method used by

ICFEP first assumes that all elements to be constructed behave as a linear elastic material
with negligible stiffness ('ghost elements'). Construction of a group of elements is essentially
performed by a gravity 'turn-on' analysis. For a broad fill, the vertical stresses are equal to
the overburden pressure while the horizontal stresses are determined by the value of
Poisson's ratio u=O.45 (K 0 =uI(1-u)=O.80) 5 . During this process the newly constructed
elements retain their negligible stiffness resulting in the spurious nodal displacements and
strains. After construction, the stresses and nodal forces (and the hardening parameters, if
the fill is modelled as an elasto-plastic material) calculated for the newly constructed elements
are retained, but the displacements and strains are set to zero. The elements are assigned
their proper stiffness before the next group of elements is constructed.

In clay fills (see Chapter 9) it is necessary to model initial suctions (negative tensile pore
pressures). To do this, after the element has been constructed the mean effective stress, p',
is reset to be equal to an appropriate initial suction, u 0 (p' =-u0 ), and the pore pressure is reset
to u=u0 +p 0', where p 0' is the mean effective stress after construction of the element. By
essentially resetting the mean effective stress, p', it is possible to control the undrained shear
strength of the clay fill material. ft is worth noting that, by changing both mean effective
stress and pore pressure, total stresses remain constant and equilibrium of forces is

5 There is an option in ICFEP to change the horizontal stresses to be equal to K,,, times the
calculated vertical stresses, using values of K0 input by the user. This option was used in
some of the analyses reported here (see Chapter 6 and Appendix) to investigate the influence
of the horizontal stresses in a newly compacted layer on the settlement profile at the end of
dam construction.

14
maintained.

2.4.4. Pore Pressures

Soil media comprise solid particles and voids which are filled with water when saturated, or
filled with both air and water when partly saturated6 . For saturated soils, instantaneously on
load application, the pore water is usually assumed to carry most of the toad in terms of
excess pore pressures. These then dissipate with time during consolidation causing changes
to the inter-particle stresses.

Throughout the loading history three situations can be identified: (i) instantaneous undrained
behaviour, (ii) time dependent consolidation, and (iii) fully drained behaviour.

Undrained analysis can be performed using either total or effective stress parameters. The
latter method is more versatile and has been used here.

The principle of effective stress is invoked to obtain the total stresses and stress-strain matrix
required in the formulation of the equilibrium equations

(2.18)

where m is a column matrix relating the pore pressure increment scalar, u, to the direct
stress terms in the total and effective stress increment vectors, to and no', respectively. For
three-dimensional analysis m T = (1,1,1 ,0,0,0).

If the bulk modulus of the pore water is denoted by K Bf , it can be shown (Naylor eta!, 1981)
that the total stress-strain matrix, D, is given by

D = D' + K Bf mmT (2.19)

where D' is the effective stress-strain matrix. For isotropic conditions, equation (2.19) is
equivalent to evaluation of the D matrix from bulk modulus, K B , and shear modulus, G, given
by

KB = K B' +K (2.20)

G = G' (2.21)

6 Clay fills are invariably partly saturated immediately after construction. However, in the
finite element analysis it is not unreasonable to idealise the c ay fill as fully saturated,
especially if it is placed wet of optimum (Naylor, 1991 b).

15
where K8' and G' are the effective bulk and shear moduli of the soil skeleton respectively.

For undrained analysis K 8f is set to a very large value such that the volumetric strain, E, is
inhibited. ICFEP has two options for defining K 8f: ( i) a constant value is prescribed, or (ii) it
is defined as a multiple of the bulk modulus of the soil skeleton, K B '. The second option is
preferred, and a value of K B, one hundred times the bulk modulus K 8 ' is used here.

The solution of the global equilibrium equation gives the strain change, Ae, and explicit values
of total, effective stress and pore pressure changes are given by

= D (2.22)

= D'te (2.23)
= Kaj.mTAE (2.24)

Equation (2.19) can also be used in a drained analysis by setting K = 0.

2.5 COUPLED CONSOLIDATION

The analysis of the intermediate time dependent consolidation is more complex than the
analyses of 'short-term' undrained and 'long-term' drained behaviour. It requires solution of
the Biot's consolidation equations coupled with the material constitutive laws and the
equilibrium equations.

2.5.1 Formulation

The following equations have to be satisfied in the process of consolidation of an elastic soil
media saturated with an incompressible pore fluid (Biot, 1941)

(i) Equations of equilibrium

øa'/8x, + 6(8pI8x) + be =0 (2.25)

(ii) Stress-strain re ationship

= De (2.26)

(iii) Equations of continuity

øv/øx = (2.27)

16
 (iv) Generalized Darcy's Law

v, = - (k1Iy)(8pIOx,) (2.28)

where

x are the rectangular Cartesian coordinates

are the components of the effective stress tensor
6 is the Kronecker delta
p is the pore fluid pressure

b, are the components of body force

E ki are the components of the strain tensor
D kI are the elastic coefficients in the generalized Hooke's Law
v1 are the components of the superficial velocity vector
E is the volumetric strain
k 1 are the coefficients of permeability in the generalized Darcy's Law
v is the unit weight of water
t denotes time

Using the principle of virtual work and incorporating the appropriate boundary conditions the
element equilibrium and continuity equations can be derived (Small eta!, 1976). After their
assembly to the global level they can be written in the matrix form

Kd + Cp = R (2.29)

CT• Wd/8t) - Qp = 0 (2.30)

where

K is the global stiffness matrix given by equation (2.6)

R is the right hand side load vector given b equations (2.7)-(2.1 1)
C is the global coupling matrix given by

C = X (BT.N)dV (2.31)

0 is the global flux matrix given by

= (1/y,,) J(ETkE)dV (2.32)

d and p are the global vectors of the unknown nodal displacements and pore

17


pressures respectively

In equation (2.31) N is the element pore pressure interpolation matrix relating the pore
pressure field, p, over an element to the nodal pore pressures, p6

p = Np 8 (2.33)

The matrix B relates the element volumetric strain, e, to the nodal displacements, d6

= Bd6 (2.34)
where

B = mB (2.35)

in which m is the column matrix (see equation (2.18)), and B is the element strain-
displacement matrix (see equation (2.2)).

In equation (2.32) k is the matrix of permeability coefficients for the generalized Darcy's Law

V = - kVp (2.36)

The matrix E relates the element pore pressure gradient vector, Vp (for three-dimensional
analysis (Vp)T = (aplax,apIay,ap/az)), to the nodal pore pressures, p8

Vp = E•Pe (2.37)
where

E = VN (2.38)

In a non-linear consolidation finite element analysis the material 'constants' D JkI will depend
on the current stress state, or on the pervious history of the particular soil element, and hence
time as well. Therefore, it is necessary to consider the following incremental form of the
global equilibrium equation (2.29) (see e.g. Small eta!, 1976)

Kd + Cp = A (2.39)

where the dot denotes differentiation with respect to time, the definition of C remains the
same and K is given by

K = (BT.DeP.B)dv (2.40)

18
in which D is the elasto-plastic stress-strain matrix given by equation (4.51), or (4.70).

If the matrix of permeability coefficients, k, is dependent on the current stress level or indeed
on the pervious history7 it is necessary to differentiate the global continuity equation (2.30)
as well. However, in the analyses reported here it has been assumed that permeability is
constant or varies with depth only (see Chapters 8 and 9) and consequently equation (2.30)
remains unchanged.

2.5.2 Time Integration

If the solution (d 1 ,p 1 ) is known at time t 1 , a marching process can then be used to obtain the
solution (d2 ,p2) at time t2 =t1 +t. Equations (2.39) and (2.30) can be approximately
integrated in the form

(2.41)
K1 (d2-d1 ) + C(p2-p1) = R2-R1
CT.(d2d1) -tø(p 2-p 1 ) = (2.42)

where Ka is the average stiffness matrix over the interval (t 1 ,t2 ) and the parameter fi
corresponds to the approximation

f t2 (øp)t Ø[fi p2 + (1-p 1 LAt (2.43)

In order to ensure stability of the marching process it is necessary to choose ^ 1/2 (Booker
and Small, 1975). p=0.55 was used in the analyses reported in this thesis.

Equations (2.41) and (2.42) may be written in the general incremental matrix form

K1 C 1d1 - (2.44)
C l. -fitø [APJ -

The system of non-linear equations (2.44) of a coupled consolidation problem can be solved
using e.g. the modified Newton-Raphson method. As before (see Section 2.3), it is convenient
for elasto-plastic calculations to take K 1 =K, where K' is the symmetric elastic stiffness

7 Quite recently these options have become available in ICFEP but they have not been used
in this thesis.

8The global continuity equation (2.30) can be further extended to account for finite
deformation (see e.g. Carter et a!, 1979).

19
matrix evaluated at the start of the increment and kept constant (after a couple of updatings)
throughout the iterations in a single time step. The 'generalized' residual load can now be
defined as

(qgi)T
(wdw') (2.45)

where qQ is given by equation (2.15) and, when the coefficients of the permeability matrix,
k, are constant

wpJ =0 (2.46)

2.5.3 Validation

Closed form solutions for consolidation problems are not easy to obtain. This is especially true
when dealing with elasto-plastic materials and finite (large) deformations. Exact solutions have
been found only to problems considering elastic materials, with simple geometry and subject
to simple boundary conditions.

Hwang et a! (1971) presented some solutions for plane strain consolidation problems by the
finite element method and compared them with available closed form solutions. The same
problems were chosen here as examples of extensive validation exercise of the coupled
consolidation formulation implemented in ICFEP.

The first problem analyzed was that of a porous elastic half-space subjected to a load of
intensity 'q' over a width 2a under conditions of plane strain. The finite element mesh is
given in Figure 2.4.i.a together with boundary conditions suggested by Hwang etal(1971).
The results have been expressed in terms of the 'adjusted time factor', r

r = (ât)/a2

and the 'adjusted coefficient of consolidation', ë

= (2Gk)Iy

where G is the elastic shear modulus, k is the coefficient of permeability, y,, is the unit weight
of water, and t denotes time.

The load 'q' was applied in a very small time step (expressed in terms of the 'adjusted time
factor' it was r= 1 .1 5 10) resulting in an undrained response. Logarithmic time increments,

20
typically five per 'log cycle', were used thereafter. Response during the first 'log cycle' was
essentially undrained, and the results for this 'log cyc'e' are not presented.

The variation of normalized excess pore pressure, u/q, beneath the centre of the loaded area
is given in Figure 2.4.ii for the specific case of 7=0.1 and Poisson's ratio u=0.0. It can be
seen that ICFEP (open squares) overpredicts the excess pore pressures particularly at distance
beneath the loaded area. Because the closed form solution, represented by the full line
(Schiffman et a!, 1969), was derived for an infinite half-space it was suspected that this
discrepancy was due to the close proximity of the bottom and/or lateral boundaries in the
finite element mesh. The effect of position of both boundaries (at bottom and lateral, see
Figures 2.4.i.b and c respectively) on the results of analysis were further investigated and,
surprisingly, it was found that the position of the lateral boundary has a much greater
influence. For b=18a (instead of b=6a) and H=9a (as before), ICFEP (shaded squares)
recovered the closed form solution except for a slight discrepancy at the base of the mesh.

In Figure 2.4.iii the variation of normalized excess pore pressure, u/q, with 'adjusted time
factor', r, is given for two specific points in the half-space (see Figure 2.4.i.c). Two
predictions were made by ICFEP depending on the particular form of the element pore
pressure interpolation matrix, N (see equation (2.33)). ICFEP has two options: (i) 'code 4' -
a linear variation of pore pressure across the element is assumed and, in the case of eight-
noded elements, pore pressure degrees of freedom, pa, are placed at the corner nodes only,
(ii) 'code 8' - a quadratic pore pressure variation across the element is assumed and a pore
pressure degree of freedom is placed at every node of an eight-noded element. It can be seen
that agreement, in both cases, with the closed form solution is good. However, it can be
argued that 'code 4' is more appropriate when dealing with eight-noded isoparametric
elements. In that case the variation of both pore pressure and effective stresses will be linear
across the element. Thus 'code 4' has been used in all analyses involving coupled
consolidation reported in this thesis.

The second problem analyzed was the consolidation of an uniformly loaded porous elastic
layer of finite thickness resting on a smooth impervious base and subjected to conditions of
plane strain. Closed form solutions were published by Gibson eta! (1970). The finite element
mesh used and appropriate boundary conditions are given in Figure 2.5.i. The variation of the
surface settlement, w, conveniently expressed in terms of the dimensionless parameter
(Gw)/(aq), with the 'adjusted time factor', r, beneath the centre of the loaded area is shown
in Figure 2.5.ii for several values of Poisson's ratio u=0.00, 0.20 and 0.40. Application of
load and time increments were the same as in the pervious problem. Again, results are not
presented for the first 'log cycle' because of the undrained response.

21
Hwang et 8/ (1971) did not specify the extent of their lateral boundary and two finite
distances were considered in the analyses performed by ICFEP: b=6a (open symbols) and
b=36a (shaded symbols). Again, a significant influence of the position of the lateral
boundary was found. With b=6a, ICFEP underpredicted the settlements especially in the
early stages of consolidation. At the end of consolidation (long-term, drained condition) the
true amount of settlement was recovered for the values of Poisson's ratio u=0.00 and 0.20
but for u=0.40 the settlement was still underpredicted. When the lateral boundary was
extended to b=36a, ICFEP gave excellent predictions. It is worth noting that slightly
different settlements were predicted at the beginning of the analysis depending on the value
of Poisson's ratio, u, employed 9 . The closed form solution yields the constant value of
immediate settlement defined by the undrained value of Poisson's ratio u=O.50 and undrained
shear modulus G=G'.

9 The analyses of the same problem using 'code 8' instead of 'code 4' also predicted
Poisson's ratio dependent initial settlements, but differences were significantly smaller. An
increase in number of layers in the finite element mesh had the same effect.

22
 True solution
01
.1i

d' ltd2

Displacement

Figure 2.1 Modified Newton-Rapson method

IJIpIdLeIIJvIuI

Figure 2.2 Accelerated Newton-Raphson method

Constitutive
Modified law
a,
C')
Euler
slre.__
path f Ii

Sub-step

Strain

Figure 2.3 Modified Euler sub-stepping

23
 Pvsi. p.,niabl. boundlty q .ubounIItIq
y/•-0.5
xl•-0.0 I y.O.B
- - - 31/1-1.0 '1i_i I I i/.-i.o till

I
Rz.d. bopm.s. bvundy
V
b-S.

(a) H=9a,b=6a 1 ----
ii

I-

(b)H=18ab=6a

q
yI.-O.S
3111-0.0

Ii
I
I -
I V

I-. b-IS.

(c) H=9a,b=18a

Figure 2.4.i Finite element meshes and boundary conditions

24


>.5

5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Normahzed Excess Pore Pressure, ufq

Figure 2.4.ii Variation of normalized excess pore pressure, u/q, beneath the centre of the
loaded area for the 'adjusted time factor' T=O.1

yla=0.5
1 - __________ ________ - - _____ xla=0.0, Schiffmanetal, 1969
- - - c'a=l .0, Schiffman et al, 1969
0 9
. xla=0.0, ICFEP,code4
0.8 . xla=1.0, ICFEP,code4
.1 .. .
o xla=0.0, ICFEP, code 8
0.7 a
o x/a=1 .0, ICFEP, code 8
0
I--S.
0- 0.6 -.
I
0- 0.5 - .-
.
0.4 ! S.
S.
0.3 . .

0.2- .
'I
0.1 . .. .
.
0 --.- .
0.0001 0.001 0.01 0.1 1 10
2
Adjusted Time Factor, (Et)/a

Figure 24. Variation of normalized excess pore pressure, ulq, with 'adjusted time
factor', T

25
 I-.
Frt PIflflSS body

'ill
12 ji
Smooth. - -

b-S's

Note: Lateral boundary was extended to b = 36-a

to examine the influence of its position
on results of the analysis.

FIgure 2.51 Finite element meshes and boundary conditions

0.2

0.25

'- 0.3

0.35
C
w
E 0) 0.4

" 0.45

0.5

0.55

0.0001 0.001 0.01 0.1 1 10

2
Adjusted Time Factor, = (E.t)Ia

Figure 2.5.ii Variation of normalized surface settlement, (Gw)/(aq}. with 'adjusted time
factor', 7

26
 CHAPTER 3
NUMERICAL MODELLING OF EMBANKMENT DAMS

3.1 INTRODUCTION

Many attempts have been made to predict the behaviour of embankment dams. The finite
element method is a powerful tool for such predictions. A finite element analysis of an
embankment dam was first carried out by Clough and Woodward (1969). Since then a large
number of finite element analyses of actual dams has been reported in the literature and these
have recently been tabulated by Duncan (1992, 1994).

The finite element analysis of stresses and movements in embankment dams is an exceedingly
complex problem. The techniques for modelling construction, first impounding and subsequent
operation are covered in various publications, most thoroughly by Naylor (1991b). Many
factors should be considered. Those relevant for the work described in the thesis will be
discussed here.

Layered analysis, stiffness of simulated layer and compaction stresses are discussed first.
Then, typical stress paths occurring in embankment dams during construction and first
reservoir filling are consdered. In the remainder of the chapter the constitutive models that
have been used in the past to describe the behaviour of fill materials are reviewed briefly.

3.2 LAYERED ANALYSIS

The need to simulate the construction sequence in the finite element analysis of embankment
dams was recognised by dough and Woodward (1967). Since embankments are constructed
in successive layers, each layer applying load to the underlying material due to its self weight
and adding stiffness to the structure as a whole, a 'layered' type of analysis must be
employed. Deformation is defined as the cumulative movement of a marker placed on the
surface of the fill due to compression of the underlying material by the weight of material
added above (Naylor and Jones, 1973). These movements are often measured in the field.

Embankment dams are built up in relatively thin horizontal layers. Consequently, there will be
a large number of layers during the construction of a large dam. The limitations of computer

27
modelling require relatively thick layers to be used in the idealization.

To provde an insight into the errors involved, Kulhawy et al (1969) and Naylor and Jones
(1973) compared the closed form solution of the incremental analysis to the finite element
'layered' analysis using a one-dimensional model which represents either a soil column or a
broad fil . Although different assumption concerning the material non-linearity was used, they
both concluded that: (i) the larger number of layers the closer is the 'layered' solution to the
'exact' one; (ii) the number of layers required to achieve a solution close to the 'exact' one
reduces as the degree of non-linearity decreases until reaching the limit for linear elastic
material, where the 'layered' solution coincides with the 'exact' one irrespective of the
number of layers'. Hamza (1976) pointed out that the above conclusions of Kulhawy eta!

(1969) and Naylor and Jones (1973) were invalid and reflect on the shortcomings of the
solution strategy they employed 2 . Using an iterative non-linear technique, Hamza (1 976)
eliminated errors other than those due to layering and concluded that one-dimensional problem
was independent of construction sequence irrespective of whether linear or non-linear material
was considered.

The same has been concluded by Naylor and Mattar (1988). They argued that for the
one-dimensional case the new layer stiffness had no effect on the settlement profile nor on
the stress distribution. However, Naylor (1991b) has noted that Poisson's ratio, or its
equivalent for non-linear models, must be the same in the new and underlying layer.
Otherwise there will be some effect on the horizontal stress 3 . Implication of these findings if
an elasto-plastic model (of the critical state type for example) is employed has not been
considered. It can be of some significance, even in case of elastic models (linear or non-
linear), when the effects of compaction are modelled. The effect of compaction is to increase
horizontal stresses in a layer as further fill is placed. This effect typically becomes insignificant
when 3-5m of fill has been placed (Ingold, 1979). For the dams analyzed in the thesis this is
the approximate thickness of a finite element layer and, to model this, the technique described
in Section 2.4.3 of the previous chapter has been used. Apparently, this way of compaction
modelling will have some influence on the distribution of horizontal stresses and consequently
the settement profile (see Footnote 3). The same is valid when inital suction in clayey fills
are modelled. This may have a consequence on Naylor and Mattar's (1988) conclusion that

1 At least three layers will be necessary to give a smooth settlement profile.

2They used the tangent stiffness method. Only one increment per layer was employed,
and the stiffness matrix was determined on either the state of stress before or after adding
the new layer or on average between these two.

3 Nayor and Mattar (1988) used the stiffness varying with depth only. If the stiffness
somehow depends on the horizontal stress, there will be some effects on the settlement
profile as well.

28
as few as five layers is usually sufficient in embankment dam analyses.

Although the one-dimensional model (no lateral strain) is a good approximation of the dam
construction (at least in the central portion of a dam), and therefore is very useful in
determining the optimum number of layers in the finite element analysis, it is important to
recognised the particular influence of the new layer stiffness in the two-dimensional analysis.
This is due to the difference in bending of one 'thick' simulated layer in the finite element
analysis when gravity is suddenly applied and progressively occurred bending of several 'thin'
layers in the field. Naylor and Mattar (1988) recognised that some stiffness between zero and
the full in-place stiffness would make the finite element layer equivalent to the real situation.
They proposed a stiffness reduction factor, f, by which the in-place stiffness should be
divided in order to obtain the equivalent stiffness giving the same deformation pattern in the
underlying layers. For fills modelled as linear elastic materials the optimum value of f =4 has
been derived. No optimum value has been determined for non-linear elastic materials.
However, they suggested the value f = 3-5 not ng that quite big variations would have little
bearing on the results, particularly if the number of layers in the analysis is increased. It
should be noted that, as in the case of the one-dimensional study, the impact of compaction
and initial suction modelling on the above conclusions and recommendations has not been
considered.

In the finite element analysis of the two embankment dams reported in this thesis no stiffness
reduction factors have been used. As it was mentioned in Section 2.4.3, a negligible stiffness
was assigned to the layer of elements during construction resulting 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.

3.3 STRESS PATHS

The deformation response of soils is known to be stress path dependent. Both measurement
of deformation parameters in the laboratory and selection of the constitutive model and its
appropriate parameters for analysis must take into account this aspect of soil behaviour.
Thus, it is of great importance to establish typical stress paths occurring in embankment dams
during construction, impounding and operation. This matter has received some attention in
the literature. Although actual stress paths are dependent on the construction sequence, the
geometry and type of the embankment dam and the stress-strain properties assumed for the
embankment fill, some general points can be made.

A comprehensive exercise of this kind was undertaken by Charles (1976). Figure 3.1 shows
typical stress paths derived from elastic finite element analyses for selected points within

29
three different types of rockfill dams. Although the stress conditions vary from one location
to another within each of three dams, in general construction involves a large increase in
mean effective stress, p' = (a" + 02' + 0 3 )13, and only gradual changes in principal stress ratio,
R=a1 '1a3 '. While the vertical stress nearly remains the major principal stress in the central
portion of the dams, there is considerable rotation of principal stress directions, 9, near the
slopes and the vertical stress ceases to be equal to the weight of soil above (overburden
pressure), due to load transfer. Reservoir impounding of the dam with an upstream impervious
membrane causes further increase in mean effective stress, p', a decrease in principal stress
ratio, R, and significant rotation of principal stresses, 0, especially in the upstream part of the
dam.

The effects of reservoir impounding on rockfill dams with clay cores are more complex. Veiga
Pinto and Maranha Das Neves (1985) plotted stress paths representative for the transition
zones of a rockfill dam with the central clay core (see Figure 3.2). During construction
maximum shear stress, t = (a-a3 )I2, increases under nearly constant principal stress ratio, R.
During reservoir impounding the transition zones are submitted to a different kind of stress
paths. A general decrease in maximum shear stress, t, is accompanied by a relief of average
effective stress, s' = (o 1 ' + a3 ')12, in the submerged upstream part of the dam and its increase
in the dry downstream part.

Mon and Pinto (1988) qualitatively considered stress paths for four points along the line
normal to the concrete face of a hypothetical rockfill dam (see Figure 3.3). During
construction, with increasing height of the dam, they noted deviation of stress paths from K 0-
line4 towards the strength envelope. This is due to lateral yield in the dam shoulders when the
stresses may approach the active state. During impounding stress paths change direction. For
shallow portions of the dam (point B) shear stress, t = (o-a3 )I2, may decrease to zero when
rotation of principal stresses through 0= 900 produces a 'reloading' stress path.

For the same cross-section Saboya and Byrne (1993) calculated by finite element analysis
stress paths during reservoir impounding for four elements (see Figure 3.4). It can be seen
that in the upstream toe area of the dam maximum shear stress, t=(a1-a3)/2, only reduces
for the initial stages of reservoir impounding. However, for the higher reservoir levels, after
rotation of principal stresses, this quantity starts increasing again.

From the all above mentioned it can be concluded that a great variety of stress paths occurs
in embankment dams. They are accompanied by a large rotation of principal stresses and are

4 Line representing the stress path during one-dimensional compression with zero lateral
strain, K 0 =a3'1a1 ' and E3=0.

30
particularly complex during reservoir impounding. 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 of their coarse nature
the tests are expansive and laborious. Consequently, their number and types are limited.
Sometimes both large triaxial and oedometer tests are performed for rockfills 5 , but quite often
only one of these two is available. Therefore, it is necessary to establish which of these is
more representative to simulate rockfill behaviour in an embankment dam.

It can be argued that oedometer test, providing a constant stress ratio path, R =a1 'Io3' = 1/K0,
can better capture rockfill behaviour, at least during the construction stage (Eisensteiri and
Law, 1979b). Although the principal stress ratio, R, is not ideally constant during the dam
construction (see Figure 3.1), in most cases its value changes within a relatively narrow limit
and usually around the assumed R = 1/K 0 value. Laboratory tests on rockfill by EI-Ruwayih
(1 975) and Charles (1976) following constant stress ratio paths at different values of A
showed that for 1.25 ^R^3.O there was no significant changes in the volumetric strain e -
mean effective stress p' relationship. That is why oedometer tests, with R typically between
2.0 and 3.0, give a reasonable prediction of overall rockfill dam behaviour during construction
(Charles, 1976). However, in spite of this triaxial tests should not be neglected. Apart from
defining the rockfill strength envelope, these tests provide an useful insight into the stress-
strain properties of rockfills too. It can be seen from Figures 3.2 and 3.3 that stress paths
during construction with increasing dam toad can approach those imposed by standard triaxial
testing. 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 rockfitl behaviour in both
oedometer and standard triaxial tests (if both types of test 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 achieve this is, however, another
question!

3.4 CONSTITUTIVE MODELS

It has been repeatedly shown (Duncan, 1972, 1992; Eisenstein, 1974; Eisenstein and Law,
1 979b) that the most influential factor in the finite element analysis of embankment dams is
the modelling of fill stress-strain behaviour by an appropriate constitutive law. However, in
spite of the diversity of the stress-strain relatonships being used, reasonable agreement was

5 Large shear box tests are often available. These tests are cheaper and simpler to perform,
and they are usually provide data defining the rockfill strength envelope. However, an
appropriate stiffness may also be deduced (Vaughan, 1994).

31
usually found when the results of the finite element analyses (typically movements) were
compared with field observations (see e.g. Duncan, 1992). 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 the after-the-event or so-called 'Class Cl' (according to Lambe, 1973)
predictions.

A review of different constitutive laws used in the numerical analysis of embankment dams
can be found in recent 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.

3.4.1 Linear Elastic Analysis

Soils are far from being either linear or elastic. 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 characterize the stress-strain behaviour of isotropic
linear elastic materials. They are usually Young's modulus, E, and Poisson's ratio, u. 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. Penman et al (1971) proposed a
'constant equivalent compressibility' method for determination of Young's modulus, E, using
data from oedometer tests. 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 equivalent compressibility, m,
determined to give the correct final displacement of a point half-way up the complete layer.
In the linear elastic analysis of a rockfill dam this suggests a Young's modulus value

E = (1 Im)[(1-2iI)I(1 -u)J (3.1)

which is dependent on the degree of non-linearity of the one-dimensional stress-strain

relationship and final height of the rockfill column at a particular location. The value of
Poisson's ratio can be estimated from the K 0-value (K=a3'/u1'), as

U = l(/(1 +K0) (3.2)

32
where K0 is related to the angle of shearing resistance q' by (Jaky, 1944)

K0 = 1-sinq' (3.3)

or, determined directly from the floating ring oedometer test for instance, which includes
measurement of lateral stress 0 3 (Penman and Charles, 1976).

This method has been used to predict displacements during construction for a series of rockfill
dams of different types constructed in Great Britain in the last 25 years (Penman eta!, 1971;
Penman and Charles, 1 973a, 1 985a, 1 985b). Comparisons with observed deformations have
shown that the method can reasonably predict rockfill behaviour under stress conditions
found during dam construction. Charles (1976) has demonstrated why this is so. The method
is less satisfactory for prediction of rockfill behaviour during reservoir impounding when stress
paths are more complex and accompanied by large principal stress rotations (see Section 3.3).

In spite of their inherent shortcomings, simple linear elastic analyses have been very useful.
They have sometimes preceded more complex non-linear elastic analyses and surprisingly, in
some cases, gave the better overall results (see e.g. Cathie and Dungar, 1978). They also
highlighted the effects of important factors,two of which, Poisson's ratio and anisotropic
deformability of compacted fill materials, will be further discussed here.

Imaizumi and Sardinha (1985) pointed out that Poisson's ratio has a decisive effect on the
general pattern of movements of concrete-faced rockfill dams. They found that a low value
of Poisson's ratio, u=0.0-0.1 5, resulted in the realistic displacement pattern observed during
reservoir impounding of large existing dams of this type. This is in agreement with
observations made by Boughton (1970) who, summarizing data from large triaxial tests on
different rockfill materials, has concluded that Po sson's ratio reduces considerably and
Young's modulus increases significantly as the principal stress ratio, R=a1 Io3, approaches
unity. The same conclusion was reached by Penman and Charles (1985a) on the basis of
small scale triaxial tests performed by Charles (1973, 1976) which will be outlined in Chapter
5 of this thesis. Employing this finding in the simple linear elastic analysis, they were able to
predict reasonably the observed movements of the upstream asphaltic membrane of Winscar
dam during reservoir impounding (see Chapter 6).

Cooke (1984) has noted that the modern practice of rockfill compaction in thin layers results
in at least a visually anisotropic rockfill mass. Indeed, it has been found at many concrete-face
rockfill dam sites that the modulus of rockfill deformation, calculated from face movement
after reservoir impounding, is greater than one measured vertically during the rockfill
placement suggesting substantially stiffer behaviour of the rockfill in the horizontal than in the

33
vertical direction (Fitzpatrick eta!, 1985). The type of anisotropy which is likely to occur and
which is the simplest to analyze is cross-anisotropy. Five elastic constants are required to
describe a linear elastic cross-anisotropic material. Detailed analysis of the influence of
anisotropy on stresses and displacements in embankments was carried out by Eisenstein and
Law (1 979a). Employing some typical reported anisotropic properties of fills they found that
the results calculated with and without consdering anisotropy were not substantially
different. A larger Young's modulus in the horizontal direction tends to underestimate vertical
and overestimate horizontal displacements during embankment construction. Imaizumi and
Sardinha (1985) reported that the larger modulus of rockfill deformation in the horizontal
direction resulted in an unrealistic prediction of the pattern of concrete-face rockfill dam slope
movements during reservoir impounding.

3.4.2 Non-linear Elastic Models

One of the most important characteristics of soil stress-strain behaviour is a lack of a linear
relationship between stresses and strains. The 'layered' analysis of enibankments permits
straightforward simulation of the stress dependent behaviour of fill materials (Clough and
Woodward, 1967). A great variety of non-linear elastic models exist in the literature. The
most widely used are briefly described below.

3.4.2.1 'Power Law' 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, (=), and axial stress, Oa' ( = 01 ' ), in a one-
dimensional compression test can be described by a power law of the form (Rowe, 1971;
Skinner, 1975)

C.(o.v/p.)D (3.4)

where p is atmospheric pressure in the same units as O', C and 0 are dimensionless
parameters. Whereas the parameter D may be considered to be constant irrespective of the
initial density, it appears that the parameter C depends linearly on the initial porosity (El-
Ruwayih, 1975; Daramola, 1978). Skinner (1975) used the above relation to predict the
settlement profile on the centre-line of an embankment° of height H constructed of rockfill of

6 Settlements are treated as one-dimensional and vertical effective stress is presumed to

be equal to 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, nd compensating.

34
placement unit weight y

= [(Ci . yD)/(D+1)JiH D * - ( H-h)°" - (3.5)

where $ is the settlement of the marker placed at a height h above the foundation and
C 1 =CIPa°.

Veiga Pinto (see e.g. Naylor eta!, 1986) in his so-caned 'E-K0' model expressed the relation
given by equation (3.4) more conventionally in the form

B. (36)
= A,P EV

where A and B are dimensionless model parameters. If the radial stress, Or' (=0 3 '), is
measured during the oedometer compression test, the K0-value can be assumed to vary
linearly with axial stress, o', according to

K0 = A0 + B0(a.'/P.) (3.7)

where A0 and B0 are dimensionless model parameters.

By differentiating equations (3.6) and (3.7), the expressions for the tangential constrained
modulus, Ect, and the tangential K0t-value can be obtained which, using equations (3.1) and
(3.2) and noting that E = 1/m u, will define the tangential Poisson's ratio, u, and the tangential
Young's modulus, E1.

Four parameters are needed to characterize the model. An additional three parameters are
required to model unloading (see e.g. Naylor, 1991a.

A similar model was advocated by Eisenstein and Law (1 979b). Arguing that the radial stress,

r' (=0 3 '), in an oedometer test cannot be successfully measured, they suggested an isotropic
compression test as complement to the oedometer test to estimate the value of Poisson's
ratio, u. They assumed that nearly identical samples of fill have the same volumetric strain
e when loaded from zero to the same mean effective stress p' in both one-dimensional and
isotropic compression tests'. Again, for loading condition, four model parameters are required.

The above models can be simplified if a constant K0-value is assumed in the oedometer test.

'This assumption will be used later in the thesis in connection with the derivation of the
parameter values for the two Lade's elasto-plastic cnstitutive models (see Chapter 5).

35


This is a frequently made assumption for soils. Then, it can be noted that both models will
'approach', to a certain extent, the simple linear elastic model described in Section 3.4.1.
Thus, the similar degree of success in predicting the constructional deformations of
embankments can be expected. It is worth mentioning that Naylor eta! (1 986) recommended
the E-K 0 model only for the central part of an embankment and its foundation arguing that
it would be unwise to use it elsewhere because of stress path deviation from that in a one-
dimensional test.

3.4.2.2 Hyperbolic Model

This model is based on the hyperbolic function proposed by Kondner (1963) to approximate
the stress-strain curve in a conventional triaxial compression test and was developed into a
full constitutive law by Duncan and Chang (1970). Using a Mohr-Coulomb failure criterion and
utilizing the experimental studies by Janbu (1963), they derived the following expression for
the tangential Young's modulus

R( 1 -sin') (a1-a3)
E = [1 - ]2Mp8(a3'/p8) (3.8)
2c'cos' + 2a3'sin'

where R f is the failure ratio with a value always less than unity, Pa S atmospheric pressure,
M is a modulus number, and n is an exponent determining the rate of variation of the initial
tangential modulus, E 1 , with confining stress, 03' (see equation 4.1).

The curved nature of the failure envelope for granular materials (c' = 0) can be accounted for
by using the following logarithmic expression for '

= '1o' - Ø'Iog(c3'/p) (3.9)

where is the angle of shearing resistance for a confining stress o 3 '=p, and is the
reduction in q' for a tenfold increase in 03'.

Initially a constant value of Poisson's ratio was used, but to model the measured volumetric
strains in a triaxial test more accurately, Kuihawy and Duncan (1972) proposed the following
expression for the tangential Poisson's ratio

u - FlO9(C3'/Pa)

Ut = (3.10)
(1 - dEa)2

36
in which u is the initial Poisson's ratio at o3'=p1, F is the parameter expressing the rate of
change of u1 with O3 d is a constant to be determined during the triaxial test, and Ea = (0i-

o 3 )/E, where E is given by equation (3.8).

Instead of using equation (3.10), Duncan eta/Cl 980) preferred to model the volume change
behaviour of soils by an exponentional relationship between the bulk modulus, KB, and
confining pressure, O3 neglecting the influence of deviatoric stress, 0103

K1 = M 1 p1 ( a3 '/p1) m (3.11)

where M 1 and m are dimensionless model parameters with the same meaning as M and n in
equation (3.8). Boscardin et a/ (1990) suggested that it is more appropriate to use the mean
effective stress, p', instead of 03 ' in equation (3.11).

It can be noted that eight (or seven, when equation (3.11) is used instead of equation (3.10))
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 (Kuthawy and Duncan, 1972; Tanaka and
Nakano, 1976; Cathie and Dungar, 1978; Seco E Pinto and Maranha Das Neves, 1985; Veiga
Pinto and Maranha Das Neves, 1 985; Naylor et 8/, 1986; Khalid et a/, 1990; Saboya and
Byrne, 1993, ...). Enormous experience with the model has been accumulated and this has
enabled Duncan eta/(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
bearing in mind the difficulties associated with the laboratory testing of rockfills.

Duncan (1992) concluded that calculated deformations from finite elemeri't analyses of
embankment dams tended to be larger than the measured deformations. He primarily
attributed this difference to the stiffer response of fill materials in the field than in the
laboratory due to various reasons. However, Eisenstein and Law (1979b) pointed out that it
was particularly so in the analyses employing the hyperbolic model whose parameters were
determined on the basis of data from conventional triaxial tests. They argue that along a
triaxial stress path the soil gradually softens whereas along a constant principal stress ratio
path (e.g. K0 stress path in oedometer test) the soil gradually stiffens. The latter takes place
in a major part of the embankment during construction. Thus, the models with the parameters
derived from oedometer test can give at least reasonably accurate predictions of vertical

37
displacements which are the dominant movements in rockfill dams during construction.

It is worth mentioning that Duncan (1992) noted the inadequacy of data from triaxial tests
for the analysis of embankments which are close to plane strain conditions ( E2 =0). Again
oedometer test has the advantage because it fulfils the requirement e2 =O, and so can be
considered a special case of plane strain (Charles, 1976).

A non-linear elastic model, similar in formulation to the hyperbolic model originally proposed
by Duncan and Chang (1970) and extended by Kulhawy and Duncan (1972), was
implemented into ICFEP (see Section 5.2.1) and has been used for numerical analyses of a
number of embankment dams at Imperial College (Hamza, 1976; Dounias, 1987). However,
problems were experienced in deriving the model parameters to satisfy both oedometer and
triaxial test data. Bearing in mind the above discussion emphasis was placed on fitting the
oedometer stress-strain curve. When results of the analyses were compared against the field
measurements, it emerged that the predicted settlements were in reasonable agreement with
the measured ones, but horizontal displacements were overpredicted (see e.g. Potts et a!,
1990). According to Eisenstein and Law (1979a) a larger Young's modulus in the horizontal
direction would further overestimate horizontal displacements during dam construction. Thus,
it seems that problem cannot be solved with typically observed anisotropic properties of
rockfill materials (see also Section 3.4.1).

3.4.2.3 'K-G' Model

It was recognised long time ago (dough and Woodward, 1 967) that modelling of soils in
terms of the bulk modulus, KB, and the shear modulus, G, had some advantages over the use
of Young's modulus, E, and Poisson's ratio, u.

The 'K B -G' model proposed by Naylor (1975) assumes a variation of the bulk modulus with
mean stress and the shear modulus with both mean and deviatoric stress. In the plane strain
version of this model (see e.g. Naylor eta!, 1986) the moduli are defined as follows

= .si + Q1' (3.10)
+ PGd (3.11)
G = G 1 + aG US

where =K 0 +G/3, O$'=(O'+O3')/2, o d = al -o3, and , G 1 , g, a, PG are the model

parameters. Imposing the condition that G must tend to zero as the strength criterion is
approached the following relations can be deduced

OGI(PG) = 2sinq' (3.12)

38

G 1 /(-fi) = 2c'cosp' (3.13)

where c' and q' are the Mohr-Coulomb strength parameters constraining G 1 , a and PG.

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). It seems that a good fit to both oedometer and triaxial
test data cannot be obtained 8. This is in agreement with the previous experience gained at
Imperial College (and confirmed during the course of this research, see Section 5.3) where
a different type of 'variable' elasticity has been used.

3.4.3 Elasto-Plastic Models

It is well known that the behaviour of soils is non-elastic. In contrast to elastic materials, the
behaviour of non-elastic materials is Stress path dependent. Strains are largely irreversible, and
strain increments depend not only on stress increments but also on the state of stress and
stress history. Total strains comprise both elastic and plastic strains. Plastic strains modify
the soil structure and hence its behaviour under continued loading. This results in a non-'inear
relationships between stresses and strains. Although the elastic moduli can be adjusted
according to the magnitudes of stresses and strains (as in the above presented 'variable'
elastic models), the plastic strains ensure that the soil deformations depend on: (i) the initial
stresses and the soil structure, and (ii) the direction in which soil is loaded, I.e. whether shear
stress and normal stress are increasing or decreasing (stress path dependency) and whether
principal stresses and/or principal stress increments are rotating.

Elasto-plastic stress-strain relationships are capable of modelling the behaviour of real soils
more closely. Taking into account the influence of stress path as well as rotation of principal
stresses on soil behaviour, the elasto-plastic finite element analysis should provide a better
prediction of stresses and especially movements in an embankment dam. Unfortunately, in
spite of the existence of different elasto-plastic 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 test ng of fill materials, especially
rockfills. It is true that there is a lack of the experimental data required for parameter
determination in order to use the advanced constitutive models, such as elasto-plastic ones,
in the numerical analysis of rockfill dams.

8 Naylor (1991a) commented that this might be true for at least the Beliche Dam rockfill
where the one-dimensional compression curve was convex regarding strain axis due to a
crushing of particle contacts at higher stress levels.

39
Nevertheless, there have been attempts to characterize the behaviour of rockfill using elasto-
plastic stress-strain relationships in the past. Tanaka and Nakano (1976) and Cathie and
Dungar (1978) modelled rockfill as linear elastic perfectly plastic material. However, their
finite element analyses gave a linear elastic solution as no (or only small) yielded zones were
discovered. Significant factors of safety usually operates in rockfill dams and the modelling
of plastic behaviour at and post-peak, even with the inclusion of hardening or softening
behaviour, is unnecessary. This is of great importance only in special circumstances when a
shear failure surface develops (see e.g. Dounias et a!, 1988; Pots et 8!, 1990).

The inclusion of plasticity pro-peak has usually been associated with the modelling of
dilatancy, i.e. the tendency of well compacted granular fills to increase or soft clayey flls to
decrease its volume on shearing. There is both laboratory and field evidence that suggest
dilatancy is unlikely to occur widely in rockfill dams. Constant stress ratio tests on rockfill up
to R=3.O (El-Ruwayih, 1975) and R=4.1 (Charles, 1976) always gave compressive volume
changes. Charles (1975) reported strains that developed in two rockfill dams during
construction which indicated that dilatancy did not take place. 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 analyzed the
behaviour during construction of a clay core embankment dam using two different
constitutive models: 'variable' elastic and critical state elasto-plastic. He showed that the
deformation behaviour was sensitive to the type of model being used even though both
models produce similar triaxial test stress-strain curves.

The basic formulation of an incremental elasto-plastic stress-strain theory will be given in the
next chapter. It will only be mentioned here that there exist a large variety of elasto-plastic
models that have been proposed in the last 20-30 years to characterize the stress-strain and
strength behaviour of soils. Most of them have been conceived from the concept of critical
state soil mechanics which is outlined in numerous textbooks (see e.g. Schofield and Wroth,
1968; Wood, 1990). 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 soft materia s present in the
foundation (see e.g. Duncan, 1992, 1994). Very few examples of the application of the
critical state models to predict the behaviour of rockfill dams have been reported (Chang and
Duncan, 1977; Seco E Pinto and Maranha Das Neves, 1985; Bertrand and Lino, 1985; Naylor
eta!, 1986; Naylor eta!, 1988).

40
Naylor (1991a) questioned the suitability of the critical state models to predict the behaviour
of fill materials. Namely, these models were originally developed for sedimentary clays.
Sedimentation at a very high initial void ratios e 0 (slurry state) is presumed with density
thereafter being a unique function of stress history (see Figure 3.5.a). However, compacted
fill, even of clay, will be placed at a much lower initial void ratio. An increase in effective
stress a' still results in normal consolidation, but the amount of compression now depends
on the initial void ratio, decreasing as initial void ratio decreases (see Figure 3.5.b). The e -
log a' lines are typically curved and only converge to a straight line representing normal
consolidation from a slurry at high stresses. A similar pattern of behaviour can be recognised
for sands. They are deposited at a wide range of initial densities which control subsequent
behaviour (see Figure 3.5.c). It is only at high normal stresses, beyond the range of
engineering interest, that the compression curves for loose and dense sand converge. It is
apparent that the elasto-plastic models developed to capture the observed behaviour of sands
will be more representative for fill materials.

It is common when considering sands to distinguish between two general aspects of their
behaviour: 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. Nova and Wood (1979) attempted to link
these two aspects of sand behaviour using the critical state concept. Reasonable predictions
of observed sand behaviour in triaxial compression was obtained. However, poor predictions
for constant stress ratio paths can be expected, particularly if a wide stress range is
considered, since the model is based on the linear volumetric strain e - log p' relationship
typically observed for sedimentary clays. A power form relationship between e and p' (see
equation (3.4)) giving a straight line in the log - log p' plot is more representative for sands
and most of fills.

Elasto-plastic models for sands usually separate the effects of consolidation and shearing. To
account for different plastic deformations during consolidation and shearing so-called 'double
hardening' models have been proposed. Three constitutive models of this kind have been
examined in detail during the course of this research. They are due to Lade (1977), Vermeer
(1978) and Molenkamp (see e.g. Smith eta!, 1988). In spite of the differences in formulation
all three models are capable of capturing the various aspects of sand behaviour, notably a
power form of stress-strain relationship dur ng isotropic (and one-dimensional) consolidation.
The Lade's double hardening model has appeared to be extremely attractive because of its
ability to fit the curves from standard triax al tests over a wide range of confining pressures.
It has been chosen for numerical modelling of rockfill behaviour here. More recently Lade and
Kim (1988a) proposed another, more general model for all types of frictional materials.
Although different in formulation, it is still capable of modelling the various aspects of rockfill

41
behaviour end has also been included for the analyses reported in this thesis.

3.5 CONCLUSIONS

The finite element method can be used to reasonable predict the behaviour of embankment
dams if the stress-strain relationships of the materials involved are known. These relationships
have generally been modelled by 'variable' elastic constitutive laws, primarily because of their
simplicity and their ability to model non-linear and stress dependent behaviour of fill materials.
Some problems have been experienced with this approach, both in deriving model parameters
from available test data, and in producing displacement patterns which agree with field
measurements. The problems are greatest with rockfills. They are coarse in nature and
difficult to test in the laboratory. Consequently, the number and type of tests are limited.
While stress dependent elastic moduli derived from triaxial tests overpredict the observed
constructional movements, those derived from the oedometer test can reasonably predict
settlements at least in the central portion of a dam. This is due to the fact that a typical
stress path in an embankment dam during construction is simulated more closely by
oedometer testing.

Ideally, the stress-strain properties should match the rockfill behaviour observed in both
oedometer and triaxial tests. However, this is not possible with 'variable' elastic models.
These models cannot account for stress path dependency which is caused by non-elastic
components of soil behaviour. Alternative constitutive models incorporating pre-peak plasticity
take into account the influence of stress path. Compacted fills have a behaviour pattern
similar to that of sand. Elasto-plastic constitutive models for sands are available in the
literature, but these have not been used for numerical analysis of embankment dams. Being
capable of capturing essential characteristics of rockfill behaviour, the two Lade's elasto-
plastic constitutive models have been incorporated into ICFEP (Chapter 4) and used for
numerical analysis of two rockfill dams with an impermeable upstream membrane (Chapters
5 and 6).

42
 4
0' for one - Omnensonal
or
zITTi -
°3

!L
0: I
120e
(a) Upstream sloprng I C
clay core -------------- I
-
. A
0.
200 400 600
p'( k N/rn2)

& for
— One-d,m,ns,on,I
o compmosron
E

<]DO3
0
mm
C
C

(b) Central Clay core

-- 100 400
I- I -

600
p'(kN/m2)

0 fOr
\.. H/ p— - drer000rorral
0 3 CompmesS,On
3

one

IL Or.pmndmg

C
20 !
(c) Upstream membrane U
-- ---
3.
H • - 200 - 400 600
- plkN/m2)

Figure 3.1 Stress paths for three types of rockfill dams: (a) upstream sloping clay core,

(b) central clay core, and (c) upstream asphaltic membrane (after Charles, 1976)
P4OHR-COULOMB STRENGTH ENVELOPE

II END OF CONSTRUCTION
Triaxial stress paths
6
z
!L

200

¼Yo

100

tOO 200 300 400 500

G1' o D3
------ (kN/m T )

FIgure 3.2 Stress paths representative for the transition zones of a rockfill dam with the

central clay core (after Veiga Pinto and Maranha Das Neves, 1985)

43


Triaxiaf stress paths .f - -

Figure 3.3 Stress paths for four points along the line normal to the concrete face of a

hypothetical rockfill dam (after Mon and Pinto, 1988)

E
0
CD

3 Triaxial stress paths

a3'=500kPa a3=1,5OOkPa
a3=1.000kPa
2.5-

31 :
0.5 ----
74 1

IJ______
0 0.5 1 1.5 2
a (MPa)

Figure 3.4 Stress paths during reservoir impounding of a concrete face rockfill dam (after

Saboya and Byrne, 1993)

44


• §

— V

88
.r:;.::r:;r.:.:;:::
o .0
C C -
C.)

U,
>

(0 C,i .- (0 (0 '
— 0 0 0 0

a o,jei pio
0 It)

U) %J
§
C,, i-

ill
-

0 0)
0
0)> L)

0
I ./
EU)
•0
- C
o €
U)
;( 0 >

I I 0
I- C
(0 4-
(0 • (0 ' j ( 6
0 0 0 0

a °!J pi0

.. >5

o
> 5).

:g
C U)

I
IIII

a 'oieJ P!OA
 CHAPTER 4
LADE'S ELASTO-PLASTIC MODELS

4.1 INTRODUCTION

Two elasto-plastic constitutive models to characterize the behaviour of granular materials, one
with a single yield surface (Kim and Lade, 1988; Lade and Kim, 1 988a, 1 988b) and the other
with two intersecting yield surfaces (Lade, 1977), are presented in this chapter. The models
are based on concepts from non-linear elasticity and isotropic work-hardening plasticity
theories. The framework for the plastic behaviour consists of a yield criterion, a flow rule and
a work-hardening (-softening) law. Both models are applicable to general three-dimensional
stress conditions, but the input parameters can be derived entirely from the results of
standard laboratory tests.

Only the basic mathematical expressions for the various components of the constitutive
models are presented here. The background and arguments for employing these expressions
are contained in the references quoted above. Using these basic equations, the incremental
elasto-plastic stress-strain relationships are obtained. Their proper implementation into ICFEP
is checked against numerous comparisons already published in the literature by Lade and
others. Finally, the models are compared in terms of their capabilities to accurately capture
the observed behaviour of granular materials for various stress paths, and in terms of the ease
with which they can be employed in a large finite element program. The procedures employed
for determination of the model input parameters from standard laboratory tests will be
presented in the next chapter in connection with derivation of parameters for a low grade
rockfill used for the construction of Roadford dam.

4.2 FORMULATION OF MODELS

Both constitutive models have the same elastic formulation and failure criterion. These two
components are therefore presented first. The more complex double hardening model will then
be presented before the simpler single hardening one.

46
4.2.1 Elastic Behaviour

The purely elastic response of both models is described by isotropic elasticity and therefore
requires two independent elastic parameters. A stress dependent Young's modulus, E, and
a constant value of Poisson's ratio, u, are adopted by Lade (Lade and Nelson, 1987; Lade,
1988b). In the present implementation of the models three alternatives for the modelling of
Young's modulus are considered.

On the basis of conventional drained triaxial compression tests, the initial tangent modulus
(Janbu, 1963), or the unloading-reloading modulus (Duncan and Chang, 1970), may be
expressed as a power function of the effective confining pressure, 0'

E = M.p.(o 3'/p1 ) n (4.1)

in which p is atmospheric pressure in the same units as E and 0 3'. M is a modulus number
and n is an exponent, both dimensionless and constant at a given initial void ratio.

In order to generallse the above expression, the minor principal effective stress, 0 3 , may be
replaced by the mean effective stress, p'

E = MPa(P'/Pa)° (4.2)

where

p' = (a1'+u2'+a3')/3 = (a'+a'+a')I3 (4.3)

It can be demonstrated that the above expressions for the elastic behaviour are non-
conservative, and that energy may, therefore, be extracted or generated from certain loading
cycles (Zytinski eta!, 1978). In practice, this fact may not be important if monotonic loading
is considered but may become significant if the loading involves many stress reversals. Lade
and Nelson (1987) derived expression for Young's modulus from theoretical considerations
based on the principle of conservation of energy. Young's modulus is expressed as a power
function invoMng the first invariant of the effective stress tensor, l, and the second invariant
of the deviatoric stress tensor, J2

E = Np.i(l1/P)2 + R(J2/p.2)? (4.4)

where
R = 611 +u)I(1-2u)
l = O I '+O 2 '+03 ' = ox '+o y'+oz' (4.5)

= J2 = 1/6 . [(o 1_o2 )2 + ( or2_a3 ) 2 + ( o31_o 1 ) 21 =

= 1 I6i(o'_o') 2 + ( O'_0') 2 + ( o'-a') 21 + T 2 + T 2 + T, 2 (4.6)

47
The parameter p1 is atmospheric pressure expressed in the same units as E, l and (J2)lfl, and
the modulus number, N, and the exponent, A, are constant, dimensionless numbers.

Because Young's modulus in equation (4.4) varies with both mean effective stress and
deviator stress, a direct comparison of material parameter values for the above three
expressions defining Young's modulus can only be made at the hydrostatc axis where the
deviator stress dependency is eliminated. Equating equations (4.1), (4.2) and (4.4) produces
the following relations between M, n, N and A

M 3'N (4.7a)

n = 2A (4.7b)

4.2.2 Failure Criterion

Although sufficient as a first approximation (Bishop, 1966), the widely used Mohr-Coulomb
failure criterion does not achieve perfect agreement with observed soil failure conditions.
Other failure surfaces have been suggested in the literature which have no corners and take
into account the value of the intermediate principal effective stress, O'2'. Lade's (Lade and
Duncan, 1975) and Matsuoka and Nakai's (1974) are the best known (see Figure 4.1) and
are in better agreement with experimental observation. In order to include curvature of the
failure surface in planes containing the hydrostatic axis, Lade (1977) extended his pervious
three-dimensional failure criterion, 1 1 31l3 =const, to the form of

( l 1 /l3 - 27 )( l i /Pa) m = (4.8)

where l is the first stress invariant given by equation (4.5), and 13 is the third stress invariant
given below

13 = 0r1 . 0r2 . 03 = O'O'U' + + TTT-

- (a'rr, + a ' T, T + (4.9)

The parameters i and m are dimensionless numbers, constant at a given initial void ratio.

The failure surface defined by equation (4.8) ts shown in Figure 4.2. The apex angle and the
curvature of the failure surface increase with the values of 4 and m respectively. For m =0
the failure surface is straight and the expresson in equaion (4.8) becomes identical to that
used previously, 1 1 3/13 =const (Lade and Duncan, 1975).

In order to include the effective cohesion and tensile strength which can be sustained by

48
concrete and rock, or compacted clay fill, a translation of the principal stress space along the
hydrostatic axis can be performed as shown in Figure 4.3 (Lade, 1984). A constant stress,
is added to the normal effective stresses, a', before substitution in equation (4.8)

a' = a' + ap (4.10)

in which 'a' is a dimensionless parameter. The value of ap reflects the effect of the tensile
strength of the material.

4.2.3 Double Hardening Model

The double hardening model (Lade, 1 977) is an extension of a previously developed model
(Lade and Duncan, 1975) which used a single yield surface. This extension involves the
employment of curved yield and failure surfaces, the use of work-hardening as well as work-
softening relationships and the addition of a cap-type yield surface.

In an elasto-plastic constitutive model the total strain increment vector, e, can be divided
into two components, an elastic component, IEe, and a plastic component, ?

(4.11)

It is assumed, in the double hardening model, that the plastic component consists of two
parts

pp1p2 (4.12)

Lade (1977) named them as plastic 'expansive' and plastic 'collapse' strains.

The plastic 'expansive' strains, a?', are governed by a conical yield surface. Like the failure
surface, it has curved meridians in any plane containing the hydrostatic axis and a smoothly
rounded triangular cross-section in the deviatoric plane (see Figure 4.2). This yield surface has
the mathematical form

F1 = (l/l3 - 27)il 1 Ip) m - H 1 = 0 (4.13)

where m is a parameter describing the curvature of the surface meridians (the same m as in
equation (4.8)), and H 1 is a work-hardening parameter which defines the size of the surface
(at falure H 1 = 4, see equation (4.8)). The first stress invariant, l, and the third stress
invariant, 1 3, are given by equations (4.5) and (4.9) respectively.

49
[BL\
LODN
UNIV
The plastic potential associated with this conical yield surface (Figure 4.4.a) has the form

G1 = '1 - [27 + 2 (Pa I'1 i) m H 3 (4.14)

where 72 is a parameter which depends on H 1 and the minimum principal effective stress, o'

172 = pH 1 + R(a3 '/p8) 1 '2 + t (4.15)

in which p. R and t are dimensionless parameters.

The work-hardening (-softening) parameter H 1 is described as

H 1 = aexp(-bW 1 )(W 1 /p 8), q ^ 1 (4.16)

where W 1 is the accumulated plastic 'expansive' work calculated from

W, 1 = Z(W,1) = (4.17)

in which (aTE1l) is the plastic work done per unit volume during the strain increment vector

&". The parameters a, b and q are dependent on the minimum principal effective stress, 03'

)1/q (4.18)
a = ,1(ep8/W
pip

b = 1/(qW1) (4.19)

q = a + fl(03 'IPa) ' q^1 (4.20)

where 'e' is the base for natural logarithms and W, 1 , is the accumulated plastic work

associated with the conical yield surface at peak strength. The value of W, 1 ,, is also dependent

on the minimum principal effective stress, O3 g and is expressed as

W p1p = PPa(03'/Pa)' (4.21)

where P and I are dimensionless material parameters.

For the sake of simplicity, the work-hardening parameter, H 1 , can be rewritten in the form

(McCarron and Chen, 1988)

H 1 = )1.[.exp(1)]h/a, q ^ 1 (4.22)

where

= (4.23)

50
The yield surface corresponding to the plastic 'collapse' strains, AEs12, forms a cap on the open
end of the conical yield surface as shown in Figure 4.2. The cap-type yield surface is shaped
as a sphere with centre at the origin of the principal effective stress space. It is described in
terms of the first stress invariant, l, defined by equation (4.5), and the second stress
invariant, 12

F2 = + 2l 2 - H 2 = 0 (4.24)

where

12 = - ( o 1 'o2' + 02 0 3 + 0 3 o) =
= T,TYX + TT + r,1 r - (o'o' + o'a' + a'a') (4.25)

H 2 is the work-hardening parameter described by

H 2 = p.2LW2/(C.p.)]' (4.26)

where C and p are dimensionless material parameters, and W 2 is the accumulated plastic
work associated with the cap yield surface

W,2 = ! ( AW) = Z(0T.P2) (4.27)

The cap yield surface uses an associated flow rule, F2=G2.

A schematic diagram of the yielding process with both yield surfaces active is shown for the
triaxial plane in Figure 4.4.b, where the plastic strain increment vectors are superimposed on
the stress space. The total plastic strain increment is calculated according to equation (4.12)
as the vector sum of the two components. In order to obtain the total strain increment for the
stress change from A to B, the elastic strain component should be added (equation (4.11)).
It is worth noting that yielding resulting from outward movement of the cap does not result
in eventual failure. Failure is controlled entirely by the conical yield surface.

It may be seen that a total of fourteen (or fifteen, in the case of materials with effective
cohesion) parameters are required to characterize the behaviour of the model. Three of these
parameters are associated with the elastic response (M or N, n or A, U); two are required for
the plastic 'collapse' strain component (C and p); and nine are required for the plastic
'expansive' strain component (i), m, p. R, t, a, p. P and I).

51
4.2.4 Single Hardening Model

The single hardening constitutive model has been developed on the basis of a thorough revew
and evaluation of data from experiments on different frictional materials such as sand, clay,
concrete and rock (Kim and Lade, 1988; Lade and Kim, 1988a, 1988b).

An important feature of the model is the use of a single isotropic yield surface that describes
the locus at which the total plastic work is constant. It can be expressed in terms of stress
invariants, I, 1 and 1 3, as follows

F= 112/l2).(l1/p1)e - H = 0 (4.28)
(wl113/13 -

The expressions for L,, 1 2 and 13 are given by equations (4.5), (4.25) and (4.9) respectively.
The parameter qi is related to the curvature parameter, m, of the failure criterion (see
equation (4.8)) according to a power function

qi 1 = O.00155m 157 (4.29)

The parameter h is constant for a given material. The parameter q varies with stress level, L,
from zero at the hydrostatic axis to unity at the failure surface according to the following
expression

q = (crL)I[1-(1-a)L] (4.30)

where L 1 is defined as

L = iIi), = (1Ii 1 )(I 1 3 /l 3 - 27)(l1/p0)m (4.31)

a is a dimensionless parameter; 4 is the value of i at failure (see equation (4.8)).

The yield surface has a smoothly rounded triangular cross-section in the deviatoric plane and
curved meridians in any plane containing the hydrostatic axis as illustrated in Figure 4.5. The
roundness is defined by the parameter qi 1 and the curvature by parameters h and q.

During work-hardening the parameter H is described as

1 Stress level, L, defined by equation (4.31) (Lade and Kim, 1 988a) is different from stress
level, S. used by ICFEP (see equation (5.4) and Figure 5.4).

52

H = (27ip1 (4.32)

where C and p are model parameters, and W is the plastic work calculated from

W = !(W) = (4.33)

During work-softening the parameter H is described by an exponentional decay function

H = Aexp[-B(W/p1 )1 (4.34)

in which A and B are positive constants to be determined by setting the slope of the softening
curve equal to the negative slope of the hardening curve at the point of failure given by
equation (4.8) as indicated in Figure 4.6. The following expressions for A and B can be
derived (see Lade and Kim, 1988a)

A= (4.35)

B = (h/p)(p/W) (4.36)

where 'e' is the base for natural logarithms, and W, is the plastic work at the point of failure.

The plastic potential function differs from the yield function and non-associative flow is
consequently obtained. It can be expressed in terms of stress invariants l, '2 and l 3 as follows

G= ( qi 1 l13 Il 3 - 112/12 + c2)('i Ip a) M (4.37)

The expressions for I, 2 and 1 3 are given by equations (4.5), (4.25) and (4.9) respectively.
The parameter ip1 is defined by equation (4.29), and the parameters q,2 and p are
dimensionless constants.

The plastic potential surface is schematically illustrated in principal effective stress space in
Figure 4.7. The parameter defines the roundness of plastic potential surface in the
deviatoric plane as in the case of the yield surface described previously. The parameter
controls the intersection with the hydrostatic axis, and the exponent p determines the
curvature of meridians. Because of the condition of irreversibility (the plastic work is positive
or zero whenever a change in plastic strain occurs), it can be shown (Kim and Lade, 1988)
that these parameters must satisfy the fo lowing requirements

p >0 (4.38a)

Y2 > - ( 27ip1+3)
(4.38b)

53
The required number of parameters to be determined for a given material is eleven (or twelve,
in the case of materials with effective cohesion): three for the elastic behaviour (M or N, n
or A, u), two for the failure criterion (, and m), two for the plastic potential function (y'2 and
p), two for the yield criterion (h and a) and two for the hardening/softening function (C and
p).

4.3 IMPLEMENTATION OF MODELS

In order to incorporate any elasto-plastic constitutive model into ICFEP, it is necessary to

obtain an incremental stress - incremental strain relationship of the form

Au = DAe (4.39)

where D' is an elasto-plastic stress-strain matrix. A general procedure for developing the
stress-strain matrix for an elasto-plastic material with multiple intersecting yield surfaces was
presented by Lade and Nelson (1984). In the following this procedure is applied to the two
models described above.

The following concepts and assumptions from classical plasticity theory provide a basis for
derivation of the incremental elasto-plastic stress-strain relationships (Duncan eta!, 1977):

(1) Yield function. For any hardening/softening material the yield function is a function
not only of stresses but also of plastic strains

Flu,?) = 0 (4.40)

(2) Plastic potential function. It may be expressed as a function of the stresses

G = G(u) (4.41)

The plastic strain increments are normal to the plastic potential and they may be
expressed as

A? = A[aG/au] (4.42)

in which A is an arbitrary constant and [OG/øu] defines the outward normal to the
plastic potential surface.

(4) Incremental elastic and plastic strain. During an incremental change in stress, the

54
 total strain increment vector, e, is divided into an elastic component, LEe, and a
plastic component, ?, according to equation (4.1 1).

(5) Relationship between stress increment and elastic strain increment. The stress
increment vector, and elastic strain increment vector, fee , are related by means
of a symmetric elastic stress-strain matrix, D

= De.? (4.43)

4.3.1 Single Hardening Model

In the case of isotropic work-hardening the yield function (equation (4.40)) may be expressed
as

F(o,H) = F'(u) - H(W) = 0 (4.44)

in which F' depends on stress only, and H contains the effect of work-hardening. If the plastic
flow is to take place along the yield surface, then the 'consistency condition' requires that

= E8F/8u]T.0 (aH/aW)W = 0 (4.45)

where Lo and are the changes in stress and plastic work during the plastic flow. Noting
that

= (4.43)
= - ?, according to equation (4.11)
= according to equation (4.33)
= (4.42)

equation (4.45) can be written as

- A[aGIau]) - (aH/aW).o.T.A.[8G/80.] = 0 (4.46)

After rearranging

- ([ôFf8g]T . De + (aH/aw)•o.T ).A.[aG/aoi = 0 (4.47)

and solving for A

55


I8FIaUIT.De.Ae
(4.48)
[ØFVØgIT.De. [ØG/8o1 + (OH/8W) [aG/aoj

Substituting this expressions for A into equation (4.42)

IaG/a].[aF/au]T.De.Ae
IE = ( 4.49)
[øFI80.] T . D• IaG/] + (aH/aW) o T [ØG/øu]

and using equations (4.11) and (4.43) the incremental stress-strain relationship may be
obtained as

= (D - ) E (4.50)
[aF/au)T .De .EaG/au] + (8H/8W) EøGf8o]

where according to equation (4.39)

De.EaG/au].[aFIau]T.De
= (4.51)
EaF/aoj T .De .EaG/ao.) + (aH/aW)oT[aG/aoi

in which DP is the elasto-plastic stress-strain matrix. It should be noted that this matrix will
be symmetric only if EØF'/øo] = [ØG/øoi, i.e. only if associated flow occurs. Furthermore, since
the matrix DP depends on the current state of stress, the strains calculated in elasto-plastic
finite element analysis are dependent not only on the magnitude of stress changes but also
on the current state of stress.

Comparing equations (4.28) and (4.44)

F'(a) = ( 4I1i1/l3 - 112/l2)(l1/p)eQ (4.52)

where q is given by equation (4.30)

q = (aL)/[1-(1-a)L] (4.30)

in which L is defined by equation (4.31)

L= f7/f 1 = ( 1/fj 1 )(I 1 3/l 3 - 27)(1i/Pa)m (4.31)

Noting that qi1 , h, a, i and m are model parameters and using the chain rule, the derivatives

56
of the function F' can be written as

-(a+o)
(1
Ii -(o+o)
I /

D1 +(
3 2
1 '127)rn(11)ml Ii 1 2 /
+ D1- -(o+o)
/
1+
D 2[ 1 110
ac 13 1 2 P. P. 12 2t
2•

(4.53)
1/2 / / 21
-(ar o -t)
/12
- (o . o - 3l3-(ao-t)I1
I 2
I -(oa-t)
/ /
3
l D3 2 )mIl ] 3I3-(ao-t) •Ii
D1 . h12 1 I
+ D1•(4i1•----)
13 Pa 1

2(ot-tr,) 12 (at-t t) I

where
D1 =
D2 = 'i'1 (h + 3) . (1 1 2 /1 3) - (h + 2)(111l2)
D3 =

The plastic potential function is given by equation (4.37)

G (çU 1 l 1 3 /I3 - 1 1 2 /1 2 + W2)(li/PaY' (4.37)

Noting again that and p are model parameters and using the chain rule, the derivatives
of plastic potential function, G, can be written as

2.

(1 I / /2
+ I -(o2a-t)
Ii - (c
,2 3 / / 2
I -(oii-r)
/ /
G I +( 11 )IJ .! 1 . 4f
(.;)
L

41
+ (_L._i -(°°)
(4.54)
3a P. 12 2t 18 13 I2(0tzxtzx)

0
2t I 2(ot-.it,,)
2t, 12(ott)

where

57


D4 = i1 u+3)(l 1 2 /l 3) - (p+2)(l 1 1l2) + (p111)•çv2

The last term in the denominator of equation (4.51) may be written as

(ØH/8W9) . UT.EØG/801 = (8H/8W)pG (4.55)

where
aH/8W = (h/(pW)J.H, during hardening (4.56)
8H/8W = -(B/p 1)H, during softening (4.57)

in which H is the parameter given by equation (4.32) during hardening and equation (4.34)
during softening.

Finally, the isotropic elastic stress-strain matrix 0 . is given by

(1-v V V 0 0 0
Iv 1-u V 0 0 0
E V 11) 0 0 0 (4.58)
D'= I
(1+u)(1-2u)I 0 0 0 (1 -2v)/2 0 0
10 0 0 0 (1 -2 u)12 0
0 0 0 0 (1 -2u)12

where E is Young's modulus and u is Poisson's ratio.

4.3.2 Double Hardening Model

If the plastic flow is to take place around the intersection of two isotropically work-hardening
yield surfaces, then the 'consistency condition' requires that

= raF1 /ao.JT .Ag - (8H 1 /8w 1 ) . Aw, = 0 (4.59a)

AF2 = [ 8F2'/øojT liu - (8H 2/8W 2)AW 2 = 0 (4.59b)

where Au is the stress change taking place during plastic flow and the subscripts '1' and '2'
denote parameters relevant to the two yield surfaces.

According to equations (4.12) and (4.42), the corresponding plastic strains are given by

A? = A 1 [8G 1 ao-] + A 2• [ 8G2 aol (4.60)

58
and, according to equations (4.11) and (4.43), the associated stress change óo is given by

Au = D0 (âc - A 1 18G 1 /øo-] - A2[8G2/8o]) (4.61)

Substituting for Ao- from equation (4.61) into equations (4.59a) and (4.59b) gives

A 1 t 11 + A2 -L12 = T1 (4.62a)
A 1 L21 + A 2 L22 = T2 (4.62b)

where
L11 = [OF1 n/8o]T .D[aG 1 /auJ + (8H1/8W91).OrT.[ØG1/øo1 (4.63)
L22 = [8F2'/øu] T D [8G 2/aoj + (8H 2/aW 2) (8G2/øgJ (4.64)
L12 = [8F1'/8u]TD[8G2f8uJ (4.65)
L21 = [aF2/ao]T.De.[aGi/au1 (4.66)
T1 = [8Fi?/8o]I.D0.AE (4.67)
T2 = [8F2/8U]T.De.AE (4.68)

Solution of equations (4.62a) and (4.62b) gives

A 1 = ( L22 T1 - L12 T2)/(L11 L22 - L12L21) (4.69aJ

A2 = (L11 T2 - L21 T1 )/(L11 L22 - L12L21) (4.69b)

Substituting these equations for A into equation (4.60) and using equation (4.61), the elasto-
plastic stress-strain matrix, D, for the constitutive law with two yield surfaces can be
written in the following form

= - (1/A).De.([aG 1 /8u1.bi T + [ 8G2/8uFb2T)D (4.70)

where

A = L11 L22 - L12L21 (4.71)

b1 = L22 18F 1 1 /8u1 - L1248F2'18u] (4.72)
b2 = L11 IøF2 '/3uJ - L21•E8F1'/8oJ (4.73)

It may be seen that this matrix will be non-symmetrical for non-associative flow on any of the
yield surfaces.

In addition to the elastic stress-strain matrix, D, given by equation (4.58), the individual
components involved in equations (4.63)-(4.68) and (4.72)-(4.73) have been given by Lade
and Nelson (1984)

59
 Ii I

III I 313-(oo-r,)11
I
ai' L3 '1)m 1J +
21 3l3-(oo-4,)11
- -2 . 1 (4.74)
8a = (-13 P. P. 10 1 P.
I0J
I 2a-ti) 11

2o-tJ 11d

-(a.°-4)
(1)
1I / 2
E27+ii2()1J -(aoy-t,) (4.75)
= E3.l+in1 2 .( .!! ) m11 1 +
Ii

0
tOJ

(H1/p)1p,/(qWp1)-bp1J (4.76)
UT .EaG 1 rnoi = 3G 1 + mi2(p/I1)m'l3 (4.77)
= 0G 2 /8o = r,2.r,JT (4.78)
8l-I2/aW = [p0/(Cp)](H2/p2)1 (4.79)
= 2G2 (4.80)

The double hardening model was initially implemented following the suggestions by Lade and
Nelson (1984) presented above. However, it was found that this procedure resulted in failure
to satisfy the 'consistency condition'. This occurred when the conical yield surface, F 1 , was
active. Inspection of equation (4.16), defining the work-hardening law employed with the
conical yield surface

H 1 = aaxp(-bWp1 )(Wp,/p)', q ^ 1 (4.16)

indicates that the hardening parameter H 1 is not only dependent on the plastic work history
but also on the stresses. Namely, the parameters a, b and q are not 'strictly' constants
because they are dependent on the minimum principal effective stress, 03', according to
equations (4.1 8)-(4.20).

Therefore, the derivatives of the conical yield function, F 1 , should be written as

60
 aF 1 /au = equation (4.74) - (aH 1 /aa)[aa/8u] - (OH 1 /8b)[øb/8 g] -

- (OH1/8q)18q/8u] (4.81a)
or

8F1/øu = equation (4.74) - (8H1/8q)øq/8u] (4.81 b)

if the work-hardening law is expressed by equation (4.22) instead of equation (4.16)

H1 = ,1[exp(1-)J, q ^ 1 (4.22)

in which

q = a + P( c73'/Pa), q ^ 1 (4.20)

The minimum principal effective stress, 03 ', can be expressed in terms of stress invariants
(Nayak and Zienkiewicz, 1972) as

03 = p' + (2/V3)Jsin(9-2n/3) (4.82)

where
p' is given by equation (4.3)
J is given by equation (4.6)
0 = (1/3)sin'[-(3V3/2)(lsI/J3)], -ff16 < 0 ^ n/6 (4.83)
in which

I SI = det s = + - - sr 2 - (4.84)

s is the deviatoric stress tensor in the form

sx txy tzx

$ = t xy S y t). (4.85)

t yz S

Using the chain rule it can be shown that

-(8H 1 l8q)[8q/8o] = ( H 1/q2 )(fl/p8) ln [ exp ( 1 -)]•L8a3/8u] (4.86)

and from Nyaoro (1989)

8a3 'iau = AX + BY + CZ (4.87)

in which

A = 1/3 (4.88)

61


C = -(1IJ2)cos(0+2n'/3)Icos30 (4.89)

B = -(1 I(2J 2 )1(3C s I-(21V3)Jsin(O+ 2,i/3fl (4.90)
X = [111100101T
(4.91)

I= (4.92)

I •+r+t,)/3

(4.93)
z=

nspection of equations (4.89) and (4.90) indicate singularities when 0= ±ir/6. Investigation
of the trend of the coefficients B and C when 0 approaches ±n-16 leads to (Nyaoro, 1989)

(1) Ate = -ff16

C = 1/(3J), B -E1/(2J2)1E3CIsI-(23)Jsin(0+2n/3)]
(2) At 9 +17/6
C = 0, B [1/(V3J)1sin(8+2,r/3)

Similarly, in deriving the derivatives of the plastic potential function G1

G1 = 1 [27 + F 2 (Pa/I i ) m ] •I 3 (4.14)

Lade and Nelson (1984) assumed that parameter is a constant. However, the parameter

'72' according to equation (4.15)

72 = pH, + R( U3 /PaYU2 + t (4.15)

depends on the minimum principal effective stress, a3'.

Therefore, the derivatives of the plastic potential function G 1 should be written as

8G,/ao = equation (4.75) + (8GI8,2)[8,2/8o] (4.86)

Using the chain rule and the work-hardening law of the form expressed by equation (4.22)
the additional term in equation (4.86) can be written as

62
 (8G, /81) 2 ) • 101) 2 180] = (8G 1 I8/j2) i(81)2/8H 1 ) ( 8H 1 /Oq) (8q/8o 3') +

+ (81) 2 /8a3 'fl• [8a3'/0o] (4.87)

where
8 G1/81)2 -(p/I1)mI3 (4.88)
81)2/aH, = p (4.89)
8H 1 /Oq = -(H1/q2)In[exp(1-)J (4.90)
8q/8o 3 ' = P/Pa (4.91)
81)2/803 = (4.92)
Do 3 /8u is given by equations (4.87)-(4.93)

Finally, adopting the simpler expression for the work-hardening law given by equation (4.22),
equation (4.76) should be written as

= (OH 1 /8fi(8/8W 1 ) = ( H1/q)[(1/)-1](1 /W1) (4.93)

where W is given by equation (4.21).

When all the above changes were included, the only difficulty found in implementing the
double hardening model was associated with the hardening parameter (equations (4.22) and
(4.23)). For low values of , the model response to shearing is essentially elastic and the
conical yield surface expands rapidly. To eliminate numerical difficulties, McCarron and Chen
(1988) suggested that the minimum initial value of should be 0.001. For the majority of
applications this value is quite appropriate. However, it has been found that in some cases
a higher value of should be adopted. The problem is connected with the 'condition of
irreversibility' and will be further discussed in the next chapter. In order to avoid the possible
difficulties with numerical instabilities of this kind, the initial value of is included as an
additional model parameter.

4.4 VALIDATION OF MODELS

The implementations of the two Lade's elasto-plastic constitutive models have been validated
by comparing predictions from ICFEP with those published in the literature. For both models
idealized (i.e. no end effects), single eement drained triaxial compression tests were
considered. In these tests, the confining pressure, 0 3', was held constant while increments
of compressive axial strain, AE, were app ied. The results are presented as plots of principal
stress ratio, O i '/o'', and volumetric strain, e, against axial strain, Ea.

For validation of the single hardening model the predictions by Lade and Kim (1 988b) for Fine
Silica sand are chosen. Model parameters are listed in Table 4.1 and comparisons for three

63
different values of confining pressures (0 3' = 100, 200 and 500kPa) are made in Figure 4.8.
It can be seen that ICFEP reproduced the same soil response as Lade and Kim (1 988b).

The double hardening model has been validated against the predictions made by McCarron
and Chen (1988) for Hostun sand. Model parameters are given in Table 4.2. Comparisons are
presented in Figure 4.9 for values of confining pressure 03' = 203, 350 and 500kPa. The lines
in this figure represent the predictions published by McCarron and Chen (1 988) and the points
are ICFEP predictions. The open points are obtained using the incrementalization procedure
as outlined by Lade and Ne son (1984). Agreement with the published predictions are perfect
indicating that McCarron and Chen (1988) followed the same incrementalization procedure
and associated expressions. The full points indicate the implications of the additional terms
in equations (4.81) and (4.86) as discussed in the pervious section. There is practically no
difference in the predictions for the lowest value of confining pressure (o31=2O3kPa).
However, the difference becomes significant as the value of confining pressure is increased
(03'= 350 and 500kPa). It was mentioned earlier that the incrementalization procedure as
outlined by Lade and Nelson (1984) occasionally failed. It is felt that this is a consequence
of failing to satisfy the 'consistency condition'. In the remainder of the thesis the 'correct'
incrementalization procedure has always been used.

4.5 COMPARISON OF MODELS

The models have been extensively used to predict the behaviour of various laboratory tests.
It has been demonstrated that both models are able to capture several essential
characteristics of the strength and stress-strain behaviour of granular soils observed in
experimental investigations (Lade, 1977, 1978, 1981, 1988a; Lade and Duncan, 1975,
1976; Lade and Kim, 1988b; McCarron and Chen, 1988): (i) non-linearity, (ii) curved failure
envelope, (iii) the effects of the intermediate principal effective stress on both strength and
stress-strain behaviour, (iv) stress-path dependency, (v) shear-dilatancy and its variation over
a range of confining stresses, (vi) strain-softening, (vii) pore pressures and effective stress-
paths for undrained conditions, (viii) coincidence of strain increment and strss increment
axes at low stress levels (elastic behaviour) with transition to coincidence of strain increment
and stress axes at high stress levels (plastic behaviour). Due to the assumption of isotropic
hardening the models cannot predict with accuracy soil response during: (i) large stress
reversals, (ii) cyclic loading, (iii) the behaviour of initially anisotropic material.

The models have been 'qualitatively' compared by Lade and Pradel (1989 . The single
hardening model employs a single yield surface describing a contour of constant plastic work.
The consequences of this are threefold. Firstly, the absence of two intersecting yield surfaces
makes the model simpler. Secondly, it is mathematically more consistent because of the direct

64
connection between yielding and plastic work resulting in a unique hardening curve. Thirdly,
it is more efficient for applications involvng large computations usually associated with real
boundary value problems. It is schematically illustrated in Figure 4.10. It can be seen that the
double hardening model involves four different combinations of active and inactive yield
surfaces, whereas the single hardening model incorporates only two possibilities (loading and
unloading). Furthermore, special care must be exercised to avoid numerical instabilities
associated with the double hardening model because of the presence of the singular corner
at the intersection of two yield surfaces.

The two models respond differently to the same increment in stress. This is due to different
definitions of yielding and different shapes of the plastic potential surfaces employed. The
single hardening model is not so sensitive to stress path directions because of the specific
shape of its plastic potential surface (very little change in gradient over a large stress level
range, see Figure 4.7.b). It has been shown that both of the models are capable of capturing
the behaviour of soils with equally good accuracy at the higher stress levels (near to failure),
where the plastic strains are large and the effect of stress-path direction is negligible.
However, at the lower stress levels where the plastic strains are rather small and stress-path
sensitive, the single hardening model exhibits poorer capabilities. Lade and Pradel (1989)
pointed out that a small contribution to the plastic volumetric strains is missing in the single
hardening model when operating near the hydrostatic axis. This point is illustrated in Figure
4.10. When the stress increment vector, Ao-, is directed outwards and towards the
hydrostatic axis, a plastic strain increment vector A?2 perpendicular to the cap is produced
in the case of the double hardening model whereas the yield surface in the single hardening
model is not even active. This can be of sgnificance in the analyses of a real boundary value
problem as will be indicated in the next two chapters where the models will be further
discussed and 'quantitatively' compared.

65
 Parameter Value Component

Modulus Number, M 11 70.0 Elastic

Exponent, n 0.53 Elastic

Poisson's Ratio, u 0.20 Elastic

Failure Constant, 24.7 Failure

Failure Exponent, m 0.10 Failure

Pl.Potent.Const., ç2 -3.69 Plastic

Pl.Potent.Exprit., p 2.26 Plastic

Work-Hard.Const., C 0.000324 Plastic

Work-Hard.Expnt., p 1 .25 Plastic

Yield Exponent, h 0.355 Plastic

Yield Constant, a 0.515 Plastic

Table 4.1 Single hardening model parameters for

Fine Silica sand (after Lade and Kim, 1988b)

Parameter
j Value Component

Modulus Number, M 1750.0 Elastic

Exponent, n 0.44 Elastic

Poisson's Ratio, u 0.27 Elastic

Collapse Modulus, C 0.000035 Pl.Col.

Collapse Exponent, p 1 .30 Pl.Col.

Failure Constant, , 43.0 Failure

Failure Exponent, m 0.093 Failure

P1 .Potent.Const., p 0.55 P1 .Exp.

Pl.Potent.Const., R -1.0 Pl.Exp.

Pl.Potent.Const., t 0.4 Pl.Exp.

Work-Hard.Const., a 3.60 Pl.Exp.

Work-Hard.Const., -0.28 Pl.Exp.

Work-Hard.Const., P 0.085 Pl.Exp.

Work-Hard.Expnt., I 1.22 Pl.Exp.

Table 4.2 Double hardening model parameters for

Hostun sand (after McCarron and Chen, 1988)

66
 Figure 4.1 Failure surfaces in the deviatoric plane
(after Gens and Potts, 1988)

Failure surface
Yield

nslon
itatic ax

(a) (bI

Figure 4.2 Double hardening model - Failure and yield surfaces in

(a) triaxial, and (b) deviatoric planes (after Lade, 1977)

67
 FAILURE SUR FACE

COMPR ESSION
HYDROSTATIC AXIS

EXTENSION

FAILURE SUP FAC

pa
P
-[-- -V (o-3 a P)
JNIAXIAL TENSILE STRENGTH
a

Figure 4.3 Translation of principal stress space along hydrostatic axis

to include effect of tensile strength in failure criterion
(after Lade, 1984)

(0) (b)
TOTAl. PLASTIC STRAIN INCREMENT
i
PLASTIC EXPANSIVE STRAIN INCREMENT
PLASTIC EXPANSIVE
/STRAIN INCREMENT

\9
<90° YIELDSURFACE
PLASTIC COLLAPSE
PLASTIC POTENTIAL STRAIN INCREMENT
SURFACE A
CURRENT STRESS POINT
/
HYDROSTATIC
7/I AXIS
/1 HYDROSTATIC
AXIS

V2 2

Figure 4.4 Double hardening model - Schematic diagrams of (a) yield and plastic potential
surfaces for plastic 'expansive' strains, and (b) yielding process with both yield
surfaces activated and combination of plastic strain increments (after Lade and
Nelson, 1984)

68
 • dc?

dq

Plane

c,

Figure 4.5 Single hardening model - (a) Schematic illustration of the yield surface in principal
stress space (after Lade and Kim, 1 988a), and (b) contours in triaxial plane (after Lade
and Kim, 1988b)
(p

d(Wp) ening Cue

H = (H),.,.
(or q = I

(W1p1)

(W/p )pk

Figure 4.6 Single hardening model - Modelling of work-hardening

and work-softening (after Lade and Kim, 1 988a)

c'. df (a)

Hydrostatic
Axis

Plane

dtj'

Figure 4.7 Single hardening model - (a) Schematic illustration of the plastic potential surface
in principal stress space (after Kim and Lade, 1988), and (b) contours in triaxial plane
(after Lade and Kim, 1988b)

69


4 I_______________________
C'.)
b
'3-
0
Ca

0 - -i
a)
C/)
Ca
0.
0
C
2 ____

Confining Pressure
• lOOkPa - ICFEP
• 200kPa - ICFEP
I A 500kPa - ICFEP
- lOOkPa - Lade&Kim, 1988b
- 200kPa-Lade&Kim, 1988b
-5
500kPa - Lade&Kim, 1988b
>
C

0
a,
E

>5
0 5 10 15 20 25
Axial Strain, ca (°"1

Figure 4.8 Single hardening model - Predicted behaviour of Fine Silica sand
in drained triaxial compression tests

70


I —. U
I T ----
- —E__ _—li:
4 - ---------- - - - I--- c ________
b

0
a,

(0
(0
a,
(I)

Ca
0.
C)
C
Confining Pressure
2 • 2O3kPa - ICFEP --
• 35OkPa - ICFEP
A 500kPa -
- 2O3kPa - McCarron&Chen, 1988
- 35OkPa - McCarron&Chen, 1988
1 I 500kPa - McCarron&Chen, 1988
2O3kPa - ICFEP, Sig3=const
____________ U 35OkPa - ICFEP, S1g3=const
, 500kPa - ICFEP, Sig3=const

-2
I I I •

- I
I
>
C., -1
C I I I
(a I I - I - i A
(I) I .-
C.)
I -I
I I •-I A I
1) O
E
- • I
- 11
A
0
>
_--e------

1 - --------------------------------

2
o 1 2 3 4 5
Axial Strain, [%]

Figure 4.9 Double hardening model - Predicted behaviour of Hostun sand

in drained triaxial compression tests

71
 a1' /
/ ®)

// /0
//@ / 0
/
----
Ii-'
1
__- __2-
'42

(a) (b)

Figure 4.10 Possibilities of yield surfaces activities in (a) double hardening model, and
(b) single hardening model (after Lade and Pradel, 1989)

72
 CHAPTER 5
NUMERICAL ANALYSIS OF ROADFORD DAM

5.1 INTRODUCTION

Roadford dam is a 41m high rockfill dam with an upstream membrane of asphaltic concrete
that has recently been constructed in the U.K. A typical cross-section of the embankment is
shown in Figure 5.1.a. The embankment was founded on broken, weathered sedimentary
rocks of the upper Carboniferous age. The same rocks from the reservoir basin were used for
the construction of the dam. The fill may be described as a low grade rockfill composed of
mudstone, siltstone and sandstone. Details of the design arid the rockfiH used in the
construction of the dam are given by Wilson and Evans (1990).

An important design detail of the dam is the junction of the asphaltic membrane and the cut-
off structure at the upstream toe of the dam. To minimise differential movements in this
region a zone of stiffer sand waste fill was used immediately against the cut-off structure (see
Figure 5.lb). A grout curtain approximately 40m deep is located below the cut-off structure.
Drainage measures included drainage layers located on top of the foundation and the
upstream face of the dam just beneath the asphaltic concrete membrane.

Large scale tests were carried out at the Building Research Establishment (BRE) to determine
the relevant properties of the low grade rockfill and sand waste fill. These included
investigations of compressibility in a im diameter oedometer and rockfill strength (and
compressibility) by drained triaxial compression tests on 0.23m diameter samples. The
stiffness and strength parameters derived from these tests have then been used in finite
element analyses to predict the behaviour of the dam both during construction and
subsequent reservoir impounding. Three constitutive models have been used. One of these
was a 'simple' non-linear elastic perfectly plastic model. The other two were the 'complex'
Lade's work-hardening (-softening) elasto-plastic models described in the previous chapter.
A comprehensive range of geotechnical instruments were installed to monitor embankment
performance (see Figure 5.la) enabling the predictions made by the finite element analys s
to be compared with the observed behaviour. This comparison highlighted the superior
performance of the elasto-plastic models.

73
5.2 DETERMINATION OF MODEL PARAMETERS

The model parameters are derived from laboratory test data for the low grade rockfill and sand
waste fill.

The model parameters for the low grade rockfill are determined using the following laboratory
tests carried out at BRE: (I) one large oedometer test on a 1 m diameter and O.5m high heavily
compacted sample using a maximum particle size of 125mm, and (ii) four drained triaxial
compression tests on O.23m diameter and 0.5m high samples using a maximum particle size
of 38mm. The apparatus, methods of sample preparation and testing procedures are
described in more detail by Penman and Charles (1976) and Charles and Watts (1980).

The rockfill was heavily compacted at approximately the field placement water content using
a hand held electric vibrating hammer to produce a dry density similar to that anticipated in
the embankment. The initial dry unit weight, Vd' and moisture content, w, of the samples are
given in Table 5.1. Results of drained triaxial compression tests are reproduced in Figure 5.2
as plots of deviatoric stress, O1O3, versus axial strain, e, and volume strain, e,, versus axial
strain, E8 . Results of the oedometer test are reproduced in Figure 5.3 as a plot of axial stress,
as', versus volumetric strain, e. In the same figure results of a large oedometer test on lghtly
compacted Roadford rockfill are also presented. The sample had a lower initial dry density
(see Table 5.1). At o' = 76OkPa this sample was flooded and a collapse compression of 2.6%
was observed. The dashed lines on the figure are estimates, for the full range of axial stress,
o-', of both the behaviour of a sample that was not inundated at all and a sample that was
inundated before any load was applied.

Only one oedometer test for sand waste fill was available. The initial condition of this large
sample is given in Table 5.1 and the results are reproduced in Figure 5.3. As no other
laboratory test data were available for this material the strength had to be assumed.

In the following the parameters will be derived for the three constitutive models.

5.2.1 Non-linear Elastic Perfectly Plastic Model

A generalized non-linear elastic perfectly plastic model incorporating a Mohr-Coulomb failure

criterion (cohesion c', angle of shearing resistance ') as a yield surface was used for both
the rockfill and sand waste fill. Non-associated plasticity was assumed with a value of the
angle of dilation i equal to " 4. However, the actual value adopted is of little significance
since none of the fill in the dam reaches failure (see Section 5.4). Thus, essentially only the
non-linear elastic part of the model will be active. In the remainder of this chapter the model

74
will be referred to as non-linear elastic only.

The elastic parameters vary according to the following equations

E = E(1 - ABS)/(1 +BS) (5.1)

E, = E0 [ ( p ' +pa)/palC (5.2)

u = (I - Liog[(p'+pe)/p1)(1 - SI + 0.49S (5.3)

where E is Young's modulus, u is Poisson's ratio, p is atmospheric pressure, p' is the mean
effective stress given by equation (4.3), E is the initial Young's modulus at p' =0, and
A,B,C,I and L are constants. The shear stress level S represents the proportion of the shear
strength mobilized (see Figure 5.4) and varies from zero under isotropic stress (J =0) to unity
at failure (plastic yielding). It is defined as

S = J/J = J/[(p'+a)G(8)] (5.4)

where
p' and J are given by equations (4.3) and (4.6)

G(9) = sin'/[cosO + (1/V3)sin9sin'] (5.5)

in which

6 = tan 1 [(2b - 1)/v'3] (5.6)

b = (a2 ' a3 )I(a1 - 03) (5.7)

a is the intercept of the failure (yield) surface with the p' axis (see Figure 5.4) and is given
by

a= c'/G(8=0) = c'/sin' (5.8)

where c' is the cohesion intercept at a Lode angle 6=0, i.e. 02 ' = (a" + a3 ')12. Since a is kept
constant in the model, c' varies with Lode angle and the intermediate principal effective
stress, a2 , to a small extent.

During unloading the Young's modulus is given by

(5.9)
Eu r = HE,

where H is a constant and E is defined by equation (5.2). Unloading is defined as a reduction

in the value of J (see equation (4.6)).

The non-linear elastic model parameters can be derived entirely from a suite of standard

75
laboratory drained triaxial tests at different confining pressure, 03'. In order to satisfy the
oedometer test data, if available, some adjustments of the model parameters are usually
necessary.

The model parameters for Roadford rockfill are derived to give the best overall fit to both
triaxial and oedometer test curves. They are listed in Table 5.2 together with the model
parameters for the sand waste fill. The latter match the compressibility observed in an
oedometer test, and the strength assumed is arbitrary.

Stifler response on unloading was adopted as is usual for soils; the unloading Young's
modulus, E,, is made independent of S (see equation (5.9)), with a value equal to the initial
tangent modulus appropriate to the current value of mean effective stress, p' (H = 1.0).

5.2.2 Lade's Double Hardening Model

The double hardening model parameters can be derived from the results of one isotropic
compression test and at least two conventional drained triaxial compression tests. The
manner in which these tests are used in evaluating the model parameters has been described
by Lade (1977, 1981, 1988a).

5.2.2.1 Elastic Parameters

To evaluate Poisson's ratio, u, and the pressure dependent Young's modulus, E, according
to any of equations (4.1), (4.2) or (4.4), unloading-reloading cycles from conventional triaxial
and/or isotropic tests are needed. In the absence of these cycles for Roadford rockfill, the
Young's modulus is estimated on the basis of the initial slope of the triaxial compression
stress-strain curves using equation (4.1). By plotting E 1/p versus O 3'/P on log-log scale, as
shown in Figure 5.5, the value of M is determined as the intercept between the fitted straight
line and the vertical line corresponding (03'/Pa) =1. The slope of the straight line corresponds
to the exponent, n. The value of Poisson's ratio has often be found to be close to 0.20 for
most granular materials (Lade and Nelson, 1987; Lade, 1988b). A higher value, u=O.25, is
adopted here principally because of the need to model a more realistic value of K 0 in one-
dimensional compression as will be dscussed later in Section 5.3.

5.2.2.2 Failure Parameters

The model parameters defining the falure criterion given by equation (4.8) can be determined
by plotting (l1l 3-27) versus (pJ1 1 at failure on a log-log diagram as shown in Figure 5.6. On
this diagram is the intercept with (p31l1 ) = 1 and m is the slope of the adopted straight line.

76
5.2.2.3 Plastic 'Collapse' Parameters

The evaluation of the work-hardening parameters associated with the spherical yield surface
(the collapse modulus, C, and the collapse exponent, p) requires an isotropic compression
test. However, this type of test was not available for Roadford rockfill. A one-dimensional
compression test may be used instead, if appropriate assumptions about the K0-value and the
compressibility of the rockfill material are made. In the first instance K0 =0.40 was assumed
and the compressibility of the rockfill was assumed to depend only on mean effective stress,
p' (see Footnote 6 of Chapter 3). However, the resulting response (after all parameters were
determined) in one-dimensional compression was too stiff. To obtain better agreement, it was
necessary to take a value of K 0 =0.30 and assume that the compressibility in an isotropic
compression test was some 20% less than in a one-dimensional compression test. There is
some experirriental evidence suggesting that the compressibility in isotropic compression is
less than in one-dimensional compression'. It may be noted that the parameters associated
with the conical yield surface did not have a significant influence on the rockfill response in
one-dimensional compression.

When the plastic 'collapse' strains are calculated from the assumed isotropic compression test
by subtracting the corresponding elastic strains the diagram in Figure 5.7 is obtained. In this
diagram the relationship defined by equation (4.26) between W 2 = 3 . Z(g3' . A63P2) ( according
to equation (4.27)) and H2 = 3 • a3 ' 2 (according to equation (4.24)) is plotted on log-log scale,
and the parameter C is given by the intercept with (H 2 1p12) = 1 and p is the slope of the
adopted straight line.

5.2.2.4 Plastic 'Expansive' Parameters

The relationship given by equation (4.16) between H, = (l i 3 /l 3 27)(l i Ipa) m (according to

equation (4.13)) and W, = Z(o T AE'') (equation (4.17)) is shown in Figure 5.8 for the triaxial
compression tests at three different confining pressures. The plastic 'expansive' strain
increments, A?', are calculated by subtracting the elastic, A?, and plastic 'collapse' strain
increments, A?2, from the total strain increments, Af, according to equations (4.11) and
(4.12). The parameter q is determined for a particular value of the confining pressure, 03',
according to (Lade, 1977)

'Charles (1976) reported the results of drained triaxial tests on a sandstone rockfill at
different stress ratio R=o 1 '/o 31 . For A between 1.25 and 3.0 the relationship between and
p' was nearly independent of R, but the compressibility was higher than for an isotropic
compression test (R=1.0).

77


q= log(W 1 /\W 160) - (1-W 1 /\W 1 )iog e

(5.10)
log(i1/H)

where 'e' is the base of the natural logarithm, and and (W 1 ,H 1 ) are two sets of
corresponding values from the relation between H 1 and W,1 in Figure 5.8. These two points
correspond to the peak point and the point at 60% of i on The work-hardening part of the
H 1 -W 1 relation as indicated in Figure 5.8. The parameter q varies with conf fling pressure, 03

according to equation (4.20). The work-hardening parameters andp are evaluated as shown
in Figure 5.9. The total plastic work associated with the conical yield surface at the point of
failure W is also dependent on confining pressure, 03 ', acoDrding to equation (4.21). The
work-hardening parameters P and I are evaluated as shown in Figure 5.10. Using the relation
between H 1 and W 1 in equation (4.16) and the appropriate values of a, b and q given by
equations (4.18), (4.19) and (4.20) respectively, the lines in Figure 5.8 are obtained. It can
be seen that the expression in equation (4.16) reasonably models the observed rockfill
behaviour for three different values of confining pressure 0 3 ' = 96, 266 and 697kPa.

The value '12 of the non-associative flow rule (see equation (4.14)) can be evaluated from the
results of triaxial compression tests using the following expression (Lade, 1981)

3(1 +u'11 1 2 - 2703 '(al ' + u°o3)

'12 = (5.11)
(Pa/l l) m [03'( 01' + c#o 3 ') - (l/l1 )m(1 +u)J

where tF is a function of the plastic 'expansive' strain increments determined by subtracting

the elastic and plastic 'collapse' strain increments from the total ones as before

= - (E3 IAe1 ) (5.12)

The variation of with H 1 is shown in Figure 5.11 for the three drained tr axial compression
tests on Roadford rockfill. The variation appears to be independent of the confining pressure,
0 3 '. and the value of H in equaton (4.15) is therefore zero. The values of p and t are indicated
in Figure 5.11.

It can be seen that the data for the plastic potential surface are well above the adopted line
at lower stress levels indicating the plastic expansive volumetric strans from the beginning
of the triaxial tests. However, these strains are very smal, arid, adopting the parameter
=0.001 (see Section 4.3.2) some of them are neglected. The condition of irreversibility

UT.EPl ^ 0

78
requires that (Lade, 1 988a)

^ [3/(3-mftH (5.13)

where m is the exponent in the failure criterion (see equation (4.8)) and H 1 is given by
equation (4.16) or (4.22). It can be shown that for f=O.001, 1),26O.O, p-O.35, t=12.O
and any value of 0 3 ' the condition expressed by relationship (5.13) is satisfied. However, in
some cases depending on the particular values of , ,), 1)2 and 03 numerical instabilities may
be caused as discussed previously in Section 4.3.2. The easiest way to overcome them is to
increase the value . That is why the value of is included in ICFEP as an additional input
parameter.

5.2.2.5 Summary of Double Hardening Model Parameters

The double hardening model parameters are not completely independent of each other. In the
absence of an isotropic compression test for rockfill, some assumptions are necessary to
make use of the data from one-dimensional compression test. Usually, unloading-reloading
cycles are not available from standard rockfill laboratory tests, and this aggravates the
problem of estimating the elastic parameters. Thus, it is hardly surprising that the model
parameters for rockfill type material cannot be determined directly but only by trial and error.
Two trials have been needed for Roadford rockfill. The fourteen model parameters are listed
in Table 5.3. The determination of thirteen of these parameters is illustrated in Figures 5.5-
5.11 and a constant value of Poisson's ratio is adopted. Some of these values are zero and
none of the parameters have dimensions. The model parameters for sand waste fill are also
presented in Table 5.3. They are based on data given in the literature (Lade, 1977, 1981,
1 988a; McCarron and Chen, 1 988), but attention has been paid to model the only available
test for this material: a large oedorneter compression test.

5.2.3 Lade's Single Hardening Model

The specific procedures employed for determination of the single hardening model parameters
are outlined by Kim and Lade (1 988) and Lade and Kim (1 988a). Again, these parameters can
be determined from results of isotropic and conventional drained triaxial compression tests.

5.2.3.1 Elastic and Failure Parameters

As mentioned previously both models have the same elastic formulation and failure criterion.
The model parameters required in the failure criterion are the same as for the double
hardening model and they are shown in Figure 5.6. An effort has been made to keep the

79


elastic parameters the same as well. However, the definitions of the yield and plastic potential
surfaces in the two models are different, resulting in different elastic responses of the models.
Lade summarized parameter values for both the double (Lade, 1977) and the single hardening
model (Lade and Kim, 1988b) for some granular materials. Although the value of Poisson's
ratio was the same (u=O.20), the elastic modulus, M, and exponent, n, were quite different
for the two models for the same material. It may be noted that the elastic response in the
single hardening model is always softer, in some cases by more than a factor of two. For
Roadford rockfill the elastic response in the single hardening model is softer too, but not to
such a great extent. A lower value of Poisson's ratio is adopted, u=O.20, instead of u=O.25
for the double hardening model. The adopted modulus number, M, and exponent, n, are
shown in Figure 5.12 together with the observed initial Young's moduli from triaxial tests.

5.2.3.2 Work-hardening Parameters

To obtain the hardening parameters, C and p, an isotropic compression test is necessary. As

such a test is not available for Roadford rockfill, a one-dimensional compression test has been
used instead, with appropriate assumptions about the K0-value and the compressibility of the
rockfill material in an isotropic compression test (see Section 5.2.2.3). Figure 5.13 shows the
plastic work W=3 (a3'Ae3 ) plotted against the value of the first stress invariant l =3o3'
for the assumed isotropic compression test. This relationship is modelled as

Wp = Cp(l 1 IP) (5.14)

in which C is the intercept at ( 1 1 1p1) = 1 and p is the slope of the adopted straight line.

5.2.3.3 Yield Parameters

The yield exponent, h, is determined on the basis that the plastic work is constant along a
yield surface defined by equation (4.28). For two stress points, A on the hydrostatic axis and
B on the failure surface, the following expression is obtained for h (Lade and Kim, 1 988a)

2
logEe( qs1 i1 3 /l3 - ' lB '2B I(27 w1 +3)]
h = (5.15)
log(l/l)

in which 'e' is the base of the natural logarithm. Using the data from large triaxial
compression tests on Roadford rockf I at three different confining pressures 0 3' = 96, 266 and
697kPa an average value of h=O.875 is obtained.

To determine the yield constant, a, the variation of q with L according to equation (4.30)

80
should be obtained. This is done in Figure 5.14 where the best-fitting value of ais determined
using the value of q at L=0.80. The points in Figure 5.14 are test data and they are obtained
calculating L according to equation (4.31) and q according to equations (4.28) and (4.32)
giving

(27ip1 +3).[W/(C.p)P

q = In (5.16)
(gv1113/13 - 112/li).(li/p)h

5.2.3.4 Plastic Potential Parameters

The last parameters to be evaluated are the plastic potential constant, p2 , and exponent, p.
This is done by using drained triaxial compression test data and a relationship of the form
(Kim and Lade, 1988)

= (lIp) - (5.17)
where
= [11(1 +u'i][(1 1 3/1 22 )(a1 ' +O3 + + qi,(1 1 4/1 32)(o 1 'o 3 ' + Ua3 )] -

- 3 1 l 1 3/l3 )+ 2.(112/12) (5.18)

= '(1/l) - ( 112/12) (5.19)

= - (E3/E1) (5.20)

The plastic strain increments, and are determined by subtracting the corresponding
elastic strain increments, e3 and Ie1 , from the total strain increments, óe3 and E1,

according to equation (4.11). Figure 5.15 shows the relationship between and for
Roadford rockfill where the points are obtained from three triaxial compression tests at
different confining pressure 0 3' = 96, 266 and 697kPa. The value of -qi2 is the intercept value
of at (=0, and the value of lip is the slope of the adopted straight line.

5.2.3.5 Summary of Single Hardening Model Parameters

The single hardening model parameters are more interdependent than those of the double
hardening model. Thus, in spite of the fact that a fewer number of parameters for the single
hardening model is needed to characterize soil behaviour, greater care must be exercised in
their derivation. They are listed in Table 5.4 together with the model parameters for sand
waste fill. The latter are adopted from the literature (Lade and Kim, '1988b) ensuring the
modelling of large oedometer compression test which is the only test available for sand waste
fill.

81
5.3 MODEL PREDICTIONS FOR IDEALIZED, SINGLE ELEMENT TESTS

The parameter values determined in the pervious section for the three different constitutive
models are used here to predict the response of Roadford rockfill and sand waste fill in
standard oedometer and triaxial laboratory tests. These predictions are then compared with
laboratory measured behaviour.

The predictions have been made assuming that the laboratory tests have no end or side
effects and that the soil behaves in a uniform manner throughout the sample. Analyses have
therefore been performed using a single axisymmetric finite element to represent the soil
sample. To model the oedometer test, no lateral strains were allowed (,=O) while
increments of compressive axial strain (Ae1 =te) were applied. The results are presented as
a plot of volumetric strain, e, against axial stress, Ca ' . To model the drained triaxial
compression tests, the confining pressure, 03 , was kept constant while increments of
compressive axial strain, Ae, were applied. The results are presented as plots of deviatoric
stress, 0-U3 ( = C.0r)s and volumetric strain, E, against axial strain, Ea.

The predicted stress-strain curves for large oedometer tests on both Roadford rockfill and
sand waste fill are compared with the laboratory measured curves in Figures 5.1 6.i, ii and iii
for the three different constitutive models: non-linear elastic, Lade's double and single
hardening elasto-plastic models respectively. The predicted and measured curves agree quite
well.

It is of interest to note the predicted values of K 0 for Roadford rockfill in the oedometer test
for the three different models. For the non-linear elastic and double hardening model these
values are nearly constant with K 0 0.35 and K 0 O.3O respectively. The double hardening
model gave an unrealistically low value of K 0 O.25 when the recommended value of
Poisson's ratio for granular materials u=0.20 (Lade and Nelson, 1987; Lade, 1988b) was
used. The single hardening model did not predict a constant value of K 0 . Its value increased
with increasing axial stress a,' (= a") from K 0 = 0.30 at tow stress level (a,' = 1 OOkPa) to more
than K 0 =0.40 at higher stresses (o,'=2000kPa). This is a consequence of the shape of the
plastic potential surface employed in the single hardening model. It can be seen from Figure
4.7.b that there is a very little change in the gradient of the plastic potential surface for quite
a large stress level change giving nearly the same plastic strain increment ratio, te3 IAe 1 , for
d fferent values of K0 =o3'1a1 ' (for one-dimensional compression, Ae3 =0 and AE3=-e3i.

A comparison of predicted and observed response in the large drained triaxial compression
tests at different values of confining pressure 03' = 25, 96, 266 and 697kPa are shown in
Fgures 5.1 7.i, ii and iii. Only the data for Roadford rockfill are presented because triaxial test

82
results were not available for sand waste fill.

The non-linear elastic model reasonable predicted the response of Roadford rockfill in triaxial
tests (see Figure 5.17.i). The pre-peak behaviour is purely non-linear elastic; while the
volumetric strains are predicted with good accuracy the stiffness is generally underpredicted.
The strength at high confining pressures (o,' 697kPa) is overpredicted because no curved
strength envelope is modelled. The expansve behaviour is modelled only at peak and post-
peak introducing the perfect plasticity with the constant angle of dilation, u,i=q'14. A slightly
better fit to triaxial test data is possible, but then oedometer test data could no longer be
recovered.

Most aspects of the behaviour of Roadford rockfill observed in the large drained triaxial
compression tests are reproduced with remarkable accuracy by both Lade's elasto-plastic
models (see Figures 5.1 7.ii and lii). Notably, the gradual variation of the volumetric strain
behaviour, from being expansive at small confining pressures to being compressive at high
confining pressures. The double hardening model slightly overpredicted the volumetric strains
at the lower stress levels (see Figure 5.17.ii). The reason for this can be found in Figure 5.11
where some of the expansive plastic strains are missing from the adopted line at lower stress
levels. The 'irreversibility condition' does not allow a line with higher intercept, t, and lower
slope, p, to be adopted. The single hardening model slightly underpredicts the stiffness in all
triaxial tests (see Figure 5.17.iii). This is due to the adopted elastic stiffness as previously
discussed (see Section 5.2.3 and Figure 5.12). However, if a stiffer elastic response was
adopted to match the initial stress-strain curve from large triaxial tests, it would not be
possible to match the observed behaviour in large oedometer test. It is interesting to note that
the parameters for both Lade's models were determined using 'only' three triaxial tests at
values of confining pressure 03 '=96, 266 and 697kPa in addition to the single oedometer
test. The predicted response of the triaxial test at a confining pressure 03' = 25kPa is also
included for comparison purposes. Both models significantly underpredicted the strength (and
stiffness) for this triaxial test while the volumetric strains were nearly the same as observed.
It can be seen from Figure 5.6 that the 'f a lure' point for this triaxial test is well above the
adopted straight line. The adopted value of exponent in the failure criterion (equation (4.8
m=O.65 is already too high (near to the upper limit for granular materials according to Lade,
1984) indicating the significant curvature of the failure envelope. It was felt that the adoption
of even higher value for the exponent, m, would not be reasonable.

A comparison of predictions from the three constitutive models with results from triaxial and
oedometer laboratory tests presented in Figures 5.16 and 5.17 show that, while both Lade's
elasto-plastic models are capable of reproducing the full range of available rockfill test data,
this is not quite possible with the non-linear elastic model. However, its performance is still

83
satisfactory, especially at the lower stress levels which usually operate in a rockfill dam.

The standard triaxial and oedometer tests impose different, but very simple stress paths on
a sample of soil. The behaviour of the soil models were therefore further examined using more
complex stress paths.

Charles (1973, 1976) presented the results of drained triaxial tests on a sandstone rockfill.
In these tests the sample was first subjected to anisotropic compression under either a
constant stress ratio of 0 1 /0 3 =2.0 or 3.0. This was followed by a second stage of loading
in which the confining pressure, 03 ', was kept constant while the axial stress, 0', was
increased. The results of these tests are shown in Figure 5.18.a as plots of principal stress
ratio, 0'/O 3 ', against mean effective stress, p'. principal strain ratio, E3/E1 , against principal
stress ratio, 1'3' and volumetric strain, e, against mean effective stress, p'. The results
of two other tests with the same initial anisotropic compression stage followed by a second
stage in which the axial stress, o,', was kept constant while the confining pressure, os', was
increased are shown in Figure 5.18.b using the same plots.

The three different constitutive models were also used to model these tests. The soil
parameters determined in the previous section to characterize the behaviour of Roadford
rockfill were used. It should be noted that the Roadford rockfill is quite similar to the
sandstone rockfill tested by Charles (1973, 1976). Again, idealized, sing e element tests were
used in which the constant stress ratio increments, U/I03 ', were applied during the first
stage of the tests (the anisotropic compression). In the second stage, the appropriate stress
increments were added (o' or Aa3') while keeping the confining stress, 03', (the first two
tests) or the axial stress, o,', (the second two tests) constant at the values reached at the
end of the anisotropic compression stages. The results are presented in Figures 5.1 9.i, ii and
iii for the non-linear elastic, Lade's double and single hardening model respectively, in the
same manner as the laboratory test results in Figure 5.18.

It can be seen that all models qualitatively captured the observed rockf II behaviour in these
tests, notably the variation of principal strain ratio, €3 /e1 , with increasing or decreasing
principal stress ratio, OI03 '. However, the Lade's elasto-plastic modes are superior. They
predicted the true manner in which the changes in stress path direction produced any changes
in the p' versus E behaviour. The non-linear elastic model did not predict the additional
compressive component of volumetric strain, E, when the stress ratio started increasing.
When the stress ratio started decreasing the response was too stiff. It is worth mentioning
that the immediate response of the single hardening model when the stress ratio started
reducing was purely elastic as was discussed in Section 4.5 of the pervious chapter (see
Figure 4.10). When the failure conditions were approached by increasing the stress ratio the

84
double hardening model showed some expansive behaviour. However, it was not so
pronounced because of the confining pressures after the first stage of anisotropic
compression were relatively high (o3'=45OkPa for a1 '/o3'=2.0 and o 3'=35OkPa for
a.1 1 Ia3' = 3.0) for Roadford rockfill to produce any significant expansive behaviour (see Figure
5.17.ii). The single hardening model even produced the increased compressive response when
the failure condition were approached. The similar behaviour was observed in the triaxial
model test where confining pressure was too high a' =697kPa (see Figure 5.1 7.iii). This may
be a consequence of the particular shape of the plastic potential surface near the failure
condition when the confining stresses are relatively high still giving the significant plastic
compressive volumetric strains.

A qualitative comparison of predictions from the two Lade's constitutive models for these
special tests on sandstone rockfill shows that the double hardening model, because it is based
on the two intersecting yield surfaces, is more sensitive to the changes in stress path
diredtion. In the next section the implications of this finding will be further discussed when
all three models are used in finite element analyses of Roadfocd dam.

5.4 FINITE ELEMENT ANALYSIS

The instrumented cross-section depicted in Figure 5.1 .a is analyzed. The finite elemeni mesh
with boundary conditions is shown in Figure 5.20.a. The mesh was refined at the upstream
embankment toe (see Figure 5.20.b) since particular attention was paid to the deflections of
the asphaltic concrete membrane during the reservoir impounding where these were measured
by instrumentation.

Six materials are modelled, namely: (i) rock foundation, (ii) concrete cut-off structure, (iii)
upstream alluvium blanket, (iv) sand waste fill and drainage material, (v) a low grade rockfill,
and (vi) asphaltic concrete membrane.

The foundation is modelled as a linear-elastic material with values of Young's modulus

E=225MPa and Poisson's ratio u=0.20. The parameters were chosen to match the observed
settlements at the toj of the foundation. Since the foundation is elastic the initial stresses in
it have no influence on the predicted deformations. They were therefore assumed to be zero.
A static water table was assumed at foundation level, and this was maintained throughout
the analysis. The increase in foundation seepage pressure ne the upstream toe (behind the
grout curtain) due to impounding was ignored as it would have negligible influence on
deformation. The cut-off structure is modelled as solid concrete with Young's modulus
E=3OGPa and Poisson's ratio u=0.15. For simplicity, the alluvium blanket upstream of the
cut-off structure and above the rock foundation is modelled as a low grade rockfill placed

85
directly on the foundation before construction with the initial stresses determined by a bulk
unit weight y=22.75kN/m 3 (sea below) and the value of K0 =0.50.

The low grade rockfill and sand waste fill are modelled using the three different constitutive
models as discussed previously. The model parameters are listed in Tables 5.2-5.4. Bulk unit
weights are calculated using the values from Table 5.1, adopting y=22.75kN/m 3 for the
rockfill and y= 1 8.35kN/m 3 for the sand waste fill.

No specific laboratory test data or field observations are available for the drainage material
in the foundation drainage layer and beneath the upstream membrane. This material has only
a small influence on deformation, and, for convenience, its properties are assumed to be the
same as those of sand waste fill.

The properties of the asphaltic concrete membrane under slow loading have been shown to
be equivalent to a granular material behaving in an undrained saturated manner (Vaughan,
1980). This undrained behaviour results from the very high viscosity of bitumen, which
inhibits drainage of bitumen from the soil pores. Thus, the asphaltic concrete of the
membrane may be modelled as an undrained granular soil. This has been done In terms of
effective stresses specifying that the equivalent bulk modulus of the pore fluid is a hundred
times higher than the effective bulk modulus for the soil (see Section 2.4.4). No deformation
properties for the aggregate of the asphaltic concrete have been measured. They have only
a limited influence on deformation. The properties in terms of effective stresses have been
assumed the same as those of the sand waste fill and the drainage material. Only the lower
part of the asphaltic concrete membrane was modelled as a separate zone (see Figure
5.20.b).

Besides the asphaltic concrete of the membrane all fill materials were kept drained with zero
pore pressure throughout construction and subsequent impounding.

5.4.1 Analysis During Construction

The modelling of the construction of embankments and fills by ICFEP is described in Section
2.4.3. The construction of Roadford dam is simulated 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 rockfull material. The loads corresponding to each construction layer
were imposed over several increments: three for the first six layers and two for the remaining
three. The layer of drainage material on the upstream face and asphaltic concrete membrane
were placed in two increments after the embankment had been built.

86
Figure 5.21 shows contours of stress level, S. (the proportion of the current strength
mobilized, see equation (5.4) and Figure 5.4) at the end of the construction for the three
different models used in the analysis. In general, low stress levels operate in the embankment.
The Lade's single hardening elasto-plastic model predicted the lowest values indicating the
higher mobilization of horizontal stresses during the dam construction than the other two
models.

The directions of the major principal stress, o', at the end of construction are shown in
Figure 5.22. They are nearly vertical, except near the embankment slopes.

The vectors of accumulated displacements during the embankment construction are shown
in Figure 5.23. The pattern of the displacement vectors is typical for a symmetric
embankment. The highest vertical displacements occur in the central portion of the dam. The
maximum settlement is predicted at approximately mid-height and is of the order of 300mm.

Results of the analyses are further examined and compared with the field measurements. The
position of the measuring devices is shown in Figure 5.1.a.

Settlements are compared in Figure 5.24: (a) on the embankment centre-line (the settlement
gauge 'Si') and (b) near the centre of the downstream part of the dam (the settlement gauge
'S2'). It can be seen that all three models give predictions in reasonable a9reement with the
field observations. The maximum settlements are overpredicted by the non-linear elastic
model but underpredicted by the Lade's elasto-plastic models. It is part cularly so for the
settlement gauge 'Si' on the embankment centre-line where the one-dimensional stress path
is representative. This may seem surprising as the parameters for all three models have been
derived to match the observed compressibility in the large oedometer test. The three models
give approximately the same amount of volumetric strains, e, in both oedorneter (see Figures
5.i6.i, ii and iii - (a)) and constant stress ratio compression tests (see Figures 5.19.i, ii and
iii) up to the maximum 'working' vertical (o' 900kPa) or mean stresses (p' 500kPa) in the
embankment. However, closer inspection of these figures indicates that irnder the stresses
of interest (vertical stresses o' 1 0O-900kPa, or mean stresses p' 1 0O-500kPa) 2 the non-
linear elastic model gives a softer response which is responsible for the larger dam
settlements.

2 They are no strains in the 'newly constructed' layers, as discussed in Section 2.4.3,
where the stresses are up to iOO-l5OkPa (depending on the thickness of adopted
construction layers), and where the non-linear elastic model predicts the st ffer response (as
a consequence of high minimum Young's modulus, E,=30,000kPa, see the parameter E1,
in Table 5.2) than the Lade's elasto-plastic models.

87
Horizontal movements along the magnetic extensometer 'H' are compared in Figure 5.25.
Inspection of this figure indicates that the analysis using the non-linear elastic model severely
overpredicted horizontal movements (by factor of two). Both Lade's elasto-plastic models
predicted horizontal movements with reasonable accuracy.

5.4.2 Analysis During Impounding

Reservoir impounding was simulated by applying the water load as an 'external' pressure on
the upstream face and alluvium blanket plus an 'internal' incree in pore water pressure in
front of the cut-off structure and grout curtain assuming an initial water table at the ground
surface. The reservoir level was raised in stages, each stage corresponded to the corner nodal
points of elements on the upstream slope. The finer element mesh on the upstream toe
enabled each of the first ten stages (up to 7.5m above the toe) to be completed in one
increment only. The next stages (11.0, 14.5, 19.75, 25.0, 30 5 and 33.5m above the toe
respectively) were completed in 5 or 10 increments per stage depending on the size of a
stage. The reservoir was raised to the top water level (TWL), 33.5m above the upstream
embankment toe, in a total of 55 increments.

Some results of analyses employing the three different constitutive models are presented in
Figures 5.26, 5.27 and 5.28 for the three impounding levels (a) 14.5, (b) 25.0 and (C) 33.5m
above the toe. All models predicted similar overall dam behaviour during impounding.

The contoured plots of stress level, S. i.e. the mobilization of the shear strength within the
fill are shown in Figures 5.26.i, ii and iii. In general, there is a decrease in strength mobilized
in the upstream part of the dam during impounding. However, the strength is becoming
mobilised to a greater extent again (after a rotation of the principal stresses, see below) in the
toe area of the embankment, especially in the drainage layer just below the membrane and
on the interface of the sand waste fill with both cut-off structure and rockf II. Impounding has
little influence on the stresses mobilization in the downstream slope.

Inspection of the directions of the major principal stress, 0 1 ' , in F gures 5.27.i, ii and iii shows
a major rotation of the principal stresses as the reservoir was filled. Initially, following
construction, the major principal stresses were approximately parallel to the upstream face
(see Figures 5.22.i, ii and iii), and the minor principal stresses normal to it. As the water load
on the membrane increases, the stress normal to the face increases at a faster rate than the
direct stress parallel to the upstream slope, and at a certain point it becomes the major
principal stress. For impounding level 14.5m above the upstream toe, the major principal
stresses have rotated to near normal to the face everywhere in the upstream toe area (except
close to the cut-off structure where arching has occurred). There are slight additional rotations

88
as the water level is raised further.

Figures 5.28.1, ii end iii show the general pattern of displacement within the dam from the
start of impounding. The main discontinuity in the displacement pattern arises around the
interface between the sand waste fill and rockfill and not on the interface with the cut off
structure. Movements develop progressively during impounding. The maximum movements
at the end of impounding are predicted in the lower part of the md-third upstream slope and
are of the order of 120mm. It is of Interest to note that, during all stages of impounding, both
Lade's elasto-plastic models predicted the displacement vectors more normal to the upstream
slope than the non-linear elastic model.

Comparison of the predicted and measured deflections of the asphaltic concrete membrane
at the upstream toe of the embankment (see Figure 5.1 .b for the extent and location of
electro-level instrumentation 'E') is given in Figure 5.29 for the three impounding levels (a)
14.5m, (b) 25.Om and Cc) 30.5m above the toe. All models predicted the magnitude of the
movements normal to the membrane reasonably. Comparing the predictions by the different
models, it can be noted that the non-linear elastic model underpredicted deflections for the
lowest impounding level, 14.5m above the upstream toe. This is a consequence of very stiff
fill behaviour predicted by the adopted non-linear elastic model, as noted in Section 5.3 (see
Figure 5.19.1.b), during a rotation of the principal stresses which have occurred in the
upstream toe of the embankment during impounding up to this water level.

5.4.3 'Improved' Predictions

The input parameters for the analyses in the pervious two sections were derived from
laboratory data obtained from large diameter oedometer and triaxial tests. It was assumed
implicitly that the tested samples were representative of the fill materials in the field.
Furthermore, no variability of the prototype fill materials was accounted for. The settlement
measurements clearly show non-uniformity of the rockfill material regarding its compressibility
(see Figure 5.24.a). The lower part of the embankment is softer than the upper part. The
explanation can be found in Figure 5.30 where a summary of the control tests for the rockfill
material is shown (see also Vaughan, 1994). It can be seen that the tower part of the fill (up
to 1 5m above the foundation level) was placed at a slightly higher water content and lower
dry density than the remainder of the fill. That is why the drier, denser upper fill has shown
lower compressibility during the dam construction. Also, in Figure 5.30, the dry densities, Vd'

and water contents, w, from two large oedometer tests on Roadford rockfill, listed in Table
5.1, are shown (lines 'A' and 'B'). It can be noted that the oedometer test on the sample
which received only light compaction in the laboratory (y=20.5kN m 3, w=4.7%, lines 'A'
in Figure 5.30) is more appropriate for the compressibility estimate of the lower part of the

89
embankment.

In this section it is attempted to make an improved prediction of Roadford dam movements

during construction and impounding taking into account both laboratory and field data. It has
been shown above that the non-linear elastic model is inferior in representing rockfill
behaviour in both the laboratory and the field. Predictions made by the Lade's elasto-plastic
models were of equal accuracy. Both models underpredcted the maximum observed
settlement (see Figure 5.24.a); the single hardening model slightly overpredicted the
horizontal movements, whereas they are slightly underpredicted by the double hardening
model (see Figure 5.25). Regarding the more compressible behaviour of the lower fill, where
the horizontal extensometer 'H' is embedded (see Figure 5.1 .a), it is reasonable to expect that
the double hardening model can better capture the observed response of the embankment
during construction. Moreover, the single hardening model is not so sensitive to the changes
in stress path direction operating during the reservoir impounding. Thus, the Lade's double
hardening elasto-plastic model was chosen in the analysis reported here.

The model parameters for the lower fill (up to 1 6.5m above the foundation level - the first
two 'construction' layers) were selected to fit the compressibility of the lightly compacted
rockfill specimen before inundation measured by BRE in the large oedometer (see Figure 5.3
and line 'A' in Figure 5.31). The model parameters for the upper fill (above 16.5m) were
determine to match the observed compressibility in the field during construction 3 . Some
simple calculations based on the assumptions that the centra4 settlements can be treated as
a one-dimensional problem, and that the vertical stress is the same as the overburden
pressure (see Section 3.4.2.1) indicated that the upper fill has a stiffness approximately three
times greater than the lower fill. It is worth mentioning that the field compression curves
showed evidence of prestress due to compaction4 . This was slight for the lower fill, but
significant for the drier and denser upper fill where it might amount to as much as 200kPa
(Vaughan, 1994).

No comparison between the observed and predicted movements of the sand waste fill was
possible during the embankment construction, as the increase in stress in this fill was small.
The properties of the sand waste fill based directly on the results of the large oedometer test
performed by BRE (see Figure 5.3) consistently overpredicted the deflections of the asphaltic

3After construction this material, however, showed signif cant collapse compression due
to delayed wetting up (see Vaughan, 1994).

4 1n heavily compacted fills the effect of compaction can be similar to prestressing and
over-consolidation. lnterparticle contacts are crushed by compaction and, when compacted
fill is loaded, they do not start to yield again until a certain stress is reached. The fill has a low
compressibility up to this point, and the stress is analogous to a pre-consolidation pressure.

90
concrete membrane located above this fill (see Figure 5.29) if the Lade's elasto-plastic models
are used. However, the control tests in the field gave a higher dry density yd=18.4kN/m3
(Wilson and Evans, 1990) than that in the available large oedometer test (Vd= 1 7.3kN/m3, see
Table 5.1) indicating that a lower compressibility of sand waste fill was achieved in the field.
The absence of large oedometer tests at different dry densities made it impossible to
interpolate (or extrapolate) the field compressibility. Thus, the findings for other granular
materials concerning the influence of density on their compressibility have been used.

El-Ruwayih (1975) and Daramola (1978) provided experimental evidence suggesting that in
equation (3.4) the parameter D may be assumed Constant irrespective of the initial density,
whereas the parameter C depends linearly on the initial porosity, n, of the sample. If the
relationship between C and n for Ham River sand (Daramola, 1975) is used it can be shown
that the field compressibility of the sand waste fill is about 25% lower than that measured
in the laboratory.

The new Lade's double hardening model parameters for the lower and upper rockfill and sand
waste fill are listed in Table 5.5. The single element model predictions of the oedometer test
are shown in Figure 5.31 where the assumed field compressibility for the rockf ill and sand
waste fill can be compared with those observed in the laboratory.

Comparing Tables 5.3 and 5.5 it can be seen that only three model parameters have been
changed: the initial modulus number, M, the collapse modulus, C, and collapse exponent, p.
Together with the exponent, n, and Poisson's ratio, u, these parameters mainly control the
material response in anisotropic compression (up to moderate stress ratios, R=o 1 'Io 3 '). No
attempt has been made to change the rest of the model parameters. They have only limited
influence on the embankment behaviour both during construction and impounding.

A comparison of the newly predicted and measured settlements during the dam construction,
horizontal movements at the end of construction and deflections during reservoir impounding
is shown in Figures 5.32, 5.33 and 5.34 respectively. Predictions are particularly good during
and at the end of construction (see Figures 5.32 and 5.33). There is an overprediction of the
membrane deflections during the reservo r impounding at the interface between the sand
waste fill and rockfill (see Figure 5.34). However, the predictions are still good bearing in mind
that this is an area where large stress reversals occur and the inability of Lade's isotropically
hardening model to capture it accurately. A comparison of the predicted and measured
deflection of the membrane during impounding at four positions (4, 6, 10 and 1 4m) up from
the toe and reservoir head above the toe are plotted against time in Figure 5.35.a. Apart from
the position 6m up from the toe, which is approximately at the interface between the sand
waste fill and rockfill, agreement is quite good in spite of the impossibility to model the creep

91
that occurred when the reservoir level was practically constant during impounding. The creep
that occurred is clearly illustrated in Figure 5.35.b where the predicted and measured
deflection is plotted against the reservoir head. It can be seen that the model predicted the
response of both rockfill and sand waste fill quite well, particularly the early stiffness at low
water loads. This is due to the reversal of principal stress direction, although, in the field, a
pre-consolidation effect caused by compaction may play some role as well.

5.5 CONCLUSIONS

It has been shown that the Lade's elasto-plastic models are capable of reproducing the full
range of available rockfill test data for Roadford dam. There is a deterministic way in deriving
the model parameters. Much will depend on the adopted elastic response, so it is advisable
to have unloading-reloading cycles from conventional triaxial and/or one-dimensional
compression tests. However, in the absence of such tests the elastic response can be
evaluated from the initial slope of the triaxial compression stress-strain curves. An isotropic
compression test is not usually available for the rockfill materials but response can be
estimated from a one-dimensional compression test.

The non-linear elastic model parameters can be derived entirely from a suite of standard
laboratory triaxial tests. However, some adjustments are necessary in order to use the
oedometer test data. The fit for the available test data is not so good as in the case of the
Lade's elasto-plastic models but is still satisfactory, especially at lower stress levels which
usually operate in rockfill dams.

When the models were employed in finite element analyses of Roadford dam the superior
performance of the Lade's elasto-plastic models was noted. The non-linear elastic model
severely overpedicted lateral movements during construction. With the adopted stiffness on
unloading-reloading it underpredicted deflections of the upstream asphaltic membrane for the
low water loads during impounding when significant rotations of the principal stresses in the
upstream toe area occurred. Further studies not reported here showed that the non-linear
elastic predictions could be improved to match the field behaviour by adjusting the input
parameters, but then the laboratory test data could no longer be recovered. Thus, the
improved predictions could not be made from the laboratory data.

Being based on two yield surfaces the Lade's double hardening model is more sensitive to the
changes in stress path directions which occur during reservoir impounding. Although more
input parameters are needed for this model they are less interconnected than the single
hardening model parameters. This is particularly relevant when subsequent analyses are
undertaken to account for differences between the fill as placed in the field and that used for

92
the laboratory tests. The effects of a change in fill density (and/or water content) can be
accounted for in the Lade's double hardening model by modification of as few as three
parameters.

Lade and Pradel (1989) claimed that the single hardening is more efficient, in terms of
computer resources, than the double hardening model when analysing boundary value
problems. However, any such efficiency will depend on the manner in which the models are
implemented in the finite element program. Comparison of the required computer resources
for the two models, expressed in terms of the equivalent central processor time units on a
Sun workstation and using the ICFEP code, shows no significant difference between the two
models when they have been employed in either the single element tests performed in Section
5.3 or analysis of Roadford dam.

In summary, it can be concluded that of the three models used 1the Lade's double hardening
elasto-plastic model is: (il more capable, and (ii) the most suitable for the use in finite element
analysis of the rockfitl dams.

93


Material Test Conf.Pres. Dry Weight Moist.Cont.

0 3' [ kPaJ Vd IkN/m3 J W 1%]

Rockfill Oedometer - 21.6 7.9

Triaxial 25 22.2 7.0

96 219 56

266 218 57
N
U 697 21.4 4.2

0edometer - 20.5 4.7

Sand waste Oedometer - 17.3 6.1

(*) - lightly compacted sample

Table 5.1 Initial conditions for large tests on Roadford rockfill and sand waste fill
(after Charles and Watts, 1985; Wilson and Evans, 1990)

Parameter Rockf ill Sand waste

Modulus Number, E0 30000.0 65000.0

Modulus Constant, A 0.05 2.90

Modulus Constant, B 6.0 0.3

Modulus Exponent, C 1.00 1.15

Unload.Mod.Param., H 1.0 1.0

Poisson's Ratio Const., I 0.25 0.25

Poisson's Ratio Const., L 0.5 0.0

Cohesion, c' 50 5

Angle of Friction, ' 36 40

Angle of Dilation, qi 9 10

Table 5.2 Non-linear elastic model parameters for Roadford rockfill

and sand waste fill

94
 Parameter Rockfill Sand waste Component

Modulus Number, M 400.0 1200.0 Elastic

Exponent, n 0.70 0.60 Elastic
Poisson's Ratio, u 0.25 0.20 Elastic
Collapse Modulus, C 0.000450 0.000020 Pl.Col.
Collapse Exponent, p 1.00 1.35 Pl.Col.
Failure Constant, , 260.0 35.5 Failure
Failure Exponent, m 0.65 0.0 Failure
Pl.Potent.Const., p 0.35 0.45 Pl.Exp.
Pl.Potent.Const., R 0.0 0.0 Pl.Exp.
Pl.Potent.Const., t 12.0 0.0 Pl.Exp.
Work-Hard.Const., a 2.6 3.0 Pl.Exp.
Work-Hard.Const., fi o.o 0.0 Pl.Exp.
Work-Hard.Const., P 0.11 0.15 Pl.Exp.
Work-Hard.Expnt, I 1.05 1.15 Pl.Exp.

Table 5.3 Double hardening model parameters for Roadford rockfill and sand waste fill

Parameter Rockfill Sand waste Component

Modulus Number, M 300.0 1200.0 Elastic

Exponent, n 0.50 0.45 Elastic

Poisson's Ratio, u 0.20 0.20 Elastic

Failure Constant, ' 260.0 45.0 Failure

Failure Exponent, m 0.65 0.10 Failure

Pl.Potent.Const., qi2 -3.00 -3.40 Plastic

Pl.Potent.Expnt., p 2.70 2.15 Plastic

Work-Hard.Const., C 0.000105 0.000090 Plastic
Work-Hard.Expnt,p 1 .80 1 .65 Plastic

Yield Exponent, h 0.875 0.650 Plastic

Yield Constant, a 0.175 0.450 Plastic

Table 5.4 Single hardening model parameters for Roadford rockfill and sand waste fill

95
 Parameter Lower Upper Sand waste
Rockfill Rockfill

Modulus Number, M 330.0 1000.0 1600.0

Exponent, n 0.70 0.70 0.60
Poisson's Ratio, u 0.25 0.25 0.20
Collapse Modulus, C 0.000450 0.000200 0.000015
Collapse Exponent, p 1.05 1.00 1 .30
Failure Constant, , 260.0 260.0 35.5
Failure Exponent, m 0.65 0.65 0.0
Pl.Potent.Const., p 0.35 0.35 0.0
Pl.Potent.Const., R 0.0 0.0 0.45
Pl.Potent.Const., t 12.0 12.0 0.0
Work-Hard.Const., a 2.6 2.6 3.0
Work-Hard.Const., fi 0.0 0.0 0.0
Work-Hard.Const., P 0.11 0.11 0.15
Work-Hard.Expnt, I 1.05 1.05 1.15

Table 5.5 Double hardening model parameters for Roadford rockfill

and sand waste fill - 'Improved' predictions

96
 Instrumentation:
Si, S2 - Settlement plates
Ii - Magnetic extensometer
E - Electro-level system
Impounding level
above dam toe
—30.Sm Si 2.25
25.Om
14.5m

S2
Scale
(a) Om Drainage
layer

Inspection gaflery

Asphaltic
membrane

Drainage layer

Alluvium
blanket
Sand Waste

(b)
Rock
foundation

Figure 5.1 (a) Cross-section of Roadford dam showing location of instrumentation, and
(b) Detail of the upstream toe

97


2,500
I I I I
- I I I I I I I
- I I I I I I I I

2.000 - - - - - L ------------------ - -
I
I I I I I
- I I I I __i____4_- __--4---- -

1,500
b
U)
U)
U)

(I)
U
1,000
CD
>
II)
a

500

-7
- Confining Pressure
-6 - - 25kPa
96kPa I I I I

-5 266kPa
697kPa
I I I I
-4 - - - I - - L - - - L - - - - - - - - - -
I I I I I I _..
I I I I I I
> I I I
/ --------I- I
- ------------4 - - - 4 - - - 4- - - -4- - - -4 - - -
C
I I / I I I
- I I I I I I I I I
CD
I I I I I I I I I
-2 - --
U I I I / I I I I I
- I I I I I I
a) I I I / I I I I_____- -
E- --------I___ ..
0 -
> 0 - I I

- - - I I - I I !- - -
- L— - I I I 4- - - I
I I I I I I I
I - I I - - - I
- I I - - -c - - I I I I
I -_[_ I I I I I
2 - '-r - - - L - - - L -------_J_ - - -1 - - -
I I I _I I I I I
- I I I I I
3 I I
0 1 2 3 4 5 6 7 8 9 10
Axial Strain, ca [°')

Figure 5.2 Results of large drained triaxial compression tests on Roadford rockfill (after
Charles and Watts, 1985)

98
 Rockflhl - Lightly Compacted
- - - No Inundation (estimate)
Inundated (estimate)
- - - - Rockflhl - Heavily Compacted
Sand Waste Fill

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

>

C')
t)

9 -Inundation at 76OkPa ....................................................

10 I I I I

0 500 1,000 1,500 2,000

Axial Stress, a [kPal

Figure 5.3 Results of large oedometer tests on Roadford rockfill and sand waste fill (after
Charles and Watts, 1985; Wilson and Evans, 1990)

(1 p

Figure 5.4 Definition of Stress level, S

99
 10,000

Confining Pressure
v 25kPa n
• 96kPa (E•iIPa)=M
266kPa
A
697kPa
adopted

A
1,000 :

:n=0.70
M=400 ......................................1
:::J .------.• -

.....• ..................... I

100
0.1 1 10

Figure 5.5 Elastic parameters M and n for Roadford rockfill - Double hardening model

1,000

-' 100

10
0.01 0.1

Figure 5.6 Failure parameters i and m for Roadford rockfihl - Double hardening model

100
 1,000

100

C.,
0

a. 10
(.,l
a.

0.1
10 100 1,000

(H2/p2)

Figure 5.1 Work-hardening parameters C and p for Roadford rockfill - Double hardening model

300
11=260

H Confining Pressure
150 b Wpl(W . 96kPa
/ H =a•e /p
1 Pla • 266kPa
100 1/q A 697kPa
a=n1 (ePa /Wpip ) - - .-.. 96kPa - adopted
(
50 - 266kPa - adopted
b=1I(qW1) 697kPa-adopted
0 I ___________________________

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

p1 'a

Figure 5.8 Variation of accumulated 'expansive ' plastic work Wi,, with hardening parameter
H 1 and confining pressures a 3 ' for Roadford rockfill - Double hardening model

101


Confining Pressure
5 -- . 96kPa
. 266kPa
- ...£ 697kPa
adopted
3 - ..........................................a=2.SO ...............
q
2 -.................................................4..................................
qa+f(a3IP)

0 I I

0 1 2 3 4 5 6 7 8 9 10

Figure 5.9 Work-hardening parameters a and/I for Roadford rockfill - Double hardening model
10.

(
a.
0

0.1

0.01 ' ii
0.1 1 10 100

Figure 5.10 Work-hardening parameters P and I for Roadford rockfill - Double hardening model

150
1/2
=pH+R (aIp a) +t

- R00

0 50 100 150 200 250 300

H1

Figure 5.11 Plastic potential parameters p, R and t for Roadford rockfill - Double hardening
model

102
 10.000

Confining Pressure - - --
v 25kPa :;H fl:
• 96kPa (EJp)M(,'/p)
• 266kPa
a697kPa . .........
adopted . .

a. 1,000
Ui

100 L
0.1 10

Figure 5.12 Adopted elastic parameters M and n for Roadford rockfill - Single hardening
model

1,000

:•
p/Pa)=C•(li/Pa)

100

:: : p=l.8O :
C',
0

m 10

Mean Effective Stress

V 200kPa
/1 :: . 500kPa
/ . • i000kPa
a 2000kPa
C=0 000105 adopted
0.1
10 100 1,000

Figure 5.13 Work-hardening parameters C and p for Roadford rockfill - Single hardening modes

103


0.8

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8
q

Figure 5.14 Yield constant a for Roadford rockfill - Single hardening model

10 - - - - -

9 - ........

8 -

3'1flfiIgPressure
3.00
. 96kPa
2 • 266kPa
A 697kPa
1 - ...adopted

0
0 1 2 3 4 5 6 7 8 9 10

Figure 5.15 Plastic potential parameters '2 and p for Roadford rockfihi - Single hardening
model

104
 N.
1 -..... . Rockfihl - Model
Rockfihl - Test
>

ca
U
a,
E

ii ... ii ..ii'i.
.
6 iitI,ji I
0 500 1000 1,500 2,000
Axial Stress, a' fkPa)

(a) Roadford rockfill

>
C

U)
U ....:....eteI1
a,
E .

0 500 1,000 1,500 2,000

Axial Stress, aa' (wa]

(b) Sand waste fill

Figure 5.1 6.i Comparison of predicted and measured behaviour in oedometer tests - Non-linear
elastic model

105
 0

>
- 2

3
0

6
0 500 1,000 1500 2,000
Axial Stress, aa [kPa]

(a) Roadford rockfill

_______________________________ . Sandwaste - Model

Sandwaste Test

0 500 1,000 1,500 2,000

Axial Stress, C5a [kPa]

(b) Sand waste fill

Figure 5.16.ii Comparison of predicted and measured behaviour in oedometer tests - Double
hardening model

106
 • Rockfill - Model
Rockfill - Test
>

(I)
U
a,
E
S

....

500 1,000 1500 2,000
Axial Stress, aa (kpa]

(a) Roadford rockfill

. Sandwaste - Model
0
C _____ Sandwaste Test
C,)
0

a,
E
0 500 1,000 1500 2,000
Axial Stress, c' [kPaJ

(b) Sand waste fill

Figure 5.16.iii Comparison of predicted and measured behaviour in oedometer tests - Single
hardening model

107
 2,500

A A A A A A

2000
A

a-
C,,
1,500
A

(I)

A
U)

/............
A a
F'°°°

•
1'

500 ..-.- ............................*.-_..,......a

/1. •
I, I -
t: V V V
• •7 v
av
0

-7
Confining Pressure
-6 v25kPa -Mode ................................
• 96kPa -Model
-5 a266kPa - Model ..................................
A 697kPa - Model - -
25kPa -Test
96kPa -Test
>
-3 .
266kPa - Test ......... /.............................................

U) -2
C-)

a,
E -1 . .....................:i
U
• S A
0

3
0 1 2 3 4 5 6 7 8 9 10
Axial Stin. Ca

Figure 5.17.i Comparison of predicted and measured behaviour in triaxial tests on Roadford
rockfill - Non-linear elastic model

108
 2500

2,000

a-
x

1,500

(I,
C.)

1,000

500

-7
Confining Pressure
v 25kPa - Model
-6 • 96kPa - Model ....................................
• 266kPa - Model y
-5 £ 697kPa - Model .................................... T.
25kPa -Test
-4 - . - . .. 96kPa - Test
266kPa-Test //
>
w
-3 697kPa - Test
C

C,)
C.)
-2 /..................................................
a,
E -1 _...-.
I-
/-
-S.-.
0

3
0 1 2 3 4 5 6 7 8 9 10
Axial Stin, Ea

Figure 5.17.ii Comparison of predicted and measured behaviour in triaxial tests on Roadford
rockfitl - Double hardening model

109
 2,500

2,000

(a

1,500
b

C,)
0 A

1,000
/ -.-.
a / A.'

Ii
500 .-

/A
'' ' " V

-7 --
V
Confining Pressure
v25kPa - Model ...............................
. 96kPa -Model
-5 . 266kPa - Model ..............................
A 697kPa - Model -
V
-4 _.. - - - - 25kPa - Test ...............,, ................................
9 6kPa -Test //
>
3 _.. -..-..-... 266kPa - Test .... y ................................................
C _697kp;Test/
U)
0
a)
E
0
>

T&1•
:

3
0 1 2 3 4 5 6 7 8 9 10
Axial Strain, Ca I%I

Figure 5.17.iii Comparison of predicted and measured behaviour in triaxial tests on Roadford
rockfifl - Single hardening model

110


--I-..,
OlD
0 Stage 2 0
C
0,
0, 0,
V
L0 '0
C V
> >
U U
4, 0,

4, 0,
C
'5
fb)
U U
C C
200 400 600 800 -06 -04 -02 0 '02 .0 a.
Mean normal effective stress (kNIm2) Principal strain ratio
200 400 600 800
0

0
J1
C
C

0,
E 2

(a) stage 1 - u,/o 3 constant; stage 2 - o' increasing, 03 constant

- '
DID

4
SJ age1
0,
1 13
U
4, o 2
4,
'5
a a
Le20
& 200 400 600 800 — 02 0 •02 +04 +06
Mean normal effective stress d(kNjm2) Principal strain ratio
-1
nn nn Rnn erm

C
'5

O'i

(b) stage 1 - g,'/o-3 ' constant; stage 2 - o-' constant, o'' increasing

Figure 5.18 Results of drained triaxial tests on sandstone rockfill (after Charles, 1976)

111
 5

4
0

c 3

0
-0.4 -0.2 0 0.2 0.4
0 Principal Strain Ratio, 3Ic1

> 1
w

C 2 Stage
• 1
Ci) 3 • 1
a, o2
4
E o2
5

6

0 200 400 600 800 1,000
Mean Effective Stress, ( a 1 '+2a3')13 [kPa)

(a) stage 1 - u1 '/o 3 ' constant; stage 2 - o' increasing, 0 3 constant

g5

4 —......................................................................................
0
3 • S . S S S •-
U,
U,
. 2
U)

- .

-0.4 -0.2 0 0.2 0.4

0 Principal Strain Ratio, c3k 1

2 Stage
• 1
3 • 1
0
---o--. 2
--0--- 2
:

6
0 200 400 600 800 1,000
Mean Effective Stress, ( a 2-c')/3 [kPa]

(b) stage 1 - u1 '1u3 ' constant; stage 2 - o,' constant, o' increasing

Figure 5.19.1 Prediction of Roadford rockfihl behaviour in drained triaxial tests - Non-linear

elastic model

112
 :
U4
0
(5

w
U)

a
0
C

-0.4 -0.2 0 0.2 0.4

0 Principal Strain Ratio, c3/c1

. 2 Stage
• 1
03 • I
0
D 2
p2

6
0 200 400 600 800 1,000
Mean Effective Stress, ( 1 '+2')/3 [kPaJ

(a) stage 1 - u,'1u3 ' constant; stage 2 - u,' increasing, O'3 constant

U4
0
(5
• . • . . S S
U)
! uI•• .
U)

a U ____
0
C

0.0

-0.4 -0.2 0 0.2 0.4

0 Principal Strain Ratio, E31E1

. 2 Stage
1! • 1
(03 • 1
0
---0-- 2
---0-- 2

6
0 200 400 600 800 1,000
Mean Effective Stress, ( '+•')/3 [kPaJ

Ib) stage 1 - u1 '1u3 ' constant; stage 2 - 01 ' constant, o' increasing

Figure 5.19.fi Prediction of Roadford rockfill behaviour in drained triaxial tests - Double
hardening model

113
 5

.2
(5

0
a
C
'C
a-0

-0.4 -0.2 0 0.2 0.4

0 Principal Strain Ratio, c3Ic1

>1
'3

. 2 Stage
• I
03 • I
C.,
'C
o2
-o 2

6
0 200 400 600 800 1,000
Mean Effective Stress, ( '+24 3')/3 [kPa]

fa) stage 1 - u1 '/u3 constant; stage 2 - o-' increasing, O 3 constant

0
(5

U)

(I)

a
C.,
C

-0.4 -0.2 0 0.2 0.4

0 Principal Strain Ratio, 3/C1

'3

. 2 Stage
(5

03 •- 1
C.) -.-- 1
--'0-- 2
---o-- 2

6
0 200 400 600 800 1,000
Mean Effective Stress, ( i "' Y3 [kPaJ

(b) stage 1 - o',"u3 ' constant; stage 2 - a,' constant, a3' increasing

Figure 5.19 Prediction of Roadford rockfill behaviour in drained triaxial tests - Single
hardening model

114
 U)

U)

U)
4-
U)

U)
U)
2
.

U)

U)
C
0
4-
. L)
I-
0
a

U)

C
0
.0
.c
C
U)

U)
U) U)
E 0
4-
4- 4-
C C
U) U)
E E
U) .
U) C
U)
U) .0
4-
C E
U)
U-

U)
E
U)
U)
4-
C U)
0.
In

eJ ;jig
U0
C
0
C.,
 _A
AA ,U —.
0.70 S 0.60
S S

a- _a A-----

(I) Non-linear elastic model

,AA_A_AA_A
—. bA__A.
A• 0.70 S
AA

L•ir_ _- — I I -

(ii) Double hardening model

A 0.60

___ LA

ScaIu
0 50m

(.)Single hardening model

Figure 5.21 Contours of stress level, S. at the end of construction

116


,'/j/,I,i ,I,I Sill

'I
H "
l,,,, ,I I
Il,
f l 1I / % % '% ' s.

? 1' (111 Hi
(i) Non-linear elastic model

iiI?

Ii , i s i s ,l,s
55,5
I l

I l
j
I I
I l %

l
I
g' ' i' ((1 I' I I 1111 'i ' ' '

(ii) Double hardening model

fig",
g i,i 1i
I'
Sill till
'11
::#/I/l4fl l i i i I I 'i 't 'i
Il /i l I S I S S i l l II / %
I I
1
çi ( ( ( 11 II I 1 '1 ' '' ''k

$c.
0 5O,ii
I

(iii) Single hardening model

Figure 5.22 Directions of major principal effective stress, u 1 ', at the end of construction

117
 * • : • : u

(I) Non-linear elastic model

It t
it s : I : I •.

(ii) Double hardening model

J414
' I
'' :
S

Victor scal
0 Im
I •

(ii) Single hardening model

FIgure 5.23 Vectors of accumulated displacements during construction

118


50 50

Nonlinear-Elasttc Model
40 - - - - - Single-Hardening Model
Double-Hardening Model
E Observed Settlement

g C
0
• 30 30-......................................................................
C
0 0

10

0' I

0I1 I I I
0 100 200 300 400 500 0 100 200 300 400 500

Settlement, s [mm] Settlement, s [mm]

(a) at settlement plates 'Si' (b) at settlement plates 'S2'

Figure 5.24 Comparison of predicted and measured settlements during construction

E
E
E

80 - ..........................................
o60-..........................................................................................:.......................................................

-20 - ...
E .... . Nonlinear-Elastic Model
- - - - - Single-Hardening Model
Double Hardening Model
- ...............................-....,.----:.................................Observed Movement
-100
-100 -80 -60 -40 -20 0 20 40 60 80 100

Upstream jm] Downstream Im)

Figure 5.25 Comparison of predicted and measured lateral movements at magnetic

extensometer 'H'

119
7<> 0.70
a

—.

-.
0.60

(a) 14.5m above the toe

—A

0 70
("'_h' 0.60
lI__i

—A
U - A

Ib) 25.Om above the toe

(aA.

A 0.60
—A

a
-iIr- _A A I _£

(C) 33.5m above the toe

Figure 5.26.i Contours of stress level, S. for three levels of reservoir impounding

- Non-linear elastic model

120


_.A —A—aa _•_

____________ -u

0.70
—A

'—A

— I U

(a) 14.5m above the toe

A
.- —A—A —a-...,

0.70

—A__A

(b) 25.Om above the toe

—A

060
0.70 — -• _
___
A

—'I I A

Scal.
0 SOm

(c) 33.5m above the toe

Figure 5.26.11 Contours of stress level, S. for three levels of reservoir impounding
- Double hardening model

121
 _,..–A A____

0.60
rt________0h1__r_

(a) 14.5m above the toe

—A
(A

A 0.60
A-)

A_._..._a______a

(b) 25.Om above the toe

'A—A--A-.'

A
0.60

A —A —A—A -

$c.

9..
(c) 33.5m above the toe

Figure 5.26.iii Contours of stress level, S, for three levels of reservoir impounding
- Single hardening model

122
 (a) 14.5m above the toe

i:?i,Ii,,';; '1':;
liii
'It
II II 11'
I 'I' I' ( % %
''' 'I '

I (I I I I i i '1 '1 ' ' ' '

(b) 25.Om above the toe

Sc.

(c) 33.5m above the toe

Figure 5.27i Directions of major principal effective stress, o', for three levels of reservoir
impounding - Non-linear elastic model

123
 ii• /l 'i I%l%
11
I l I / l l %
(till I '1 '1 '1 '1 ' '\

(a) 14.5m above the toe

(b) 25.Om above the toe

r?
I
:7 j t
% %% %
I

t1 I liii
II II I % %

II1I1I1I1I1I1Ij

' '1 '1 ' '1 '\

Scali

0 SOni

(c) 33.5m above the toe

Figure 5.27.1 Directions of major principal effective stress, o-', for three levels of reservoir
impounding - Double hardening model

124
 <ii7/
I
\ 'I"
I ¼' '' 't'
1k? ( Itt ' '1

(a) 14.5m above the toe

(b) 25.Om above the toe

i' l ! '\\Z'iI
1
'I i
jIll 1111 II II 11
II I I I I I l l I I ll lI %
lIit

$G.
0 I •
50m

(c) 33.5m above the toe

Figure 5.27. Directions of major principal effective stress, os', for three levels of reservoir
impounding - Single hardening model

125


(a) 14.5m above the toe

•.. ...
I :

(b) 25.Om above the toe

• -. - -
I

Scm.
9

Victor .ca.
0 Im
I • I

(C) 33.5m above the toe

Figure 5.28.1 Vectors of accumulated movements during impounding for three reservoir levels

- Non-linear elastic model

126
 (a) 14.5m above the toe

(b) 25.Om above the toe

$c.l.
0 50m

V.ctor scI.
0 Im
I I

(C) 33.5m above the toe

Figure 5.28.1 Vectors of accumulated movements during impounding for three reservoir levels
- Double hardening model

127
 (a) 14.5m above the toe

...
: :

(b) 25.Om above the toe

\W'\ '\ - -
' \ % .. .. -
: I

Scal.
5m

Vsctor ucI.
0 Im
I I

(C) 33.5m above the toe

Figure 5.28.iii Vectors of accumulated movements during impounding for three reservoir levels
- Single hardening model

128
 Distance from Toe tm]
0 2 4 6 8 10 12 14 16 18 20
0 I I I I I I

20
E
.40
g
Nonlinear-Elastic Model -
60
a, - — - - Single-Hardening Model
0 Double-Hardening Model
80 (a)
Observed Deflection
100

20
E
40
C
0
60
a,
0
80

100

20
E
C
0
60
a,
0
80

100

Figure 5.29 Comparison of predicted and measured deflections of asphaltic concrete

membrane at electro-level system 'E' for three levels of reservoir impounding:
(a) 14.5m, (b) 25.Om, and (c) 30.5m

129


Dry Unit Weight Water Content

--.-- Field
50 ____________ --.-- Field 50
Lab. - 21.6kN/m3 Lab. - 7.9%
Lab.- 4.7%
L205J
• /
40 - ..........40

30 - .........30 - ..................
•
-'-S

2 . -. 20

10
Dry Unit Weight, d [kN/m3] Water Content, w [%J

(a) Dry densities, I'd (b) Water contents, w

Figure 5.30 Summary of control tests for Roadford rockflhl

Rockfif I
(a) • Model - Softer Response
--s- Model - Stiffer Response -
0 ., Test - Heavily Compacted
Test - Lightly Compacted
' 1- ...............................................................

- 2 - ...........
C

o 500 1,000 1,500 2,000

Axial Stress, 'a' [Ia)

Sandwaste
Model - Stiffer Response
0 Test
C

0 500 1,000 1,500 2,000

Axial Stress, a' [kPa]

Figure 5.31 Behaviour in oedometer tests for (a) Roadford rockfill and (b) sand waste fill

130


Observed - End of Construction, H=44.Om

Predicted - End of Construction, H=44.Om
Observed - During Construction, H305m
Predicted - During Construction, H=30.5m
Observed - During Construction, H=22.5m
Predicted - During Construction, H23.5m

50

40
E

• 30

0
LL

20
0)
I0)
10

0 I I

0 100 200 300 400 500

Settlement, s [mml

Figure 5.32 Comparison of predicted and measured settlements during construction at

settlement plates 'Si' - 'Improved' prediction

E
E
E
10c_____________
80 Predicted - End of Construction, H=44.Om
Observed - End of Construction, H=44.Om

-80- ................................................................................................................................................
.-100
20 40 60 80 100
-100 -80 -60 -40 -20 0

Upstream [ml Downstream [mj

Figure 5.33 Comparison of predicted and measured lateral movements at magnetic

extensometer 'H' - 'Improved' prediction

131
 Distance from Toe tml

0 2 4 6 8 10 12 14 16 18 20
0

20
F
.40
=
0
60
a,
Predicted Deflection
80 .. . (a)
- Observed Deflection

100

20
E
.4o
C
0
0
a,
a,

80

100

20
F
40
C
0
60
a,
0
80

100

Figure 5.34 Comparison of predicted and measured deflections of asphaltic concrete

membrane at electro-level system 'E' for three tevels of reservoir impounding:
(a) 14.5m, (b) 25.Om, and (c) 30.5m - 'Improved' predictions

132
 40

30
I-
25

.g20
U) Distance from Toe
. 15 4.Om - Observed .
6.Om -Observed
ito lOOm-Observed .
14.Om-Observed
v 4.Om - Predicted
0 I o 6.Om -Predicted -
o lOOm - Predicted
0 I 4.Om - Predicted -

20 0
0

40
0 •0

60
C)

80

(a)
100L
1000 200 300 400 500
Time [days)

Reservoir Head above Toe Em]

0 10 20 30 40
0 . -.-.-,=..-..., I I

V. - --.. -

2O Sand Waste Fill

- ........................
C

(b)

Disnce ;rom Toe ocll

4.Om - Observed
100 - . - -.. 14.Om - Observed
v 4.Om - Predicted
14.Om-Predicted

Figure 5.35 Comparison of predicted and measured deflections of asphaltic concrete

membrane (a) as function of time and (b) reservoir head

133
 CHAPTER 6
NUMERICAL ANALYSIS OF WINSCAR DAM

6.1 INTRODUCTION

During the impounding of Roadford reservoir the deflections of the upstream asphaltic
membrane were only measured in the area of the embankment toe. Lade's elasto-plastic
models successfully predicted these movements for different impounding levels. It was also
noted that, regarding the membrane movements further up from the toe which were not
measured, Lade's elasto-plastic models predicted displacement vectors more normal to the
slope than the non-linear elastic model. Movements of the membrane normal to the slope
have been observed for other rockfill dams of this type built in the U.K. (Charles and Penman,
1988). Thus, it was decided to check the capabilities of both Lade's elasto-plastic models
once more. The availability of field measurements of movements, adequate laboratory test
results and deformation predictions using a simple linear elastic finite element analysis made
the Winscar dam a logical choice for a further numerical study using the Lade's elasto-plastic
constitutive models.

The 53m high Winscar dam was the first in the U.K. to use an upstream asphaltic concrete
membrane. Three horizontal plate gauges (labelled A,B and C) were installed on the 50m high
section of the dam during construction at the positions indicated in Figure 6.1, so that dam
performance during construction and subsequent impounding could be monitored. The
carboniferous sandstone rockfill used in the construction of the dam was described by
Penman ate! (1982). Large diameter oedometer and triaxial tests on scaled down samples of
the rockfill were carried out at BRE. The equipment, methods of sample preparation and test
procedures were the same as for Roadford rockfill described in Chapter 5. The initial
conditions of the samples are given in Table 6.1 together with those achieved in the field. The
test results are shown in Figures 6.2 and 6.3 for oedometer and triaxial tests respectively.
For comparison, the test data for Roadford rockfill from Figures 5.2 and 5.3 are
superimposed. Finite element analyses of the dam behaviour during construction and reservoir
impounding using a 'constant equivalent compressibility' approach (see Section 3.4.1) were
reported by Penman and Charles (1975, 1 985a) where they compared predictions with the
observed behaviour.

134
6.2 DETERMINATION OF MODEL PARAMETERS

The parameters for the two Lade's elasto-plastic models were determined on the basis of the
laboratory test data on the sandstone rockfill using the procedures described in the previous
chapter. They are listed in Tables 6.2 and 6.3 for the double and single hardening model
respectively. The model parameters for Roadford rockfill are inøluded too. They considerably
helped in the derivation of model parameters for Winscar sandstone rockfill.

Predictions for the idealised, single element oedometer and triaxial tests are compared with
the observed behaviour in Figures 6.4 and 6.5. Agreement is quite good apart from the triaxial
test at low confining pressure (o 3 =92kPa) where the stiffness is underpredicted and
volumetric strain severely overpredicted. This is so for both models. An attempt to capture
the 'stiff' behaviour of the sandstone rockfill in triaxial tests at low confining stresses
(O3'=27 and- 92kPa) resulted in an underestimate of compressibility observed in the
oedometer test.

6.3 FINITE ELEMENT ANALYSIS

The finite element mesh used in the analysis is shown in Figure 6.6. Mesh discretization was
chosen in such a way that some nodal points coincided with the location of the
instrumentation used to monitor embankment deformations (see Figure 6.1).

Only two materials were modelled: the sandstone rockfll and the rock foundation. The
upstream membrane was not modelled; it was implicitly assumed to be infinitely thin and
flexible but impermeable.

Sandstone rockfill was modelled using the two Lade's elasto-plastic constitutive models. The
model parameters are derived in the pervious section and are listed in Tables 6.2 and 6.3. The
adopted value of bulk unit weight was y=21.5kN/m 3 (see Tabe 6.1).

The foundation was modelled as a 20m thick layer of linear elastic material with a Poisson's
ratio u=O.1 5. Its stiffness increased with depth: E= 14OMPa for the first 7m and E=700MPa
for the remaining 1 3m of the foundation bedrock (see Penman and Charles, 1 985a).

Construction of the dam was simulated by building in 8 layers. The load of each layer was
placed in two increments. Reservoir impounding was simulated by increasing the water load
in 8 stages. To avoid numerical instability, the water load for each stage was applied over
several increments. The top water level (TWL) was reached after 68 increments. Throughout
the analysis drained cond tions were assumed in the fill materal.

135
A comparison of the predicted and measured movements at the end of construction is shown
in Figure 6.7. The first prediction is made by Penman and Charles (1985a). They used a
simple linear elastic finite element analysis. The non-linearity was introduced through a
'constant equivalent compressibility' approach assigning a modulus value which was
dependent on the final height of the fill (see Section 3.4.1). Using the data from a large
oedometer test and adopting the value for Poisson's ratio u=O.33, the derived Young's
moduli varied between E=6,200kN/m 2 at the embankment toe area to E=17300kN/m 2 at
the centre-line of the dam. The second prediction was obtained using the Lade's double
hardening elasto-plastic model. The single hardening model predicted a similar dam response
during construction and, for the sake of clarity, this prediction is omitted in Figure 6.7.

The finite element analysis using the two different constitutive models (a 'simple' linear elastic
and 'complex' non-linear elasto-plastic) underestimated the embankment movements during
construction. It was especially so at the higher levels and it can be argued that these large
observed settlements may well be due to creep in the sandstone rockfill which both analyses
did not model (Penman and Charles, 1985a), or the difference between the compressibility
of maximum size sandstone rockfill tested in the laboratory and the compressibility of the
larger material in the field (Charles, 1991 b). However, it is of a greater interest here to
compare the predictions using the two different constitutive models because their parameters
were derived using the same large oedometer test 1 . It can be noted that the Lade's double
hardening model predicted somewhat smaller displacement vectors. Further studies (see
Appendix) showed that this was due to the relatively high horizontal stresses in a 'newly'
constructed layer modelled by ICFEP 2 . This is of importance when Young's modulus, E, is
estimated using the minimum principal effective stress, 0 3' (in this case the horizontal stress,
apart from the finite elements at the embankment slopes). In this study the Lade's elasto-
plastic models used equation (4.1) for modelling Young's modulus resulting in a stiffer initial
response than observed in a large oedometer test which was successfully modelled by the
single element test (see Figures 6.4.i and ii) but started from nearly zero initial stresses. If a
sufficiently large number of construction layers had been used resulting in low construction
initial stresses, or if a K 0 -value predicted by a single element oedometer test had been
employed to determine the 'horizontal' stresses in a 'newly' constructed layer, better
agreement with the linear elastic analysis could have been obtained (see Appendix).

A comparison of the predicted and measured deflections of the upstream membrane during

'The Lade's double harden ng model parameters are determined using both large
oedometer and triaxial tests.

2 While the elements are being constructed they are assumed to be linear elastic with
negligible stiffness and Poisson's ratio u=O.45, essentially behaving as a dense liquid (see
Section 2.4.3).

136
full reservoir impounding is shown in Figure 6.8. The merrJbrane deflection measurements
were made at the three levels of the horizontal plate gauges (labelled A, B and C in Figure
6.1) and the crest (the survey reference pillar).

Predictions presented in Figure 6.8.a were made by Penman and Charles (1975, 1 985a). They
used the same method of analysis (a 'constant equivalent compressibility' approach) but the
appropriate 'elastic constants' were selected differently. n the first, 'before-the-event'
prediction (Penman and Charles, 1975) the values of Young's moduli, E, of the rockfill were
obtained from the large oedometer test results by estimating the increase in vertical effective
stress, Ao', caused by the reservoir water. In the upstream rockfill Young's modulus, E,
varied from 12,lOOkNIm2 at the toe to 18,500kNIm 2 at the embankment centre-line. The
same value of Poisson's ratio, u=O.33, as used during construction was assumed. This
analysis significantly overpredicted the membrane movements because the 'elastic constants'
E and u were derived from the one-dimensional compressoi test. This way of deriving the
elastic parameters works well for prediction of movements during construction because the
value of the stress ratio, R=o 1 'Io 3 ', in a large zone of fill is nearly constant while the mean
effective stress, p' =(o 1 ' + 0 2' + 03 )13, is increasing. However, during the reservoir impounding
the stress changes are quite different. In general, the reservoir water pressure on the
membrane will still increase the mean stress, p', but reduce the stress ratio, R .

Charles (1976) simulated these stress changes in the laboratory (see Section 5.3). It can be
concluded from Figure 5.18.b that (Penman and Charles, 1985a): (i) the relationship between
mean effective stress, p', and volumetric strain, , was sirn liar in the two stages of the tests,
(ii) regarding the shear stress, O-O3, the second, 'impounding' stage of the tests represented
an unloading situation resulting in much greater shear modulus, G, than in the first,
'construction' stage of the tests, and (iii) the strain increment, óe 1 , normal to the stress
increment, O3', was very small during the second, 'impounding' stage of the tests. In terms
of the elastic parameters these results imply that the value of the bulk modulus, KB, should
remain constant with a Poisson's ratio u=O.O and a higher value of Young's modulus, E. By
employing such modifications Penman and Charles (1985a) were able to obtain smaller
predictions of membrane movement (see Figure 6.8.a). M3reover, the true pattern of the
membrane behaviour with the movement vectors nearly normal to the membrane was
recovered. This is a consequence of the low value of Poisson's ratio, u, adopted for their
revised analysis of reservoir impounding.

Predictions made by ICFEP using the two Lade's elasto-plastic constitutive models are

3 At the upstream toe of the dam under the higher levels of impounding the stress ratio,
R, can be increased again after rotation of the principal stresses as discussed in Section
5.4.2.

137
presented in Figure 6.8.b. Both models gave good predictions of the membrane movements.
There is an underprediction of the magnitude of movements close to the crest and a slight
overprediction elsewhere, but they are at least of the same quality as the revised prediction
obtained by Penman and Charles (1985a). This is not surprising bearing in mind that both
Lade's elasto-plastic models are capable of capturing the experimentally observed rockfill
behaviour under the 'typical' stress path imposed by reservoir impounding (see Section 5.3).
Notably, they realistically predict the volumetric strains of the rockfill both during construction
and reservoir impounding. They also provide on increased shearing stiffness due to reservoir
impounding without reducing the value of Poisson's ratio. This is clear from Figure 6.9, where
the diagrams from Figures 5.19.ii.b and 5.19.iii.b of the pervious chapter and obtained for
Roadford rockfill 4 are replotted, substituting principal stress ratio, OIO'3 , and principal strain
ratio, /e 1 , with shear stress, a1 - a3, and shear strain, e1 -e3, respectively. It can be seen that
there are no significant changes in the mean effective stress, p', versus volumetric strain, es,,

behaviour during both stages of the tests. The shear strain, E1 -E3, is increasing during the first,
'construction' stage, but the response during the second, 'impounding' stage is very stiff.

6.4 CONCLUSIONS

The simple linear elastic 'constant equivalent compressibility' approach to modelling of rockfill
behaviour is less satisfactory when analysing the effect of reservoir impounding on a rockfill
dam with an upstream impermeable membrane than it is for analysing its behaviour during
construction. To predict the observed membrane behaviour during reservoir impounding
Young's modulus, E, has to be 'adjusted' to give the bulk modulus, K B , estimated from the
one-dimensional compression test adopting very low value of Poisson's ratio, u. This is not
the case with the analyses using the Lade's elasto-plastic constitutive models which are
capable of reasonably reproducing the rockfill behaviour both during embankment construction
and reservoir impounding without changing the model parameters. The predicted movement
vectors which are nearly normal to the upstream membrane during reservoir impounding and
in agreement with available field observations are a consequence of the plastic behaviour of
the underlying rockfill. To predict such a pattern of movements, the elastic finite element
analyses (linear or non-linear) have to use a value of Poisson's ratio much lower than that
required to account for movements during construction.

4The similar results can be expected for Winscar sandstone rockfill.

138
 Condition Test Dry Weight J Moist.Cont. Porosity
______________ _______________ Vd [kN/m3J w [%] n [%]
J
Laboratory Oedometer 20.2 7.0 21 .0

*Triaxial 20.0 7.0 20.0

Fi&d - 19.9 6.0 22.0

(*) - average for four tests at different 03'

Table 6.1 Initial conditions for the large laboratory tests on Winscar sandstone and
average achieved in the field (after Penman and Charles, 1976, 1985; Charles and

Watts, 1 980)

139


Parameter Winscar Roadford Component

Sandstone RockfiH

Modulus Number, M 330.0 400.0 Elastic

Exponent, n 0.70 0.70 Elastic

Poisson's Ratio, u 0.20 0.25 Elastic

Collapse Modulus, C 0.00080 0.00045 Pl.Col.

Collapse Exponent, p 0.70 1.00 Pl.Col.

Failure Constant, , 400.0 260.0 Failure

Failure Exponent, m 0.70 0.65 Failure

Pl.Potent.Const., p 0.40 0.35 Pl.Exp.

Pl.Potent.Const., R 0.0 0.0 Pl.Exp.

Pl.Potent.Const., t 0.0 12.0 Pl.Exp.

Work-Hard.Const., a 3.0 2.6 Pl.Exp.

Work-Hard.Const., fi 0.0 0.0 Pl.Exp.

Work-Hard.Const., P 0.095 0.11 Pl.Exp.

Work-Hard.Expnt, I 1.15 1.05 Pl.Exp.

Table 6.2 Double hardening model parameters for Winscar sandstone

and Roadford rockfill

Parameter Winscar Roadford Component

Sandstone Rockfill

Modulus Number, M 330.0 300.0 Elastic

Exponent, n 0.75 0.50 Elastic

Posson's Ratio, u 0.15 0.20 Elastic

Failure Constant, , 400.0 260.0 Failure

Failure Exponent, m 0.70 0.65 Failure

Pl.Potent.Const., qi2 -3.05 -3.00 Plastic

Pl.Potent.Expnt., p 2.75 2.70 Plastic

Work-Hard.Const., C 0.000150 0.000105 Plastic

Work-Hard.Expnt,p 1.70 1.80 Plastic

Yield Exponent, h 0.975 0.875 Plastic

Yield Constant, a 0.175 0.175 Plastic

Table 6.3 Sngle hardening model parameters for Winscar sandstone

and Roadford rockfill

140
 TwL
w
Asphaltic concrete
17 Scale
membrane
1 0 20m

Sandstone rockfihl

Figure 6.1 Cross-section of Winscar dam with position of horizontal plate gauges A, B
and C (after Charles and Penman, 1988)

0 -

. Winscar RockfiII
N ---0-- Roadford RockfHI
>

2-.............................................................................................
C N

U) 3 -
0

E
0
>
:1
6 I I I

0 500 1,000 1,500 2,000
Adal Stress, c' [kPaJ

Figure 6.2 Results of large oedometer tests on Winscar rockfill (after Penman and Charles,
1976) and Roadford rockfill (after Charles and Watts, 1985)

141
 2500

2,000

0
x

1,500

1,000

500

-7

-6

-5

-4
>
-3

C,) -2
C)

w
E -1
0
>
0

3
0 1 2 3 4 5 6 7 8 9 10
Axial Strain, Ea [1

Agure 6.3 Results of large drained triaxial compression tests on Winscar rockfill (after Charles
and Watts, 1980) and Roadford rockfihl (after Charles and Watts, 1985)

142
 0

>
w
2
C

U) 3
U

6
0 500 1,000 1,500 2,000
Axial Stress, aa rkPa

Figure 6..4.i Comparison of predicted and measured behaviour in oedometer test - Double
hardening model

2
C

U) 3
U

E 4
0
>
5

6
0 500 1,000 1,500 2,000
Axial Stress, a' (kPal

Figure 6.4.ii Comparison of predicted and measured behaviour in oedometer test - Single
hardening model

143
 2500

TA
2000 /.
£

1,500

I
•-'•-
1,000
I
7 U,
U

500 .. ---_,_ - -. -

Confining Pressure
•
. 96kPa -Model
0è I I • 282kPa-Model
A 695kPa - Model
-7 92kPa -Test
282kPa-Test
-6 695kPa-Test

-5

-4
>
(3
-3 ....
C

(/) -2
U • •-.-. .
• ,
E -1 S..................................
0
>
0

3
0 1 2 3 4 5 6 7 8 9 10
AciaI Strain, La [%]

Figure 65.i Comparison of predicted and measured behaviour in triaxial tests - Double
hardening model

144
 2,500

! 1500 -

•
Cl) A
.
1,000
U
a
I'
I

500 / - -I ••

I
Confining Pressure
• 96kPa -Model
• 282kPa - Model I
o -
A 695kPa - Model
-7 - - - -. - ----- 92kPa -Test
282kPa-Test
- ...........................695kPa - Test

5 - ....................................

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

>
- ......
C
U) -2-.........................................................
0
a)
E .•.. ...

Axial Strain, Ea L%1

Figure 6.5ii Comparison of predicted and measured behaviour in triaxial tests - Single
hardening model

145
 (I)
0
4-

'5
C
0

• (0
N
(5

(I,
U)
E
4-
C
E
U)

U)

LL

1.0
'a
E

0
Lt

U,

0
U,

a
U,
0
U,
0
 U,
0)
I--

C
C
•0
0
C
N
C
C.
LU
C

o. .

*0

CI
U,
S..
0

>
 0 h

eçoe

'1

çE jp -'

//
• /1 •.

,1
S '• • - - //
0
o Toe •- ,,- -

--

$
V
Observed Movements
'3 g _____ Predicted by Penman&Charjes, 1975
-. Predicted by Penman&Charles, 1985

(a) predictions by Penman and Charles (1975, I985a) using

a 'constant equivalent compressibility' method

Crest
p

c4i \,,, -

- \ /1
p

•_\ - -
0
0

?e. ' - ____ Observed Movements

Predicted by Single-Hardening Modes
'P , Predicted by Double-Hardening Model

(b) predictions by ICFEP using Lade's elasto-plastic models

Figure 6.8 Comparison of predicted and measured movements of upstream membrane during
reservoir impounding

148
 800

0
6O0

U) 400

U)

200
C,)

0
0 1 2 3 4 5
- 0 Shear Strain, c 1 —c 3 N]

Stage
c 2

3 • I
0
--o-- 2
---0--- 2
1 (Cont.)
1 (Cont.)
6
0 200 400 600 800 1000
Mean Effective Stress ( '+2-3')/3 [kPa]

(a) Double hardening model

800

0.
600

400
(n
U)

200
U)

0
0 1 2 3 4 5
0 Shear Strain, E1 C3 [%]

•- 2 Stage

3 • 1
0
--o--- 2
---0--- 2
1 (Cont.)
1 (Cont.)
6
0 200 400 600 800 1,000
Mean Effective Stress ( '+2a3')/3 [kPaJ

(b) Single hardening model

Figure 6.9 Prediction of rockfill behaviour in drained triaxial test: stage 1, 0 1 '/0 3 ' constant;
stage 2, o' constant, 03 increasing

149
 CHAPTER 7
FIRST-TIME SLIDES IN STIFF PLASTIC CLAYS

7.1 INTRODUCTION

Many U.K. road cuttings and embankments constructed in stiff plastic clays have suffered
delayed superficial sliding. Railway cuttings in similar strata have developed deep-seated slides
up to 100 years after construction. These 'delayed' slides have taken place in previously
unsheared material and can be classified as 'first-time' slides (Skempton and Hutchinson,
1969).

In both cases slides have occurred with an average operational strength significantly less than
the peak strength measured in the laboratory. A number of explanations of this phenomenon
have been proposed in the literature. The clays involved are brittle, and the discrepancy
between the strengths has often been attributed to progressive failure.

Recent enhancements to ICFEP enable both strain-softening and coupled consolidation (and
swelling) to be included in an analysis. The study presented here has involved the application
of these techniques to typical railway (and road) cutting and road embankment slopes to
examine the mechanism of progressive failure. The initial part of the study is of cutting
slopes, as the method of analysis could then be calibrated against the well established history
of delayed slides in railway cuttings (Chapter 8). Subsequently, the mechanism by which
shallow slides occur in embankment slopes and the risk of larger deep-seated slides in older
embankment slopes will be examined using the same method of analysis (Chapter 9).

The present chapter initially discusses various processes involved in delayed slides of man-
made slopes in stiff plastic clays. Special attention is devoted to the mechanism of
progressive failure and parameters influencing It. A brief review of methods used to analyze
progressive failure effects then follows, together with a description of the soil model used
during the coarse of this research. Soil properties of relevance for the analyses reported in this
thesis and the way how they have been determined are discussed in some detail. The chapter
concludes with the hydraulic boundary conditions and slope surface effects typically
encountered in a 'wet' British climate.

150
7.2 DELAYED SLIDES AND PROCESSES INVOLVED

The delayed sliding of slopes cut In stiff plastic clays has been a subject of considerable soil
mechanics interest for a number of years. There is a well established history of delayed sliding
of the slopes of railway cuttings formed In London Clay during the nineteenth and early
twentieth centuries (Skempton, 1948, 1964, 1970, 1977; DeLory, 1957; Henkel, 1957;
James, 1970; Vaughan and Walbancke, 1973). They are deep-seated and have generally
occurred many years after a cutting had been made. The principal reason for this delay is the
very slow rate of equilibration of negative excess pore pressure following excavation through
a process of 'swelling'. Analysis of the slips in these slopes by conventional limit equilibrium
methods has shown that the average operational drained strength is less than the peak
strength as measured in the laboratory. Many explanations have been offered for this
difference. A reduction of the strength through a process of 'softening' and progressive failure
was mostly advocated in the past, but other processes have been proposed tool.

Cut and embankment slopes for motorways built in the U.K. in the last 30 years have shown
a history of delayed sliding which still continues. This history has been established in a major
survey conducted by the Transport and Road Research Laboratory (TRRL) (see e.g. Perry,
1989). Delayed stability problems have arisen predominantly in stiff plastic clays. Slips have
been shallow, typically less than 2m deep, and more pronounced in embankment slopes,
which are usually steeper than cut slopes. It was found that the stability of these shallow
slips was controlled by the critical or fully softened strength, which is much less than the
peak strength measured in the laboratory (Crabb and Atkinson, 1991; Perry, 1991).

7.2.1 Swelling

Because of low permeability, an excavation in saturated stiff plastic clay is undrained. It

unloads the soil in the slope and pore pressures depress both beneath and adjacent to the
excavation. With time the soil swells as pore pressures equilibrate to the long-term steady
seepage regime imposed by the new hydraulic boundary conditions provided by the
excavation. With swelling or equilibration, mean effective stresses reduce and the stress state
approaches the failure envelope. A detailed discussion of this process is given by Bishop and
Bjerrum (1960).

The field evidence for the pore pressure depression immediately following excavation in

1 Tavenas and Leroueil (1981) suggested a rheological 'time-dependent limit state' in which
the strength of the clay ultimately decays to the critical state at a constant shear stress. More
recently, Morgenstern (1990) put forward the 'mechanics of loosening' following excavation
as a possible explanation for a reduction in strength.

151
London Clay is given by Burland and Hancock (1977). Vaughan and Walbancke (1973)
reported measurements of pore pressure changes In cuttings in London Clay that revealed
their slow equilibration. Moreover, calculations indicated that the deduced coefficient of
swelling, c,, was comparable in magnitude to the values measured in the laboratory on large
undisturbed samples.

Numerical studies of the pore pressure changes due to excavation of a slope and their
subsequent equilibration have been undertaken by Vaughan and Walbancke (1973), Eigenbrod
(1975) and Walbancke (1976). Walbancke (1976) showed that the pore pressure changes
after undrained excavation, in the central portion of the slope, can be reasonable
approximated by

= - yh (7.1)

where y is the unit weight of the soil and Ah is the depth of clay removed from above the
point on the slope. However, this equation does not hold for very steep cuts and will
overestimate the pore pressure change under the base of the excavation. It will also
underestimate pore pressure changes adjacent to the crest of the cutting. Pore pressure
equilibration is a swelling process whose rate is governed by the permeability and stiffness
of the soil, and the hydraulic boundary conditions. Eigenbrod (1975) concluded that full
equilibration of pore pressures was obtained for a time factor

T = (c1 t)/H 2 = 0.33 (7.2)

where H is the height of the slope, t is the period of time required for pore pressure
equilibration, and c1 is the coefficient of swelling given by

C3 = k/(my,) (7.3)

in which k is the coefficient of permeability, rn is the coefficient of volume change in swelling

and y is the unit weight of water. Laboratory values of c, for London Clay are rare but for
over-consolidated clays c5 is comparable to the coefficient of consolidation, c. For
c,< 1 .0m2 /yr (see Walbancke, 1976) and a typical cut height H = 1 Om, equation (7.2) gives
an elapsed time t>33yrs for the equilibration of pore pressures. This is comparable to the
time which has been reported between excavation and collapse of many cut slopes in London
Clay. Therefore, it was unnecessary to presume a significant drop in drained strength with
time to explain delayed slides in London Clay cuttings (Vaughan and Walbancke, 1973).

Data on pore pressure equilibration in slopes excavated in the weathered brown London Clay

152
are given by Skempton (1977) and Chandler (1984a), and they are shown in Figure 7.1.b.
Direct measurements of pore pressures in cutting slopes of various ages agree well with
average pore pressure ratios, i, obtained from back-analysis of slope slips by a conventional
limit equilibrium method, assuming the strength parameters c' -1 kPa and 200. It can be
seen that a period of 40 to 50 years is required for completion of the swelling process.
Beyond that period long-term pore pressures have become established and other factors must
control the occurrence of slope slips (Chandler, 1984a).

The stability of clay embankments has been studied less extensively than the stability of cut
slopes. During embankment construction the soil is loaded undrained. In saturated clay
foundations excess positive pore pressure are generated. Stability then improves with time
as these pore pressures dissipate. Placement and compaction of clays as fill typically
generates quite high pore water suctions, particularly in more plastic clays. 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 then decreases with time
as the pore pressures equilibrate and there is a risk of delayed slides.

Field evidence of negative pore pressures at the end of construction of a clay embankment
was given by Bridle eta/(1985). Farrar (1978) and Crabb etal(1987) reported negative pore
pressures measured in road embankments shortly after construction. Time for critical long-
term seepage pore pressures to develop may take several decades. Walbancke (1976)
presented observations showing that negative pore pressures still existed in the upstream
shoulder of a 1 6m high water retaining embankment built of plastic clay 8 years after
impounding. Atkinson and Farrar (1985) reported some dissipation of positive excess pore
pressures in the lower part of a London Clay road embankment 9 years after construction, but
little change was observed in the negative excess pore pressures (suctions) in the upper part
of the embankment.

7.2.2 Softening

The first explanation of the softening process was given by Terzaghi (1936). He recognised
the fissured nature of stiff plastic clays and observed that fissures remain essentially closed
as long as the clay is in its natural state. Following stress relief during excavation some
fissures open, ground water infiltrates into them weakening the cay mass. The process
continues with time and may lead to the sliding.

As a consequence of softening the shear strength of stiff fissured clays deteriorates with

153
time2 . Skempton (1948) investigated many cut slope failures in London Clay in terms of total
stresses and presented the variation of the average undrained shear strength along the slip
surface against the age of the cutting. Similarly, Henkel (1957) plotted the gradual decrease
with time of effective cohesion intercept obtained from the back-analysis of several slips in
London Clay cuttings in terms of effective stresses. James (1970) analyzed a number of case
histories of slope slips in various English stiff plastic clays advocating the time dependency
of softening through a decay in effective cohesion intercept.

Skempton (1970) pointed out that the end result of the softening process is a cohesionless
clay (c'=O) giving a drained strength close to the normally consolidated, 'fully softened', or
critical state value (Schofield and Wroth, 1968). Chandler and Skempton (1974) suggested
that in the brown London Clay and weathered Upper Lias Clay a small cohesion intercept of
c' 1 .5kPa can be relied on in the limit equilibrium method of analysis.

Morgenstern (1977) noted that the result of softening may be a decrease in dilatancy of the
fissured clay at low effective stresses. In the usual working stress range this is consistent
with the fact that softening typically results in a greater variation in effective cohesion, c',
than angle of shearing resistance, ' (see e.g. James, 1971). This concept has been used by
Yoshida et a! (1990) and Yoshida and Adachi (1992) to simulate softening numerically.
Unfortunately, their analyses are dependent on parameters which can only be reaIitically
determined from field observations or back analyses of case histories.

Chandler (1984a, b) regarded softening as 'secondary swelling', i.e. an increase in void ratio
under constant effective stress resulting in a reduction of the available drained strength. Since
softening involves a void ratio increase in excess of that required for swelling, he suggested
that significant softening is unlikely until swelling within the slope is complete. Furthermore,
on the basis of Skempton's (1977) observation that more or less similar values of drained
strength apply to all the brown London Clay first-time slides irrespective of their age at
collapse (see Figure 7.1 .b), he concluded that time dependent softening does not contribute
significantly in reducing the strength towards the critical state value. The good agreement
between field strengths and the 'fully softened' or critical state strength is to some extent
fortuitous (Bishop, 1971a; Chandler, 1984a). It appears that the displacements preceding a
first-time slide through the process of swelling are sufficient to trigger the progressive failure
mechanism (Skempton, 1964,1977).

Terzaghi's concept of softening described above implies that the process acts more or less

2The term 'softening' used in this section should be differentiated from the term 'strain-
softening', i.e. a decrease of peak strength with further shear-straining.

154
uniformly within the cut slope giving rise to generally deep-seated slides. The behaviour of
both cut and embankment clay slopes is more complicated at shallow depth in which summer
shrinkage and winter swelling occurs. This may lead to cracking, increased permeability and,
when given access to water, faster loss of strength through the process of softening in
addition to that predicted by monotonic swelling. The depth of the slips so far observed in
motorway slopes is typically close to or within the zone influenced by these effects. Crabb
and Atkinson (1991) were able to 'recover' these planar slips by the limit equilibrium method
of analysis using the critical or 'fully softened' friction angle, and pore pressures close
to zero. However, a layer of higher permeability on the slope surface, which swells more
quickly, can cause a strain discontinuity and a rupture surface through the mechanism of
progressive failure as well (see Section 8.4 and Chapter 9).

7.2.3 Progressive Failure

Progressive failure refers to the non-uniform mobilization 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. Final collapse of the soil mass can occur before the rupture surface has fully
developed. 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
less than the peak strength and greater than the residual strength.

The mechanism of progressive failure was recognised a long time ago (Terzaghi and Peck,
1948; Taylor, 1 948). The significance in stability problems of the different peak and residual
strengths of soils was emphasized by Skempton (1964), who introduced a residual factor R
as a measure of progressive failure

R = (p)1&pr) (7.4)

where i- is the average shear stress at collapse and i and i, are the average peak and residual
shear strengths at the current normal effective stress along the eventual rupture surface.
Turnbull and Hvorslev (1967) and Peck (1967) associated progressive failure with non-
uniform Stress and Strain conditions within a slope. Bjerrum (1967) discussed how an
excavation in stiff plastic clay leads to overstress at the toe and the development of a slip
surface from the toe inwards. He suggested that the release of strain energy, stored in over-
consolidated plastic clays and clay shales, caused by weathering, provides a driving force for
a progressive failure mechanism. Bishop (1967) distinguished possible progressive fa lure
mechanisms in drained and undrained conditions. Subsequently, Vaughan and Hamza (1977)

155
deduced that progressive failure could be more significant in undrained slope stability
problems, as shear stress increases with depth, while undrained strength may not do so. On
the other hand, in drained failure the strength would typically increase with depth, and so
progressive failure should be less significant. Bishop (1987) considered progressive failure to
play a major role in long-term stability problems, particularly delayed slides. He identified
brittleness, the reduction in strength in passing from peak r to residual r,, as the most
important factor controlling progressive failure and defined a brittleness index 'B as the
maximum percentage error arising due to progressive failure in a brittle soil

'B = (r-T,)/r. in % (7.5)

However, the brittleness index alone Is not adequate to quantify the material susceptibility
to progressive failure. The rate of post-peak strength reduction is of equal importance. If it
is low, then progressive failure may be unimportant even when the brittleness index is
significant (Vaughan and Hamza, 1977). Duncan and Dunlop (1969) and Dunlop and Duncan
(1970) explored the non-uniformity of stress distribution in slopes and indicated a
considerable influence of the initial horizontal stresses on the magnitudes of shear stresses
following excavation. The stiffness of the materials involved will determine the distribution
and magnitude of strains and therefore will effect progressive failure as well. Palmer and Rice
(1973) studied the factors controlling the growth of a flat slip surface in the progressive
failure of an infinite clay slope and illustrated the importance of the energy release and thus
the role of unloading stiffness. They also noted the presence of a strong size effect in their
analyses. Lo and Lee (1973b) concluded that progressive failure effects for a slope under
drained conditions and a constant shape increase with height and inclination. Dounias et a!
(1989) came to the same conclusion for the height but found the opposite to be true for the
slope inclination. Namely, a flatter slope may be less stiff than a steep one and, deforming
more non-uniformly, it suffers more progressive failure.

Both Bjerrum (1967) and Bishop (1967) suggested that the discrepancy between average
operational strengths deduced from back-analysis of actual slides in stiff plastk clays by limit
equilibrium methods and peak strengths measured in laboratory tests might be due to
progressive failure. This is even so when the 'bulk' strength s is considered and appropriate
allowance for the effects of strain-rate on strength is made (Skempton, 1977; Chandler,
1 984b). However, there has been little field data from which the effect of progressive failure
can be deduced unambiguously.

3Peak strength of a mass of soil obtained from laboratory samples sufficiently large to
contain a representative number of discontinuities.

156
 Burland et a! (1977) provided rare field evidence showing that horizontal stress relief as a
result of sequential excavation in stiff plastic Oxford Clay can generate a progressive slip
surface having a strength at, or close to, the residual value. Morgenstern (1990)
acknowledged the existence of progressive deformations in this important study, but he
questioned whether or not these deformations resulted in failure of previously unsheared
material.

More recently, the end-of-construction failure of the Carsington embankment provided another
field example of a possible progressive failure mechanism. The field studies and the limit
equilibrium analyses have been described by Skempton and Coats (1985) and Skempton
(1985). The limit equilibrium analyses showed that the average strength at failure was
approximately mid-way between peak and residual for the materials involved along the slip
surface. The use of peak strength parameters overestimated the factor of safety by more than
- 20%. Since the failure was through brittle plastic clays, this difference was attributed to
progressive failure. However, in order to quantify this, it was necessary to recover the effect
of progressive failure directly from analysis. A constitutive model incorporating strain-
softening was developed and incorporated into ICFEP, and finite element analyses were
performed to recover the effect of progressive failure independently (Dounias, 1987; Potts
et a!, 1990). These and other analyses (Douriias et al, 1988, 1989; Vaughan et a!, 1989)
clearly showed that the effect of progressive failure can be reproduced arid quantified.
However, all of their analyses were applied to large embankments which are subjected to
loading and consolidation, rather to cuttings and road embankments which invoke unloading
and swelling. These have not yet been examined by strain-softening finite element analysis
of this type.

7.3 METHODS OF ANALYSIS FOR PROGRESSIVE FAILURE

In the absence of clear field evidence, an enormous effort has been made in the past to
reproduce the effects of progressive failure numerically. However, the mechanism of
progressive failure is complex and analysis is difficult. In the following the available methods
of analysis are discussed briefly with emphasis on recent advances in numerical modelling of
progressive failure and delayed slides in genera. The soil model used for the work described
in this thesis to examine the effects of progressive failure in typical railway cuttings and road
embankments is outlined as well.

7.3.1 Available Methods of Analysis

Dounias (1987) summarized various attempts to analyze progressive failure effects and
identified three different approaches, namely: (i) limit equilibrium analysis, (ii) analysis using

157
purpose built models, and (iii) finite element analysis.

7.3.1.1 Limit Equilibrium Analysis

The effect of progressive failure is strain-dependent end it cannot be recovered from

conventional limit equilibrium stability analyses. Nevertheless, it was common in the past to
reduce the effective cohesion c' to zero and the angle of shearing resistance from peak 'P.'
to a lower residual p,' value over parts of the assumed slip surface (see e.g. James, 1970,
1971). Bishop (1971a) recommended the use of an empirical distribution factor based an
Skempton's residual factor A (see equation (7.4)), and suggested that the analysis of delayed
slides based on this factor is of greater generality than the postulate of an uniform
mobilization of a 'fully softened' strength.,, c' =0 (Skempton, 1970). However, he noted that
predictive value of such an analysis would be no better until backed by analytical methods
taking into account the following factors: (1) the rate of post-peak strength reduction, (ii)
swelling characteristics of the soil, (iii) the pre-peak stiffness of the soil under the appropriate
conditions of stress change, (iv) the value of K0 before the formation of the slope, (v) the
geometry and scale of the slope, and (vi) the long-term flow pattern of the ground water.

7.3.1.2 Purpose Built Models

Some of the above factors have been investigated by Bjerrum (1967) and Christian and
Whitman (1969) using one-dimensional models to study the development of a continuous
sliding surface in an infinitely long slope. More realistic two-dimensional models have been
proposed by Blacker (1977), Bernander and Olofsson (1981), Bernander and Gustaas (1984),
Bernander (1985).

Although the purpose built models are useful in providing an insight into the effects of
progressive failure on the stability of slopes, they can only account for the parameters
affecting progressive failure in a crude way. In addition, the exact failure mechanism must be
postulated in advance.

7.3.1.3 Finite Element Analysis

It was recognised by Peck (1967) and Bishop (1971 a) that a better understanding of the
problem of progressive failure would require a finite element solution for the strain-softening
soil. However, the numerical difficulties aris ng from the softening process are significant and
have hampered the finite element analysis of progressive failure.

Approximate analyses involving a combination of limit equilibrium stability analyses with an

158
elastic finite element stress analysis were performed by Lo and Lee (1973a,b). They were also
able to model the decrease of drained shear strength with ttme and predicted the time to
failure for several cuttings in London Clay.

Hoeg (1972) developed an incremental elasto plastic strain-softening model and, by means
of the finite element method, analyzed the undrained bearing capacity of soft clay. The
theoretical justification for using the classical theories of plasticity to simulate strain-softening
behaviour was later provided by Prevost and Hoeg (1975), who discussed the stability and
uniqueness of the solution. Analyses of this kind using a variety of strain-softening models
were also undertaken by Sture and l(o (1976), Biondi et 81(1976), Kalteziotis and Kyrou
(1985).

Gates (1 972) was the first to analyze a slope in strain-softening soil. The model was elasto-
plastic with strength dropping instantly from peak to residual. Fifteen years later Chan and
Morgenstern (1987) employed a similar model to study the magnitude and migration of a slip
caused by strain-softening of a thin bentonite layer 24m below the base of 22m deep
excavation. More recently, Yoshida and Adachi (1 992) used an elastic, post-peak plastic
strain-softening model to analyze a cutting in clay shale. Unfortunately, the emphasis in their
analyses was on 'softening' as a time-dependent process leading to a reduction in shear
strength (see Section 7.2.2) rather than on 'softening' in the post-peak stress-strain response.
They mentioned the possibility of coupling the swelling process in their analyses (see also
Yoshida et a!, 1990), but no such analyses have been presented in either publications.
Griffiths and Li (1993) described an elasto-plastic finite element technique that takes account
of coupled swelling/consolidation and investigated the transient stability of excavated soil
slopes. However, a liner elastic perfectly plastic stress-strain law was employed so that the
effects of progressive failure could not be modelled.

Numerous applications of critical state models in the finite element analyses of embankments
have been reported in the last 20 years (see e.g. Duncan, 1994).. Coupled consolidation has
often been included. Only short-term problems involved in construction on weak alluvial clays
have been considered. This is beyond the scope of this thesis. Nevertheless, it is worth
mentioning that critical state models can be used to study progressive failure effects, but
critical state does not usually coincide with residual conditions (see Section 7.4). They can
only model softening due to dilation but not due to particle orientation.

It has been recognised for a long time that strain-softening behaviour usually leads to the
formation of so-called 'shear bands' of 'softened' material within a soil mass along which the
deformation tends to localize. Pietruszczak and Mroz (1981) pointed out that the finite
element solution of strain-softening materials suffers from sensitivity to mesh size because

159
of this localized deformation. In order to avoid this sensitivity they modified the finite element
formulation of the stiffness matrix to account for the existence of a shear band within an
element through a sub-element approach. A similar approach has been adopted by Tanaka and
Kawamoto (1989) in their numerical study of cutting slope failure in London Clay at Sudbury
Hill (Skempton, 1977). The analyses were performed by increasing the gravity load acting on
the slope until large displacements developed. They found zones of high shear strains in
locations that corresponded to the position of the actual slip surface. By introducing the shear
band width and its ratio to the characteristic element dimension, their analyses became
essentially insensitive to the mesh characterization. However, the load factor at which large
displacements developed was related to the assumed width of the shear band. More recently,
Pietruszczak and Niu (1992) presented a more rigorous formulation of the above approach,
which was invariant with respect to the thickness of the shear band. However, the analysis
of strain-softening soil becomes much involved requiring identification of material
characteristics of the shear band. It should be pointed out that the above approach is only one
of several approaches for the modelling of localized deformation which have emerged in the
last fifteen years. The analysis of shear band behaviour is still the subject of much research
and debate, from both an experimental and numerical point of view.

Dounias (1987) has concluded that the finite element method incorporating non-linear elastic
strain-hardening/softening plastic constitutive models offer the most versatile and efficient
way of analysing progressive failure effects. The size of the finite elements involved will
inevitably affect the resulting progressive failure, but if uniform finite elements of comparable
size are employed in the area of interest and great care is exercised in the determination of
the rate of post-peak strength reduction, realistic results can be expected.

7.3.2 Soil Model Used in the Thesis

The constitutive model employed here to analyze the effect of progressive failure in delayed
slides of cut and embankment slopes is the same as the model used previously by Dounias
(1987) and Potts et a! (1990) in their finite element analyses of progressive failure of
Carsington embankment. Some details of the model have already been described in Section
5.2.1. Softening behaviour is introduced here by allowing the angle of shearing resistance,
', and the cohesion intercept, c', to vary with the deviatoric plastic strain invariant, ED, as
shown in Figure 7.2. The deviatoric plastic strain invariant, E0 , is defined as

( EDrI 2 = 2i(E 1 -e2 ) + ( E2 -E3 ) + (e3-e1)]I3 (7.6)

The angle of dilation, y', may also be varied with E0.

160


It can be seen that the model requires the specification of peak (', c a') and residual strength
(' Cr'), pre-peak stiffness (E), stiffness in unloading (E) and the rate at which strength is
lost with post-peak strain 4 ed). Swelling and consolidation are 'coupled' arid are
modelled by allowing flow according to a specified permeability 5, the soil stiffness and the
hydraulic boundary conditions. Steady state seepage conditions are adhieved at the end of
consolidation (swelling) and a consolidation solution using a linear elastic soil model and a
long time history can be used to obtain it.

In previous analyses examining progressive failure (Dounias, 1987; Potts et a!, 1990) it has
been found that stability is not strongly affected by the pre-failure stiffness model used. A
simple elasto-plastic soil model is used here, with stiffness prior to failure given by an elastic
model in which Poisson's ratio, u, is constant, and Young's Modulus, IE, varies with mean
effective stress, p', but not with shear Stress level, S. No distinction is made between
unloading and loading. Consolidation and swelling follow approximately the same modulus in
stiff clays and, in the solutions presented here, volumetric strains are based on the
appropriate swelling modulus.

Most slope stability studies have been based on limit equilibrium analyses, n which an internal
force distribution is assumed. There is an inherent difference between the strength on the
rupture surface assumed in these analyses and the strength assumed in the finite element
analyses, even when the same strength envelope is specified. In the former the strength on
the slip surface is, according to the Coulomb failure criterion

Tf = c' + a' tan' (7.7)

The finite element analysis is of a continuum, and failure is according to the Mohr-Coulomb

4The post peak behaviour of strain-softening soils in laboratory experiments, such as the
ring or direct shear devices, is expressed in terms of displacements rather than strains. There
is a provision for this in ICFEP and, by specifying the softening function in terms of nominal
displacements the size of the element can be accounted for to some extent. However, it is
more common to express the softening function in terms of shear strains. h order to evaluate
the shear strains, the displacement necessary to bring residual conditions may be divided by
the minimum height of the element that can be independently deformed. For eight-noded
elements with 2x2 integration scheme this is the minimum distance between two nodes, i.e.
half the thickness of an element. This approximate procedure is a convenient way of
incorporating strain-softening effects into a finite element analysis.

5 Constant or spatially varying permeability is used in all analyses reported here. For more
realistic modelling of soil behaviour the soil permeability should vary as the void ratio of the
soil changes (see e.g. Vaughan, 1989). This option has only recently become available in
ICFEP.

161


criterion

a1 ' - a3 ' = (a1 1 +a3t)sinp' + 2c'cosp' (7.8)

The solution indicates failure surfaces (a set of velocity characteristics along which there is
no linear strain, see e.g. Davis, 1968) which are generated at angles ± (45°-qi/2) to the major
principal stress direction, where u is the specified angle of dilation. In associated plasticity,
4V=, and the two failure criteria are consistent, with the strength along velocity

characteristics in accordance with the Coulomb criterion. With çv<Ø' the strength along the
velocity characteristics predicted by the finite element analysis is less than that predicted by
the Coulomb criterion

c'cosp'cosç +

7., = (7.9)
1 - sin'sinq'

If the rupture surface is constraned by the geometry to occur in a particular place, then the
finite element analysis will predict collapse at lower shear stresses than the limit equilibrium
analysis in which the same assumptions for c' and ' are made. Consistent predictions of
collapse can then only be achieved by modifying the shear strength envelopes assumed in one
of the analyses (Potts et al, 1990). If there is no constraint on the location of the rupture
surface, then collapse is likely to be predicted by both analyses at the same stresses, but with
different rupture surfaces.

In real clays, failure surfaces are typically observed at angles of about ± (45°-'I4) to the
major principal stress direction, which is approximately equivalent to cv="I2. As discussed
subsequently, it is possible to specify çu—Ø'/2 in the model 1 and initially this was done for
peak strength, with qi decreasing linearly to zero as the strength dropped from peak to
residual. However, for reasons d scussed in Section 8.3.2, more realistic results were found
when qi=0 was assumed in the analyses. With this assumption, the velocity characteristics
predicted are at ±45° to the direction of the major principal stress and, at rupture, the peak
strength on the failure surface is given by

7, C'COS' + c7'SIfl' (7.10)

For the adopted strength parameters (see Table 7.2) and the normal effective stresses of
interest (see Figure 7.5) the difference in strength on the failure surfaces from the Coulomb

6This combines the Coulomb failure criterion with the assumption that the critical failure
surface is that on which T/O",, is a maximum (the plane of maximum stress obliquity).

162
criterion is about 5%. Since there is little geometric constraint, the positions of the failure
surfaces from limit equilibrium and finite element analysis will differ, but the effect on the
prediction of collapse will be slight. No adjustment of the strength parameters has therefore
been made.

7.4 RELEVANT AND ASSUMED SOIL PROPERTIES

Three characteristics of clay behaviour control delayed collapse. The first is low permeability,
which gives rise to delayed pore pressure changes over long periods sometimes in excess of
100 years. The second is large strains during swelling and shrinkage. The third characteristic,
especially for the more plastic clays, is low residual strength and high brittleness, and a
consequent potential for progressive failure.

The influence of clay content on brittleness and progressive failure, and on the deterioration
of a slope surface is summarized in Table 7.1.

Figure 7.3 shows how critical state and residual angles of shearing resistance vary with
increasing plasticity and clay content for some British c ays. Peak strength will be higher than
critical state strength due to density and dilation and, in in-situ soils, by cementation and
other bonding effects developed with geological age. Critical state and residual strength
depend only on granulometry (Lupini et a!, 1981). When clay content is low, strength is
controlled by the rotund granular particles present, which do not orientate significantly during
shear. Residual and critical state strengths coincide, and are relatively high. When there is a
high content of platy, low-friction particles, these orientate during shear, leading to polished
shear surfaces and to low residual strength. Transitional behaviour occurs in soils of
intermediate clay content.

The coefficient of volume change in swelling, m, increases with clay content, although in in-
Situ clays, swelling may be inhibited by cementation and bonding. The potential for
progressive failure increases with the magnitude of the strains induced by swelling.

Strains due to swelling and shrinkage of the surface soil which is subject to seasonal changes
in pore water Suction also increase with increasing clay content. These strains can cause
cracking, increased permeability and higher seepage pressures due to perched water table
effects. They may cause a reduction in peak bulk strength. The risk of shallow sliding will be
increased by these effects. Thus the potential for superficial instability increases with
increasing clay content due to these additional factors.

163
It is known that stiff, low plasticity glacial tills1 have higher permeability, smaller strains
during swelling, and higher residual strength than stiff plastic sedimentary clays (Vaughan and
Walbancke, 1975). Thus, the potential for progressive failure and surface deterioration is
limited. Limit equilibrium methods of slope stability analysis, based on strengths measured in
the laboratory, work quite well. Skempton and Brown (1961) showed that the cohesion and
friction as measured in the laboratory were mobilized in the field. This was further supported
by observations of Vaughan and Walbancke (1975).

The properties adopted for the analyses presented in this thesis for both cut and fill slopes
are those of a typical stiff plastic clay of the type causing most stability problems. They are
based on published data for London Clay. There is a well established history of delayed slides
of the slopes of railway cuttings in London Clay. Information about its soil properties, both
in the laboratory and in the field, is more complete than for any other clay strata. Thus this
strata is best suited for validation of the methods of analysis adopted. The properties of many
other English stiff plastic clays are essentially similar to those of London Clay, and the
analyses presented here are thus representative of both cut and embankment slopes in these
materials.

The properties assumed in the analyses are summarised in Table 7.2. The derivation of these
properties is now discussed.

7.4.1 Cut Slopes

7.4.1.1 Peak Strength

In-situ stiff plastic clays frequently contain discontinuities, principally fissures. These are
usually non-planar, approximately random in orientation, and have matt surfaces. The
discontinuities reduce 'bulk' peak strength below 'intact' peak strength. There is very little
data from which the reduction can be deduced. A simplified analysis (Skempton and Vaughan,
1993) suggests that the bulk strength should approximate to the average of the peak intact
strength and the strength on the fissures. This assumption is consistent with the results of
Sandroni (1977), who performed undrained triaxial tests with pore pressure measurement on
samples of brown London Clay with closely spaced fissures. Various samp'e sizes were used,
the largest being 250mm diameter and 500mm long. These samples probably contained
representative patterns of fissures. The results are shown in Figure 7.4. The peak strengths
used for the in-situ soil in the present investigations (see Table 7.2) are taken directly from

7Composition of glacial tills is very variable. More plastic glacial tills will probably behave
in a manner comparable with that of stiff plastic clays.

164
Sandroni's results and include the effect of discontinuities.

7.4.1.2 Residual Strength

The residual strength of London Clay measured on planar surfaces in the laboratory ring shear
apparatus is typically of the order of c,'=O, r11OO (Lupini et a!, 1981). Back-analysis of
existing slides (which involves approximations) usually indicates a residual strength greater
than this, possibly because field rupture surfaces are not planar. Results from back-analyses
from Bromhead (1978) and Chandler (1984a) are shown in Figure 7.5,, together with ring
shear results. Back-analyses at normal effective stresses a,' = 30-5OkPa indicate a strength
of c.'=O, r' = 14°. A residual strength of c,'=2kPa, Ø'=1O was used in the initial finite
element analyses, but it was apparent that collapse was predicted more readily than it
occurred in the field. Subsequently, the higher strength of c,'=2kPa, Ør' = 13° was adopted.
This strength is shown in Figure 7.5, together with the average normal stress range operating
on the rupture surfaces predicted by the finite element analyses. It is slightly higher than that
indicated by back analysis.

7.4.1.3 Brittleness

Brittleness is simulated in the analysis by adopting a flat peak. i.e. the strength remains
constant up to a certain deviatoric plastic strain, followed by a linear drop from peak to
residual strength with increasing deviatoric plastic strain. Definitions are given in Section
7.3.2. The results of typical drained and undrained triaxial tests predicted by the soil model
are presented in Figure 7.6.1. The predictions for undrained triaxial tests are compared in
Figure 7.6.ii with the results of large triaxial tests on brown London Clay by Sandroni (1977).
With isotropic elastic properties, the stress path predicted for the undrained triaxial
compression test involves no pore pressure change due to shear, although pore pressure
increases during loading because of the increase in mean total stress. Post-peak, with the
angle of dilation qi=O, the pore pressure drops as mean total stress decreases. The volume
changes predicted in the drained triaxiat test are consistent with the pore pressure changes
in the undrained test. The predicted undrained strength is a function of mean effective stress,
p', as in the real soil.

One or more shear surfaces are likely to form as peak strength s mobilised in real plastic
clays, and loss of strength post-peak occurs as a consequence of sliding on these surfaces.
Such thin discontinuities are not reproduced in the finite element analys s, in which the
minimum thickness of a rupture zone is approximately half the thickness of an element. This
must be taken into account when specifying the rate at which strength is lost post-peak. The
strength-displacement plot for a shear zone of thickness H in simple shear is shown in Figure

165
7.7. Typically, half an element is 0.5m minimum. A displacement scale equivalent to this is
shown in the figure. Residual strength of the in-situ soil is reached after a total displacement
of 0.1 m. Ring shear tests on remoulded clays show that strength drops rapidly post-peak and
that a strength about 2° above residual is reached after 50-100mm displacement. The
strength continues to drop slowly until a displacement of the order of 500mm is reached.
Tests on London Clay (Garga, 1970; Bishop eta!, 1971) show that the undisturbed clay is
more brittle than the remoulded clay.

7.4.1.4 Pro-peak Stiffness

This is modelled on behaviour during swelling, as this has a direct influence on strains, strain-
softening and progressive failure. The amount of swelling depends predominantly on the pre-
peak stiffness assumed. The assumptions used in the analyses are based primarily on
modelling reasonable strains in one-dimensional swelling over the appropriate stress range.
Typical stress paths are shown in Figure 7.8.a and the matching strains in Figure 7.8.b. Strain
increases with decreasing mean effective stress, p', in a non-linear manner because of the
non-linear function relating the Young's modulus E to p'. There is some strain-softening at low
stresses.

When plotted in a conventional manner against vertical effective stress (Figure 7.8.c),
relationships depend on the initial value of K 0 assumed, and a discontinuity occurs between
elastic and elasto-plastic behaviour. Typical re'ationships are compared with results from
laboratory oedometer tests on undisturbed samples which had been pre-loaded to 3,000kPa
(Apted, 1 977), and with an average field swelling curve obtained by plotting water contents
from borehole profiles against estimated vertical effective stress, a,'. The predictions generally
lie between the laboratory and the field curves. This is reasonable, as the field curves are very
long-term and may include some weathering effects, and the laboratory curves are short-term
and do not allow for creep. The same stiffness was specified in consolidation and swelling,
as discussed in Section 7.3.2.

7.4.1.5 Coefficient of Earth Pressure at Rest

The lateral stress in the ground prior to excavation of a cutting slope controls the average
total stress change on excavation and hence the pore pressure changes predicted after
undrained excavation. This controls the increase in pore pressure during swelling, the
volumetric strains and the release of strain energy, and the potential for progressive failure.
When an elastic model is used to model pre-peak stiffness, the initial effective stress path
during one-dimensional swelling is controlled by the value of Poisson's ratio, u, adopted. With
sufficient unloading the yield Criterion S satisfied, and the effective stress path then moves

166
down a passive failure line towards the origin of stress space. The paths according to the
assumptions made for the analyses are shown in Figure 7.8.a. The initial value of K 0 is
specified in the analyses.

The value of I( in heavily over-consolidated plastic clays is usually quite high, but it is difficult
to estimate and can vary with ground slope, degree of weathering, tectonic stress changes,
reloading by recent deposition of alluvium, head, etc. In the analyses performed the value of
K,, was varied in the bulk of the soil to determine its significance. However, in all analyses a
value of K, =1.0 was adopted at initial ground level, changing linearly with depth to the
specified value at the level 1 m below, where the ground water table was assumed. This was
to simulate the effect of surface weathering.

7.4.1.6 Undrained Strength Profile

The undrained strength is only invoked in the early stages of the analyses. The undrained
strength predicted by the model is mobilised when the linear stress path arising from the
assumption of isotropic elasticity pre-peak intersects the failure envelope in terms of effective
stress. The undrained strength thus depends on the initial effective stress and on the changes
in the intermediate principal stress (plane strain or triaxiat conditions). If the clay is sampled
at constant effective stress, the initial isotopic effective stress is p0' = o'(1 + 2K0)13, and the
undrained strength, S, of the sample when tested in compression is

S, = (c'/sin' ^ p0')sin' (7.11)

S,, = E(c'/sin' + p0')sin.p']I(1 - sin'/3) (7.12)

for plane strain and triaxial conditions respectively.

The predictions from the soil model depend on K0 . The predicted undrained strength profiles
are shown in Figure 7.9, for two different values of Kc,, and for triaxial and plane strain tests
on perfect samples. Figure 7.9.a shows the profiles for the usual ground water conditions
assumed in the analyses of cuttings. This undrained strength was found insufficient to
support a lOm high embankment constructed on an undrained foundation, and the profile
shown in Figure 7.9.b was adopted for the embankments built on a clay foundation. In this
profile an increased suction was adopted near the surface to give an increase in undrained
strength sufficient to support the lOm high embankment.

7.4.1.7 Coefficient of Permeability

The coefficient of permeability assumed in the analyses controls the rate of swelling. Its

167
spacial variation controls the long-term equilibrium pore pressures to which seepage pressures
tend. The permeability of any particular soil varies with void ratio, and hence with effective
stress. When flow is in the direction in which effective stress changes, the distribution of pore
water pressures is affected to a significant extent (Bromhead and Vaughan, 1980; Vaughan,
1 989). The permeability decreases with depth and increasing effective stress according to a
relationship of the form

ln(kIk0) = (7.13)

where k is the coefficient of permeability in [rn/si, o,' is the vertical effective stress in
[kN/m 21, k,, is the permeability at zero stress in [mis], and B is a coefficient in [m2ikNJ.

If there is downward flow, significant perched water table effects develop, from which the
change in permeability with depth can be deduced. The value deduced for the in-situ London
Clay (Vaughan eta!, 1983; Vaughan, 1989) was B =0.0075-0.01 m 2/kN. This relationship has
only recently been incorporated in ICFEP and a linear variation in permeability with depth was
specified. The variation of permeability with depth assumed in the ana'yses is shown in Figure
7.10, together with some in-situ measurements of permeability for the London clay, and for
the brecciated Upper Lias clay. The variation of permeability with depth deduced by Vaughan
eta! (1983) for the London Clay is shown. The permeability specified in the finite element
analysis remains constant throughout swelling.

The coefficient of consolidation is not specifically invoked in the analyses. However, the
permeability specified above, together with the soil model used, give values for the coefficient
of swelling, c8 . Using the compressibility shown in Figure 7.8 for K 0 = 1.5 and equation (7.3),
values of c8 for one-dimensional swelling to half initial vertical effective stress are 2.7m2/yr
at 5m depth, 0.87m 2/yr at lOm depth, and 1.3m 2/yr at 20m depth. Walbancke (1976) quotes
values deduced from pore pressure observations in the London Clay of c=3.2m 2/yr for the
brown clay and of c8 =0.8m 2 /yr for the blue clay below. These values are somewhat higher
than are usually obtained from laboratory tests. They are similar to the assumptions used in
the analyses.

Changes in pore pressure during swelling and consolidation are best displayed as a
proportionate change between the initial values immediately after slope excavation and the
ultimate ones given by steady seepage conditions. Where appropriate, long-term equilibrium
pore pressures in the slopes analyzed here have been predicted by analysing an elastic slope
of the same geometry with the same permeability distribution and hydraulic boundary
conditions, and allowing this to swell to equilibrium.

168
7.4.2 Fill Slopes

7.4.2.1 Peak Strength

Comparative test studies have shown that the structure of field compacted fills differs from
the structure of samples of the same clay compacted in the laboratory (Vaughan, 1977). This
difference in structure causes significant differences in undrained pore pressure response and
undrained strength and stiffness. However, peak strength in terms of effective stress depends
more on dry density than structure (Vaughan et a!, 1978). Typical results from triaxial
compression tests on samples taken from field compacted plastic clays are shown in Figure
7.11. Fills derived from four different plastic clays give very similar results.

Figure 7.12 shows another set of data for various British plastic clay fills (Perry, 1991; Crabb
and Atkinson, 1991). Six different clays show similar strengths. Strength at low stress, which
controls highway embankment stability (particularly in shallow slides) was examined
specifically, and tests at low stress (s'<5OkPa) were failed by increasing the pore pressure
at constant deviatoric stress. It can be seen that the strengths observed were close to the
tensile failure limit.

Atkinson and Farrar (1985) found that the strength envelope at low stress was expressed
best by a failure criterion of the form

= A.(c) b (7.14)

where A and b are experimental parameters. Such a failure criterion can be used directly in
limit equilibrium analysis (Charles and Soares, 1984; Perry, 1991). For a finite element
analysis it can be transformed into a Mohr-Coulomb equivalent, which may be done in terms
of a' and t as

St = Q' + (7.15)

t = rpIcosp' (7.16)

where 4' varies and is given by

tan' = 8T,JøO' = (7.17)

The result is complicated and the difference between the strength on the plane of maximum
stress obliquity and on the failure plane (when '^') varies with stress level. A criterion of

169
this form is not used in the current soil model, and an equivalent linear failure criterion has
been adopted.

The peak strength adopted for the analyses reported here is that deduced from laboratory
tests on typical field compacted London clay. No reductions have been made to allow for
anisotropy of strength or the presence of other weaknesses. The strength envelope adopted
is ;'=l2kPa, '=2O°.

7.4.2.2 Residual Strength

Residual strength is a fundamental soil property as t develops only after any initial structure
has been destroyed by large strains, and when the void ratio remains constant with further
shearing. Thus the same value as that adopted for the in-situ clay was adopted for fill slopes,
namely Cr' = 2kPa, ,' 13°.

7.4.2.3 Brittleness

Laboratory test results indicate that the post-peak displacement on a discontinuity required
to reduced strength from peak to residual is greater in plastic clay fills than in the in-situ clays
from which they are formed, probably because the bonded structure present in the in-situ clay
is progressively destroyed by excavation and compaction. This displacement is increased
further by remoulding in the laboratory (Vaughan, 1977). The assumptions made in the
analysis are summarised in Table 7.2, and the equivalent triaxial compression test curves are
shown in Figures 7.6.i and iii, where they may be compared with the assumptions for the in-
situ clay. The stress displacement relationship for a tialf-element thickness of fill is shown in
Figure 7.7.

7.4.2.4 Pre-peak Stiffness

The breakdown in structure referred to above also reduces pre-peak stiffness. Test results
typically show a failure strain in fills which is three times that observed for the intact clay.
The assumptions made for the analysis are summarised in Table 7.2 and Figures 7.6.1 and iii.

The pre-peak stiffness controls the volumetric expansion during swelling, according to the
changes in effective stress, and thus it controls the strains to which the fill is subjected, and
therefore the amount of progressve failure. The stress paths and the volume changes during
swelling are summarised in Figure 7.13.

170
7.4.2.5 Undrained Behaviour

During construction the embankment fill will behave largely undrained. Real fill would be
partly saturated. It can be shown (Vaughan, 1978) that effective stresses in partly saturated
soil of relatively high degree of saturation can be calculated in the usual way by subtracting
pore water pressures from total stresses. This will be assumed in the present work. Pore
pressures change during undrained loading according to a relationship of the form

Au = BAp (7.18)

where B increases with pore pressure, and is water content dependent. A typical British fill,
from a saturated borrow pit, would be sufficiently wet for B to approach 1.0 at high stress.
This effect can be reproduced in the soil model by incorporating a non-linear compressible
pore fluid. However, such a model was not available at the time of this research, and the fill
has been modelled as if it were saturated, with 8=1.0. This gives an undrained strength
which is independent of mean total stress, p.

The properties assumed have been adopted to give reasonable in-situ pore pressures at the
end of construction, as these control the amount of subsequent swelling. Therefore, an initial
suction of 1 2OkPa has been adopted. In one-dimensional loading, this gives a constant vertical
effective stress of l2OkPa. With a bulk unit weight of 18.8kNIm3, the pore pressure thus
increases to zero under 6.5m of fill. The predicted pore pressures at the end of construction
are reasonably consistent with the observations of Bridle eta! (1985). Farrar (1978) and
Crabb et a! (1987) show pore pressures measured some time after the end of construction
which are also consistent with the predictions made. Swelling to zero pore pressure, as might
occur near the surface, gives an effective stress change of l2OkPa, and thus a volume
change (from Figure 7.13) of 5%, and an increase of water content of about 3.5%.

There is little data from which the increase in volume may be validated. Swelling of partly
saturated fills is not normally reported in terms of eftective stress, due to the problems of
measuring the suctions which are present in the sample. Chaudhury (1955) reports volume
changes when samples of London Clay, compacted in the laboratory and loaded in the
oedometer 'dry', were flooded with water. The subsequent swelling was measured. It is
probably more than would occur in a comparable field-compacted fill. The effective stress
after swelling is known but not that before swelling. However, a comparison with the
assumptions of the present analyss may be made, as follows

171
 volume change on soaking
swelling from a 6kPa 1 O7kPa

assumption in FE analysis 5.0% 0.6%

compacted at opt. -4% 13.5% 3.6%
compacted at optimum 12.0% 2.4%

Thus, the amount of swelling adopted in the analysis probably underestimates swelling both
near the surface, which influences the shallow slips, and at depth.

7.4.2.6 Initial Stresses due to Compaction

The stress history of an embankment fill during construction involves loading only and is
thus equivalent to normal consolidation. However, the effect of compaction is to induce
higher lateral stresses while the fill is still at shallow depth. Some field observations from
rather wet fills are shown in Figure 7.14. It can be seen that honzontal stresses do not
become less than vertical ones until a significant thickness of clay has been placed. -

Since the fill is modelled in the finite element analyses as a saturated soil, the lateral
pressure developed under undrained one-dimensional loading is nearly equal to the increase
in vertical stress. However, it is of interest to examine the actual increase within the
embankment modelled. Figure 7.15 shows the stress paths for two points within a lOm
high embankment built on an undrained clay foundation. The points are at the back of a
potential slip surface. There is sufficient lateral yield for the horizontal stresses to reduce
to realistic values by the end of construction. Thus, the total amount of swelling, which is
likely to affect progressive failure, is reasonable.

7.4.2.7 Coefficient of Permeability

The permeabdity of plastic clay fills is low. Vaughan et a! (1978) give values for the
coefficient of swelling, c, back calculated from field observations of pore pressure in the
range 0.5-2.0m2 /yr. A value of the coefficient of permeability k=1.510 10m/s was
adopted for the present analyses. With the volume change in swe ling shown in Figure
7.13, this gives c3 1.5m2 /yr. No variation with depth was assumed in the analyses.

172
7.5 HYDRAULIC BOUNDARY CONDITIONS AND SURFACE EFFECTS

The ground surface seepage boundary condition in fine grained soils is controlled by
climate. The boundary pore water pressure is zero when there is a supply of free water
and the surface is 'wet'. When the surface is 'dry' and there is outflow due to evapo-
transpiration, the surface pore water pressure is a suction, which will vary. In the U.K. the
surface pore Water pressure is zero for much of the winter, when evaporation is low, and
is a variable Suction during the summer, when evaporation is high. The depth to which
seasonal variation penetrates is some 3m.

The U.K. has a 'wet' climate with annual infiltration in excess of evapo-transpiration. The
net effect in clay soils (where water available for infiltration exceeds permeability) is of
recharge, with an apparent 'water table' typically about 1 m below the surface. At this
depth seasonal variations still occur. However, an average condition may be identified
readily from piezometric measurements below the depth of variation, where pressures are
positive. This average pressure controls swelling at depth. Thus, swelling in response to
the surface boundary condition may be considered in two parts: (i) a monotonic change at
depth in response to the average boundary condition, and (ii) a surface zone of cyclic pore
pressure change (mainly in the negative range) where the magnitude of the changes
decreases with depth, and the period of variation increases with depth.

Observations of the pore water pressures within motorway embankments have been
presented by Farrar (1978), Anderson and Kneale (1980a,b) and Crabb eta! (1987). These
generally indicate that equilibrium pore water pressures have not been reached, and that
significant suctions persist at depth. Transient pore water pressures may become quite
high near the slope surface during severe winter rainfall.

The average surface boundary pore water pressures are difficult to evaluate from these
observations. This is more easily done from observations in much older slopes in which
near hydrostatic pore water pressures exist at depth, as observations at depth, which do
not fluctuate, can be extrapolated to give the average surface pore water pressure. Typical
observations compiled by Walbancke (1976) are summarised in Figure 7.16. Figure 7.16.a
is derived mainly from slopes at or near seepage equilibrium. Maximum pore water
pressures in the top 2m are very high, and are equivalent to flow parallel to the slope.
Below about 3m the pressures are equivalent to hydrostatic with en average surface
suction of 1 m of water. This is equivalent to the assumption made in most of the present
analyses. These observations are indicative of a cracked zone typically about 2m deep, in
which behaviour is unlike that at depth. There is evidence of a small increase in average
suction as slopes became steeper, and a small decrease as the grass cover becomes

173
rougher (see Figure 7.1 6.b). However, the differences are not signifcant. No differences
were observed between cut and fill slopes.

The effect of vegetation is to increase evapo-transpiration. By inference, average surface

boundary pore water pressure must be reduced. Grass increases surface pore water
suctions as compared to bare earth. Trees have a substantial effect. Problems caused by
shrinkage of clays induced by trees, and by swelling following tree removal, are well
known. Tree felling causes slope instability. Root growth acts as reinforcement to the
layer of superficial soil which is vulnerable to shallow movement. Yim eta! (1988) discuss
this effect in detail and conclude that the increase in shear strength in the upper 2m of a
slope soil profile due to the effect of tree roots may be in the range 1-l5kPa. The growth
of trees and shrubs may well increase the depth of cracking. Greenwood et a! (1985)
consider that tree planting may temporarily reduce stability. For vegetation to be an
effective means of stabilisation, it is necessary for the effect of root reinforcement to
extend below the depth of cracking and high transient pore water pressure. Apparently,
controlled growth of trees or shrubs may be an effective and cheap way of maintaining
the stability of clay slopes, provided they are established at an early stage of swelling
before stability is threatened.

Conditions within the surface zone subjected to seasonal changes in pore water suction
are complex. Cracks can be seen in most in-situ plastic clays and clay fils in dry weather.
Permeability due to cracks will be increased during wetting, but the effect will diminished
as the clay swells and the cracks close. Permeability will not be affected much by dry
cracks during drying. These complex cyclic effects cannot be accurately modelled, and the
effect of a cracked surface layer has been reproduced in the present work by monotonic
swelling with an increased surface permeability. This may not produce the full surface
effects experienced in the field.

In the analyses presented here, a typical British surface hydraulic condition of a pore water
pressure of -lOkPa (equivalent to a 'ground water table' im below ground surface) has
generally been assumed, although the effects of a transient increase in this pressure to
zero, and a decrease to -2OkPa, as might be induced by vegetation, has been examined.

174
 content of low-friction piety clay minerals
Characteristic___________________________ __________________________
low high

Plasticity low high

Residual Strength high low

Brittleness low high

Permeability low very low

Swelling Modulus lower higher

Potential for Progressive Failure smaller greater

Potential for Surface Deterioration reduced increased

Table 7.1 Influence of clay content on potential for instability in clay slopes

Property In-situ Clay Compacted Fill Drained Foundation'

Bulk Unit Weight, v [kN/m3J 18.8 18.8 18.8

Peak Strength (Intact) c0' 1 4kPa, •' 20° ci,' -1 2kPa, •,' 20° ci,' - 5kPa, Ø,' = 30°

Discontinuity Strength c1' = OkPa, • - 20° no discontinuities no discontinuities

Bulk Peak Strength c' - 7kPa, gi' 20° as peak intact as peak intact

Residual Strength c,' - 2kPa, •,' - 100 c,' - 2kPa, •,' - 130 not brittle,
c,'= 2kPa, Q,'-13° as peak

Plastic Stran at Peak e=O-5% as in-situ clay -

Plastic Strain at Residual ,,j=20% =50% -

Poisson's Ratio, u 0.20 0.30 0.20

Young's Modulus, E (kPa) 2,500ip' +100)/100 1 ,000(p' + 100)/100 2,500(p' +100)/100

(mean eff. stress p' in kPa) (mm. 4,000) (mm. 2,000) (mm. 4,000)

Angle of Dilatation, , 1 1 0 or 10 0 0

Permeability, k lm/sJ varies, see F,g.7.10 1.5 x 10 always drained

Coeff. Earth Pres. at Rest, l( varies not relevant 1.0 (as foundation)

'Also granular fiN for embankment repair

Table 7.2 Properties of London Clay assumed in the analyses

175
 páezometric height -- --
PQi1 01 WQIQr

T iivt weight ol clay

aveoge value of r, orourid
slip cur face

0 depth foci or

Figure 7.1 (a) Definition of the pore pressure ratio, r

(after Skempton, 1977)

- intermediate term 1 back—calculated

long term J from failures
04l

/ : r.
, _______

04 measured
pore pressures
/0 (stable slopes I
I I 1
//
/20 4 60 100 120 160
Age of cutting years

'VT
Figure 7.1 (b) Variation of the average pore pressure ratio, r,
with time in slopes excavated in brown London Clay
(after Chandler, 1 984a)

176


0 p

Figure 7.2 Variation of the angle of shearing resistance, ', and the cohesion intercept, c',
with the deviatoric plastic strain invariant, E0 (after Potts et a!, 1990)

__ "o.,,
Ioalc I
30 \_ ' ISIR I
'

\
20
\• \
'
\ •.-
10 a •

0
0 20 40 60
Plasticity Index, l %

Figure 1,3 Values of ,' and 0,' for some British clays as a function of plasticity index, l,
(after Vaughan. 1990)

177
 ISO a1 a, vertIcal
265mm dli. simple,
2 inciine

kPa 0 vertIcal
150mm dla. samples
-
100
0
• 71mm dli. samples

X 35mm dli. samples

PosslDis Intact strength XX.
c.2OkPii'2V -
Avenge st-rength. 265mm dl. samples
50 - 7kPsi s 20

Possible discontinuity strength

- 0, a - 20

o 50 100 150 200 250 300 350

C+ 3
kPa
2

Figure 7.4 Peak strength measured in brown London Clay by undrained triaxial tests with
pore pressure measurement on samples of various sizes (Sandroni. 1977); fissure
spacing 20-50mm, l, =35-60%

Shear strength,n, kPa

I
H
• Back analysed slides
IChandler, 1984 ; Bromhead, 1978] ]
x Ring shear test
.4
(LupIni, 1981]
I H

50 '

I Stress nge
In analyses
Final assumption
2kPa

InItial assumption
c 2kPa
••. •

J _-
—' K
Ring shear
C 0
I
—
y. .'JI
I—

0 100 200
Normal effective stress, a', kPz

Figure 7.5 Residual strength of London Clay as measured in the ring shear apparatus (Lupini,
198 1) and back-calculated from slides (Bromhead, 1978; Chandler, 1 984a), and as
assumed in the analysis

178


100
C l G3 Fill

kPa
'-S -

I
10 20 3 0€/O 150 250
I I t I I .
200
o
- +
° kPa
-v
[al

- 0 c.•\__.___

2 - Fill

kPa 50
In-situ -

0 I I-
________
I I
0 10 20 30 150 200
50
+ a'
kPa
2
u kPa
- [ b]
/
0- 1 ' I -.
0 30
10 20 CA%

Figure 7.6.1 Typical triaxial compression tests predicted by the numerical model: (a) Drained
tests and (b) Undrained tests

179
 - 03

2 Predictions
kPa

50

,1
Ii
B

/
Unconsolidated-undrained trlaxiat tests
.
(265mm dia. sampiesj Sandroni [1977] [b]

0.1

0 5 10 15 100 150
C +
CA kPa
2
400 U
I-
0
)E SET
! 350
0
[c]
-1
I- - -'
0 5 10 15

Figure 7.6.ii Predictions for undrained triaxial tests on in-situ clay compared with test results

0-
-
1)

I
( Ii Numerical model -...
Average test curve
Triaxial compression - lOOm dia. samples
Field compacted Upper Lias clay

o _!
0 10 20 30
A%

Figure 76.iii Predictions for an undrained triaxial test on fill compared with normalised test
result for field compacted upper Lies Clay; undrained strength S = 1 5OkPa

180


a', l37kPa

47
kPa

50 I -
Fill

In-situ clay
0
o 20 Shear straIn 60 A/H %
I- • • I I

o 0.1 0.2 0.3

Displacement accross a 0.5m thick layer
(½ the thickness of an element] m

Figure 7.7 Stress-strain and stress-displacement relationships of in-situ clay and fill in simple
shear

0 100 200 300 C•,, + O'

1
o4
-S.-. F) 2
Uaf K5 = 1.0
kPa
50 K0 = 1.5
a,- (
Strain-softening
2 K, 2.0
100 [a]
-5-. Swelling in the field
kPa -S
(from water content/
10 10 depth profiles]

AV% 1%
\ Laboratory oedometer tests
V (Apted, 1977]

5 Predictions
5
K5 1.0
'.

[C]

o- • I -Th _•__I 0 1 •

0 200 300 0 100 200

joo
vertical effective stress, Ør', kPa
2

Figure 78 Predictions for one-dimensional swelling of in-situ clay: (a) Stress paths,
(b) Volumetric strain, e = Aviv, as a function of average effective stress, (0,,' + (rh')!2,

and (ci Volumetric strain, s = Av v, as a function of vertical effective stress, 0,,';

predictions compared with laboratory and field evidence

181


Vertical stress, NB K 1.0 at ground surface

Pore pressure, u kPa
Undrained strength,; kPa
-100 0 100 200 300 0

La] 5\ \'. Sample

Plane strain
- sample
10
Depthm \\
15 %Tr'ax'a',100
Vertical stress, a K0 1.0 I( 2.0
Pore pressure, u kPa Undrained strength,; kPa
-100 0 _100 200 300 0
.4150 100
z/i ... Triaxlal
Ib] sample
Plane strain
ample

10-
Depth m
15\,

K01-° K02.0

Figure 7.9 Undrained strength profiles from the assumptions made in the analyses: (a)

Profiles for hydrostatic ground water with lOkPa surface suction, the state before
excavation (the profiles vary with )(), and (b) the stronger profiles for road

embankment analyses with the clay foundation ( K0 =2.O was used)

Coefficient of permeability , k rn/s

10.12 1011 10-10 10'

0 1 ' /1

Brecciated Upper Lias clay, Empingham. . .. , II

Laboratory and in-situ tests. / .
(Chandler, 1974] • $ 0 •
I •• /
London Clay Wraysbury.
Laboratory test on large sample. • •• •.
(Garga, 1970] /
10 ii

London Clay Wraysbury. '

In-situ tests [Garga, 1970]
London Clay, UniversIty 01 Kent.
• / / Change In permeability with depth
•
/ / (AssumIng K - 104m/s at GLJ
.. S
(Vaughan et al, 1983]
________________ ________________ / /

20
C
Assumption made In analyses

. I
Depth m
Figure 7.10 Variation in permeability with depth assumed In the analyses of cuttings

compared with field and laboratory data

182


- °3kPa
200 - = 2 Upper Lias clay 100mm
0

Indicates l < 25% 0

100 -
Kellaways Clay 100mm J Weald Clay 100mm
Kellaways Clay 38mm) Mean strength
[higher plasticity Samples]
C' = l2kPa, i' = 21°
- Ci +
..
U
Wadhurst CIaylOOmm

0
0 100 200 300 400 500

Figure 7.11 Peak strengths for various field-compacted fills from plastic clays (after Vaughan

eta!, 1978)

°1°3kPa
50 2

V-O
O'/ 'Ali results [s5'2OkPa]
c' = 6.7kPa, ,' = 26.7°

Tests at low stress tailed by Increasing the pore pressure.

0
- I I

0 50 100

l% ,0
A b + ° kPa
C
Gault (Cambsj 42 23 1.93 0.76 SI =
2
. Gault IDunton] 42 23 2.25 0.69
Gault (Nepicar] 43 23 1.88 0.73
x Kimmeridge 27 22 1.81 0.71
+ London 27 25 3.71 0.58
0 Oxford 28 25 2.76 0.63
0 Reading 32 19 2.84 0.58
V Weald 23 24 2.23 0.69

Figure 7.12 Peak strengths measured in various British field-compacted clays (Perry, 1991;

Crabb and Atkinson, 1991); consolidated drained tests on 38mm triaxial specimens

183
 100

C
0
4)

0
a

>0
0 100 200
-100
Vertical effective stress o' kPa

C
+
(-I, kPa
C)
2

0 100 200

Figure 713 Swelling of the fill as modelled in the analyses

184


200
(a)
Vt
0 Balderhead Dam core - Low plasticity till
placed at about optimum water content -
embedded pressure cell [Kennard,
Vt Penman & Vaughan, 1967]
Vt
Vt
I-
fbj
.: 100
Vt
0 Low plasticity very wet Norwegian clay
placed and compacted by dozer (Ostlid,
Vt iiaj
0
P4
I..
(c
0 Cow Green Dam core - low plasticity tilt -
placed about 2% wet of optimum - pressure
against rigid end of concrete dam (Vaughan
0 & Kennard, 1972]

0 100 200
Overburden pressure, a,, kPa

Figure 714 Some lateral pressures observed in wide compacted fills

200

Vt
//
B

0)
/.
Vt

100 /
0)

,pressure 1Dm
0 J
B

/
0 -'
0 100 200
Vertical stress a,, kPa

Figure 7.15 Lateral pressure developed in an embankment according to the finite element

analysis

185


Pore pressure, m ol water

.7 -S -5 -4 -3 -2 -1 o 1 2 3 4 5

\iI
o 0 0 0 0 0 I
o coo Flow parallel
2 to slope
co1p2.5
mm. max. o&*A £ ,f cotp 4.0
£
•
Natural slopes o • A *
Iwith head] I
.4 4 * 1 \ '
Cutting J
Embankment a
Cow Green (
C
twet upIand J
Ia] 6 Hydrostatic equilibrium -
j
I surface suction Im
Depth m 7 j Edgewarebury I
(probably not equilibrated]
Slope cotp

0 10 20
I_____

— —a
-
.-'- — o -
ODo
I
1 *°
- Increasing roughness
[b]
I .—

Natural slope o
Depth m Cutting
Embankment n
2

Figure 7.16 Superficial pore water pressures measured in slopes (after Walbancke, 1976):

(a) Maximum and minimum pore water pressures in slopes near long-term equilibrium,

(b) Estimated depth of zero pore water pressure line below grassed slopes as a
function of slope angle and surface roughness

186
 CHAPTER 8
ANALYSIS OF CUT SLOPES

8.1 INTRODUCTION

Table 8.1 summarises details of all the analyses performed for cut slopes. A typical finite
element mesh is shown in Figure 8.1. Meshes for other slope geometries had a similar
pattern.

The different geometries examined in the present investigation are summarised in Figure 8.2.
An axis of symmetry was assumed in the centre of the cutting, the width of which was
approximately that of a double-track railway, and thus the slopes are representative of the
railway slopes for which there is a well established history of sliding1.

Excavation of the 1 Om high slopes was simulated at a uniform rate over 10 increments, in
an overall time of 0.25 years. Each of the upper three rows of elements was excavated in two
stages. There was no significant swelling during excavation. Slopes of other heights were
excavated at comparable rates.

After excavation had been performed, a specified surface pore water suction was applied and
swelling (consolidation) allowed according to the new hydraulic boundary conditions. The first
time step after excavation was ót=0.25yrs, followed by ât=O.5yrs. The constanttime step
of at =1 .Oyr was adopted thereafter until collapse was approached.

Monitoring of convergence is difficult for analyses involving strain-softening. A procedure

similar to that employed by Potts et a! (1 990) was used here. Convergence norms for both
iterative displacements and nodal pore pressures (1 % of the norm of the incremental values,
see Section 2.3) were always satisfied. However, this procedure was inadequate when
considering iterative loads, and thus the maximum residual stress at any integration point was
checked at the end of each increment. A typically observed value was 2kPa. As collapse was
approached, time steps were decreased (At =0.25, 0.05, 0.01, O.0025yrs) and the number

1 The analyses show that the assumed width of the cutting should have no significant
influence on collapse of the slope.

187
of iterations increased (typically 200), reducing the maximum residual stress at any
integration point to less than 0.1 kPa. 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 solution for a small time step (t=0.0025yrs) and a large number of
iterations (typically 400). Instability was indicated when the deformations increased according
to the number of iterations without an improvement in convergence.

The primary set of analyses were performed for a slope of 3:1, 1Dm deep. The effect of
varying ,', ç and K0, and the effect of increasing the surface hydraulic boundary pore water
suction from 1 OkPa to 2OkPa was examined. This latter change would be equivalent to a
modest increase in evapo-transpiration due to a change in vegetation.

The lower boundary to the mesh was assumed to be impermeable in most of the analyses.
This is equivalent to a deep bed of clay without underdrainage. An impermeable boundary
gives slow pore pressure equilibration. In reality drainage boundary conditions may be variable
and much more complex, and they may involve a drained lower boundary. Pore pressures
behind the slope may then be lower, but pressures near the toe of the slope may be artesian
relative to the bottom of the cutting. Several analyses were made to investigate the influence
of base drainage.

Slopes of 4:1, lOm and 15m deep, and slopes of 3:1 and 2.5:1, 6m deep, were analyzed
with K0 =2.0. All except four of the analyses were performed without varying the soil
properties near the slope or applying cyclic boundary pore water pressures to the slope
surface. No shallow slips such as are observed in the field were reproduced in these analyses,
indicating that such slips probably involve surface effects. As a final step, analyses were
made of the 3:1 slope, lOm high, with K 0 =1 .0 and 2.0, and with the permeability of the
outer 2m deep layer of clay on the slope surface increased by approximately an order of
magnitude, to simulate an increase in permeability due to shrinkage cracks, and to examine
whether this caused shallow instability. As will be shown, shallow sliding is reproduced in
some of these analysis.

8.2 TYPICAL ANALYSIS

The results of a typical analysis with K0 =1.5, q,'=13° and q,=0 and a pore water pressure
condition at the exposed soil surface of -lOkPa throughout are shown in Figure 8.3. They are
presented for the three stages of analysis, namely: (a) the end of excavation, (bI 9 years after
excavation, when local failure along the base of the slope has become significant, and Ic)
14.5 years after excavation, just before the predicted collapse.

188
Figure 8.3.i shows: (a) the deformations predicted during excavation; maximum lateral
deformation of the crest of the slope due to excavation is about 150mm. which is reasonable,
but probably an overestimate because the pre-failure stiffness adopted in the analysis does
not include the high stiffness present at small strains; (b) the post-excavation deformations
after 9 years swelling; there is a maximum combined heave and outward movement at the
toe of about 200mm; (c) the post-excavation movements just prior to collapse; movements
at the toe are now about 400mm, and there is major lateral movement within the slope.

Figure 8.3.ii shows the incremental vectors of displacement at the three stages. These
vectors show the current mechanism of movement most clearly, and they indicate the
eventual collapse mechanism. Their absolute magnitude is of no significance. These show that
the outer part of the slip has developed after 9 years, and that the complete slip has
developed after 14.5 years.

Pore pressure development is shown as contours in Figure 8.3.iii, with some vertical profiles
on which both changes towards equilibrium and equilibrium pressures are shown. In the
middle of the slope, pore pressure changes due to excavation are almost constant with depth.
For this value of K 0, they change by an amount almost equal to the weight of soil removed
from above the slope (see equation (7.1)). At the crest of the slope the reduction is also
nearly linear. It varies from 2OkPa to 3OkPa 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 pressure is
significantly less than the weight of soil removed. The interaction between changes in total
stress and pore pressure is clear. After 9 years, the pore pressures at depth have reduced due
to undrained unloading as the slope deforms outwards releasing the locked-in lateral stresses,
rather than increased due to swelling.

Figure 8.3.iv shows the magnitude of the major principal effective stress, or,'; Figure 8.3.v
shows its direction, and Figure 8.3.vi shows the magnitude of the minor principal effective
stress, 0 3'. The magnitude of the stresses within the slipping mass decrease substantially as
the slip develops, and an abrupt change develops across the rupture surface. The value of K0

at the ground surface is taken to be one increasing linearly to K0 =1 5 at 1 m depth, where

the zero pore pressure line is assumed (see Section 7.4.1.5). Thus, near ground level the
horizontal stress is initially close to the vertical one. A small reduction in lateral stress causes
principal stress rotation with the vertical stress becoming 01'. Perhaps surprisingly, this effect
extends some 50m from the crest of the excavation. The principal stresses within the slope
also show significant rotation as the rupture surface develops, with the major principal stress
tending to become vertical as the base shear develops and strain-softens, and the lateral
pressures decrease. A zone of tension (in terms of effective stress) on near vertical surfaces

189
develops (see Figure 8.3.v and vi), about im deep at ground level, to about 35m from the
slope crest. Both minor principal effective stress and pore water pressure are zero about 1 m
down, which is the theoretical maximum depth of a tension crack according to the cohesion
intercept adopted. There is a small tensile force exerted on the potential sliding mass, of the
order of 1 OkN per m run, which would not operate if the soil were unable to take tension2.

The major principal stress before excavation is horizontal, and, given the bulk unit weight of
soil y=18.8kN/m3 and water y9.81kN/m 3, depends on the value of K 0. Below im it is
a,' =11 8.8 + 8.99(z-1 )1K0 in [kPaJ, where z is depth in Imi. The values of a' at depth on the
right boundary in Figure 8.3.iv, which are horizontal stresses, are equal to these after
excavation. Thus, there is no change in stress on the boundary at depth due to excavation,
and the position of the boundary does not influence the slope.

Figure 8.3.vii shows the shear stress level contours, S (see equation (5.4) and Figure 5.4),
representing the proportion of the current strength mobilised. The only local failure which has
occurred due to excavation is below the bottom of the cutting. This has no bearing on
subsequent events. The rupture zone propagates into the slope from the toe as swelling
proceeds, and virtually all the clay in the slip is near failure when collapse occurs.

Figure 8.3.viii shows the velocity characteristics (see Section 7.3.2) as 'crosses' indicating
the two conjugate potential rupture planes. These are at right angles because a zero angle of
dilation has been assumed. They are generated at Gauss integration points of elements (see
Section 2.2.6) where local failure is occurring currently. (b) shows a rather general area of
failure near the inner end of the extending rupture surface. (c) shows that this area has
become much narrower by the time collapse occurs; namely, as the strength has reduced with
strain-softening, failure is concentrated in a narrow zone, and the clay on either side has been
unloaded and is no longer failing. At collapse there is a more general active failure zone at the
back of the slip, where the final rupture surface has not started to strain-soften (see also
Figure 8.3.ix.b).

Figure 8.3.ix shows contours of the deviatoric plastic strain invariant, e (see equation (7.6)),
for the second and third stage (plastic strains are nearly zero after excavation). The 5% and
20% contours represent the start and end of strain-softening. The base shear rapidly drops
from peak to residual as the rupture zone develops.

Figure 8.3.x shows the long-term equilibrium pore pressures in the slope.

2 may be noted that in other analyses in which lower values of K 0 are assumed, the
rupture surface near the ground surface forms as a post-collapse event, when tension will
have no influence on stability.

190
These figures show that, just before collapse, when the slope is still in static equilibrium, all
the eventual rupture surface has reached peak strength in terms of effective stress, and much
of it has started to strain-soften. Thus, the average shear strength in terms of effective stress
reduces with displacement. The slope can only remain in equilibrium because the pore
pressures are also reducing due to undrained unloading, and the total shearing resistance is
not decreasing.

Figure 8.3.xi shows the volumetric strains which develop due to swelling. Strains after
excavation are very small. The strains within the slip are quite large, and equivalent to a
change in water content at the toe of the slope before collapse of the order of 1.5%. Farrar
(1990) reported increases in water content in a slipped area of the order of 10% at the
ground surface, and 5% at a depth of I .5m. Thus, perhaps, the analysis underestimates the
volumetric strains at low stresses due to swelling. The analysis predicts an abrupt change in
volumetric Strains at the rupture surface. Although the changes in water content predicted
are significant, they would be rather difficult to measure and use as a way of monitoring
incipient collapse.

Figure 8.3.vii shows that a complete rupture surface has developed at collapse, with the
stresses lined up to allow rupture on a single surface. 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 average operational strength along the eventual rupture
surface at collapse is given by /' = 0.328, which is equivalent to c' =0, 0' = 18.2° '. The
average pore pressure ratio on the final rupture surface has increased from r = O.044 after
excavation to , = 0.238 at collapse. The long-term equilibrium pore pressure is equivalent to
r = 0.375. Thus, on the rupture surface, the pore pressure has changed 60% of the way
towards final equilibrium.

Figure 8.4.b shows the horizontal displacement at mid-height of the slope as it develops with
time. The deformation is quite small as the base shear develops. It then increases more rapidly
(note the discontinuity on the displacement/time plot in Figure 8.4.b) as the back of the s'ope
swells. Final collapse occurs abruptly, with no warning from accelerating deformation. The
average pore pressure on the rupture surface changes quite rapidly as the base shear
develops. There is only a relatively small increase in pore pressure in the five years preceding

3This is not necessarily the strength that would be obtained by limit equilibrium analysis
of the slope with the pore pressures as predicted by the analysis. The location of the rupture
surface predicted by the finite element analysis depends on K0. Limit equilibrium analysis of
this surface, with internal force assumptions compatible with the finite element analysis, will
produce a similar average strength. However, this may not be the 'critical' surface found by
search using limit equilibrium analysis. The average strength on the critical surface,
determined by limit equilibrium analysis, will then be higher.

191
collapse. Undrained decreases in pore pressure due to decreasing average total stress are
compensating for increases due to swelling.

The complex behaviour is shown further in Figure 8.5.b, where the stress paths predicted for
two points on the eventual rupture surface are plotted. Point 'b' is on the horizontal sliding
surface formed at an early stage. Point 'e' is on the rupture surface at the back of the slip,
which forms subsequently. After excavation there is a small Suction at 'b', which is quite
close to failure. There is a positive pore pressure at 'e', which is some way from failure. After
3 years point 'b' reaches peak strength. Point 'a' has changed little. After 7 years the strain
at 'b' has become sufficient for the strength to start dropping to residual. Little change has
occurred at 'e'. After 9 years the strength at 'b' has reached residual. The strength at 'e'
reaches peak after 12 years. The strength at both points then drops due to the increasing
pore pressure until collapse occurs at 14.5 years; at 'b' the strength is at residual and at 'e'
it has remained at peak.

The predicted deformations are consistent with the limited precedents for field observation
of London Clay slopes. The best recorded data is of the Kerisat Green retaining wall
(Skempton, 1977), which moved nearly 0.5m before failing some 29 years after its
construction. The movements were about 100mm after 26 years.

8.3 PARAMETRIC STUDIES

Table 81 summarises the results of the twelve analyses performed on the 3:1 slope, lOm
high with varying residual strength, angle of dilation and K0.

8.3.1 Effect of Residual Strength

Analyses [1], [4] and [5] utitised a residual strength of c,' = 2kPa, 4r' = 10 0, with an angle of
dilation 'p = 10° at peak, dropping to zero at residual. These values were deduced directly from
laboratory tests. For K 0 = 1 .0, 1.5 and 2.0 the average pore pressure ratios at collapse were
r=0.1 57, 0.108 and 0.091 respectively, and collapse occurred after 28.4, 11.4 and 12.6
years. Pore pressures at collapse were slightly lower than indicated by field measurements
(see Figure 7.1.b). As reviewed in Section 7.4.1.2, c,'=2kPa, p'=10° is typical of results
from laboratory tests on London Clay, but back analysis of slips suggests higher operational
vaJues in the field. A residual strength of C,' 2kPa, ,' 13° was adopted for analyses [2],
[6) and [7]. For K0 values of 1.0, 1.5 and 2.0, the values of at collapse, shown in Figure
8.6, were 0.127, 0.156 and 0.264, after delays of 33.4, 15.9 and 34.6 years. The higher
residual strength was used thereafter.

192
8.32 Effect of Angle of Dilation

The pore pressures predicted in the analyses in which iu1O° was assumed showed
significant fluctuations with time. This may be seen in Figure 8.7 for analyses [2] and [7],
which may be compared with analyses 18] and 110], in which 4u = 0 was assumed. There are
no major differences between the two pairs of analyses in other respects, except that there
was a major drop in pore pressure in analysis (7] just prior to collapse. The analyses with
çu 100 show slightly more progressive failure: R=0.743 as against R0.623 with K0 = 1.5,
and R=0.601 as against R=0.500 with K 0 2.0 (see Table 8.1). Examination of the results
from these and other analyses showed the following:

(I) The shear surface developing horizontally from the toe caused a reduction in horizontal
stress in the undrained clay at the back of the slip, and a consequent undrained drop in pore
pressure. This effect operated irrespective of the value of qi adopted.

(ii) Additional drops in undrained pore pressure occurred in the analyses with yi10°, due to
the effects of dilation in the relatively thick 'shear zone' developed in the analyses. The
formulation of the analysis gives a minimum thickness for such a zone of half the thickness
of an element. In these analyses the elements were typically im thick. The influence of the
angle of dilation on the generation of undrained pore pressure and undrained strength are
shown in Figure 8.8. This effect caused the large drop in pore pressure just before collapse
in analysis [71, which occurred in the shear zone developed in the back of the slip. This can
be seen in Figure 8.9, in which the pore pressures just prior to collapse from analysis [7] with
p= 100 are shown. Most of the rupture surface is at residual, where dilation is zero. At the
back of the slip it is part way from peak to residual. Since strength is lost with further
movement, this state would be post-collapse, except that the drop in pore pressure is
compensating for the loss of strength. The effect may be seen in Figure 8.7, where the
deformations increase significantly during pore pressure fluctuations, even though average
pore pressure may be decreasing and safety factor apparently increasing. The rupture surface
for analysis [10] with qi0 is shown in Figure 8.9 for comparison. Here the back of the slip
has reached peak, but has not started to strain-soften.

(iii) In reality, dilation would be restricted to the clay adjacent to a few thin rupture surfaces,
and equilibration of the pore water pressure would occur much more rapidly than occurred
in the analyses in which qi = 1 00. Thus, if it is assumed that dilation occurs post-peak and only
on thin shear surfaces, which is supported by laboratory observations by Sandroni (1977),
the analyses in which çu 0 are more likely to give a better representation of undrained affects
in clay in which discontinuities develop post-peak.

193
An angle of dilation of zero was adopted for the remaining analyses. A consequence of
adopting q'=O is shown in Figure 8.9, where the two rupture surfaces from analyses [7] (with
u-10°) and [10] (with v=0) are shown. The same horizontal surface is obtained from both
analyses, but the inclined length at the back of the sip is different. The rupture surfaces make
angles of ±(45°-/2) to the direction of the major principal stress, o,'. Thus different rupture
surfaces, or different amounts of stress rotation before rupture, should be shown by the two
analyses. The common horizontal length of rupture surface involves different amounts of
stress rotation. The ground surface is a principal plane, and it can be seen that the two
rupture planes make the appropriate angles to it of 45° and 50°. The rupture surface predicted
by analysis [10] is thus deeper than that predzcted from analysis (7], although the
compression strength of the soil is the same. That predicted by analysis (7] is likely to be
more like that occurring in the field, although the influence of this on predictions of collapse
should be small. It may be noted that the problem can only be avoided if many smaller
elements are used in the analysis. Slip elements cannot be used as the rupture surface must
then be known in advance.

8.3.3 Effect of Coefficient of Earth Pressure at Rest

Analyses [91, [111, [10], (1 2], [8] and (16] were performed with the higher residual strength,
with zero angle of dilation, and with values of K 0 increasing from 1 .0 to the limiting value of
K0 =2.15(K). The variation of average pore pressure ratio and mid-height horizontal
displacement with time for these analyses are plotted in Figure 8.4. Average pore pressure
ratio at collapse are plotted against time to collapse in Figure 8.6. It may be noted that the
steady state pore pressures ratios plotted in Figure 8.4 are from the same seepage solution,
but differ because the rupture surfaces are different. The pore pressures predicted after
excavation are summarised in Figure 8.10. The average shear stress at collapse along the
rupture surface is plotted against average normal effective stress in Figure 8.11. Note that,
since ç=0, the strengths at peak and residual are given by equation (7.10). The effect of K0

on the residual factor R at collapse is shown in Figure 8.12. The shapes of the rupture
surfaces are summarised in Figure 8.13, where they are compared with field examples, and
they are shown in more detail in Figure 8.14. Stress paths for particular points on the rupture
surfaces are shown in Figure 8.5.

Figure 8.4 shows that time to collapse is strongly dependent on K 0,, and reaches a minimum
of about 13 years when K 0 is about 1 .25. Maximum pore pressure ratios preceding collapse
are similar, ranging from =0.225 to i=0.275. However, two types of behaviour may be
seen: (i) pore pressures reaching close to their maximum values soon after excavation when
K0 < 1 .5, and (ii) with K0 > 1.5, there is a rapid equilibration of pore pressure over about 5
years, followed by a much slower increase until collapse. The slow increase in pore pressure

194
is accompanied by an initially increasing rate of displacement, which then slows down. Rates
of displacement only accelerate again a few months prior to collapse. Thus, there is little
warning of collapse from observations of either pore pressure or movement. Collapse occurs
when pore pressures are significantly lower than final equilibrium values.

Figures 8.t3 and 8.14 show the shapes of the rupture surfaces predicted. These are quite
strongly influenced by the value of K 0. They become deeper as l( increases from 1.0 to 1 .75.
As K0 increases further the back of the slip moves back towards the slope. Some rupture
surfaces observed in the field are compared with the predictions in Figure 8.1 3.b, where the
field surfaces have been normalised to the same slope and slope height as the analyses. The
predicted and observed rupture surfaces are in broad agreement.

Figure 8.1 0.b shows that the decrease in pore pressure due to rapid undrained excavation
increases with K0, reflecting the change in average total stress. However, the value of , on
the eventual rupture surface (see Figure 8.10.a) increases as K 0 increases from 1.0 to 1.5.
It then reduces as K 0 increases to 2.15. The reason may be seen in Figure 8.3.iii. Pore
pressure after excavation varies almost linearly with depth, and r increases with depth. The
rupture surfaces become significantly deeper as K0 increase from 1 .0 to 1 .5, and thus on
the rupture surface increases with K0 even though the pore pressure at a given depth is
decreasing. As K0 changes from 1.5 to 2.15 the rupture surface stays in approximathly the
same place, and decreases as the pore pressure decreases. Figure 8.10.b shows that the
average decrease in pore pressure due to excavation is approximately equal to the reduction
predicted from the weight of soil removed when K0 =1.O (see equation (7.1)). It is greater
when K0 is higher.

The pore pressures after excavation summarised in Figure 8.10 are dependent on the
assumptions explicit in the formulation of the soil model. Prior to peak an isotropic elastic
material is assumed, and thus pore pressure changes, u, are due to changes in mean total
stress, Ap, only. Most real stiff clays are anisotropic, and pore pressure changes are then a
function of both changes in mean total stress and changes in the magnitude and direction of
the shear stress (see e.g. Symes eta!, 1984). Figure 8.3.v shows that there is a 45° rotation
of principal stresses over most of the eventual rupture surface. Thus, pore pressure
predictions would change if anisotropy were assumed.

It may be noted that the variation of pore pressure after excavation with K0 is quite significant
and, subject to the effect of anisotropy, observation of pore pressure before and after
excavation may be a good indicator of initial K0.

Figure 8.6 shows predicted pore pressure ratios r., on the rupture surface at collapse, which

195
are typically higher than those estimated from field measurements. This may be due in part
to the manner in which , was estimated from the field measurements'. Analyses (9J and
(11), which give shallower rupture surfaces, give results for at collapse which are more
consistent with past field observations. In general, considering the difficulties in estimating
the coefficient of 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 representative of stiff clay) are assumed. As the
present analyses show, time to collapse varies significantly with factors other than
permeability, slope angle and slope height. The value of K,, angle of dilation, residual angle
of shearing resistance, rate of softening, surface and base hydraulic boundary conditions also
affect times to collapse.

Figures 8.11 and 8.12 show that the operational strengths and the amounts of progressive
failure predicted from all the analyses are essentially similar. However, the influence of
progressive failure changes when K 0 exceeds 1.75, as shown by the values of A in Figure
8.12; the influence increases with K 0 initially, then it increases again. The cause of this
change is shown in Figure 8.14, in which the various rupture surfaces predicted from the
analyses are shown in more detail. The length of the initial contained rupture surface which
develops horizontally increases with K,. At the higher values of K0 it extends beyond the crest
of the slope. The behaviour is like that observed by Burland et a! (1 979) in the Oxford Clay.
The back of the slip subsequently develops 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 this occurs, and the amount of progressive failure is therefore reduced.

The nature of the rupture surfaces is also shown in Figure 8.14. In all the cases shown only
a short length is at a strength between peak and residual. When K 0 is low then part of the
back of the slip forms as a post-collapse event. The operational strengths deduced from field
studies by Chandler and Skempton (1974) are also shown in Figure 8.11. The operational
strengths deduced from the finite element 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. These surfaces are
produced by progressive failure, and are not necessarily those on which TIO' is a maximum.
Normal effective stresses are higher and strengths lower than those on the critical surfaces
according to limit equilibrium, on which r/a' is a maximum. With K 0 1.0 the rupture surface

4Three-four piezometers were typically installed into the slopes and mean values
representative for the whole slope (or part of it) were deduced. in some cases the average
value of along a hypothetical (or observed) slip surface was calculated (see Skempton,
1977; Chandler. 1 984a).

196


predicted by the finite element analysis is similar to the critical rupture surface. In this case,
as can be seen in Figure 8.11, both the strength and the normal effective stress are similar
to those of Chandler and Skempton (1974).

The stress paths in the two parts of the slip are illustrated for K0 1 .0 and 2.0 in Figure 8.5.c
and d, where they may be compared with those for the typical analysis (Figure 8.5.b). The
mechanisms are similar for all three values of K0, with the horizontal shear at the base
reaching residual well before collapse, and then slowly loosing strength as the pore pressure
increases and the effective normal stress and the strength reduces. The clay in the back of
the slip does not strain sufficiently to strain-soften, although there is a loss of strength due
to swelling and decrease in normal effective stress, which becomes more significant as K0
increases.

8.3.4 Effect of Rate of Softening

The influence of progressive failure must increase with the abruptness of the drop from peak
to residual strength. The magnitude of this effect was examined in analyses [31 and (301 (see
Table 8.1 and Figure 8.7), in which the drop from peak to residual was linear from 5% to
35% of deviatoric plastic strain, ED. These analyses may be compared with analyses [2] and
[101:

Analysis [10) [301 [21 [31

___________ 1.5 1.5 2.0 2.0

D, 20 35 20 35
R 0.623 0.430 0.601 0.598
0.238 0.240 0.264 0.276

t,. Fyrs) 14.5 23.8 34.6 38.3

The amount of progressive failure is decreased and the pore pressure at collapse and the
time to collapse consequently increased. E' = 20% was assumed in all the other analyses.

8.3.5 Effect of Surface Boundary Suction

Analysis (13] was on a 3:1 slope, lOm high, with K0 =2.0 and a surface suction raised to
2OkPa, to examine the effect of an increase in suction such as might be produced by the
management of slope vegetation. The results may be compared directly with analysis (81.
Figure 8.15 shows how the average pore pressure ratio on the rupture surface and the
horizontal displacement at mid-height develop in the two analyses. Collapse with the

197
increased boundary suction occurs after 100 years. Analysis [13] was rerun without
strain-softening. No collapse occurred, and the changes in behaviour as the base shear
develops about 5 years after excavation were eliminated.

Figure 8.16 shows the pore pressures predicted just before collapse. The three vertical
profiles show that pore pressure is close to final equilibrium, but has not reached it. In all
the analyses the permeability profile remains constant below original ground level, and
permeability does not change with effective stress and swelling, as it should. To
investigate the influence of permeability distribution on predicted equilibrium pore water
pressure, a prediction was made with a uniform permeability. The results are shown on
the profiles in Figure 8.16. Differences are insignificant. This is Consistent with the
findings of Bromhead and Vaughan (1980), who showed that when flow is normal to the
direction in which permeability changes, as is the case in a slope without underdrainage,
the effect of a variable permeability should be small.

The nature of the collapse is shown in Figure 8.17; (a) shows the shape of the slide; (b)
shows the deviatoric plastic strains just prior to collapse; (c) shows the velocity
characteristics just prior to collapse. The rupture surface is marked on (c). The broken line
on (b) is the rupture surface predicted in analysis [8], with the same assumptions but a
surface suction of lOkPa. The rupture surface from analysis [131 is slightly deeper. The
average pore pressure ratio at collapse on the rupture surface of analysis [131 is
=0.243, compared with r = O.264 for analysis [81. The amount of progressive failure
increases to compensate, from R=0.500 for analysis [81.to R=0.693 for analysis [13],
which is a drop in average strength of some 12%. These differences are not of major
significance.

Although there is more progressive failure, the small increase in boundary suction (from
1 OkPa to 2OkPa) is effective in improving stability. Extrapolation indicates that a further
increase in boundary suction to 3OkPa would give an ultimate pore pressure equivalent to
=O.22 on the rupture surface, which would be just sufficient to avoid collapse with
equilibrium pore pressures.

Figure 8.17 shows that a deep-seated contained failure zone has propagated from the toe
of the slope towards and along the base of the clay, against the rigid boundary. This has
not been incorporated into the final rupture surface. A repeated analysis (not quoted here)
of the same slope with the rigid boundary lowered by 1 Om showed a similar mechanism at
about the same level. Thus, this feature was not caused by the rigid boundary. A similar
contained failure zone was seen in all the analyses with high values of K 0, and is discussed
further in the next section.

198
8.3.6 Effect of Slope Geometry

8.3.6.1 Flatter arid Deeper Slopes

Two analyses were made of a 4:1 slope with an undrained lower boundary, wlth a surface
suction of lOkPa and K 0 =2.0; analysis (141 for H10m and analysis (29] for H15m.
Figure 8.18 shows the development of the average pore pressure ratio and mid-height
horizontal displacement with time. The 1 Om high slope collapsed after 203 years, with
=0.396 and R=0.680 (compared with t s=31.6yrs, =0.264 and A =0.500 for analysis
(81 of the equivalent 3:1 slope). The pore pressures were close to final equilibrium, Li =

0.425 (see Figure 8.19). There was more progressive failure than in analysis [8]. This is
equivalent to a reduction in operational strength of 11%, compared to the 3:1 slope. The
conditions predicted just before collapse are shown in Figure 8.20.

The 15m high slope collapsed after 145 years, with r=0.361 (some way from final
equilibrium, •r=0.445) and with R=0.606. Thus, there was slightly less progressive
failure than in the lOm high slope. Pore pressures at collapse are shown in Figure 8.21,
and conditions just before collapse in Figure 8.22. The shorter time to collapse predicted

for the higher slope arises because the smaller amount of pore pressure equilibration
required for collapse compensates for the longer drainage paths.

Like the 3:1 slopes, the 4:1 slopes showed the rapid development of a horizontal shear
zone from the toe of the slope, accompanied by discontinuities in the plots of the average
pore pressure ratio and displacement against time (see Figure 8.18). Figures 8.20 and
8.22 show the slips in the two slopes. They are almost the same shape. Both show the
pronounced extension of the horizontal shear into the slope, leaving a 'blind' end which is
not incorporated into the final rupture surface, which was characteristic of the 3:1 slope
with K0 =2.0. A second deeper shear develops from the toe down towards the rigid base.
This shear is not incorporated in the final rupture surface. Analysis 114E) was performed
to investigate the influence of the rigid boundary on the shearing mechanism. The depth of
the clay was increased by lOm (see Figure 8.2). Figure 8.23 shows the rupture surface
predicted. It is virtually identical to that from analysis 1141. However, the additional yield
allowed by the deeper boundary reduces the extent to which both the upper and lower
horizontal surfaces penetrate into the slope. As a consequence the amount of progressive
failure is slightly reduced, and the time to collapse increased to 300 years. In the field an
increase in strength and a reduction in K 0 with depth are likely to inhibit the development
of deep-seated shears. Deformations between excavation and collapse of the two slopes
(see Figure 8.24) are large, although swelling of the floor of the cutting is only about 4%
of the thickness of the clay below.

199
The analyses of the 4:1 slope show that it behaves in a similar manner to a 3:1 slope.
They indicate that such a slope is nearly stable when 1 Om high, even in the very long-
term, but a 1 5m high slope fails with pore water pressures some way from predicted
equilibrium values. While predicted times to collapse are very long, such that the eventual
collapse may be considered rather academic, such slopes may be vulnerable if geologic
features such as sand or silt layers within the slope lead to more rapid swelling.

8.3.6.2 Steeper end lower Slopes

Analyses [15) and [171 were performed to examine the stability of steeper 2:1 and 2.5:1
slopes, lOm high, with l(=2.0, and the results may be compared with analysis [81 for the
3:1 slope. The development of pore pressure and deformation with time are shown in
Figure 8.25. The rupture surfaces are shown in Figure 8.26 (the 2:1 slope) and Figure
8.27 (the 2.5:1 slope). Both slopes fail in less than 10 years, with relatively low average
pore pressure ratios on their rupture surfaces of r=0.070 and 0.087, and with R=0.686
and 0.636. There was slightly more progressive failure than for the 3:1 slope, for which
R=O.500. These represent drops in operational strength of 13% and 9% respectively.

Analysis 118] was made of a 3:1 slope and analysis [191 of a 2.5:1 slope, both 6m high
with K0 =2.0 and a surface suction of lOkPa. The 3:1 slope had not failed when pore
pressures had reached equilibrium after 1 50 years. Figure 8.28.a shows the contours of
deviatoric plastic strain at this time and Figure 8.28.b shows the velocity characteristics.
Two lateral shear zones had developed, one just below the toe and a much deeper one.
The velocity characteristics show that a rupture zone has developed in the form of a slip,
but the deviatoric plastic strains show that strain-softening has not started in the back of
a slip. Examination of the velocity characteristics shows that they are not 'lined up' in the
horizontal portion of the potential rupture surface, as they have to be for a slide to occur.
This seems to be the principal cause of a slip not developing. The predicted stability of this
slope is marginal.

The 2.5:1 slope collapsed after 36 years, with an average pore pressure ratio on the
rupture surface of r=0.236, quite close to the equilibrium value of 0.285, and with
R=0.591. Figure 8.29 shows the behaviour just before collapse. Two horizontal shears
have developed, with signs of a third deeper one. The final rupture surface incorporates
the upper one. The behaviour is similar to that shown in Figure 8.23 for the 4:1 slope. The
relevance of the lower shears to field behaviour depends whether there is a substantial
depth of clay of similar strength below the base of the cut. There is slightly less
progressive failure than for the 2.5:1 slope, lOm high (analysis (171).

200
8.3.7 Effect of Drainage at Base Boundary

All the analyses described previously involved swelling from the surface with an
impermeable lower boundary, which is approximately equivalent to a deep clay layer.
Swelling may occur from a permeable stratum at a relatively shallow depth, as well as
from the ground surface. Analyses [31] - (35] were run to examine this effect. Analyses
[31] and [32] were of the 3:1 slope, lOm high, with swelling from the base of the mesh
1 Om below the bottom of the cutting. A pressure of 49kPa was adopted on this boundary,
representing a head 5m below the bottom of the cutting. Analysis 1331 was made of the
4:1 slope, lOm high, and analysis [34) of the 4:1 slope, 15m high, with the same base
boundary condition. Analysis [351 was as analysis [341, but with the base boundary
pressure increased to 1 47kPa, equivalent to a head at the level of the bottom of the
cutting.

For each analysis it was necessary to establish the equilibrium pore pressure prior to
excavation, according to the base boundary condition and the variation of permeability
with depth. This was done by setting up initial hydrostatic pore pressures with an initial
constant value of K 0 and a surface Suction of 1 OkPa, then performing a consolidation
analysis with the modified lower boundary seepage pressure. This involved an increase in
vertical effective stress, and a reduction in K0 according to the value of Poisson's
ratio, u: tUh /fO' V =U/(l-U). The value of K 0 after consolidation is no longer constant with
depth. The equilibrium pore pressures and values of K0 after consolidation are shown in
Figure 8.30. The value of K0 before consolidation was varied to give the results shown

The development of average pore pressure ratio and mid-slope displacement with time for
these analyses is shown in Figures 8.31 and 8.32. Analysis 1311 predicts collapse after
35.5 years. The vectors of incremental displacement, the contours of accumulated
deviatoric plastic strain, and the velocity characteristics are shown in Figure 8.33.
Comparison may be made with analysis (10), with swelling from the ground surface only,
which had a similar average value of K 0. The rupture surface, shown in Figure 8.33, is
very similar to that shown in Figure 8.13, although it is slightly deeper. on the rupture
surface at collapse has reduced from 0.238 to 0.210, and R has increased from 0.623 to
0.715.

Analysis [321, with the same assumptions as analysis (31], except that an average K0 of
nearly two was induced, showed no collapse, and equilibrium pore pressure was reached
(see Figure 8.31). Figure 8.34 shows the predictions when equilibrium pore pressure was
reached. The equilibrium pore pressure ratio on the potential rupture surface was
i,=0.230, higher than the pore pressure at collapse from analysis (31). Thus the slope

201
was more stable, despite the higher value of K 0 . The explanation is as before, and may be
seen by comparing Figures 8.33 and 8.34. With the higher value of K 0. a deeper shear
surface propagates from the toe, whereas this is only a muted feature with the lower
value. The lateral stresses in the slope, where the rupture surface would eventually
develop, are relieved by this deep shear, and there is less progressive failure. However, the
slope is quite close to collapse.

The results from the analyses of the 4:1 slope are shown in Figures 8.35, 8.36 and 8.37.
tn each case no collapse occurs, and equilibrium pore pressures are reached.

8.4 DETERIORATION OF SURFACE CLAY IN CUTTING SLOPES

Shallow slides have been commonly observed on both road cutting and embankment
slopes. No such slides were reproduced by analyses of cutting slopes with constant
surface boundary conditions and soil properties. As discussed in Section 7.5, 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 pressures such as are generated in a saturated cracked zone (Crabb and
Atkinson, 1991; Perry, 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. A higher permeability zone due to cracking was incorporated on the
slope surface, and a wet surface boundary (zero suction) was sometimes used to
reproduce the worst limiting end-of-winter ground water conditions.

Most of the analyses in which a high permeability zone was simulated have been made for
embankment slopes and will be described in the next chapter. Analyses (201 - [23] (see
Table 8.1) were performed on 3:1 cutting slopes, lOm high, with a high permeability zone
approximately 2m thick (three rows of finite elements, see Figure 8.1). These analyses are
described here. An initial K 0 =2.O (analyses 120) and 121]) and K 0 =1.O (analyses (221 and
[23]) was adopted, with surface boundary suction of lOkPa (analyses [20] and 122]) and
zero (analyses [211 and [231).

The permeability adopted for the 'cracked' layer was k =1 0m/s. This is the same as the
permeability of the 'un-cracked' clay at the top of the slope, and is seven times higher
than the 'un-cracked' clay at the bottom of the slope. The layer only produces a
discontinuity in the pore pressure distribution during swelling. As long-term equilibrium is

202
reached the pore pressure distribution approaches a hydrostatic condition, and is
insensitive to the distribution of permeability with depth.

Analyses [20], [21] and 122) all showed deep-seated sliding similar to that shown by the
equivalent analyses without the cracking zone. The development of average pore pressure
ratio on the rupture surface and mid-height horizontal displacement with time are shown in
Figure 8.38. Results from analyses [81 and [9] for the equivalent slope without the
permeable superficial layer are shown for comparison. As can be seen, collapse was
significantly accelerated by the permeable surface layer, which promoted swelling. The
results in Table 8.1 show a slight increase in progressive failure due to the less uniform
seepage pressure distribution induced by the permeable layer. The residual factor R
increases from 0.500 (analysis [81) to 0.632 (analysis [2011, and from 0.436 (analysis [9])
to 0.490 (analysis [221). The general nature of the three deep-seated slides is shown in
Figure 8.39.

Analysis [23], with K 0 =1.0 and zero boundary suction developed a shallow slip shortly
after excavation. The slip developed from the bottom of the slope. Figures 8.40 and 8.41
shows the development of the slip with time. Figure 8.42 shows the development of the
deformations with time.

The predicted time-scale for the shallow slip to develop is shorter than is usually observed
in the field. However, the prediction involve 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.

Two aspects of the predicted slipping are noteworthy. The first is the delay in the
development of the slip, once it starts, which takes nearly a year to form fully. Figure 8.43
shows the deviatoric plastic strain contours, the pore pressures and the contours of the
minor total principal stress for 1 .0 and 1.55 years after excavation. There is a tensile force
supporting the slide on its up-slope boundary when it develops beyond a certain length.
The pore pressures, where the progressive failure is occurring, are decreased by the tensile
stresses. The delayed slip is controlled by the equilibration of these reduced pore pressures
in the vicinity of where the rupture is propagating. In the field such tensile forces are
unlikely to be sustained, and vertical cracking is likely to occur.

The second point of interest is why the shallow slide developed with an initial K 0 = 1.0,
whereas a deeper seated slide developed after a longer delay when K0 = 2.0 was assumed.
Figure 8.44 shows the development of the pore pressure along the potential shallow

203
rupture surface in the two analyses. The amount of swelling and the final pore water
pressure is greater with K0 =2.O, although no shallow slip developed. The probable cause
is shown in Figure 8.45, where the predicted deviatoric plastic strains and pore pressures
1 year after excavation are shown. A deep-seated horizontal rupture surface has
developed with K0 -2.O, under virtually undrained conditions. As with the previous
analyses, the stress relief caused by this deep-seated rupture surface seems to reduce
subsequent swelling and the potential for progressive failure. This prevents a shallow
slide.

8.5 FURTHER DISCUSSION AND CONCLUSIONS

The parametric studies of the 3:1 slope, tOrn high, show that the probability of collapse is
relatively insensitive to the value of K 0 assumed. It is predicted for values of K0 from 1 .0 to
K 0 =2.15(K) at rather similar pore pressures. The value of K0 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 K0 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 strain-softening to start; and, only when K0 is low, (iv) an inclined
section near the original ground surface on which rupture only occurs post-collapse. The
location of the back of the slip depends ort the angle of dilation adopted in the analysis,
but the horizontal surface does not.

The deep-seated nature of slips predicted at the high values of K0 generally present in in-
situ London Clay is reasonably typical of field experience. The analyses show that, at the
higher values of )(, the rupture surface is developed by progressive failure at a greater
depth than the critical surface which would be determined by limit equilibrium analysis on
which r/q1,' is a maximum. As a consequence the operational strength on the rupture
surface is less than that which would be determined for the same slope with the same
pore pressures by limit equilibrium analysis.

The two analyses of the 4:1 slope with high K0 showed similar behaviour to the 3:1 slope.
Both failed eventually, after more than 100 years. The lOm high slope was near pore
pressure equilibration. The 1 5m high slope was not, but time to collapse was long because
of the long drainage paths. However, swelling times in real slopes could be shortened by

204
geologic features such a sand layers within the clay.

6m high slopes were more stable. While the 3:1 slope was just stable in the long-term, the
behaviour of the 2.5:1 slope was similar to that of the 3:1 slope, lOm high. Equilibrium
long term pore pressures were decreased, and the amount of progressive failure was also
slightly reduced.

The analyses carried out with swelling from the lower boundary and with varying degrees
of underdrainage indicate that the probability of collapse depends primarily on the ultimate
pore pressures developed around the potential rupture surface. Bromhead and Vaughan
(1980) and Vaughan (1989) have shown that, when permeability decreases with depth,
the pore pressures in the upper part of the soil profile are largely protected from the effect
of underdrainage. This effect is illustrated in Figure 8.30. However, underdrainage has a
reduced effect at depth and stability is improved by underdrainage. The amount of
progressive failure is increased slightly by swelling with underdrainage.

There is considerable variability in the predicted time to collapse of the slopes analyzed,
bearing in mind that a constant permeability is assumed. The parametric studies on the
3:1 slope gives 11 <t1 <45yrs. The steeper slopes fail in less than a decade. The 1 Sm
high, 4:1 slope failed after 145 years, and the lOm slope with increased surface suction
after 100 years. The field experience also shows considerable variability in time to
collapse.

Figure 8.46 shows values of at collapse and after equilibration, both on the eventual
rupture surface, plotted against slope inclination for all the analyses. The pore pressures
required for collapse increase with decreasing slope angle, but so do the equilibrium pore
pressures. The analyses show that collapse is sensitive to the magnitude of the pore
pressures in the slope, and thus to the equilibrium pore pressures to which these pressures
tend.

There will be seasonal variation in superficial pore pressures, not represented in the
analyses, superimposed on the general trend (Vaughan and Walbancke, 1973; Walbancke,
1976; Chandler, 1 984b), and thus real collapse will tend to occur in winter when surface
suction is zero.

The surface hydraulic boundary condition has a strong effect on stability. Figure 8.46
shows that the increase in surface suction from lOkPa to 2OkPa in analysis [13] goes
more than half way to stabilise the 3:1 slope. It has almost as much effect as
underdrainage from a layer I Om below the cutting bottom to a head 5m below the cutting

205
bottom. Surface pore pressures can be reduced by increasing evapo-transpiration through
the controlled use of vegetation, or by surface drainage which improves pore pressures
below the depth of the drains. Farrar (1990) presents observations of the effects of
counterfort drains in a slope in London Clay. These observations suggest an average
reduction at depth midway between drains of about 11 kPa, and a consequent mean
reduction of about 2OkPa. The development of suitable drilled drainage systems could also
reduce pore pressures sufficiently to eliminate the risk of long-term deep-seated sliding.

No shallow slips are predicted in analyses in which uniform soil properties were assumed.
This confirms that alternation of the surface soil is necessary for these slips to occur. if a
superficial zone of higher permeability is introduced in a clay cutting slope, such as might
be caused by shrinkage cracks, then high pore pressures develop quite quickly in this
zone. This thin layer swells, and can fail by progressive failure from the bottom up. No
reduction in the peak strengths measured in the laboratory is required to reproduce
collapse. This mechanism was inhibited by a high initial value of I(, apparently because of
the early formation of a deep-seated rupture surface and the stress relief which this then
caused, which reduced subsequent swelling and the potential for progressive failure.

It is worth noting that pore pressures at depth may be reduced by remedial works for
shallow slips. Details of such works are given by Greenwood et 8/(1985) and Johnson
(1985). The effect of slope repair of embankments is examined subsequently. Repairs
involving replacement of a shallow slip by free-draining granular material, or by any
stabilised material with an effective drain below it, has the effect of establishing a zero
pressure boundary at the base of the drain. Thus replacement with 2.5m of granular fill
has an effect on deep-seated sliding equivalent to increasing the boundary suction to
25kPa. This suggests that, if and when higher slopes which may be at risk of long term
deep-seated sliding suffer shorter term superficial sliding and are repaired, then the repairs
should be of sufficient depth and include good underdrainage, so that the risk of long term
deeper seated sliding is eliminated.

206


= = = = — = — = = = = = = = = =

o
z

- in
— N
CO CO
10 (I 10 CO N N
CO CI C
N 10
O
N
d
N
0

00 000000 d .-ddo - d

- UI a in o 0 o 0 N 0 0• 0
IdmN -
UI
E.
- N ('I N .- .- -. .- N . C) )
'
1 UI C)

CO C') C') C') C') .- 0 0 0 CD CO C') CO

- C) CO N 0 0 0 0 0 1') 0) 0 0)
1 U, N CO CO CD in IS) CO CO CO CD CO U,
00 00 00000000 0

U) 0 U) IS)
N CO
0
CO CO N
0 CD 0
0
U) IS)
CO
(') C') C') C') C') C') C') I C-1
00 0 a 0 aa a a a

N N 0 CD CO CO IS) .- 1 CD CD 0 N C') CO
U, N 1 CD 0 U, C') CO 0) (0 (0 CO 0) N CD (0 C')
0 - .- N — - N ("1 0 N N N C') 0 0 N ('I Dl
-
° 00000a00000a oddd a a
• 0) N U, . N N CD N 0) CD CO a N
CO CS) — U, 0) — C') C') N
0. 00 o•q q 0-:-° 0
99 900 999 o9990 9

ON 0) IS) 10 CD NO).- Dl (0 IS)

• 0 N .- N CO N N N in 0 CD CO N
IS) CD CO (0 CO CO CO CD N N CD 0) C')
0

ONCO 0O 0-IflNN N
CO CD 0 0 - Dl N N - C') Dl 0 C')
Dl N N Dl N N N N Dl N C')

22 0 0 2 2 0 0 2 2 0 0 0 0 0 0 0 0 0

0 o, m a C m m ', co C., c., C' UI

0)
0.
0
CS)

0 0 0 IS)
000 NIS) U,IS, N
0 0 0 UI 0 0 0 IS)00-00000
000.- 0 0 0 0 0 0 0
.- - - .- N Dl Dl N Dl Dl N N N Dl Cl U
'I-
0
UI
-
0. - 0 0 .-
.- ,— 0,—0 ,- 0 0
—0.- . 0 0 0 0 0 0 0 0 0
N
0 0.-
0 (0
U)
>-
CD

CO

E
('1 5') C') ('I C') F) C') 5') C') C') F) C') I N C') C') 01
.5-
()

CS)

222222222222 2222o (0

cx
CD
H100 N2 0
CD
I-


= = = — = = =

0) c N U) I) N N 0
4E U i I E !
- -. 0 0 0 0 0 ci o

U) . ..,

2 r-

____ —
N 0 0 in =
0
C') 0) a, U) u
U) * N
o a ao a
C)
____ —

'-S
- U) U) 0 0 0 0 0
U) in C') C') N E
C C') NCINN C')
0 0 0 0 0 0 0
____ —
0•
0) in 0) 0 0 4'-
a, N 0
00. .- N N C' N
05 0 0 0 0
0 0
— — .
____
0) N CD — N 0 N -
> N 1 U) CD 0 0 0) (0 - 0
-: -: o q ci
0 0 0 0 0 0 0 0 o
____ g
. •Eo
U)C 0) •
-(I)
- 0 (0• 0
0) IS) N N
LI) (0 II) N (0
0
_____ — — S C'4
E
N 0 (0 CD N
c- -
N E g
N
____ — —
o•
• 0

0 0 0 0 0 2 U)
V a,
U. 0.
_____ ----------- - a- V
0
(O U)
,,
C') a, a, C') C') C') C') C') C') 0)
U)

C.)
'4-
0
0• 00• 0 ow .-N—— 0 U,
N N . — N a,
C,,
_____ S S -
— --
a,
:2
0
a,
2 0 0 0 2 2 2 2 2 2 2 - E
0
I-
____ — -- .4-
0)
5 —
(I) (I) () () I') (') C') C') * (I)
0 a,
S
_____ • U
0 a
4-
C
0 0 0 0 0 0 0 0 in U)
0 0
, e e r e ,- ,- ,- .- ,- U) C.,
—

0 - N U) 0 — N U) in 0)
- N N N N C') C') U) I') U) .0
C
____ __ __ = = I-
 Elements 'excavated' In layers
'5
.0
.0 '5
'5 '5
E E
'5
0.
0. E
E
'5

'5
'5
C
0
N

U
I-
Fixed boundary . Impermeable or constant pressure 0
x
0 Scale 25rfl
I I I I I

Figure 8.1 Typical mesh for analysis of cutting slopes

ANALYSES
2:1 2.5:1
3:1
I 1,2, 3, 4, 5, 6, 7, 8, 9, 10, 11,12,
13, 16, 20C, 21C, 22C, 23C, 30, 31D, 320
cEPT10m 2.5:1 2i1
17 15
____ 3:1
2M
18 19
rPTh6i
- Cracked surface layer
- Underdrainage

I—'TH10m
E Extra depth
1 14,14E, 330

29,340,350
DEPTH 15m

0 50

Figure 8.2 Mesh geometries used in analyses

209
 1..

(a) deformations during excavation

•
. .:
¶
••

(b) deformations since excavation, 9.0 years afterwards

• • ...

o Scale 25m
I 1 1 I

0 Vector scale 5m
I I I I I I

(C) deformations since excavation, 14.5 years afterwards - just before collapse

Figure 8.3.1 Typical analysis of 3:1 cutting slope, lOm high, K 0 = 1.5, surface suction lOkPa
- Development of deformation

210
 _, .-

--

(al after excavation

Ib) 9.0 years after excavation

• ./

0 Scale 25m
L I I I

(c) 14.5 years after excavation - just before collapse

Figure 8.3.ii Typical analysis of 3:1 cutting slope, lOm high, K 0 = 1.5, surface suction lOkPa
- Incremental displacement vectors

211


Pore pressure
before excavation
Pore pressure Pore pressure kPa
after excavation
Por. pressure __._•-•- 0 ss
after excavatice _-.--,- 25 -'-
if Au • y.Az
__-.- 50 -. _-
-50 -.- 75 --
- .- 100 -'---
_-.-- 125
-- 150

(a) after excavation

Pore pressure liPs

Change since
excavation
, .-- 25 .-.-
_.,____._____-.- 50 -.-
w— 75 I

100 --
125
- ___-150 I-

(b) 9.0 years after excavation

Pore pressure liPs

_-..-,- 25 •-.-
_______ 50
________ 75

100
125
150

0 Scale 25m
I I a I

Pressure scale
0 100 200 kPa
I I I

(c) 14.5 years after excavation - just before collapse

Figure 8.3i11 Typical analysis of 3:1 cutting slope, lOm high, K0 = 1.5, surface suction lOkPa
- Contours of pore water pressure

212


= c1kPa

- 50 --

i oo

4_44_4_.J__4__ 1U..._____• -
__- - 150 -.-
- ' __4_—'
_________________
200
.-. N— —U-

_,_,___,_p____,_._c— ________ -.-. a — •_ 250 —a-

________________________________________________________________________________________________ - - - F- -

(a) after excavation

ckPa
50 _.-
----- 0

U___U__N___._____4I..____._______.lI_____-NU 100 —'-

_- 150 —a-
-:- -7—..
4
_________4__________ __________
200
:
__q_____•_._.v____,,.____.,_____,_____,•••____.._•__________I
... -- —, ________ _________p 250 —.
p

(b) 9.0 years after excavation

o1kPa

50

100
—1.
150

a
_v — _...1..._.N 200 —' —
- - U
s .— . 250 _,_
_,_..._a—' - ._..._...._----..... ._ -

0 Sca'e 25m
L I

(C) 14.5 years after excavation - just before collapse

Fagure 8.3 iv Typical analysis of 3:1 cutting slope, lOm high, K 0 = 1.5. surface suction lOkP
- Contours of major principal effective stress, a,'

213


I : (i' t
2Ze>',,/,'J (1
' ' ,' , / . -- — ..'- ——
r _
- - =
-
= =
-----___________ = = = =
________-- -- -- --
__ -- __ -- - - -
- -
-
___________ ____ -- -- -- - - * - - - - - - - - - -
flflflfl -- -- -- -- -
_______________*______ -- -- -- -- - - -
- * __________ -- -- -- -- - -

(a) after excavation

-f IPIIIIIulIISIlSSlSSISSlIII',--
nupiuresuisaC #11:55 , , , , III,,,,,
liii Is II IS II IIl###-- - ----------
_,,,,/,, 111I111I ,;,#* -------------

' /$4'(
J//j
== = / .f f = = =
=
------- -- __ __ __ - - - = =
---------- =
------- -- __ -- __ - ———————————-
- __ -- -- -- __ __ - - - - - ----------
__________ __ -- -- -- - - - ----------
************ __________ -- -- -- -- - - - ----------
- ***********
- __________ -- __ -- __ - - -----------

(b) 9.0 years after excavation

Rupture surface

*** __ -- -
_ *- -_ _ - - * - _
c;17
--------- = = =
*****-- _- - -
-_ - -- _ - _ ---------
***____ __ __ __ - - _ - -----------
* ______ -- __ -- -- - ------------
--- -- -- _ _ - - - - ---------------
*-- __ __ -- -- -- - - - ------------

0 Scale 25m
I_ i I I I

(ci 14.5 years after excavation - just before collapse

Figure 8.3.v Typical analysis of 3:1 cutting slope. 1Gm high, K 0 = 1.5, surface suction lOkPa
- Directions of major principal effective stress, o

214
 ____.u.___________ _ 150 ..
_l,__..t-l4.*.4..''
-I
,-. _I
________

(a) after excavation

°TLEE1TJTT

_----'--
__N__
___.c

.1

(b) 9.0 years after excavation

100 -.

-' ' -

k - _.- 150 -*

-t L _I .1
•___________j______
-- - _. - -I-S LI -J

0 Scale 25m
I I I I I

(c) 14.5 years after excavation - just before collapse

Figure 8.3.vi Typical analysis of 3:1 cutting slope, lOm high, 1( 0 = 1.5, surface suction lOkPa
- Contours of minor principal effective stress, 03

215

0.89 079 0.69 __*__
. 9 _6' \ "N.,
"- ..._ •- 0.49
. S
". .—S._c
'-I

(a) after excavation

Rupture surface

.L
\ 0.49

0.59
:

) 0.69 \\
i
0.79 \
s. 0.89

(bI 9.0 years after excavation

0.49
'-'S..
5-

o Scale 25m
t I I

(c) 14.5 years after excavation - just before collapse

Figure 8.3.vii Typical analysis of 3:1 cutting slope, lOm high, K 0 = 1.5, surface suction lOkPa

- Contours of mobilised shear strength, S

216
 lih
(a) after excavation

E ++ ++
XXXX X%X
xx

Ii,) 9.0 years after excavation

xxxxx
xx xx.x
xx x.x.
xx xx
Rupture suace

x xs IC
xxxxxxIC
xxx
x xx
IC

0 Scale 25m
C I

(C) 14.5 years after excavation - just before collapse

Figure 8.3.viii Typical analysis of 3:1 cutting slope. 1 Om high, l( =1.5, surface suction 1 OkPa
- Velocity characteristics

217
 Note: Accumulated devietoric plastic strains
during excavation ED< 5%.

Note: Strain-softening starts when e0=5%.

It is complete when e1P=20%.

Rupture surface

(a) 9.0 years after excavation

•'•;Pture Surface

e9=5O% eD=2O%

1W 14.5 years after excavation - just before collapse

Figure 8.3.1* Typical analysis of 3:1 cutting slope, 1 Om high, K 0 = 1.5, surface suction lOkPa
- Contours of accumulated deviatoric plastic strain, ED

Pore pressure kPa

Equipotentials

0 100 0 100 0 100

Pore pressure on vertical profile

o Scale 25m

Figure 8.3.x Typical analysis of 3:1 cutting slope, lOm high, K 0 = 1.5, surface suction lOkPa
- Contours of equilibrium pore water pressure

218
(a) after excavation

-2.5 -2.0 -0.5
,1t.
.1.

- t
-1 0 .____._, _______, -.- .__.
___.?_. -.----
_* ___, _____

(b) 9.0 years after excavation

__. .__i7
-0.5
-1.0 Av/v%
_._ -3.0 7
-, -2.0
-2.5
.- ___.__.
_.l_.___S ___.0
.-'-c-- _1s-'
'.__._._ ._.....&•' -0.1
e__
e-••'_•_

0 Scale 25m
I I 1

Note: in a saturated clay of y = 18.8IcN/m3, water content 32%, a volumetric

strain of 1.0% is equivalent to a water content change of 0.7%.

(c) 14.5 years after excavation - just before collapse

Figure 8.3.xi Typical analysis of 3:1 cutting slope, lOm high. K 0 = 1.5, surface suction lOkPa
Contours of volumetric strains, e= Av/v

219
 0.4 Ultimate equilibrium pore pressures
j12J 1.75 T [8] 2.00 [16) 2.15
[11) 1.25 Trioi 1.50 T T
0.3 TI JT]
--
0.2
a
0. 0.1 = 2kPa, R = 13°, = 0 In all analyses

0 I I

10 20 30 40 50
-0.1 Time from end of excavalion, 1 years

10 20 30 40 50
0
E -S
a [11] 1 2
0 0.2 191 1.00
C Number of analysis
K0 assumed
10.4
112] 1.75
0.6

C b 116] 2.15
0.8
0 [81 2.00

1.0

Figure 8.4 3:1 slope, lOm high, with varying K 0, surface suction lOkPa - Development
of average pore pressure ratio on the rupture surface. r, and mid-slope horizontal
displacement, 6h' with time

220


Time alter excavation, t years

0 10 20 30
E0___
£ = 1.0
0.2
K0 = 1.5 a
E 0.4
0 1<0 2.0
O.6

j:
b,C e,1
U)

0 100 200 300

0_s.. I I

-. 14.5 S = + a/2, s' = [o' + O' 3)/2 kPa

20
b/
40 14.5 12 esj1
j _ - - -
60 b
0 100 200 300
0.
0. _I____1

'I 20. - + = + c' 3)/2 kPa

b
23
19 E C
is = 1.0 4
U
o 100 200 300
0
.6
S = (a + c3 1/2 , S = ('7' + a'31/2 kPa
20 ._-.-. __'.t,t 7

40
1
" C E
60 6
= 2.0 -
Time, years
60 d
Legend:

- — - Total stress paths Numbers denote time after
excavation in years.

— Effective stress paths End of excavation

Figure 8.5 3:1 slope, 1 Om high, l( =1.0, 1.5 and 2.0, surface suction 1 OkPa - Development
of mid-slope horizontal displacement, 6h' with time, and typical stress paths

221


0.5

Number of analysis (14]

0.4 measured in the held (291 4:1
C 3:1.(
[12) [8] 116] + 2.5:1_
Ii 0.2
. . 0 ' + S
Range of equilibrium values
ru. + [19] (13]

00(17] ) 4:1 lime alter excavation, t years

0 (15] I I I

4
3:1 5 100 iso 201)

-0.2 +
J ) 2.5:1
Range of values after excavation - - -.
•
• 4:1 slope
o <3:1 slOpe
All analyses with impermeable tower boundary

Figure 8.6 Predicted average pore pressure ratios on the rupture surface, r, at collapse
versus time to collapse, and field observations

Surface suction lOkPa throughout

6 0.3
--

0.2
a
a!
0.1
/
ED'2O%

0
4 10 20 30 40
0 lime after excavatIon, 1 years
• • I
0
E [30) K0 = 1.5, 'p
' 0.2

0.4

j [3)Ka 2.0, 100,

0.6
E,,35%
E (8] a 2.0, $ = 0, e020%

V
0.8
[21 ic a 2.0, 'p a lo, e,'-'20%

Figure 8.7 3:1 slope, lOm high. K 1.5 and 2.0, surface suction lOkPa - Predicted effect
of changes in angle of dilation, sp, and rate of post-peak strain-softening -

Development of average pore pressure ratio on the rupture surface, r,, and mid-slope

honzontal displacement, 6h' with time

222
 a 1 - a3 a1-a3 kPa
______ kPa
2 2
.-

100

0 +
r
10 20 200 300 400 2

A %

I I

10 20
a -50
.E

-100

Figure 8.8 Undrained strength predicted by the model with angle of dilation q, = 100

0 Scale 25m
L i

Surface at residual
- */2
500 450
- - - Surface between peak and residual
(7) (10)

__ ,__1_.L
_-. -. -

Pore pressure
Analysis 7 - K 1.5. H - lOis. CotØ - 3. - 10. kPs
Analysis 10 - 1.5. H - lOis. Cot$ - 3. - 0

Figure 8.9 3:1 slope, 1Dm high, K 0 = 1.5, surface suction lOkPa - Depression of pore water
pressures at back of slip with a peak angle of dilation of q'= 10° (analysis [71) and
influence of angle of dilation on the rupture surface predicted (analysis [7] and [10])

223


0.4
Steady see Ps LThT

C-I- H
0.3 -___+ +
10 20
, 0.2 Collapse 0
0
- -- 0.11 AuMi.y
Collapse (1' io°j7

1.0 FE
g F° : j From
After •xcavatiOfl

Au/y.H A
0.2 0.4
Surface suction IOKPa Change in pore pressure
due to excavation
[a)

Figure 8.10 (a) Average pore pressure ratio on the rupture surface. r, after excavation, at
collapse and long-term steady state; (bi Change in pore pressure due to excavation
as a function of K0

50 Slope cotp
+4

I
.3
I,. o 2.5
*••, -.------
.2 -
Field back-analysis esv _.-- ..- o.s .--- +
I c1kPa,.'2o°
(Chandler & Skempton, 1974)
.

I a. j3'

- _—;_---
0
0 50 100
Average normal effective stress, i'0 lIPS

Figure 8.11 Operational strength predicted from analyses - Average shear stress on the
rupture surface at collapse, i, versus average normal effective stress on the
rupture surface, o,'

224


1.0
Values for iç 2.0
3:1, 100 3:1, Underdralned
3:1,p_100 R-0.601
3:1, suction 2OkPa R = 0.693 x)
3:1, flange of
3:1, cracked 0.632 J predictions
3:1, ED'35% fl - 0.598 0.5
4:1 - 0.680
3:1, E935%
4;1,H= ISM R -0.606 'I 3:11=0
2.5:1 Ti = 0.636
2.5:1,H= Sm -0.591
Ti
2.1 0.686
Ti - 0
H lOm, suction = lOkPa, 1.0 1.5 2.0 K0
unless stated otherwise

Figure 8.12 Change in residual factor, R, with the initial coefficient of earth pressure at rest.
K0

1.75[

0 20
I- 4

Scale m

Ibi flange 01 predictions from !a)

observation [normallsedj

Northolt IMddx.J. Base on

am
blue and

Figure 8.13 Ia) Summary of the rupture surfaces predicted by the analyses on the 3:1 slope.
lOm high, with varying K0 and surface suction lOkPa; Ib) Comparison of prediction
with some field observations of slips in London Clay (James. 1970)

225
 K0 = 1.25111)113 = 0.583]

-
-_---

- ,1
1.5[loflR = 0.623] 1
/
/
I

1.75112J(R = 0.643] I
F
/
/
/
,
•1'

K0 2.O(RJ(R = 0.500]

'p
'p

'-'I-

--.-1I

K0 2.15[16][R = 0.508]

- -I--I

Rupture surface not formed

Rupture surface at peak
Rupture surface between peak and residual
Rupture surface at residual

Figure 8.14 Rupture surfaces predicted by the analyses on 3:1 slope, lOm high, with varying
K0 and surface suction lOkPa, showing the strength mobilised just prior to collapse

226
 ______ _•_..- Equulbrtum por. pressures
6 0.4
I AnalysIs (8] - surface suction lOkPa
0.3 (for comparison]

çAnall$ as (13] b with no strsln4often(ng

Equilibrium por. pressure reached after 250 years.
0.1 AnalysIs [131 - surface suction 2OkPa

0
20 40 60 80 100
0 Time since excavation, t years
0 20 40 60 80 100
E

C
0
Equilibrium por. pressure reached after 250 years.
p0.5
Analysis (13] - surface suction 2OkPa
a
1.0 AnalysIs (8] - surface suction lOkPa
(for comparison]

Figure 815 Analysis [13] - 3:1 slope, lOm high, K0 =2.0 - Effect of increasing surface
suction to 2OkPa on the development of average pore pressure ratio on the
rupture surface, r, and mid-slope horizontal displacement, d, with time

Pore pressure u kPa

Equlpotentlats __..—s—.--.--..--.—.---i-.--..--'.--i 0

a a 50
U

_________ 100
_________ I.

____
150

Hydrostatic e—. Hydrostatic

Pressures on vertical sections
Steady state - variable Ii 0 Scale 25m
Just betor collapse I a

Steady state - uniform Ii

Figure 8.16 Analysis [131 - 3:1 slope. lOm high, 1(0 = 2.0, surface suction 2OkPa - Pore water
pressure at collapse and equilibrium pore water pressure

227
 • • • . I - - - - •. - - - S - S • S •
• •S_____ S S S S - S
• . . • . • S I - S. - - - - - S - S S • •

(a) Incremental displacement vectors showing shape of the slip

(b) Contours of accumulated deviatoric plastic strain, E0

*n n
k* S

"Rupture surface

++ +

+
33 +3 3+
$ $3 +. ++
444+4+444 43

0 Scale 25m
I I I I

(c) Velocity characteristics

Figure 8.17 Analysis [131 - 3:1 slope, lOm high. K0 = 2.0, surface suction 2OkPa - Conditions
just before collapse, 100 years after excavation

228
 Steady seepage

'-' 0.4 [29J H = 15m 1
[14] H = lOm

I:::
0.1 Surface suction lOkPa throughout

0
0 50 100 150 200
Time since excavation, t years
0 50 100 150 200

E
:0.2
C
0 14H 1Dm
; 0.4

0
0.6
(29]H=lSm
0.8

Figure 8.18 Analysis [141 and [291 - 4:1 slopes. lOm and 15m high, K 0 2.O, surface
suction 1 OkPa - Development of average pore pressure ratio on the rupture

surface, and mid-slope horizontal displacement, 6h' with time

pore pressure kPa

Pore pressure on verticat sections ____.-..S-S-.--S-I 0 -s--

-t-----.----.----.-.-.--.-

S
çeadystht::

150 -

h Y drostatic /f'\\ Just belore collapse '\ Hydrostatic

o Scale 25m
I I

Pore prsure scale

0 100 200kPa
l_____I I

Figure 8.19 Analysis [141- 4:1 slope, 1Dm high. K 0 =2.0, surface suction lOkPa-Pore water
pressure just before collapse, 203 years after excavation

229


/'Y

• • . . - - - - - - - - - a . .
• • e. _ - a . .
• S - - - — — - - - a - - S

(a) Incremental displacement vectors showing shape of the slip

(b) Contours of accumulated deviatoric plastic strain, E0

K K
KX K* KK KK K K
K XK KK KK XK

K *

++ + i II , 5.,, + + +

K
*"
K $44$3
NKII ..KK K

o Sc.). 25m
( I I I I I

(c) Velocity characteristics

Figure 8.20 Analysis [14) - 4:1 slope, lOm high, l(=2.0, surface suction lOkPa - Conditions
just before collapse, 203 years after excavation

230
 Pore pressur. icel.

o too 200k?. Pore pressure

I I I J kP.

Pore pressure on vertical sections

0 Scale 25m
I I I

Figure 8.21 Analysis [291 - 4:1 slope, 1 Sm high, K0 2.0, surface suction 1 OkPa - Pore water
pressure just before collapse, 145 years after excavation

• . . I - -
I - - - -. - -
' - •-. - •- -

(a) Incremental displacement vectors showing the shape of the slip

0 Scale 25m
I I I I I

LI
•

Z*.a3
a__a e,'-BO% e'..2O%

(b) Contours of accumulated deviatoric plastic strain, e

Figure 8.22 Analysis [29] - 4:1 slope, 1 5m high, 1( 0 2.0, surface suction lOkPa - Conditions
just before collapse, 145 years after excavation

231
 XKKXX%XXXX
** XX XX K
** XX XX X
XX XX

++ + + + + + + +
XX X*4$4.
XX - X%V.4.+44$
XX4CI$)CI W K I'*$+++++ $
4.4.4.4.4. 4. +4. 44. 33

(c) Velocity characteristics

Figure 8.22 Analysis 129] -4:1 slope, 15m high, lC=2.O. surface suction lOkPa - Conditions
just before collapse, 145 years after excavation

P35% e
0 0 2O% D

C,

Figure 8.23 Analysis [14Ej for deep clay bed - 4:1 slope, lOm high, K 0 =2.O. surface
suction 1 OkPa . Contours of deviatoric plastic strain, E0 - just before collapse

,' S S S S - -. - - - - - - - - - .

4 5 5 5 . S. - • S_ •S. - - - - - - - -
S. .DsI.â........... — — — ----—— - - -

0 Scale 25m
I I I

0 Vector scale 5m

I I I I I__I

(a) 1 Om high slope, t, = 2O3yrs

Figure 8.24 Analysis [14] and 1291 - 4:1 slopes, lOm and 15m deep, K 0 =2.O, surface
suction 1 OkPa - Deformations since excavation, just before collapse

232
 --.-.--.--
..-..-.-.-.-
-. - .- - - - - --

l
I l . 11111 I .
= ._.._-: = -

Iii itt,,, s
I I I I S - - - . . - .

o Scale 25m
I- I

0 Vector scale 5m
t___i I I I I

Ib) 1 5m high slope, ; =1 45yrs

Figure 8.24 Analysis [14] and [29] - 4:1 slopes, lOm and 15m deep, K 0 =2.O, surface
suction 1 OkPa - Deformations since excavation, just before collapse

0.3

0
0.2 (9] 3:1
-u

0.1
(17] 251
• 0. (15] 2:1
0..
• .
I'll 10 20 30
° -0.1 .4/ TIme since excavation 1 t years
4
0 10 20 30
0

0.2
E
£
So

0
1 0.6
0
0.8

Figure 8.25 Analyses [151 and [171 - 2:1 and 2.5:1 slopes. 1Cm high. K 0 =2.O, surface
suction 1 OkRa - Development of average pore pressure ratio on the rupture
surface. r, and mid-slope horizontal displacement, 5h' with time

233
(a) Incremental displacement vectors showing shape of the slip

E0 5%
- ^-

-*

e0 '=35% e0=2O%
E1 ' 50%

(b) Contours of accumulated deviatoric plastic strain. e'

x xx xx
xx x
xx
xx

+++4- +4

xx

0 Sc&e 25m
I I

(c) Velocity characteristics

Figure 826 Analyses [15) - 2:1 slope, lOm high, l(=2.O. surface suction lOkPa -
Conditions just before collapse

234
(a) Incremental displacement vectors showing shape of the slip

- .5O%\ e0=2O%
E= 35%

(b) Contours of accumulated deviatoric plastic strain,

,( XX xx
xx xx x
lxxx
xxx
x xx x

*
-x
xxx XR -xxx ,,.
xx x%x,'$$x,'
x xx
x *X
xx

0 Scale 25m
I I

(c) Velocity characteristics

Figure 8.27 Analyses L17J - 2.5:1 slope, lOm high, K 0 =2.O, surface suction lOkPa
Conditions just before collapse

235
 XX XX X
K
[aJ

4-4• +4+4

-.-'_'3 L
XX K
—:--z-.., ED =20% / * *
I *3 K
XX. 3333 3,'
K *3+3333
K
XXX XX 33333
XXXX*X 'K 3 33
K 3

Figure 8.28 Analysis [18) - 3:1 slope, 6m high, K =2.O, surface suction lOkPa - (a)
Contours of accumulated deviatoric plastic strain, E0 . and (b) Velocity
characteristics, 150 years after excavation - no collapse

(a) Incremental displacement vectors (b) Contours of accumulated deviatoric

showing shape of the slip plastic strain, e"

K XX XX

Jk7'<

++ $
K
XX K
XX XX
XX K
XXX

0 Scale 25m
I I I I

(c) Velocity characteristics

Figure 8.29 Analysis (191, 2.5:1 slope. 6m high. K 0 = 2.0, surface suction 1 OkPa - Conditions
just before collapse

236


iç 302.O 1.0 K0 3.0 2.0 1.0

0 1 _____ . 0

Analysis L34
An.iviiIs4\ 10 10
(8]
18] [35j
(29]
-201 ' 20
Depthm 100 2b0
lnWal por. pressure kPa

30
Depth
Initial pore pressure kPa
Figure 8.30 SoIl profiles with underdrainage- Initial pore water pressures and K0

, j 0.3 [8] Steady state values

[103 -- ___-.
6 -- -
0.2 (31] K(J*1.5
[32] K0 a 2.0
— Analysis (321 reached equilibrium
0.1 I, pore pressure ci a 0.230 alter
100 years without collapse.
a 0 I 1
2 20 30 40 50 90 100
Time since excavation, t years
/ 10 20 30 40 50 100
90
______________
4 Of.... I
I I

E
0.2 For comparison [10] iç = 1.50
(83 K0 = 2.0
o 0.4
I0
(1OJ..
0.6 (31) K a 1.5
(32] I( a 2.0
Q 0.8
Figure 8.31 Analyses (31] and [321 - Soil profiles with underdrainage - 3:1 slope, lOm high
- Development of pore water pressure and displacement with time

'- 0.4 -- _..._.[14]

(35] u = l47kPa, H = 15m
0.3 —-
/ - _________ [33] Ub 49kPa, H a lOm
,
0.2 __, /_ [34] Ub 49kPa, H = 15m
ail
For comparison (143 K0 = 2.0, H = lOm
0.1 129] K0 = 2.0, H = 15m

50 100 150 200

e -0.1 Time since excavation, t years
50 100 150 200
0
E (33) U1 , 49kPa, H lOm
0.5
C [34] U b = 49kPa, H a 15m
1.0
E -- -414]
1.5 —— (35] u = l4TkPa, H = 15m
[29]

Figure 8.32 Analyses [33], [34) and [351 - Soil profiles with underdrainage - 4:1 slope, lOm
and 1 5m high - Development of pore water pressure and displacement with time

237
(a) Incremental displacement vectors showing shape of the slip

e' =5%

ED50%

(b) Contours of accumulated deviatoric plastic strain, ED

xx xx xx
xx xx xxx
,( xx xx
x xx )(
xxx

+4
..*,
- - K,'
K K
"K -KK
"K 4 44444*
KK4$K .' 4 4
*4

0 Scale 25m

(c) Velocity characteristics

Figure 8.33 Analysis [31] - 3:1 slope, lOm high. K 0 1.5, surface suction lOkPa, boundary
pressure at the bottom 49kPa - Conditions just before collapse, 35.5 years after
excavation

238
 - — -----

-u-- - - - -.

____________ =

• 1_V - - V — - - - - - - -

.- — — — — - - - - - -

- — - - - - - - .

(a) Incremental displacement vectors showing shape of the slip

-'O ,

e0'=50%
• •1

.. e2O%
-•

-.-•- •..__s_._._

(b) Contours of accumulated deviatoric plastic strain, ç

)c xx xx
xx xx *
*
k

x X ++ ++ *I( x

* x
*1.
xx
* $3
x+ +* 3+ ++ +3 I.

0 Scale 25m
I__ t I

(c) Velocity characteristics

Figure 8.34 Analysis [32) - 3:1 slope, lOm high, K0 'i 2.0, surface suction lOkPa, boundary
pressure at the bottom 49kPa - Conditions when long-term steady state pore water
pressure is reached, 100 years after excavation

239
(a) Incremental displacement vectors

-7

ED 50 a - 0
0 -20%

(b) Contours of accumulated deviatoric plastic strain, E0

X4

4+ ++ 4+ ++ + + + +
xx
x
xx x
x

0 Scale 25m
I i I I I

(C) Velocity characteristics

Figure 8.35 Analysis (33]- 4:1 slope, lOm high, I( * 1.5, surface suction lOkPa, boundary
pressure at the bottom 49kPa - Conditions when long-term steady state pore water
pressure is reached, 200 years after excavation

240
(a) Incremental displacement vectors

4p = 5%

ED5O%

(b) Contours of accumulated deviatoric plastic strain, E

II
I,
/

j j /I!

++ ++ ++ ++ +
- x ,'' ,'-
,'
(,

o Scale 25rn
1 I I I I

o VectOr scale 5m

I I I I I I

(C) Velocity characteristics

Figure 8.36 Analysis [34] - 4:1 slope, 15m high, K0 1.5, surface suction lOkPa, boundary
pressure at the bottom 49kPa - Conditions when long-term steady state pore water
pressure is reached, 200 years after excavation

241
 • • S
* .. — — — S — — — — — — — - - - . •
- ..___r-_t_,'_a________ — — — — — — — — —

(a) Incremental displacement vectors

—I —

(b) Contours of accumulated deviatoric plastic strain, E0

XXXkXXXXXX
x ick X,c ,c*
%XXc*)c **
XX* rkX

H—
—
- -, _.
-'I,

w ++++++$ I, -

0 Scale 25rn
I i . I I - I
0 Vector scale 5m
I I I I I I

(c) Velocity characteristics

Figure 8.37 Analysis [351 - 4:1 slope, lOm high, K 0 " 1.5, surface suction lOkPa, boundary
pressure at the bottom l47kPa - Conditions when long-term steady state pore water
pressure is reached, 230 years after excavation

242
 0.4 Ultimate equilibrium pore pressures

T I
0.3

121] [8] No permeable layer, = 2.0

0.2 [20]
Boundary suction 1 OkPa

e 0.1

10 20 30 40
4
-0.1 Time since excavation, t years

10 20 30 40

E 0
£
0
0.2
E
U

04

(2011(0=2.0 .-_--.
0.6 Boundary suction lOkPa
[21] K0 2.0
2 Boundary suction 0
(8] No permeable layer, K.,, = 2.0
0.8
Boundary Suction lOkPa

Ultimate equilibrium pore pressures

0
- 0.3
T
(22] j _- - -
I
[9] No permeable layer, K0 = 1.0
Boundary suction lOkPa
u,v,

10 20 30 40
4 -0.1 Time since excavation, t years
E 10 20 30 40

E [9] No permeable layer, = 1.0

2 0.2 (22] (0 1.0 Boundary suction 10k Pa
Boundary suction lOkPa
0,
'0 0.4
0,
C
0
N 0.6 Note: Analyses (81 and (91 are the equivalent analyses without
0
x the surface pemeable layer shown for comparison.

Figure 8.38 Analyses [201. [211 and [22] - 3:1 slope, lOm high with a permeable surface

layer, K 0 =2.0 - Deep-seated slide - Development of average pore prassure

ratio on the rupture surface, r, and mid-slope horizontal displacement. ôh

with time

243
 =

(a) Incremental displacement vectors showing shape of the slip

•------ Rupture surface

___,) Ci —ö —.1=1
•
'- ,—' '—.
E0— 50%

(b) Contours of accumulated deviatoric plastic strain, eD

* xx xx xx
XX XXX
x
lx

N
Rupture surface

4.
,cnxn_ - , .3 $
xxx,'- - .
*,' ,#.. .$

0 Scale 25m
I I I

(c) Velocity characteristics

Figure 839.i Analysis [201 - 3:1 slope, lOm high with a permeable surface layer, )(=2.0,
surface suction 1 OkPa - Conditions just before collapse

244


Ia) Incremental displacement vectors showing shape of the slip

Rupture surface

e=2O%

(b) Contours of accumulated deviatoric plastic strain, ED

xx xx xx x
xxxxx xxxxxx
*X**)(X I xx XX
x xx xx x•
x xx *2

++ ++
* Rupture surface
xxx x,cx* - ,'*,'**,'
xx .x*x'
* **%$* *L
xx xx
*x**x- xxx . x

-- -

0 Scale 25m
I I I

Ic) Velocity characteristics

Figure 8.39ii Analysis [211 - 3:1 slope, 1 Om high with a permeable surface layer, K 0 = 2.0,
surface suction OkPa - Conditions just before collapse

245
 -, .- -
1

T:: •.;.•

(a)Incremental displacement vectors showing shape of the slip

Rupture suace

-,- _IJ.
=5%

E02O%

(b)Contours of accumulated deviatoric plastic straIn, e'

XX
XXX *
*

Rupture surface

0 Scale 25m
L i I I

Ic) Velocity characteristics

Figure 8.39.iii Analysis [22] - 3:1 slope. lOm high with a permeable surface layer, K 0 = 1.0,
surface suction 1 OkPa - Conditions just before collapse

246
(a) 0.5 years after excavation (d) 1.5 years after excavation

!.TTTTI.

(b) 1 .0 year after excavation (e) 1 .55 years after excavation

0 Sca'e 25m
I I I I

(C) 1.25 years after excavation (1) 1 .6 years after excavation

Figure 8.40 Analysis [23] - 3:1 slope, lOm high with a permeable surface layer. K 0 = 1.0,
surface suction OkPa - Incremental displacement vectors showing the
development of the superficial slip

247

(a) 0.5 years after excavation (dl 1.5 years after excavation

E0 20% e050%

(b) 1 .0 year after excavation (e) 1 .55 years after excavation

ED' 5%

20%
E0 50%
E0 20%

Scale 25m
I I I I

(c) 1 .25 years after excavation (f) 1.6 years after excavation

Figure 8.41 Analysis [231 - 3:1 slope, lOm high with a permeable surface layer. Kc,=l.O,
surface suction OkPa - Contours of accumulated deviatoric plastic strain, ED.
showing the development of the superficial slip

248
(a) 0.5 years after excavation Id) 1.5 years after excavation

TT

(b) 1 .0 year after excavation le) 1.55 years after excavation

Scale 25rn
a a e
o Vector scale 5m
t I I I Ii

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

Figure 8.42 Analysis (231 - 3:1 slope, lOm high with a permeable surface layer, K 0 = 1.0,
surface suction OkPa - Deformations since excavation

249


D 5%
,P_20%

(a) Contours of acc. deviatoric plastic (b) Contours of acc. deviatoric plastic
f0P,
strain, year after excavation strain, ED, 1.55 years after excavation

___—;-- <I

- 50
__'• ' - —I--- __—•
.._a_
- .— .-
pa-
--

(c) Contours of pore water pressure (d) Contours of pore water pressure

1.0 year after excavation 1.55 years after excavation

—I —.-----
0O
_______J
_L_t _••_••L ___4_ __..R
-'
—u — —U
—u —p P
200(pa
.__I___, -
_d__j _4_J __ ___U•
I ____U
I __I__I-
—I-- I __l.___I'

0 Sca'e 25m
I I I

(e) Contours of minor princ. total stress, G 3 u (1) Contours of minor princ. total stress, 03,

1.0 year after excavation 1.55 years after excavation

Agure 843 Analysis [23] - 3:1 slope, 1Dm high with a permeable surface layer, K 0 =1.0,
surface auction OkPa - Contours of accumulated deviatoric plastic strain, D'

pore water pressure and minor pnncipal total stress, 0 3, showing the role of
tension in the development of the superficial slip

250
 Ultimate equilibrium pore pressures
20 11.5 years ____________________

year after excavation

! -20

years

end of excavation

-80

-100
(a) Analysis [211 - K0=2.0

1.6 years -
20 Ultimate equilibrium pore pressures
just belore collapse

0.ml - _ _
0
r after excavation
:
yea--05yearS
I
excavation

0
-AD
1
0.
-60

(b) Analysis [23] - K 0 =1.0

0 10 20 30 4Dm
Scale

shallow rupture surface

surface layer

I I

Figure 8.44 Analysis [21) and [23] - 3:1 slope. lOm high with a permeable surface layer.

K 0 = 2.0 and 1 .0, surface suction OkPa - Development of pore pressure along
the shallow rupture surface with time

251
 e09=20% T" o'' -
.

-
:
-s---0

(a) Analysis [211, K 0 =2.0 - Contours of acc (ci Analysis [21], K = 2.0 - Contours of
deviatoric plastic strain. €0 pore water pressure

---z-- -U •_____
=.- \ ___0
ED-5% _.o__I._0

D-2O% -.- -, U-' - I---

__0 -I-
0___t
_z
_•__- ,
0
_N - _

0 Scale 25m
I I L I

(b) Analysis [23], K 0 = 1.0 - Contours of acc. (di Analysis [23), 1( 0 = 1.0 - Contours of
deviatoric plastic strain,
ED pore water pressure

Figure 8.45 Analysis [21) and [23] - 3:1 slope, lOm high with a permeable surface layer,
= 2.0 and 1.0, surface suction OkPa - Contours of accumulated deviatoric
plastic strain, E0 , and pore pressure after 1.0 year. showing the development
of the deep and shallow slides

252
 15m
Change in steady state value
0.4
due to increasing surface
suction from lOkPa to 2OkPa
-

0
.s

0.3
vIu
6mX ,'

I-
0
6m

0.2
o1 steady state
values with underdralnage

0.1 /
.
C

0
I 1
2 3 4
Slope inclination. colJI

Figure 8.46 The variation of average pore pressure ratio predicted on the rupture surface, r,,

at collapse and at equilibrium with slope inclination

253
 CHAPTER 9
ANALYSIS OF ROAD EMBANKMENTS

9.1 INTRODUCTION

The geometry of embankments is potentially more complex than that of cutting slopes, as a
clay embankment may sit on a foundation of varying strength and permeability, which will
affect its performance. The extreme conditons of undrained clay and of a drained granular
material have been examined here. The analyses are summarised in Table 9.1. The meshes
used for the lOm and 6m high embankments are shown in Figures 9.1 and 9.2 respectively.
The analyses were mainly of 1 Om high embankments, which were thought most likely to
show deep-seated slips. An external slope of 2.5:1 was adopted, representative of the
steeper embankment slopes adopted in practise. The lOm high embarikments were
constructed in 7 layers, in 10 increments of (oading taking a total construction time of 0.25
years. The 6m high embankments were constructed in 5 layers, in 6 increments taking 0.15
years. Consolidation was allowed during construction.

After embankment had been constructed, a specified surface suction was applied and swelling
(consolidation) allowed according to the new boundary conditions. Time steps used and
monitoring of solution convergence were generally the same as those employed in analyses
of cuttings (see Section 8.1).

The permeability model used did not allow permeability to vary with the void ratio (and
therefore effective stress) of the soil being modelled. Neither could the undrained
characteristics of partly saturated fill be fully modelled (see Section 7.4.2.5).

The effect of permeability varying with effective stress was examined in one dimension by
a closed form solution (see Vaughan, 1989). This is illustrated in Figure 9.3, which shows
vertical flow in a slope to either a fully drained or a partly drained foundation. The relationship
between permeability and effective stress is appropriate for a clay fill (Vaughan, 1989).
Significant excess pore pressures are predicted even when the slope is fully underdrained
(lines A). With partial underdrainage (lines B) the pore pressures are increased, although the
effect of permeability changing with depth is reduced. The partial drainage condition
illustrated in Figure 9.3 is equivalent to a foundation of clay on which there is a superficial

254
layer, say 2m thick, thirty times more permeable than the clay fill at the slope surface. Such
foundations are likely to occur in practice below some embankments.

Steady-state pore pressures can only be observed in old clay embankments. High pore
pressures have been observed in the downstream slopes of old embankment dams of clay fill
(Wa$bancke, 1976).

The variation of permeability with effective stress, and therefore depth, in a fill is not easy
to model by varying the permeability with depth, as was done with the cutting slopes. A
uniform permeability was therefore adopted (Section 7.4.2.7). However, a shallow 'cracked'
zone of higher permeability at the slope surface was incorporated in all the analyses. Thus,
some pore pressures are developed in the long-term in an underdrained embankment as a
consequence of this superficial layer. Higher pore pressures have been generated in analyses
[E2J - [E51 by assuming an impermeable boundary on top of a drained foundation. The pore
pressure in the fill then increased with time in a manner similar to that predicted in Figure 9.3
for a partly-drained foundation. This allowed the pore pressure required to cause collapse to
be generated in a realistic manner.

A permeability of k=10 9 m/s (6.7 times the permeability of the fill) was assumed for the
cracked zone in all the analyses except [E4J and [E51, for which k=l.5 x 10 9m/s was
adopted. The cracked zone is three elements thick (Figures 9.1 and 9.2) and varies in
thickness from 1 .8m at the top of the slope to 2.2m at the bottom. A surface suction of
lOkPa was generally adopted, representing average annual conditions. The limiting surface
suction of zero (extreme winter conditions) was adopted for analyses [E7J, [E8J and [E91.

Three groups of analyses were performed; analyses [Eli and [ElO] with a clay foundation,
analyses [E2] - [E51 with a drained granular foundation but with an impermeable boundary
between fill and foundation, and analyses EE61 - [E9] with a drained foundation without the
mpermeable boundary at the base of the fill. Analysis 1E31 was a continuation of analysis
[E21. The shallow slide generated in analysis [E21 was 'excavated' and 'replaced' with
granular fill, and the swelling process continued until a deep-seated slip was generated.

9.2 EMBANKMENT ON CLAY FOUNDATION

In the first analyses (not included in the thesis) a clay foundation was adopted with the same
properties as the clay in the cutting slopes. Undrained foundation failure occurred during
construction when the embankment was about 6m high. Analyses [Eli and [ElO] were then
performed, with a foundation with a higher undrained strength as shown in Figure 7.9.b. The
initial stresses were given by K 0 = 2.0. The development of average pore pressure ratio and

255
mid-slope horizontal displacement with time is shown in Figure 9.4. The lOm high
embankment collapsed after 9.7 years on a deep-seated rupture surface running through the
top of the clay foundation. The 6m high embankment did not fail., and pore water pressures
are calculated for the hypothetical rupture surface shown.

Figure 9.5i shows the development of the sup in the lOm high embankment. Figure 9.5.ii
shows the development of deviatoric plastic strain and Figure 9.5.111 shows the velocity
characteristics. The initial rupture surface develops rapidly along the foundation to just beyond
the crest of the embankment. Only the part nearer the slope, on which the strength has
dropped to residual (e1,=50%), is utilised in the final collapse. The inclined rupture leading
to final collapse develops later as the embankment swells. The length of rupture surface on
which the strength lies between peak (E7 5%) and residual (E =50%) is much greater than
in the equivalent cutting slope analyses, due to the flatter stress-strain curve adopted. Despite
this the amount of progressive failure is greater (R=0.753, see Table 9.1) than is typical for
the analyses of the cutting slopes. The average operational strength is thus virtually the same
as was found in the cutting slope analyses (see Table 8.1 and Figure 8.11), despite the higher
peak strength adopted for the fill.

A rupture surface develops along the surface of the foundaton below the 6m high
embankment, but the inclined surface has only just started to form when pore prssure
equilibration is complete after 100 years.

Passive failure occurs below the toe of the embankments during construction (see Figure
9.5.iii.a), as extra lateral thrust from the embankment is added to the already high lateral
stress due to the value of K0 adopted. However, this plays no part in the collapse.

Figure 9.5.iv shows that pore water pressures increase rapidly in the top of the foundation
below the high permeability cracked zone at the toe of the slopes. This promotes failure here,
where there is a lower peak drained strength than in the fill. The horizontal rupture surface
develops some way beyond the depth to which swelling from the ground surface has
progressed, although there is redistribution of pore pressure between the fill and the
foundation. The rapid development of displacement accompanying this during the first year
after embankment construction is shown in Figure 9.4. The horizontal rupture surface is
mainly at residual (Figure 9.5.ii). It loses strength further as swelling from the ground surface
proceeds. Figure 9.4 shows that horizontal displacements are of the order of 0.6m before
final collapse occurs, which happens with little warning. Analyses W1I and (E1OJ show no
sign of the development of a shallow slide.

256
9.3 EMBANKMENT ON DRAINED FOUNDATION
WITH IMPERMEABLE SURFACE BOUNDARY

Analysis (E21 is of a slope Identical to that analyzed in analysis FE 11, except that the
foundation is drained and 'strong', and that the boundary between fill and foundation is
impermeable. Failure can only occur in the fill, and pore pressures in the toe of the fill become
higher than in analysis [Eli. A shallow slip develops in the high permeability layer, as shown
in Figure 9.6.i. The post-construction deformation is &own in Figure 9.6.ii. Displacements
at the toe just prior to collapse are about O.6m.

Figure 9.6.iii and 9.6.iv show the development of deviatoric plastic strain and the velocity
characteristics with time. There is some active failure in the foundation, occurring with major
principal stresses which are nearly vertical (Figure 9.6.iv). This has no bearing on eventual
collapse. Figure 9.6.iv shows that two rupture zones develop after about 1 year, one up from
the toe along the base of the cracked layer, and one horizontally in the fill just above the
foundation. The inclined rupture zone has always propagated further than the horizontal one,
arid has much larger plastic strains. Figure 9.6.iii shows that the deviatoric plastic strains on
the horizontal rupture zone are small and there is only limited strain-softening.

Two factors seem to control the generation of the shallow slide, rather than a deep-seated
one like that in analysis [Eli. First, the undrained strength along the base of the fill is about
5OkPa under plane strain conditions. The undrained strength of the clay foundation of analysis
EEl] is about 4OkPa (see Figure 7.9). Thus there is more resistance to the early propagation
of a base rupture surface in analysis [E2J. Second, the shallow rupture surface at the base of
the high permeability layer on the slope forms in advance of the deeper horizontal rupture
surface at the base of the fill. Figure 9.6.iii shows the base rupture starting to propagate after
3 years. At this time the shallow surface has propagated nearly half way up the slope. The
strain energy in the cracked zone has been released by the shallow surface, and is not
available to promote progressive failure on the deep horizontal rupture surface.

Analyses [E4] and [E51 (see Table 9.1) were performed to investigate the sensitivity of the
results of analysis [E21 to the permeability of the cracked layer, and to the abruptness of the
drop from peak to residual strength. The displacement/time plots for these analyses are
shown in Figure 9.7. A shallow slide developed in each case, although the time to collapse
varied.

The high permeability layer in which the shallow slide had developed in analysis [E2J was
'excavated' and 'replaced' with granular fill, with zero pore pressure at its base, in analysis
(E31. The granular fill was given the same properties as the drained foundation. Swelling was

257
continued in order to examine whether a deep-seated slide would develop. The introduction
of the layer of granular fill with a hydraulic boundary pressure of zero is approximately the
same as increasing the surface suction of the unaltered slope to 2OkPa. Figure 9.8 shows the
development of the average pore pressure ratio along the eventual rupture surface and of mid-
height horizontal displacement with time. A deep-seated slide was predicted after 32 years.

The current displacement vectors in Figure 9.9.i show (a) rebound following excavation and
(b) undrained settlement following replacement with granular fill. Swelling of the embankment
(C end d) follows. After 25 years (e) the bottom of the sUp has formed. Figure 9.9.ii shows
the development of deviatoric plastic strain, and Figure 9.9.iii shows the velocity
characteristics. The latter indicate where failure is occurring currently. The state of failure
indicated in the upper part of the granular fill is a function of the properties assumed and
plays no part in the slide. The deviatoric plastic strain shown at the toe in Figure 9.9.ii.a is
a relic of the previous failure (Figure 9.6.iii.f), as no velocity characteristics are shown at this
location in Figure 9.9.iii.a. Failure is re-generated when the granular fill is placed (Figure
9.9.iii.b). Failure is generated through the granular fill toe at a late stage (Figure 9.9.iii.e). Just
before collapse about 30% of the final rupture surface is at residual (ED5 O %, see Figure
9.9.ii.f) and about 20% has not reached peak (no velocity characteristics, see Figure 9.9.iii.f).

Figure 9.9.iv shows the development of the pore pressures. At collapse (Figure 9.9.iv.b) the
pore pressure distribution has a similar shape to that predicted in Figure 9.3. However, Figure
9.3 indicates that pore pressures sufficiently high to cause the collapse would be unlikely with
a fully drained foundation. The deep-seated slide shows a residual factor R = 0.750, indicating
a substantial amount of progressive failure; more than was predicted for the cutting slopes.

9.4 EMBANKMENT ON DRAINED FOUNDATION

Analyses [E6J - [E9J were made of embankments on a fully drained granular foundation (see
Table 9.1). Results are presented in Figures 9.10, 9.11 and 9.12. Figure 9.10.i shows the
current displacement vectors for analyses [Eli and 1E81 of lOrn and 6m high embankments,
with the surface suction of zero. A shallow slide develops in the lOm high slope after 2.3
years (Figure 9.10.i.dl). A slide has started to develop in the 6m high slope after 30 years
(Figure 9.1O.i.d2) but has not fully formed. Figure 9 10.ii shows the development of
deviatoric plastic strains for these cases.

Figure 9.11 shows the development of pore pressure along the base of the high permeability
layer (the rupture surface of analysis [E71). Figure 9.1 l.a is for analysis EE61, a control
analysis performed with a surface suction of lOkPa, in whch no slide was predicted. Figure
9.11 .b is for analysis [El) of the lOm high slope, in which collapse and a shallow slide were

258
predicted. Pore pressures come close to zero after about 1 year, when the rupture surface has
just started to propagate (Figure 9.1O.ii.b). As the rupture surface develops up the base of
the cracked zone, changes in pore pressure similar to those observed in the high permeability
zone on the cutting slopes (see Figure 8.43) can be seen. The lower part of the slope yields
and creates tension and a reduction in pore Water pressure by undrained unloading higher up
the slope. The reduced pore water pressure then increases and the slide progresses up the
slope.

The rate of propagation is controlled by the rate of swelling of the cracked layer. This could
be significantly quicker in the field than in the analyses, as the permeability of the cracked
zone is likely to be higher than that assumed. It may be noted that the pore water pressures
at the base of the cracked zone in Figure 9.11 generated with zero boundary Suction are not
as high as might occur if a higher permeability in the cracked zone had been used, and not
as high as has been observed in the field (see Figure 7.16). Thus the tendency towards
shallow sliding may be underestimated.

A shallow slide did not develop in the 6m high slope (analysis [E81), apparently due to there
being insufficient strain energy in the shorter slope to create the required progressive failure.
Analysis [E9J was run on this slope with the deviatoric plastic strain at residual reduced to
e'=2O% (one third of the strain required to get from peak to residual in analysis [E81). The
current vectors of displacement are shown in Figure 9.12.i, and the deviatoric plastic strains
in Figure 9.12.ii. Analysis [E8J is shown for the same elapsed times for comparison. The
results of analysis [E8) at longer times are shown in Figures 9.1O.i and 9.1O.ii. A complete
shallow slide is developed after one year. Thus the prediction of the shallow slide is quite
strongly dependent on the rate of strain-softening assumed, which controls the amount of
progressive failure.

9.5 DISCUSSION AND CONCLUSIONS

The analyses indicate that progressive failure occurs in embankments in both shallow and
deep-seated slides. It is encouraged by the large swelling strains which occur due to the low
initial pore pressures (suctions) in compacted plastic clay fi Is. More swelling occurs in the
outside of embankment slopes than in cutting slopes, which, apart from steeper slopes
usually adopted for embankments, may account for the greater amount of superficial sliding
in embankment slopes. The amount of progressive failure in deep-seated slides is large with
operational strength more than half way from peak to res dual. It is typically greater in
embankments than in cutting slopes.

Deep-seated slides develop after long delays. The assumpt on of saturated soil behaviour

259
made in the analyses may result in significant overestimation of the rate of swelling. Thus,
it is likely that delayed collapse of ernbankments will take longer than delayed sliding of
cutting slopes.

Deep-seated slides may occur wholly through the fill, or along the surface of a clay foundaton
if this is weaker than the fill and subject to undrained pore pressures. They are only likely in
the higher road embankments. Deep-seated slides wholly In fill are unlikely in embankments
built on foundations sufficiently permeable to remain fully drained (zero excess pore pressure),
even in higher slopes. Deep-seated slides are less likely in embankments than in cutting slopes
of the same height and in the same clay.

Deep-seated slides involve the early development of a near horizontal rupture surface. There
is interaction between shallow and deep-seated sliding, and the early formation of a horizontal
rupture zone may reduce the strain energy stored in the superficial 'cracked' layer, and so
inhibit the subsequent development of a shallow slide by progressive failure.

As in the cutting slopes, the analyses of embankments show that a shallow slide can be
generated by progressive failure driven by differential swelling of the surface 'cracked' layer.
However, higher permeability due to cracking and progressive failure due to swelling may not
be the only mechanism causing shallow slides. The effects of cyclic stress changes and
possible material deterioration are not included in the analyses. The predicted time taken for
shallow slides to develop is arbitrary. It will depend on climate, the development of cracking
and the occurrence of a 'wet' period of sufficient length to develop high pore pressure at
shallow depth, as well as on the time taken for swelling to occur.

In both cutting and embankment slopes the analyses predict that shallow slides develop from
the bottom up. A rupture surface at or near residual develops horizontally from the toe and
then up the slope at the base of the high permeabi ity layer. If it reaches a critical point, then
the slide develops relatively rapidly. The toe provides passive support, even when it is sliding
at residual, and a critical active length of slide up the slope is required for collapse to occur.
Over this length there must be a net down-slope force generated by the slide, i.e. the weight,
resolved down the slide, must be greater than the shear force on the base acting up the slide.
When it is of critical length the net down-slope force from the active length exceeds the
passive resistance at the toe. Thus a critical length of slope is required for collapse to occur.
This length will depend on the swelling properties of the clay, on the rate at which strength
is lost post-peak, and on the residual strength relative to the slope angle.

It is of interest to consider the average operational strengths predicted by the analyses of

shallow slips and to compare this with conclusions drawn from back analysis of field slides.

260


This has been done for the analyses [E71 and [ES) by limit equilibrium analysis of the final
rupture surfaces (postulated for analysis [E81 in which collapse did not occur) shown in Figure
9.11. The pore pressures shown in this figure are used in the analyses. Three analyses were
performed; the first a two wedge analysis of the whole rupture surface, the second and third
being for the infinite slope with sliding at the average depth predicted by the finite element
analysis, with average and maximum predicted pore pressures assumed. The operational
strength is obtained assuming F= 1.0, with c' assumed equal to zero. Results are:

wedge analysis infinite slooe analysis _______

a y . ' ( 0 1 ay. V( 0 1 max. i1 , ' 1
0)

lOm slope, 2.35
24.3 0.10 22.5 0.10 0.20 27.5
'rs old at collapse

6m slope, infinite 0.26 >27.4 0.23 >28.6 0.31 >32.0
time, no collapse

These results are broadly similar to the conclusions of Crabb and Atkinson (1991) and Perry
(1991), who have concluded that the operational strength governing shallow sliding is
equivalent to the critical state strength. However, the operational strengths predicted above
are rather higher than critical state strength.

It must be noted that, when progressive failure occurs, operational strength is not a material
property, but is problem dependent. This was demonstrated in the analyses of cutting slopes.
It is also shown by the above analyses. The operational strength required to support the 6m
slope, which did not fail, is higher than the strength demonstrated by the 1 Om slope, which
failed. Thus analysis using operational strengths based on critical state may prove to be of
limited value In predicting susceptibility to shallow slides. Variations in behaviour between
different strata and within the same strata are likely to be due to many other factors apart
from changes in critical state strength.

The higher permeability zone which generates shallow slides by progressive failure is
apparently caused by cyclic stress changes at shallow depth, pr manly due to shrinkage and
swelling. Elimination of these cyclic stress changes by placing a layer of free-draining gravel
below the top soil, which acts as a capillary break and eliminates drying and cyclic stress
changes, seems to prevent superficial sliding on clay embankment dams. However, this is a
rather expensive technique for road embankment slopes.

According to the analyses, the shallow slides are promoted by progressive failure. Repair and
prevention can therefore involve:

261
(a) Increasing peak and residual strength, by drainage, by increasing evaporation and
reinforcing by roots from selected vegetation, and perhaps by mechanical reinforcement. To
be effective root reinforcement to below the depth of cracking is probably required.

(b) Decreasing the factors causing progressive failure, in particular the amount of swelling of
the superficial layer. Replacement by granular fill will stop superficial swelling. Lime or cement
stabilisation may do the same, and may also increase peak and residual strength. In
embankment construction it is possible that the use of wetter fill on the outside of the slope
might reduce initial auctions and subsequent swelling, although the fill would still remain
subject to large volume changes, and peak strength would be reduced.

Cc) Since rupture starts from the bottom of the slope, this is indicated as the most important
location for stabilising works.

Cd) Compaction to maximum possible density increases the peak strength of clay flU in
embankments. However, it is also likely to increase swelling strains, which can increase
progressive failure and may offset the gain in strength.

It has been shown that a deep-seated slip can follow the repair of a shallow slide in a high
embankment slope if further swelling 'eads to high enough pore pressures. Thus repairs to
shallow slides in higher slopes should include drainage to control this risk by reducing pore
pressures at depth. This has already been discussed in Section 8.5 in connection with cut
slopes.

262


- = = = — = -

4- - - ID CD N 03 CD
2 - -. - - -
Pd
. .c
CD C
ID ID
r C4
CD C
- -. 0 0 0 0 CIS)
- CD
øi•
_______ '4.

N NO0 ID

., -. .2
C3 Cfl,

.2
D

______ - - - - - - - - - - ''U
___

C1
0 0 o
I
.5 0

_______ C
_____ -

I a

CD 0 (0
o 0) CN .- -
-:
9 .. 9
- a
CD 0) 8o
.2

- - -- - 4-
0) 0
. CC

CD
03 _ 4-
0 C
o—C CD

- .E
a a C
CD
a a • 0 0 • - • .0
0 0 0 ' 0 0 0 0 E
CD
44-

- 0

______ CD
CD

-. 't —Cs1 Ca)

CD
C
- CD

- - - - - - - - - - a C E
4- C
0
S 'l-

2 2 2 2 2 2 4-4-CD
CD
4-

x
(I)
CD
______ -
I-

0 C) ' ID CD N CD CD
UJ UJ LU LU LU LU LU LU LU - CD

.0
CD
— = = = = I-
 Embankment 'constructed' In layers
a
.0
(5
a a,
.0
(5 C,
a C.
a E
C.
E C
(5
I-
C (5
IS
I') (5
a C
IS 0
N
C 0
Fixed boundary - Impermeable

o Scale 25m
I I I

Figure 9.1 Typical mesh for analysis of 1 Om high embankment

Embankment 'constructed' In layers

a a
.0 .0
CS
(S
C, a,
E
I-
C, C,
0.
E E

E
(5
I-

a
CS
C
0
N Fixed boundary- Impermeable
I-
0
I

Figure 9.2 Typical mesh for analysis of 6m high embankment

1*

'.4

lOm
B
Bj
- • • -• II\ \
iJA \ B A
Predicted pore pressure, ID downward flow,
Ii Ice. exp[-B.c'j, B D.01m2/kN, a' - y.z - u

FIgure 9.3 Pore pressure in a fill with permeability decreasing with increasing effective stress,

wet upper boundary, one-dimensional downward flow, drained and partly drained

foundation

264


0.3

d 0.2 -UItlmate equilibrium pore pressures

T
J [1] H=lOm 1101 H=Sm

0 I 1 I
5 10 15 95 100
Time since construction, t years
o —0.1
4 'c (1]H10fl1
H=6m

In-situ clay o
I 1Cm
fill
in-situ clay
• Scale

Time since construction, * years

5 1P 15 95 100
E 0
£

(10] H6m

I:
(13 H=lOm

Note: Rupture surface for 6m high embankment (Analysis jlOJ)

is hypothetical.

Figure 9.4 Analyses [Eli and [E1OJ - lOm and 6m high embankments on undrained
foundation, permeable surface layer, surface suction lOkPa - Development of average
pore pressure ratio on the rupture surface, r, and mid-slope horizontal displacement,
oh, with time

265


!/JId'IJ11

• . . . S S - -. - - - -
• 4_s_p_I_..,- - - - - - -
• - S - - - - -

(a) end of construction

(b) 1.0 year after construction

iI1Sn • 1
' 4 .. 4 ' ..
S
S
I
•444•S*S• S S S S S S U

o Scale 25m
E , i I I I

(ci) 9.75 years after construction (c2) 100.0 years after construction
- just before collapse

Figure 9.5.1 Analyses [Eli and EE1OJ - lOm and 6m high embankments on undrained
foundation, permeable surface layer, surface suction lOkPa - Incremental
displacement vectors

266
 (a) end of construction

(b) 1.0 year after construction

€0 1U1j7o €obU% 'D0%

EP50% D

0 Scale 25m
a I

(ci) 9.75 years after construction (c2) 100.0 years after construction
- just before collapse

Figure 9.5.1 Analyses [El] and IE1O] - lOm and 6m high embankments on undrained
foundation, permeable surface layer, surface suction lOkPa - Contours of
accumulated deviatoric plastic strain,

267
%%
%%

(a) end of construction

(b) 1 .0 year after construction

0 Scale 25m
I I I

(ci) 9.75 years after construction (c2) 100.0 years after construction
just before collapse

Figure 9.5.iI Analyses [E1 and (E101 - lOm and 6m high embankments on undrained
foundation, permeable surface layer, surface suction 1 OkPa - Velocity characteristics

268


Por, pressures on vertical sections

- After construction -

LL

—I— 50 -1 100-.
•.. _hL—L'
-'T J—. —.--- 50-' -
; -za ---.
—! _._—__._N _—' _. ._. 50 . -
S j___• ___?
— I 50 ,100-
S •
-,—
200kP9 : ¶5OkPa-

Befors construction

(a) end of construction

'—I—S—u—I—.

__- _-50
......dJ
—' 0—u—
__N •___•_• 50-I-
_I N__a. __.
, ___q
_I I—$

_S •
_.S
N__N
___,_._ui__N __.____,
___SS,_,_p __I

__I

- __,___—,___•__—.--_________.
_ -

150-.----
-
100-i-
::EEzEiiE _._ 100 kPa
i__V
i__S .
—, — - —s -:-- --:.--•-- 2OO1cpa

(b) 1.0 year after construction

Pore pressures on vertical sections

____— After construction
—
Just before collapse state
r—'{
\ \t__-_I_.
I
_•___•
ady state

_.-4--c 50

a 100 kPa
\ I

o Scale 25m
I a I I I

Pore pressure scale

0 100 200kPa
I I I

(ci) 9.75 years after construction (c2) 100.0 years after construction
- just before collapse

Figure 9.5.iv Analyses [Eli and [E101 - lOm and 6m high embankments on undrained
foundation, permeable surface layer, surface suction 1 OkPa - Contours of pore water

pressure

269
 T..TTI7

(a) 0.5 years after construction (d) 3.0 years after construction

(b) 1 .0 year after construction (e) 4.0 years after construction

o Scale 25m
I I

Ic) 2.0 years after construction (1) 4.25 years after construction

- just before collapse

Figure 9.6.i Analyses [E21 - 1 Om high embankment on drained foundation with impermeable

boundary, permeable surface layer, surface suction 1 OkPa - Incremental displacement

vectors showing the development of the superficial slip

270


(a) 0.5 years after construction (d) 3.0 years after construction

--

(b) 1 .0 year after construction (e) 4.0 years after Construction

0 Scale 25m
I I I I

0 Vector scale 5m
I J I IJ

(c) 2.0 years after construction (1) 4.25 years after construction
- just before collapse

Figure 9.6.i Analyses [E21 - 1 Om high embankment on drained foundation with impermeable
boundary, permeable surface layer, surface suction lOkPa - Accumulated
displacement vectors since the end-of-construction

271
 = 20%
e= 50%

(a) 0.5 years after construction (dl 3.0 years after construction

0=5%
e02O%

e0=50%
e0=20%
E0 100%

(hI 1.0 year after construction (e) 4.0 years after construction

E0 100%

0 Scale 25m
I I I I I

Ic) 2.0 years after construction (1) 4.25 years after construction
- just before collapse

Figure 9.611 Analyses [E21 1 Om high embankment on drained foundation with impermeable
boundary, permeable surface layer, surface suction 1 OkPa - Contours of accumulated
deviatoric plastic strain, showing the development of the superficial slip

272

(a) 0.5 years after construction (d) 3.0 years after construction

(b) 1.0 year after Construction (e) 4.0 years after construction

Scale 25m
I I I I

Ic) 2.0 years after construction It) 4.25 years after construction
- just before collapse

Figure 9.6.iv Analyses 1E21 - 1 Om high embankment on drained foundation with impermeable
boundary, permeable surface layer, surface suction 1 OkPa - Velocity characteristics
showing the development of the superficial slip

273
 Time since construction, t years
5 10 15 20
E 0
£
'0
0.2
4-
C 0.4
'\ [2J
E lOxlObOm/s
U [5] E= 0.95
0.8 k = 15x1010m/S
[4] E: 15x10'°m/s
0 1.0

Figure 9.7 Analyses 1E21. [E4J and [E51 - 1 Om high embankment on drained foundation with
impermeable boundary, permeable surface layer 1 surface suction 1 OkPa - Development
of mid-slope horizontal displacement with time

0.3
Ultimate equilibrium pore pressures
0.2 T
[3]
2u0
04)

! 0
0.0 10 20 30 40
Time since replacement, t years
0 0.1
00 replaced by ,- rupture surface
granular mater I
fill
—0.2
drained foundation Impermeable boundary
E Time since replacement, t years
10 20 30 40
'00
0.2
[3]

1:::

Figure 9.8 Analyses [E3] - 1 Om high embankment on drained foundation with impermeable
boundary - 'Excavation' of shallow slide from analysis (E21, and replacement with
granular fill with zero pore water pressure on lower boundary - Development of
average pore pressure ratio on the rupture surface, r. and mid-slope horizontal
displacement, 6, with time

274


'
.- - .. - ..•. ..

. . #1 1 I I I I I •
S #1191St III..

(a) just after removal of superficially (d) 5.0 years after replacement
failed slope

-- .- % % ' ..

11 ##Il111 1 a sill IIII

I g , • I I I
..,,s,,I. •. ..........

(b) just after replacement by tel 25.0 years after replacement

granular soil

S
.—.--. .- .,. 'S • •.. /
—

- :-

iS:.:.:

o Scale 25m
L i p

(c) 1.0 year after replacement (f) 31.75 years after replacement
- just before failure

Figure 9.9.1 Analyses [E3] - 1 Om high embankment on drained foundation with impermeable
boundary - 'Excavation' of shallow slide from analysis 1E2], and replacement with
granular fill with zero pore water pressure on lower boundary - Incremental
displacement vectors showing the development of the deep-seated slide

275

E020% eo= 20%

(a) just after removal of superficially (d) 5.0 years after replacement
failed slope

ffP=5%

e0=20% E0 50%

Ib) just after replacement by (a) 25.0 years after replacement

granular soil

E0 =5%

= 20%__ -

ED =20% 50%

0 Scale 25m
I I I .1

(c) 1 .0 year after replacement (1) 31.75 years after replacement
- just before failure

Figure 9.9i1 Analyses EE3] - 1 Om high embankment on drained foundation with impermeable
boundary - 'Excavation' of shaflow slide from analysis [E21. and replacement with
granular fill with zero pore water pressure on lower boundary - Contours of
accumulated deviatoric plastic strain, , showing the development of the deep-
seated slide

276

(a) just after removal of superficially (d) 5.0 years after replacement
failed slope

oc

(b) just after replacement by (e) 25.0 years after replacement

granular soil
Rupture surface

Scale 25m
I I I J

(c) 1 .0 year after replacement (1) 31 .75 years after replacement

- just before failure

Figure 9.9.iii Analyses 1E31 - lOm high embankment on drained foundation with impermeable

boundary - 'Excavation' of shallow slide from analysis 1E2], and replacement with
granular fill with zero pore water pressure on lower boundary - Velocity characteristics
showing the development of the deep-seated slide

277
 (Pa

(a) just after replacement by granular soil

Pors pressures on vertical sections

,—Att.r replacement
( Just before collapse
\ Steady state , )

—I
I 44N.I—N
— ' —' —i
I.-4 1
r r.r— ! -
-1 0
N 25
—s—s—•--.--._-e_._._._s —a •
—•—F_•—._,.-_._,_,_._.. _._e S 7

(b) 31.75 years after replacement - just before failure

0 Scale 25rn
I I I I I
Pressure scale
0 100 200 kPa

(c) long-term steady state condition

Figure 9..9iv Analyses [E31 - 1 Om high embankment on drained foundation with impermeable
boundary - 'Excavation' of shallow slide from analysts [E2j, and replacement with
granular fill with zero pore water pressure on lower boundary - Contours of pore water

pressure

278
 1,11111:: 4

I.

(a) 0.5 years after construction

(b) 1.0 year after construction

_AL:;:

(c) 2.0 years after construction

___T -!!r7 :-i: I

o Scale 25m
I I -t I

(dl) 2.35 years after construction (d2) 30.0 years after construction
- just before collapse

Figure 9.10.1 Analyses [E71 and [E81 - lOm and 6m high embankments on drained
foundation, permeable surface layer, surface suction OkPa - Incremental displacement
vectors

279
 (a) 0.5 years after construction

(b) 1 .0 year after construction

(ci 2.0 years after construction

0 Scale 25m
I I I I

(dli 2.35 years after construction (d2) 30.0 years after construction
- just before collapse

Figure 9.10 ii Analyses [E71 and EE81 - lOm and 6m high embankments on drained

foundation, permeable surface layer, surface suction OkPa - Contours of accumulated

deviatoric plastic strain, En"

280


20
term
I-J
!-20 I 2.0 years after construction
I I
1.0 year after construction
-4O

-60
I ears after consUUctiofl
-80
On$truction
-100

-120
(a) H =1 Om, Surface suction 1 OkPa

2.35 years - 2.0 years after

20
just before co coflstr,....Zr
(5
0.
,

-20
1.0 year after construction -r"
U)
U)
0)
I-

0)
0
0.
-60
years after construction
-80
end of construction

-100
/1C\\J k4N\
-120
(b) H = I Om, Surface suction OkPa (c) H = 6m, Surface suction OkPa

Scale
-I,
0 10 20 300m 10 20
I
permeable surface layer
shallow rupture surface-

fill
fill

I drained foundation

Figure 9.11 Analyses [E61, 1E71 and [E81 - I Om and 6m high embankments on drained
foundation, permeable surface layer, surface suction lOkPa (Analysis [E61) and OkPa
(Analyses [E7] and [8]) - Development of pore water pressure along the shallow
rupture surface with time

281
 (a) 0.5 years after construction

__._:!?.y:i:

(bi) 1.0 year after construction (b2) 1.0 year after construction
- just before collapse
Figure 9.12.1 Analyses [E81 and [E9] - 6m high embankment on drained foundation,
permeable surface layer, surface suction OkPa - Incremental displacement vectors

(a) 0.5 years after construction

0 Scale 25m
I I I

(bi) 1.0 year after construction (b2) 1.0 year after construction
- just before collapse
Figure 9.12.i Analyses [E81 and 1E91 - 6m high embankment on drained foundation.
permeable surface layer, surface suction OkPa - Contours of accumulated deviatonc
plastic strain, ED

282
 CHAPTER 10
CONCLUSIONS AND SUGGESTIONS FOR FURTHER RESEARCH

10.1 INTRODUCTION

The research work presented in this thesis can be divided into two parts. The first part deals
with the finite element analyses of rockfill dams. To model the pre-peak plastic behaviour of
rockfills, two Lade's elasto-plastic models, one having a double and the other a single yield
surface, have been incorporated into ICFEP. A wide range of available laboratory test data on
rockfills has been reproduced using these models and the behaviour of two rockfill dams with
an upstream asphaltic concrete membrane has been investigated.

In the second part of the thesis the modelling of progressive failure in delayed slides of cut
and embankment slopes made in and from stiff plastic clays is undertaken. The soil model
used was a non-linear elastic strain-softening plastic constitutive model which was
successfully employed in the finite element analysis of progressive failure of Carsington dam.
To simulate delayed sliding, coupled consolidation (swelling) was included in the present
work.

The main conclusions drawn from the thesis are summarized in this chapter. Some areas for
further research are also identified.

10.2 CONCLUSIONS

10.2.1 RockfilI Dams

To achieve reasonable accurate and useful predictions of stresses and deformations of rockfill
dams, the rockfill stress-strain behaviour has to be represented by an appropriate constitutive
law. 'Variable' elastic models have been widely used in the past, primarily because of their
simplicity and ability to model non-linear and stress dependent behaviour of fill materials. A
considerable amount of experience with this approach has shown some difficulties, both in
deriving model parameters from available test data, and in producing disp acement patterns
which agree with field measurements.

283
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
non-elastic components of rockf ill behaviour. In this respect elasto-plastic constitutive models
offer some advantages. However, the complexity of these models, in both formulation and
subsequent application, together with the difficulties associated with testing of the rockfill
materials has hampered their wider use In the finite element analysis of rockfill dams.

To account for pre-peak plasticity of rockfill behaviour, only elasto-plastic constitutive models
of the critical state type have been used to date. However, 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 clay. Rockfill can be considered as an
extreme granular material. Several characteristics of the stress-strain and strength behaviour
of rockfills (i.e. non-linearity, a higher stiffness on unloading and reloading, a curved failure
envelope, the effects of the intermediate principal effective stress on both strength and
stiffness, stress-path dependency, shear-dflatancy, more plastic response at higher stress
levels...), which are of importance in numerical analyses of dams, can be captured by the two
Lade's elasto-plastic constitutive models. It was therefore decided to incorporate these models
into ICFEP and employ them in the finite e ement analyses of two rockfill dams with an
upstream asphaltic concrete membrane.

Analyses of Roadford dam have been performed using three different constitutive models. In
addition to the two 'complex' Lade's elasto-plastic constitutive models, a 'simple' non-linear
elastic perfectly plastic model has been used for comparison purposes. Input parameters for
the analyses were derived from both large oedometer arid triaxial tests on the low grade
rockfill used for the dam construction. There is a deterministic way in derving Lade's model
parameters and it has been shown that they are capable of reproducing the full range of the
observed rockfill behaviour in the available laboratory tests. Although the non-linear elastic
perfectly plastic model produced a satisfactory fit to the available test data, the fit was not
so good as in the case of the Lade's models. Further comparison of the three constitutive
models when used to predict the rockfill behaviour along more complex and, for rockfill dams,
more representative stress paths has also revealed the superior performance of the Lade's
elasto-plastic constitutive models.

Comparison of the predictions with the observed behaviour during both construction and
reservoir impounding of Roadford dam indicated that while the predictions using both Lade's
models were in reasonable agreement with the field data this was not so for the analysis

284
using the non-linear elastic model'. The latter severely overpredicted lateral movements during
construction and underestimated membrane deflections during the initial stages of reservoir
impounding. These predictions could be improved by adjusting the non-linear elastic model
parameters, but then the laboratory test data could no longer be recovered.

Comparison of the two Lade's elasto-plastic constitutive models has shown that the double
hardening model is more suitable for the use in the finite element analyses of rockfill dams
than the single hardening model. The double hardening model is more sensitive to the changes
in stress path directions which occur during reservoir impounding. This is particularly so at
lower stress levels which usually operate in rockfill dams. Although more input parameters
are needed to characterize this model they are less interconnected than the single hardening
model parameters. For example, as few as three double hardening model parameters require
modification in order to account for differences in the initial conditions (dry unit weight,
moisture content) between the laboratory samples and rockfill as placed in the field.

Finite element analyses of Winscar dam have confirmed the capabilities of both Lade's
models. Again, they reasonable predicted dam movements during both construction and
reservoir impounding without adjusting the input parameters derived from the available large
laboratory test data. Moreover, the Lade's models are capable of predicting a pattern of
membrane movement during reservoir impounding which is nearly normal to the upstream
slope and in agreement with available field observations. This is in spite the fact that, due to
the assumption of isotropic hardening, the Lade's models cannot predict with accuracy the
rockfill response during large stress reversals which take place in the upstream toe area of a
dam for the higher levels of reservoir impounding.

10.2.2 Cut Slopes

The finite element analyses incorporating strain-softening and coupled consolidation (swelling)
reproduce delayed deep-seated sliding in slopes cut in stiff plastic clays which matches the
well established history of delayed s ps in railway cuttings. Collapse invohes significant
progressive failure and reduction in operational strength. This is induced by the large strains
resulting from swelling. These conclusons are drawn from analyses using peak and residual
strengths as measured in the laboratory, and by back-analysis of slips which are at residual
strength. No adjustment of these strengths to obtain a fit with field behaviour has been
necessary, and the use of the finite element analyses to reproduce deep-seated progressive
failure is validated.

'During the analysis of Roadford dam the plastic part of the model was not 'active'.

285
The parameters adopted for these analyses are based on London Clay, but are reasonably
representative of the range of British stiff plastic clays, and thus the conclusions drawn
should be of some generality.

The analyses indicate the following preliminary guidelines for susceptibility to deep-seated
sliding. The 3:1 slope, lOm high is unstable. The parametric studies show that the location
of the shear surface and/or the time to collapse depends on many factors, particularly the
value of K0 assumed. The 4:1 slopes behave in a similar manner. While the lOm high slope
is nearly stable, even in the very long-term, the 1 5m high s ope fails with pore pressures some
way from the predicted equilibrium values. For slopes 6m high, 2.5:1 is potentially unstable.
3:1 should be just stable. These conclusions arise when swelling is from the surface to near
hydrostatic conditions. Stability is improved by underdrainage.

The magnitude of the pore pressures which develop in the long-term is the principle cause of
uncertainty in the above predictions, and minor changes in pore pressure have a major effect
on stability. The time predicted for collapse is generally between 10 and 100 years. Since
predicted pore pressures change very slowly as equilibrium is approached, and the analyses
do not include the perturbations of pore pressure due to exceptionally wet weather, collapse
in the field may occur earlier than is predicted by an analysis incorporating monotonic
swelling.

The analyses suggest that measuring pore pressure and observing the generation of the shear
surface from the toe of the slope are the best ways of monitoring the development of
deep-seated slides. The analyses also predict that deep-seated slides will occur abruptly,
without much warning from increasing deformation.

No shallow slides are predicted in analyses 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 K0 , when early formation of a horizontal rupture surface probably reduced the
strain energy stored in the superficial 'cracked' layer.

Control of surface pore pressures looks to be the most powerful way of controlling the risk
of deep-seated slips in the higher slopes. Drainage can be built into repairs of shallow slips
which should remove the risk of future deep-seated ones developing later in the same slope.
Pore pressure can be controlled by surface trench dra ns, drilled in pipe drains, or by
controlled use of vegetation to increase evapo-transpiration.

286
10.2.3 Road Embankments

The analyses indicate that progressive failure occurs in embankments in both shallow and
deep-seated slides. The progressive failure predicted depends on the volumetric expansion in
the fill, which, in turn, depends on the modulus of swelling, and the water content and pore
water Suction at which plastic clay fill is placed.

Deep-seated slides can occur in high clay embankments if sufficiently high pore pressures are
generated, as in the case of a poorly drained embankment foundation. The amount of
progressive failure is significant and typically greater than in cutting slopes. It also occurs
abruptly, without much warning from increasing deformation. Swelling rates are likely to be
slower than in cutting slopes, and therefore the time scale for delayed failure longer.

Shallow slides are more complex, as they occur towards the base of a superficial layer of clay
in which summer shrinkage and winter swelling occurs. Th s may itself produce down-slope
movement, and loss of strength. Cyclic stress changes and loss of strength have not been
reproduced in the analyses. However, as in the cutting slopes, shallow slides have been
reproduced in analysis by incorporating a layer of higher permeability on the slope surface,
which swells more quickly and develops a strain discontinuity and a rupture surface at its
base. Superficially higher permeability has been observed in the field due to shrinkage cracks.
Shallow slides occur by similar mechanisms in both embankment and cutting slopes. Collapse
starts from the bottom of the slope and involves substantial progressive failure.

Shallow sliding will be inhibited by flatter and lower slopes, by increased strength, and by
reduced superficial shrinkage and swelling. The latter is prevented by a surface layer of free-
draining material, which prevents drying. It may well be that shallow sliding is also inhibited
by the controlled use of vegetation to reinforce the surface layer with roots and to decrease
pore pressures by increased evapo-transpiration. Increased surface evapo-transpiration or
drainage also has a beneficial effect at depth and inhibits deep-seated sliding.

Shallow slides are generally repaired with inert material wh ch is of higher strength than the
original clay, in which shrinkage and swelling no longer occur. Future superficial sliding is
therefore avoided. It is possible for a deep-seated slide to occur in a higher slope as swelling
continues after a shallow slide has occurred and been repaired. This risk should be avoided
if the repair incorporates drainage.

The risk of longer term deep-seated sliding in embankments can be monitored as for cutting
slopes, by pore pressure and lateral deformation measurement.

287
10.3 SUGGESTIONS FOR FURTHER RESEARCH

10.3.1 Rockfill Dams

One of the principal causes of uncertainty in the finite element analyses of rockfill dams is a
modelling of the effects of compaction. Field evidence suggests increased horizontal stresses
due to compaction of a layer as further fill is placed. In the analyses reported in this thesis the
increased horizontal stresses in a 'newly' constructed layer have been modelled. However,
it has been demonstrated that in this case the Lade's elasto-plastic constitutive models predict
a settlement profile which is dependent on the number of construction layers used in an
analysis. To overcome this dependency, either a large number of layers should be used, or
the lower horizontal stresses, determined by the K0-value predicted by a single oedometer
test, should be modelled (i.e. neglecting compaction induced stresses). The field compression
curves for heavily compacted rockfills can also show evidence of pre-consolidation due to
compaction. There is an option in ICFEP (not used in the present work) to take this into
account by adopting a value of overconsolidation ratio (OCR). This value can be estimated
from a pre-consolidation pressure which may, in turn, be related to the compactive effort,
initial conditions of the rockfill as placed in the field, compressive strength of a rockfill
particle... Unfortunately, a pre-consolidation pressure cannot be easily deduced, especially
from the laboratory tests, and this is an area where both further field and laboratory
investigations are needed.

Only two rockfill dams have been analyzed in this thesis. Both dams were of a 'homogeneous'
type with an upstream asphaltic concrete membrane which has little bearing on rockfill
deformation. In the future it is necessary to check the capabilities of the Lade's elasto-plastic
constitutive models by predicting the behaviour of 'more complex' rockfill dams with internal
clay cores, transition and filter zones for which extensive field measurements are available.
In their present form, both Lade's models can reasonable account for the laboratory behaviour
of a wide range of fill materials used for the construction of rockfill dams, and it is believed
that, using these models, a more realistic insight into the behaviour of these important
geotechnical structures can be achieved.

10.3.2 Cut Slopes and Road Embankments

In this thesis delayed slides in cuttings and road embankments have been analyzed on a
parametric basis. Good case histories of both deep-seated sliding in railway cuttings and
shallow sliding in motorway embankments exist for which a reasonable amount of field (and
laboratory) data are available. It would be useful to analyze some of these specific cases in
more detail.

288
The present finite element analyses of cuttings formed in London Clay strata have predicted
the generation of the shear surface from the toe of the slope. Field evidence of slopes cut
through both brown and blue London Clays shows the slips occurred at the junction between
these materials. Laboratory tests suggest different strengths of the brown and blue London
Clay. Preliminary analyses have shown that the introduction of the higher strength for the
unweathered, blue London clay is not enough to reproduce the final rupture surface
propagating entirely through the weathered, brown clay. However, it appears that a
discontinuous change in stiffness between these two strata is also necessary in order to
recover the observed collapse at the brown/blue London Clay junction. This should be further
investigated.

The time predicted for collapse varies directly with permeability adopted. A permeability
varying with depth (or constant for fills) has been used in all analyses reported here. At
present, there is an option in ICFEP to vary the soil permeability with the void ratio (or the
mean effective stress). Although there is little experimental or observational evidence for the
amount of volumetric swelling in both cuttings and road embankments it would be of benefit
to a employ a non-linear permeability in future analyses.

Embankment fills are partly saturated, and their behaviour is not fully reproduced in the
current finite element analyses. Inclusion of a non-linear compressible pore fluid In the
numerical model used here to characterize the behaviour of embankment fill material would
prove valuable.

Progressive failure in both cuttings and road embankments depends strongly on the strains
generated by swelling. There is little data on swelling rnoduli at the appropriate low stress
range, especially in partly saturated field compacted fills. Laboratory studies in this area would
be of considerable value. Field observations of the changes in water content accompanying
pore pressure equilibration and ultimate pore water pressures which can be deduced from
measurements in very old slopes would be of equal value. These laboratory and field studies
would be of paramount importance in calibrating numerical predictions.

289
 REFERENCES

ANDERSON, M.G. and KNEALE, P.E. (1980a), "Pore water pressure changes in a road
embankment", J. lnstn. Highway Engrs., May, pp.11-17

ANDERSON, M.G. and KNEALE, P.E. (1 980b), "Pore water pressure and stability conditions
on a motorway embankment", Earth Surface Processes, Vol.5, pp.37-46

APTED, J.P. (1977), "Effects of weathering on some geotechnical properties of London Clay",
PhD Thesis, Imperial College, University of London

ATKINSON, J.H. and FARRAR, D.M. (1985), "Stress path tests to measure soil strength
parameters for shallow landslips", Proc. 11th. Int. Conf. Soil Mechs. Found. Eng., San
Francisco, Vol.2, pp.983-986

BATHE, K. (1982), "Finite Element Procedures in Engineering Analysis", Prentice-Hall Inc.,

New Jersey

BERNANDER, S. (1985), "On limit criteria for plastic failure in strain-rate softening soils",
Proc. 11th lnt. Conf. Soil Mechs. Found. Eng., San Francisco, Vol.2, pp.397-400

BERNANDER, S. and OLOFFSON, I. (1981), "On formation of progressive failure in slopes",

Proc. 10th Int. Conf. Soil Mechs. Found. Eng., Stockholm, Vol.3, pp.357-362

BERNANDER, S. and GUSTAAS, H. (1984), "Consideration of in-situ stresses in clay slopes

with special reference to progressive failure analysis", Proc. 4th Int. Symp. Landslides,
Toronto, Vol.2, pp.235-240

BERTRAND, V. and LINO, M. (1985), "TRIADH and its application: Calculating to predict the
behaviour of Verney Dam during first impounding", Proc. 11th Int. Cont. Soil Mechs. Found.
Eng., Can Franc sco, Vol.5, pp.2621

BIONDI, P., MANFREDINI, F., MARTINETTI, S., RIBACCHI, R. and RICCIONI, R. (1976), "Limit
load of a foundation in a strain-softening soil", Proc. 2nd Int. Conf. Num. Meth. Geomechs,
Blacksburg, Vol.1, (ed. Desai), pp.591-596

290
BlOT, M.A. (1941), "General theory of three-dimensional consolidation", J. AppI. Phys.,
Vol.12, pp.155-i 64

BISHOP, A.W. (1966), "The strength of soils as engneering materials", Geotechnique,

Vol.16, No.2, pp.91-i 30

BISHOP, A.W. (1967), "Progressive failure - with special reference to the mechanism causing
it", Proc. Geotech. Conf., Oslo, Vol.2, pp.3-10

BISHOP, A.W. (1971a), "The influence of progressive failure on the choice of the method of
stability analysis", Geotechnique, Vol.21, No.2, pp.1 68-172

BISHOP, A.W. (1971b), "Shear strength parameters for undisturbed and remoulded soil
specimens", Stress-Strain Behaviour of Soils, Proc. Roscoe Memorial Symp., Cambridge
University, Foulis, pp.3-i 39

BISHOP, A.W. and 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.437-451

BISHOP, A.W., GREEN, G.E., GARGA, V.K., ANDERSEN, A. and BROWN, J.D. (1971), "A
new ring shear apparatus and its application to the measurement of residual strength",
Geotechnique, Vol.21, No.4, pp.273-328

BJERRUM. L. (1967), "Progressive failure in slopes of over-consolidated plastic clay and clay
shales", J. Soil Mechs. Found. Div., ASCE, Vol.93, No.SM5, pp.3-49

BLACKER, P. (1977), "Progressive failure: Its role in governing some aspects of soil
behaviour", MSc Dissertation, Imperial College, London

BOOKER, J.R. and SMALL, J.C. (1975), "An investigation of the stability of numerical
solutions of Biot's equations of consolidation", lnt. J. Sol ds Structures, Vol.11, pp.907-917

BOSCARDIN, M.D., SELIG, E.T., LIN, R.S. and YANG, GY. (1990), "Hyperbolic parameters
for compacted soils", J. Geot. Eng. Div., ASCE, Vol.116, No.GT1, pp.88-i 04

BOUGHTON, N.O. (1970), "Elastic analysis for behavour of rockfill", J. Soil Mechs. Found.
Div., ASCE, Vol.96, No.SM5, pp.1715-i 733

291
BRIDLE, R.C., VAUGHAN, P.R. and JONES, H.N. (1985), "Empingham Dam - Design,
construction and performance, Proc. lnstn. Civ. Engrs., London, Part 1, Vol.78, pp.247-289

BROMHEAD, EN. (1978), "Large landslides in London Clay at Herne Bay, Kent", Quart. Jour.
Engng. Geol., Vol.11, pp.291-304

BROMHEAD, E.N. and VAUGHAN, P.R. (1980), "Solutions for seepage in soils with an
effective stress dependent permeability", Proc. Conf. Num. Meth. Non-Linear Problems,
Swansea, Vol.1, pp.567-578

BURLAND, J.B. and HANCOCK, R.J.R. (1977), "Underground car park at the House of
Commons, London: Geotechnical aspects", Struct. Engng., Vol.55, pp.87-i 00

BURLAND, J.B., LONGWORTH, T.I. and 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

CARTER, J.P., BOOKER, J.R. and SMALL, J.C. (1979), "The analysis of finite elasto-ptastic
consolidation", mt. J. Num. Anal. Meth. Geomechs., Vol.3, pp.107-i 29

CATHIE, D.N. and DUNGAR, R. (1978), "Evaluation of finite element predictions for
constitutional behaviour of a rockfill dam", Proc. lnstn. Civ. Engrs., London, Vol.65, Part 2,
pp.551-568
Discussion (1979), Vol.67, Part 2, pp.233-234

CHAN, D.H. and MORGENSTERN, N.R., (1987), "Analysis of progressive deformation of the
Edmonton Convention Centre excavation", Can. Geotech. J., Vol.24, No.4, pp.430-440

CHANDLER, R.J. (1974), "Lies Clay: The long-term stability of cutting slopes, Geotechnique,
Vol.24, No.1, pp.24-38

CHANDLER, R.J. (1 984a), "Recent European experience of landslides in over-consolidated

clays and soft rocks, Proc. 4th lnt. Symp. Landslides, Toronto, Vol.1, pp.61-81

CHANDLER, R.J. (1 984b), "Delayed failure and observed strengths of first-time slides in stiff
clay", Proc. 4th lnt. Syrnp. Landslides, Toronto, Vol.2, pp.19-25

CHANDLER, R.J. and SKEMPTON, A.W. (1974), "The design of permanent cutting slopes in
stff fissured clays", Geotechnique, Vol.24, No.4, pp.457-464

292
CHANG, C.S. and DUNCAN, J.M. (1977), "Analysis of consolidation of earth and rockfill
dams", Geotechnical Engineering Report No. TE-77-3, US Army Corps of Engineers,
Waterways Experimental Station, Vicksburg, M ssissippi

CHARLES, J.A. (1 973), "Correlation between laboratory behaviour of rockfill and field
performance with particular reference to Scammonden Dam", PhD Thesis, Imperial College,
University of London

CHARLES, J.A. (1975), "Strains developed in two rockfill dams during construction",
Geotechnique, Vol.25, No.2, pp.321-331

CHARLES, J.A. (1976), "The use of one-dimensional compression tests and elastic theory in
predicting deformations of rockfill embankments", Can. Geotech. J., Vol.13, No.3, pp.189-
200 -

CHARLES, J.A. (1991a), "Laboratory shear strength tests and the stability of rockfill slopes",
Advances in Rockfill Structures, (ed. Maranha das Neves), pp.53-72

CHARLES, J.A. (1991b), "Laboratory compression tests and the deformation of rockfill
structures", Advances in Rockfill Structures, (ed. Maranha das Neves), pp.73-95

CHARLES, J.A. and WATTS, K.S. (1980), "The influence of confining pressure on the shear
strength of compacted rockfill", Geotechnique, Vol.30, No.4, pp.353-367

CHARLES, J.A. and SOARES, M.M. (1984), "The stability of slopes in soils with nonlinear
failure envelopes", Can. Geotech. J., Vol.21, No.3, pp.397-406

CHARLES, J.A. and WATTS, K.S. (1985), "Report on testing of rockfill for Roadford
Reservoir", Job No. W84/203, Building Research Establishment, Garston, UK

CHARLES, J.A. and PENMAN, A.D.M. (1988) "The behaviour of embankment dams with
bituminous watertight elements", Proc. 1 6th mt. Cong. Large Dams, San Francisco, Vol.2,
pp.693-705

CHAUDHURY, B. (1955), "Compressibility characteristics of compacted soils", MSc

Dissertation, Imperial Co lege, University of London

CHRISTIAN, J.T. and WHITMAN, R.V. (1969, "A one-dimensional model for progressive
failure", Proc. 7th Int. Conf. Soil Mechs. Found. Eng., Mexco City, Vol.2, pp.541-545

293
CLOUGH, R.W. and WOODWARD, R.J. (1967), "Analysis of embankment Stresses and
deformatIons", J. Soil Mechs. Found. Div., ASCE, Vol.93, No.SM4, pp.529-549

COOKE, J.B. (1984), "Progress in rockfill dams", J. Geot. Eng. Div., ASCE, Vol.110, No.10,
pp.1381-1414

CRABB, G.I., WEST, G. and O'REILLY, M.P. (1987), "Groundwater conditions in three
highway embankment slopes", Proc. 9th. Eur. Conf. Soil Mechs. Found. Eng., Dublin, Vol.1,
pp.401-406

CRABB, G.I. and ATKINSON, J.H. (1991), "Determination of soil strength parameters for the
analysis of highway slope failures", Proc. lnt. Conf. Slope Stability Engineering, lnstn. Civ.
Engrs., London, pp.13-18

DARAMOLA, 0. (1978), "The influence of stress-history on the deformation of a sand", PhD

Thesis, Imperial College, University of London

DAVIS, E.H. (1968), "Theories of plasticity and the failure of soil masses", Chapter 6 in 'Soil
Mechanics: Selected Topics', (ed. Lee), Butterworths, pp.341-380

DE LORY, F.A. (1957), "Long-term stability of slopes in over-consolidated clays", PhD Thesis,
Imperial College, University of London

DE MELLO, V.F.B. (1977), "Reflections on design decisions of practical significance to

embankment dams, Geotechnique, Vol.27, No.3, pp.281-354

DOUNIAS, G.T. (1987), "Progressive failure in embankment dams", PhD Thesis, Imperial
College, University of London

DOUNIAS, G.T., POTTS, D.M. and VAUGHAN, P.R. (1988), "Finite element analysis of
progressive failure : Two case studies", Computers and Geotechnics, Vol.6, pp.t55-1 75

DOUNIAS, G.T., POTTS, D.M. and 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 7/7/222, Build ng Research Establishment

DUNCAN, J.M. (1972), "Finite element analysis of stresses and movements in dams,
excavations and slopes", Proc. Symp. Applicat on of the FEM in Geotech. Engng., US Army
Corps of Engineers, Waterways Experimental Station, Vicksburg, pp.267-326

294
DUNCAN, J.M. (1992), "Static stability and deformation analysis", Proc. Spec. Conf. Stability
and Performance of Slopes and Embankments - II, Goat. Eng. Div., ASCE, Vol.1, (eds. Seed
and Boulanger), pp.222-266

DUNCAN, J.M. (1994), "The role of advanced constitutive relations in practical applications",
Proc. 13th mt. Conf. Soil Mechs. Found. Eng., New Delhi, Vol.5, pp.31-48

DUNCAN, J.M. and DUNLOP, P. (1969), "Slopes rn stiff-fissured clays and shales", J. Soil
Mechs. Found. Div., ASCE, Vol.95, No.SM2, pp.467-492

DUNCAN, J.M. and CHANG, C.Y. (1970), "Nonlinear analysis of stress and strain in soils",
J. Soil Mechs. Found. Div., ASCE, Vol.96, No.SM5, pp.1629-i 653

DUNCAN, J.M., OZAWA, V., LADE, P.V. and BOOKER, J.R. (1977), "An elasto-plastic stress-
strain relationship for cohesioniess soil", Specialty session 9: Constitutive equations in soils,
Proc. 9th Int. Conf. Soil Mechs. Found. Eng., Tokyo, pp.45-50

DUNCAN, J.M., BYRNE, P.M., WANG, K.S. and 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.UCB/GT/80-01, University of
California, Berkeley

DUNLOP, P. and DUNCAN, J.M. (1970), "Development of failure around excavated slopes",
J. Soil Mechs. Found. Div., ASCE, Vol.96, SM2, pp.471-493

EIGENBROD, K.D. (1975), "Analysis of the pore pressure changes following the excavation
of a slope", Can. Geotech. J., Vol.12, No.3, pp.429-440

EISENSTEIN, Z. (1974), "Application of finite element method to analysis of earth dams",

State-of-the-Art Report, Proc 1St Brazilian Seminar on Application of FEM in Soil Mechanics,
Rio do Janeiro, pp.457-528

EISENSTEIN, Z. and LAW, S.T.C., (1979a), "Influence of anisotropy on stresses and

displacements in embankments", Proc. 3rd lnt. Conf. Num. Meth. Geomechs., Aachen, Vol.2,
(ed. Wittke), pp.709-715

EISENSTEIN, Z. and LAW, S.T.C., (1979b), "The role of constitutive laws in analysis of
embankments, Proc. 3rd Int. Conf. Num. Meth. Geomechs., Aachen, Vol.4, (ed. Wittke),

pp.141 3-1 430

295
 EL-SOHBY, MA. (1969), "Deformation of sands under constant stress ratios", Proc. 7th mt.
Conf. Soil Mechs. Found. Eng., Mexico City, Vol.1, pp.111-119

EL-RUWAYIH, A. (1975), "Stress-strain characteristics of rockfill, and of clays under high

pore water tension", PhD Thesis, Imperial College, University of London

FARRAR, D.M. (1978), "Settlement and pore pressure dissipation within an embankment built
of London clay", Proc. Conf. Clay Fills, lnstn. Civ. Engrs., London, pp.101-106

FARRAR, D.M. (1990), "Observations of slope drainage at Romford, Essex", Transport and
Road Research Laboratory, Research Report No.302

FITZPATRICK, M.D., COLE, B.A., KINSTLER, F.L. and KNOOP, B.P., "Design of concrete-
faced rockfill dams", Proc. ASCE Symp. Design, Construction and Performance of Concrete-
Face Rockfill Dams, Denver, pp.410-434

GARGA, V.K. (1 970), "Residual strength under large strains and the effect of sample size on
the consolidation of fissure clay", PhD Thesis, Imperial College, University of London

GATES, R.H. (1972), "Progressive failure model for clay shale", Proc. Symp. Application of
the FEM in Geotech. Engng., US Army Corps of Engineers, Waterways Experimental Station,
Vicksburg, pp.327-348

GIBSON, R.E., SCHIFFMAN, R.L. and PU, S.L., (1970), "Plane strain and axially symmetric
consolidation of a clay layer on a smooth impervious base", Q.J. Mechs. AppI. Maths., Vol.23

GREENWOOD, J.R., HOLT, D.A arid HERRICK, G.W. (1985), "Shallow slips in highway
embankments constructed of over-conso'idated clay", Proc. Conf. Failure in Earthworks,
"Instn. Civ. Engrs., London, pp.79-92

GRIFFJTHS, D.V. and LI, C.O. (1993), "Analysis of delayed failure in slopng excavation", J.
Geotech. Eng., ASCE, Vol.119, No.9, pp.1 360-1 377

HAMZA, M.M.A.F. (1976), "The analysis of embankment dams by nonl near finite element
method", PhD thesis, Imperial College, University of London

HENKEL. D.J. (1957), "Investigations of two long-term failures in London Clay slopes at
Wood Green and Northolt", Proc. 4th mt. Corif. Soil Mechs. Found. Eng., London, Vol.2,
pp.378-381

296
HOEG, K. (1972), "Finite element analysis of strain-softening clay", J. Soil Mechs. Found.
Div., ASCE, Vol.98, No.SM1, pp.43-58

HWANG, C.T., MORGENSTERN, N.R. and MURRAY, D.W. (1971), "On solutions of plane
strain consolidation problems by finite element method", Can. Geotech. J., Vol.8, pp.109-118

IMAIZUMI, H. and SARDINHA, A.E. (1985), "A study of deformations in concrete faced
rockfill dams", Proc. ASCE Symp. Design, Construction and Performance of Concrete-Face
Rockfill Dams, Denver, pp.528-540

INGOLD, T.S. (1979), "The effects of compaction on retaining walls", Geotechnique, Vol.29,
No.3, pp.265-283

JAKY, J. (1944), "The coefficient of earth pressure at rest", J. Soc. Hungarian Architects and
Engineers, pp.355-358

JAMES, P.M. (1970), "Time effects and progressive failure in clay slopes", PhD thesis,
Imperial College, University of London

JAMES, P.M. (1971), "The role of progressive failure in clay slope", Proc. 1st Australia-New
Zealand Conf. Geomechs., Melbourne, Vol.1, pp.344-348

JANBU, N. (1963), "Soil compressibility as determined by oedometer and triaxial tests", Proc.
European Conf. Soil Mechs. Found. Eng., Wiesbaden, Vol.1, pp.19-25

JANBU, N. (1977), "Slopes and excavations in normally and lightiy over-consolidated clays",
General report: Slopes and excavations, Proc. 9th Int. Conf. Soil Mechs. Found. Eng., Tokyo,
Vol.2, pp.549-566

JOHNSON, P.E. (1985), "Maintenance and repair of highway embankments: Sthdies of seven
methods of treatment", Transport and Road Research Laboratory, Research Report No.30

KALTEZIOTIS, N.A. and KYROU, K. (1985), "Undraned bearing capacity of strain-softening

clay", Proc. 11th Int. Cont. Soil Mechs. Found. Eng., San Francisco, Vol.2, pp.773-776

KENNARD, M.F., PENMAN, A.D.M. and VAUGHAN, P.R. (1967), "Stress and strain
measurements in the clay core at Balderhead Dam", Proc. 9th Int. Cong. Large Dams,
Istanbul, Vol.3, pp.129-151

297
 KHALID, S., SINGH, B., NAYAK, G.C. and JAIN O.P. (1990), "Nonlinear analysis of concrete
face rockfill dam", J. Geotech. Eng., ASCE, Vol.116, No.5, pp.822-837

KIM, M.K. and LADE, P.V. (1988), "Single hardening constitutive model for frictional
materials, I. Plastic potential function", Computers and Geotechnics, Vol.5, pp.307-324

KONDNER, R.L. (1963), "Hyperbolic stress-strain response: Cohesive soils", J. Soil Mechs.
Found. Div., ASCE, Vol.89, No.SM1, pp.115-143

KULHAWY, F.H., DUNCAN, J.M. and SEED, H.B. (1969), "Finite element analyses of stresses
and movements in embankments during construction", Geotechnical Engineering Report No.
TE-69-4, University of California, Berkeley

KULHAWY, F.H. and DUNCAN, J.M. (1972), "Stresses and movements in Oroville Dam", J.
Soil Mechs. Found. Div., ASCE, Vol.98, SM7, pp.653-665

LA ROCHELLE, P., LEROUEIL, S. and TAVENAS. F. (1 989), "The short- and long-term stability
of clay slopes", De Mello Volume, Edgard Blucher Ltd, Sao Paulo, pp.240-251

LADE, P.V. (1977), "Elasto-plastic stress-strain theory for cohesionless soil with curved yield
surfaces", Int. J. Solids Structures, Vol.1 3, pp.1019-1035
S

LADE, P.V. (1978), "Prediction of undrained behaviour of sand", J. Geot. Eng. Div., ASCE,
Vol.104, No.GT6, pp.721-735

LADE, P.V. (1981), "Elasto-plastic stress-strain model for sand", Limit equilibrium, plasticity
and generalized stress-strain analysis in geotechnical engineering, ASCE, (eds. Yong and Ko),
pp.628-648

LADE, P.V. (1984), "Failure criterion for frictional materials", Mechanics of Engineering
Materials, (eds. Desai and Gallagher), Wiley, New York, pp.385-402

LADE, P.V. (1 988a), "Double hardening constitutive model for soils, parameter determination
and predictions for two sands", Constitutive Equations for Granular Non-cohesive Soils, (eds.
Saada and Bianchini), Balkema, pp.367-382

LADE, P.V. (1988b), "Model and parameters for the elastic behaviour of soils", Proc. 6th lnt.
Conf. Num. Meth. Geomechs., lnnsbruck, Austria, Vol.1, (ed. Swoboda), pp.359-364

298
LADE, P.V. and DUNCAN, J.M. (1975), "Elasto-plastic stress-strain theory for cohesionless
soil", J. Geot. Eng. Div., ASCE, Vol.101, No.GT1O, pp.1037-1058

LADE, P.V. and DUNCAN, J.M. (1976), "Stress-path dependent behaviour of cohesionless
soil", J. Geot. Eng. Div., ASCE, Vol.102, No.GT1, pp.51-68

LADE, P.V. and NELSON, R.B. (1984), "Incrementalization procedure for elasto-plastic
constitutive model with multiple, intersecting yield surfaces", mt. J. Num. Anal. Meth.
Geomechs., Vol.8, pp.311-323

LADE, P.V. and NELSON, R.B. (1987), "Modelling the elastic behaviour of granular materials",
Int. J. Num. Anal. Meth. Geomechs., Vol.11, pp.521-542

LADE, P.V. and KIM, M.K. (1988a), "Single hardening constitutive model for frictional
material, II. Yield criterion and plastic work contours", Computers and Geotechnics, Vol.6,
pp.13-29

LADE, P.V. and KIM, M.K. (1988b), "Single hardening constitutive model for frictional
materials, Ill. Comparisons with experimental data", Computers and Geotechnics, Vol.6,
pp.31-47

LADE, P.V. and PRADEL, D. (1989), "Comparison of single and double hardening constitutive
models for frictional materials", Proc. Int. Symp. Num Mod. Geomechs., Niagara Falls,
pp.147-154

LAMBE, T.W. (1973), "Predictions in soil engineering", Geotechnique, Vol.23, No.2, pp.14 .9-
202

LAMBE, T.W. and WHITMAN, R.V. (1969), "Soil Mechanics", Wiley, New York

LO, K.Y. and LEE, C.F. (1973a), "Stress analysis and slope stability in strain-softening
materials", Geotechnique, Vol.23, No.1, pp.1-li

LO, K.Y. and LEE, C.F. (1 973b), "Analysis of progressive failure in clay slopes", Proc. 8th Int.
Conf. Soil Mechs. Found. Eng., Moscow, Vol.1, pp.251-258

LUPINI, J.F., SKINNER, A.E. and VAUGHAN, P.R. (1981), "The drained residual strength of
cohesive soils", Geotechnique, Vol.31, No.2, pp.181-213

299
MATSUOKA, H. and NAKAI, T (1974) 1 "Stress-deformation and strength characteristics of
soil under three different principal stresses", Proc. JSCE, 232, pp.59-70

MORGENSTERN, N. (1977), "Slopes and excavations ifl heavily over-consolidated clays",

General report: Slopes and excavations, Proc. 9th Int. Cont. Soil Mechs. Found. Eng., Tokyo,
Vol.2, pp.567-581

MORGENSTERN, N. (1990), "Instablity mechanism in stiff soils and weak rocks", Proc. 10th
South Asian Geotech. Cont., Taipei, Vol.2, pp.27-36

MORGENSTERN, N. (1992), "The evaluation of slope stability - a 25 year perspective", Proc.

Spec. Conf. Stability and Performance of Slopes and Embankments - II, Geot. Eng. Div.,
ASCE, Vol.1, (eds. Seed and Boulanger), pp.1-26

MORI, R.T. and PINTO, N.L.de S. (1988), "Analysis of deformations in concrete face rockfill
dams to improve face movement prediction", Proc. 16th Int. Cong. Large Dams, San
Francisco, Vol.2, pp.27-34

McCARRON, W.F. and CHEN, W.F. (1988), "An elastic-plastic two surface model for non-
cohesive soils", Constitutive Equations for Granular Non-cohesive Soils, (eds. Saada and
Bianchini), Balkema, pp.427-445

NAYAK, G.C. and ZIENKIEWICZ, O.C. (1972), "Convenient form of stress invariants for
plasticity", J. Struct. Div., ASCE, Vol.98, No.ST4, pp.949-954

NAYLOR, D.J. (1974), "Stresses in nearly incompressible materials by finite elements with
application to the calculation of excess pore water pressure", mt. J. Num. Meth. Engng.,

Vol.8, pp.443-460

NAYLOR, D.J. (1975), "Numerical models for clay core dams", Proc. Int. Symp. Criteria and
Assumptions for Numerical Analyss of Dams, Swansea, (eds. Naylor, Stagg and Zienkiewicz),
pp.489-514

NAYLOR, D.J. (1991a), "Stress-strain laws and parameter values", Advances in Rockfill
Structures, (ed. Maranha das Neves), pp.269-290

NAYLOR, D.J. (1991 b), "Finite element methods for fills and embankment dams", Advances
in Rockfill Structures, (ed. Maranha das Neves), pp.291-340

300
 NAYLOR, D.J. and JONES, D.B., "The predictions of settlement within broad layered fills",
Geotechnique, Vol.23, No.4, pp.589-594

NAYLOR, D.J., PANDE, G.N., SIMPSON, 6. and TAAB, R. (1981), "Finite Elements in
Geotechnical Engineering a Pineridge Press, Swansea

NAYLOR, D.J., MARANHA DAS NEVES, E., MATTAR, D.Jr. and VEIGA PINTO, A.A. (1986),
"Prediction of construction performance of Beliche Dam", Geotechnique, Vol.36, No.3,
pp.359-376

NAYLOR, D.J. and MATTAR, D.Jr. (1988), "Layered analysis of embankment dams", Proc.
6th mt. Conf. Num. Meth. Geomechs., lnnsbruck, Vol.2, (ed. Swoboda), pp.1199-i 205

NAYLOR, D.J., KNIGHT, D.J. and DING, D. (1988), "Coupled consolidation analysis of the
construction and subsequent performance of Monsavu Dam", Computers and Geotechnics,
Vol.6, pp.95-i 29

NOVA, R. and WOOD, D.M. (1979), "A constitutive model for sand in triaxial compression",
lnt. J. Num. Anal. Meth. Geomechs., Vol.3, pp.255-278

NYAORO, D.L. (1989), "Analysis of soil-structure interaction by finite elements", PhD thesis,
Imperial College, University of London

OSTLID, H. (1978), Discussion, Proc. Conf. Clay Fills, lnstn. Civ. Engrs. London, pp.224-227

PALMER, A.C. and RICE, J.R. (1973), "The growth of slip surfaces in the progressive failure
of over-consolidated clay", Proc. Roy. Soc., London, A332, pp.527-548

PECK, R.B. (1967), "Stability of natural Clays", J. Soil Mechs. Found. Div., ASCE, Vol.93,
No.SM4, pp.403-417

PENMAN, A.D.M. (1986), "On the embankment dam", Geotechnique, Vol.36, No.3, pp.303-
348

PENMAN, A.D.M., BURLAND, J.B. and CHARLES, J.A. (1971), "Observed and predicted
deformations in a large embankment dam during construction", Proc. Instn. Civ. Engrs.,
London, Vol.49, pp.1-49
Discussion (1972), Vol.5 1, pp.729-756

301
PENMAN, A.D.M. and CHARLES, J.A. (1 973a), "Constructional deformations in rockfill dam",
J. Soil Mechs. Found. Div., ASCE, Vol.99, No.SM2, 139-1 63
Discussion (1973), Vol.99, No.SM12, pp.1170-1172

PENMAN, A.D.M. and CHARLES, J.A. (1973b), "Effect of the position of the core on the
behaviour of two rockfill dams", Proc. 11th lnt.Cong. Large Dams, Madrid, Vol.3, pp.315-319

PENMAN, A.D.M. and CHARLES, J.A. (1975), "Predicted deformation of the upstream
membrane of a rockfill dam", Ground Engineering, Vol.8, No.6, pp.47-48

PENMAN, A.D.M. and CHARLES, J.A. (1976), "The quality and suitability of rockfill used in
dam construction", Proc. 12th mt. Cong. Large Dams, Mexico City, Vol.1, pp.533-556

PENMAN A.D.M., CHARLES, J.A. and HUMPHREYS, J.D. (1982), "Sandstone rockfill in two
dams", Proc. 14th mt. Cong. Large Dams, Rio de Janeiro, Vol.4, pp.279-291

PENMAN, A.D.M. and CHARLES, J.A. (1985a), "Behaviour of rockfill dam with asphaltic
membrane", Proc. 11th mt. Conf. Soil Mechs. Found. Eng., San Francisco, Vol.4, pp.2011-
2014

PENMAN, A.D.M. and CHARLES, J.A. (1985b), "A comparison between observed and
predicted deformations of an embankment dam with central asphaltic core", Proc. 1 5th mt.
Cong. Large Dams, Lausanne, Vol.1, pp.1373-i 389

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

PERRY, J. (1991), "The extent and an analysis of shallow failures on the slopes of highway
earthworks", PhD Thesis, University of Durham

PIETRUSZCZAK, S. and MROZ, Z. (1981), "Finite element analysis of deformation of strain-

softening materials", mt. J. Num. Meth. in Engineering, Vol.17, pp.327-334

PIETRUSZCZAK, S. and NIU, X. (1992), "Numerical evaluation of bearing capacity of a

foundation in strain-softening soil", Computers and Geotechnics, Vol.13, pp.187-i 98

POTTS, D.M. and GENS, A. (1984), "The effect of the plastic potential in boundary value
problems involving plane strain deformation", mt. J. Num. Anal. Meth. Geomechs., Vol.8,
pp.259-286

302
POTTS, D.M. and GENS, A. (1985), "A critical assessment of methods of correcting for drift
from the yield surface in elasto-plastic finite element analysis", mt. J. Num. Anal. Math.
Geomechs., Vol.9, pp.149-159

POTTS, D.M., DOUNIAS, G.T and VAUGHAN, P.R. (1987), "Finite element analysis of the
direct shear box test", Geotechnique, Vol.37, No.1, pp.11-23

POTTS, D.M., DOUNIAS, G.T. and VAUGHAN, P.R. (1990), "FInite element analysis of
progressive failure of Carsington embankment", Geotechnique, Vo.4O, , pp.79-i 01

PREVOST, J.H. and HOEG, K. (1975), "Soil mechanics and plasticity analysis of strain-
softening", Geotechnique, Vol.25, No.3, pp.279-297

ROSCOE, K.H. and BURLAND, J.B. (1968), "On the generalized stress-strain behaviour of wet
clay", Engineering Plasticity, (eds. Heyman and Leckie), Cambridge University Press,
Cambridge, pp.535-609

ROWE, P.W. (1962), "The stress dilatancy relation for static equilibrium of an assembly of
particles in contact", Proc. Roy. Soc., London, A269, pp.500-527

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

SABOYA, F.Jr. and BYRNE, P.M. (1993), "Parameters for stress and deformation analysis of
rockfill dams", Can. Geotech. J., Vol.30, No.4, pp.690-701

SANDRONI, S.S. (1977), "The strength of London Clay in total and effective stress terms",
PhD Thesis, Imperial College, University of London

SCHIFMAN, R.L and GIBSON, R.E. (1964), "Consolidation of nonhomogeneous clay layers",
J. Soil Mechs. Found. Div., ASCE, Vol.90, No.SM5, pp.1-30

SCHIFMAN, R.L., CHEN, A.T-F. and JORDAN, J.C. (1969), "An analysis of consolidation
theories", J. Soil Mechs. Found. Div., ASCE, Vol.95, No.SM1, pp.285-312

SCHOFIELD, A.N. and WROTH, C.P. (1968), "Critical State So Mechanics", McGraw-Hill,
London

303
SECO E PINTO, P.S. and MARANHA DAS NEVES, E. (1985), "Hydraulic fracturing in zoned
earth and rockfill dams", Proc. 11th mt. Conf. Soil Mechs. Found. Eng., San Francisco, Vol.4,
pp.2025-2030

SEED, H.B., DUNCAN, J.M. and IDRISS, P.M. (1975), "Criteria and methods for static and
dynamic analysis of earth dams", Proc. mt. Symp. Criteria and Assumptions for Numerical
Analysis of Dams, Swansea, (eds. Naylor, Stagg and Zienkiewicz), pp.564-588

SKEMPTON, A.W. (1948), "The rate of softening in stiff fissured clay, with special reference
to London Clay", Proc. 2nd mt. Conf. Soil Mechs. Found. Eng., Rotterdam, Vol.2, pp.50-53

SKEMPTON, A.W. (1964), "Long term stability of clay slopes", Geotechnique, Vol.14, No.2,
77- 101

SKEMPTON, A.W. (1970), "First-time slides in over-consolidated clays", Geotechnique,

Vol.20, No.3, pp.320-324

SKEMPTON, A.W. (1977), "Slope stability of cuttings in brown London Clay", Proc. 9th mt.
Conf. Soil Mechs. Found. Eng., Tokyo, Vol.3, pp.261-270

SKEMPTON, A.W. (1985), "Geotechnical aspects of the Carsington dam failure", Proc. 11th
mt. Conf. Soil Mechs. Found. Eng., San Francisco, Vol.5, pp.2581-2591

SKEMPTON, A.W. and BROWN, J.D. (1961), "A landslide in boulder clay at Selset,
Yorkshire", Geotechriique, Vol.11, No.4, pp.280-293

SKEMPTON, A.W. and HUTCHINSON, J.N. (1969), "Stability of natural slopes and
embankment foundations", Proc. 7th mt. Conf. Soil Mechs. Found. Eng., Mexico City, State-
of-the-Art Volume, pp.291-340

SKEMPTON, A.W. and COATS, D.J. (1985), "Carsington dam failure", Proc. Conf. Failures
in Earthworks, Instn. Civ. Engrs., London, pp.203-220

SKEMPTON, A.W. and VAUGHAN, P.R. (1993), "The failure of Carsington Dam: causes and
mechanisms", Geotechnique, Vol.43, No.1, pp.1 51-173

SKERMER, N.A. (1973), "Finite element analysis of lnfiernillo Dam", Can. Geotech. J.,
Vol.10, No.2, pp.129-144

304
SKINNER, A.E. (1975), "The effect of high pore water pressures on the mechanical behaviour
of sediments". PhD Thesis, Imperial College, University of London

SLOAN, S.W., (1987). "Substepping schemes for numerical integration of elasto-plastic

stress-strain relations", mt. J. Mum. Meth. Eng., Vol.24, pp.893-911

SMALL, J.C., BOOKER, J.R. and DAVIS, E.H. (1976), "Elasto-plastic consolidation of soils",
hit. J. Solids Structures, Vol.12, pp.431-448

SMITH, LM., SHUTTLE, D.A., HICKS, M.A. and MOLENKAMP, F. (1988), "Prediction number
32", Constitutive Equations for Granular Non-cohesive Soils, (eds. Saada and Bianchini),
Balkema, pp.647-664

STURE, S. and KO, H.Y. (1976), "Stress analysis of strain-softening clay", Proc. 2nd mt.
Conf. Num. Meth. Geomechs., Blacksburg, led. Desai), Vol.1, pp.580-589

SYMES, M.J.P.R., GEMS, A. and HIGHT, D.W. (1984), "Undrained anisotropy and principal
stress rotation in saturated sand", Geotechnique, Vol.34, No.1, pp.11-27

TAYLOR, D.W. (1948), "Fundamentals of Soil Mechanics", Wiley, New York

TANAKA, T. and NAKANO, R. (1976), "Finite element analysis of Miyama rockfill dam", Proc.
2nd lnt. Conf. Num. Math. Geomechs., Blacksburg, Vol.2, (ed. Desal), pp.650-661

TANAKA, T. and KAWAMOTO, 0. (1989), "Plastic collapse analysis of slopes of strain-

softening materials", Proc. lrit. Symp. Num. Mod. Geomechs., Niagara Falls, pp.667-674

TEDD, P., PRICE, G., WILSON, A.C. and EVANS, J.D. (1991), "Use of the BRE electro-level
system to measure deflections of the upstream membrane of Roadford Dam", 3rd hit. Symp.
Field Measurements in Geomechanics, Oslo, Vol.1, pp.261-271

TERZAGHI, K. (1936), "Stability of slopes of natural clays", Proc. 1st mt. Conf. Soil Mechs.
Found. Eng., Harvard, Vol.1, pp.161-165

TERZAGHI, K. and PECK, R.B. (1948), "Soil Mechanics in Engineering Practice", Wiley, New
York

THOMAS, J.N. (1984), "An improved accelerated initial stress procedure for elasto-plastic
finite element analysis", mt. J. Num. Anal. Math. Geomechs., Vol.8, pp.359-379

305
TURNBULL, W.J. arid HVORSLEV, M.J. (1967), "Special problems in slope stability", J. Soil
Mechs. Found. Div., Vol.93, No.SM4, pp.499-528

VAN EEKELEN, H.A.M. (1980), "Isotropic yield surfaces in three dimensions for use in soil
mechanics", mt. J. Num. Anal. Meth. Geomechs., Vol.4, pp.89-101

VAUGHAN, P.R. (1971), "Undrained failure of clay embankments", Stress-Strain Behaviour

of Soils, Proc. Roscoe Memorial Symp., Cambridge University, Fou s, pp.683-69 1

VAUGHAN, P.R. (1977), "Stability of 8lopes of fill", Panel discussion, Main session 3: Slopes
and excavations, Proc. 9th lnt. Conf. Soil Mechs. Found. Eng., Tokyo, Vol.3, pp.411-413

VAUGHAN, P.R. (1978), "Engineering properties of clay fills", General report, Proc. Conf.
Clay Fills, lnstn. Civ. Engrs., London, pp.283-293

VAUGHAN, P.R. (1980), "Asphaltic concrete cores for dams", Unpublished Report for
Wimpey Laboratories Ltd

VAUGHAN, P.R. (1982), "Design and construction with wet fills", Special Lecture, Pub.
Associacao Brasileira de Mecanica dos Suelos - Nucleo Rehional de Sao Paulo

VAUGHAN, P.R. (1985), "Mechanical and hydraulic properties of tropical lateritic and
saprolitic soils, particularly as related to their structure and mineral components", 1st Int.
Conf. Geomechs Tropical Lateritic and Saprolitic Soils, Vol.3, pp.231-263

VAUGHAN, P.R. (1989), "Non-linearity in seepage problems - Theory and field observations",
De Mello Volume, Edgard Blucher Ltd, Sao Paulo, pp.501 -516

VAUGHAN, P.R. (1990), MSc Lecture Notes, Imperial College, University of London

VAUGHAN. P.R. (1991), "Stability analysis of deep slides in brittle soil - Lessons from
Carsirigton", Proc. mt. Corif. Slope Stability Engineering, lnstn. Civ. Engrs., London, pp.1-il

VAUGHAN, P.R. (1994), "Criteria for use of weak and weathered rock for embankment fill,
and its compaction control", Proc. 13th mt. Cont. Soil Mechs. Found. Eng., New Delhi, Vol.6,
To be pubi shed

306
VAUGHAN, P.R. and KENNARD, M.F. (1972), "Earth pressures at a junction between an
embankment dam and a concrete dam", Proc. 5th European Conf. Soil Mechs. Found. Eng.,
Madrid, Vol.1, pp.215-222

VAUGHAN, P.R. end WALBANCKE, H.J. (1973), "Pore pressure changes and the delayed
failure of cutting slopes in over-consol dated clay", Geotechnique, Vol.23, No.4, pp.531-539

VAUGHAN, P.R. and WALBANCKE, H.J. (1975), "The stability of cut and fill slopes in boulder
clay", Proc. Symp. Engineering Behaviour of Glacial Materials, Midland Soc. Soil Mechs.
Found. Eng., Birmingham, pp.209-219

VAUGHAN, P.R. and HAMZA, M.M. (1977), "Clay embankments and foundations: Monitoring
stability by measuring deformations", Specialty session 8: Deformation of earth-rockfill dams,
Proc. 9th mt. Conf. Soil Mechs. Found. Eng., Tokyo, pp.37-48

VAUGHAN, P.R., HIGHT, D.W., SODHA, V.G. end WALBANCKE, H.J (1978). "Factors
controlling the stability of clay fills in Britain", Proc. Conf. Clay Fills, lnstn. Civ. Engrs.,
London, pp.205-218

VAUGHAN, P.R., KENNARD, R.M. and GREENWOOD, D.A. (1983), "Squeeze grouting of stiff
fissured clay after a tunnel collapse", Proc. 5th Eur. Conf. Soil Mechs. Found. Eng., Helsinki,
Vol.1, pp.171-176

VAUGHAN, P.R., DOUNIAS, G.T. and POTTS, D.M. (1989), "Advances in analytical
techniques and the influence of core geometry on behaviour", Proc. Conf. Clay Barriers for
Embankment Dams, lnstn. Civ. Engrs.. London, pp.35-56

VEIGA PINTO, A. and MARANHA DAS NEVES, E. (1985), "Prediction of Beliche Dam
behaviour during reservoir filling", Proc. 1 Ith mt. Conf. Soil Mechs. Found. Eng., San
Francisco, Vol.4, pp.2021-2024

VERMEER, P.A. (1978), "A double hardening model for sand", Geotechriique, Vol.28, No.4,
pp.413-433

WALBANCKE, H.J. (1976), "Pore pressure in clay embankments and cuttings", PhD thesis,
Imperial College, Unversity of London

WILSON, A.C. and EVANS, J.D. (1990), "The use of low grade rockfilI at Roadford Dam",
The British Dam Society, 6th Conference, The Embankment Dam, London, pp.1 5-21

307
WOOD, D.M. (1990), "Soil Behaviour and Critical State Soil Mechanics. Cambridge
University Press, Cambridge

WROTH, C.P. and HOULSBY, G.T. (1985), "Soil mechanics - Property characterization and
analysis procedures", Proc. 11th mt. Conf. Soil Mechs. Found. Eng., San Francisco, Vol.1,
pp.1-55

YIM, K.P., HEUNG L.K. and GREENWAY D.R. (1988), "Effect of root reinforcement on the
stability of three fill slopes in Hong Kong", Proc. 2nd lnt. Conf. Tropical Soils, Singapore,
Vol.1, pp.293-299

YOSHIDA, N., MORGENSTERN, N.R. and CHAN, D.H. (1991), "Finite element analysis of
softening aspects in fissured, over-consolidated clays and mudstories", Can. Geotech. J.,
Vol.28, No.1, pp.51-61

YOSHIDA, N. and ADACH), T. (1992), "FE analysis of time-dependent instability of cut slopes
in clay shale", Proc. Spec. Conf. Stability and Performance of Slopes and Embankments - II,
Geot. Eng. Div., ASCE, (eds. Seed and Boulanger), Vol.1, pp.429-444

ZYTYNSKI, M., RANDOLPH, M.F., NOVA, R. and WROTH, C.P. (1978), "Short
communication on modelling the unloading-reloading behaviour of soils", lnt. J. Num. Anal.
Meth. Geomechs., Vol.2, pp.87-94

ZIENKIEWICZ, O.C. (1977), "The Finite Element Method", 3rd ed., McGraw-Hill, London

308
 APPENDIX

To provide an insight into the effect of modelling compaction stresses in a 'newly'

constructed layer on settlement profile during dam construction, the finite element 'layered'
analysis using a one-dimensional model (no lateral strain) has been performed.

A 50m high soil column is constructed in four, ten and twenty layers respectively. Fill is
modelled using the Lade's double hardening model' with the parameter values derived for the
Winscar sandstone rockfill (see Table 6.2).

In the first set of analyses the horizontal stresses in a 'newly' constructed layer were
determined by the 'standard' ICFEP procedure resulting in the value of K0 0.80 (see Section
2.4.3). In the second set of analyses the horizontal stresses were reduced to be equal to
K0 0.25 times the calculated vertical stresses, where K0 = 0.25 is a typical K0-value predicted
by the idealised, single element oedometer test on the sandstone rockfill using the Lade's
double hardening model (see Figure 64.1).

The results of this exercise are presented in Figure A.1. It can be seen that the second set
of analyses (broken lines with open symbols) predicts settlement profiles which are nearly
independent on the number of construction layers used. This is not so when the higher
horizontal stresses due to compaction were modelled as in the first set of analyses (full lines
with shaded symbols). However, the larger the number of layers employed, the more the
settlement profile approaches to that predicted by the second set of analyses.

For comparison purposes the settlement profile calculated using equation (3.5) is also included
(the thick full line). The stress-strain curve from a large oedometer test on Winstar sandstone
rockfill (see Figure 6.2) was fitted by equation (3.4) assigning the following values for the
dimensionless parameters C and D: C =0.0137 and D=0.52 (see Section 3.4.2.1).

'Similar results have been obtained using the Lade's single hardening model.

309
 50

40

30
E

0)
a,
I

20

10

0 100 200 300 400 500

Settlement, s [mm]

Figure A.1 Effect of number of layers and compacted stresses on settlement of a soil column

310

/ E'flL
( LC::
)
\ U: /